Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

The Central Pacific El Niño and its impact on weather and forest fire patterns in western North America Banholzer, Sandra 2012

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

Item Metadata

Download

Media
24-ubc_2012_fall_banholzer_sandra.pdf [ 6.86MB ]
Metadata
JSON: 24-1.0072998.json
JSON-LD: 24-1.0072998-ld.json
RDF/XML (Pretty): 24-1.0072998-rdf.xml
RDF/JSON: 24-1.0072998-rdf.json
Turtle: 24-1.0072998-turtle.txt
N-Triples: 24-1.0072998-rdf-ntriples.txt
Original Record: 24-1.0072998-source.json
Full Text
24-1.0072998-fulltext.txt
Citation
24-1.0072998.ris

Full Text

The Central Pacific El Niño and its impact on weather and forest fire patterns in western North America by Sandra Banholzer BSc. Geography, University of Zurich, Switzerland, 2010 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF Master of Science in THE FACULTY OF GRADUATE STUDIES (Geography) The University of British Columbia (Vancouver) August 2012 c© Sandra Banholzer, 2012 Abstract The El Niño/ Southern Oscillation (ENSO) is known to influence the weather in western North America through teleconnections. Several studies have established a relationship between ENSO and forest fire occurrence. However, a recently dis- covered variant of ENSO, called Central Pacific El Niño, may cause a different teleconnection and forest fire pattern. Investigating and classifying past El Niño events and their possible influence on weather and forest fire patterns in western North America from 1981-2010 was the objective of this study. The analysis revealed that current El Niño classification methods are subopti- mal and that a binary distinction leads to misclassification of events. It, however, confirms that the two types show a different warming pattern as well as differ- ent wind and precipitation patterns. These characteristics of the Central Pacific El Niño can cause different extra tropical teleconnections in western North America than the canonical El Niño. Variation of teleconnections within the events and the limited amount of events, however, complicate a clear conclusion. Further, other oscillations such as the Arctic Oscillation play a major role in impacting the climate in western North America. Exploratory analysis of natural forest fires of North America identified hot spots of annual area burned in central Alaska, north-west and central Canada and western United States. Further, singular value decomposition and spatial correla- tion analysis revealed a different teleconnection response in summer drought pat- terns over western North America related to the two types of El Niño. The drought pattern is significantly related with forest fire frequency and area burned in certain regions across western North America. A clear connection between the different El Niño types and the forest fire pattern however remains inconclusive. ii Table of contents Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii Table of contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii List of tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi List of figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.1 Impetus for study . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Project overview . . . . . . . . . . . . . . . . . . . . . . . . . . 3 2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 2.1 Sea surface temperature data . . . . . . . . . . . . . . . . . . . . 4 2.2 Weather data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 2.3 Forest fire data . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 3 Identification of Central Pacific El Niño events: A critique of current methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 3.1 Introduction and background . . . . . . . . . . . . . . . . . . . . 7 3.1.1 Emergence of a new type of El Niño . . . . . . . . . . . . 7 3.1.2 Current classification methods and associated problems . . 8 iii 3.2 Research objectives . . . . . . . . . . . . . . . . . . . . . . . . . 11 3.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 3.4.1 Inter-annual sea surface temperature variation in the equa- torial Pacific . . . . . . . . . . . . . . . . . . . . . . . . 14 3.4.2 Identification of Central Pacific and Eastern Pacific El Niño events by sea surface temperature based indices . . . . . . 15 3.4.3 Case study of the Central Pacific El Niño event in 2002/03 22 3.4.4 Extended analysis of identified Central Pacific and Eastern Pacific El Niño events . . . . . . . . . . . . . . . . . . . 24 3.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 4 Teleconnection of Central Pacific and Eastern Pacific El Niño events 38 4.1 Introduction and background . . . . . . . . . . . . . . . . . . . . 38 4.2 Research objectives . . . . . . . . . . . . . . . . . . . . . . . . . 40 4.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 4.4.1 Geopotential height . . . . . . . . . . . . . . . . . . . . . 42 4.4.2 2 m air temperature . . . . . . . . . . . . . . . . . . . . . 46 4.4.3 Precipitation . . . . . . . . . . . . . . . . . . . . . . . . 48 4.4.4 Teleconnections in southeastern British Columbia . . . . . 50 4.4.5 Possible influence of other climate oscillations . . . . . . 50 4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 5 Forest fires in North America . . . . . . . . . . . . . . . . . . . . . . 56 5.1 Introduction and background . . . . . . . . . . . . . . . . . . . . 56 5.2 Research objectives . . . . . . . . . . . . . . . . . . . . . . . . . 58 5.