UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Aerosol type analysis with single wavelength, dual polarization elastic LIDAR Cottle, Paul Wesley 2016

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


24-ubc_2016_may_cottle_paul.pdf [ 53.23MB ]
JSON: 24-1.0225867.json
JSON-LD: 24-1.0225867-ld.json
RDF/XML (Pretty): 24-1.0225867-rdf.xml
RDF/JSON: 24-1.0225867-rdf.json
Turtle: 24-1.0225867-turtle.txt
N-Triples: 24-1.0225867-rdf-ntriples.txt
Original Record: 24-1.0225867-source.json
Full Text

Full Text

Aerosol type analysis with singlewavelength, dual polarization elasticLIDARbyPaul Wesley CottleB.Sc., Cornell University, 1999M.Sc., Johns Hopkins University, 2003A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinThe Faculty of Graduate and Postdoctoral Studies(Atmospheric Science)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)February 2016c© Paul Wesley Cottle 2016AbstractAerosols play an important role in many atmospheric processes but their highly heteroge-neous nature makes them difficult to study. Thus, new advancements in the field frequentlyfocus on finding ways to more accurately determine more information about aerosols as theyoccur. LIght Detection And Ranging (LIDAR) systems have become an important tool inthe study of aerosols because they can provide high resolution vertical profiles of quantitiesof interest (including aerosol concentrations, temperature, and wind speed, among others)over several kilometres of altitude. But on the other hand, the data can be ambiguousor difficult to correctly interpret and LIDAR systems can be costly and usually require agreat deal of technical expertise to maintain and operate. In recent years, technologicaldevelopments in lasers and detectors have led to the development of relatively inexpensiveLIDAR systems that are robust and simple to operate, but to date these single-wavelengthelastic LIDARs have provided only basic analysis products, such as determining the heightsof clouds or qualitative monitoring of aerosol layers. There is a need for more extensiveanalyses using these simpler LIDARs.To this end, an algorithm has been developed that employs ground-based, single-wavelength elastic LIDAR to create high resolution maps of aerosol and cloud types aswell as backscatter and extinction coefficients. Applications for maps such as these includestudies of long-range transport of aerosols, air quality within the planetary boundary layer,cloud-aerosol interactions, and visibility. Algorithms similar to this have been developed inthe past, but they have required either multi-wavelength LIDAR systems or have stoppedshort of differentiating between aerosol and cloud types. This algorithm also includes anovel utilization of depolarization ratio profiles for sub-layer discrimination. Thus far, thealgorithm has been applied to limited number of cases, resulting in a high degree of uncer-tainty compared to some more complex systems. The algorithm is thus merely a first step,and further refinements are suggested as a way to continue to improve its performance.iiPrefaceDescriptions of supporting data sourcesThe descriptions of supporting data sources in Section 3.2 are all based on descriptionsprovided by the data providers. The specific contributions are as follows:• Section 3.2.1 includes information about the HYbrid Single-Particle Lagrangian Inte-grated Trajectory (HYSPLIT) software provided by Air Resources Laboratory (ARL)for a similar description written by the author and published in [23]• Section 3.2.2 was written by me based on descriptions of the Navy Aerosol Analysisand Prediction System (NAAPS) model originally written by Dr.Douglas L Westphaltaken from [Westphal]Variance and wavelet analysisThe variance and wavelet analysis described in Section 4.1.2 is derived from the algorithmdeveloped for the STRucture of the ATmosphere (STRAT) LIDAR in [72].Figures taken from external sourcesSome of the figures in this document are reproductions of figures obtained from otherpublished papers or online resources. These are listed below:• Figure 4.5 is a reproduction of a figure shown in [44]• Figure 4.22 shows NAAPS and Navy Operational Global Atmospheric PredictionSystem (NOGAPS) results obtained from [Westphal]• Figure 5.1 shows an image generated from MODIS real colour imagery by Jeff Schmaltzand the MODIS Land rapid Responses Team [96]iiiPreviously published works• Figure 5.2 is a multi-panel figure including NAAPS results from [NRL] and HYSPLITresults generated by me using ARL software.• Figure 5.7 shows NAAPS results obtained from [NRL]• Figure 5.14 shows two images generated from MODIS real colour imagery by JeffSchmaltz and the MODIS Land rapid Responses Team [97, 98]• Figure 5.27 shows two images generated from MODIS real colour imagery by JeffSchmaltz and the MODIS Land rapid Responses Team [94, 95]• Figure 5.28 contains two panels of NAAPS data originating from [NRL] and oneNational Oceanic and Atmospheric Administration (NOAA) Hazard Mapping System(HMS) Fire and Smoke Product Map obtained from [UMBC-ALG]• Figure 5.29 contains three panels of NAAPS data originating from [NRL] and threeNOAA HMS Fire and Smoke Product Map obtained from [UMBC-ALG]• Figure A.1 is a reproduction of a figure from [34]Previously published worksA pervasive and persistent Asian dust event over North America duringspring 2010: LIDAR and sunphotometer observationsThe results reported in Section 5.1 are based in part on results published in [23]. I wasthe lead author on this paper and wrote most of the manuscript. The co-authors were Dr.Ian McKendry, Dr. Norm O’Neill, Auromeet Saha, and Dr. Kevin Strawbridge. With theexception of figures showing MODerate resolution Imaging Spectro-radiometer (MODIS)True-Color imagery and NAAPS data provide by Naval research Laboratory (NRL) (seelist of figures in previous section) all text, figures, analysis, and results reproduced fromthis paper were contributed solely by me.Long-range transport of Siberian wildfire smoke to British Columbia:LIDAR observations and air quality impactsThe results reported in Section 5.2 are based in part on results published in [22]. I wasthe lead author on this paper and wrote most of the manuscript. The co-authors were Dr.ivPreviously published worksIan McKendry and Dr. Kevin Strawbridge. With the exception of figures showing MODISTrue-Color imagery and NAAPS data provide by NRL (see list of figures in previous sec-tion) all text, figures, analysis, and results reproduced from this paper were contributedsolely by me.Studying Taklamakan aerosol properties with LIDAR (STAPL)The results reported in Section 4.2.3 are based in part on results published in [21]. I wasthe lead author on this paper and wrote most of the manuscript. The co-authors wereDr. Detlef Mueller, Dong-Ho Shin, Dr. Xiao-Xiao Zhang, Dr. Guanglong Feng, Dr. IanMcKendry, and Dr. Kevin Strawbridge. This paper includes research performed by Dr.Zhang and myself in Aksu, Xinxiang Province, P.R. China. This research performed wassupported by the National Natural Science Foundation of China (Grant No. 41171019),One Hundred Talented Researchers Program of Chinese Academy of Sciences and WestLight Foundation of Chinese Academy of Sciences (No. XBBS201104). The majority ofthe field work was performed by Dr. Zhang. All figures, analysis, and results presentedherein were generated solely by me.vTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ixList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xList of Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiiiList of Acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xviAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xixDedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxi1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Thesis goals and objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.1 Light propagation and scattering in the atmosphere . . . . . . . . . . . . . 52.1.1 Extinction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.1.2 Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.2 The standard LIDAR equation . . . . . . . . . . . . . . . . . . . . . . . . . 112.2.1 The LIDAR ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152.3 Optical properties of clouds and aerosols . . . . . . . . . . . . . . . . . . . 162.3.1 Scattering in clouds . . . . . . . . . . . . . . . . . . . . . . . . . . . 16viTable of Contents2.3.2 Scattering in aerosols . . . . . . . . . . . . . . . . . . . . . . . . . . 182.3.3 Aerosol summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273 Data collection methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . 293.1 LIDAR systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293.1.1 CORALNet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293.1.2 Mini-Micropulse LIDAR . . . . . . . . . . . . . . . . . . . . . . . . 303.2 Supporting data sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313.2.1 HYSPLIT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313.2.2 NAAPS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324 Layer discrimination algorithm development . . . . . . . . . . . . . . . . 344.1 Layer analysis techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344.1.1 Step #1 - Pre-processing . . . . . . . . . . . . . . . . . . . . . . . . 354.1.2 Step #2 - Variance and wavelet analysis . . . . . . . . . . . . . . . 424.1.3 Step #3 - Layer type discrimination . . . . . . . . . . . . . . . . . . 464.1.4 Step #4 - Calculation of backscatter and extinction coefficient pro-files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534.2 Evaluation of results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 554.2.1 Quantitative tests using idealized and empirical profiles . . . . . . . 564.2.2 Qualitative tests using empirical data . . . . . . . . . . . . . . . . . 724.2.3 Ongoing issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 854.3 Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 925 Research applications for layer discrimination . . . . . . . . . . . . . . . . 935.1 Case Study #1: A Pervasive and persistent dust episode in North America- 2010 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 945.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 945.1.2 NAAPS and HYSPLIT Model Results . . . . . . . . . . . . . . . . 975.1.3 CORALNet observations and LDA results . . . . . . . . . . . . . . 1005.1.4 Analysis of trends in depolarization . . . . . . . . . . . . . . . . . . 1185.1.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1205.2 Case Study # 2: Siberian wildfire smoke transport - 2012 . . . . . . . . . . 1235.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1235.2.2 Smoke emission and transport . . . . . . . . . . . . . . . . . . . . . 125viiTable of Contents5.2.3 CORALNet observations and LDA results . . . . . . . . . . . . . . 1285.2.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1425.3 Case Study #3 - Analysis of wildfire haze events in Vancouver, July andAugust 2015 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1475.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1475.3.2 Smoke emission and transport . . . . . . . . . . . . . . . . . . . . . 1485.3.3 Mini-MPL observations and layer analysis . . . . . . . . . . . . . . 1525.3.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1655.4 Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1666 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1686.1 Summary of results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1686.1.1 Use of layer masks for interpretation of LIDAR results . . . . . . . 1686.1.2 Calculation of extinction and backscatter using layer-specific LIDARratios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1696.1.3 Layer masks applied to long-term analysis of aerosol trends . . . . . 1706.2 Plans for future research and development . . . . . . . . . . . . . . . . . . 1706.2.1 Software development . . . . . . . . . . . . . . . . . . . . . . . . . . 1706.2.2 Study incorporating atmospheric sounding . . . . . . . . . . . . . . 1706.2.3 ”Coastal” Setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1716.2.4 Long-term comparison to CALIPSO VFM . . . . . . . . . . . . . . 1716.2.5 Improving thresholds and PBL edge detection . . . . . . . . . . . . 1716.2.6 Iterative LIDAR ratio optimization . . . . . . . . . . . . . . . . . . 1726.2.7 Examples of potential future applications . . . . . . . . . . . . . . . 1756.3 Closing remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177AppendixA LIDAR technical details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190A.1 Mini-micropulse lidar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190A.2 CORALNet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194A.2.1 CORALNet data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194viiiList of Tables2.1 Values for dust depolarization and LIDAR ratios . . . . . . . . . . . . . . . 202.2 Values for smoke depolarization and LIDAR ratios . . . . . . . . . . . . . . 222.3 Values for urban depolarization and LIDAR ratios . . . . . . . . . . . . . . 242.4 Values for marine depolarization and LIDAR ratios . . . . . . . . . . . . . . 262.5 Table of layer identification categories used in this analysis . . . . . . . . . 275.1 Smoke layer depolarization and normalized relative backscatter statistics forJuly, 2015 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1575.2 Smoke layer depolarization and normalized relative backscatter statistics forJuly-August, 2015 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161A.1 Table comparing polarization states along the Mini-MicroPulse LIDAR (mMPL)light path . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193ixList of Figures3.1 Map of locations where LIDAR data were collected for the various studiesincluded in this thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304.1 Examples of wavelets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444.2 Sample wavelet analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 454.3 Histogram of cloud and aerosol NRB values . . . . . . . . . . . . . . . . . . 484.4 Histogram of cloud and aerosol NRB values for depolarization > 0.25 . . . . 494.5 Typical CALIPSO Cloud signatures . . . . . . . . . . . . . . . . . . . . . . 514.6 NRB vs. Depolarization for three cloud types . . . . . . . . . . . . . . . . . 524.7 Example of idealized LIDAR profile . . . . . . . . . . . . . . . . . . . . . . 584.8 Test results from single layer: βp = 1e−3 km−1sr−1,R = 65 sr . . . . . . . . 604.9 Test results from single layer: βp = 1e−2 km−1sr−1,R = 65 sr . . . . . . . . 614.10 Test results from single layer: βp = 1e−4 km−1sr−1,R = 65 sr . . . . . . . . 624.11 Results of Monte-Carlo simulations of inversions with noise with no layers . 644.12 Results of Monte-Carlo simulations of inversions with noise with one layer . 664.13 Results of Monte-Carlo simulations of inversions with noise with one layerand LIDAR ratio error of +50% . . . . . . . . . . . . . . . . . . . . . . . . . 674.14 Results of Monte-Carlo simulations of inversions with noise with one layerand LIDAR ratio error of -50% . . . . . . . . . . . . . . . . . . . . . . . . . 684.15 LIDAR ratio sensitivity analysis - July 2015 . . . . . . . . . . . . . . . . . . 704.16 LIDAR ratio sensitivity analysis - July 2015 . . . . . . . . . . . . . . . . . . 714.17 Mini-MPL data from Whistler-March 27, 2014 . . . . . . . . . . . . . . . . 734.18 Backscatter and extinction coefficients from Whistler-March 27, 2014 . . . . 754.19 Mini-MPL data from Ucluelet-May 03, 2014 . . . . . . . . . . . . . . . . . . 774.20 Backscatter and extinction coefficients from Ucluelet-May 03, 2014 . . . . . 794.21 Mini-MPL data from Ucluelet-May 06, 2014 . . . . . . . . . . . . . . . . . . 814.22 NAAPS Results for May 06, 2014 . . . . . . . . . . . . . . . . . . . . . . . . 83xList of Figures4.23 Backscatter and extinction coefficients from Ucluelet-May 06, 2014 . . . . . 844.24 Mini-MPL data from Aksu, April 25-27, 2013 . . . . . . . . . . . . . . . . . 884.25 Box plot showing range of depolarization values for the lower 4 km of altitudeover Aksu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 904.26 Layer type mask for Aksu, April 25-26, 2013 . . . . . . . . . . . . . . . . . . 915.1 MODIS image of dust over East China from Mach 12, 2010 . . . . . . . . . 965.2 NAAPS global dust model results alongside frequency plots of HYSPLITback trajectories for peak days during primary dust transport events of 2010 995.3 CORALNet false colour plots for dust event on March 17-20, 2010 . . . . . 1015.4 Backscatter and extinction for March 17-20 dust event . . . . . . . . . . . . 1035.5 CORALNet false colour plots for dust event on March 24-25, 2010 . . . . . 1055.6 Backscatter and extinction for March 24-25 dust event . . . . . . . . . . . . 1075.7 NAAPS-NOGAPS time-height sections from Cheeka Peak, WA . . . . . . . 1095.8 LIDARobservations for Vancouver, BC from 09-12 April, 2010 . . . . . . . . 1115.9 Backscatter and extinction for April 09-12 dust event . . . . . . . . . . . . . 1135.10 LIDARobservations and layer mask for Egbert, ON from 18-21 April, 2010 1155.11 Backscatter and extinction for April 18-21 dust event . . . . . . . . . . . . . 1175.12 2-D histograms of dust layer depolarization vs. altitude for 2010 dust events 1215.13 2-D histograms of dust layer depolarization vs. NRB for 2010 dust events . 1225.14 MODIS-Terra images of smoke layers observed on both sides of the PacificOcean, summer 2012 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1245.15 NAAPS and HYSPLIT model results for select days from summer 2012 . . 1265.16 CORALNet UBC LIDARobservations for July 04-14, 2012 . . . . . . . . . . 1295.17 CORALNet UBC LDA for July 05-10, 2012 . . . . . . . . . . . . . . . . . . 1315.18 CORALNet UBC backscatter and extinction plots for July 05-10, 2012 . . . 1335.19 CORALNet UBC LIDARobservations for August 04-16, 2012 . . . . . . . . 1355.20 CORALNet UBC LDA for August 09-15, 2012 . . . . . . . . . . . . . . . . 1375.21 CORALNet UBC backscatter and extinction plots for August 09-15, 2012 . 1395.22 Vertical profiles taken from CORALNet UBC LIDARobservations on July07, 09 and 14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1435.23 Vertical profiles taken from CORALNet UBC LIDARobservations on August06, 12, and 14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1445.24 2-D histogram of depolarization vs. altitude for 2012 smoke events . . . . . 145xiList of Figures5.25 2-D histogram of depolarization vs. BR for smoke events in July, August 20121465.26 Histogram of depolarization ratios for smoke events in July, August 2012 . . 1465.27 MODIS images of smoke over Vancouver on July 05 and July 06, 2015 . . . 1475.28 NAAPS and HMS Smoke Product data: July 07, 2015 . . . . . . . . . . . . 1505.29 NAAPS and HMS Smoke Product data: August 11-25, 2015 . . . . . . . . . 1515.30 Mini-MPL measurements of smoke event in Vancouver, July 06-10, 2015 . . 1535.31 Daily depolarization ratios and NRB values for smoke layers in Vancouver-July 2015 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1565.32 Mini-MPL measurements from Vancouver, August 10-14, 2015 . . . . . . . 1585.33 Backscatter and extinction coefficients from Vancouver, August 10-14, 2015 1605.34 Daily depolarization ratios for smoke layers in Vancouver-August 2015 . . . 1625.35 2-D histogram of depolarization vs. altitude for smoke events in July, August2015 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1635.36 2-D histogram of depolarization vs. NRB for smoke events in July, August2015 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1645.37 Histogram of depolarization ratios for smoke events in July, August 2015 . . 165A.1 mMPL layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191xiiList of SymbolsA Area of receiver entrance aperture km2B Normalized relative backscatter counts∗km2/µs∗µJB1 Normalized relative backscatter channel 1(B⊥(φ = 0))counts∗km2/µs∗µJB2 Normalized relative backscatter channel 2(B⊥(φ = pi/2))counts∗km2/µs∗µJC Composite correction factor for lidar profilesthat is range- and time-independent (C =A∆zη)-D Atmospheric molecular number density cm−3Ds Standard molecular number density(2.547× 1019 cm−3)cm−3E Transmitted laser pulse energy µJE0 Laser pulse energy µJG Unitless scaling factor used to represent RMSnoise in lidar profiles in terms of shot noises-k Imaginary component of complex index of re-fraction-L Radiance W/m2srL‖ Radiance polarized parallel to laser pulse W/m2srL⊥ Radiance polarized perpendicular to laserpulseW/m2srm Complex index of refraction (m = n+ ik) -N Number of scatterers (particles or molecules)within a specified volume#xiiiList of Symbolsn Real component of complex index of refrac-tion-ns Real component of complex index of refrac-tion for air at STP-O Lidar overlap function -P Atmospheric pressure PaPs Standard pressure (101.325 kPa) PaQabs Absorption efficiency -Qext Extinction efficiency -Qscat Scattering efficiency -R Lidar ratio (extinction to backscatter ratio) srr Particle radius µmRm Molecular (Rayleigh) scattering lidar ratio srS Measured signal at the detector counts/µsS0 Signal at the detector from backscatteredlaser pulsecounts/µsSB Background signal counts/µsSD Dark current signal counts/µsS‖ Signal at the detector from scattered light po-larized parallel to laser pulsecounts/µsS⊥ Signal at the detector from scattered light po-larized perpendicular to laser pulsecounts/µsT Atmospheric temperature KT r0 Optical transmittance from rage 0 to r -Tm Molecular optical transmittance -Tp Particulate optical transmittance -Ts Standard temperature (273.15 K) KU Natural log of the range-corrected lidar signal -V Natural log of the normalized relativebackscatter B-qvar Threshold beyond which variance in a lidar isassumed to be due to particulate scattering-x Scattering size parameter -z Range kmxivList of Symbolsα Extinction coefficient km−1αm Molecular extinction coefficient km−1αp Particulate extinction coefficient km−1β Scattering coefficient km−1sr−1βpi 180◦ Backscatter coefficient km−1sr−1βm Molecular backscatter coefficient km−1sr−1βp Particulate backscatter coefficient km−1sr−1β‖ Backscatter polarized parallel to laser pulse km−1sr−1β⊥ Backscatter polarized perpendicular to laserpulsekm−1sr−1βT Total backscatter coefficient (sum of βm andβp)km−1sr−1δ Depolarization ratio -δMPL MPL depolarization ratio -η Depolarization factor for Rayleigh scattering -Γ Backscatter Ratio -γ Lidar system efficiency constant counts/µs∗µJκ Exponent in lidar ratio power law relationship(typically assumed to be unity)-λ Wavelength µmΦ Radiant flux Wφ Phase angle radΨ Molecular scaling factor -σB Shot noise from background signal SB counts/µsσD Shot noise from dark current signal SD counts/µsσS Shot noise from lidar signal S0 counts/µsσT Total RMS detector noise counts/µsτ Optical depth -θ Scattering angle radω Variance of a lidar signal around a normalizedvalue representing pure Rayleigh scattering-xvList of AcronymsAERONET AEROsol RObotic NETworkANSI American National Standards InstituteAOD Aerosol Optical DepthAPD Avalanche Photo-DiodeARL Air Resources LaboratoryAVHRR Advanced Very High Resolution RadiometerBR Backscatter RatioBTD Back Trajectory DurationCALIPSO Cloud-Aerosol LIDAR and InfraredPathfinder Satellite ObservationsCCN Cloud Condensation NucleiCIMSS Cooperative Institute for MeteorologicalSatellite StudiesCORALNet Canadian Operational Research Aerosol LI-DAR NetworkCWT Continuous Wavelet TransformFLAMBE Fire Locating and Modelling of BurningEmissionsFOV Field Of ViewFT Free TroposphereGDAS Global Data Assimilation SystemGFED Global Fire Emissions DatabaseGOES Geostationary Operational EnvironmentalSatelliteGSFC Goddarrd SpaceFlight CenterHMS Hazard Mapping SystemxviList of AcronymsHSRL High Spectral Resolution LIDARHYSPLIT HYbrid Single-Particle Lagrangian IntegratedTrajectoryLCR Liquid Crystal RetarderLDA Layer Discrimination AlgorithmLIDAR LIght Detection And RangingmMPL Mini-MicroPulse LIDARMODIS MODerate resolution Imaging Spectro-radiometerMPE Maximum Permissible ExposureMPL Micro-Pulse LIDARNAAPS Navy Aerosol Analysis and Prediction SystemNCDC National Climactic Data CentreNCEP National Centre for Environmental PredictionNESDIS National Environmental Satellite Data andInformation ServiceNOAA National Oceanic and Atmospheric Adminis-trationNOGAPS Navy Operational Global Atmospheric Pre-diction SystemNRB Normalized Relative BackscatterNRL Naval research LaboratoryORA Office of Research and ApplicationsPBL Planetary Boundary LayerPMT Photo-Multiplier TubePRF Pulse Repetition FrequencyRMS Root Mean SquaredSNR Signal to Noise RatioSPCM Single Photon Counting ModuleSTRAT STRucture of the ATmosphereTSP Total Suspended ParticulateUBC University of British ColumbiaUSGS United States Geological SurveyVFM Vertical Feature MaskxviiList of AcronymsWF ABBA WildFire Automated Biomass Burning Algo-rithmxviiiAcknowledgementsFirst and foremost, I must acknowledge the multiple professional and personal contributionsof my adviser, Dr. Ian McKendry. Ian is nothing short of the epitome of what a univer-sity professor and researcher should be: intelligent, down-to-earth, engaging, attentive,helpful when critical and inspirational in his achievements and his scrupulous intellectualstandards. Not every brilliant researcher is good at working with students, and not ev-ery gifted teacher has the time and inclination to stay at the forefront of his/her field ofresearch. I have been lucky enough to work with someone who excels at both, and hiscontribution to my success cannot be overstated.As a close second, I would like to acknowledge the contributions of the other (current)members of my Ph.D. committee: Dr. Phil Austin and Dr. Douw Steyn. It would bedifficult to accurately assess how many times, and in how many ways these men havesteered my efforts away from pointless distractions and toward useful areas of inquiry,both in and out of the classroom.I would like to acknowledge Dr. Allan Bertram and the CREATE-AAP program hefounded for providing me with support that reached far beyond the financial contributionsof the scholarship. CREATE-AAP is one of those programs that succeeds in reclaiming abuzzword like “multi-disciplinary” by making an earnest and (even more rare!) successfuleffort to merge people with similar interests from very different fields and get them toactually share ideas and collaborate. With me.I gratefully acknowledge the NOAA ARL for the provision of the HYSPLIT trans-port and dispersion model used repeatedly in this publication. I thank the team at NRL-Monterrey and the Office of Naval research for developing NAAPS and continuing to main-tain the NRL/Monterrey Aerosol website (http://www.nrlmry.navy.mil/aerosol web/). Iwould like to thank Michael Travis and Bernard Firanski of the CORALNet LIDAR teamat Environment Canada for valuable support and feedback in producing and interpretingCORALNet LIDAR images. I would like to thank Roland Stull and Roland Schigas ofthe University of British Columbia Geophysical Disaster Computational Fluid DynamicsxixAcknowledgementsCentre for providing BlueSky data for the Siberian Smoke transport study. Special thanksalso go to SigmaSpace Corp. and the Chinese Academy of Sciences for donating the use ofthe mMPL for the Aksu study, and especially to Yunhui Zheng and Savyasachee Mathurwithout whose tireless support it would not have been possible.All data processing and analysis programs were written using the Python v2.7 program-ming language, and the SciPy, NumPy, and Pandas packages were particularly helpful.Unless otherwise specified, figures were generated using the matplotlib library [48].xxDedicationI dedicate this ... whatever it is ... to my daughter Margot. Years from now, when youare explaining to your therapist why your father missed the first year of your life, you cannow add that at least you were mentioned in a long-winded report on a mildly interestingtopic that no-one ever read. You’re welcome, Margot!xxiQuote”In streams of light I clearly sawThe dust you seldom see,Out of which the Nameless makesA Name for one like me.”- Leonard CohenxxiiChapter 1IntroductionLIght Detection And Ranging (LIDAR) is a common name for a wide range of remotedetection instruments. In simplest terms, LIDAR instruments comprise three elements: apulsed laser light source, a receiver module that consists of a telescope and an aft-opticsassembly used to filter and focus the received light, and a detector assembly usually consist-ing of some number of Photo-Multiplier Tubes (PMTs), Avalanche Photo-Diodes (APDs)or other high-gain photoelectric modules. Using these components, LIDARs collect time-resolved measurements of reflection of pulsed laser light from remote targets. Dependingon the application, the reflected light could be from a hard target (as for laser altimeters)or a combined signal from multiple scattering particles within a given volume of integrationdefined by the receiver Field Of View (FOV) and the integration period of the detector(s)(as for aerosol LIDAR). The time that elapses between the emission of the laser pulseand the detection of the reflected light provides information about the range and otherinformation about the scattering target(s) is provided by the measured signal strength inthe detector channel(s). The focus of this thesis is specifically on single-wavelength, dualpolarization elastic LIDAR. This is a system that emits laser light in one wavelength anddetects scattered light in that same wavelength, but in two orthogonal polarizations.Today, aerosol LIDAR systems are unique among instruments used to investigate theatmosphere, primarily because they carry with them the promise of generating a cross-section of cloud and aerosol particles from a large range of altitudes with high temporal andspatial resolution. Among the many applications for LIDARs in atmospheric science are:monitoring emission plumes, investigating long-range aerosol transport, and studying theoptical properties of different cloud types. Researchers have also used aerosols in LIDARdata as tracers to monitor mixing behaviour in elevated layers, in low-density clouds, andat the edge of the Planetary Boundary Layer (PBL).Historically, the use of this kind of instrument has been limited by two factors: theexpense of purchasing and operating such systems and the highly difficult task of prop-erly analysing and reporting the results. The former is due to the need for components1Chapter 1. Introductionsuch as high-power lasers and extremely sensitive optical receivers, combined with delicateoptical alignment requirements. The latter is the result of the difficulty in solving thenecessary light scattering equations, which become increasingly complex in the presenceof inhomogeneous distributions of unidentified scattering species. Unfortunately, efforts toresolve the latter issue typically involve more technically advanced systems including addi-tional wavelength bands, sometimes extending to collection of inelastically scattered light,which tends to exacerbate the former issue. Advances in optical and photonic technologyare making some LIDARs less expensive and easier to operate, but the results from thesetend to be limited in scope to single-wavelength elastic systems for which aerosol analysisbecomes difficult.Products like the Vertical Feature Mask (VFM) generated with data from the Cloud-Aerosol LIDAR and Infrared Pathfinder Satellite Observations (CALIPSO) satellite andthe aerosol type profiles generated by NASA’s airborne High Spectral Resolution LIDAR(HSRL) are of particular use to researchers who are not necessarily experts in light prop-agation and scattering because it allows them to monitor aerosol layers of interest andto perform long-term statistical analyses showing the distribution of aerosols and linkingthese trends to topics of interest such as climatological effects, atmospheric processes, andair quality. The generation of this type of qualitative data product, while not uncommonfor airborne and spaceborne systems has proven elusive for most ground-based LIDARsystems that do not include at least two emission wavelengths and one Raman channelthat can be used to directly measure extinction in the lower troposphere.Uses for this kind of data product extend to many areas of aerosol research. TheVFM has been used repeatedly to investigate climatological impacts of different aerosoltypes [46, 83, 86, 89]. LIDAR data are especially useful when the vertical extent of thelayers plays an important role in the study. LIDAR data have been used to investigate theindirect climate effects of dust acting as ice cloud nuclei [89, 91]. Such studies could beenhanced by the application of a layer analysis tool. Layer masks have also been used inlong-term tracking of aerosol types and their effects on weather patterns [60]. The uses ofthese tools extends even beyond pure atmospheric research. For example, studies trackinganthropogenic impacts on iron deposition to oceans from dust [53, 66] could benefit fromground-based LIDAR layer analysis near source regions or in coastal locations. Of coursethis does also have an indirect climactic effect because nutrients such as iron are a limitingfactor in photosynthetic production in the oceans.21.1. Thesis goals and objectives1.1 Thesis goals and objectivesThe goal of this thesis is to develop and evaluate an algorithm that can be used to findand analyse elevated aerosol and cloud layers and to demonstrate its utility in atmosphericaerosol research. More specifically, the algorithm presented focuses on LIDAR data col-lected from any single wavelength elastic LIDAR with dual polarizations. Due to theirsimplicity, systems such as these are now being sold in relatively inexpensive off-the-shelfstand-alone packages, making them among the most robust and accessible forms of LIDARinstruments. Therefore the algorithm is well suited to enhance the role of LIDAR in studiesof clouds and aerosols in remote locations and for research teams that do not have accessto more complex LIDAR systems. This type of analysis serves two purposes, first it makesthe images easier to interpret by providing masks that can be easily read and used as filtersfor layer analysis. This makes it especially useful as a tool for long-term analysis of aerosollayers. Secondly by generating a profile of LIDAR ratio values, it allows for an improvedcalculation of aerosol optical properties such as extinction and backscatter coefficients, andparticulate depolarization ratios. The results of the algorithm are used in both of theseways To this end this document contains the following chapters:Chapter 1 - Introduction A top-level overview of the thesis including a description ofthe Layer Discrimination Algorithm (LDA), an explanation of why it wasdeveloped and how it can provide value to aerosol research efforts, and anoverview of the thesis goals and objectives.Chapter 2 - Theory An exploration of the theoretical underpinnings behind LIDARlayer analysis. This begins with a brief introduction to the physics of lightpropagation and scattering in the atmosphere. Then the LIDAR equation isintroduced in its most general form and the more specialized equation usedfor single wavelength elastic scattering is derived. The chapter ends withan in-depth review of observed scattering properties for major aerosol andcloud types that are used as a basis for the development of the LDA.Chapter 3 - Data Collection Here, the various sources of data used herein are describedin detail. This includes the empirical data from LIDAR systems used in theLDA as well as various sources of empirical and modelled supporting dataused to verify the layer identifications and to provide context for aspects such31.1. Thesis goals and objectivesas the origins of the observed layers, how they were transported, estimatesof the time spent in transit, and any measurable impacts on local air quality.Chapter 4 - Algorithm Development An in-depth look at the fundamental innerworkings of the LDA. Descriptions include calculation of the primary LIDARdata products (normalized relative backscatter for the Mini-MicroPulse LI-DAR (mMPL), backscatter ratio for the Canadian Operational ResearchAerosol LIDAR Network (CORALNet)), methodology for finding layer edges,and the use of depolarization, signal and Signal to Noise Ratio (SNR) thresh-olds to classify these layers into broad categories. Following this is an ex-planation of the inversion methodology used to calculate backscatter andextinction coefficients from the LIDAR data. The results of the algorithmare tested in two ways: quantitatively against idealized profiles of simulatedLIDAR signals, and qualitatively using example scenes from the LIDARdata record. The results of both kinds of test are reported here.Chapter 5 - Research Applications This chapter demonstrates some of the capabilitiesof the algorithm to add value to aerosol research. This is accomplished bydescribing three studies of medium to long-range aerosol transport eventsin which ground-based LIDAR systems using the LDA, in combination withexternal data sources, were employed. Two of these studies involve resultsthat were published previously without the use of the LDA. In these casesthe nature of the elevated layers was well-established prior to the use of theLDA and the results of the LDA are shown to be in agreement. The LDA isthen used to enhance the analysis by isolating layers of interest for furtherinvestigation. The final study involves recently collected results of an aerosoltransport event that occurred in July and August 2015 that have yet to bepublished. In this case the LDA serves as the initial determination of thenature of the elevated layers as well as being used as an analysis tool.Chapter 6 - Conclusions The final chapter provides a summary of the LDA and itscontributions to aerosol research to date. It also includes a description ofsome known issues with the algorithm and plans for future development.4Chapter 2TheoryIn atmospheric LIDAR, one attempts to learn as much as possible about the vertical struc-ture of the atmosphere by measuring the optical properties of successive volumes of inte-gration within a (usually) vertical profile. There is an inherent difficulty in disentanglingthe measured values at a given range from the extinction caused by intervening scatteringspecies. It thus becomes important to employ certain simplifying assumptions combinedwith as much empirical knowledge as possible of the scattering behaviour of the backgroundatmospheric gases as well as the aerosols and cloud particles typically encountered in theatmosphere.2.1 Light propagation and scattering in the atmosphereThe interaction of light with atmospheric particles and gases results in attenuation of theinitial beam as well as scattered radiation in all directions at the original wavelength (elasticscattering) and other wavelengths (inelastic scattering, fluorescence). The full calculationof atmospheric extinction would include the combined effects of elastic scattering, inelasticscattering, and absorption. However, in this study, the focus is on single-wavelength elasticLIDAR and the effects of inelastic scattering are orders of magnitude smaller than the elas-tic scattering, so these are not included in the calculations of LIDAR transmission. Unlessotherwise noted, the assumption is made that the emitted light is at a single wavelength λand the scattered light is at the same wavelength in all the following equations.2.1.1 ExtinctionFor a collimated monochromatic beam (such as a laser) with initial radiant flux Φ0 atwavelength λ, the attenuation at range z is described by the Beer-Lambert-Brouger Lawas follows:52.1. Light propagation and scattering in the atmosphereΦ(z) = Φ0 exp−τ(z, λ), (2.1)where:τ(z, λ) =z∫0α(z′, λ)dz′.Although the standard form of the Beer-Lambert-Brouger Law is defined in terms ofradiant flux, the same relationship applies to the total energy of a laser pulse E, which isa more commonly used quantity in the field of LIDAR:E(z) = E0 exp−z∫0α(z′, λ)dz′. (2.2)In these equations, τ(z, λ) is referred to as the vertical optical depth and α(z, λ) isthe extinction coefficient. The value of α is determined by the number of scatterers in agiven volume of integration and their scattering and absorption efficiencies: Qscat(r,m, λ)and Qabs(r,m, λ), respectively. These combine to determine the total extinction efficiency(Qext = Qscat + Qabs). These are unitless parameters determined for a given wavelengthof incident light by a scattering particle’s radius r and the complex index of refraction m.The complex index of refraction is defined as: m = n+ ik where the real part n deals withscattering and is based on the speed of light in a medium relative to that in a vacuum andthe imaginary component k is related to the absorptivity of the medium.For N spherical scatterers of a given size within some volume of integration the extinc-tion coefficient is:α(λ, r,m) = pir2Qext(λ, r,m)N (2.3)Scattering particles are not all the same size, but rather include a distribution of sizesthat is often approximated by a multi-modal log-normal distribution [9]. In this case, the62.1. Light propagation and scattering in the atmosphereintegrated scattering with the volume would be expressed asα(λ,m) =∞∫r=0pir2Qext(λ, r,m)N(r)dr (2.4)Often the limits of integration for r are restricted to a reasonable range that reflectsthe majority of expected particle sizes. The minimum and maximum radii (rmin andrmax respectively) will vary depending on the application, the wavelength of light, and thetype of scattering being investigated. Of course, in a real atmosphere any given volume ofintegration will contain several different scattering species with their own optical properties.When summing over these species the equation for extinction becomesα(λ) =∑irmax∫rminpir2Qexti(λ, r,mi)N i(r)dr. (2.5)2.1.2 ScatteringThe scattering in a given volume of integration (β) can be expressed in terms analogous tothose used for extinction in Equations 2.3-2.5. The difference is that with scattering thereis an additional variable to indicate the scattering angle θ measured from the direction ofthe incident light ray. The equation for scattering by N particles of the same size and typewithin a volume of integration isβ(λ, θ, r,m) = pir2Qscat(λ, θ, r,m)N. (2.6)For one type of particle with a range of sizes this becomesβ(λ, θ,m) =∞∫r=0pir2Qscat(λ, θ, r,m)N(r)dr. (2.7)72.1. Light propagation and scattering in the atmosphereFor a polydisperse combintation of particle types the equation isβ(λ, θ) =∑irmax∫rminpir2Qscati (λ, θ, r,mi)N i(r)dr. (2.8)Note that due to the assumption of a particle that is spherical, the scattering is inde-pendent of the azimuthal angle φ. It is common in scattering theory to make use of thephase function p(θ). This unitless quantity is normalized such that the integral throughall 4pi steradians is unity and it describes the angular dependency of scattered radiance Las a function of the scattering angle θ.p(θ) =L2pi∫ pi0 Lsinθdθ(2.9)or, because L ∝ β(θ)p(θ) =β(θ)2pi∫ pi0 β(θ)sinθdθ. (2.10)The phase function can be interpreted as a probability distribution function describingthe likelihood that an incident photon will be scattered in a given direction.A common quantity of interest in LIDAR analysis is the backscatter coefficient βpi,which is the scattering coefficient for 180◦ scattering.