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 5.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 5.4.1 Forest fires history of North America from 1981-2010 . . 63 5.4.2 El Niño types and drought patterns of North America . . . 68 5.4.3 Drought and forest fire patterns in western North America 72 5.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 iv 6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 v List of tables Table 3.1 EP and CP El Niño events identified by several indices . . . . . 21 Table 3.2 EP and CP El Niño events identified by only NINO3, NINO3.4 and NINO4 indices . . . . . . . . . . . . . . . . . . . . . . . 21 Table 3.3 Selected EP El Niño and CP El Niño event years for the com- posite analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 25 Table 4.1 Selected events for the teleconnection analysis . . . . . . . . . 41 Table 5.1 Forest fire statistics of North America from 1981-2010 . . . . . 64 Table 5.2 Correlation statistics of SVD analysis between SSTA and PDSI 70 Table 5.3 Monthly forest fire frequency statistics of three selected regions along the coast of western North America . . . . . . . . . . . 73 Table 5.4 Monthly area burned statistics of three selected regions along the coast of western North America . . . . . . . . . . . . . . . 73 Table 5.5 Correlation statistics of fire frequency and area burned of three selected regions along the coast of western North America . . . 74 Table 5.6 Correlation statistics of area burned, forest fire frequency and PDSI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 Table 6.1 Suggested classification of ENSO events from 1981-2010 for teleconnection analysis . . . . . . . . . . . . . . . . . . . . . 86 vi List of figures Figure 3.1 Regions of different El Niño indices . . . . . . . . . . . . . . 8 Figure 3.2 Regions of EMI and IEMI . . . . . . . . . . . . . . . . . . . 10 Figure 3.3 Dominant modes of EOF analysis on equatorial SSTA . . . . 14 Figure 3.4 time-series of EP El Niño indices . . . . . . . . . . . . . . . 19 Figure 3.5 time-series of CP El Niño indices . . . . . . . . . . . . . . . 20 Figure 3.6 Indices during the CP El Niño event in 2002/03 . . . . . . . . 23 Figure 3.7 SSTA of the three boxes A,B and C during the CP El Niño event in 2002/03 . . . . . . . . . . . . . . . . . . . . . . . . 23 Figure 3.8 Averaged SSTA over NDJF during the CP El Niño event in 2002/03 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 Figure 3.9 Composites of seasonal SST evolution . . . . . . . . . . . . . 26 Figure 3.10 Evolution of equatorial SSTA of EP El Niño events along the equator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 Figure 3.11 Evolution of equatorial SSTA of CP El Niño events along the equator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 Figure 3.12 Equatorial precipitation anomalies during EP and CP El Niño events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 Figure 3.13 Zonal wind anomalies during EP- and CP El Niño events . . . 30 Figure 3.14 Wind composites for EP an CP El Niño events (DJF) . . . . . 31 Figure 3.15 Comparison of latitudinally averaged SST, precipitation and rain anomalies of CP and EP El Niño events . . . . . . . . . . 34 Figure 4.1 Geopotential height December, January, February (DJF) com- posite of CP- EP El Niño events . . . . . . . . . . . . . . . . 44 vii Figure 4.2 Geopotential height DJF mean composite and standard devia- tion of CP-, EP El Niño and all ENSO events . . . . . . . . . 45 Figure 4.3 2m air temperature DJF mean composite and standard devia- tion of CP-, EP El Niño and all ENSO events . . . . . . . . . 47 Figure 4.4 Accumulated precipitation DJF mean composite and standard deviation of CP-, EP El Niño and all ENSO events . . . . . . 49 Figure 4.5 Correlation between DJF GPH and AO . . . . . . . . . . . . 51 Figure 4.6 Temperature anomalies during winter 2009/10 . . . . . . . . . 53 Figure 5.1 Selected regions of western North America for focused forest fire analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 Figure 5.2 Trend of area burned and frequency by forest fires for the U.S and Canada . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 Figure 5.3 Trend of area burned and frequency by natural forest fires for the U.S and Canada . . . . . . . . . . . . . . . . . . . . . . . 65 Figure 5.4 Hot spots of area burned in North America . . . . . . . . . . 66 Figure 5.5 Hot spots of fire frequency in North America . . . . . . . . . 67 Figure 5.6 SVD analysis on PDSI and SST . . . . . . . . . . . . . . . . 68 Figure 5.7 Time series of SSTA modes of SVD analysis . . . . . . . . . 69 Figure 5.8 Spatial correlation patterns between ENSO timeseries and PDSI pattern . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 Figure 5.9 Composites of summer PDSI patterns of North America after the two ENSO types . . . . . . . . . . . . . . . . . . . . . . 72 Figure 5.10 Drought and forest fire pattern in central Alaska . . . . . . . . 75 Figure 5.11 Drought and forest fire pattern in southeastern BC . . . . . . . 76 Figure 5.12 Drought and forest fire pattern in western U.S. . . . . . . . . 77 Figure 5.13 Scatter plots of JJA area burned and PDSI . . . . . . . . . . . 