βpi(λ) = p(θ = pi)β(λ). (2.11)The light scattering behaviour for a particle can fall into one of three domains: Rayleighscattering, Mie scattering, and geometric scattering. The scattering behaviour is deter-mined by the size of the particle relative to the wavelength. The size parameter for scat-terers is defined as x = 2pirλ where r is the particle diameter.Rayleigh ScatteringWhen x << 1, (less than 0.1 is a standard threshold) the particle can be approximated bya dimensionless point dipole and a simplified solution to the scattering equations can beused that is independent of particle shape. This is referred to as Rayleigh scattering. At82.1. Light propagation and scattering in the atmospherevisible wavelengths, this applies to atmospheric gases, vapours and very fine mode particlesprimarily in the nucleation phase.Assuming a dry atmosphere, the complex index of refraction m in the molecular at-mosphere is a function of air density. In the absence of appreciable amounts of ozone,atmospheric absorption is negligible for visible wavelengths. Thus for the purposes of thisinvestigation, which deals exclusively with scattering of visible light in the troposphere, onecan focus on the real part of the index of refraction n. At standard temperature and pres-sure (Ts=273.15 K,Ps=101.325 kPa) the real part of the index of refraction nsis as follows[28, 29]108(ns − 1) = 8342.3 + 2406030130− ξ2 +1599738.9− ξ2 , (2.12)where ξ = 1/λ. Based on this quantity the index of refraction can be calculated forother temperatures and pressures as follows [82](n− 1) = (ns − 1)(1 + 0.00367Ts1 + 0.00367T)PPs. (2.13)Ignoring for the moment the effects of depolarization (more on this in Section 2.1.2)the Rayleigh scattering coefficient is as follows:βm(λ, θ) =pi2(n2 − 1)2D2Ds2λ4(1 + cos2θ) (2.14)where Ds is the number density for a standard atmosphere (Ds = 2.547× 1019 cm−3).The total integrated scattering for all angles is thenβm(λ) =8pi3(n2 − 1)2D3Ds2λ4. (2.15)Based on this equation it is clear that the scattering efficiency of a particle in theRayleigh scattering regime is strongly dependent on wavelength in that it is proportionalto λ−4. It is also the case that the amount of scattered light is symmetrical so that equalamounts of light are scattered forward and backward.Geometric ScatteringWhen x >> 1 the particle is large enough for the scattering to be described by the re-flection, refraction, and diffraction equations of geometric optics. In this case the shape92.1. Light propagation and scattering in the atmosphereand orientation of the particle have a large impact on the resultant scattering, but theresults are largely independent of wavelength. The most common examples of this in theatmosphere is large water droplets in clouds. Geometric scattering also comes into playin the case of large, horizontally oriented ice platelets in cirrus clouds. These cases aredescribed in more detail in Section 2.3.1.Mie ScatteringWhen x ≈ 1 Mie scattering theory comes into play. This is an umbrella term used for aseries of solutions for scattering of an electromagnetic plane wave by uniform or stratifiedspheres, infinite cylinders, and elongated spheroids. Dust, pollen, smoke, marine sulphates,and urban aerosols are some examples of aerosols that fall into this category. In Miescattering the scattering coefficient and phase functions are expressed as an infinite seriesof spherical multipole partial waves. The result is an approximation of the scattering,extinction, and absorption efficiencies as well as the phase function for a given particletype. Although real particles may not adhere to these simple shapes (spheres and cylinders)the approximation is good enough for most applications. Computational Mie scatteringcodes have been developed to calculate the scattering properties of particles that adhereto these simple shapes, and ones that can be approximated by combinations of them.For cases where more generalization is required, a computational solution called T-matrixtheory can be employed [115]. Black carbon, for example is known to have a fractal shapeand for applications where precise understanding of scattering coefficients is key, T-matrixsolutions are often employed.DepolarizationLasers emit highly linearly polarized light, so it is easy to identify how this initial polar-ization state is altered by the scattering species by measuring the scattered light in theoriginal polarization and the orthogonal polarization. The depolarization ratio is typicallydefined asδ =L⊥L‖, (2.16)where L‖ is the measured radiance with polarization parallel to the emitted laser pulseand L⊥ is the radiance that is polarized in the orthogonal direction. In scattering, these102.2. The standard LIDAR equationradiance values are of course proportional to the backscatter coefficients, so the expressionδ = β⊥β‖ is also used. A value of δ = 0 indicates no change in polarization at all, whereasa value of δ = 1 indicates completely randomly polarized light. Spherical particles suchas water droplets and aerosols formed by condensation of organic and inorganic vapoursscatter light with little or no alteration to the polarization state, resulting in very lowdepolarization ratios. This is also the case when specular reflection dominates - as inhorizontally oriented ice plates in clouds [92] Conversely, elongated irregular particles suchas dust and ice crystals cause changes in polarization. The integrated scattering of anumber of randomly oriented particles such as these within a given volume of integrationresults in a large degree of depolarization.Depolarization in Rayleigh scattering Due to anisotropy in molecular scattering,some degree of depolarization also occurs in the molecular atmosphere. The amount ofdepolarization is dependant on atmospheric temperature and pressure as well as a depo-larization factor η. Thus the complete formulation of Equation 2.15 is actually [9, 117]βm‖(λ) =8pi3(n2 − 1)2D3Ds2λ4(6 + 3η6− 7η)(PPs)(TsT). (2.17)The value of η has been calculated in a variety of ways over the years [82, 117] and ispartially dependant on the spectral bandpass of the detection system. This is because evenin what is ideally a purely elastic LIDAR system, the reality is that there is some spectralbroadening of the scattered light due to rotational Raman scattering. If the bandpass of thereceiver is wider than ≈ 1 A˚, some portion of the energy from this inelastic scattering willpass through and the depolarization factor will be affected. For purely elastic scattering,the currently accepted value is η = 0.00714, but if the effects of Raman scattering areincluded the value is η = 0.0279 [117].2.2 The standard LIDAR equationWhen one combines Equations 2.2 and 2.11 with the inverse square law, it is possibleto formulate the amount of energy from a laser pulse that is transmitted to a scatteringvolume of thickness ∆z at a range z, and is elastically scattered back to a receiver with112.2. The standard LIDAR equationentrance pupil area A:E(z, λ) = E0(λ)Az2βpi(z, λ)∆z exp[− 2z∫0α(z′, λ)dz′](2.18)where: E(z, λ) = energy entering the entrance pupil [µJ ]E0(λ) = transmitted laser energy [µJ ]z = range [km]∆z = thickness of the volume of integration [km]λ = wavelength [µm]A = usable area of receiver entrance pupil [km2]1βpi(z, λ) = volume backscatter coefficient [km−1sr−1]α(z, λ) = volume extinction coefficient [km−1)].Equation 2.18 parametrizes the measured signal of a LIDAR receiver as a functionof range and atmospheric scattering properties, and as such it sits at the heart of allatmospheric LIDAR data analysis. When it is combined with several scaling quantitiesrepresenting the properties of the receiving optics and detection sensor, the result is thestandard LIDAR equation.S(z, λ) = E0(λ)Az2γO(z)βpi(z, λ)∆z exp[− 2z∫0α(z′, λ)dz′]+ SB (2.19)Where: S(z, λ) = signal measured at the detector [counts/µs]SB = background signal [counts/µs]γ = overall system efficiency constant [counts/µs∗µJ]O(z) = incomplete overlap correction function [unitless].There are several implicit assumptions built into this particular formulation of theLIDAR equation. First, the aforementioned assumption of elastic scattering implies thatthe scattered wavelength is identical to the emitted wavelength. Second, it is assumedthat the scattering angle is 180◦ backscatter. In addition, it is assumed that the scatteringproperties within the volume of integration are homogeneous, or at least that there are1Many find it more convenient to express this value in units of m2 or cm2 with an appropriate conversionmultiplier122.2. The standard LIDAR equationno sharp changes within the range ∆r, and that the range r must be sufficiently largethat r >> ∆r so that the difference between r2 and (r + ∆r)2 is negligible. Finally, thisequation is only representative of single-scattering events.Equation 2.19 comprises two terms, the first representing the LIDAR signal and thesecond representing the radiative background. The second term is self-explanatory, but oneway of understanding the first is by breaking it into four quantities that together determinethe probability that one of the photons that are emitted from the transmitter will find itsway to the receiver to be counted.Term #1: Az2This is based on the inverse square law and represents the likelihood that a scatteredphoton at range z will enter the receiver aperture. This quantity can be thought ofas representing the solid angle occupied by the scattered light that enters the receiveraperture.Term #2: γO(z)This represents the combined effects of system optical losses, including detector re-sponsivity and the zone of incomplete overlap between the transmitter and the re-ceiver. It can be thought of as the likelihood that a photon emitted by the laser willpass through the system to the detector.Term #3: βpi(z, λ)∆zThis is the range integrated backscatter coefficient of all particles and gases in thescattering volume and represents the probability that an incident photon will be scat-tered in the direction of the receiver (θ = pi).Term #4: exp[− 2z∫0α(z′)dz′]This is the two-way transmissivity of the atmosphere at the emission wavelengthλ and represents the likelihood that a photon emitted by the transmitter will passthrough the intervening atmosphere to reach the scattering volume and be returnedto the receiver132.2. The standard LIDAR equationIn this equation, the backscatter and extinction terms (βpi and α respectively) representthe integrated contributions of all gases and scattering particles within the volume ofintegration defined by the detector field of view at the altitude in question and the laserpulse width. For simplicity here on out, the term βpi will be written simply as βT since the180◦ backscatter is the only quantity of interest. Since the contribution to scattering andextinction from the molecular atmosphere can be readily calculated based solely on profilesof temperature and pressure (or, failing that, an assumed atmospheric profile), Equation2.19 can be rewritten to separate them from the scattering of atmospheric aerosols andcloud particles.S(z) = E0Az2ηO(z)(βp(z, t) + βm(z, t))∆z exp[− 2z∫0(αp(z′) + αm(z′))dz′]+ SB (2.20)where:βp(z) = contribution to backscatter from particles [km−1sr−1]βm(z) = contribution to backscatter from the molecular atmosphere [km−1sr−1]αp(z) = contribution to extinction from particles [km−1]αm(z) = contribution to extinction from molecular atmosphere [km−1].For the sake of simplicity several of the factors in Equation 2.20 that are held constantfor a given LIDAR profile can be combined into one scaling term and the contributions ofparticulate and molecular extinction can be expressed in terms of atmospheric transmis-sivity:S0(z) =1z2CE0O(z)(z)(βp(z) + βm(z))TpTm (2.21)where:S0(z) = S(z)− SBC(z) = A∆zηTp = exp[− 2z∫0αp(z′)dz′]Tm = exp[− 2z∫0αm(z′)dz′].142.2. The standard LIDAR equation2.2.1 The LIDAR ratioA cursory examination of the above equation reveals one of the most important challenges inusing LIDAR data to investigate aerosols in the atmosphere. The particulate backscatterand extinction coefficients are both unknown. This is typically addressed by assumingsome sort of relationship between the two quantities. It is common to assume a power lawrelationship as in Equation 2.22, where the constant R is referred to as the extinction tobackscatter ratio or simply the “LIDAR ratio”. However in this formulation the uncertaintyin the value of the exponent κ is so high that in practice most researchers use a defaultvalue of κ = 1:α(λ) = R(λ)βTκ(λ) (2.22)The LIDAR ratio for Rayleigh scattering (assuming negligible amounts of absorption)can be calculated directly from Equations 2.14 and 2.15Rm =8pi3(n2−1)2D3Ds2λ4pi2(n2−1)2D2Ds2λ4(1 + cos2pi)=8pi3(2.23)For aerosol and cloud particles in the Mie scattering regime there is no theoreticalbasis for a linear relationship between extinction and backscatter except in cases whereboth the size distribution and particulate material properties are constant, which is rarelythe case in a real atmosphere. This is because the phase function, and therefore thebackscatter coefficient, varies strongly with size parameter and complex index of refraction,as well as particle shape and orientation in the case of non-spherical particles. However,in LIDAR applications where polydispersion of many different particle types is the norm,and the values for extinction and backscatter coefficients represent integrated volumes,this variation is greatly reduced. In other words, some degree of smoothing occurs as thescattering of many particles within a volume of integration are combined together [26, 84].Although there is error in this assumption, popular inversion methods make use of thisformulation to obtain stable analytical inversions of the LIDAR equation to determineprofiles of extinction and backscatter coefficients from LIDAR data. [e.g.: 31, 57, 58]. Whatis known about the optical properties for these particles, including the LIDAR ratios, is152.3. Optical properties of clouds and aerosolsdetermined almost exclusively from empirical studies.2.3 Optical properties of clouds and aerosolsAccurately calculating the combined results of scattering from multiple particles in the Miescattering regime within a volume of integration is a challenging task even given simplifyingassumptions such as simple shapes and homogeneous and well-defined optical properties.Under real conditions in the atmosphere, aerosol and cloud layers are typically composedof inhomogeneous mixtures of different particle types and size distributions and the opticalproperties such as index of refraction are rarely known in advance. In situ measurementsof scattering provide the best information about the light-particle interactions for differentaerosol and cloud types. What follows is an extensive, but by no means comprehensive,review of published results from empirical studies of the optical properties of aerosols andclouds that pertain to this work. Since the two quantities of interest for analysis withdepolarization LIDAR are the depolarization ratio and the LIDAR ratio, these are thequantities upon which this review will be focused.2.3.1 Scattering in cloudsCloud particles comprise two major types with very distinct scattering properties: waterdroplets and ice crystals. They both tend toward the larger end of the typical size distri-bution ranges for atmospheric particles (1 µm to 20 µm for water droplets, 20 µm to 100 µmfor larger drops, and 50 µm to 1 mm for ice crystals) and exhibit high total scattering crosssections for a given volume of integration as compared to aerosols, which leads to relativelylarge values for integrated backscatter and extinction coefficients, with the exception ofsome types of cirrus clouds.In the case where single scattering events dominate the LIDAR signal, water dropletsand ice crystals are easily differentiable by their depolarization ratios. Water droplets arespherical and homogeneous in composition - perhaps as close an approximation to theidealized spheres of Mie scattering theory as exist in nature - and the scattering from suchparticles is known to preserve polarization state almost perfectly2. On the other hand,ice crystals are irregular in shape, randomized in orientation, and scatter light through2As droplets grow larger than ≈ 20µm they begin to fall and are deformed into more complex “teardrop”shapes resulting in somewhat higher depolarization162.3. Optical properties of clouds and aerosolsa series of internal reflections that tend to alter the polarization of the scattered light.Even in the absence of knowledge of the specific shape and orientation of the crystals,it is easy to see that a large degree of depolarization can be expected [88]. This wouldseem to make differentiating between these particle types with a polarization sensitiveLIDAR a straightforward task, but in the presence of multiple scattering, these seeminglywell-defined properties can become less distinct [13, 17, 44, 80, 88].Water dropletsAs mentioned above, water droplets are spherical (or nearly so) and homogeneous. Ac-cording to Mie scattering theory, scattering from one of these spheres will preserve thepolarization of the incident light ray perfectly. In practice, empirical studies have shownthat the depolarization ratio in clouds or haze consisting of only water droplets will typi-cally not exceed a threshold of 0.05. [88]. Empirical studies have shown that the extinctionto backscatter ratio for water clouds is ≈ 15.7 regardless of droplet size for size parametersup to 500 [84]. These properties can be considered to be definitive for single scatteringevents, but actual LIDAR measurements in clouds have been shown to be significantly im-pacted by multiple scattering. In some studies involving LIDARs [e.g.: 44, 88], a roughlylinear relationship of increasing depolarization with cloud depth was demonstrated frommultiple empirical observations. The specific relationship is dependant on many factorsincluding the laser beam divergence, the receiver FOV, cloud base height and particle sizedistributions. This phenomenon is not part of the cloud phase discrimination portion ofthe LDA due to the very small FOV of the mMPL instrument (more on this in Section4.1.3)Ice particlesScattering from water droplets in LIDAR applications was once thought to be straight-forward, but has since been shown to include some unforeseen complications. In contrast,scattering by ice crystals has always been known to be a complex and poorly understoodphenomenon. Ice crystals come in a wide array of sizes and shapes and their orientationscan be either highly randomized or highly uniform depending on the conditions. In additionto this, the surface composition of the crystals can vary due to micro-physical processessuch as partial freezing, melting, and riming. As a result of all of these conditions, a widerange of depolarization results has been observed from ice clouds including values of < 0.03172.3. Optical properties of clouds and aerosolsfor horizontally oriented platelets to values > 1.0 (a result not previously thought to bepossible) for other orientations. Despite these anomalies, when taken as a whole the depo-larization signatures of icy clouds fall primarily within a range of 0.20 to 0.80 with medianvalues between 0.40 to 0.50. [88].As with water clouds, multiple scattering effects lead to large variations in depolar-ization with cloud depth. In the case of ice crystals the relationship is in the oppositedirection to that of water droplets (decreased depolarization with increased backscatter)and is also less linear [44, 45]. In addition to this, it has also been observed that sometypes of ice crystals, specifically horizontally oriented planar crystals that tend to form inthe −10 ◦C to −20 ◦C temperature range, exhibit very low depolarizations in backscatterdue to specular reflections dominating the scattering phase function [88, 92]. As for themultiple scattering effects in water clouds, these are not part of the existing algorithm.Variations in the size, shape, and orientation of ice crystals can lead to large variationsin LIDAR ratio for cirrus clouds. Some researchers find effective LIDAR ratios in opticallythin cirrus clouds for altitudes up to 15 km have ranged from 2 sr to 20 sr with median valuesnear 10 sr) [4, e.g.] while others have observed ranges from 10 sr to 40 sr with median valuesof 29 sr [15, 118, e.g.]. Anomalous cases have been observed where sub-visual cirrus cloudshave exhibited LIDAR ratios of over 200 sr [90]. This is clearly an issue for calculatingextinction and backscatter for any LIDAR profile that includes one or more cirrus clouds.The LDA currently assumes an effective LIDAR ratio for cirrus clouds of 25 sr with a lidarratio of 20 sr for mixed-phase clouds. Future iterations of the algorithm may incorporateprofiles of temperature and relative humidity, where available, into the LDA to improveperformance with respect to inversions including clouds (this is discussed further in Scattering in aerosolsAtmospheric aerosols exhibit a wide array of scattering properties, and not only becauseof the range of materials, and particle shapes and sizes. Similar aerosol types exhibitsubstantial variations depending on the location of origin [27]. Humidity can have a largeeffect on smoke, urban, and marine aerosols as the result of an increase in real and imaginaryparts of index of refraction as well as a shift toward larger particle sizes [1]. The time spentin transport from source regions leads to increased variations and changes in mean valuefor depolarization as well as LIDAR ratios [2, 63, 73].182.3. Optical properties of clouds and aerosolsMineral dustMineral dust is the most prominent atmospheric aerosol by volume, comprising roughly 75% of all particulate matter. It is estimated that the global dust cycle lofts some 1 Tg to 2 Tgof particles into the atmosphere annually [119]. Atmospheric dust consists of irregularlyshaped particles of soil, rock and clay that are advected into the atmosphere by surfacewinds. Most dust particles that travel more than a short distance have a mean diameterin the range of 1 µm to 10 µm. Once lifted into the free troposphere, dust particles cantravel thousands of miles, even circling the globe [63, 113]. The primary sources of dust inthe atmosphere are in the so-called “Dust Belt”, which ranges from North Africa throughthe Middle East to Central Asia. [85]. Dust particles are the primary source of iron to theworld’s oceans, but can also carry harmful bacteria over long distances. [53, 66]Dust particles exhibit very high degrees of depolarization as a result of their irregularshape (see Table 2.1) however, unlike ice crystals there is no observable variation in depo-larization ratio a a result of multiple scattering [47]. Empirical measurements of dust-richaerosol layers reveal depolarization ratios ranging from 0.20 to 0.40. LIDAR ratios rangefrom ≈30 to 70 depending on the location of the source region, the age of the aerosol layer,and the degree of mixing with other aerosols. Although dust particles are chemically verystable as compared to other kinds of aerosols, there are some drops in depolarization forolder layers, usually as a result of mixing with other aerosol types, but also due to NO−13attachment to the surface of dust particles [63, 71]. Due to their relatively large particlesize, the scattering properties of dust particles exhibit a low degree of wavelength depen-dence [38]. There are many types of dust, but unless otherwise noted the term “dust” inthis document will be assumed to imply “mineral dust” specifically.192.3.OpticalpropertiesofcloudsandaerosolsLocationDepolarization ratio LIDAR ratio [sr]SourceMean Range Mean RangeNorth America - 0.30-0.33 - 45-51 [12]Cape Verde 0.30 0.29-0.31 62 54-70 [38]Northwest Africa 0.31 0.30-0.32 41 35-47 [64]Northwest Africa 0.30 0.23-0.37 38.5 29.3-47.7 [65]Tarim Basin 0.34 0.27-0.41 - -[63]Gobi Desert 0.28 0.22-0.34 - -Northwest Africa 0.30 0.22-0.38 - -Northwest Africa - - 56.4 -[99]Non-Sahel Africa - - 57.7 54.5-60.8African Sahel - - 55.2 52.7-56.4Middle East - - 46.9 39.5-50.2Sahara (PBL) - 0.30-0.35 55 50-60[73]Sahara (FT) - 0.10-0.25 59 48-70Gobi (PBL) - - 35 30-40Saudi Arabia (FT) - - 38 33-43Tokyo > 0.20 - 49 - [76]Global - - 42 38-46 [14]Table 2.1: Values for dust depolarization and LIDAR ratios. All values are for λ = 532nm unless otherwise specified.202.3. Optical properties of clouds and aerosolsSmokeSmoke from biomass burning is an important factor in climatology due to direct forcing[10, 43], indirect forcing [54], and changes to the albedo of snow and ice [39]. Smoke fromwildfires and, in more rural regions, from cookstoves also have a marked effect on air qualityand human health, even after long-range transport [50, 52]. And of course, these impactsare worsening as the climate continues to change [10].Aerosol emissions from biomass burning consist largely of elemental carbon (soot) con-glomerates which have a large imaginary component to the index of refraction, makingthem highly absorptive. The result is a large LIDAR ratio and, despite the irregularity ofthe particles, a very low depolarization ratio, typically < 0.05. However, in cases where thefire burns particularly hot, the depolarization ratio can be increased by ash and even dustlifted into the plume by the strong convection currents resulting in depolarization ratioson the order of 0.10− 0.15 [73].212.3.OpticalpropertiesofcloudsandaerosolsLocationDepolarization ratio LIDAR Ratio [sr]SourceMean Range Mean RangeNorth America - 0.04-0.09 - 55-73 [12]Siberia/Canada - < 0.05 53 42-64 [73]Cape Verde 0.16 0.15-0.17 87 79-95 [38]Tokyo - 0.05-0.08 ≈ 65 - [76]Germany - 0.03-0.05 51 - [67]Global - - 60 52-68 [14]Table 2.2: Values for smoke depolarization and LIDAR ratios. All values are for λ = 532nm unless otherwise specified.222.3. Optical properties of clouds and aerosolsUrban aerosolsUrban aerosols include both liquids and solid and are generated by combustion and otherindustrial processes. They comprise mainly sulphates, nitrates, hydrocarbons, and soot.Urban aerosols can be emitted in particulate form directly into the atmosphere (primaryaerosols) or formed in the atmosphere through chemical processes (secondary aerosols).Their size distribution can range from 1 µm to 100 µm. Some variation in Table 2.3 is dueto mixing with dust and other aerosols [73].232.3.OpticalpropertiesofcloudsandaerosolsLocationDepolarization ratio LIDAR Ratio [sr]SourceMean Range Mean RangeNorth America - 0.03-0.07 - 53-70 [12]Central Europe(PBL) - < 0.05 53 42-64[73]South-west Europe (FT) - < 0.05 45 36-64North America (FT) - < 0.05 39 29-49Chicago - - 47 35-59 [2]Global - - 71 61-81 [14]Table 2.3: Values for urban depolarization and LIDAR ratios. All values are for λ = 532nm unless otherwise specified.242.3. Optical properties of clouds and aerosolsMarine aerosolsMarine aerosols consist primarily of sea salt, sulphates, and some other gases that arereleased into the atmosphere from the oceans as a result of wind and waves. They existalmost exclusively within the PBL over the oceans. Although salts dominate the compo-sition of emissions from ocean water by mass, they rarely rise more than a few feet abovethe surface and it is almost exclusively the sulphates that rise higher and form droplets.Depending on the winds, they can be advected up to 10 km from the coast, but are rarelyfound any further inland. Marine aerosols are an important factor in global radiative forc-ing causing a global average Aerosol Optical Depth (AOD) of 0.07 to 0.15 and contributingCloud Condensation Nuclei (CCN) of 60 cm−3 [55, 100]. Marine aerosols are largely spher-ical in nature exhibit depolarization ratios within a similar range as water clouds (0.00to 0.05). Raman LIDAR measurements of LIDAR ratios for marine aerosols have shownaverage values from 17 sr to 27 sr. It’s worth noting that this value has been shown to varyinversely with wind speed by as much as ±5 sr [24].252.3.OpticalpropertiesofcloudsandaerosolsLocationDepolarization ratio LIDAR Ratio [sr]SourceMean Range Mean RangeCape Verde 0.02 0.00-0.04 19 17-21 [38]Global - - 26 17.5-33.5 [24]North America - 0.04-0.09 - 17-27 [12]North Atlantic (PBL) - - 23 20-26[73]Indian Ocean (PBL) - - 23 18-28Indian Ocean (FT) - - 29 21-37Global - - 28 23-33 [14]Table 2.4: Values for marine depolarization and LIDAR ratios. All values are for λ = 532nm unless otherwisespecified.262.3. Optical properties of clouds and aerosolsType Sub-TypeDepolarizationRatioLIDARRatio [sr]CloudWater Cloud 0 - 0.10 15.3Mixed Cloud 0.10 - 0.35 20Ice Cloud > 0.35 25AerosolSmoke / Urban 0 - 0.10 65Polluted Dust 0.10 - 0.20 55Dust 0.20 - 0.55 40Unidentified Aerosol N/A 30Table 2.5: Table of layer identification categories used in this analysis2.3.3 Aerosol summaryIn the absence of multi-spectral data, it is not possible to determine properties such assize distribution, complex index of refraction or A˚ngstrom coefficient3 the LDA must relyon the depolarization ratios, backscatter strength and location of the layers to distinguishbetween layer types and assign LIDAR ratios. As a result, the cloud and aerosol categoriesdefined above must be combined into some major groups for which at least one of theseproperties is sufficiently distinct. By combining the results from the above sections, threeaerosol and three cloud categories have been identified for use in the LDA (see Table 2.5)It is important to note that these categories are not definitive as quantities such asdepolarization and backscatter strength are only a proxy for particle shape, not a directmeasurement of aerosol type, and from Tables 2.1-2.4 it is clear that there is a non-trivialamount of overlap between the aerosol types. For example, the categories of smoke frombiomass burning and urban aerosols have been combined into one category, partially be-cause there is no useful distinction between their depolarization ratio ranges. Fortunately,there is also a great deal of overlap between the ranges of LIDAR ratio for the two cate-gories. These categories form the core of the layer identification portion of the LDA, whichis explained in detail in Chapter 4. As is evidenced by the tables in this section, there isa substantial range of variation in the observed LIDAR ratios of the various aerosol andcloud types. Nevertheless, the designation of even roughly calculated layer LIDAR ratiosis sufficient to improve the accuracy of calculations such as extinction and backscatter3The A˚ngstrom coefficient is a term used to describe the exponent in the power-law relationship betweenAOD and wavelength. It is inversely related to particle size and can be used as an indication of aerosol sizedistribution.272.3. Optical properties of clouds and aerosolscoefficients, AOD and particulate depolarization ratios. Note that there is currently nospecific category for marine aerosols due to the fact that they are typically only observedin specific circumstances (within 1 km to 2 km of the coast) and their depolarization ra-tio profile is not sufficiently distinct from the lower end of the range for urban aerosols.In future iterations of the algorithm, a “coastal” setting may be implemented for use inlocations such as these (more on this in Section 6.2.3).28Chapter 3Data collection methodologyThe primary data used for the various investigations contained herein include two differentLIDAR systems as well as a variety of supporting data sources that were employed toprovide context and support to the analysis of the LIDAR results.3.1 LIDAR systemsFigure 3.1 shows six locations around the world where LIDAR data were collected as apart of several research efforts led by the candidate from 2010-2015. The details of thetwo systems (designated CORALNet and mMPL) are provided in Sections 3.1.1 and 3.1.2respectively. The locations marked in the figure include four locations (Aksu, Ucluelet,Vancouver and Whistler) where the candidate was the sole system operator, personallycollecting data using the portable mMPL system , two semi-permanent CORALNet LIDARinstallations (Vancouver and Egbert) which were operated by Environment Canada andraw data were collected and processed by the candidate, and one location in Lanham wheredata were provided by the manufacturer of the mMPL system.3.1.1 CORALNetThe Canadian Operational Research Aerosol LIDAR Network is a semi-autonomous net-work of ground-based LIDARs. These remotely controlled facilities are housed in cargotrailers with modifications including a roof hatch assembly, basic meteorological tower,radar interlock system, climate control system and levelling stabilizers. The units can beoperated via an internet link and require an external power source. A precipitation sen-sor is used to operate the roof hatch and three pan/tilt web-cams capture sky conditionsand monitor the LIDAR system’s health. A remote control interface is used to control allvital components of the system, including the ability to provide hard resets of the laserelectronics.293.1. LIDAR systemsFigure 3.1: Map of locations where LIDAR data were collected for the various studiesincluded in this thesisThe CORALNet LIDAR system has been used for several years in a variety of aerosolstudies [e.g. 68, 70, 103]. The analysis of these data sets has been accomplished in previ-ous studies by using the Fernald method of selecting a lidar ratio and holding it constantthroughout the scene [31], with the addition of a range-dependent correction factor deter-mined by a member of the CORALNet team specifically for each event to correct for timeand situation dependant variations in detector responsivity. By using this methodology, theCORALNet team determines what they refer to as the backscatter ratio, which is a unitlessquantity described as the ratio between the measured signal and that expected from anaerosol-free standard atmosphere. It is in some ways similar to the “attenuated backscat-ter” product provided by the CALIPSO satellite. The CORALNet system is describedfurther in Appendix A.2 and full details of the CORALNet system, including calibrationprocedures, determination of overlap function, and data product descriptions can be foundin [105].3.1.2 Mini-Micropulse LIDARThe mMPL is an elastic, polarization sensitive LIDAR operating at 532 nm. It is a small,portable and self-contained system suitable for flexible deployment in adverse conditionsand on limited schedules. As a micropulse LIDAR system, the laser emits thousands of low-303.2. Supporting data sourcesenergy pulses that are integrated into a single profile. Earlier versions of the system had arepetition rate of 2.5 kHz whereas the current system uses 4 kHz. The time of integrationcan technically range from 5 s up to 60 min, but the most common usage involves a rangeof 15 s to 60 s. The vertical resolution can be set by the user to values ranging from 5 mto 75 m. Because of the pulsed nature of the laser and the fact that it utilizes a sharedinput-output “optical transceiver” style layout resulting in a beam that is expanded to adiameter of 7.62 cm, the system is rated eye safe at a range of 3.5 m, simplifying regulatoryhurdles and making the system easier to deploy in urban areas or near flight paths.The primary data product of the mMPL system is the normalized relative backscatter.This is a range-corrected data product that also includes normalization by laser outputenergy, and corrections for incomplete overlap and known detector artefacts (afterpulseand dead time corrections). Like the backscatter ratio calculated by CORALNet, the nor-malized relative backscatter is analogous to the attenuated backscatter of CALIPSO. Byusing an actively controlled liquid crystal retarder on the shared light path, the mMPLalternately transmits linearly and circularly polarized light, allowing it to generate a lineardepolarization ratio using a single Single Photon Counting Module (SPCM) detector. Thisprovides the advantage that the ratio is not affected by the detector field of view, overlapcorrections, or any other range-dependent corrections. This approach does have the dis-advantage that the two quantities are not measured at exactly the same time, so if theintegration time is long compared to the time scale of atmospheric change, the accuracycan be negatively affected. An in-depth description of the mMPL system, including theoptical layout, the calculation of the normalized relative backscatter and depolarizationratio, and other operational details are provided in Appendix A.1. Further information isavailable in [34]3.2 Supporting data sources3.2.1 HYSPLITBack-trajectories were calculated using the HYbrid Single-Particle Lagrangian IntegratedTrajectory (HYSPLIT) model Version 4. HYSPLIT 4 is the current version of a completesystem for computing simple air parcel trajectories to complex dispersion and depositionsimulations for any location and date (depending on data availability) using a variety ofstandard data input products (e.g. the National Centre for Environmental Prediction313.2. Supporting data sources(NCEP) Reanalysis 1948-present). All meteorological data used for this study were takenfrom Global Data Assimilation System (GDAS) 1◦ 3P weekly files generated by the NationalClimactic Data Centre (NCDC). In this study, the layers of interest were observed over arange of altitudes and individual events continued for several days. In order to give a fullpicture of the paths taken by the air parcels in question, an array of back trajectories werecalculated for each event. For each day during which dust was detected, back trajectorieswere originated in 6 hour intervals throughout the altitude regions where dust layers wereobserved, in 200 m increments. The time spans for which parcels were traced back wereselected on a case-by-case basis, but for each trajectory calculated the time step was 1 hour.Due to the large number of trajectories calculated, the results were plotted in “frequencymode”, which is essentially a 2-D histogram in 1◦ increments, rather than individually.3.2.2 NAAPSThe Navy Aerosol Analysis and Prediction System (NAAPS) is an Eulerian system forpredicting the distribution of tropospheric aerosols based upon the work of Christensen[18]. It uses global meteorological fields from the Navy Operational Global AtmosphericPrediction System (NOGAPS) analyses and forecasts on a 1◦ by 1◦ grid, at 6 hour intervals,for 24 vertical levels reaching 100 mbar. NAAPS provides global 120 hour forecasts ofsmoke, sulfate, and dust distributions in near-real time.Dust Emissions Dust emission in the model occurs whenever the friction velocity fromwind exceeds a threshold value (currently set at 0.6 m s−1, snow depth is less than a criticalvalue (current value is 0.4 cm), and the surface moisture is less than a critical value (criticalvalue set to 0.3). When these conditions are met, the particle flux is calculated andinjected into the bottom two layers of the model. In NAAPS, the particle flux is scaled toinclude only particles with radii less than 5 µm. The dust prediction model is based upon acombination of observed and predicted weather patterns with a global map of known dustemission areas. Dust emission areas are derived from eight of the 94 land-use types used inthe United States Geological Survey (USGS) Land Cover Characteristics Database, whichwas developed from Advanced Very High Resolution Radiometer (AVHRR) data and has1 km resolution. Some subjective modifications were made to the land use data basedon observable evidence when dust source regions were selected (e.g., regions designated“low sparse grassland” were defined as source regions only in China and Mongolia despite323.2. Supporting data sourcesexisting in other regions such as New Zealand and North America). The flux from a given1◦ by 1◦ grid cell is scaled based upon the fraction of land use that fits within one of theeight designated land use categories. For all areas not designated as dust source regions,the friction velocity threshold is set to infinity.Smoke Emissions The NAAPS smoke emission algorithm is based upon a global smokesource function incorporating results from a joint project of the US Navy, NASA, NOAAand university collaborators designated Fire