78 viii Glossary AO Arctic Oscillation AMO Atlantic Multidecadal Oscillation AMIP Atmospheric Model Intercomparison Project CNFDB Canadian National Fire Database CP El Niño Central Pacific El Niño CPC Climate Prediction Centre DJF December, January, February DOE Department of Energy EMI El Niño Modoki Index ENSO El Niño/ Southern Oscillation EOF Empirical Orthogonal Function EPA United States Environmental Protection Agency GPCP Global Precipitation Climatology Project GPH Geopotential height HadISST Hadley Centre Sea Ice and Sea Surface Temperature IEMI Improved El Niño Modoki Index ix JJA June, July, August MAM March, April, May NARR North American Regional Reanalysis NCDC National Climatic Data Centre NCEP National Centers for Environmental Prediction NOAA National Oceanic and Atmospheric Administration PNA Pacific North American Pattern PDO Pacific Decadal Oscillation PDSI Palmer Drought Severity Index RDAS Regional Data Assimilation System SCF Square Covariance Fraction SST Sea Surface Temperatures SSTA Sea Surface Temperature Anomalies SVD Singular Value Decomposition TNI Trans Niño Index USGS United States Geological Service x Acknowledgments Many people assisted me in writing and completing this thesis. Special thanks goes to my supervisor Dr. Simon Donner, who inspired and accompanied me throughout the entire process. I would also like to thank my two committee members Dr. Ian McKendry and Dr. Dan Moore. The encouragement and time commitment of my committee over the last two years was crucial and I am very thankful for their support. I also wish to thank the individuals of the Climate and Coastal Ecosystem Lab- oratory. Special thanks goes to Justin Lau and Doris Leong, who, along with many online bloggers, patiently introduced me to the world of coding. Thanks goes to Sophie Weber for countless proofreading. Further thanks goes to the other fellows of the UBC Geography Department and in particular to Jenna Keane who supported me throughout and made these last two years an unforgettable experience. Furthermore, I would like to thank my family and TerreWEB, this research would not have been possible without their financial support. xi Chapter 1 Introduction 1.1 Impetus for study In the last few decades, scientists have identified a new type of El Niño. The main warming of Sea Surface Temperatures (SST) of this new type occurs in the central equatorial Pacific (Ashok et al., 2007). This type has therefore been called Central Pacific El Niño (CP El Niño) (Kao and Yu, 2009) among other names. During a canonical or Eastern Pacific El Niño (EP El Niño) the main warming of SST is placed just off the South American west coast in the eastern Pacific. The dynami- cally different CP El Niño might cause different global teleconnections compared to the teleconnections of the EP El Niño (Weng et al., 2007, 2009). Studying this new type is important because of its broad socioeconomic and environmental im- pacts. Further, it is also important because its frequency and intensity might be linked to climate change (Lee and McPhaden, 2010; Yeh et al., 2009). Recent studies have shown that using the NINO3.4 index to characterize a canonical El Niño event can lead to misclassification, as this region of the equato- rial Pacific tends to pick up warming signals from the two different El Niño types (Larkin and Harrison, 2005a,b; Weng et al., 2007). Several authors claim that past CP El Niño events have been misclassified as the canonical El Niño type (Weng et al., 2007; Larkin and Harrison, 2005a,b). This misclassification emerges from the fact that the scientific community has not yet agreed on a single way to cat- egorize an event and to differentiate between EP- and CP-El Niño events. Not 1 differentiating between two El Niño types may obscure distinct teleconnection sig- nals and possibly even cancel them out. In impact analysis it may be crucial to classify the events separately according to their different characteristics. Large-scale climate oscillations, like El Niño/ Southern Oscillation (ENSO), greatly impact climate around the globe and hence indirectly forest fire activity, as climate is a strong natural top-down driver of forest fire occurrence. This is primarily because climate determines forest conditions, which strongly relates to how susceptible a forest is to ignition (Trouet et al., 2006). The influence of climate oscillations on forest fire regimes in western North America has been demonstrated by several studies: Alaska (Duffy et al. 2005), Pacific Northwestern U.S. (Hessl et al., 2004), Southwestern U.S. (Swetnam and Betancourt, 1990). In particular, it has been determined that the high frequency ENSO phenomenon and its temporal and spatial impacts (teleconnections) have an influence on forest fire occurrence by altering preliminary fire circumstances such as the average temperature and the total amount of precipitation (Duffy et al., 2005). Forest fires play important economic, social and environmental roles (Walker et al., 2008). They have important implications regarding the carbon cycle (Netz et al., 2007), air pollution and visibility (Jaffe et al., 2008). Moreover, forest fires might increase with climate change (Flannigan et al., 2000; Swetnam and Ander- son, 2008). Understanding and consistently classifying past CP El Niño events and their associated teleconnections and possible influence on forest fire patterns in western North America is in great need, especially given that CP El Niño events may become more common in the future (Yeh et al., 2009). This knowledge can help us to understand current and future aspects associated with the CP El Niño. One objective of this research is, therefore, to analyse past ENSO types from 1981 to 2010, to investigate several classification methods (see Chapter 3) and to as- sess their different teleconnections on western North America (see Chapter 4). Analysing the forest fire pattern of North America over the last three decades as well as investigating the influence of the two types of ENSO on drought occurrence and related forest fire occurrence is another objective of this research (see Chapter 5). 