Locating and Modelling of Burning Emissions(FLAMBE), and the Geostationary Operational Environmental Satellite (GOES) WildFireAutomated Biomass Burning Algorithm (WF ABBA), which is a collaboration betweenthe National Oceanic and Atmospheric Administration (NOAA) National EnvironmentalSatellite Data and Information Service (NESDIS) Office of Research and Applications(ORA) and the University of Wisconsin - Madison Cooperative Institute for MeteorologicalSatellite Studies (CIMSS).33Chapter 4Layer discrimination algorithmdevelopmentThis chapter describes the inner workings of the Layer Discrimination Algorithm (LDA),how it was developed and tested, and how it is applied. It begins with a step by stepbreakdown of how LIDAR profiles are processed into layer masks and then how theseare used as inputs to an inversion that generates backscatter and extinction coefficientprofiles. The next section contains a quantitative evaluation of results when applied toidealized LIDAR profiles for which the backscatter and extinction values are predetermined.Finally, the LDA is applied to a number of actual scenes of LIDAR data to demonstratehow it performs on empirical results. In these cases the true backscatter and extinctioncoefficients are not known ahead of time so the evaluation is more qualitative in nature,demonstrating how the LDA performs under various conditions.4.1 Layer analysis techniquesWhat is described here is an algorithm that takes as input a profile generated as rawoutput of a LIDAR system and generates as output a profile of layer type designationswith associated LIDAR ratios4The layer analysis proceeds in three basic steps, each of which will be discussed in thesections below:Step 1 - The returns are pre-processed to reduce noise and apply corrections for back-ground radiation, range, and other external factors such as overlap and detector4The LDA was initially developed with and for data from the mMPL system, and unless otherwise noted,the assumption for this section is that these are the data used as inputs to the process. However a modifiedversion was then generated to apply to CORALNet data, with some variations in the initial processing stepsdue to the differences in how each system generated its primary data product. The important aspects of theLDA can be applied universally to any LIDAR system, but wherever there are differences the methodologyused for CORALNet data will be explained.344.1. Layer analysis techniquesartefacts. The result is a normalized profile that is proportional to the atten-uated backscatter. Associated values such as SNR and depolarization ratio arealso calculated.Step 2 - Variance and wavelet analyses are used to find layer edges within each normal-ized profile. This includes using variance analysis to find regions of molecularscattering, using wavelets to estimate the altitudes of the upper edge of thePBL and of the edges of elevated aerosol layers, and the application of waveletsto depolarization ratio profiles within layers to find edges within complex layerstructures.Step 3 - Layer types are determined based on a number of criteria and LIDAR ratios aredesignated for each layer.Step 4 - Backscattter and extinction coefficient profiles are calculated.4.1.1 Step #1 - Pre-processingIn this step, all of the basic inputs to the LDA are calculated starting from the raw outputof the LIDAR systems. These calculated inputs include normalized relative backscatter(or in the case of CORALNet, backscatter ratio)Calculation of normalized relative backscatterRaw data files collected by the mMPL instrument undergo three separate corrections toaccount for intrinsic LIDAR and detector effects as a pre-processing step. These effectsinclude the region of incomplete overlap between the transmitter and receiver, as wellas second-order non-linearities in the response functions of the SPCM detectors. Thecorrection factors for each of these steps are provided by the LIDAR manufacturer as partof their initial calibration efforts, and are assumed to remain constant unless unexpectedreductions in data quality arise5.Dead time The nature of SPCM detectors is that after the detector is irradiated by apulsed electron current, the potential distribution is deformed such that subsequent electronmultiplication is suppressed for a period of time until it is neutralized by a strip current5These correction curves can be recalculated in the field if necessary354.1. Layer analysis techniquesflowing through the channel wall. During this time period, called the dead time, the gainfor incoming signals is unusually low. A correction function for this effect is provided bythe detector manufacturer in the form of a polynomial function that approximates signalloss due to photons arriving during the dead time. This correction applies for all signals,but it is close to negligible except for unusually high signal input levels. Equation 4.1 showsthe dead time correction polynomial for the mMPL unit owned by University of BritishColumbia (UBC) (unit MMPL5012).y = 6.335554× 10−8x7 − 4.714599× 10−6x6 + 1.389143× 10−4x5− 2.025010× 10−3x4 + 1.523593× 10−2x3 − 5.243415× 10−2x2+ 1.126443× 10−1x+ 9.714449× 10−1 (4.1)Afterpulse When operated in Geiger Mode (a.k.a. photon-counting mode), APD detec-tors are held at a bias slightly above the breakdown voltage. This results in a meta-stablestate that will break down with the introduction of a single electron-hole pair, resultingin a strong avalanche effect. This can be caused by the arrival of an incident photon, orby dark counts resulting from random fluctuations. When such random events coincidewith the arrival of an incident photon, the additional signal is referred to as the afterpulse.Because the mMPL uses APD detectors in photon counting mode, these afterpulse currentscan cause measurement errors that must be corrected.Background Each detector measurement includes a certain amount of signal-independentbackground. Contributions to the background include dark current, thermal current, andsolar background radiation. Prior to applying any range-based multipliers such as theoverlap correction or the range squared correction, the background noise level must besubtracted out from the signal. This can be easily achieved. One can assume that due toa relative lack of clouds and aerosols near the top of the LIDAR range (12 km to 15 km)combined with the total optical depth up to that range that the only measured signal fromthis part of the range is from the background. This can be averaged to obtain an estimateof the background level for that profile. Alternatively, in the case of mMPL data, theheader for each file contains a background signal value for each measured profile that canbe applied. Many systems (e.g.: CALIPSO) take a background signal measurement priorto the laser pulse being fired.364.1. Layer analysis techniquesOverlap Even though the mMPL uses an “optical transceiver” arrangement where thesame telescope is used for both the output and input light paths, it is still susceptibleto a region of incomplete overlap. This is addressed by applying an overlap correctionfunction, which is determined empirically by firing the laser on a clear day in a horizontalconfiguration where the backscatter and extinction coefficients are assumed to be constantthroughout the relevant portion of the profile. The mMPL achieves complete overlap within1 km so it is prudent to ensure roughly twice this range is clear of obstructions or variationsin backscatter to correctly calibrate the overlap function. The overlap correction factor isa unitless number representing the portion of the return signal that is seen by the receiverand ranges from 0-1. The correction is applied by dividing the signal by this factor afterthe afterpulse, dark current, and background corrections have been applied.These three corrections are combined in Equation 4.2 to determine the quantity knownas normalized relative backscatter. This quantity has units of [counts∗km2/µs∗µJ], whichis proportional to the attenuated integrated backscatter cross-section within the receiverFOV at a given range. This quantity can be used on its own for a variety of qualitativeanalyses (observing relative changes in AOD over time, distinguishing clouds from aerosols,etc.), but it can also be used as an input to estimating quantitative values for backscatterand extinction coefficients, provided the aerosol and cloud types within a given profile areknown, or at least a value is assumed for their extinction-to-backscatter ratios.B(z) =S(z) ∗ d(S)− a(z)− SB(z)E0 ∗O(z) ∗ z2, (4.2)where:z = range [km]B(z) = normalized relative backscatter [counts∗km2/µs∗µJ]S(z) = total measured signal at the detector [counts/µs]d(S) = dead time correction factor [unitless]a(z) = afterpulse correction factor [counts/µs]SB(z) = background signal at the detector [counts/µs]E0 = laser output energy [µJ ]O(z) = overlap correction factor [unitless].This value is calculated for both of the detected polarization states (values identifiedas B1 and B2). Because the same detector and optical path is used for both, the dead374.1. Layer analysis techniquestime and overlap correction functions are the same for both channels, but because thedetected pulse strength is affected by the state of the liquid crystal retarder, the afterpulsecorrection is slightly different.Once initial corrections are made for the laser pulse energy, overlap and z2 factors, thenormalized relative backscatter can be expressed in similar terms to the LIDAR equation(Equation 2.20).B(z) = C[βp(z) + βm(z)]exp−2z∫z′=0[αp(z′) + αm(z′)]dz′ (4.3)=S0(z)z2E0O(z)(4.4)This is simplified toB(z) = C[βp(z) + βm(z)]T r0 , (4.5)where T r0 represents the integrated optical depth from the surface to the range zT r0 = exp−2z∫z′=0[αp(z′) + αm(z′)]dz′ (4.6)As described in Equation 2.21, the remaining system constant (C) represents systemlosses and detector responsivity and is independent of range and signal. Thus the Nor-malized Relative Backscatter (NRB) is proportional to the attenuated backscatter and issuitable as an input for any stable inversion methodology based on the first derivative ofthe signal profile [e.g.: 31, 57, 58]. The NRB is also the primary data product for themMPL making it an uniquely useful quantity for this system in particular.Calculation of backscatter ratioThe primary data product of the CORALNet system is the backscatter ratio. In simpleterms, this is defined as the ratio of the measured signal to the signal that would bemeasured from a purely molecular atmosphere. In concrete terms the Backscatter Ratio(BR) can be derived from the LIDAR equation as follows:384.1. Layer analysis techniquesΓ(z) =S0(z)1z2E0CO(z)βm(z) exp{−2z∫z′=0αm(z′)dz′}=[βp(z) + βm(z)]βm(z)exp−2z∫z′=0αp(z′)dz′ . (4.7)Like the normalized relative backscatter, this product is analogous to attenuated backscat-ter and can be used in any of the aforementioned inversion strategies, however in the caseof the backscatter ratio, the inversion requires an extra step, which will be described inmore detail in Section 4.1.4.Calculation of additional data products:Some other quantities are required by the LDA in addition to the primary data product.These include the linear depolarization ratio, the signal variance, and the SNR.Linear depolarization ratio The quantity known as the linear depolarization ratio fora LIDAR signal is defined as the ratio between the scattered light at a given range forwhich the polarization is perpendicular to that of the emitted laser pulse and the light thatis polarized parallel to the pulse.δlinear =S⊥S‖. (4.8)As mentioned in Appendix A.1, the mMPL uses a novel combination of an optical-transceiver-style layout and a liquid crystal retarder to monitor the depolarization ratiousing a single detector. One result of this method of polarization detection is that thesystem does not switch between detecting two orthogonal linear polarizations. Rather,the mMPL only passes light which is polarized perpendicularly to the initial laser pulsethrough to the detector, and switches between light that is in phase with the outgoingpulse (φ = 0) and light that has a φ = pi/2 phase shift. The linear depolarization ratio canbe related to the mMPL depolarization ratio as in Equation 4.9. The associated Stokesvectors and Mueller matrices to support this are described in detail in [34].394.1. Layer analysis techniquesδ =δMPLδMPL + 1, (4.9)whereδMPL =B2B1=B⊥(φ = pi/2)B⊥(φ = 0).In the case of CORALNet, the calculation of depolarization ratio is ostensibly simple,because the two quantities S‖ and S⊥ are measured directly. However, there is a technicalissue that can arise with this ratio because unlike the mMPL the CORALNet system usesseparate detectors for the two channels. The result is that a range-dependant calibrationconstant must be calculated and applied to the ratio in order to correct for variations indetector responsivity. To some extent this is achieved by the CORALNet team as partof their data processing procedure, however, since this difference has not been quantified,one must consider errors in the correction as a source of uncertainty in the depolarizationmeasurements.δ = Cδ(z)S⊥(z)S‖(z). (4.10)Variance and SNR calculation Most of the layer masking operations that follow relyon the ability to distinguish actual aerosol and cloud layers from the random noise com-ponent inherent in any photoelectric signal. Thus, it is important to calculate a varianceand an SNR value for each data point in a given LIDAR profile and to derive a thresholdSNR value below which it is not feasible to apply these methods to the data. It is knownthat for properly operated SPCM detectors, the primary noise component is shot noise,or Poisson noise. This is the random variance of any light signal arising from the discretenature of photons and the probability of conversion of that photon to a photoelectron bythe dynode. As shown in Equation 4.11, the total noise at a range r for a profile taken attime t can be thought of as the Root Mean Squared (RMS) combination of shot noise fromthe LIDAR signal, the background radiation, and the dark current:σT (z) =√σS(z)2 + σB(z)2 + σD(z)2, (4.11)404.1. Layer analysis techniqueswhere:σT = total RMS noise [counts/µs]σS = noise from measured LIDAR signal [counts/µs]σB = noise from background radiation [counts/µs]σD = noise from dark current. [counts/µs].The probability distribution for photon detection can be described by a Poisson distri-bution, which means that as the number of detected photons increases, the noise (repre-sented by the standard deviation of the distribution) approaches a value that is proportionalto the square root of the mean signal level:σT (z) = G√S(z) = G√S0(z) + SB(z) + SD(z), (4.12)where:G = a unitless scaling factorS = total signal measured by the detector [counts/µs]S0 = signal from LIDAR pulse [counts/µs]SB = signal from background radiation [counts/µs]SD = signal from dark current [counts/µs].The scaling factor can be calculated by examining a portion of the LIDAR range whereit is reasonable to assume there is negligible signal from the LIDAR pulse (S0 = 0). FormMPL data, the default assumption for this range is the topmost 50 values, which typicallyrepresent the 13.5 km to 15 km range. For this range of the LIDAR return, Equation 4.12simplifies to:σT (z) = G√SB(z) + SD(z). (4.13)By finding the mean and standard deviation of the signal in this region, the scalingfactor can be calculated as:G(z) =σT (z)√SB(z) + SD(z). (4.14)This scaling factor G is necessary to account for the impact on noise statistics as aresult of averaging multiple measurements together over a period of integration. G can414.1. Layer analysis techniquesnow be applied to calculate the noise level at any point in the profile by substituting itinto Equation 4.12. Thus the SNR is calculated with Equation 4.15:SNR =S0σT (z)√SB(z)+SD(z)√ST (z). (4.15)This equation can be applied to either of the LIDAR channels. In the case of CORAL-Net backscatter ratio values, the signal has already been processed to remove contributionsfrom the background and dark currents, so the SNR can be calculated directly:SNR =S0(z)σT (z). (4.16)4.1.2 Step #2 - Variance and wavelet analysisMolecular profile - variance analysisOnce the initial corrections have been applied to calculate the normalized relative backscat-ter and the SNR has been characterized, the next step in identifying aerosol and cloudlayers is distinguishing them from the molecular background. For ground-based LIDARs,it is always a challenge to apply a simple threshold-based analysis to the data in order tofind layer edges because signal calibration is challenging and even in cases where it canbe accomplished, the extinction from the PBL and other intervening layers within a pro-file is unknown. It is thus a more effective and universally applicable solution to apply avariance-based approach. The variance analysis used here is based upon the one developedfor the STRucture of the ATmosphere (STRAT) LIDAR system in [72]. The basis of thisapproach hinges on the idea that even if the absolute values of the retrieval are affectedby extinction, for regions where molecular scattering dominates, the profile shape shouldclosely match an idealized clear-air profile derived from the profile of atmospheric tempera-ture and pressure6. Furthermore, in regions of primarily molecular scattering, the varianceof the LIDAR signal around such a profile should be determined by the shot noise as de-scribed in Section 4.1.1. Substantial deviation from this variance is taken as an indicationof the presence of a layer of particles providing additional scattering signal.In order to compare a given profile of attenuated normalized relative backscatter valuesto an idealized profile of molecular backscatter coefficients, it is first necessary to devise a6In the absence of actual temperature and pressure measurements for a given location, or another user-defined profile, the default clear-air profile used is the US Standard Atmosphere, 1976 [77]424.1. Layer analysis techniquesscaling factor that can be used to bring them into a common scale. This value is calculatedand averaged over a sliding window centred around each measurement, as in Equation 4.17:Ψ =z+N∑z′=z−Nβm(z′)B(z′)2N + 1, (4.17)where: βm = molecular backscatter coefficient [km−1sr−1].βm can be calculated directly for a given profile of atmospheric temperature and pres-sure under the assumption of Rayleigh scattering as described in Section 2.1.2 Once thevalue of Ψ is established, the variance of the LIDAR signal around this normalized valuecan be expressed as:ω(z) =z+N∑z′=z−N{1z′2[B(z′)− 1Ψ(z)βm(z′)]}22N + 1. (4.18)If the variance in normalized relative backscatter for a region where molecular scatteringdominates can be assumed to be σT2 then one can define a threshold value qvar such that ifω(z) ≥ qvarσT 2, one can infer that the additional variance could be the result of increasedscattering from species other than the molecular background, and these regions are flaggedas layers to be examined more closely.Molecular detection with CORALNet backsactter ratios One of the benefits ofCORALNet data is that it is not necessary to employ the variance analysis as definedin Equation 4.18 to find regions of molecular backscatter. In these regions, according toEquation 4.7 the value of a properly calculated backscatter ratio will obviously approachunity while the slope appraoches zero. Therefore once the backscatter ratio has been calcu-lated, a simple threshold is applied to the ratio to determine regions where the particulatebackscatter is negligibly low. The downside to this method is that it breaks down in regionsof low SNR, especially at higher altitudes or above optically thick layers. Thus, in order toavoid difficulties, regions where the SNR falls below a predetermined threshold are maskedout and assumed to be free of particulates for the purposes of performing inversions withthese data.434.1. Layer analysis techniques(a) ”Mexican Hat” wavelet (b) First derivative of Gaussian waveletFigure 4.1: These two wavelets are the most commonly used in layer analysis in LIDARdata. In this instance the units of the axes are irrelevant as they can be applied to anyfunction. It is only the shapes of the wavelets that is important.Layer edge detectionAlthough the variance analysis is extremely sensitive, it does have some inherent flaws.Firstly, and most importantly, it has difficulty identifying the upper edges of the PBL.Secondly, in regions of low SNR, it can lead to false positives. Finally, it can lead to theidentification of regions that, while they can not be classified as purely molecular, alsoexhibit an SNR in the depolarization ratio that is too low to be further identified as aparticular aerosol type. Therefore, in conjunction with the variance analysis, layer edgesare located using wavelet analysis to determine which regions of a given profile exhibitsufficiently strong scattering signals to be categorized. Regions that exhibit variance thatexceeds the defined threshold but do not have sufficient SNR are labelled as “UnidentifiedAerosols”.Wavelet analysis The wavelet analysis technique used here is based upon the algorithmdeveloped by [11] and used again by [72]. The process involves using a Continuous WaveletTransform (CWT) to identify sharp changes in slope. This method was found to be superiorto the slope threshold techniques used by some LIDAR analyses (e.g.: CALIPSO) becauseof its resistance to the effects of noise in the signal and its applicability to a range of LIDARmeasurement techniques. The most commonly used wavelet function for finding layer edgesin LIDAR signals is the second derivative of the Gaussian function (the so-called “mexicanhat” wavelet shown in Figure 4.1) A).Once the CWT is calculated for a range of dilation values, the modulus of the CWTcoefficients is calculated. The maxima of this function represent the layer peaks and the444.1. Layer analysis techniquesFigure 4.2: Analysis of a LIDAR profile using a continuous wavelet transform. PanelA shows the initial profile. Layer bases are marked with red circles, peaks with greentriangles and tops with blue x’s. Panel B shows the continuous wavelet transform of theprofile. Panel C shows a skeleton plot of peaks (black) and valleys (red) in the CWTminima represent the bottom and top edges of layers. There is a trade-off here betweenprecision and noise tolerance. Using larger wavelet dilation values will avoid markingrandom signal spikes from noise as layers, but will result in lower precision in the locationof layer edges and peaks. The ideal wavelet dilation value, in number of data points, is afunction of the SNR. The current algorithm uses a constant value of four data points forall profiles, but in a future iteration an adaptive value based on SNR may be implemented.The use of this technique is demonstrated in Figure 4.2.Determination of PBL height The core purpose of the PBL height determinationfunction in the LDA is to identify cases where a well-mixed boundary layer is dominatedby surface-emitted aerosols, thereby affecting the LIDAR ratio. The “Mexican hat” waveletis not ideal for identifying the top edge of the PBL because the shape of the signal profile454.1. Layer analysis techniquesis different. For this region the first derivative of the Gaussian is used (See Figure 4.1 B).Some previous applications of the wavelet technique have also applied a Haar wavelet forthis purpose [e.g.: 93, Cohn and Angevine]. The detection of PBL height using LIDAR isfraught with difficulties and carries with it a large degree of uncertainty (for more on this,see Section 4.2.3), so this is an optional feature of the LDA.Wavelet analysis in depolarization ratio profiles One novel aspect of the LDA isthe use of depolarization ratio profiles to find edges between layers. Most other layeridentification algorithms to date are based purely on the backscatter profiles [e.g.: 72, 79],or when such information is available, on returns from multiple spectral channels in HSRLsystems [e.g.: 12]. Using this methodology, it is possible to miss the boundary betweentwo adjacent layers for which the backscatter strength does not differ significantly. Onecommon example of this occurs within cloud layers near the point of transition from waterto ice or vice-versa. Another is in stable atmospheric conditions where layers of differentaerosol types coexist without fully mixing.The method used for edge detection within the depolarization profile is almost com-pletely analogous to the wavelet analysis described above for the normalized relativebackscatter profile. However, as the ratio of two signals, both with a substantial noisecomponent, the SNR of the depolarization ratio is too low to reliably be used as a primarybasis for layer demarcation. Therefore, in the LDA its application is limited to findingedges within already defined layers where the SNR levels tend to be higher due to he in-creased backscatter. Once a layer is identified, the next step is to apply a wavelet analysisto the depolarization ratio profile within the layer to determine if there are any changes indepolarization that are both large enough to exceed the background noise threshold in thedepolarization ratio profile, and sustained enough to indicate a real change in depolariza-tion and not a random change in value. If edges are found that pass these two tests, thelayer is divided into sub-layers at that point.4.1.3 Step #3 - Layer type discriminationThreshold analysisOnce layers have been identified within a given profile, the next goal is to classify themunder one of the seven predetermined categories shown in Table 2.5. These layer iden-tifications are used to define the layer LIDAR ratios used in generating backscatter and464.1. Layer analysis techniquesextinction coefficients.In addition to the categories in Table 2.5, there are two additional categories used inthe LDA: “PBL” is an optional category sometimes used to represent aerosols within thewell-mixed PBL. The LIDAR ratio for this category is set to 30 sr and is meant to representa mixture of various unidentified aerosol types. The final category is “Insufficient Signal”which is reserved for regions where the SNR is too low to properly make any determinationabout whether a layer exists or not, or for when the LDA returns nonsensical or physicallyimpossible values (such as negative depolarization ratios).Cloud vs. aerosol The first step in layer analysis is to differentiate between clouds andaerosols. The primary method for doing so is based upon the difference between the meannormalized relative backscatter value of a layer and the normalized relative backscatter ofthe background molecular scattering in the surrounding region. Analysis of this value forlayers detected from multiple locations over a range of circumstances reveals a distinctlybi-modal distribution resulting from the generally higher optical depth of clouds as opposedto aerosol layers (Figure 4.3). From examination of this distribution, it is clear that thereis a marked difference between the normalized relative backscatter distributions of aerosolsand clouds.Based on the results shown in Figure 4.3 a basic threshold normalized relative backscat-ter value of 1.0 seems to be sufficient to avoid most cases of aerosols being identified asclouds, but not vice versa, so this alone is not sufficient. A large portion of the overlapat lower normalized relative backscatter values is due to thin cirrus clouds, which occupymuch the same regions of the normalized relative backscatter range as aerosols. Luckily,these clouds can be identified by their altitude and high depolarization ratio signatures.When one compares the range of normalized relative backscatter values for clouds andaerosols with depolarization ratios higher than 0.25, a different threshold emerges at annormalized relative backscatter of 0.2 (see Figure 4.4). This value is employed as a baselinefor discriminating between optically thin aerosol layers and cirrus clouds.In addition to the pure signal analysis technique, there are some constraints that canbe imposed on the process of layer identification in order to speed up the process and avoidfalse results. One example of this kind of constraint is the addition of an altitude threshold.High-altitude cirrus clouds can be so optically thin as to be confused with aerosols, evengiven the high-depolarization threshold described above. In this instance, the algorithmrelies on the fact that aerosols and optically thin cirrus clouds are rarely found at the same474.1. Layer analysis techniquesFigure 4.3: Histograms of normalized relative backscatter values collected from over 4,000hours of mMPL operation from multiple locations for altitudes up to 10 km and for alldepolarization ratios. These results show the distinct differences in normalized relativebackscatter distributions for the two major categories. Note that the x-axis is cut off at annormalized relative backscatter of 5.0, but the tail of the range of cloud normalized relativebackscatter values extends to over 100484.1. Layer analysis techniquesFigure 4.4: Histograms of normalized relative backscatter values collected from over 4,000hours of mMPL operation from multiple locations for altitudes up to 10 km. These resultsshow the distinct differences in normalized relative backscatter distributions for the twomajor categories for depolarization ratios over 0.25. Note that the x-axis is cut off at annormalized relative backscatter of 1.0, but the actual range of cloud normalized relativebackscatter values extends much higher494.1. Layer analysis techniquesaltitudes. Thus, highly depolarizing layers at altitudes higher than 10 km are automaticallyassumed to be clouds, even if their mean backscatter is below both aerosol-cloud thresholds.This is also true for layers at any altitude for which the mean depolarization ratio exceedsa value of 0.45. Depolarization of this magnitude is rarely observed in aerosols of any kindand is highly indicative of ice crystals, so layers with mean depolarization values this highare automatically identified as clouds.Cloud types Table 2.5 identifies three distinct cloud types based on their depolarizationratios: water clouds, ice clouds, and mixed-water-and-ice clouds (mixed clouds for short).Water droplets, being essentially spherical, alter the polarization state of scattered photonsvery little, if at all, resulting in perhaps the lowest mean depolarization ratio of any observedlayer. Conversely, ice crystals tend to be more strongly depolarizing than any commonlyoccurring aerosols7. In real clouds, these two easily distinguishable particle types are oftenmixed together in cloud layers, which necessitates the third designation of mixed cloudsfor layers where the depolarization ratio falls between these two extremes.In many LIDAR systems, the distinction between cloud types becomes more difficultas a result of multiple scattering effects within cloud layers. When multiple scatteringis taken into account, there is a strong non-linear relationship seen between backscatterand depolarization within cloud layers [e.g.: 17, 44]. As shown in Figure 4.5, there areclearly definable differences in this relationship for ice and water clouds, but the values ofdepolarization change drastically as the beam traverses the cloud and multiple scatteringeffects become more prominent. This non-linearity does make cloud type analysis morecomplicated, but it can actually be utilized in the difficult task of differentiating betweencirrus clouds and dusty aerosol layers, as dust does not exhibit the same multiple scatteringrelationship that ice does [121].However, for the mMPL LIDAR, the FOV is so small that these multiple-scatteringeffects are not detected. This can be seen when one performs a similar analysis of depolar-ization vs. normalized relative backscatter for the mMPL. In Figure 4.6) there is clearly nodiscernible relationship between depolarization ratio and normalized relative backscatterfor any of the three cloud types. As a result it is possible to employ a simple depolarizationratio threshold to distinguish between cloud types when the LDA is applied to the mMPLor any other system with a sufficiently small FOV.7The notable exception to this statement is ice platelets that can occur under certain conditions at highaltitudes and can result in specular reflection with correspondingly low depolarization ratios [92].504.1. Layer analysis techniquesFigure 4.5: Figure reproduced from [44] showing summary statistics for depolarization-backscatter relationships for opaque clouds detected by CALIPSO during July (left panel)and October (right panel) of 2006. The green dashed line represents the theoretical re-lationship for spherical water droplets, the red dashed line shows the relationship for icecrystals.Aerosol types Layers that are not identified as clouds using one of the three thresholdsare assumed to be aerosols. For these layers, the aerosol type is determined by the layermean depolarization ratio. Table 2.5 shows the depolarization ratio thresholds for the threeaerosol types currently used in the LDA: “dust”, “polluted dust”, and “smoke / urban”.The categories are necessarily broad due to the limitations of using only depolarizationratios to distinguish between aerosol types.The term “smoke / urban” is used as a blanket term for mostly spherical aerosoldroplets or black carbon conglomerates that exhibit generally low depolarization ratios.Sulphates and nitrates would fall into this category as would many forms of pollen andaerosols generally referred to as “urban aerosols”. Marine aerosols would also fall intothis category, but these are limited to certain geographical areas and are not part of thelayer designations at this time. Dust, with it’s elongated crystalline structure, is knownto be among the more highly depolarizing of aerosols and is also easily identified by thedepolarization ratio. The “Polluted Dust” category is used for any mixture of dust withmore spherical aerosols. The name is taken from one of the CALIPSO VFM categories, butin fact it can be somewhat misleading because the range it covers includes mixtures thatmostly comprise more nearly spherical particles (such as from particularly intense wildfires514.1. Layer analysis techniquesFigure 4.6: Two-dimensional histograms of layer-integrated depolarization ratio vs. NRBfor clouds collected by the mMPL instrument in 2014. None of the included panels showsthe distinct relationships between depolarization and backscatter observed in the CALIPSOdata as a result of multiple scattering effects. Panel (a) shows the results for all cloud layers,Panel (b) shows just layers identified as water clouds, Panel (c) shows layers identified asmixed ice and water, and Panel (d) shows layers marked as ice clouds.524.1. Layer analysis techniqueswhen soil particles are lifted into the smoke plumes by the strong convection currents - fora concrete example of this, see Section 5.3)The label “unidentified aerosol” is reserved for layers for which the SNR is too low toreliably determine the mean layer depolarization ratio. For such layers a default LIDARratio of 30 sr is assumed4.1.4 Step #4 - Calculation of backscatter and extinction coefficientprofilesThe most direct result of the layer identification process is the most important: the calcu-lation of backscatter and extinction coefficient profiles. These profiles are valuable becausethey bring a LIDAR scene from an essentially qualitative tool where layer values can onlybe interpreted in comparison to the background of the scene itself, to quantitative results,in well-defined units, that can be compared to results from other sources and consideredin absolute terms. That being said, the LIDAR equation is, by its very nature, not analyt-ically solvable due to the fact that both the extinction and backscatter coefficients at eachrange value are unknowns, and even stable solutions based upon simplifying assumptionslike a power-law relationship between the two unknown quantities can produce fairly largeerrors. There is no one method that has been developed to accomplish this task that worksin all conditions. Some, such as those proposed in [57] and [58] are particularly well-suitedto highly turbid atmospheres but run into difficulties when visibility is high. Others workfor a wider range of visibilities but rely on a predetermined and constant LIDAR ratiofor the profile (e.g.: [31]) but this assumption can result in compounding errors due tomultiple different cloud and aerosol layers where changes of up to ±200% in LIDAR ratioare possible.For this study a variation on the inversion proposed in [57] is used because it allowsfor variation in LIDAR ratio for different layers within the profile, utilizing the LDA toimprove backscatter and extinction calculations. This inversion relies on first convertingthe standard LIDAR equation into a solvable differential equation through a simple re-formulation of the standard LIDAR equation from 2.20 by taking the natural log of therange-corrected signal:U(z) = ln(z2S0(z)), (4.19)where S0 is defined in Equation 2.21. Is is clear that this formulation would also534.1. Layer analysis techniqueswork when applied to the normalized relative backscatter quantity B(z) calculated in 4.5(V (z) = ln(B(z))) since the normalized relative backscatter is proportional to the range-corrected signal.In Equation 2.22, the LIDAR ratio R(z) is defined as a range-dependant quantity thatrelates the extinction α to the backscatter β. If one defines R′ as the inverse of this ratio(so β = R′ακ), the change of the logarithm of the normalized relative backscatter V withrange can be expressed as a differential equation:dVdz=1R′dR′dz+καdαdz− 2α. (4.20)Klett shows that the stable solution to this equation takes the form:α(z) =(R′ref/R′)1/κexp [(V−V ref )/κ]{α−1ref +2κzref∫z(R′ref/R′)1/κexp [(V−V ref )/κ] dz′} . (4.21)In Equation 4.21, the values R′ref ,V ref , and αref represent the values for each of theserange-dependant quantities at a reference range zref that has been selected and for whichit is assumed that the extinction and LIDAR ratio are well-defined. Since this solution isapplied from the top of the profile down in an effort to improve stability, it is commonto try to set the reference range to a region of the profile free of aerosols for which pureRayleigh scattering is a valid assumption. In the absence of better information on the valueof κ, it is also common to start with the assumption κ = 1.Once the extinction profile is determined using this inversion, it is a simple matter touse the range-dependant LIDAR ratio to calculate the associated backscatter.There is a complication with using this methodology in the context of the proposedLDA. The LIDAR ratios defined by the defined layers are specific to the particulate scatter-ing species in the layer. The actual LIDAR ratio for a given volume of integration would bedetermined by this as well as the Rayleigh scattering component. For optically thick layerswhere βp >> βm, this difference is negligible, but for thinner layers it is more important.For this reason, it becomes necessary to separate the particulate and molecular com-ponents as in Equation 4.22:β(z) = R′p(z)αp(z) +R′m(z)αm(z). (4.22)Klett notes that this equation can be substituted into 2.20 to derive a new differential544.2. Evaluation of resultsequation, the solution of which is described in detail in [57]. Only the final result isreproduced here as Equation 4.23:β(z) =exp(∆V ′)βm−1 + 2zref∫zexp(∆V ′dz′)R′p, (4.23)where:∆V ′ = V − V ref + 2R′mzref∫zβmdz′ − 2zref∫zβmdz′R′p. (4.24)To apply this inversion, the particulate backscatter ratio cannot be equal to zero,making it difficult to apply in regions of high visibility where the particulate backscatteris near zero making the particulate LIDAR ratio R′p indeterminate. To avoid this issue theparticulate LIDAR ratio is set to the same value as the molecular LIDAR ratio R′p = R′m =3/8pi in regions identified as “clear air”. These regions have been identified either throughvariance analysis (for mMPL data) or slope threshold analysis (for CORALNet data) tovary only slightly from the profile expected for the standard molecular atmosphere. Theassumption here is akin to saying that any deviation from the backscatter of the standardatmosphere in these regions is the result of variations in temperature and pressure ratherthan the presence of aerosols.4.2 Evaluation of resultsIt has been pointed out before that the primary limitation of the LDA is the fact that,by working with a limited amount of data, it must rely on some broad generalizations toinfer specific parameters from mean layer depolarization and integrated backscatter mea-surements, which are essentially only proxies for aerosol type. Furthermore, the tables inSection 2.3 show conclusively that even within specific categories the key optical proper-ties (LIDAR ratio, depolarization ratio, and extinction and backscatter coefficients) canvary considerably, even overlapping each other in many instances. For example, dependingon the conditions and the specific kind of layer, LIDAR ratios can and do vary by up to150% even within a single category. It is therefore important to perform, to the extent554.2. Evaluation of resultspossible, an evaluation of the performance of the LDA in the presence of the unavoidableuncertainties inherent in the methodology.This evaluation proceeds along two paths. The first is quantitative and is aimed specif-ically at attempting to determine the performance of the profile inversion algorithm usedto calculate backscatter and extinction coefficients in the presence of three root causes oferror: uncertainty in the assigned LIDAR ratio, random noise in the profile, and uncor-rected background signal. This is