2 1.2 Project overview Chapter two describes the various data sets that were used for this analysis. Chap- ters three, four and five represent the core research chapters. All research chap- ters start by introducing the background, the research objectives and the methods to then present the results and end with a discussion. The first research chapter focuses on the identification of the new CP El Niño and critiques the current clas- sification methods. The second chapter is dedicated to the different winter tele- connections of the two El Niño types. The third chapter assesses the forest fire history of whole North America, the influence of the two ENSO events on drought patterns and the possible links of drought and ENSO event on forest fire patterns along western North America. The analysis ends with general recommendations in Chapter six. 3 Chapter 2 Data The time period of this study ranges from January 1981 to December 2010. This period was selected for two reasons: First, it has been shown that during this time period CP El Niño events occurred more frequently and with increasing intensity (Lee and McPhaden, 2010; Yeh et al., 2011). Second, data quality and availability are limited before 1980. Data quality of Sea Surface Temperatures (SST) before the use of satellite-based measurements is less reliable (Vecchi and Wittenberg, 2010). In the case of forest fire records, comprehensive modern data records for both Canada and the U.S. are not available before 1980. 2.1 Sea surface temperature data For the SST analysis the Met Office Hadley Centre Sea Ice and Sea Surface Temperature (HadISST) data set (Rayner et al., 2003) was used 1. This gridded data set is a com- bination of in situ observations and satellite measurements and contains monthly reconstruction of SST. The spatial resolution is 1◦ by 1◦ in longitude and latitude. SST data are available from January 1870 until present. For the in-depth analysis of the different ENSO events, monthly surface wind (1000 hPa) and monthly mean precipitation were utilized. Both data sets have a 2.5◦ by 2.5◦ resolution and are available since 1979. Wind data were acquired from National Centers for Environmental Prediction (NCEP) Department of Energy (DOE) 1HadISST downloaded from the Met Office Hadley Centre website http://www.metoffice.gov.uk/ hadobs 4 Atmospheric Model Intercomparison Project (AMIP) II (Reanalysis 2) (Kanamitsu et al., 2002) 2. Precipitation data were obtained from Global Precipitation Clima- tology Project (GPCP) Version 2.2 combined precipitation data set (Adler et al., 2003). The GPCP combined precipitation data were developed and computed by the NASA/Goddard Space Flight Center’s Laboratory for Atmospheres as a contri- bution to the GEWEX Global Precipitation Climatology Project 3. 2.2 Weather data In order to assess possible teleconnections of different types of ENSO, historical climate variables from the North American Regional Reanalysis (NARR) data set were used (Mesinger et al., 2006) 4. The NARR high resolution data set (32 km and 29 pressure levels) is an extension of NCEP global reanalysis that focuses only on North America. The reanalysis is based on the Eta model and the Regional Data Assimilation System (RDAS). Monthly data are available since January 1979. For this analysis Geopotential height (GPH) at 500 hPa, 2 m air temperature and accumulated precipitation were extracted from the NARR data set. To incorporate the influence of other large scale climate oscillations, indices of the Arctic Oscillation (AO) 5, the Pacific Decadal Oscillation (PDO) 6, and the Pacific North American Pattern (PNA) 7 were acquired as well. 2.3 Forest fire data Forest fire records from Canada were obtained from the Canadian National Fire Database (CNFDB), which was compiled by the Canadian Forest Service from fire 2NCEP Reanalysis 2 data provided by NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, down- loaded from their Web site at http://www.esrl.noaa.gov/psd/ 3GPCP Precipitation data provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, from their Web site at http://www.esrl.noaa.gov/psd/ 4Downloaded from National Oceanic and Atmospheric Administration (NOAA) National Climatic Data Centre (NCDC) access page: http://nomads.ncdc.noaa.gov/cgi-bin/ncdc-ui/ define-collection.pl?model sys=narrmon&model name=narrmon-a&grid name=221 5Downloaded from the NOAA Climate Prediction Centre (CPC) website: http://www.cpc.ncep. noaa.gov/products/precip/CWlink/daily ao index/monthly.ao.index.b50.current.ascii 6Downloaded from http://jisao.washington.edu/pdo/PDO.latest 7Downloaded from the NOAA CPC website: http://www.cpc.ncep.noaa.gov/products/precip/ CWlink/pna/pna index.html 5 management agencies all over Canada (Canadian Forest Service, 2010) 8. For the United States, forest fire records were acquired from the Federal Fire Occurrence 9 of the United States Geological Service (USGS). The data set is also a compilation of data from several federal agencies 10 that belong to the United States Department of Interior and the United States Department of Agriculture. Forest fires records are in both cases geospatially referenced and available as shapefiles. This allows for an advanced spatial