achieved through the use of idealized profiles with one ormore layers of known backscatter and extinction, which are then inverted using the samemodified Klett algorithm used in the LDA under a variety of error conditions.The second path is qualitative and involves applying the full LDA to several test casesin various locations and under various conditions to determine how well it performs. Thesecases are qualitative because the exact nature of the aerosol and cloud layers is not knownso the analysis is based on inferences from what are considered likely or expected outcomesand, where available, supporting information from external sources. These test cases serveto highlight some of the strengths and weaknesses of the LDA in practice as well as someongoing issues that have yet to be resolved.4.2.1 Quantitative tests using idealized and empirical profilesThe performance of the modified Klett inversion algorithm used in the LDA was testedunder a range of different circumstances using theoretical profiles as a starting point. Thetwo primary sources of error in these inversions are uncertainty in LIDAR ratio and theeffects of noise, particularly at high altitudes.Generation of idealized profilesIn these tests, profiles of molecular backscatter, extinction, and the resulting signal atthe detector were generated based on temperature and pressure profiles using the USStandard Atmosphere, 1976 [5]. The scattering and extinction coefficients are calculatedfrom these values according to the assumption of Rayleigh scattering. Then, if an aerosollayer is to be inserted, it is added to the profile at the desired altitude with fixed valuesfor backscatter coefficient and LIDAR ratio. The flux at the aperture is calculated bydetermining the backscatter and total integrated two-way extinction for each altitude stepalong the profile, taking into account the output radiance and “range-squared” losses. If abackground value and noise component are to be considered, they are added to the profile at564.2. Evaluation of resultsthis point8. Factors such as multiple scattering, Ozone absorption, and optical system lossesincluding incomplete overlap and detector responsivity are not taken into consideration.A simulation of the normalized relative backscatter data product is generated from theat-aperture radiance by performing a “range-squared” correction and normalizing by theoutput radiance. As with many LIDAR inversion algorithms, the absolute value of theat-aperture radiance is not important, just the shape of the profile, so for these tests, thevalues are all normalized to an output radiance of unity.Effects of errors in LIDAR ratioEven moderate errors in LIDAR ratio have been shown to have significant impact on inver-sion results [e.g.: 59] Given the aforementioned uncertainties in assigned LIDAR ratio fora given layer, it is worth investigating how much of an improvement is actually achieved byapplying the LDA to make layer-specific LIDAR ratio assignments. The simplest inversionmethodology for lidars operates under the assumption of a single, constant LIDAR ratiothroughout the profile that represents both molecular and particulate scatterers [31]. Thismethod is particularly useful in turbid atmospheres where a single layer dominates themajority of the usable range. For areas of intermittent layers interspersed among regionsof relatively clear air, an inversion was later developed that separated the contributionsfrom the background atmosphere and aerosols [57]. The method employed in the LDAgoes a step further by applying separate LIDAR ratios to each layer and also to regionsidentified as clear air [57]. These methods will be referred to as the single-component,two-component, and layer-specific inversions, respectively.The first test case involves a single elevated layer from 7.5 km to 9.5 km in an otherwiseclear atmosphere. The test was repeated for a range of backscatter coefficients from βp =1e−4 km−1sr−1 up to βp = 1e−2 km−1sr−1 and for LIDAR ratios ranging from R = 15 srto R = 85 sr. For each test input LIDAR ratios of 8pi/39, 35, 65, and 85 sr are applied andthe results are collected to determine how each strategy performs under erroneous LIDARratio assumptions. Representative examples of the results are shown in Figures 4.8,4.9,and 4.10. In these tests, the focus is on the extinction results form the inversion, as theseare both the most useful and the most sensitive to inversion errors.Note that when a LIDAR ratio is applied in the single-component inversion, it is as-8The noise is approximated as a zero-mean random Gaussian component despite the fact that the actualstatistical distribution of random shot noise is technically a Poisson distribution.9This is the LIDAR ratio for Rayleigh Scattering574.2. Evaluation of resultsFigure 4.7: Plot of an example idealized profile with a simple layer added. Panels A andB show the profiles of temperature and pressure from the US Standard atmosphere 1976.Panels D and E are profiles of backscatter and extinction coefficients based on the atmo-spheric density derived from the temperature and pressure with the addition of a simulatedlayer from 2.5 km to 4.5 km with an aerosol backscatter coefficient of 1× 10−3 km−1 sr−1and a LIDAR ratio of 35 sr. Panel C shows the signal at the aperture, normalized by theoutput radiance. Panel F shows that same signal after “r-squared” correction, which isequivalent to the normalized relative backscatter product provided by the mMPL584.2. Evaluation of resultssumed to represent the combined LIDAR ratio of all scattering species in the profile. Whenit is used in the two-component algorithm it is also applied to the entire profile, but it isassumed to represent only the contribution of aerosol scatterers with the contribution ofmolecular scattering considered separately. In the layer-specific algorithm, the selectedLIDAR ratio is applied only to the layer(s) and a molecular LIDAR ratio is applied toregions that are assumed to be aerosol free.The results of this test show conclusively that the layer-specific algorithm outperformsthe others in any scenario where there is a large range of LIDAR ratios represented inthe profile. This includes any profile where layers of aerosols exist against a backgroundof regions of clear air, which is the most common mode of operation for LIDAR in mostconditions. For the single-component and two-component algorithms, if the LIDAR ratiois set to match the clear regions it runs the risk of underestimating the extinction ofthe aerosol layer(s), and when it is set to match the layers, the result is a substantialoverestimation of the extinction in the rest of the profile. The two-component solutionperforms better in this regard, but not as well as the layer-specific algorithm. In addition,the layer-specific algorithm is more robust in the case of unidentified layers, and of courseit performs better for profiles with multiple layers of differing LIDAR ratios.The maximum error in a profile is determined by the altitude, backscatter coefficient,and LIDAR ratio of the layer in addition to the actual error in the assigned LIDAR ratio.However, it is interesting to note that these influences are small compared to the LIDARratio error. This is due to the fact that all three algorithms are applied using a far pointboundary condition and once errors in calculated extinction occur, they propagate andcompound down through the rest of the profile as long as the error persists.This is evident when one compares the results in Figure 4.9 to those in Figure 4.10. Inthese cases the backscatter coefficients vary by two orders of magnitude, and this does havean important effect on the errors within the layer itself, but the maximum errors resultingfrom applying a LIDAR ratio of 85 sr to the profile, are virtually identical.Inverting layers with noiseIn any real profile of LIDAR data, noise is always an important consideration. Noisein LIDAR profiles arises primarily from standard shot noise and thermal noise and isproportional to the square root of the measured signal, including background components.This can lead to very low SNR values toward the top of the usable LIDAR range where594.2. Evaluation of resultsFigure 4.8: Example test results involving LIDAR ratio errors for three different inversionstrategies. The original input data includes a single layer from 7.5 km to 9.5 km with aparticulate backscatter coefficient of βp = 1e−3 km−1sr−1 and a LIDAR ratio of R = 65 sr.In each case, input LIDAR ratios ranging from 8pi/3 up to 85 sr are used. Panel A showsthe results of the single-component inversion. In this instance the chosen LIDAR ratio isapplied as a constant value for the entire profile. Panel B shows the results of the two-component inversion, where the LIDAR ratio represents only the aerosol component - butis constant throughout the profile. Panel C shows the result of the layer-specific inversionwhere the chosen LIDAR ratio is used to represent only the layer in question and theLIDAR ratio is held at 8pi/3 for the rest of the profile. The original profile of extinctionvalues is plotted as the solid black line in all three panels.604.2. Evaluation of resultsFigure 4.9: Example test results involving LIDAR ratio errors for three different inversionstrategies. The original input data includes a single layer from 7.5 km to 9.5 km with aparticulate backscatter coefficient of βp = 1e−2 km−1sr−1 and a LIDAR ratio of R = 65 sr.In each case, input LIDAR ratios ranging from 8pi/3 10up to 85 sr are used. Panel A showsthe results of the single-component inversion. In this instance the chosen LIDAR ratio isapplied as a constant value for the entire profile. Panel B shows the results of the two-component inversion, where the LIDAR ratio represents only the aerosol component - butis constant throughout the profile. Panel C shows the result of the layer-specific inversionwhere the chosen LIDAR ratio is used to represent only the layer in question and theLIDAR ratio is held at 8pi/3 for the rest of the profile. The original profile of extinctionvalues is plotted as the solid black line in all three panels.614.2. Evaluation of resultsFigure 4.10: Example test results involving LIDAR ratio errors for three different inversionstrategies. The original input data includes a single layer from 7.5 km to 9.5 km with aparticulate backscatter coefficient of βp = 1e−4 km−1sr−1 and a LIDAR ratio of R = 65 sr.In each case, input LIDAR ratios ranging from 8pi/3 up to 85 sr are used. Panel A showsthe results of the single-component inversion. In this instance the chosen LIDAR ratio isapplied as a constant value for the entire profile. Panel B shows the results of the two-component inversion, where the LIDAR ratio represents only the aerosol component - butis constant throughout the profile. Panel C shows the result of the layer-specific inversionwhere the chosen LIDAR ratio is used to represent only the layer in question and theLIDAR ratio is held at 8pi/3 for the rest of the profile. The original profile of extinctionvalues is plotted as the solid black line in all three panels.624.2. Evaluation of resultsthe laser pulse is almost completely attenuated, especially when optically thick layers existlower down in the profile or when the signal from background solar radiation increases11.For a discussion of sources of noise in LIDAR profiles see Section 4.1.1.All of the standard LIDAR inversion methodologies that utilize a far-end boundary con-dition are sensitive to noise toward the upper end of the profile. Therefore, it is importantto investigate the impact of random noise on the inversion results in order to develop limitson acceptable noise levels for using the inversion to calculate backscatter and extinctioncoefficients. In order to achieve this, a Monte-Carlo simulation was used where a level ofrandom noise was added to a given profile to achieve a prescribed SNR value for the top1 km of the profile and the result is then inverted. The process is then repeated 500 times.The first case utilizes a profile representing pure molecular scattering with no aerosollayers. The distributions for some example SNR values are shown in Figure 4.11. Theresults of this analysis reveal a clear breakpoint in the effect of increasing SNR at a valueof 2.0. Below this point, the mean value of the Monte-Carlo runs falls well below the originalprofile and the standard deviation actually increases with increasing SNR. In this range,the profiles actually include non-physical negative extinction coefficient values. At an SNRof 2.0, the mean of the inverted profiles begins to roughly match that of the original, butthe standard deviation reaches a maximum. Above the break point, the error between themean profile and the original drops exponentially (the maximum error drops from roughly20% for an SNR of 2.0 to 0.02% for an SNR of 20.0) and the standard deviation drops ata similar rate.With the introduction of an aerosol layer, the behaviour is analogous in that the SNRbreakpoint is the same. (see Figure 4.12) In this case, the error in the mean profile is farlarger within the layer than the rest of the profile for SNR values below the threshold.Above the threshold, the same exponential reduction in error is followed as for the purelymolecular profile. As before, the standard deviation reaches a maximum at the thresholdSNR. For a given SNR value, the highest standard deviation value is reached at the topedge of the layer immediately after the sudden change in slope. At this point the standarddeviation is nearly double the value of the layer backscatter, which means that even withan SNR of 2.0 a step change of this magnitude would result in at least some negativeextinction values in roughly 20% of cases12. From this point on, the standard deviations11Most lidars are designed to restrict solar input as much as possible with very limited FOV and narrowbandpass filters in the receiver optics path, but it is never completely eliminated.12This test is designed to push the limits of the inversion algorithm. In real profiles, changes in backscatter634.2. Evaluation of resultsFigure 4.11: These profiles represent 500 Monte-Carlo runs where the input profile repre-sents purely molecular scattering with no aerosol layers and various levels of noise added.In each case, the standard deviation of the noise component is constant throughout theprofile and the noise level is defined in terms of the resulting SNR for the topmost 1 km.In Panel A this SNR value is 1.0, in Panel B it is 2.0, in Panel C it is 5.0, and in Panel D itis 20.0. In each panel the black dashed line is the original profile and the blue solid line isthe mean of the runs. The shaded blue area represents one standard deviation away fromthe mean, and the green area, two standard deviations. Note that due to the variationsin the scale of the results there is not a constant vertical scale between the figures. Thisis especially true of Panel B where the range of variation is much larger than in the otherpanels. 644.2. Evaluation of resultsdrop exponentially with increasing SNR as in the first case, meaning that by the time theSNR value reaches 10, the ratio of the range of profile values to the signal level is reducedto under 1%.Inverting layers with both errors and noiseCombining errors in LIDAR ratio with noise is the next step in the profile analysis. Thisexercise reveals that the two sources of error are essentially orthogonal in that neither oneseems to have a substantial effect on the other. Figures 4.13 and 4.14 show essentiallythe same behaviour as a function of SNR as in Figure 4.12. The differences in the meanprofile arise from the ±50% error in the assigned LIDAR ratio, but the ratio of the errorwithin the layer to the original profile is not altered significantly by the presence of thenoise. Furthermore, the ratio of the standard deviation to the mean of the inverted profilesremains the same for both cases.To summarize, for profiles for which the SNR in the upper 1 km is 2.0 or lower, theerror in the profile is too high to be considered valid, regardless of the accuracy of theLIDAR ratio. For profiles with lower noise values in this region, the error is determinedprimarily by the LIDAR ratio uncertainty for the intervening layers. As compared to otherinversion strategies, the layer-specific inversion used in the LDA, largely constrains theerror introduced by a given amount of LIDAR ratio uncertainty to within the layer itself.The benefits of this are particularly apparent when inverting largely clear profiles with oneor more layers, especially if two layers have different LIDAR ratios.LIDAR ratio sensitivity analysisAs a final test of the performance of the modified Klett inversion used in the LDA, empiricalprofiles were inverted with a range of LIDAR ratio values to demonstrate the sensitivity ofthe inversion to LIDAR ratio errors. Two profiles were selected from the data collected bythe mMPL instrument located in Vancouver, one at 12:00 AM on July 08, 2015 and theother at 11:00 PM on August 08, 2015. These example profiles were selected from datathat were gathered as part of the case study described in full in Section 5.3.In Figures 4.15 and 4.16 the initial backscatter and extinction profiles are generatedusing the initial LIDAR ratios as assigned by the LDA. Once the inversion is performed, itis repeated with the LIDAR ratios for each identified layer ranging from 25% to 200% ofof this magnitude usually take at least 5 altitude steps654.2. Evaluation of resultsFigure 4.12: These profiles represent 500 Monte-Carlo runs where the input profile includesa single layer from 7.5 km to 10.5 km with a backscatter strength of βp = 10−3 km−1sr−1and LIDAR ratio of R = 65 sr with various levels of noise added. In each case, the standarddeviation of the noise component is constant throughout the profile and the noise level isdefined in terms of the resulting SNR for the topmost 1 km. In Panel A this SNR valueis 1.0, in Panel B it is 2.0, in Panel C it is 5.0, and in Panel D it is 20.0. In each panelthe black dashed line is the original profile and the blue solid line is the mean of the 500runs. The shaded blue area represents one standard deviation away from the mean, andthe green area, two standard deviations. Note that due to the variations in the scale of theresults there is not a constant vertical scale between the figures. This is especially true ofPanel B where the range of variation is much larger than in the other panels. 664.2. Evaluation of resultsFigure 4.13: These profiles represent 500 Monte-Carlo runs where the input profile includesa single layer from 7.5 km to 10.5 km with a backscatter strength of βp = 10−3 km−1sr−1and LIDAR ratio of R = 65 sr with various levels of noise added. In this case the inversionassumes a LIDAR ratio that is 50% higher than the actual value.In each case, the standarddeviation of the noise component is constant throughout the profile and the noise level isdefined in terms of the resulting SNR for the topmost 1 km. In Panel A this SNR valueis 1.0, in Panel B it is 2.0, in Panel C it is 5.0, and in Panel D it is 20.0. In each panelthe black dashed line is the original profile and the blue solid line is the mean of the 500runs. The shaded blue area represents one standard deviation away from the mean, andthe green area, two standard deviations. Note that due to the variations in the scale of theresults there is not a constant vertical scale between the figures. This is especially true ofPanel B where the range of variation is much larger than in the other panels.674.2. Evaluation of resultsFigure 4.14: These profiles represent 500 Monte-Carlo runs where the input profile includesa single layer from 7.5 km to 10.5 km with a backscatter strength of βp = 10−3 km−1sr−1and LIDAR ratio of R = 65 sr with various levels of noise added. In this case the inversionassumes a LIDAR ratio that is 50% lower than the actual value.In each case, the standarddeviation of the noise component is constant throughout the profile and the noise level isdefined in terms of the resulting SNR for the topmost 1 km. In Panel A this SNR valueis 1.0, in Panel B it is 2.0, in Panel C it is 5.0, and in Panel D it is 20.0. In each panelthe black dashed line is the original profile and the blue solid line is the mean of the 500runs. The shaded blue area represents one standard deviation away from the mean, andthe green area, two standard deviations. Note that due to the variations in the scale of theresults there is not a constant vertical scale between the figures. This is especially true ofPanel B where the range of variation is much larger than in the other panels.684.2. Evaluation of resultsthe original. The results of these variations show that the inversions, while they do dependstrongly on LIDAR ratio, remain stable despite large variations in this quantity.As one would expect, in both cases the extinction varied in direct proportion to theLIDAR ratio and the backscatter varied inversely. This relationship holds true despite thefact that the levels of backscatter vary by nearly two orders of magnitude between the twoprofiles. Although it is not directly related to the sensitivity analysis, it is also interestingto note that the region from 1.6 km to 2.2 km in Figure 4.15 appears to be part of anaerosol layer that is missed by the wavelet analysis. This may be an indication that theparameters of the LDA could be improved to better identify the top edges of layers suchas these where the transition zone has a more gradual slope.The analysis of LIDAR ratios for observed aerosol layers found in Section 2.3 shows thatLIDAR ratios can frequently vary by up to 100% from the median values depending on thespecific aerosol type and the circumstances of the observation. This is especially true ofcirrus clouds. The LIDAR ratio ranges for the profiles in the sensitivity analysis have beenselected to conservatively represent the largest of these ranges. For these ranges of variation,the sensitivity analysis shows that despite variations in extinction and backscatter withinindividual layers on the order of the variation in LIDAR ratio, the modified Klett inversionutilized by the LDA remains stable for typical LIDAR ratio range.694.2. Evaluation of resultsFigure 4.15: These profiles were generated by varying the LIDAR ratio for a profile collectedin Vancouver at 12:00 AM on July 08, 2015. Panel A shows backscatter, Panel B showsextinction. The black line represents that LIDAR ratio as assigned by the LDA. The restof the lines show the results of varying the LIDAR ratio from 25% to 200% of the original.704.2. Evaluation of resultsFigure 4.16: These profiles were generated by varying the LIDAR ratio for a profile collectedin Vancouver at 11:00 PM on August 13, 2015. Panel A shows backscatter, Panel B showsextinction. The black line represents that LIDAR ratio as assigned by the LDA. The restof the lines show the results of varying the LIDAR ratio from 25% to 200% of the original.714.2. Evaluation of results4.2.2 Qualitative tests using empirical dataThese tests are meant to show the LDA in action in a range of circumstances and to revealsome of the strengths and weaknesses of the algorithm in the context of real empirical data.In most of these cases there is no definitive evidence from other sources to support or refutethe layer designations, but the purpose of these cases is qualitative in nature. This meansit is not necessarily meant to prove the validity of the results but rather to show that itsperformance is consistent in a variety of conditions and that the results are plausible. Theother purpose is to demonstrate some ongoing unresolved issues with the performance ofthe LDA.Mountain Location: Whistler, BCThe first demonstration case was collected in Whistler, BC on March 27, 2014 ( see Figure4.17). The mMPL was operated in Whistler, B.C. from March 20 through June 07, 2013as part of a joint study between the UBC Chemistry and Geography Departments toinvestigate soot particles in the free troposphere. On the day in question layers are observedat altitudes ranging from 1 km to 4 km throughout the day, with a weaker layer above from6 km to 8 km from midnight to 6:00AM. To an experienced observer, the primary layerscan be identified as clouds by the combination of their high normalized relative backscattervalues and the high contrast in the depolarization ratios within the layers. The nature ofthe secondary layer, on the other hand, is not as easily determined.724.2.EvaluationofresultsFigure 4.17: Mini-MPL data from Whistler-March 27, 2014 Panel A shows normalized relative backscatterreturns. Panel B shows linear volume depolarization ratios. Panel C shows LDA results. In this instance the mMPLas operating in a mountain valley in cloudy conditions. The results include two layers: a primary layer of stronglyscattering clouds from 2 km to 4 km that includes multiple interleaving sub-layers including water, mixed ice andwater, and ice. The secondary layer occurs from 6 km to 8 km with much lower backscatter for which the layerdesignation is less conclusive.734.2. Evaluation of resultsThe LDA identifies the primary layer as a combination of interleaving cloud types withall three categories (water, mixed, and ice) represented. The layer edges match well withthe edges observed in the normalized relative backscatter and depolarization ratio plots.This example highlights the novel capability of the LDA to discriminate between adjacentlayers through changes in depolarization ratio even when no edges in normalized relativebackscatter exist, allowing an increase in detail within the cloud layers that would nototherwise be achievable.The secondary layer is identified by the LDA as a combination of mixed cloud anddusty aerosols. While it is possible that this designation is accurate, there is no evidenceon this day of dust incursion from long-range transport into the region and no immediatesources of dust within the valley - especially not that would send dust to this altitude. Itis therefore considered to be an erroneous designation and an example of one of the keyflaws in the LDA algorithm involving the difficulty in distinguishing between optically thinice clouds and dusty aerosols (more on this topic in Section 4.2.3). Furthermore, it is clearthat the upper layer is obscured to a varying degree by the extinction caused by the layerbelow, leading perhaps to an increased error in the layer backscatter coefficient. While thisis exactly the kind of effect that is meant to be at least partially corrected in the profileinversion, the LIDAR ratio designation is not currently iterative so these corrections are nottaken into account. The implementation of an iterative process including this correction isamong the next steps planned for the LDA as described in Section 6.2.6.The LDA seems to have performed well in identifying the top edge of the PBL. In thisinstance, the mean height of the PBL was 567 m with a standard deviation of 210 m anda maximum value of 990 m. Without direct measurements of the altitude of the inversionmarking the top of the PBL, it is difficult to determine whether this is accurate, but it ispossible that the persistent gap between the top of the PBL and the bottom of the primarycloud layer represents a small error in the height of the PBL.Backscatter and extinction coefficients were generated by inverting the normalized rela-tive backscatter profiles using the layer identifications shown in Figure 4.17 and the resultsare shown in Figure 4.18. Decreased SNR during daylight hours resulted in a notable in-crease in variance in the inversion results, which is consistent with the results in Section4.2.1 and indicates that under these conditions, the results of the inversion must be takenwith an increased scepticism. The results during the night have much lower noise and aretherefore more stable and more likely to be accurate.744.2.EvaluationofresultsFigure 4.18: Backscatter and extinction coefficients from Whistler-March 27, 2014 Panel A shows backscat-ter coefficients. Panel B shows extinction coefficients. Backscatter coefficients are calculated using the modified Klettalgorithm, extinction coefficients are calculated from the backscatter using the assumed LIDAR ratios for each layer.754.2. Evaluation of resultsThis case is an interesting one because the primary layer provides a straightforwardcase for the LDA with high SNR and strong demarcations in depolarization, but the PBLand the secondary layer are examples of borderline cases that demonstrate some of thepotential shortcomings of the LDA and highlight the need for both experience and cautionin employing this algorithm. Also demonstrated in this case is the notable impact elevatednoise levels have on the success of the inversion methodology, which is discussed in moredetail in Section 4.2.3.Coastal location: Ucluelet, BCA coastal location such as Ucluelet presents a number of challenges for layer identification.One reason for this is that the algorithm does not currently include a designation for marineaerosols and this is one location where they could be relevant, especially within the PBLwhen the wind is onshore. This is also a location where difficulties are likely to arise in theidentification of the PBL due to the influence of the ocean on boundary layer formationas well as the possibility of insufficient aerosol concentration near the surface to produce adetectable edge.Two cases were selected from the Ucluelet data. The first was May 03, 2014 (see Figure4.19). This instance included a somewhat unusual cloud formation where a deck of waterclouds was observed from 3 km to 4 km while at the same time a layer of highly depolarizingparticles, appears higher up and over the course of several hours, the separation betweenthem reduces until they begin to overlap at approximately 5AM. After this time, theclouds at 3 km are dominated by higher depolarization layers. The LDA identifies thehigh depolarization layers as a combination of dusty aerosols and ice/mixed clouds. It ispossible that this is a case of erroneous layer designation, but if it is accurate it is potentiallyinteresting result due to the fact that it would be unusual to see ice clouds at this altitudeat this coastal location in May. Temperature soundings at nearby Quillaute, WA (roughly126 km away, south-southeast) show temperatures at this altitude of only −15 ◦C, near thewarmer end of the range at which ice clouds typically form, which could be indicative ofthe presence of dust or some other similar aerosol resulting in an increase of heterogeneousfreezing. At the lower altitudes layers of water clouds appear intermittently near the topof a persistent mixing layer of low depolarization aerosols and from 4AM to 6AM severallayers of very optically thin low depolarization aerosols are observed ranging up to 2 km764.2.EvaluationofresultsFigure 4.19: Mini-MPL data from Ucluelet-May 03, 2014 Panel A shows normalized relative backscatterreturns. Panel B shows depolarization ratios. Panel C shows the results of the LDA. This represents a complex casewith a number of different challenges for the LDA. The PBL is capped in some places by a low-lying layer of whatappears to be water clouds and in others by a mixing layer of aerosols directly above the PBL. An optically thicklayer of clouds with another layer above it, and a layer for which the discrimination between cirrus cloud and elevateddusty aerosol together add additional challenges.774.2. Evaluation of resultsThe LDA identifies the aerosols in the lower free troposphere as ’smoke/urban’. It’simportant to note that the optical properties of these layers are consistent with those ofmarine aerosols, and if a “coastal” setting were implemented that they could potentially beidentified as such. This case also has some things in common with the Whistler case in thatthey both have optically thinner ice clouds above thicker cloud layers, and in this instancethe LDA has the same difficulty in distinguishing the upper layer from dusty aerosols.The backscatter and extinction coefficients generated for this case exhibit much moreconsistency and lower variance that in the Whistler case primarily due to the fact that themajority of the scene was observed before sunrise (see Figure 4.20). there is no way todirectly verify the specific values generated for extinction and backscatter but the valuesare within reasonable limits for the observed layers. The reduced resolution in the layeranalysis is apparent in the extinction image where the values are amplified by the highLIDAR ratio.784.2.EvaluationofresultsFigure 4.20: Backscatter and extinction from Ucluelet-May 03, 2014 Panel A shows backscatter coefficients.Panel B shows extinction coefficients. These were calculated based on the assigned LIDAR ratio profiles from thelayer analysis.794.2. Evaluation of resultsThe second case collected from Ucluelet is very different from the first. In this instance,there are no clouds present, just multiple optically thin aerosol layers distributed from 2 kmto 9 km (see Figure 4.21 ). The purpose of this example is to test the capability of theLDA to detect these layers and to provide plausible values for the associated extinctionand backscatter coefficients.804.2.EvaluationofresultsFigure 4.21: Mini-MPL data from Ucluelet-May 03, 2014 Panel A shows normalized relative backscatterreturns. Panel B shows linear volume depolarization ratios. Panel C shows the LDA results. The challenge in thiscase is to correctly separate the optically thin layers from 2 km to 9 km from the background noise. This was achievedreasonably effectively prior to approximately 1:00AM when an optically thick layer of water cloud developed at thetop of the PBL resulting in too much signal loss for the algorithm to reliably identify layer edges.814.2. Evaluation of resultsThese layers can be divided into two rough categories. One group, with layers occurringprimarily between 2.5 km to 7.5 km has higher normalized relative backscatter values andlower depolarization ratios. The depolarization ratios of these layers falls neatly into the“smoke/urban” aerosol category. Another group, with layers that occur intermittentlyfrom 5.0 km to 9.5 km, shows comparatively lower normalized relative backscatter valuesfor which the depolarization is too high to reasonably considered marine, urban, or mosttypes of smoke. The layer designation for these regions is “polluted dust” indicating somemixture of dust - or some other highly depolarizing aerosol) into a series of layers otherwisedominated by more spherical particles. Although these layers exhibit depolarization ratiosand NRB levels that are also consistent with very optically thin cirrus clouds, this isconsidered to be unlikely in this case due to the altitudes of the layers and the coastallocation.This is an intriguing result given that NAAPS results show trans-pacific transport ofboth dust and sulphates on this day, but not at this location (see Figure 4.22. It is feasiblegiven the very low normalized relative backscatter levels that these layers represent tracelevels of dust that would otherwise go undetected. Of course further investigation wouldbe required to verify the presence of dust in these layers. It should be pointed out thatthe aforementioned normalized relative backscatter comparisons are relative since all of thelayers in this scene are optically thin and near the lower end of the SNR range for whichthe LDA can reliably operate.In the second half of the image (after approximately 1:00 AM) an optically thick cloudlayer appears just above the PBL resulting in drastic loss of signal above. After this pointin the data set, the LDA is no longer capable of isolating and identifying the aerosol layers.This is to be expected given that they were near the noise threshold level even before theintroduction of the cloud layer. The PBL algorithm was presented with a simple case herewith strong backscatter within the PBL and a layer of cloud forming at what appears tobe the top of the PBL.824.2. Evaluation of resultsFigure 4.22: Panel A shows NAAPS model results for AOD over North America at midnighton May 06, 2014. Panel B shows a time-height section of dust concentrations over CheekaPeak Atmospheric Research Station, roughly 90 km SSE from Ucluelet. The optical depthresults in Panel A don’t show any measurable quantities of dust over Ucluelet, but thelowest detectable level for the model is 0.1 and these layers are likely well below thatthreshold. The time-height sections in Panel B are more consistent with the LDA analysis,showing elevated dust layers in the vicinity for several days. NAAPS figures obtained from[NRL]834.2.EvaluationofresultsFigure 4.23: Backscatter and extinction coefficients from Ucluelet-May 06, 2014 Panel A shows backscattercoefficients. Panel B shows extinction coefficients. In this case the low resolution of the layer-specific LIDAR ratiosbecomes very clear. Despite the appearance, the application of these layer-by-layer optical properties serves to improvethe general performance of the inversion844.2. Evaluation of resultsThe results of the inversion for this example are shown in Figure 4.23. Despite therelatively low resolution in the assigned layer-specific LIDAR ratios, the extinction valuesare in the expected range for the type of aerosols and the normalized relative backscatterstrength.These cases have been selected to demonstrate some of the strengths and weaknessesof the current version of the LDA. The algorithm is best at identifying layers with highSNR but is also capable of finding and correctly identifying very optically thin layers. Thelayer-specific LIDAR ratios generated by the LDA are likely to reduce the error in inversionof profiles to obtain backscatter and extinction coefficients. At the same time, there areseveral unresolved issues that will be discussed in more detail in the next section.4.2.3 Ongoing issuesIn this section some of the most important limitations of the application of the LDAare discussed, and examples are given where the algorithm fell short of expectations byproviding results that were misleading or inconsistent with information collected fromexternal sources. Where possible, suggestions are made for methods to be employed indevelopment of future iterations of the algorithm that could mitigate or eliminate theissues.Known adverse effects of noiseThis is perhaps the most pervasive issue with the calculation of extinction and backscattercoefficients. As mentioned in Section 4.2.1 even standard levels of shot noise in the originalmeasured LIDAR signal can result in significant error in the inverted profile on the order of100%. This level of error occurs regardless of the level of uncertainty in the LIDAR ratio,and idealized profile results reveal that it can occur even when there is no error in theLIDAR ratio of any aerosol layers in the profile. This error in backscatter and extinctioncoefficient profiles reflects the cumulative effect of noise on the Klett inversion when thestandard deviation of the signal approaches or surpasses the mean signal level for the upperreaches of the profile. This effect is particularly pronounced under clear conditions andduring daylight hours when background radiation increases. (see Figure 4.18)The current strategy for mitigating these effects is to use a smoothing window (thewindow size is typically 5x5) to reduce the noise levels, and to include an optional featurethat masks out regions above optically thick clouds as well as other regions where the SNR854.2. Evaluation of resultsdrops dramatically, but the results are not consistent. A more comprehensive approachwould involve an adaptive averaging algorithm that increases the size of the averagingwindow based on the SNR of the region, combined with a hybrid or iterative methodologyfor dealing with regions of high visibility where the standard Klett algorithm is less thanoptimal.Cloud-aerosol thresholdsAs described previously in Section 4.1.3, there are currently three threshold tests beingused to discriminate between clouds and aerosols in the LDA: altitude, backscatter, anddepolarization. These thresholds have been chosen based on data collected over roughly ayear of operation of the mMPL system in various locations and they are sufficient for mostcases, but there remains the possibility of mis-categorization, especially for particularlythin ice clouds (for examples of this, see Figures 4.17 and 4.19) and for the upper edges ofsome optically thick clouds layers(as in Figure 5.6)No threshold value has been identified to date that can reliably differentiate betweenoptically thin ice clouds and dusty aerosols in all cases based solely on depolarization ratioand backscatter strength. As discussed in Section 4.1.3, other LIDAR systems with largerfields of view may be able to use multiple scattering effects to improve this distinction, butwith the mMPL and other micro-pulse systems, this option is not available. It is a topic offuture investigation to collect more data and try to look for additional patterns that canbe exploited to mitigated or overcome this issue.Determining the height of the PBLOver the years there have been many attempts to employ LIDAR profiles to discern thetop of the PBL and the associated entrainment zone thickness, including methods basedon the slope of the LIDAR profile [e.g.: 40], ones based on threshold values [e.g.: 6],ones utilizing a multi-variate full-profile curve fitting technique [e.g.: 103], and ones thatcombine elements of multiple techniques [e.g.: 93]. With the exception of water vapourLIDAR systems, the process of determining the top of the PBL using LIDAR, regardlessof the method, hinges upon the assumption that the well-mixed boundary layer carieswith it a higher load of aerosols than the air directly above it. As it turns out, whileeach of the above methods has its strengths under certain conditions, there are conditionsunder which each of them fail and the one employed by the LDA is no different. There864.2. Evaluation of resultsare multiple conditions under which this method can encounter difficulties including thepresence of an aerosol rich layer just above the PBL, a very clear day when the PBL isessentially free of aerosols, and (perhaps most concerning of all) nocturnal boundary layerswhere stable conditions beneath the capping inversion can cause locally emitted aerosolsto stratify well beneath the capping inversion resulting in confusing or erroneous results.Tests comparing the results of LIDAR-derived PBL heights to those taken from radiosondeprofiles have shown a marked inability of the LIDAR results to track diurnal variations aswell as a consistent tendency to over-estimate PBL heights [93]. Even water-vapour LIDARsystems, directly measuring a quantity much more closely related to PBL hight have showna large degree of variance as compared to radiosonde profiles [42] In addition to this, in thecase of insufficient aerosol content, the LDA wavelet analysis becomes susceptible to edgeeffects. In the future it is worth looking into other well-known methods of PBL detectionto see if any of them outperform the wavelet method.Turbid atmospheresThe first step in the layer identification process is to use a variance analysis to identifyregions of molecular scattering (for more information see Section 4.1.2) This method isextremely well-suited to finding layers of even slightly increased scattering against a back-ground of clear air, but in the unusual case of highly turbid atmospheres where there isa well-mixed haze of aerosols throughout the observable range with no well-defined edges,this step fails. An example of this was observed when the LIDAR was deployed to observefresh dust particles from the Taklamakan Desert in Aksu, Xinxiang Provence, P.R. China.In this case, all the supporting evidence points to the presence of dust in at least the lower4 km of the atmosphere, but the LDA fails to properly identify it.In late April, 2013 a mMPL LIDAR system was operated for six days at the AksuNational field research facility on the edge of the Taklamakan desert (40.62◦N, 80.85◦E)where dust from this region was observed in its unadulterated state shortly after emission(see Figure 3.1). NAAPS models of the region reveal observable quantities of dust fromthe Taklamakan region were present at the location of the observations. [21]874.2.EvaluationofresultsFigure 4.24: mMPL data from Aksu, April 25-27, 2013 Panel A shows normalized relative backscatter re-turns, Panel B shows linear volume depolarization ratios. The data reveal a persistent optically thin haze of highlydepolarizing aerosols, most likely dust, in the lower troposphere.884.2. Evaluation of resultsIn Figure 4.24, a consistent aerosol layer can be seen over Aksu for altitudes from thesurface up to at least 4 km from April 25-27, 2013. The depolarization ratio for this layeris consistently > 0.30 indicating highly non-spherical particles consistent with fresh dustfrom the Tarim Basin [63]. The dust within the boundary layer is likely to be largely fromlocal sources near Aksu. Dust storm and aerosol transport models reveal that the dust inthe free troposphere is likely to have originated further west near Kashgar. On April 26,an influx of another layer of more spherical particles can be seen entering the dust layerfrom above and mixing down into it over a course of several hours. On subsequent days,this layer is seen mixing into the dust, temporarily lowering the volume depolarizationratios. The nature and origin of this intervening layer is currently unknown, but it servesas evidence that the observed depolarization ratios are not artificially elevated as a resultof calibration errors or some other technical flaw in the observations .The statistical distribution of depolarization ratios for the lower 4 km over Aksu isrepresented as a box plot in Figure 4.25. This plot reveals a fairly consistent and narrowlydistributed range of values for the six days of observation. The mean depolarization ratiosvary from a minimum of 0.266 on August 26 to a maximum of 0.344 on August 28. Thisresult is on the high end, but well within the observed range for pure dust from this region(see Table 2.1). The effect of the introduction of the aforementioned layer of more sphericalparticles on April 26 is clear as the mean depolarization ratio dropped and the standarddeviation increased on this day.Based on these observations alone, and given the location of the LIDAR, a strong casecan be made for identifying the observed aerosols as dust, but when the LDA is employed,the result is very different. The well-mixed nature of the dust throughout the visible rangeof the LIDAR stymies the variance analysis and the LDA fails to distinguish these regionsfrom clear air. The only layers that are distinguishable from the background, and thereforethe only ones that appear on the layer type mask are a few clouds and the influx of morespherical aerosols on April 26. (see Figure 4.26)894.2. Evaluation of resultsFigure 4.25: Box plot showing range of depolarization values for the lower 4 km of altitudeover Aksu for April 22-27, 2013. In this plot the top and bottom whiskers represent 1.5times the upper and lower interquartile ranges, respectively. Outlying values have beenleft out because of the large number of measurements.904.2.EvaluationofresultsFigure 4.26: Layer type mask for Aksu, April 25-26, 2013 In this case ,the layer analysis fails to identify thedust layers in the lower 4 km, incorrectly labelling the entire region as “Clear Air”914.3. Summary and conclusionsThis is an unusual case for the areas where the LIDAR is typically deployed, but for aplace such as Aksu, in foggy conditions, or possibly in a city with high levels of pollution,this could be a serious impediment to the proper operation of the LDA.4.3 Summary and conclusionsIn this chapter, the building blocks of the LDA are described in detail. The first section isdevoted to describing the algorithm itself, step by step. This begins with the pre-processingused to calculate the fundamental inputs to the algorithm, including profiles of normalizedrelative backscatter (or in the case of CORALNet backscatter ratio) and depolarizationratio. Following this is a description of the variance and wavelet analysis used to find layeredges and to discriminate between layers and the molecular background in a profile. Theresults of these two steps are both used as inputs to the final step where the processes andcriteria of layer classification are described. Finally the assigned LIDAR ratio values foundas a result of the layer identification are used in range-dependant inversions to calculatebackscatter and extinction coefficient profiles.The second section is devoted to an in-depth evaluation of the algorithm. This includesa quantitative analysis of the sensitivity to errors in LIDAR ratio as well as noise incalculating backscatter and extinction profiles, as well as a qualitative analysis of resultsgenerated by applying the LDA to a number of empirical data sets under varying conditions.The section then concludes with a brief description of some known issues of sub-optimalperformance including issues arising from less than ideal SNR levels, difficulties in dealingwith highly turbid atmospheres, and the uncertainties involved in attempting to determinethe depth of the PBL using the LDA.92Chapter 5Research applications for layerdiscriminationTo date, the LDA described in Chapter 4 has been successfully employed in three fieldstudies. Each of these cases deals in some way with aerosols - specifically dust and smoke- observed in Vancouver, B.C. after short, medium or long-range transport. Thanks toadvances in aerosol monitoring and modelling transport events such as these are known tobe a common phenomenon throughout the world and they can have far-reaching impactson meteorology, air quality, and climate as well as multiple global bio-geo-chemical cycles.However, they are still largely poorly understood due in part to their highly anisotropic andunpredictable nature, but also because most of the transport occurs in the free tropospherewhere monitoring and identifying aerosols remains difficult and data collection is muchmore sparse than near the surface. This forces a reliance on models and, where available,mountain top, aircraft, or satellite monitoring. This is precisely where ground-based lidar,which can provide continuous high-resolution profiles of aerosols at a range of altitudes,can make an important contribution to understanding these aerosol transport phenomena.The range of the transport in the following case studies varies from local to regional tointercontinental. The first case (Section 5.1) deals with an unusually strong and persistentevent that occurred in 2010 involving long-range transport of dust from China to NorthAmerica. This involved CORALNet lidars located in Vancouver, BC and Egbert, ON. Ini-tial results of this study were published in [23] prior to the development of the LDA whereinthe nature of the elevated layers was inferred from inspection of the LIDARdata in com-bination with a range of supporting information including NAAPS models and HYSPLITtrajectory analysis and AEROsol RObotic NETwork (AERONET) data. The second case(Section 5.2) deals with another long-range transport event from Asia to North America,but in this case it is wildfire smoke from Siberia. As in the first case, initial results werepublished without the benefit of LDA results [22]. In both of these cases, an abbreviated935.1. Case Study #1: A Pervasive and persistent dust episode in North America - 2010summary of the relevant results from these publications is presented alongside the resultsof the LDA to show that the results agree and to demonstrate how the LDA facilitatesand (in some cases) improves the analysis of LIDARdata. Then in each case, an additionalanalysis of layer properties is performed using the LDA layer mask as a filter - somethingthat was not possible in the original published papers.The third study, described in Section 5.3 deals with an extended period in July andAugust of 2015 where smoke from forest fires was present almost continuously in the LowerFaser Valley airshed for over six weeks. This is a case where careful inspection and collectionof supporting evidence for each day of the event would be onerous, so the use of the LDAis taken a step further by utilizing it as the primary tool for layer identification with aminimum of supporting evidence and employing it to automate the collection and analysisof data for these days. This demonstrates that by using the LDA it becomes possibleto perform long-term collection and statistical analysis of aerosol layers whereas in theprevious studies it was necessary to focus on a few days at a time.Taken as a whole, these studies demonstrate the efficacy of the LDA and its ability toincrease the value of ground-based LIDARas a tool for aerosol research projects. The casesexamined in this chapter deal with an important but relatively narrow field of research(transport of smoke and dust) compared to the range of possible applications, but theydo provide a thorough vetting of the algorithm including operation with complex layerstructures, borderline identification cases, and a large range of layer AODs. It is not hardto imagine the LDA being useful to a wide range of research efforts involving lidars.5.1 Case Study #1: A Pervasive and persistent dustepisode in North America - 20105.1.1 IntroductionSpringtime trans-Pacific transport of crustal dust from the deserts of central Asia to NorthAmerica has been well documented over the past decade. Noteworthy studies includecomprehensive analyses of the significant 1998 [49] and 2001 events [111], summaries ofmultiple events [32, 56, 69], model-based climatologies of trans-pacific transport processes[e.g.: 37, 104], and assessments of impacts of Asian dust on North American air quality[16, 30, 41, 51, 120]. Thanks to years of continuous data collection by multiple, overlappingglobal and regional networks of aerosol monitoring stations, what was once considered to945.1. Case Study #1: A Pervasive and persistent dust episode in North America - 2010be a relatively rare occurrence is now known to be a regular annual event.The dust transport events of Spring 2010 were an extraordinary example of this phe-nomenon. Details of the temporal sequence of dust mobilization in the source regions ofChina, and the meteorological pathways and processes of trans-Pacific transport of theAsian dust during April 2010 are described by Uno et al. [113]. In summary, at least eightmajor dust mobilization events were shown to have occurred in the Taklamakan and Gobidesert regions between March 1 and April 30, 2010. Of these, the third, fifth, and seventhApril events were the largest and all of the events were observed to result in transport ofdust to North America. Dust was observed to travel in multiple layers with elevations from2 km to 10 km and was observed to split into two distinct pathways. Of particular note inthe context of this study is dust following the northern pathway across Canada. A slowmoving anticyclone over the central continent during the period 15-22 April resulted in sig-nificant subsidence and stagnation, temporarily trapping the dust over the north-westernportion of the continent. These observations are consistent with observations from Mt.Bachelor Observatory in Oregon, showing increased aerosol loading on April 10 and April19 corresponding to layers identified as dust or polluted dust by concurrent CALIPSOoverpasses [33]. While Uno et al. [113] documented the transport of dust from the largestevents of April, plumes have been observed arriving on the west coast of North Americain late March as well. These are clearly the result of smaller, but still substantial Marchdust storms documented by Uno et al. [113] and Li et al. [62] (MODIS True Colour im-agery of one such storm is shown in Figure 5.1). Prior to this investigation the presenceof Asian dust in North America during March, 2010 had not been well documented, withthe exception of Fischer et al. [33].This study extends and complements the satellite and global model based analyses ofUno et al. [113], Li et al. [62] and Fischer et al. [33] and focuses specifically on ground-based LIDARobservations of the aerosol layers as they passed over the continent. LI-DARobservations were collected from CORALNet stations located in Vancouver, BC andin Egbert, ON (see Figure 3.1). Layer analyses performed on these data revealed multipleoverlapping elevated layers of dust and polluted dust in both locations. Collected datawere analysed in combination with results from the NAAPS and HYSPLIT models to pro-vide a more complete picture of the scope and distribution of this event throughout NorthAmerica, including documentation of additional dust transport events reaching the westcoast as early as 17 March.955.1. Case Study #1: A Pervasive and persistent dust episode in North America - 2010Figure 5.1: True colour image from 12 March, 2010 showing dust layers sweeping outover the East China Sea. Image from MODerate resolution Imaging Spectro-radiometer(MODIS) instrument on NASA’s AQUA satellite is thanks to Jeff Schmaltz and the MODISLand Rapid Response Team, NASA GSFC [96]965.1. Case Study #1: A Pervasive and persistent dust episode in North America - 20105.1.2 NAAPS and HYSPLIT Model ResultsNAAPS models from the spring of 2010 predicted aerosol optical thickness in excess of0.8 for dust in the air above central Asia, including the Gobi and Taklamakan regions, formultiple sustained periods during March and April. Concentrations this high occur almostexclusively over dust producing regions under conditions of high winds. This is a strongindication that these regions were the source of the dust layers that spread out acrossthe Pacific throughout March and April, in agreement with Uno et al. [113] and Li et al.[62]. Furthermore, these models predicted that the total integrated AOD from dust in thetroposphere over CORALNet installations in Vancouver and Egbert was high enough tobe detectable. In order to confirm the connection between this dust and layers observedin LIDARdata, HYSPLIT back trajectories were calculated for the regions and times inquestion to show the progress of the air parcels containing the dust as they approached theLIDARinstallations.It is often the case that for trajectories of more than a couple of days in duration,the majority of parcels come from a generally westward direction for locations such asVancouver and Egbert, in the mid-latitude westerlies, and this is indeed the case for mostof the days reported here. More important to note is how tightly the trajectories aregrouped and how many of them pass directly over the regions where the aforementioneddust storms were occurring (see Figure 5.2).NAAPS predictions show dust first reaching the west coast of North America on 15March. Throughout the periods in March during which dust was detected by the lidar, theNAAPS models consistently showed dust AOD levels of 0.1-0.2 over the west coast of NorthAmerica from southern British Columbia down to northern California. A representativeexample of this from 19 March is seen in Figure 5.2(a).Based on LIDARobservations, a starting altitude range of 3 km to 9 km was selectedfor HYSPLIT back trajectories for this event. Tightly grouped back trajectories for thisaltitude range and date (Figure 5.2(a)) are consistent with an Asian dust source and confirmthat this event occurred one week prior to the earliest event identified in Fischer et al. [33]and almost a full month earlier than the events described in Uno et al. [113]. For this day,air parcels were tracked backwards for a duration of 180 hours.As the dust storms in Asia continued into April, the NAAPS results predicted thecontinued presence of high concentrations of dust in the aforementioned desert regions. Atthe same time, layers continued to arrive over North America. For the first 10 days of975.1. Case Study #1: A Pervasive and persistent dust episode in North America - 2010April the majority of the dust followed a more southern path across the Pacific than wasobserved during the events of March. The highest concentrations for early April were nearthe Canada-USA border, engulfing almost its entire length, with lower concentration layersat times extending as far south as the Gulf of Mexico and north to Baffin Bay. This can beseen in Figure 5.2(b), which shows the back-trajectory frequency plot for 11 April basedon 210 hours back trajectory duration and a starting altitude of 3 km to 6 km (consistentwith Vancouver LIDARobservations).After 13 April, the NAAPS predictions showed the highest concentrations of dust wellnorth of the Canada-USA border and over the next two days, the layers of dust spreadto engulf nearly all of Canada. This large dust layer persisted for another 10 days beforefinally starting to dissipate on 24 April. On some days, lower concentration regions alsoextended to cover substantial portions of the US as well. A representative example ofthis distribution can be seen in Figure 5.2(c), taken on 20 April. This layer of dust wasunusually large due to the extraordinary size and duration of dust storms in the desertsof central Asia, and it persisted over North America as a result of a large high pressuresystem that dominated over Canada for several weeks.Back trajectories from Egbert, Ontario showed greater spread in the medium rangecompared to Vancouver back trajectories, with air parcels covering large portions of westernNorth America and the Pacific. But, as exemplified in the trajectory frequency plot for20 April (Figure 5.2(c)), a significant portion of trajectories converge in the long rangeto pass over the deserts of Asia. Based on LIDARobservations from Egbert, the startingaltitudes for these back trajectories was 3 km to 9 km. After considering the HYSPLITback trajectories for this area, we conclude that the observed dust originated from theAsian dust storms in early April.985.1. Case Study #1: A Pervasive and persistent dust episode in North America - 2010Figure 5.2: NAAPS global dust model results (left) alongside frequency plots of HYSPLITback trajectories (right) for peak days during the three primary dust transport eventsobserved by the CORALNet lidars. Dust models represent 24-hour average values for thedays in question. HYSPLIT plots include back trajectories that were calculated in 200 mincrements throughout the altitudes of interest and repeated every 6 hours for the day inquestion. HYSPLIT plot labels include the location from which trajectories were generatedand the Back Trajectory Duration (BTD), or the number of hours air parcels were trackedfor that day. In each figure, the red ellipses indicate the source region in Asia and thered stars indicate the location where LIDARmeasurements were collected. NAAPS resultswere provided by Naval research Laboratory (NRL) [NRL]995.1. Case Study #1: A Pervasive and persistent dust episode in North America - 20105.1.3 CORALNet observations and LDA resultsMultiple aerosol layers were clearly visible throughout the troposphere in CORALNet LI-DARdata in March and April.March 2010Specific observations of elevated aerosol layers were made at UBC from 17-20 March andagain on 23-25 March. These events pre-dated the observations of long-range transportmade by previous authors [e.g.: 33, 63, 113], but coincided with mid-March wind stormsin that region [62].The first event occurred between 17-20 March. As shown in Figure 5.3, false colourimages of backscatter ratio from the 1064 nm channel show layers entering the LIDARfield-of-view at an altitude of 6 km to 9 km at noon on 17 March. At this point the mean volumedepolarization ratio of the elevated layers was around 0.25. Combined with backscatterratios well below what would be expected for clouds, this is a strong indication of dust. Thedepolarization ratio differentiates this layer clearly from the well-mixed layer covering thebottom 2.5 km of the troposphere, which maintains a consistent depolarization ratio of <0.1. Later in the day, the elevated layers appear at lower altitudes, which could be the resultof subsidence or possibly the entrainment of lower altitude dust over time. Interestingly, forthe layers observed at lower altitudes and later in the day the depolarization was steadilyreduced to match that of the lower layers. The results of the LDA are used to investigatethis trend is further in Section 5.3: LIDARobservations for 17-20 March, 2010 showing 1064 nm backscatter ratios (top) and masked 532 nmdepolarization ratios (centre). The red line divides the backscatter ratio plot into Region A, which is dominatedby aerosols, and Region B, which is mostly clouds. Arrows marked C point to areas that are typified by highdepolarization ratios, likely to contain high concentrations of dust. Arrows marked D point to areas that showreduction in depolarization as compared to the aerosols above probably due to mixing with fine-mode aerosols. Thebottom panel shows the results of the LDA for this scene. The regions of aerosol and cloud that were demarcated ingeneral terms as Region A,B,C, and D are clearly identified as aerosol and cloud layers in the layer mask plot.1015.1. Case Study #1: A Pervasive and persistent dust episode in North America - 2010The LDA results for this event (shown in the bottom panel of Figure 5.3) are largelyconsistent with the analysis based on the backscatter and depolarization data. The highestaltitude layers are identified as dust, then as the depolarization ratios drop for lower altitudelayers, the designation shifts to “polluted dust” and finally to the “smoke/urban” aerosoldesignation for the layers just above the PBL. The LDA shows the cloud layers that appearfrom 5 km to 11 km starting at around 12:00 on 18 March to be a complex mixture of ice,mixed ice and water, and dust-rich aerosols. Although it would require direct measurementof temperature, pressure, and humidity profiles within the clouds to verify this result, itis plausible given the altitude of the clouds and the range of depolarization ratios. Thereappears to be a certain amount of aerosol mixture in these cloud layers but it is unclearwhether these returns are actually the result of aerosols or if they are simply very opticallythin sections of cloud. Given that the CORALNet system has a larger FOV than themMPL system, it is also possible that the results have been affected by some degree ofmultiple scattering.Based on the layer designations, the modified iterative Klett algorithm described inSection 4.1.4 was used to invert the LIDARprofiles to generate the backscatter and extinc-tion coefficient plots shown in Figure 5.4. While these results are difficult to verify, thebackscatter and extinction values are well within the range expected for the aerosol andcloud types present.1025.1.CaseStudy#1:APervasiveandpersistentdustepisodeinNorthAmerica-2010Figure 5.4: LDA inversion results for March 17-20 dust event. Panel A shows backscatter coefficients and Panel Bshows extinction coefficients. These results are calculated based on the layer mask in Figure 5.3 using the modifiedKlett inversion for varying LIDARratios.1035.1. Case Study #1: A Pervasive and persistent dust episode in North America - 2010The dust layer observed in the lower troposphere in Vancouver on 24-25 March, shownin Figure 5.5 is anomalous among these examples in that it is not seen to include a seriesof multiple optically thin layers in the middle to upper altitudes of the free tropospherewith gradually increased mixing in the lower altitudes. In this instance, a single contiguousand relatively short-lived layer of highly depolarizing particles is observed near the surfacebelow a consistent cloud deck and is clearly distinguishable from the aerosols that surroundit in the backscatter and depolarization ratio plots as well as the layer identification mask.This layer exhibits far higher optical depth than any other dust layer observed in thisstudy, as evidenced by the unusually high backscatter ratios. The depolarization ratiosexhibit the typical characteristics of dust, with a relatively high mean depolarization ratioexceeding 0.22 although there is little apparent mixing with the surrounding aerosols whilethe layer is within the LIDARFOV. The time period from 15:00, 24 March - 09:00, 25March (PDT) is also the only time that the LIDARdetected a strong signal from unmixeddust in the lower 1 km above the ground. The layer analysis for this event is shown in thebottom panel of Figure 5.5. In this case, the layer analysis recognizes the aforementioneddust-rich aerosol layer in the regions of strong backscatter, but it is accompanied by regionsof unidentified aerosol on the edges of the layer where the SNR is too low for an adequatelayer designation to occur. In addition to this, the cloud layer from 6 km to 9 km on March24 appears in the layer mask to contain a large portion of dust but it is much more likelythat this region is actually cloud. This, like similar cloud/aerosol uncertainty in Figure5.3 is an example of a recurring issue with optically thick clouds with high depolarizationvalues and is discussed further in Section 5.5: LIDARobservations for 24-25 March, 2010 showing 1064 nm backscatter ratios (top) and masked 532 nmdepolarization ratios (centre). The red line divides the backscatter ratio plot into Region A, which is dominated byaerosols, and Region B, which is mostly clouds. The arrows marked C point to areas that are suspected to containdust due to high depolarization ratios. Arrows marked D point to the background aerosols surrounding the dustlayer. Arrows marked E point to artefacts above optically thick layers that are not real features. The bottom panelshows the results of the LDA for this scene. The regions of aerosol and cloud that were demarcated in general termsas regions A and B, and with arrows C and D are clearly identified as aerosol and cloud layers in the layer maskplot, with some notable exceptions. Specifically, portions of the primary dust layer are marked as having insufficientsignal for layer analysis and a large portion of the upper cloud layer is marked as dust when it is perhaps more likelyto be composed of ice crystals. the artefacts in region E appear as mixed cloud in the layer mask plot.1055.1. Case Study #1: A Pervasive and persistent dust episode in North America - 2010Based on the LIDARratios assigned by the layer analysis, backscatter and extinctioncoefficients were calculated for the scene. The results are shown in Figure 5.6. Basedon this analysis the maximum extinction values for the dust rich layer are in excess of0.1 km−1 This value is very high for an aerosol layer so far from its source.1065.1.CaseStudy#1:APervasiveandpersistentdustepisodeinNorthAmerica-2010Figure 5.6: LDA inversion results for March 24-25 dust event. Panel A shows backscatter coefficients and PanelB shows extinction coefficients. As expected from the backscatter ratio values in the top panel of Figure 5.5, thebackscatter and extinction coefficients for the dust layer are very high, especially for an aerosol layer so far from itssource1075.1. Case Study #1: A Pervasive and persistent dust episode in North America - 2010These observations were accompanied by an increase in PM10 concentrations to 28 µg m−3at Vancouver International Airport. Surface level pressure maps of the area show the pass-ing of a ridge of high pressure aloft with subsiding winds during this time period, whichis consistent with the observation of relatively high optical depth dust layers close to theground as shown in Figure 5.7. With the passing of the front on 25 March, the dust overthe LIDARwas quickly removed.1085.1. Case Study #1: A Pervasive and persistent dust episode in North America - 2010Figure 5.7: NAAPS-NOGAPS time-height sections from nearby Cheeka Peak, roughly100 km south-west of Vancouver, showed a similar structure to that observed by the LI-DARwith increased concentrations of dust near the surface from 10:00, 23 March to 04:00,25 March, at which point surface concentrations of dust sharply declined. Image collectedfrom [NRL]1095.1. Case Study #1: A Pervasive and persistent dust episode in North America - 2010April 2010Layers with similar characteristics were again observed by CORALNet in Vancouver onApril 9-12, 2010. Investigation of the LIDARbackscatter ratio plots indicates that, similarto March 17-20, by the time the dust had reached Vancouver it had developed a complexvertical structure comprising multiple sub-layers, the thickest of which were concentratedat altitudes from 3 km to 6 km (Figure 5.8 top panel). By April 11, the 6 km layer hadsubsided and appears to have been entrained into the mixed layer as well.1105.1.CaseStudy#1:APervasiveandpersistentdustepisodeinNorthAmerica-2010Figure 5.8: LIDARobservations for Vancouver, BC from 09-12 April, 2010: 1064 nm backscatter ratios (top) andmasked 532 nm depolarization ratios (centre). The red line divides the backscatter ratio plot into Region A, whichis dominated by aerosols, and Region B, which is mostly clouds. Areas marked C are typified by high depolarizationratios, likely to contain high concentrations of dust. Areas marked D show reduction in depolarization probably dueto mixing with other aerosols. The areas marked E are artefacts due to optically thick layers below and are notreal features. In clouds note the distinct difference between dark blue areas (water) and red-orange areas (ice). Thebottom panel shows the results of the LDA for this scene. The regions of aerosol and cloud that were demarcatedin general terms as Region A,B,C, and D are clearly identified as cloud and aerosol layers. The artefacts marked asregion E appear in the layer mask as mixed cloud.1115.1. Case Study #1: A Pervasive and persistent dust episode in North America - 2010Just as in the events of March, the depolarization ratios are high for the higher altitudelayers, with mean ratios for layers above 3 km approaching 0.20 (Figure 5.8 center panel).Also as in the March case, as the layer altitudes decrease toward the mixed layer, the de-polarization ratios decreased. For the primary dust layer, the mean volume depolarizationratio dropped over the course of two days. This trend is discussed in more detail in Section5.1.4. The layer analysis clearly identifies these layers as “polluted dust” (Figure 5.8) andas before the designation changes from “polluted dust” to “smoke/urban” as the altitudesand the depolarization ratios decrease.Backscatter and extinction coefficient plots (Figure 5.9) show these layers to be opticallythicker than those observed during the similar event in March, especially for higher altitudeswhere the extinction coefficients for the layers peak at 0.025km−1.1125.1.CaseStudy#1:APervasiveandpersistentdustepisodeinNorthAmerica-2010Figure 5.9: LDA inversion results from April 09-12 dust event. Panel A shows backscatter coefficients and Panel Bshows extinction coefficients.1135.1. Case Study #1: A Pervasive and persistent dust episode in North America - 2010The LIDARobservations from Egbert show a similar pattern five days later, but withmany important differences (Figure 5.10) that pose an interesting challenge for layer iden-tification. At this location, elevated layers are originally seen briefly at around midnight on18 April distributed over altitudes from 5 km to 8 km. An elevated layer was detected priorto this on 18 April, but the depolarization ratios for this layer are low, remaining under 0.1(Region B in middle panel of Figure 5.10). This is an indication that for this time period,this layer is dominated by aerosols other than dust. This observation was corroboratedby back trajectory analysis for 18 April showing trajectories that were widely distributed,not focused over the Asian source regions. After a break during the first half of the dayon 19th, the primary layer is observed by the lidar. Backscatter ratios show that unlikethe Vancouver events, which were separated into multiple distinct layers, this dust wasembedded within a surrounding layer of more spherical aerosols from 3 km to 8 km. After20 April, the highest concentration occurred from 3 km to 4 km with the layer remainingoptically thinner above. It is also clear that by the time the dust layers travelled acrossNorth America to Egbert, the mean volume depolarization ratios within the layers hadbeen reduced substantially. The maximum depolarization ratio observed within the dustlayers over Egbert was 0.15, as compared to 0.27 over Vancouver. This could have beenthe result of continued loss of dust through deposition, external mixing with the aforemen-tioned background layer or more likely a combination of both. Less likely is the possibilityof continued internal mixing as dust particles were coated with nucleated gases such assulfates or nitrates, because although this type of atmospheric process for dust has beenobserved and documented repeatedly for relatively young dust particles [e.g. 62, 109] thereis no evidence the dust particles would continue to accrue further coatings after so manydays aloft. Although lower than that observed during the Vancouver events, the depolar-ization ratio for this layer is clearly within the lower end of the range that indicates thepresence of dust. The contrast in the depolarization ratios clearly distinguishes this layeras having a different composition than its surroundings. Considering the large distancebetween the source and the observation and the likelihood of mixing in the interim, it isto be expected that the depolarization ratios would be lower than those observed on thewest coast. It is clear that this layer contains a much higher dust content than the localaerosols observed in the boundary layer below it.1145.1.CaseStudy#1:APervasiveandpersistentdustepisodeinNorthAmerica-2010Figure 5.10: LIDARobservations for Egbert, ON from 18-21 April, 2010: The top two panels show the familiar1064 nm backscatter ratios (top) and masked 532 nm depolarization ratios (centre). Depolarization ratios identifyRegion A as dust. Region B indicates a larger contribution from background aerosols with low de-polarization ratiosfrom other sources. Region C is the atmospheric boundary layer. Areas marked D are clouds. The area marked E isan artefact of optically thick layers below and is not a real feature. The bottom panel shows the results of the LDAfor this scene. The regions of aerosol and cloud that were demarcated in general terms as Region A,B,C, and D areclearly identified as cloud and aerosol layers. The artefacts marked as region E appear in the layer mask as a regionof insufficient signal.1155.1. Case Study #1: A Pervasive and persistent dust episode in North America - 2010Perhaps more interestingly, along with a net reduction in volume depolarization ratio,the dust layers appear to show more consistency in optical properties over time and altitudethan those observed in Vancouver. With the exception of the unique event on 24 March,each of the Vancouver events showed a range of depolarization ratios from 0.20 to 0.10 orbelow during the course of the four day observation period and over the range of observedaltitudes. This is a good indication that the composition of the layers varied substantially.For the layers observed over Egbert, the initial depolarization ratio was lower, but the ratiowas consistent for all altitudes and remained almost totally constant during the entire fourday period for which it was visible (Figure 5.10). This qualitative observation is consistentwith Ryder et al. [87] who observed what they called a “weakly exponential” trend in theevolution of mean particle diameter, single scattering albedo, and other optical propertiesas a function of age.The layer analysis for this event shown in the bottom panel of Figure 5.10 revealsthat the aerosols observed from 2 km to 9 km comprise a complex mixture of dust andmore spherical aerosols (most likely urban aerosols as there were no known sources ofsmoke in the region at the time). This is a case where the capability of the LDA todistinguish between layers based on depolarization ratio in addition to backscatter strengthis particularly useful. This allows the embedded layer of dust to be separated from thesurrounding aerosols where more traditional algorithms that find layer boundaries basedsolely on backscatter profiles would see them as a single layer. This mixture of aerosoltypes is in agreement with both the initial LIDARobservations as well as HYSPLIT backtrajectories showing a portion of the air parcels passing over dust source regions in Chinaand others deviating to the north or south. Based on the layer analysis, backscatter andextinction coefficients were calculated for the event (Figure 5.11).1165.1.CaseStudy#1:APervasiveandpersistentdustepisodeinNorthAmerica-2010Figure 5.11: LDA inversion results for April 18-21 dust event. Panel A shows backscatter coefficients and Panel Bshows extinction coefficients. In this case, the lower 2 km above the surface exhibits by far the highest backscatter,but the extinction coefficients show a much different distribution due to the differing LIDARratios for the layer typesidentified by the LDA.1175.1. Case Study #1: A Pervasive and persistent dust episode in North America - 20105.1.4 Analysis of trends in depolarizationUsing the results of the LDA layer mask, the layers identified as “dust” or “polluted dust”were isolated and examined for trends in depolarization ratios. When these layers wereexamined, strong correlations were observed between depolarization ratio and altitude forthe events in Vancouver on March 17-20 and April 09-12, with mean depolarization ratioswell below 0.1 for altitudes below 1 km and ranging up to mean values of 0.23 for 6 km to8 km. This quantitative result provides support for the qualitative assessment of variationsin depolarization discussed in Section 5.1.3. There is a number of possible reasons for thisvariation. Perhaps the most likely explanation is that the lower altitude layers exhibitmore spherical aerosols. This is further supported by the fact that there are continuouslayers marked “smoke/urban” in the lower 500 m of both scenes13. In addition, there is aconsistent increase in the variance around the mean as the altitude increases. This is likelydue, at least in part, to portions of ice clouds being erroneously identified as dust (for moreon this phenomenon see Section 4.2.3)The event on March 24-25 shows a very different pattern, especially for the loweraltitudes where the majority of the dust was observed. In this instance the altitude plot islimited to 0 km to 6 km because the dust layers were confined to this range. As mentionedpreviously, it is likely that any layers marked as dust above this altitude are ice cloudsmisidentified as dust. For this event, there were dust layers observed primarily below 1 kmand again from 2 km to 3 km with a reduced dust presence in between. This is reflected inthe altitude plot in Figure 5.12 (b) where the median depolarization ratio at 500 m is 0.20.The minimum depolarization is reached at 1 km where the layer structure is dominated by“polluted dust”. From there it increases with altitude reaching a median value of 0.18 atan altitude of 3 km where a prominent layer of “dust” is identified embedded within the“polluted dust” layers.Figure 5.12 (d) shows the histogram from the data collected during the Egbert event onApril 18-21. As expected, this instance shows a much different pattern with altitude thanthe Vancouver events. In the Egbert