analysis in ArcGIS. Each fire incident is marked as a point in the shapefile and contains attributes such as the start date, the cause and the final area burned. The level I ecoregions of North America (Commission for Environmental Co- operation, 2009) were applied to group the forest fire incidents according to similar vegetation and climate influences 11. The ecoregions are also available in shape- files, which facilitates the spatial analysis. To determine the influence of the teleconnections and to link them with the oc- currence of forest fires, the self-calibrated Palmer Drought Severity Index (PDSI) was included in the analysis (Dai, 2011a,b). This index represents an improved version of the commonly used PDSI as it incorporates local conditions instead of a fixed coefficient proposed by Palmer and Bureau (1965), which was computed only by data from the central United States. A further improvement is the use of the more physically-based Penman-Monteith equation instead of the Thornthwaite equation for computation. These and other improvements make the self calibrated PDSI a better measurement of drought and also more comparable over greater spa- tial extent (Dai, 2011b). 8Downloaded from http://cwfis.cfs.nrcan.gc.ca/en CA/datamart 9Downloaded from: http://wildfire.cr.usgs.gov/firehistory/index.html. 10Bureau of Land Management, Bureau of Indian Affairs, U.S. Fish and Wildlife Service, National Park Service, U.S. Forest Service 11Downloaded from http://www.epa.gov/wed/pages/ecoregions/na eco.htm#Downloads 6 Chapter 3 Identification of Central Pacific El Niño events: A critique of current methods 3.1 Introduction and background 3.1.1 Emergence of a new type of El Niño El Niño is a large-scale ocean-atmosphere coupled phenomenon that recurs every 3-8 years. An El Niño event is characterized by warmer than normal SST in the eastern equatorial Pacific and is known to impact the climate around the world through teleconnections 1 (Rasmusson and Carpenter, 1982; Philander, 1990). Despite the naturally occurring variability in a large-scale phenomenon, some past El Niño events have been shown to have different characteristics than the tra- ditional El Niño phenomenon described by Rasmusson and Carpenter (1982). Re- cent studies have found that the origin and evolution of these El Niño events differ from those of the traditional El Niño (Trenberth et al., 2002; Ashok et al., 2007; Yeh et al., 2009). Sea Surface Temperature Anomalies (SSTA) of these non tra- 1Teleconnections are weather anomalies that are experienced at one place but caused by a phe- nomenon that occurs at a different place (Flannigan and Wotton, 2001). 7 ditional El Niño events show a substantially different pattern than conventional El Niño events. During a conventional El Niño, maximum warming occurs in the eastern equatorial Pacific, off the west coast of South America and spreads west- ward across the equator. During the other type of El Niño, maximum warming of SST occurs shifted west, toward the central equatorial Pacific, and then propagates eastward. Further, the positive SSTA are flanked by negative SSTA on both sides and form a so called horseshoe pattern (Ashok et al., 2007). As a consequence, these recent studies have suggested that El Niño events with an altered SSTA pattern describe a different El Niño type. According to the westward displacement of the maximum SSTA, the new type has been variously named: Dateline El Niño (Larkin and Harrison, 2005b), El Niño Modoki (Ashok et al., 2007), Warm pool El Niño (Kug et al., 2009) or CP El Niño (Kao and Yu, 2009). In this research the name CP (Central Pacific) El Niño is used for the new El Niño type and EP (Eastern Pacific) El Niño or canonical El Niño is used for the traditional type. 3.1.2 Current classification methods and associated problems The majority of ENSO classification methods consists of indices that are based on SSTA of different regions in the equatorial Pacific.  120 ° E  140 ° E  160 ° E  180 ° W  160 ° W  140 ° W  120 ° W  100 ° W   80 ° W  20 ° S  10 ° S   0 °  10 ° N  20 ° N NINO1+2 NINO4 NINO3 NINO3.4 Figure 3.1: Regions of different El Niño indices. Indices are computed by area averaging SSTA. The NINO1+2 index consists of the most spatially restrained area, which is located just off the west coast of South America. NINO3, NINO3.4 and NINO4 indices include larger areas centred around the equator but with a gradually in- creasing western extent, closer toward the international dateline (see Figure 3.1). 8 Besides purely SSTA based indices, there exists a variety of other indices, such as the Multivariate ENSO Index that incorporates variables like pressure and wind. In spite of the variety of different indices, there was no general scientific agree- ment on which index should be used to classify ENSO events and what SSTA threshold and time period should be applied to classify events. In 2003, NOAA tried to alleviate this ambiguity and released an official defi- nition that achieved U.S. consensus on how to classify an El Niño event: If SSTA in the NINO3.4 region reach an averaged value equal or greater than 0.5◦ C over three consecutive months, this phenomenon is considered a canonical El Niño or La Niña event, depending on the sign of the SSTA (NOAA, 2003). This classi- fication method has been adopted by several organizations, including the World Meteorological Organization Region IV (Larkin and Harrison, 2005a) Since the emergence and recognition of the CP El Niño event, the application of the NINO3.4 index has been questioned by several studies (Trenberth and Stepa- niak, 2001; Larkin and Harrison, 2005a; Weng et al., 2007; Kao and Yu, 2009; Li et al., 2010) as this index was not developed to distinguish between EP and CP El Niño events. As a consequence, NINO3.4 captures warming of both event types and classifies EP and CP El Niño events as a single type (Larkin and Harrison, 2005a,b; Weng et al., 2007). In order to classify CP El Niño events separately from EP El Niño events Ashok et al. (2007) developed the El Niño Modoki Index (EMI): EMI = 1.0Ta,A−0.5Ta,B−0.5Ta,C (3.1) This index is based on SSTA (Ta) of three areas in the equatorial Pacific (see Figure 3.2). The three areas (A, B, C) serve to capture the distinct characteristics of the CP El Niño with its horseshoe pattern of positive SSTA flanked by negative SSTA (Ashok et al., 2007). Besides the EMI, other indices, such as the NINO4 or Trans Niño Index (TNI) have also been used to classify CP El Niño events (Li et al., 2010). Despite the new CP El Niño indices, there is no scientific agreement on which index and what threshold should be applied to classify the two types. As a con- sequence, different studies that have scrutinized historical El Niño events, have 9  120 ° E  140 ° E  160 ° E  180 ° W  160 ° W  140 ° W  120 ° W  100 ° W   80 ° W  20 ° S  10 ° S   0 °  10 ° N  20 ° N A C B Figure 3.2: Area averaged SSTA of regions A, B and C are used to compute the EMI and the IEMI. reached different classifications depending on what index and threshold they ap- plied. Some studies defined events based on SSTA surpassing an absolute thresh- old (Kim et al., 2011), while others classified events if SSTA exceed their standard deviation (0.7 seasonal std (Ashok et al., 2007) or 1.0 overall std (Li et al., 2010)) for a certain time period. This time period also varies across studies from a certain range of months SONDJF (Yu and Kim, 2010) to JFM (Yeh et al., 2011) or DJF (Lee and McPhaden, 2010; Yoon et al., 2012)) to a total amount of months (three (Kim et al., 2011), five (Li et al., 2010) consecutive months). The variety of classification methods explains, for example, why some studies classified occurring SSTA during the winter 2006/07 as an EP El Niño event (Yeh et al., 2009; Lee and McPhaden, 2010; Kim et al., 2011; Yoon et al., 2012), whereas others did not classify these anomalies at all (Li et al., 2010; Yu and Kim, 2010). Some studies classified the SSTA during the winter in 2002/03 as an EP El Niño event (Larkin and Harrison, 2005a; Wang and Fiedler, 2006; Yoon et al., 2012) while other classified it as a CP El Niño event (Ashok et al., 2007; Yu and Kim, 2010; Kug et al., 2009). In summary, the majority of studies agrees on really strong events, such as the EP El Niño events in 1982/83 and 1997/98, but disagree on classification in some other years. Several authors confirmed this ambiguity that results from a lack of a common classification method, and claimed that different past CP El Niño events have been misclassified as the canonical El Niño (Weng et al., 2007; Larkin and Harrison, 2005a; Li et al., 2010). Li et al. (2010) compared several El Niño indices to identify past events from 10 1979-2008. They focused, in particular, on indices’ ability to differentiate between an EP- and a CP-El Niño event. They found that the NINO3 index is compara- tively better at capturing canonical events than the other three compared indices (NINO3.4, NINO1+2, NINO4). To monitor and detect past CP El Niño events, they further suggested a new index called the IEMI (see Equation 3.2). This index is based on the EMI but with slightly altered weighting of the different regions A,B and C. IEMI = 3.0Ta,A−2.0Ta,B−1.0Ta,C (3.2) According to Li et al. (2010), this weight adjustment fixed the problem of un- derestimating CP El Niño events with the EMI. They further presented that the IEMI showed significantly better results than other indices that were commonly utilized to classify CP El Niño events (TNI (Trenberth and Stepaniak, 2001) and EMI (Ashok et al., 2007)). 3.2 Research objectives As outlined in Section 3.1, capturing SSTA differences and identifying the two types of events with indices is a challenge. Moreover, depending on the index and the threshold applied, the events can be misclassified. Grouping different types together into one El Niño type can lead to misinterpreted impact analyses. This is especially important when teleconnections are analyzed, as they are thought to be substantially different depending on the El Niño type (Larkin and Harrison, 2005b; Kao and Yu, 2009; Weng et al., 2009; Yeh et al., 2009). Possibly differ- ent teleconnections can lead to different climate, ecological and societal impacts. Not distinguishing between the two types can obscure or even possibly cancel out different teleconnection signals (Larkin and Harrison, 2005a; Weng et al., 2007). Additionally, Yeh et al. (2009) has suggested that the CP El Niño has occurred more frequently in the recent decades and that it may be linked to anthropogenic induced climate change. A clear system for classifying CP El Niño events is nec- essary for interpreting the influence on climate change on the ENSO phenomenon and its teleconnections. The main purpose of this chapter is to use the analysis of SST of the last three decades to identify the best way of classifying the two different events and thus 11 explore the ability of different indices to uniquely identify EP El Niño and CP El Niño events. Further, each