data, the “polluted dust” layers are concentratedprimarily from 2.5 km to 7 km and are embedded completely within layers identified as“smoke/urban” (there were no “dust” layers in this instance). The histogram reveals nomeasurable variation in median depolarization ratio with altitude. Furthermore the median13It is important to note that given the location and the wind direction for these cases, and the verylow depolarization values for these low-altitude layers, it is impossible to rule out the possibility that theycould be marine aerosols1185.1. Case Study #1: A Pervasive and persistent dust episode in North America - 2010value is 0.12 throughout the range. These values were observed at only the lowest end ofthe range of depolarization in the Vancouver events. All of these observations are generallyconsistent with those made above based on an interpretation of the backscatter ratio anddepolarization ratio plots in Figures 5.3, 5.5, 5.8, and 5.10, but the use of the layer maskallowed a much clearer and quantitative analysis of these trends in depolarization.In Figure 5.13 the LDA layer masks were also used to produce two-dimensional his-tograms of depolarization with backscatter ratio, which revealed trends that are moredifficult to determine from viewing the backscatter ratio and depolarization ratio plotsthan trends with altitude. In Figure 5.13 (a), the event in Vancouver from March 17-20is revealed to have two distinct modes for the higher backscatter ratio layers with onemedian value at 0.12 and the other at 0.24. At the lower backscatter ratio values, themodes persist but the separation between them becomes less distinct. A similar trend wasobserved for the data in Figure 5.13 (c). The difference in this case is that the higherdepolarization mode has a median value of only 0.17 and the variance is much higher witha wider distribution toward the higher depolarization range for all backscatter ratio valuesthat in Panel (a). In Figure 5.13 (b) the range of depolarization ratios as well as backscat-ter ratio values extend much higher and the variance is much greater than in any of theother events. To the extent that any multi-modal character can be discerned, it is onlyapparent at backscatter ratio values above 2.5. In this region the higher depolarizationlayer is much higher than in any of the other events at 0.38 but the median of the lowermode is close to that of the other events at 0.13. In all of these events it is clear thatthe two modes are consistent with the standard definitions of “polluted dust” and “dust”respectively. As in the altitude analysis, the Egbert event in Figure 5.13 shows a smalldegree of variance around the 0.12 median depolarization ratio throughout the range ofbackscatter ratio values with no higher depolarization “dust” mode in evidence. What isless expected based on inspection of the backscatter ratio plots in Figure 5.10 (d) is thatthe distribution of backscatter ratio values is quite high as compared to the Vancouverevents in Figure 5.10 (a) and (c), with a median backscatter ratio value of 1.55 and veryfew values below 1.2. It seems counter-intuitive that the layers would have such strongbackscatter this far from the source. This is evidence to support the conclusion that thegeneral reduction in depolarization ratio in Egbert as compared to Vancouver is as a resultof mixture with higher concentrations of more spherical aerosols rather than as a result ofdeposition of more irregular particles over time.1195.1. Case Study #1: A Pervasive and persistent dust episode in North America - 20105.1.5 ConclusionsIn this study modelled as well as empirical observations of an extraordinary example ofspringtime dust transport from Asia to North America were presented. The event wasunusual among similar annual dust transport events in its size, spatial extent, and du-ration. The extent of the dust distribution was made evident with NAAPS global dustmodels. More detailed observations of the vertical structure of dust layers in Vancouver,BC and Egbert, ON were made with the use of CORALNet LIDARimagery. The EgbertLIDARobservations constituted the first recorded LIDARobservation of Asian dust at thislocation. The layers were identified as a combination of dust and polluted dust and weredistinguished from local urban aerosols through the application of the LDA. The specificlayers in question were traced back to the regions where dust storms occurred through theuse of multiple HYSPLIT back trajectories. This combination of NAAPS models and HYS-PLIT back trajectories along with the LDA was used to show that dust from earlier stormsin the Gobi region in March also reached the west coast of North America in observablequantities nearly a month prior to previous observations. Despite variations due to sourceregion, layers with high concentrations of dust were easily distinguished from other aerosoltypes by their high depolarization ratios (see Section 2.3.2). This task was somewhat com-plicated by the fact that the depolarization ratios for layers tend to decrease over time incases of long-range transport due to the combined effects of gravitational settling of thelargest particles and the introduction of more spherical aerosols (primarily black carbon,sulphates, and nitrates), in this case, the intermixed aerosols came from industrial centresin China, Korea, and Japan [63]. This behaviour is supported by several other studies aswell [e.g.: 19, 62, 71].In each case, the LDA that was performed on the CORALNet LIDARdata that gen-erated additional clarity with regard to the nature of the elevated layers and facilitatedaccurate calculation of layer optical properties. Taken together, these results complementthe results of Uno et al. [113], Fischer et al. [33], Li et al. [62] and Liu et al. [63] regardingthis event, as well as those of Cottle et al. [23], by revealing new details of the structureand dynamics of these layers over North America after weeks spent in the atmosphere.1205.1. Case Study #1: A Pervasive and persistent dust episode in North America - 2010Figure 5.12: 2-D histograms of dust layer depolarization vs. altitude for 2010dust events. Data are collected from layers identified as “dust” or “polluted dust” fromeach of the events described in Section 5.1.3. Panel (a) represents the data from the event inVancouver from March 17-20. In this instance a distinct trend of increasing depolarizationratio with altitude was observed. Panel (b) represents data from the event in Vancouverfrom March 24-25 where a single, highly concentrated layer with varying concentrationsof dust was observed in the lower 3 km of the atmosphere. Panel (c) represents data fromthe event in Vancouver from April 09-12. This event shared a great deal in common withthe event in Panel (a) with a trend of increasing depolarization with altitude. Panel (d)represents data from the event in Egbert from April 18-21. In this instance there was nodiscernible variation in depolarization with altitude.1215.1. Case Study #1: A Pervasive and persistent dust episode in North America - 2010Figure 5.13: 2-D histograms of dust layer depolarization vs. Normalized RelativeBackscatter for 2010 dust events. Data are collected from layers identified as “dust”or “polluted dust” from each of the events described in Section 5.1.3. Panel (a) representsthe data from the event in Vancouver from March 17-20. In this event a distinct bi-modaldistribution emerges with increasing backscatter ratio. Panel (b) represents data from theevent in Vancouver from March 24-25. In this instance, there is a much larger amount ofvariation in depolarization ratio, but for the higher end of the backscatter ratio range aseparation between the higher depolarization “dust” layers and the lower depolarization“polluted dust” layers is evident. Panel (c) represents data from the event in Vancouverfrom April 09-12. This event shared a great deal in common with the event in Panel(a) but in this case, the higher depolarization mode is not as separated from the lowerdepolarization mode. Panel (d) represents data from the event in Egbert from April 18-21.In this instance there was no discernible variation in depolarization with backscatter ratio.1225.2. Case Study # 2: Siberian wildfire smoke transport - 20125.2 Case Study # 2: Siberian wildfire smoke transport -20125.2.1 IntroductionIn July and August 2012, a combination of dry weather and record-breaking temperaturesled to an unusually intense wildfire season in Boreal Asia. During the peak months ofJuly and August, over 17,000 wildfires were detected in this region. The smoke emissionsfrom these fires were sufficient to be seen clearly in MODIS true-colour imagery not onlyover Siberia but also over south-western British Columbia (see Figure 5.14) where an ex-tensive aerosol monitoring program, including two CORALNet lidars, was in place for thedetection of long-range transport of aerosols. CORALNet LIDARobservations taken inVancouver during these months revealed aerosol layers in the free troposphere followed byincreases in backscatter ratio within the boundary layer peaking on July 7-10 and againon August 9-15. Depolarization ratios in the boundary layer and for layers in the free tro-posphere during this period were consistently low, and LDA performed on the LIDARdataindicate the presence of smoke. Throughout July and August, Total Suspended Particulate(TSP) monitors throughout the lower Fraser Valley of British Columbia revealed severaldays with a significant increase in PM2.5 concentrations and nine of the twenty highestdaily average PM2.5 concentrations of 2012 coincide with increases in backscatter in the LI-DARobservations indicating that these events were accompanied by a substantial increasein particulate concentrations near the surface.It is expected that the frequency and severity of wild fire activity will increase as aconsequence of global climate change in many regions [81]. This has important implicationsfor local air quality in downwind regions. Although trans-Pacific transport is less commonfor smoke than it is for dust, it is not unprecedented. Reports of smoke from Siberiainfluencing air quality in North America in previous years include those of [110] and Jaffeet al. [51]. The objective of this study is to provide a basis for further research into theglobal impacts of wildfire emissions by documenting a case of long-range transport that hada significant impact on local air quality in southwestern British Columbia. And althoughthe LIDARdata alone provide substantial value in researching the event, this is anothercase where the application of a layer identification mask allowed aerosol layers to be moreclearly identified in the process of determining their optical properties.1235.2. Case Study # 2: Siberian wildfire smoke transport - 2012Figure 5.14: MODIS-Terra images of smoke layers observed on both sides of the PacificOcean. Panel A depicts Siberian wildfires emitting smoke over the Sea of Okhotsk onJune 29, 2012 (source: [97]). Panel B shows smoke traces reaching south-west BritishColumbia from the west on July 07, 2012 (source: [98]). All MODIS-Terra true colourimages provided courtesy of Jeff Schmaltz and the MODIS Land Rapid Response Team,NASA GSFC1245.2. Case Study # 2: Siberian wildfire smoke transport - 20125.2.2 Smoke emission and transportAccording to the Global Fire Emissions Database (GFED) [Giglio] [36], it is estimated,that 9.1±0.23×106 Ha were consumed by wildfires within “Boreal Asia” in 2012. Satellitefire detections gathered by the MODIS instruments on the AQUA and TERRA satellitesrevealed that during the months of July and August alone, 17,000 individual wildfiresoccurred in the boreal forests of Siberia covering an estimated 4.75± 0.08× 106 Ha. Sincethe GFED began in 1998, this represents the second largest two-month total area burnedfor this region, exceeded only by the severe summer forest fire season of 2003.NAAPS ground-level smoke models show large concentrations of smoke traversing thePacific Ocean from Siberia to North America intermittently throughout July and August.According to these models, the initial transport took the form of a substantial plume ofsmoke that traversed the Pacific starting in late June and first reached the west coast ofNorth America on July 6. The highest concentrations first reached Vancouver on July 7(Figure 5.15 A), persisting through July 9. The arrival of this plume correlates well withthe timing of the activity in the free troposphere shown in Figure 5.16. HYSPLIT backtrajectories originated from Vancouver from altitudes of 600 m to 4000 m on July 8 areshown in Figure 5.15 B. These trajectories were tracked backwards in time for 150 hoursand reveal that a high percentage of the air parcels originated over Vancouver passed overthe source regions in Siberia approximately six days earlier. Notably, air parcels are shownto have arrived on the west coast of North America well south of the Canadian borderbefore turning north to travel along the coast to British Columbia. It is conceivable thatif any sufficiently large wildfires existed near this coastal region along Washington, Oregonand northern California that they could have contributed to the smoke plumes observed inVancouver.The NAAPS results show that pulses of smoke continued to emanate from Siberiaintermittently throughout the rest of July and into August. The NAAPS results fromAugust 10 (Figure 5.15 C) reveal concentrations of smoke over the Pacific on the orderof 8 µg m−3 to 16 µg m−3 . These layers extend to British Columbia and beyond into theinterior of Canada, but this time the main plume passed just to the north of Vancouver andconcentrations over the lower Fraser Valley, although not negligible, were still noticeablylower than those during July. By August 13 (Figure 5.15 E), the Siberian smoke appearsto have split into two paths: a Northern route that remained dispersed over the Arcticand a renewed southern plume that again brought concentrations of up to 16 µg m−3 to1255.2. Case Study # 2: Siberian wildfire smoke transport - 2012Figure 5.15: NAAPS and HYSPLIT model results. A), C) and E) show NAAPS surfacesmoke concentrations in µg m−3 B), D) and F) show HYSPLIT frequency plots of 150hour-long back trajectories originated for altitudes of 1600 m to 4000 m over Vancouver,BC.1265.2. Case Study # 2: Siberian wildfire smoke transport - 201232 µg m−3 to south-western British Columbia. As in July, both of these events correlatewell with the arrival of layers in the free troposphere shown in Figure 5.19.HYSPLIT back trajectories for August 10 showed a much more diverse array of tra-jectory paths than those seen for the July event (Figure 5.15 D). Back trajectories weredistributed from south to north with most reaching Siberian air space via a more northerlyroute through the Arctic, others diverging toward South East Asia and still others confinedto the west coast of North America. The spread of trajectories is consistent with the moredistributed nature of the layers exiting Siberia in the NAAPS models and the trajectoryspread overlaps with the smoke layers over the ocean, Alaska and the eastern regions ofSiberia as well. In contrast, HYSPLIT back trajectories originating on August 13 showa more concentrated path for the southern plume observed in the NAAPS model for thisday. The trajectories were very tightly packed, passing between a low pressure system overthe Gulf of Alaska and a high pressure system off the Siberian coast. These pressure sys-tems were well developed leading to steep pressure gradients at the surface level along thepath indicated by HYSPLIT, which is consistent with the relatively high concentrations ofsmoke reaching North America as predicted by NAAPS.It is notable that the plumes transported from Siberia were not the only potentialsource of smoke in the region at this time. July and August of 2012 was also a veryactive season for wildfires in North America and even a cursory inspection of the NAAPSmodels in Figure 5.15 reveals multiple large emanations of smoke from regional forest fires(e.g.: California, Idaho and the interior of British Columbia). Although these fires werein much closer proximity to the region, and there were multiple episodes during July andAugust when mixing could have occurred, HYSPLIT back trajectories clearly show thatprior to August 17, any smoke observed in the lower Fraser Valley during the specifictime periods in question for this study would have originated primarily from Siberianwildfires, as opposed to the aforementioned fires in the interior of North America. Theseresults were confirmed through high resolution smoke dispersion models generated using theBlueSky framework [61], which revealed no measurable ground-level smoke from regionalfires reaching Vancouver prior to August 17. BlueSky results predict some intrusion ofsmoke from regional fires into southwestern British Columbia on August 17 and 18, butwith maximum concentrations of no more than 4 µg m−3.1275.2. Case Study # 2: Siberian wildfire smoke transport - 20125.2.3 CORALNet observations and LDA resultsSmoke intrusion into the atmospheric boundary layer from the free troposphere was ob-served by CORALNet lidars multiple times in the summer of 2012. The first signs ofsmoke from long-range transport occurred from July 6-8. In Figure 5.16, distinct layerscan be seen in the free troposphere for this time period, initially at altitudes of 2 km to4 km. The aerosol layers in the free troposphere resumed late on July 8 after a brief hiatusand continued through July 10. For several days prior to these events, the LIDARshowedrelatively low optical depth from aerosols even within the atmospheric boundary layer. Astime progressed, the backscatter from the observed layers above 1800 m lessened while adistinct mixed layer became visible above the boundary layer and the optical thickness ofthe boundary layer below steadily increased as more aerosols were apparently entrainedinto it, peaking on July 9 (this indicated by black arrows in Figure 5.16). The LIDARdataalso reveal a strong diurnal cycling of the aerosol content of the boundary layer during thistime period with periods of relatively high levels of backscatter near the surface followed byperiods when the boundary layer is so free of aerosols it becomes almost indistinguishablefrom the free troposphere in the LIDARdata.1285.2.CaseStudy#2:Siberianwildfiresmoketransport-2012Figure 5.16: CORALNet UBC LIDARobservations for July 04-14, 2012. Panel A shows the ratio of measuredbackscatter to that of clear air. Panel B shows the volume depolarization ratio. N1 indicates the time for whichNAAPS models and HYSPLIT back trajectories are shown in Figure 5.15 A. P1-P3 indicate the times for whichprofiles are shown in Figure 5.22.1295.2. Case Study # 2: Siberian wildfire smoke transport - 2012In Figure 5.17 the LDA was applied to a subset of the data from Figure 5.16 duringwhich the majority of the activity in the free troposphere took place: July 5-10. Duringthis time, the LDA found multiple layers in the lower free troposphere and identified themajority of them as “smoke/urban”. This corresponds well to the proposed nature andorigin of the layers as smoke from long-range transport. There are some exceptions to this.Most notably, on July 8-9 the uppermost layers had higher depolarization ratios and crossedthe threshold into the “polluted dust” category. Also there was an intermittent layer in thelowest 200 m to 300 m for which the depolarization ratio was also high enough to cross thisthreshold. This lower layer was also occasionally strong enough to pass the “aerosol/cloud”threshold and be identified as cloud, but given the altitude and the conditions, it seemslikely that this is an invalid designation.1305.2.CaseStudy#2:Siberianwildfiresmoketransport-2012Figure 5.17: CORALNet UBC LDA layer type analysis results for July 05-10, 2012. Layers observed in the freetroposphere in Figure 5.16 clearly correspond to layers identified as “smoke/urban” in the layer mask.1315.2. Case Study # 2: Siberian wildfire smoke transport - 2012It is difficult to determine whether the increased depolarization ratios for some of thehigher altitude layers is actually due to an unusually large proportion of irregular parti-cles in the smoke layers, or the introduction of highly depolarizing aerosols from anothersource14.Based on the results of the layer identification, backscatter and extinction coefficientswere calculated for this event. The results are shown in Figure 5.18. While the extinctionvalues peak at 1.06 km−1 at an altitude of 600 m on July 5, this is an extreme outlier. Themedian extinction value for the layers identified as “smoke/urban” or “polluted dust” is6e−4 km−1 with 92% of values falling below 0.002 km−1.14Particularly energetic fires sometimes incur a large proportion of fly ash and lofting of dust into thesmoke plume. This phenomenon will be discussed in more depth in Section 5.31325.2.CaseStudy#2:Siberianwildfiresmoketransport-2012Figure 5.18: LDA inversion results for July 05-10, 2012. Panel A shows backscatter coefficients and Panel B showsextinction coefficients. The layers observed during this event had low backscatter coefficients, making some layersvirtually indistinguishable rom the molecular background in Panel A. However, the high extinction to backscatterratio for smoke resulted in a much more distinct presence for these layers in Panel B.1335.2. Case Study # 2: Siberian wildfire smoke transport - 2012Two similar events were observed in the free troposphere in August (see Figure 5.19);the first occurred on August 10, the second on August 12-13. These episodes, indicatedby red arrows in the figure, included increases in backscatter ratio in the free tropospherethat were of a strength comparable to those observed in July. These were followed byentrainment into a persistent mixing layer above the boundary layer (indicated by the whitearrows in Figure 5.19). In contrast to the events of July, these layers were accompanied bymuch more activity in the 4 km to 8 km range. Additionally, twice during August increasesin backscatter within the boundary layer consistent with measurable impacts on air qualityoccurred, but in August these events did not seem to be as closely correlated to activityin the free troposphere. The first of these increases peaked on August 07 and the secondpeaked on August 14.1345.2.CaseStudy#2:Siberianwildfiresmoketransport-2012Figure 5.19: CORALNet UBC LIDARobservations for August 04-16, 2012. Panel A shows the ratio of measuredbackscatter to that of clear air. Panel B shows the volume depolarization ratio. Markers N2 and N3 indicate timeswhen NAAPS models and HYSPLIT back trajectories are shown in Figure 5.15. P4-P6 indicate the times for whichvertical profiles were calculated for Figure 5.23.1355.2. Case Study # 2: Siberian wildfire smoke transport - 2012Although the backscatter strength of the observed layers was far lower in August thanin July, when the LDA was applied to this event the results were in many ways similar tothose of the July event, with persistent layers of low depolarization aerosols identified as“smoke/urban” in the lower 3 km and intermittent layers of higher depolarization aerosolsidentified as “polluted dust” appearing above. (see Figure 5.20). The primary differencesbetween the two events are that the August event had a higher proportion of aerosols inthe region above 3 km and the July event had more higher depolarization aerosols near thesurface.It is worth noting that the LDA results from the August event also include manyspurious results as a result of signal attenuation (such as the areas marked “dust” directlyabove the occasional cloud or particularly thick aerosol layer). Comparing the image inFigure 5.20 to Figure 5.19 reveals this.1365.2.CaseStudy#2:Siberianwildfiresmoketransport-2012Figure 5.20: CORALNet UBC LDA results for August 09-15, 2012. Smoke layers are clearly shown in the analysisto correspond to observed layers in Figure 5.191375.2. Case Study # 2: Siberian wildfire smoke transport - 2012The LIDARratios generated by the LDA were used to calculate backscatter and extinc-tion coefficients for the August event (results are shown in Figure 5.21). As expected fromthe backscatter ratio results, the range of backscatter and extinction coefficients are farlower in August than in July, with the exception of August 13 when the when the strengthof the layers clearly peaked.1385.2.CaseStudy#2:Siberianwildfiresmoketransport-2012Figure 5.21: LDA inversion results for August 09-15, 2012. Panel A shows backscatter coefficients and Panel B showsextinction coefficients.1395.2. Case Study # 2: Siberian wildfire smoke transport - 2012Detailed Layer AnalysisFigure 5.16 B and Figure 5.19 B both show very little apparent variation of depolarizationratio in the observed layers. This is a strong indication that the aerosols consisted almostexclusively of aged aerosols from biomass burning (see Section 2.3.2). A closer examina-tion of the vertical profiles of backscatter and depolarization ratios for various key momentsfrom these events is provided in Figures 5.22 and 5.23. The timing of the individual profilesis indicated by lines marked P1-P3 on Figure 5.16 and P4-P6 on Figure 5.19. The depo-larization ratio profiles in particular reveal some variation with altitude beyond what canbe seen in the false-colour plots. Specifically, the depolarization ratios consistently reach aminimum value in the region from 500 m to 1500 m, with slightly higher ratios appearingfor altitudes below and above this region. This is an indication of some degree of variabilityin the aerosol content of these layers with altitude, most likely due to varying degrees ofmixing between layers of different aerosol type, but the fact that the depolarization ratiosfor all altitude regions remains, in most cases, well below the threshold value of 0.10 is anindication that the smoke content remains high for all observed layers.Figure 5.24 shows the distribution of depolarization ratios with altitude for layers iden-tified as “smoke / urban” or “polluted dust” during the two smoke events in Vancouverin 2012. Both cases exhibit a strikingly similar pattern with median depolarization ratiovalues near the ground of 0.14 and a steady decrease to a minimum of 0.05 at 1 km. Abovethis altitude, both cases exhibit a steady increase in depolarization ratio with altitude, bothreaching maximum values for the median depolarization of 0.20 at 6 km. The increaseddepolarization near the ground is most likely the result of local soil particles within thewell-mixed PBL. In contrast, the increases at higher altitudes are harder to explain. Asmentioned previously, this could be the result of higher proportions of dust or ash particlesin the higher elevation layers, a reduction in spherical droplets with altitude, perhaps as aresult of lower humidity, or a mixture of other aerosols from different sources than biomassburning.In Figure 5.25, the distribution of depolarization ratios with backscatter ratio values isexamined for the layers identified as “smoke/urban” or “polluted dust”. As in the altitudeanalysis, there are many similarities between the two events. In both cases, the lowestbackscatter ratiovalues (near unity) show a broad distribution with a high concentration ofvalues near 0.04 but with values ranging up to 0.20. Both cases also exhibit a strong modecentred around 0.05 that extends throughout the backscatter range with little variation.1405.2. Case Study # 2: Siberian wildfire smoke transport - 2012This is consistent with smoke from biomass burning. Finally, both events exhibit a thirdmode that becomes most distinct in the higher backscatter ranges.In order to better understand the distinction between these layer types, a histogramof depolarization ratio values was generated where the contributions of the three primarylayer types were separated out (’Smoke/Urban’, ’Polluted Dust’ below 1 km, and ’PollutedDust’ above) (see Figure 5.26). As expected from the previous results, the histograms showa multi-modal distribution with three distinct peaks. The character of the distributionsremains almost identical for the two events. For the July event the peak values occurred at0.041, 0.078, and 0.11; for the August event the peaks were 0.048, 0.083, and 0.11. Inter-estingly, for both events, the first two peaks appear to both belong to the “smoke/urban”layers. Furthermore, this bi-modal distribution was found to be virtually invariant withaltitude. This is a strong indication that these layers comprise a hetergeneous externalmixture of layers, some of which are dominated by standard biomass burning particleswhile others contain a higher proportion of more irregular particles, possibly form anothersource. The main peak is consistent with the results of previous studies of biomass burn-ing as described in Section 2.2, which includes reference to a similar case wherein smokeparticles from Siberia were measured to have depolarization ratios of approximately 0.03after undergoing long-range transport to Leipzig [67]. The secondary peak is consistentwith more irregular particles such as smoke with an unusually high proportion of fly ashor a mixture of smoke and dust. It appears from Figure 5.26 that the “polluted dust”layers located at altitudes above 1 km are almost the sole contribution to the third peakcentred at 0.11. This is an indication that these layers contain an homogeneous mixtureof particles not entirely dissimilar to those that contribute to the second peak discussedpreviously, the difference being either that they contain a higher proportion of irregularparticles, or that the particles are more highly irregular. Although there are differencesin the the layers labelled as “polluted dust” found in the lower 1 km are shown to have awide distribution of depolarization values with a bi-modal distribution including peaks at0.081 and 0.11. These layers, although clearly a relatively small contribution to the whole,seem to contain a heterogeneous mixture of aerosols of the type that comprise peaks twoand three from the other layers.Although there is much that remains unclear about the nature of the observed layersduring these two events, the LDA has revealed some conclusive results. First, it is clearthat the layers observed in July and in August are nearly identical, which is strong evidencethat they are not only the same aerosol types but also from the same source. It is also1415.2. Case Study # 2: Siberian wildfire smoke transport - 2012clear that the aerosols observed over Vancouver consist of three distinct aerosol mixtures:a primary one that is almost exclusively smoke as per the standard definition, and twosecondary types that comprise some mixture of smoke with more irregular particles (mostlikely fly ash and dust). The origin of these layers and the path of transport are shownwith the use of NAAPS and HYSPLIT model results.5.2.4 ConclusionsAccording to the GFED database, the unusually active forest fire season of 2012 resulted inthe largest area burned during summer in Siberia since 2003. NAAPS models indicate thatthe smoke produced by these fires was transported across the Pacific to North Americaduring July and August in quantities sufficient to degrade air quality. During the same timeperiod, the CORALNet LIDARin Vancouver revealed two separate occasions in July andAugust of 2012 where layers of aerosols were observed in the free troposphere accompaniedby increased scattering in the PBL. The LDA identified the layers as either “smoke/urban”or “polluted dust”. More detailed analysis identified the observed layers as a mixture ofthree distinct layer types. The layers identified as “smoke/urban” were found to be aheterogeneous mixture of one layer type that is primarily smoke with a mean depolarizationratio of 0.04-0.05 and a second that is a mixture of smoke and more irregular particles suchas fly ash or dust. The layers identified as “polluted dust” actually occur in two differentgroups: one near the ground that is likely to be local in origin and a second in the freetroposphere that has a mean depolarization ratio of 0.11 and is a homogeneous mixture ofsmoke and more irregular particles similar to the one described before but with a somewhathigher proportion of dust and/or fly ash. HYSPLIT back trajectories originating from thelower free troposphere in Vancouver showed the air parcels carrying these aerosols to theregion to have passed over Siberia and the North Pacific rather than areas to the east andsouth where North American fires were active (e.g. California, Idaho and the interior ofBritish Columbia).1425.2. Case Study # 2: Siberian wildfire smoke transport - 2012Figure 5.22: Vertical profiles taken from CORALNet UBC LIDARobservations on July 07,09 and 14, corresponding to markers P1-P3 in Figure 5.16. For each profile pair, subscript(a) indicates a profile of backscatter ratios and subscript (b) indicates a profile of volumedepolarization ratios.1435.2. Case Study # 2: Siberian wildfire smoke transport - 2012Figure 5.23: Vertical profiles taken from CORALNet UBC LIDARobservations on August06, 12, and 14, corresponding to markers P4-P6 in Figure 5.19. For each profile pair,subscript (a) indicates a profile of backscatter ratios and subscript (b) indicates a profileof volume depolarization ratios.1445.2. Case Study # 2: Siberian wildfire smoke transport - 2012Figure 5.24: 2-D histogram of depolarization vs. altitude for 2012 smoke eventsPanel (a) shows the distribution of depolarization with altitude for the event in July 05-10.Panel (b) shows the distribution of depolarization with altitude for the event in August09-15.The distributions of depolarization ratios with altitude are very similar for the twoevents. They both exhibit an initial decrease in depolarization near the ground, reachinga minimum at around 1 km and a subsequent increase.1455.2. Case Study # 2: Siberian wildfire smoke transport - 2012Figure 5.25: 2-D histogram of depolarization vs. backscatter ratio for smokeevents in July, August 2012 Panel (a) shows the distribution of depolarization withbackscatter ratio for the event in July 05-10. Panel (b) shows the distribution of depolariza-tion with backscatter ratio for the event in August 09-15. The variation of depolarizationratio with backscatter strength reveals two distinct modes in the distribution for regions ofstronger backscatter and a third, broader distribution with a larger variance for backscatterratio values near unity.Figure 5.26: Histograms of depolarization ratios for layers identified as “smoke/urban” or“polluted dust” for smoke events in Vancouver. Data is separated into three categories:“smoke/urban”, “polluted dust” under 1 km and “polluted dust” over 1 km. Panel (a)shows data from the July event, Panel (b) shows data from the August event.1465.3. Case Study #3 - Analysis of wildfire haze events in Vancouver, July and August 20155.3 Case Study #3 - Analysis of wildfire haze events inVancouver, July and August 20155.3.1 IntroductionThe combination of a dry winter and unusually high temperatures for the preceding 12months in western North America led to a particularly severe summer forest fire seasonin Canada and the Western United States in 2015, with notable impacts on AOD and airquality throughout Canada and the US. During July and August of 2015 air quality inmany cities throughout North America was severely affected by smoke from these fires.One notable event, although far from the largest, occurred in Vancouver in early July. Inthis case, large amounts of smoke impacted the Lower Fraser Valley causing record highdaily values for PM2.5 concentrations, peaking at values of over 200 µg m−3 on the morningof July 05, and resulting in air quality warnings and advisories throughout the region. (seeFigure 5.30)Figure 5.27: Panel A: True colour image from July 05, 2015 showing large amounts of smokegathered over the Strait of Georgia with Boulder Creek and Elaho Canyon fires marked(source: [94]). Panel B: True colour image from July 06, 2015 showing smoke flowing overthe lower mainland from the Elaho Valley and Boulder Creek fires (source: [95]). Imagesfrom MODIS instrument on NASA’s AQUA satellite is thanks to Jeff Schmaltz and theMODIS Land Rapid Response Team, NASA GSFC1475.3. Case Study #3 - Analysis of wildfire haze events in Vancouver, July and August 2015LIDARdata collected in July and August revealed that smoke layers lingered in thelower troposphere over Vancouver long after the air quality warnings ceased. Smokeemission and transport models from NAAPS and satellite data collected and reportedby NOAA’s Hazard Mapping System (HMS) were found to be in strong agreement thatmultiple fires spread throughout Washington, Oregon, Idaho and northern California con-tributed to these layers despite the fact that depolarization ratios for the elevated layerswere somewhat higher than what is normally expected for smoke from biomass burning.Through the use of the LDA, the distributions of backscatter strength and depolar-ization ratio for elevated layers in the lower 3 km from July and August are analysed andcompared to determine what differences can be discerned between the smoke from short- tomedium-range transport observed in July and the long-range transport observed in August.5.3.2 Smoke emission and transportThe smoke observed throughout the Lower Mainland in July is thought to have originatedprimarily from two nearby forest fires located 100 km to 150 km to the north of the city,one in the Elaho Valley and the other near Boulder Creek. Both fires were started bylightning strikes in mid-June and had remained small and relatively well-contained untilJuly when gusty winds and increasing temperatures led both to begin growing rapidly.By July 7, the Elaho Valley fire and its associated scar covered over 20 000 ha and firemanagers reported crown fires resulting in uncontrolled growth on all fronts. During thesame period, the Boulder Creek wildfire grew from less than 500 ha over 1500 ha in a singleday eventually reaching a size of 5500 ha by the time of the smoke event in Vancouver.[source: BC-Wildfire-Service]. On July 04-06, 2015 extraordinarily dense haze from thesefires was observed in Vancouver (see Figure 5.27). After July 06 the wind direction shiftedfrom the northwest to the southeast, resulting in a rapid decrease in PM2.5 at the surfacefrom a daily maximum of 118 µg m−3 on July 06 to one of 18 µg m−3 on July 07 [source:BC-MOE]. However, LIDARdata taken from July 06-10 reveal that smoke layers lingeredin the lower free troposphere from 1 km to 3 km for several days afterwards. NAAPS modelresults as well as HMS observations confirm that the region was inundated with smoke, butthere were so many fires in the region that the Elaho Canyon and Boulder Creek fires aredifficult to see in the model results (see Figure 5.28). However given the proximity of thefires, their strong smoke output, and the direction of the prevailing winds, it is reasonableto assume that the smoke observed in Vancouver was from these local sources as opposed1485.3. Case Study #3 - Analysis of wildfire haze events in Vancouver, July and August 2015to the more distant ones in the US.LIDARdata collected in Vancouver from August 10-26, show continuous aerosol layersin the lower 3 km with depolarization signatures consistent with smoke (more on this inSection 5.3.3). In contrast to the July event, there were no uncontrolled local fires thatwere rapidly growing during this time period. Figure 5.29 shows the NAAPS and HMSresults from a selection of days in this time period. These results indicate that multiplefires spread throughout the Pacific Northwest, including Washington, Oregon, Idaho, andNorthern California were releasing large amounts of smoke that was spreading north intoregions of Canada extending from British Columbia to Manitoba, including the LowerMainland.Based on these results and the depolarization signatures observed in the observed layers,it is reasonable to conclude that these layers were smoke from biomass burning that hadbeen transported an average of several hundred kilometres to reach Vancouver, likely takingseveral days in transit, as compared to the smoke observed in July that was more likely tobe little more than 24 hours old.1495.3. Case Study #3 - Analysis of wildfire haze events in Vancouver, July and August 2015hFigure 5.28: Smoke emissions data products for July 07, 2015 Panel A: NAAPSsurface level smoke concentration map. Panel B: NAAPS Smoke AOD map. Panel C:NOAA HMS Fire and Smoke Product map. July model and satellite data show strongagreement as to the nature and origin of the layers observed in Vancouver in early July.1505.3. Case Study #3 - Analysis of wildfire haze events in Vancouver, July and August 2015hFigure 5.29: Smoke emissions data products for August 11-25, 2015 Panel A:NAAPS surface level smoke concentration map for August 11, 2015. Panel B: NOAA HMSFire and Smoke Product map for August 11, 2015. Panel C: NAAPS surface level smokeconcentration map for August 17, 2015. Panel D: NOAA HMS Fire and Smoke Productmap for August 17, 2015. Panel E: NAAPS surface level smoke concentration map forAugust 25, 2015. Panel F: NOAA HMS Fire and Smoke Product map for August 25, 2015.These days were selected as representative of the extended period in August during whichsmoke was observed in Vancouver. The source for Panels A,C, and E is [NRL], and thesource for Panels B,D, and F is [UMBC-ALG].1515.3. Case Study #3 - Analysis of wildfire haze events in Vancouver, July and August 20155.3.3 Mini-MPL observations and layer analysisThe LDA results are shown in the bottom panel of Figure 5.30. The LIDARreturns fromthe observed smoke layers on these days are particularly interesting due to their unusuallyhigh depolarization ratios for smoke, leading the LDA to identify them as “polluted dust”rather than “smoke/urban” Table 2.2 shows that for most cases of long-range transport,the depolarization ratio for smoke layers is < 0.1015 but in the case of these layers thedepolarization ratio range was consistently above that threshold. This is much like theelevated “polluted dust” layers observed in Section 5.2 that were identified through detailedanalysis as a mixture of smoke and more irregular particles.15Many researchers have found the depolarization ratio of smoke to be so consistently low that in theabsence of direct measurement of depolarization ratios, it is often assumed to be < 0.05.1525.3.CaseStudy#3-AnalysisofwildfirehazeeventsinVancouver,JulyandAugust2015Figure 5.30: Mini-MPL measurements of smoke event in Vancouver, July 06-10, 2015 Panel A showsnormalized relative backscatter, Panel B shows volume depolarization ratios, and Panel C shows the results of theLDA. Normalized relative backscatter values in Panel A reveal a consistent layer of elevated aerosols ranging from1 km to 2 km The layers show an unusually high levels of depolarization for smoke, indicating some mixture of soilparticles and large amounts of fly ash in the smoke. This is all consistent with relatively young smoke layers andthose generated by intense fires accompanied by strong upward convection. Therefore in this instance despite strongevidence that the elevated layers originated from biomass burning, the mean depolarization ratios for the layers exceedthe threshold for “polluted dust” in the LDA. Note that in this instance the PBL detection function was turned offto allow viewing of layer details within the PBL1535.3. Case Study #3 - Analysis of wildfire haze events in Vancouver, July and August 2015Aerosols in smoke layers are actually a complex mixture of ash, black carbon (soot),spherical droplets formed from a variety of organic and inorganic gases, and water vapour.The exact mixture and the size distribution are both determined by the temperature of thefire, the type of fuel, and the availability of oxygen during combustion. Fires that have ahigh oxygen availability burn hotter and tend to have a higher proportion of ash and fewerof the other constituents. Smouldering fires will produce more soot and hydrocarbons.