classified event is analysed regarding their SSTA evolu- tion, precipitation and wind anomalies to more precisely identify the strengths and weaknesses of the different classification methods. This detailed analysis aids in testing if the classified events are coherent and if the binary system of classification is justified. The advantage of this analysis is that it can include the most recent CP El Niño event in 2009/10, which is the strongest event of this type in recent records (Lee and McPhaden, 2010; Kim et al., 2011). This chapter starts with a detailed Empirical Orthogonal Function (EOF) anal- ysis of equatorial SST variation, to identify dominant spatial patterns of SSTA during the two types of events. Second, advantages and disadvantages of different ENSO indices are investigated by testing which indices best classify the domi- nant patterns correctly. This is then followed by a detailed analysis of each of the classified events. The chapter ends with a discussion of problems of a binary classification of ENSO events and a possible alternative approach. 3.3 Methods Anomalies of all investigated variables were computed by removing the monthly mean climatology, which is based on the entire study period (1981-2010). The SST data set was not de-trended (similar to other studies Ashok et al. (2007); Kao and Yu (2009) and NOAA), as statistical analysis resulted in no significant trends in the equatorial Pacific over the three decades, except in the western Pacific (results not shown here). This trend in the western equatorial Pacific has also been established by other studies (Cravatte et al., 2009). De-trending SST that do not show a trend would bias the index computations. The main methods can be described in three steps: First, a combined linear regression EOF analysis was carried out to identify the dominant spatial modes of variability of monthly SSTA in the equatorial Pacific region (20◦N-20◦S, 120◦E- 70◦W) (see Section 3.4.1). EOF is commonly used to analyze the variability of a single, scalar variable. It identifies the dominant spatial patterns and the corre- sponding temporal variation over a time period. It is important to note that the EOF analysis only extracts spatial modes from the data. These modes do not necessarily 12 represent real world modes (Bjornsson and Venegas, 1997). The combined linear regression EOF method is an expanded version of the basic EOF and was adopted for this analysis from Kao and Yu (2009). They first applied a simple EOF anal- ysis on monthly equatorial SSTA and found that the two different evolutions of ENSO events are hidden in phases of the first two modes, as previously reported by Trenberth and Stepaniak (2001). As a consequence, Kao and Yu (2009) applied a combined regression EOF method. This method aids in extracting the charac- teristic signal of each type of event without including any warming signal from the other type of event. To identify the spatial pattern of the EP El Niño (CP El Niño) each grid cell of monthly equatorial SSTA was regressed with the averaged SSTA of the NINO4 (NINO1+2) region (see Figure 3.1 for the location of the in- dices). The EOF analysis was then carried out on the residual anomalies after the regression (Kao and Yu, 2009). Subsequently, in Section 3.4.2, different ENSO indices were computed to eval- uate how well they identify years with SSTA patterns that reflect the dominant modes obtained from the combined regression EOF analysis. The focus lies on identifying warming events (El Niño) and not La Niña events. All indices except the EMI and IEMI were computed by area averaging SSTA of individual regions of the equatorial Pacific (see Figure 3.1 for the regions). The EMI and IEMI were computed according to Equations 3.1 and 3.2, as suggested by Ashok et al. (2007) and Li et al. (2010). All indices were normalized by the overall standard deviation. In addition, a binomial one quarter filter (0.25, 0.5 0.25) was applied to reduce the amount of month to month noise, in accordance with the method of Li et al. (2010). Applying this filter or different filter (e.g. running three month average) did not change the results substantially. The index computation resulted in time vectors with 360 monthly steps, which equalled the 30 year study period. To iden- tify an event, the threshold from Li et al. (2010) was applied: if the standardized indices reach one standard deviation for at least five consecutive months, this time period was classified as an event. Finally, the identified ENSO events were further evaluated regarding their ge- ographical centre of SST, their wind and their precipitation anomalies (see Section 3.4.4). This aided in evaluating the classification of the events by the different indices. Each variable was analyzed for its temporal evolution along the equator 13 from June before an ENSO event until June after the event. This analysis identified the months when the anomalies peaked. These identified months were then utilized for a composite analysis and a meridional averaged analysis. 3.4 Results The results section is split into four parts: The first part contains results of the general inter-annual variation in SST of the equatorial Pacific (see Section 3.4.1). The second part presents the index computation and their classifications outcome (see Section 3.4.2), followed by a case study (see Section 3.4.3). The final part describes the results of the extended analysis of each classified event (see Section 3.4.4). 