[25] It has also been shown that particularly intense wildfires will produce strong upwardconvection currents that can mix a substantial amount of dust into the layers entering thefree troposphere [e.g. 75, 114] Ash and dust are larger, more irregular particles and theirpresence will tend to increase the layer depolarization ratio. Soot occurs as a randomlyoriented fractal aggregate of smaller particles. Mie scattering models indicate that throughinternal and external multiple scattering effects, such conglomerates will tend to induce acertain amount of depolarization as well, although the amount is difficult to predict highlydependant on the shape, orientation, and number of these particles ([101, 108]).It has also been observed that smoke particles undergoing long-range transport tend toincrease in size over time due to a combination of processes including hygroscopic growth,conversion of organic and inorganic vapours to particles during transport, coagulation ofparticles, and photochemical processing [74]. Based on the nature of these processes, it isreasonable to assume that they would tend to increase the ratio of scattering from sphericalparticles to aggregates in a given layer as the number and size of spherical droplets increase.In addition, wet and dry deposition processes would tend to scavenge coarse mode dustand ash particles during the early stages of long-range transport. Most measurements ofsmoke plumes provided in the literature are at least several days old. Given the proximityof the fires and the wind speed, HYSPLIT back trajectories indicate that the smoke layersobserved in Vancouver were less than a day old. It is reasonable to assume that theywould contain a higher proportion of coarse mode particles such as ash and dust than theolder plumes that form the basis of most depolarization measurements for biomass burningreported in Section 2.3.2. In this way, the Vancouver observations are similar to those of[107] who observed mean depolarization ratios of 0.15 for smoke plumes transported fromnorthern Mongolia to Tokyo in 2007, or the observations of [38] for smoke from biomassburning mixed with desert dust.Based on these findings, the unusually high depolarization ratios for the smoke observedby the Vancouver LIDARcould be due in part to the fact that the fires of origin were burningat a particularly high temperature, with associated strong convection (a conclusion that is1545.3. Case Study #3 - Analysis of wildfire haze events in Vancouver, July and August 2015supported by the high rate of growth) resulting in a large proportion of ash and dust inthe layers as they were emitted. It is also possible that due to the fact that the observedplumes of smoke were less than a day old, fewer of the larger, more irregular particles hadbeen scavenged from the layers and less coagulation had occurred as compared to the older“standard” smoke plumes observed in the bulk of the literature on the subject. It is thusconcluded that the designation of “polluted dust” for these layers is not erroneous, but israther a reflection of the high content of dust and ash in the young smoke plume.Figure 5.31 in combination with Table 5.1 shows the daily ranges of depolarization ratiosand normalized relative backscatter values for layers marked “smoke/urban” or “polluteddust” in the lower 3 km of the layer mask for July 6-10. What they reveal about thedepolarization of the elevated layers is a relatively narrow distribution around mean valuesthat were consistently well above the threshold of 0.10 that separates the “smoke/urban”designation from “polluted dust” in the LDA, but the median values were clearly decreasingfrom day to day. This is an indication that the ratio of spherical coagulates and blackcarbon particles is increasing as a proportion of the whole over time even though thenumber of soil particles and fly ash remains consistently high as compared to typical smokeplumes. This is all consistent with smoke generated by particularly intense fires and withfresh smoke that has spent a relatively short time in transport, but in this instance, whetherthe decreasing trend is a result of a reduction of the intensity of the fires or of the smokespending more time in transit or a combination of both is unclear.Table 5.1 also shows this the percentage of the air in the PBL and lower free tropo-sphere that contains some amount of smoke. The percentages are calculated as the ratioof the number of volumes of integration below 3 km identified as either “smoke/urban”or “polluted dust” to total volumes of integration. This is just an indication that thesevolumes contain some amount of smoke and it is not scaled in any way by the optical depthor strength of backscatter measured in these volumes of integration. These results showthat as the median depolarization ratios decrease, the proportion of smoke-containing airparcels also decrease. On July 6, over two thirds of the air parcels below 3 km containedsome amount of smoke, but by July 10, as the wind direction was starting to shift, theportion was reduced to below one third. Interestingly, the normalized relative backscatterstatistics reveal that the total integrated normalized relative backscatter peaked on July 7,when the percent of air parcels containing smoke reached a local minimum. This indicatesthat the backsactter within those parcels must have been particularly intense, which is1555.3. Case Study #3 - Analysis of wildfire haze events in Vancouver, July and August 2015Figure 5.31: Box-plots of daily mMPL measurements for layers identified as “smoke/urban”or “polluted dust” below 3 km from July 6-10, 2015. Panel A shows depolarization ratiodistributions and Panel B shows distributions of normalized relative backscatter values.Median depolarization ratios are consistently well above the 0.10 threshold for “polluteddust” but also show a decreasing trend from day to day. normalized relative backscattervalues peak on July 7 and consistently show large standard deviations as compared to themedian values.1565.3. Case Study #3 - Analysis of wildfire haze events in Vancouver, July and August 2015DateDepolarization ratio Normalized relative backscatter Coverage below3 kmMedian Std Median Std07/06 0.156 0.050 0.101 0.072 67.7%07/07 0.124 0.039 0.150 0.267 40.3%07/08 0.132 0.046 0.142 0.137 40.5%07/09 0.112 0.040 0.089 0.093 48.7%07/10 0.106 0.037 0.092 0.297 29.5%Average 0.125 0.043 0.113 0.176 45.4%Table 5.1: Smoke layer depolarization and normalized relative backscatter statistics forJuly, 2015consistent with the high AOD levels from smoke predicted by NAAPS in Figure 5.2816.In this instance the LIDARsignal was too attenuated by the smoke to make a reliablecalculation of backscatter and extinction coefficients possible. The SNR levels above theselayers was below the threshold for profile inversion set out in Section 4.2.1, especially duringthe day when background signal levels were high.Smoke plumes were observed again over Vancouver from August 10-26, finally endingwith a rainstorm followed by a shift in wind direction moving the majority of the smokeplume to the Midwest. The date range of August 10-14 was selected as a representativesample and the results are shown in Figure 5.32. On multiple occasions throughout themonth, layers were observed entering the free troposphere at altitudes ranging from 2 kmto 6 km while nearly continuous layers were observed in the lower 1 km. In this case thelayers had much lower mean normalized relative backscatter values than in the July event,wherein the optical depth of the layers was exceptionally high. In the August event, thesmoke was present for an extended time but in much lower quantities. The depolarizationratio for these layers was much more in line with those in the literature for smoke. Itappears from preliminary smoke maps (e.g. those shown in Figures 5.28 and 5.29), theseplumes had been transported from fires in Washington, Oregon and northern California andit is thus reasonable to assume that they were several days old at the time of observation.16The LIDARwas rendered inoperable for a large portion of the day on July 07 due to a power outage1575.3.CaseStudy#3-AnalysisofwildfirehazeeventsinVancouver,JulyandAugust2015Figure 5.32: Mini-MPL measurements of smoke event in Vancouver, August 10-14, 2015 Panel A showsnormalized relative backscatter, Panel B shows volume depolarization ratios, and Panel C shows the results of theLDA. Normalized relative backscatter values in Panel A reveal a consistent layer of elevated aerosols ranging fromnear the ground up to 2 km with intermittent injections of smoke layers reaching up to 6 km. In Panel B the layersshow depolarization ratios that are more typical of smoke, but still on the upper end of the range, indicating somepresence of soil particles and fly ash but substantially less than for the July event. Note that in this instance thePBL detection function was turned off to allow detailed viewing of layer details within the PBL1585.3. Case Study #3 - Analysis of wildfire haze events in Vancouver, July and August 2015The Panel C in Figure 5.32 shows the LDA results for the August event. As would beexpected for depolarization ratios as low as those observed for this event, the majority ofthe air parcels were identified as “smoke/urban” with only a fraction crossing the thresholdinto the “polluted dust” category. As in previous cases, based on the layer designationsalone it is not possible to determine whether the “polluted dust” layers have elevateddepolarization ratios due to a mixture of irregular particles from an external source or ifthe smoke layers themselves contain a higher than normal proportion of fly ash and dustparticles.The optical depth of the layers in August was not so high as to preclude inversionsto calculate extinction and backscatter coefficients, although the SNR levels during thedaylight hours are low enough that the uncertainty in the calculations is high. (see Figure5.331595.3.CaseStudy#3-AnalysisofwildfirehazeeventsinVancouver,JulyandAugust2015Figure 5.33: LDA inversion results from August 10-14, 2015. Panel A shows backscatter coefficients and Panel Bshows extinction coefficients. In this case, stray light during daylight hours caused a substantial decrease in SNRresulting in noticeable variations in the results between day and night.1605.3. Case Study #3 - Analysis of wildfire haze events in Vancouver, July and August 2015DateDepolarization ratio Normalized relative backscatter Coverage below3 kmMedian Std Median Std08/10 0.059 0.035 0.045 0.109 50.7%08/11 0.080 0.033 0.051 0.022 58.0%08/12 0.092 0.027 0.059 0.017 66.8%08/13 0.094 0.028 0.055 0.024 67.8%08/14 0.076 0.036 0.052 0.746 54.2%08/15 0.052 0.051 0.066 1.000 23.6%08/16 0.062 0.036 0.039 0.351 26.0%08/17 0.064 0.028 0.036 0.006 35.1%08/18 0.072 0.026 0.035 0.005 36.4%08/19 0.074 0.029 0.033 0.008 36.0%08/20 0.060 0.032 0.039 0.723 37.0%08/21 0.059 0.031 0.038 0.569 39.3%08/22 0.071 0.025 0.035 0.007 33.5%08/23 0.083 0.037 0.039 0.012 46.1%08/24 0.093 0.045 0.043 0.028 30.1%08/25 0.095 0.037 0.043 0.009 34.3%08/26 0.106 0.046 0.040 0.008 37.5%Average 0.077 0.034 0.045 0.354Table 5.2: Smoke layer depolarization and normalized relative backscatter statistics forAugust, 2015Figure 5.34 in combination with Table 5.2 shows the daily range of depolarization ratiosand normalized relative backscatter values for layers marked “smoke/urban” or “polluteddust” in the lower 3 km of the layer mask for August 10-26. The mean depolarization ratioscycle up and down slightly over the course of the month, ranging from 0.052-0.011 witha mean value of 0.08. As for the July event, Table ?? also includes the percentage of airparcels that contained dust in some form. For the August event, this ranged from a maxi-mum of 68% on August 13 to a minimum of 23 % on August 15. As before, it’s importantto note that an air parcel containing smoke is not an indication of the concentration ofsmoke particles in the layer, just a count of the portion of the air parcels that contain someamount of smoke. For the August event, the median values were consistently less than halfof the July event often with a markedly higher standard deviation as well.1615.3. Case Study #3 - Analysis of wildfire haze events in Vancouver, July and August 2015Figure 5.34: Box-plots of daily mMPL measurements for layers identified as “smoke/urban”or “polluted dust” below 3 km from August 10-26, 2015. Panel A shows depolarization ra-tio distributions and Panel B shows distributions of normalized relative backscatter values.During August median depolarization ratios were markedly lower than in July, as weremedian normalized relative backscatter values, which is consistent with the fact that thesmoke was being transported much further from fires in Washington, Oregon, and Califor-nia. There was also a noticeable variation in depolarization over the course of the month,indicating variations in the composition of the smoke layers, and possibly also in origin.1625.3. Case Study #3 - Analysis of wildfire haze events in Vancouver, July and August 2015Figure 5.35: 2-D histogram of depolarization vs. altitude for smoke events inJuly and August of 2015 Panel (a) shows the distribution of depolarization with altitudefor the event in July 06-10, 2015. Panel (b) shows the distribution of depolarization withaltitude for the event in August 10-26. In Panel (a) there was almost no change at all inthe depolarization distribution with altitude, aside form a higher overall concentration inthe lower altitudes.Detailed layer analysisAs was done in the previous cases, the depolarization ratios were investigated further forany correlation to altitude or normalized relative backscatter, using the layer mask as afilter. In Figure 5.35 (a) the depolarization values for the July event are plotted againstaltitude. For this event there is almost no variation in depolarization distribution withaltitude in the lower 3 km. For the August event, shown in Figure 5.35 (b), there is amarked difference in mean value compared to July. In addition, the August event shows aslight increase in mean depolarization and standard deviation both above and below the500 m mark.The distributions of depolarization ratios with normalized relative backscatter wereplotted in Figure 5.36 The results from the July event are shown in Panel (a). In thisinstance, as in previous examples of this analysis, there is a region of low normalizedrelative backscatter with values below 0.1counts∗km2/µs∗µJ for which the variance is muchhigher than for the rest of the normalized relative backscatter range. There is also a verydense concentration of values near the mean depolarization of 0.125 for this event. Likewise,there is a high concentration of values near a normalized relative backscatter value of 0.1.1635.3. Case Study #3 - Analysis of wildfire haze events in Vancouver, July and August 2015Figure 5.36: 2-D histogram of depolarization vs. normalized relative backscatterfor smoke events in July and August of 2015 Panel (a) shows the distribution ofdepolarization with normalized relative backscatter for the event in July 06-10, 2015. Panel(b) shows the distribution of depolarization with normalized relative backscatter for theevent in August 10-26. In both cases, very little variation was observed in depolarizationratios for normalized relative backscatter values above 0.1counts∗km2/µs∗µJFigure 5.36 Panel (b) shows an even higher degree of concentration around he meanvalues than in July. Also, in this case, the mean normalized relative backscatter anddepolarization ratio values is quite a bit lower than in Panel (a)These distributions reveal important differences between the smoke layers observed inVancouver in 2015 as compared to those observed in 2012. In this case, despite the fact thatthe layer identifications are split between “polluted dust” and “smoke/urban”, there is littlesign of a bi-modal or tri-modal distribution emerging in the higher backscatter regions ineither the July or the August data. This property of the distribution is investigated furtherin the histograms in Figure 5.37.During the July, 2015 event, when “polluted dust” dominated the layers, the layers withthis designation are shown to have one dominant mode with a mean value at 0.117 withsecondary mode with a mean value of 0.152. This bi-modal distribution indicates a slightlyheterogeneous distribution of air parcel types within the layers identified as “polluted dust”.Although the layers identified as “smoke/urban” in this event are small in number, it isstill possible to discern a distinctly bi-modal distribution for them as well.1645.3. Case Study #3 - Analysis of wildfire haze events in Vancouver, July and August 2015Figure 5.37: Histograms of depolarization ratios for layers identified as “smoke/urban”or “polluted dust” for smoke events in Vancouver in July and August of 2015. Datais separated into two categories: “smoke/urban”, and “polluted dust”. Panel (a) showsdata from the July event, Panel (b) shows data from the August event. In Panel (a) thedistribution is dominated by the “polluted dust category” that exhibits a slightly bi-modaldistribution. In Panel (b) the primary contribution is from uni-modal “smoke/urban”layers.In Figure 5.37 Panel (b), the August event is shown to be dominated by “smoke/urban”layers and in this case the distribution is clearly uni-modal. The layers identified as “pol-luted dust” still show a bi-modal distribution as did the ones in July. This is an indicationthat for these layers, the lower depolarization layers were more homogeneously mixed thanany of the others observed in this case, more consistent with a larger proportion of sphericalparticles and older smoke plumes that have not been mixed with other aerosol types enroute. The “polluted dust” layers in this event remain bi-modal indicating some degree ofmixing with more irregular particles.5.3.4 ConclusionsIn July and August of 2015, north-western North America saw an usually high amount offorest fire activity, incurring evacuations and large amounts of property damage in manyregions and measurably affecting air quality in many locations throughout the continent.Vancouver was just one of several municipalities to see record high PM2.5 levels. In earlyJuly, fires in Boulder Creek and Elaho Canyon only 100 km to 150 km north of Vancouverquickly intensified and grew by an order of magnitude in just a few days. This, combinedwith north-westerly winds resulted in an influx of smoke into the city from July 04-10,peaking on July 5. LIDARdata collected in Vancouver from July 06-10 revealed intense1655.4. Summary and conclusionssmoke layers throughout the lower troposphere with the highest concentrations ranging inaltitude from 300 m to 2000 m The layers detected during this event exhibited unusuallyhigh depolarization ratios for smoke, resulting in layer designations of “polluted dust”.Based on the fact that the smoke originated from particularly intense fires and that it waslikely to have spent less than a day in transit, it is likely that the increased depolarizationwas due to a combination of dust lifted into the plume due to convection from the intenseheat and a high proportion of scattering from fly ash as opposed to spherical coagulatesdue to the very young age of the smoke. Detailed analysis of these layers showed thedistribution of depolarization ratios to be very consistent with altitude, indicating that allof the layers originated form the same source leaving little doubt that the observed layerswere entirely smoke from the two local fires.Although the smoke event in early July was widely reported, LIDARdata show thatsmoke layers persisted in the lower troposphere over Vancouver throughout August. Thesedid not exhibit anything like the intensity of the July events, but they lasted much longer.In the August event, the LDA identified the majority of layers as “smoke/urban” dueto the relatively low depolarization ratios, more in keeping with typical expectations ofsmoke. Preliminary model results appear to show the smoke from these layers originatingfrom fires in Washington, Oregon and northern California. As opposed to July, the layersobserved in August were dominated by a single homogeneous aerosol mixture. The higherdepolarization layers identified as “polluted dust” were bi-modal in nature indicating aheterogeneous mixture of two aerosol types.This capability to perform detailed analysis of aerosol properties (depolarization inthis case) is among the most important contributions of the LDA to the use of LIDARforaerosol research.5.4 Summary and conclusionsIn this chapter three cases are presented that highlighted the application of the LDA inaerosol analysis. The first case involves a case of Asian dust transport to North Americain March and April of 2010 where dust layers were observed in Vancovuer, BC and Egbert,ON. Dust transport of this kind is now known to be an annual event, but this case wasunusual both in its persistence and it extent. The second case deals with plumes of wildfiresmoke generated in Siberia that were transported to western North America in July andAugust of 2012 (an event much more rare than the annual dust transport, but not unheard1665.4. Summary and conclusionsof). The third case deals with another example of smoke from biomass burning beingtransported to Vancouver in 2015, but in this instance, the origins of the smoke were muchcloser: in July of 2015 the smoke originated from local fires just outside of Pemberton,BC, while in August the origins were likely ot be fires in Washington, Oregon, Idaho andNorthern California.In each of these cases, the LDA was employed to further the investigation of the aerosolactivity in question, either as a means of identifying layers of interest, or as a tool forfiltering LIDARscenes to allow more in-depth analysis of the optical properties of thelayers in question, or both. CORALNet data were analysed in the first two cases, andmMPL data in the third. Together, these example cases serve to show the efficacy of theLDA in real-world research scenarios and to demonstrate a few of its possible applications167Chapter 6ConclusionsIn this thesis, a novel Layer Discrimination Algorithm (LDA) has been developed andthen evaluated under a range of dust and smoke transport events in multiple geographiclocations. This is not the first algorithm of its kind, but it is the first to be developedfor a ground-based single-wavelength elastic LIDAR that attempts to discriminate not justbetween clouds and aerosols but also between sub-types within these two categories.6.1 Summary of resultsIn this thesis, the LDA has been employed in a number of ways to contribute to aerosolresearch. Some of those contributions are included in the three Case Studies described inChapter Use of layer masks for interpretation of LIDAR resultsPerhaps the most useful application of the LDA thus far has been in isolating layers, or layertypes, of interest for closer examination. This feature allows for analysis of layer propertiesthat is simultaneously more detailed and more extensive than is achievable otherwise.More detailed because isolating layers for analysis allows distributions of optical propertiesspecific to a given layer to be seen, and more extensive because the analysis allows statisticsto be gathered for specific layer types over several days or weeks, or even longer. The casestudies in this document focus primarily on the distributions of depolarization ratios forthese layers, but of course the layer mask can be used as a tool for investigating any aerosolproperty, including ones that are not directly measured by the LIDAR itself. The fact thatlayers can be identified in a scene quickly and reliably, and when used appropriately, withminimal need for user intervention, allows for much larger and more complex datasets tobe analysed than could ever be attempted otherwise.The efficacy of this approach was demonstrated repeatedly in each of the three Case1686.1. Summary of resultsStudies in Chapter 5. In each case, layers of interest were separated from the rest of the dataand the distributions of depolarization were examined as a means of better understandingthe nature of the layers in question. This included plotting histograms of depolarizationratios, and looking into variations in polarization with altitude and correlations betweenpolarization and backscatter. To produce analyses such as these without a layer maskwould be at best exceedingly difficult.One example where the layer masks were particularly effective involved the event in-volving long-range transport of smoke from Siberia to North America in July 05-10, 2012(Section 5.2.3). In a case such as this, where aerosols from biomass burning had trav-elled thousand of miles before being detected, one might normally expect to observe layersthat were both homogeneous and relatively low in depolarization, but this was not thecase. Once the layers were isolated, a series of histograms of layers identified as either“smoke/urban” or “polluted dust” revealed complex, heterogeneous mixture of at leastthree separate aerosol sub-types within the layers. This process cast new light on whatappeared at first glance to be a largely homogeneous mixture.6.1.2 Calculation of extinction and backscatter using layer-specificLIDAR ratiosOne of the primary purposes of the LDA is to assign LIDAR ratios to the layers for thepurposes of improving the accuracy of the profile inversions used to calculate backscatterand extinction coefficients. Section 4.2 shows that the introduction of a layer-by-layerprofile of LIDAR ratios results in an appreciable improvement in the accuracy of theinversion over the use of a single arbitrary LIDAR ratio for the whole profile despiteuncertainty in the assigned LIDAR ratios and the challenges inherent in attempting toaccurately invert LIDAR profiles in the presence of noise. This method is particularlyeffective for profiles with multiple layers of different type, especially if one or more of themis optically thick, so it is useful in complex aerosol profiles or when aerosol layers and cloudscoexist. Wherever possible, extinction and backscatter inversions were performed on thedata presented in Chapters 4 and 5. The only exceptions are cases where the SNR valuesin the upper atmosphere are too low to support successful profile inversion. The results ofthe inversion were most consistent for the data from the dust transport events in Section5.1.1696.2. Plans for future research and development6.1.3 Layer masks applied to long-term analysis of aerosol trendsAnother important contribution of the LDA (and other similar algorithms) to the study ofaerosols is the additional functionality of being able to work with much larger sets of data.This is due to the increased level of automation that can be implemented as a result ofhaving clearly defined layer boundaries. For example, in the analysis of the smoke eventsof July and August 2015 in Section 5.3 daily box plots of depolarization ratios for layersidentified “smoke/urban” or “polluted dust” below 3 km were calculated for 32 separatedays. A similar process could be used to investigate average aerosol types, altitudes, andoptical properties over the course of months or years.6.2 Plans for future research and development6.2.1 Software developmentIn its current form, the software used to implement the LDA is in a prototype stage andhas been developed with essentially one user in mind: the author. The code in its currentform, although functional, is neither robust enough nor user-friendly enough to be used bymultiple people on various platforms without inevitably encountering difficulties. Perhapsthe most important step to be taken to increase the value of the LDA to the broadercommunity is to develop the software to a point where it can be used by a wider rangeof people. This process would include bundling the various software tools into a singlepackage, the implementation of improved error detection and reporting schemes, a moreintuitive user interface, and a set of written operating procedures.6.2.2 Study incorporating atmospheric soundingThe work on the LDA thus far has revealed that the majority of erroneous layer designationsfall into two basic categories. The first involves differentiating between optically thin cirrusclouds and layers of dusty aerosols, and the second involves properly identifying regionsof water, ice, and mixed-phase clouds. Both of these issues can be further investigated byco-locating a mMPL with a radiosonde, or other atmospheric sounding instrument, andincorporating profiles of temperature and relative humidity into a study aimed at refininghow the LDA makes these layer identifications.1706.2. Plans for future research and development6.2.3 ”Coastal” SettingOne of the first improvements planned for the LDA will be to develop and add an optional“coastal” setting. When this setting is employed, aerosols found with depolarizationsbelow 0.05 would be designated “marine” and these layers would be assigned a LIDARratio of 23. Marine aerosols only occur over land within a short range of the ocean andthey also occur almost exclusively in the lower atmosphere. In order to avoid using thissetting in inappropriate locations, and to avoid mislabelling higher altitude aerosols withthis designation, a survey of CALIPSO Vertical Feature Mask (VFM) results in coastallocations will be used to determine appropriate ranges for distance from the coast andaltitude.6.2.4 Long-term comparison to CALIPSO VFMThe most well-known example of feature masking for LIDAR data is of course the CALIPSOVFM. A very useful exercise would be to collect data from the VFM as it passed over oneor more locations where mMPL 17 LIDARs were operating to perform a cross-validation ofthe results. The CALIPSO satellite is in the “A-Train” of earth observing satellites, whichmeans it has a 90 min orbit and a 16 day revisit time. With a radius of 10 km aroundeach ground-based instrument, after a year of data collection a sizeable data set could becollected with which to compare the results of the LDA to the VFM. The purpose of thiswould be to verify the layer designations of the VFM against a well-established standardand possibly to find variations in performance in different locations, at different altitudes,or for different layer or cloud types. This would be especially useful for results measuredeither above or on the upper edges of clouds where a downward looking LIDAR wouldhave a much clearer view. It would also be a great help in developing the aforementionedcoastal setting where a category is added for marine aerosols in the LDA.6.2.5 Improving thresholds and PBL edge detectionThe threshold settings developed for the LDA, although well-researched and fully func-tional, have been developed over a relatively short period of time and a under fairly lim-ited circumstances. Now that the algorithm is in place, as data continue to be collectedthey can serve as a basis for continued improvement of the algorithm itself. This includes17Technically, this study could be performed for most LIDAR systems as long as they collected therequisite elastic backscatter and depolarization information1716.2. Plans for future research and developmentthresholds at every level of the process, including those dealing with aerosol and cloudsub-type designations, separation of clouds and aerosols, and SNR limits. In addition tothis, the process currently used for determining the upper edge of the PBL can be refinedthrough a similar process of iteration.6.2.6 Iterative LIDAR ratio optimizationIn the current version of the LDA, once initial LIDAR ratios are assigned to each regionof a profile, a profile of backscatter and extinction coefficients is calculated using theKlett inversion process for profiles with variable LIDAR ratios (see Section 4.1.4). Thisinversion can be performed directly on the normalized relative backscatter profile withno requirement for absolute calibration, but it does hinge on the assumption that thetopmost portion of the profile is always clear air free of aerosols where molecular scatteringdominates, and as shown in Section 4.2.3 it is adversely affected by large amounts of noisetoward the top of the profile. This is a stable inversion in that errors resulting fromnoise effects and LIDAR ratio uncertainty do not tend to increase exponentially as onesteps through the profile, but the accuracy of its results do depend on the accuracy of theLIDAR ratios, and there is a large amount of uncertainty in these values (as is shown inTables 2.1, 2.2, 2.3, and 2.4) The result is that, although the use of the current layer maskresults in an improvement over a single blanket LIDAR ratio value applied to the entireprofile, it is still prone to errors as a result of uncertainty in the assigned LIDAR ratio (formore on this, see Section 4.2.1).In future versions of the LDA an optimization process is proposed where each layer isexamined one by one starting from the top of the PBL in an attempt to refine the initialLIDAR ratios to better match the observed extinction of the layer.Layers that are bounded on both sides by regions of aerosol-free molecular scatteringare called “simple layers”. The optimization process for these kinds of layers is relativelystraightforward. When layers overlap each other, or are embedded within other layers,these are classified as “complex layers”. In these cases, finding an unique solution becomesdifficult or impossible and assumptions or constraints are introduced to limit the solutionspace. Although this process allows for a degree of optimization in some cases, in manyinstances the process still fails to converge on a value that is physically reasonable and theinitial LIDAR ratio guesses must be used, resulting in an increase in uncertainty for thebackscatter and extinction coefficient profiles.1726.2. Plans for future research and developmentSimple layersFor layers that are bound both above and below by regions of molecular scattering, New-ton’s method is used to refine the initial LIDAR ratio. The premise is that the amount ofextinction across a given range due to purely molecular scattering can be calculated usinga measured or assumed profile of atmospheric temperature and pressure. Thus, any addi-tional reduction in signal beyond this is assumed to be due to the intervening layer. Theintegrated optical depth of the layer can be calculated in two ways. The first method usesthe measured normalized relative backscatter from the bottom and top molecular layers,as in Equations 6.1 - 6.4The normalized relative backscatter for molecular scattering above and below the layer isaveraged over 2N values to reduce noiseBm = C(1)2Nrm+N∑r′=rm−Nβmol(r′)T r′0 (6.1)The ratio of these two averaged values is proportional to the extinction of the layer betweenthemBtopBbot=βm(r2)Tr20βm(r1)Tr00(6.2)=βm(r2)βm(r1)T r2r1 (6.3)Solving for the layer extinctionT r2r1 =BtopBbotβm(r1)βm(r2)(6.4)The second method uses the backscatter profile of the layer itself combined with the1736.2. Plans for future research and developmentestimated LIDAR ratio as in Equation 6.5T r2r1 = exp−2r2∫r′=r1[αpart(r′) + αm(r′)]dr′= exp−2r2∫r′=r1[L0βpart(r′) + Lmβm(r′)]dr′ (6.5)If the assumption that the bounding top and bottom layers are dominated by molecularscattering is valid, and the estimated LIDAR ratio is correct, Equations 6.4 and 6.5 shouldgive the same result to within a reasonable tolerance. By equating them, it is a simplething to use these two values in an iterative root-finding scheme to determine the LIDARratio that provides the correct optical depth for the layer.Complex layersComplex layers are those that are bounded on one or more sides by other aerosol orcloud layers. In such circumstances, it is no longer possible to perform the relativelystraightforward Newton’s method-based optimization used on simple layers because thebackscatter strength of the bounding layers is not known beforehand. In these cases, it isnot possible to perform the relatively straightforward process of root finding used for simplelayers, but in certain cases, it is possible to apply constraints that allow optimization toproceed.Marine aerosol layers In the event that the “coastal setting” is implemented in combi-nation with this optimization scheme, it is assumed that layers with mean depolarizationratios below 0.05 could fall into either the category of marine aerosols or the previouslydefined “smoke/urban” category. In the case where one layer within a complex structurehas a depolarization ratio below this threshold, and the nature of the layer is indetermi-nate based on the depolarization ratio alone, the LIDAR ratio for the other layers is heldconstant and the LIDAR ratio of the marine/smoke layer is varied in a process similar tothat of a simple layer. The reason for this is the difference in LIDAR ratio between the twopotential types is so large (55 sr for “smoke/urban” vs 15 sr for marine aerosols) that evenif the other layers in the structure introduce some degree of error, it should be possible tocorrectly distinguish between smoke and marine aerosols in this way.1746.2. Plans for future research and developmentMixed cloud layers It is common to encounter cloud layers where some portions ofthe cloud are frozen and others are not. For cloud layers that are identified as predomi-nantly water or ice, the LIDAR ratio is more certain than for mixed ice and water layers.Therefore, in such instances, the optimization process operates on the assumption that theLIDAR ratios for areas identified as water or ice can be held constant and only mixed cloudlayers are varied. It is thought that in this way, the largest errors in LIDAR ratio will becorrected, providing the least total deviation in the resulting backscatter and extinctionprofiles.For complex layers that do not fall into one of the above categories, the initial LIDARratios are not altered. In such cases, the layers that cannot be optimized and all higherlayers in the profile are marked with a flag indicating their reduced level of certainty.Particulate depolarization ratiosPart of the process of this iterative refinement of extinction and LIDAR ratio would bethe calculation of particulate depolarization ratios for layers based upon the results of theinversion at each step of the iterative algorithm. Optically thin layers can exhibit erroneousdepolarization ratios as a result of the influence of molecular scattering, so based upon thecalculated aerosol backscatter for a given layer, the depolarization ratio for the particulatematter alone can be estimated. This would provide a more accurate depolarization ratiodata product and could even lead to the re-classification of some layers.6.2.7 Examples of potential future applicationsThe applications for the LDA are nearly unlimited, but what follows is just a few specificexamples of studies for which it could be particularly useful.Detection and monitoring of pollution or volcanic ash plumesA LIDAR used to monitor the emissions of any source of pollution, including volcanicemissions, would clearly benefit from the capability provided by the LDA to isolate andanalyse the optical properties of layers in the profile associated with the emissions inquestion. If used in combination with other instruments, the benefits would be compoundedas the LDA results could be used as a spatio-temporal filter for the other instruments toensure their readings corresponded with the layers of interest.1756.3. Closing remarksMonitoring long-term trends in aerosol loading at a given locationA LIDAR situated in a location of interest for a particular kind of aerosol activity (e.g.monitoring long-range transport of Saharan dust in the Bahamas, or urban and industrialemissions in a city in eastern China) could employ the LDA to generate statistical dataover a period of months or even years to look for trends in the occurrence of a particularlayer type. Measured quantities could include frequency, altitude, optical depth, and timeof day among others.6.3 Closing remarksAlthough there is room for further development, the LDA in its current form has beenshown to be robust enough to accurately identify various kinds of layers and clouds undera variety of circumstances, and for two different LIDAR systems. The resulting layer masksadd value to aerosol investigations in a number of different ways. As a method of displayingLIDAR results, they provide a clearer, more easily interpreted image than normalized rela-tive backscatter or depolarization ratio plots alone. As a basis for inversion of LIDAR data,the layer-specific LIDAR ratios provide additional precision in the calculation of backscat-ter and extinction coefficients and associated quantities such as particulate depolarizationratios and AOD. As a tool for investigating layer properties, the layer masks are a handytool for selecting specific layers of interest for deeper investigation. Finally, in the analysisof long-term trends, the LDA provides a tool for compiling statistics on quantities suchas layer height, composition, frequency of occurrence and layer optical properties. Despitelimitations arising from single-wavelength operation and a great deal of room for futuredevelopment, the LDA in its current form produces valuable analysis products that can beuseful to investigations of aerosol transport events as well as long-term analyses .These improvements combined with the added value that comes along with effective (ifimperfect) layer filtering make the LDA a worthwhile endeavour and a useful tool, but itis in no way a substitute for thoughtful and well-informed analysis of LIDAR results. Itslimitations are important to understand and proper judgement must be employed wheninterpreting the results.176Bibliography[1] Ackermann, J. (1998). The extinction-to-backscatter ratio of tropospheric aerosol: Anumerical study. Journal of atmospheric and oceanic technology, 15(4):1043–1050.[2] Anderson, T. L., Masonis, S. J., Covert, D. S., Charlson, R. J., and Rood, M. J. (2000).In situ measurement of the aerosol extinction-to-backscatter ratio at a polluted continen-tal site. Journal of Geophysical Research: Atmospheres (1984–2012), 105(D22):26907–26915.[3] ANSI Z136.1-1986 (1986). American National Standard for the Safe Use of Lasers.[4] Ansmann, A., Wandinger, U., Riebesell, M., Weitkamp, C., and Michaelis, W. (1992).Independent measurement of extinction and backscatter profiles in cirrus clouds by usinga combined raman elastic-backscatter lidar. Applied Optics, 31(33):7113–7131.[5] Atmosphere, U. S. (1976). Nasa tm-x 74335. national oceanic and atmospheric admin-istration. National Aeronautics and Space Administration and United States Air Force.[6] Batchvarova, E., Steyn, D., Cai, X., Gryning, S.-E., and Baldi, M. (1997). Modellingthe internal boundary layer over the lower fraser valley, british columbia. In EURASAPWorkshop, pages 141–144.[BC-MOE] BC-MOE. Bc air data archive website. http://envistaweb.env.gov.bc.ca.Accessed: 2015-09-05.[BC-Wildfire-Service] BC-Wildfire-Service. Coastal fire centre: All current wild-fires. http://bcwildfire.ca/hprScripts/WildfireNews/Fires.asp?Mode=normal&AllFires=1&FC=2. Accessed: 2015-07-17.[9] Bohren, C. F. and Huffman, D. R. (1983). Absorption and scattering of light by smallparticles. John Wiley & Sons.177Bibliography[10] Bowman, D. M., Balch, J. K., Artaxo, P., Bond, W. J., Carlson, J. M., Cochrane,M. A., DAntonio, C. M., DeFries, R. S., Doyle, J. C., Harrison, S. P., et al. (2009). Firein the earth system. science, 324(5926):481–484.[11] Brooks, I. M. (2003). Finding boundary layer top: Application of a wavelet covariancetransform to lidar backscatter profiles. Journal of Atmospheric and Oceanic Technology,20(8):1092–1105.[12] Burton, S., Ferrare, R., Vaughan, M., Omar, A., Rogers, R., Hostetler, C., and Hair,J. (2013). Aerosol classification from airborne hsrl and comparisons with the calipsovertical feature mask. Atmospheric Measurement Techniques, 6(5):1397–1412.[13] Carswell, A. and Pal, S. (1980). Polarization anisotropy in lidar multiple scatteringfrom clouds. Applied optics, 19(24):4123–4126.[14] Cattrall, C., Reagan, J., Thome, K., and Dubovik, O. (2005). Variability of aerosol andspectral lidar and backscatter and extinction ratios of key aerosol types derived from se-lected aerosol robotic network locations. Journal of Geophysical Research: Atmospheres(1984–2012), 110(D10).[15] Chen, W.-N., Chiang, C.-W., and Nee, J.-B. (2002). Lidar ratio and depolarizationratio for cirrus clouds. Applied Optics, 41(30):6470–6476.[16] Chin, M., Diehl, T., Ginoux, P., and Malm, W. (2007). Intercontinental transport ofpollution and dust aerosols: implications for regional air quality. Atmospheric Chemistryand Physics, 7(21):5501–5517.[17] Cho, H.-M., Yang, P., Kattawar, G. W., Nasiri, S. L., Hu, Y., Minnis, P., Trepte,C., and Winker, D. (2008). Depolarization ratio and attenuated backscatter for ninecloud types: analyses based on collocated calipso lidar and modis measurements. Opticsexpress, 16(6):3931–3948.[18] Christensen, J. (1997). The danish eulerian hemispheric model - a three-dimensionalair pollution model used for the arctic. Atmospheric Environment, 31(24):4169–4191.[19] Clarke, A., Shinozuka, Y., Kapustin, V., Howell, S., Huebert, B., Doherty, S., Ander-son, T., Covert, D., Anderson, J., Hua, X., Moore, K., McNaughton, C., Carmichael,G., and Weber, R. (2004). Size distributions and mixtures of dust and black carbon178Bibliographyaerosol in asian outflow: Physiochemistry and optical properties. Journal of GeophysicalResearch-Atmospheres, 109(D15):D15S09.[Cohn and Angevine] Cohn, S. A. and Angevine, W. M. Boundary layer height and en-trainment zone thickness measured by lidars and wind-profiling radars. Journal of Ap-plied Meteorology, 39(8):1233–1247.[21] Cottle, P., Mueller, D., Shin, D.-H., Zhang, X. X., Feng, G., McKendry, I., and Straw-bridge, K. (2013). Studying taklamakan aerosol properties with lidar (stapl). In SPIERemote Sensing, pages 88940X–88940X. International Society for Optics and Photonics.[22] Cottle, P., Strawbridge, K., and McKendry, I. (2014). Long-range transport of siberianwildfire smoke to british columbia: Lidar observations and air quality impacts. Atmo-spheric Environment, 90:71–77.[23] Cottle, P., Strawbridge, K., McKendry, I., O’Neill, N., and Saha, A. (2012). Apervasive and persistent asian dust event over north america during spring 2010: li-dar and sunphotometer observations. Atmospheric Chemistry and Physics Discussions,12(11):30589–30618.[24] Dawson, K., Meskhidze, N., Josset, D., and Gasso´, S. (2014). A new study of sea sprayoptical properties from multi-sensor spaceborne observations. Atmospheric Chemistry &Physics Discussions, 14:213–244.[25] DiNenno, P. J. (2008). SFPE handbook of fire protection engineering. SFPE.[26] Dubinsky, R., Carswell, A., and Pal, S. (1985). Determination of cloud microphys-ical properties by laser backscattering and extinction measurements. Applied optics,24(11):1614–1622.[27] Dubovik, O., Holben, B., Eck, T., Smirnov, A., Kaufman, Y., King, M., Tanre, D., andSlutsker, I. (2002). Variability of absorption and optical properties of key aerosol typesobserved in worldwide locations. Journal of the Atmospheric Sciences, 59(3):590–608.[28] Edlen, B. (1953). The dispersion of standard air. Josa, 43(5):339–344.[29] Edle´n, B. (1966). The refractive index of air. Metrologia, 2(2):71.179Bibliography[30] Fairlie, T. D., Jacob, D. J., and Park, R. J. (2007). The impact of transpacific transportof mineral dust in the united states. Atmospheric Environment, 41(6):1251–1266.[31] Fernald, F. G. (1984). Analysis of atmospheric lidar observations- some comments.Applied optics, 23(5):652–653.[32] Fischer, E. V., Hsu, N. C., Jaffe, D. A., Jeong, M. J., and Gong, S. L. (2009). A decadeof dust: Asian dust and springtime aerosol load in the us pacific northwest. GeophysicalResearch Letters, 36:L03821.[33] Fischer, E. V., Perry, K. D., and Jaffe, D. A. (2011). Optical and chemical propertiesof aerosols transported to mount bachelor during spring 2010. Journal of GeophysicalResearch-Atmospheres, 116:D18202.[34] Flynn, C. J., Mendoza, A., Zheng, Y., and Mathur, S. (2007). Novel polarization-sensitive micropulse lidar measurement technique. Optics Express, 15(6):2785–2790.[Giglio] Giglio, L. Global fire emissions database. http://www.globalfiredata.org/index.html. Accessed: 2012-09-05.[36] Giglio, L., Randerson, J. T., and Werf, G. R. (2013). Analysis of daily, monthly, andannual burned area using the fourth generation global fire emissions database (gfed4).Journal of Geophysical Research: Biogeosciences.[37] Gong, S. L., Zhang, X. Y., Zhao, T. L., Zhang, X. B., Barrie, L. A., McKendry, I. G.,and Zhao, C. S. (2006). A simulated climatology of asian dust aerosol and its trans-pacific transport. part ii: Interannual variability and climate connections. Journal ofClimate, 19(1):104–122.[38] Groß, S., Tesche, M., Freudenthaler, V., Toledano, C., Wiegner, M., Ansmann, A., Al-thausen, D., and Seefeldner, M. (2011). Characterization of saharan dust, marine aerosolsand mixtures of biomass-burning aerosols and dust by means of multi-wavelength depo-larization and raman lidar measurements during samum 2. Tellus B, 63(4):706–724.[39] Hansen, J. and Nazarenko, L. (2004). Soot climate forcing via snow and ice albe-dos. Proceedings of the National Academy of Sciences of the United States of America,101(2):423–428.180Bibliography[40] Hayden, K., Anlauf, K., Hoff, R., Strapp, J., Bottenheim, J., Wiebe, H., Froude, F.,Martin, J., Steyn, D., and McKendry, I. (1997). The vertical chemical and meteorologicalstructure of the boundary layer in the lower fraser valley during pacific’93. AtmosphericEnvironment, 31(14):2089–2105.[41] Heald, C. L., Jacob, D. J., Park, R. J., Alexander, B., Fairlie, T. D., Yantosca, R. M.,and Chu, D. A. (2006). Transpacific transport of asian anthropogenic aerosols and itsimpact on surface air quality in the united states. Journal of Geophysical Research-Atmospheres, 111(D14):D14310.[42] Hennemuth, B. and Lammert, A. (2006). Determination of the atmospheric bound-ary layer height from radiosonde and lidar backscatter. Boundary-Layer Meteorology,120(1):181–200.[43] Hobbs, P. V., Reid, J. S., Kotchenruther, R. A., Ferek, R. J., and Weiss, R. (1997).Direct radiative forcing by smoke from biomass burning. Science, 275(5307):1777–1778.[44] Hu, Y., Vaughan, M., Liu, Z., Lin, B., Yang, P., Flittner, D., Hunt, B., Kuehn, R.,Huang, J., Wu, D., et al. (2007). The depolarization-attenuated backscatter relation:Calipso lidar measurements vs. theory. Optics Express, 15(9):5327–5332.[45] Hu, Y., Winker, D., Vaughan, M., Lin, B., Omar, A., Trepte, C., Flittner, D., Yang,P., Nasiri, S. L., Baum, B., et al. (2009). Calipso/caliop cloud phase discriminationalgorithm. Journal of Atmospheric and Oceanic Technology, 26(11):2293–2309.[46] Huang, J., Fu, Q., Su, J., Tang, Q., Minnis, P., Hu, Y., Yi, Y., and Zhao, Q. (2009).Taklimakan dust aerosol radiative heating derived from calipso observations using thefu-liou radiation model with ceres constraints. Atmospheric Chemistry and Physics,9(12):4011–4021.[47] Huang, Z., Huang, J., Bi, J., Wang, G., Wang, W., Fu, Q., Li, Z., Tsay, S.-C., and Shi,J. (2010). Dust aerosol vertical structure measurements using three mpl lidars during2008 china-us joint dust field experiment. Journal of Geophysical Research-Atmospheres,115:D00K15.[48] Hunter, J. D. (2007). Matplotlib: A 2d graphics environment. Computing In Science& Engineering, 9(3):90–95.181Bibliography[49] Husar, R. B., Tratt, D., Schichtel, B. A., Falke, S., Li, F., Jaffe, D., Gasso, S., Gill,T., Laulainen, N. S., Lu, F., et al. (2001). Asian dust events of april 1998. Journal ofGeophysical Research: Atmospheres (1984–2012), 106(D16):18317–18330.[50] Jaffe, D., Bertschi, I., Jaegle´, L., Novelli, P., Reid, J. S., Tanimoto, H., Vingarzan, R.,and Westphal, D. L. (2004). Long-range transport of siberian biomass burning emissionsand impact on surface ozone in western north america. Geophysical Research Letters,31(16).[51] Jaffe, D., Price, H., Parrish, D., Goldstein, A., and Harris, J. (2003). Increasingbackground ozone during spring on the west coast of north america. Geophysical ResearchLetters, 30(12):1613.[52] Jalava, P. I., Salonen, R. O., Ha¨linen, A. I., Penttinen, P., Pennanen, A. S., Sillanpa¨a¨,M., Sandell, E., Hillamo, R., and Hirvonen, M.-R. (2006). In vitro inflammatory andcytotoxic effects of size-segregated particulate samples collected during long-range trans-port of wildfire smoke to helsinki. Toxicology and applied pharmacology, 215(3):341–353.[53] Jickells, T., An, Z., Andersen, K. K., Baker, A., Bergametti, G., Brooks, N., Cao, J.,Boyd, P., Duce, R., Hunter, K., et al. (2005). Global iron connections between desertdust, ocean biogeochemistry, and climate. science, 308(5718):67–71.[54] Kaufman, Y. J. and Fraser, R. S. (1997). The effect of smoke particles on clouds andclimate forcing. Science, 277(5332):1636–1639.[55] Kaufman, Y. J., Tanre´, D., and Boucher, O. (2002). A satellite view of aerosols in theclimate system. Nature, 419(6903):215–223.[56] Kim, S.-W., Yoon, S.-C., Kim, J., and Kim, S.-Y. (2007). Seasonal and monthly vari-ations of columnar aerosol optical properties over east asia determined from multi-yearmodis, lidar, and aeronet sun/sky radiometer measurements. Atmospheric Environment,41(8):1634–1651.[57] Klett, J. D. (1985). Lidar inversion with variable backscatter/extinction ratios. AppliedOptics, 24(11):1638–1643.[58] Kovalev, V. A. (1993). Lidar measurement of the vertical aerosol extinction profileswith range-dependent backscatter-to-extinction ratios. Applied optics, 32(30):6053–6065.182Bibliography[59] Kovalev, V. A. (1995). Sensitivity of the lidar solution to errors of the aerosolbackscatter-to-extinction ratio: Influence of a monotonic change in the aerosol extinctioncoefficient. Applied optics, 34(18):3457–3462.[60] Kuhlmann, J. and Quaas, J. (2010). How can aerosols affect the asian summer mon-soon? assessment during three consecutive pre-monsoon seasons from calipso satellitedata. Atmospheric Chemistry and Physics, 10(10):4673–4688.[61] Larkin, N. K., O’Neill, S. M., Solomon, R., Raffuse, S., Strand, T., Sullivan, D. C.,Krull, C., Rorig, M., Peterson, J., and Ferguson, S. A. (2010). The bluesky smokemodeling framework. International Journal of Wildland Fire, 18(8):906–920.[62] Li, J., Wang, Z., Zhuang, G., Luo, G., Sun, Y., and Wang, Q. (2012). Mixing ofasian mineral dust with anthropogenic pollutants over east asia: a model case study of asuper-duststorm in march 2010. Atmospheric Chemistry and Physics, 12(16):7591–7607.[63] Liu, Z., Fairlie, T. D., Uno, I., Huang, J., Wu, D., Omar, A., Kar, J., Vaughan, M.,Rogers, R., Winker, D., et al. (2013). Transpacific transport and evolution of the opticalproperties of asian dust. Journal of Quantitative Spectroscopy and Radiative Transfer,116:24–33.[64] Liu, Z., Omar, A., Vaughan, M., Hair, J., Kittaka, C., Hu, Y., Powell, K., Trepte,C., Winker, D., Hostetler, C., et al. (2008). Calipso lidar observations of the opticalproperties of saharan dust: A case study of long-range transport. Journal of GeophysicalResearch: Atmospheres (1984–2012), 113(D7).[65] Liu, Z., Winker, D., Omar, A., Vaughan, M., Trepte, C., Hu, Y., Powell, K., Sun,W., and Lin, B. (2011). Effective lidar ratios of dense dust layers over north africa de-rived from the caliop measurements. Journal of Quantitative Spectroscopy and RadiativeTransfer, 112(2):204–213.[66] Mahowald, N. M., Baker, A. R., Bergametti, G., Brooks, N., Duce, R. A., Jickells,T. D., Kubilay, N., Prospero, J. M., and Tegen, I. (2005). Atmospheric global dust cycleand iron inputs to the ocean. Global biogeochemical cycles, 19(4).[67] Mattis, I., Muller, D., Ansmann, A., Wandinger, U., Murayama, T., and Damoah,R. (2004). Siberian forest-fire smoke observed over central Europe in spring/summer2003 in the framework of earlinet. In Pappalardo, G and Amodeo, A, editor, 22ND183BibliographyINTERNATIONAL LASER RADAR CONFERENCE (ILRC 2004), VOLS 1 AND 2,volume 561 of ESA SPECIAL PUBLICATIONS, pages 857–860. ICLAS; Int CoordinatGrp Laser Atmospher Studies; Ist Metodol Analisi Ambientale; Consiglio Nazl Ric; esa;NASA. 22nd International Laser Radar Conference (ILRC 2004), Matera, ITALY, JUL12-16, 2004.[68] McKendry, I., Strawbridge, K., Karumudi, M., O’Neill, N., Macdonald, A., Leaitch,R., Jaffe, D., Cottle, P., Sharma, S., Sheridan, P., et al. (2011). Californian forestfire plumes over southwestern british columbia: lidar, sunphotometry, and mountaintopchemistry observations. Atmospheric Chemistry and Physics, 11(2):465–477.[69] McKendry, I. G., Macdonald, A. M., Leaitch, W. R., van Donkelaar, A., Zhang, Q.,Duck, T., and Martin, R. V. (2008). Trans-pacific dust events observed at whistler,british columbia during intex-b. Atmospheric Chemistry and Physics, 8(20):6297–6307.[70] McKendry, I. G., Strawbridge, K. B., O’Neill, N. T., Macdonald, A. M., Liu, P. S.,Leaitch, W. R., Anlauf, K. G., Jaegle, L., Fairlie, T. D., and Westphal, D. L. (2007).Trans-pacific transport of saharan dust to western north america: A case study. Journalof geophysical research, 112(D1):D01103.[71] Mori, I., Nishikawa, M., Tanimura, T., and Quan, H. (2003). Change in size distribu-tion and chemical composition of kosa (asian dust) aerosol during long-range transport.Atmospheric Environment, 37(30):4253–4263.[72] Morille, Y., Haeffelin, M., Drobinski, P., and Pelon, J. (2007). Strat: An automatedalgorithm to retrieve the vertical structure of the atmosphere from single-channel lidardata. Journal of Atmospheric and Oceanic Technology, 24(5):761–775.[73] Mu¨ller, D., Ansmann, A., Mattis, I., Tesche, M., Wandinger, U., Althausen, D., andPisani, G. (2007a). Aerosol-type-dependent lidar ratios observed with raman lidar. Jour-nal of Geophysical Research: Atmospheres (1984–2012), 112(D16).[74] Mu¨ller, D., Mattis, I., Ansmann, A., Wandinger, U., Ritter, C., and Kaiser, D. (2007b).Multiwavelength Raman lidar observations of particle growth during long-range trans-port of forest-fire smoke in the free troposphere. GEOPHYSICAL RESEARCH LET-TERS, 34(5).184Bibliography[75] Mu¨ller, D., Wagner, F., Wandinger, U., Ansmann, A., Wendisch, M., Althausen, D.,and von Hoyningen-Huene, W. (2000). Microphysical particle parameters from extinctionand backscatter lidar data by inversion with regularization: experiment. Applied optics,39(12):1879–1892.[76] Murayama, T., Muller, D., Wada, K., Shimizu, A., Sekiguchi, M., and Tsukamoto, T.(2004). Characterization of Asian dust and Siberian smoke with multiwavelength Ra-man lidar over Tokyo, Japan in spring 2003. GEOPHYSICAL RESEARCH LETTERS,31(23).[77] NASA, U. (1976). standard atmosphere, 1976. US Government Printing Office, Wash-ington DC.[NRL] NRL. Nrl/monterey aerosol page. http://www.nrlmry.navy.mil/aerosol. Ac-cessed: 2015-09-05.[79] Omar, A. H., Winker, D. M., Kittaka, C., Vaughan, M. A., Liu, Z., Hu, Y., Trepte,C. R., Rogers, R. R., Ferrare, R. A., Lee, K.-P., Kuehn, R. E., and Hostetler, C. A.(2009). The calipso automated aerosol classification and lidar ratio selection algorithm.Journal of Atmospheric and Oceanic Technology, 26(10):1994–2014.[80] Pal, S. and Carswell, A. (1976). Multiple scattering in atmospheric clouds: lidarobservations. Applied Optics, 15(8):1990–1995.[81] Parry, M. L. (2007). Climate Change 2007: Impacts, Adaptation and Vulnerability:Working Group II Contribution to the Fourth Assessment Report of the IPCC Intergov-ernmental Panel on Climate Change, volume 4. Cambridge University Press.[82] Penndorf, R. (1957). Tables of the refractive index for standard air and the rayleighscattering coefficient for the spectral region between 0.2 and 20.0 µ and their applicationto atmospheric optics. JOSA, 47(2):176–182.[83] Penner, J. E., Charlson, R., Schwartz, S., Hales, J., Laulainen, N., Travis, L., Leifer,R., Novakov, T., Ogren, J., and Radke, L. (1994). Quantifying and minimizing uncer-tainty of climate forcing by anthropogenic aerosols. Bulletin of the American Meteoro-logical Society, 75(3):375–400.185Bibliography[84] Pinnick, R., Jennings, S., Chy`lek, P., Ham, C., and Grandy, W. (1983). Backscatterand extinction in water clouds. Journal of Geophysical Research: Oceans (1978–2012),88(C11):6787–6796.[85] Prospero, J. M., Ginoux, P., Torres, O., Nicholson, S. E., and Gill, T. E. (2002). Envi-ronmental characterization of global sources of atmospheric soil dust identified with thenimbus 7 total ozone mapping spectrometer (toms) absorbing aerosol product. Reviewsof Geophysics, 40(1):1002.[86] Ramanathan, V., Ramana, M. V., Roberts, G., Kim, D., Corrigan, C., Chung, C., andWinker, D. (2007). Warming trends in asia amplified by brown cloud solar absorption.Nature, 448(7153):575–578.[87] Ryder, C. L., Highwood, E. J., Rosenberg, P. D., Trembath, J., Brooke, J. K., Bart,M., Dean, A., Crosier, J., Dorsey, J., Brindley, H., Banks, J., Marsham, J. H., McQuaid,J. B., Sodemann, H., and Washington, R. (2012). Optical properties of saharan dustaerosol and contribution from the coarse mode as measured during the fennec 2011 air-craft campaign. Atmospheric Chemistry and Physics Discussions, 12(10):26783–26842.[88] Sassen, K. (1991). The polarization lidar technique for cloud research: A review andcurrent assessment. Bulletin of the American Meteorological Society, 72(12):1848–1866.[89] Sassen, K. (2002). Indirect climate forcing over the western us from asian dust storms.Geophysical Research Letters, 29(10):103–1.[90] Sassen, K. and Cho, B. S. (1992). Subvisual-thin cirrus lidar dataset for satelliteverification and climatological research. Journal of Applied Meteorology, 31(11):1275–1285.[91] Sassen, K., DeMott, P. J., Prospero, J. M., and Poellot, M. R. (2003). Saharan duststorms and indirect aerosol effects on clouds: Crystal-face results. Geophysical ResearchLetters, 30(12).[92] Sassen, K., Kayetha, V. K., and Zhu, J. (2012). Ice cloud depolarization for nadir andoff-nadir calipso measurements. Geophysical Research Letters, 39(20).[93] Sawyer, V. and Li, Z. (2013). Detection, variations and intercomparison of the plan-etary boundary layer depth from radiosonde, lidar and infrared spectrometer. Atmo-spheric Environment, 79:518–528.186Bibliography[94] Schmaltz, J. Modis earth observatory. http://eoimages.gsfc.nasa.gov/images/imagerecords/86000/86190/britishcolumbia_tmo_2015186_lrg.jpg. Ac-cessed: 2015-07-17.[95] Schmaltz, J. Modis earth observatory. http://eoimages.gsfc.nasa.gov/images/imagerecords/86000/86190/britishcolumbia_tmo_2015187_lrg.jpg. Ac-cessed: 2015-07-17.[96] Schmaltz, J. Modis image gallery. http://modis.gsfc.nasa.gov/gallery/images/image03162010_250m.jpg. Accessed: 2010-05-10.[97] Schmaltz, J. Modis image gallery. http://modis.gsfc.nasa.gov/gallery/images/image06292012_250m.jpg. Accessed: 2012-07-17.[98] Schmaltz, J. Modis image gallery. http://modis.gsfc.nasa.gov/gallery/images/image07072012_250m.jpg. Accessed: 2012-07-17.[99] Schuster, G. L., Vaughan, M., MacDonnell, D., Su, W., Winker, D., Dubovik, O.,Lapyonok, T., and Trepte, C. (2012). Comparison of calipso aerosol optical depth re-trievals to aeronet measurements, and a climatology for the lidar ratio of dust. Atmos.Chem. Phys, 12(10.5194).[100] Smirnov, A., Holben, B. N., Kaufman, Y. J., Dubovik, O., Eck, T. F., Slutsker, I.,Pietras, C., and Halthore, R. N. (2002). Optical properties of atmospheric aerosol inmaritime environments. Journal of the Atmospheric Sciences, 59(3):501–523.[101] Sorensen, C. (2001). Light scattering by fractal aggregates: a review. Aerosol Science& Technology, 35(2):648–687.[102] Spinhirne, J. D. (1993). Micro pulse lidar. Geoscience and Remote Sensing, IEEETransactions on, 31(1):48–55.[103] Steyn, D. G., Baldi, M., and Hoff, R. (1999). The detection of mixed layer depthand entrainment zone thickness from lidar backscatter profiles. Journal of Atmosphericand Oceanic Technology, 16(7):953–959.[104] Stohl, A., Eckhardt, S., Forster, C., James, P., and Spichtinger, N. (2002). On thepathways and timescales of intercontinental air pollution transport. Journal of Geophys-ical Research-Atmospheres, 107(D23):4684.187Bibliography[105] Strawbridge, K. (2013). Developing a portable, autonomous aerosol backscatter lidarfor network or remote operations. Atmospheric Measurement Techniques, 6(3):801–816.[106] Strawbridge, K. B. (2012). Developing a portable, autonomous aerosol backscatterlidar for network or remote operations. Atmospheric Measurement Techniques Discus-sions, 5(6):8609–8652.[107] Sugimoto, N., Tatarov, B., Shimizu, A., Matsui, I., and Nishizawa, T. (2010). Opticalcharacteristics of forest-fire smoke observed with two-wavelength mie-scattering lidarsand a high-spectral-resolution lidar over japan. SOLA, 6:93–96.[108] Sun, W., Liu, Z., Videen, G., Fu, Q., Muinonen, K., Winker, D. M., Lukashin, C.,Jin, Z., Lin, B., and Huang, J. (2013). For the depolarization of linearly polarizedlight by smoke particles. Journal of Quantitative Spectroscopy and Radiative Transfer,122:233–237.[109] Tang, Y., Carmichael, G., Kurata, G., Uno, I., Weber, R., Song, C., Guttikunda,S., Woo, J., Streets, D., Wei, C., Clarke, A., Huebert, B., and Anderson, T. (2004).Impacts of dust on regional tropospheric chemistry during the ACE-Asia experiment:A model study with observations. JOURNAL OF GEOPHYSICAL RESEARCH-ATMOSPHERES, 109(D19).[110] Tanimoto, H., Kajii, Y., Hirokawa, J., Akimoto, H., and Minko, N. P. (2000). Theatmospheric impact of boreal forest fires in far eastern siberia on the seasonal variation ofcarbon monoxide: Observations at rishiri, a northern remote island in japan. GeophysicalResearch Letters, 27(24):4073–4076.[111] Thulasiraman, S., O’Neill, N. T., Royer, A., Holben, B. N., Westphal, D. L., andMcArthur, L. J. B. (2002). Sunphotometric observations of the 2001 asian dust stormover canada and the us. Geophysical Research Letters, 29(8):1255.[UMBC-ALG] UMBC-ALG. Us air quality: The smog blog. http://alg.umbc.edu/usaq/archives/2015_07.html. Accessed: 2015-08-25.[113] Uno, I., Eguchi, K., Yumimoto, K., Liu, Z., Hara, Y., Sugimoto, N., Shimizu, A.,and Takemura, T. (2011). Large asian dust layers continuously reached north americain april 2010. Atmospheric Chemistry and Physics, 11(14):7333–7341.188[114] Wandinger, U., Mu¨ller, D., Bo¨ckmann, C., Althausen, D., Matthias, V., Bo¨senberg,J., Weiß, V., Fiebig, M., Wendisch, M., Stohl, A., et al. (2002). Optical and mi-crophysical characterization of biomass-burning and industrial-pollution aerosols from-multiwavelength lidar and aircraft measurements. Journal of Geophysical Research: At-mospheres (1984–2012), 107(D21):LAC–7.[115] Waterman, P. (1965). Matrix formulation of electromagnetic scattering. Proceedingsof the IEEE, 53(8):805–812.[Westphal] Westphal, D. L. Description of naaps (navy aerosol analysis and prediction sys-tem). http://www.nrlmry.navy.mil/aerosol_web/Docs/globaer_model.html. Ac-cessed: 2011-09-05.[117] Young, A. (1980). Revised depolarization corrections for atmospheric extinction.Applied Optics, 19(20):3427–3428.[118] Young, S. A. (1995). Analysis of lidar backscatter profiles in optically thin clouds.Applied optics, 34(30):7019–7031.[119] Zender, C. S., Miller, R., and Tegen, I. (2004). Quantifying mineral dust massbudgets: Terminology, constraints, and current estimates. Eos, Transactions AmericanGeophysical Union, 85(48):509–512.[120] Zhao, T. L., Gong, S. L., Zhang, X. Y., and Jaffe, D. A. (2008). Asian dust storminfluence on north american ambient pm levels: observational evidence and controllingfactors. Atmospheric Chemistry and Physics, 8(10):2717–2728.[121] Zhou, T., Huang, J., Huang, Z., Liu, J., Wang, W., and Lin, L. (2013). Thedepolarization–attenuated backscatter relationship for dust plumes. Optics express,21(13):15195–15204.189Appendix ALIDAR technical detailsThe two systems that provided the most data for the endeavour were the Mini-MicroPulseLIDAR (mMPL) and Canadian Operational Research Aerosol LIDAR Network (CORAL-Net).A.1 Mini-micropulse lidarThe mMPL is an elastic, polarization sensitive micropulse LIght Detection And Ranging(LIDAR) instrument operating at 532 nm. It is a small, portable and self-contained systemsuitable for flexible deployment in adverse conditions and on limited schedules. Dependingon the model used, the laser Pulse Repetition Frequency (PRF) of the system is either2.5 kHz or 4 kHz, with an average energy per pulse of 3 µJ to 4 µJ. The vertical resolutioncan be set by the user to 5,15,30, or 75 m. Pulses are averaged together for a time periodthat can range from 1 s up to 10 min to generate each profile 18 The mMPL is a self-contained unit, but the full system includes a weather-proof enclosure allowing the LIDARto be operated remotely in extreme weather conditions for extended periods with minimalmaintenance required. Data from multiple detection channels, and the calculated ratiosbetween them, are a key component for generating value-added analysis products fromLIDAR data. However, the calculation of such ratios from LIDAR signals is often plaguedby detector-specific non-linear effects that can alter or obscure the true ratios. The mMPLachieves polarization sensitivity without the necessity for difficult cross-detector calibrationprocedures by detecting both polarizations using a single detector.Micropulse lidar The first micropulse LIDAR system was demonstrated by James Spin-hirne in 1993. [102] The motivation behind the innovation was to provide an eye-safe alter-18In order to achieve the desired Signal to Noise Ratio (SNR) for generating analysis products, it is rareto set the averaging time to less than 30 s. This setting also avoids unnecessarily large files for longer periodsof data collection190A.1. Mini-micropulse lidarFigure A.1: Basic optical layout of the mMPL. More detail on the polarization stated atkey points along the light path is shown in Table A.1native to the LIDAR systems typically used for atmospheric research and environmentalmonitoring. Typical atmospheric LIDAR systems operate at a pulse repetition frequencyof 1 Hz to 10 Hz with pulse energies of 0.1 J to 1 J and laser beam diameters at the exitaperture of 1 cm to 2 cm, thus generating pulses that exceed American National StandardsInstitute (ANSI) eye safety regulations by several orders of magnitude. 19 The micropulseLIDAR achieves eye safety by reducing laser pulse energy to the 1 µJ to 5 µJ range and atthe same time expanding the beam diameter to 20 cm. The loss of per-pulse energy is par-tially compensated by an increase in the pulse repetition frequency to the kHz range andintegrating for multiple seconds. Clearly, this still results in a reduction in total signal irra-diance at the receiver aperture of 2 to 3 orders of magnitude from the typical atmosphericlidar. This results in the necessity for strict narrowband filtering of the input signal toreduce the impact of background radiation and the use of Avalanche Photo-Diode (APD)detectors in photon-counting mode as opposed to the more typical Photo-Multiplier Tube(PMT) detectors to boost signal sensitivity. Input from background atmospheric scatteringis further limited by reducing the receiver field of view to the µrad range. This narrow fieldof view is possible due to the reduction in laser beam angular dispersion that occurs natu-rally as a result of the expanded beam diameter due to preservation of etendue. Although19The [3] laser exposure safety standard cites the Maximum Permissible Exposure (MPE) for the 520 nmto 530 nm wavelength range as 5× 10−7 J cm−1. For pulsed sources, this is reduced by a factor of N−.25where N is the number of pulses incident on the eye. For visible light the standard states that the naturalaversion response limits exposure to 0.25 s. For example, a PRF of 4 kHz results in a count of 1000 pulsesreducing the permissible per-pulse energy to 8.89× 10−8 J cm−1191A.1. Mini-micropulse lidarthis is not a requirement of the technology, most modern Micro-Pulse LIDAR (MPL) sys-tems, including the mMPL, achieve the necessary beam expansion by utilizing a sharedinput-output light path and matching the laser beam diameter and angular dispersion tothe aperture and field of view of the receiving telescope. The mMPL combines the inputand output optical paths using a polarizing beamsplitter cube.Liquid crystal retarder The mMPL utilizes this polarizing beamsplitter in combina-tion with an actively controlled Liquid Crystal Retarder (LCR) acting as a quarter waveplate 20 on the shared input-output light path to achieve polarization sensitivity. Whenthe LCR is not activated, the system emits linearly polarized light, some portion of whichhas its polarization shifted through scattering, resulting in what amounts to a rotation ofthe polarization plane. On the return path, the polarizing beamsplitter rejects the portionof the light that is polarized parallel to the initial laser pulse, passing the depolarized com-ponent to the detector. When the LCR is activated, the system emits circularly polarizedlight. When circularly polarized light is scattered from a non-depolarizing particle (e.g. asphere), the scattered light is still circularly polarized but with the chirality shifted (i.e.from right-handed to left-handed 21) The portion of the light that undergoes polarization-altered scattering (e.g. from crystals or other elongated particles) retains the originalchirality. The combination of the two states effectively results in elliptical polarization.When this scattered light encounters the LCR, it is converted to linearly polarized lightat an angle that is shifted, similarly to the previous case, but because of the two passesthrough the LCR a net pi/2 phase shift is introduced. The result of this is as if the twopolarization axes had been transposed. Therefore, the pulse perpendicular light that ispassed through the beamsplitter to the detector in this case is actually proportional tothe pulse parallel scattering component. The steps alogn the light path and associatedpolarization states are summarized in Table A.1.Through Mueller matrix calculus, the ratio between the two quantities measured by themMPL can be converted to the linear depolarization ratio commonly used to distinguishbetween aerosol particle types (Equation A.1). Furthermore, the two measured quantities20Wave plates are optical elements made of a material that exhibits a different index of refraction fordifferent polarization states according to orthogonal “fast” and “slow” axes. In a quarter wave plate, thedifference between the indices is such that if the axis of polarization of linearly polarized incident light isat 45◦ to these axes, the resultant output light is circularly polarized. Similarly, if the incident light iscircularly or elliptically polarized, the output light is linearly polarized.21the most commonly-used convention classifies counter-clockwise rotation as right-handed and clockwiseas left-handed192A.1. Mini-micropulse lidarLCR turned off LCR turned onEmitted from systemaperture (1a)Linear, pulse parallel RH circularScattered into systemaperture (1b)Linear, combination ofparallel and perpendicu-larElliptical, combinationof RH and LHEmitted from LCR (2) Unaltered Linear, combination ofparallel and perpendicu-lar, pi/2 phase shiftedTransmitted troughbeamsplitter to detector(3)Linear, pulse perpendic-ularLinear, pulse perpendic-ular, pi/2 phase shiftedTable A.1: Table comparing polarization states at various points along the transmis-sion/detection light path (see Figure A.1). Thanks to the polarizing beamsplitter, thepolarization of light that is transmitted to the detector is always pulse perpendicular.When the activated, backscattered light is phase shifted by pi/2, effectively transposing thepolarization axes.can be combined to calculate the total and the pulse-parallel linear backscatter components(Equations A.2 and A.3).δlinear =δMPLδMPL + 1(A.1)|Btotal(θ = 0)| = |B⊥(θ = 0)|+ 2 ∗ |B⊥(θ = pi/2)| (A.2)|B‖(θ = 0)| = |B⊥(θ = 0)|+ |B⊥(θ = pi/2)| (A.3)Because they are measured with a common light path and detector, detector responsiv-ity, field of view, and overlap functions cancel out when calculating the ratio, but it doescome at a cost. Major drawbacks to this method include a reduction in temporal resolutionarising from the fact that both polarizations are not measured simultaneously and a sub-stantial loss of signal though the polarizing beamsplitter. There is also the possibility ofsome degree of contamination between polarization states as the LCR is switching betweenmodes. Studies with mMPL instruments have measured this effect to be < 1%. [34]193A.2. CORALNetA.2 CORALNetThe CORALNet is a series of autonomous, remotely operated dual wavelength elastic LI-DAR systems housed in trailers at various locations across Canada. A Q-switched Nd:YAG22 Continuum Inlite III laser operating at 1064 nm and 532 nm simultaneously with a pulserepetition rate of 10 Hz is the foundation of the system. The energy output is approxi-mately 150 mJ at 1064 nm and no more than 150 mJ at 532 nm. This is not within eyesafety requirements and in order to operate continuously without supervision, the systemmust be protected by a locked cage and is fitted with a radar interlock that shuts downthe laser when aircraft are detected nearby. The upward pointing system measures thereturn signal in three channels (1064 nm, and two polarization channels at 532 nm). Thereis also a prototype channel measuring Raman scattered light at 607 nm, but this is stillunder development. Backscatter information is collected at 3 m vertical resolution with10 sec averaging over a usable range from near ground to 18 km. Signal correction pro-cedures for each channel include subtraction of background signal from dark current andstray light, application of an overlap correction curve, and an iterative calibration process.With a combination of APD and PMT detectors for the 1064 nm and 532 nm channelsrespectively, CORALNet systems maintain a usable range of 15 km. [105] However, be-cause these two types of detector experience different non-linear residual effects toward theupper end of their dynamic range, CORALNet systems encounter difficulties in calculatingstable ratios between the two wavelength bands so these ratios are not included in anyof the analysis products. Depolarization ratio calculations are more reliable, but run intosome difficulties at the lower end of the dynamic range where background and thermalnoise effects begin to dominate the cross-polarization channel. This is largely due to thelack of cross-calibration between the channels and as a result a lower bound threshold isused to avoid spurious results.A.2.1 CORALNet dataData reported here from the CORALNet LIDARs include backscatter ratios and linearvolume depolarization ratios. Using Equation 2.20 as a basis, the backscatter ratio (γ(r))is defined as the ratio of the total measured backscatter for a given altitude to the signalexpected from purely molecular scattering.22Nd:YAG is a crystal made of yttrium, aluminium and garnet doped with triply ionized neodymium andis commonly used as a cavity crystal for visible and near-infrared lasers194A.2. CORALNetγ(r) =S0(r)1r2E0CO(r)βmol(r) exp{−2r∫r′=0αmol(r′)dr′}=[βpart(r) + βmol(r)]βmol(r)exp−2r∫r′=0αpart(r′)dr′ (A.4)The linear volume depolarization ratio (δ) is calculated as the linear ratio of the cali-brated attenuated backscatter of the two polarization channels.δ(r) = ξS⊥(r)S‖(r)(A.5)The correction factor ξ is used to correct for differences in throughput and detector sensi-tivity for the two channels. The procedure for determining this correction is described inStrawbridge [106]. The volume depolarization ratio represents the average depolarizationof all the particles and gases within the measurement volume, not just the particles.Even after calibration and background correction, the calculation of stable depolariza-tion ratios from LIDAR data involving two or more detector channels is often difficult toachieve for high altitudes and areas of low aerosol optical depth. This is due primarily tothe fact that in these regions the strength of the LIDAR signal is small compared to thebackground noise floor of the detectors, especially for the depolarized channel, resulting ina ratio that is highly unstable. Areas of low aerosol optical depth can be identified by acalibrated backscatter ratio near unity. In order to avoid reporting spurious or misleadingdepolarization ratios, a mask was generated for these regions using the backscatter ratiofrom the 1064 nm channel, which provided the cleanest signal. This mask was then appliedto the depolarization ratio maps and regions for which the aerosol optical depth was insuf-ficient to calculate a reliable depolarization ratio were masked out. These are regions forwhich the system is not sensitive enough to provide sufficient signal to calculate the depo-larization ratio. Furthermore, these areas are of sufficiently low aerosol content as to beirrelevant to the topic at hand so rather than make any assumption about them, depolar-ization ratios were simply not calculated in these areas. This is in contrast to times whereno data were taken, which appear as vertical white stripes in the CORALNet backscatterand depolarization ratio plots.195


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items