3.4.1 Inter-annual sea surface temperature variation in the equatorial Pacific To detect the different spatial patterns of SSTA of the CP and EP El Niño, the inter-annual SST variation in the equatorial Pacific region from 1981-2010 was analysed with a combined regression EOF analysis. This analysis identified the different spatial characteristics of the two types (see Figure 3.3).   120E 180 120W 70W 20S Eq 20N −0.01 −0.005 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04   120E 180 120W 70W 20S Eq 20N −0.01 −0.005 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 Figure 3.3: Dominant modes of two combined regression EOF analyses on equatorial SSTA residuals after regression with the NINO4 (left) and the NINO1+2 (right) indices. The dominant mode after regression with the NINO4 index explains 49.6% of the residual variance in SSTA. The spatial pattern resembles the conventional EP 14 El Niño event, described by Rasmusson and Carpenter (1982) (see left panel in Figure 3.3). The main warming centre is spatially restricted to the NINO1+2 and NINO3 regions, close to the South American west coast, whereas colder SST can be found to the west (Philander, 1990). The spatial pattern resulting from the EOF analysis after regression with the NINO1+2 index explains 44.8% of the residual SSTA variance. The dominant mode shows the characteristic horseshoe pattern of the CP El Niño with the main warming centre between the dateline and 120◦W and extending arms towards the subtropics (Ashok et al., 2007). This warming centre spans the areas of the NINO4 and NINO3.4 regions. Also detectable are the colder SSTA to each side of the warming centre (see right panel in Figure 3.3). To support the spatial patterns obtained from this analysis, a cluster analysis was performed as well (following the procedure of Kao and Yu (2009)). It iden- tified similar SSTA patterns than the combined EOF analysis (results not shown), which reduces the possibility that the previously identified dominant modes are artificial statistical results of the method itself (Kao and Yu, 2009). 3.4.2 Identification of Central Pacific and Eastern Pacific El Niño events by sea surface temperature based indices As a next step, several indices were computed and analyzed for the thirty year time period in order to determine, which indices correctly identify the two dominant spatial patterns from the combined EOF analysis in Section 3.4.1. The time series of the two dominant modes (named: regEOF EP and regEOF CP) were added as an index to the commonly known indices shown in Figure 3.2 and 3.1. Using them as indices was suggested by Kao and Yu (2009) in their study. To assess the indices’ ability to capture the dominant modes, the time-series of the indices were tested for correlation with the time-series of the dominant modes. To evaluate their capacity to identify different signals, they were also tested for correlation between the indices themselves. All significance tests were carried out at the α=1% level. The NINO1+2 and NINO3 index time-series show the strongest correlation with the time series of the dominant mode that describes the EP El Niño, com- pared to correlation with other time-series. Similarly, EMI, IEMI and NINO4 all 15 showed significant positive correlations with the dominant mode that represents the CP El Niño event. This confirms that NINO1+2, NINO3 are suitable to cap- ture the EP El Niño signal and IEMI, EMI and NINO4 are suitable to capture the CP El Niño signal. The NINO3.4 index showed significant correlations with the time-series of both dominant modes, which confirms that this index tends to be in- fluenced by both types of ENSO. The correlation value of the NINO3.4 was higher with the dominant mode that represents the CP El Niño than with the mode that represents the EP El Niño event. Both correlation values between the NINO3.4 and the dominant modes were lower compared to the correlation values that other indices (NINO3, NINO1+2 etc) reached with the dominant modes. Correlation between the different indices themselves further confirmed the ex- istence of the two different events; the time-series of NINO3 and IEMI are not correlated significantly. The EMI and the IEMI display a very similar pattern, despite the different weighting in the equations. These two indices show almost perfect correlation. NINO1+2, NINO3 and the index from the combined regression EOF are tested to classify past EP El Niño events (see Figure 3.4), whereas NINO4, EMI, IEMI and the regression EOF index are tested to identify historical CP El Niño events (see Figure 3.5), according to their correlation with the dominant modes of the EOF analysis. The time-series of the NINO3.4 index is displayed in both figures, so that its behaviour can be compared with both types of indices. The EP El Niño indices reach generally higher magnitudes than the CP El Niño indices (see Figures 3.4, 3.5). For example, the two strong EP El Niño events in 1982/83 and 1997/98 clearly stand out. NINO1+2, NINO3 and the index from the combined EOF analysis reach the highest values during these two events, captur- ing well the real difference in SSTA. The NINO3.4 also peaks during these two event years but the peak magnitudes are lower. The magnitude of the index does hence not reflect the true relative difference in SST warming that occurred during these two events compared to other events identified by this index. Therefore, the NINO3.4 index did not capture the peak warming centre. This issue is amplified when looking at the magnitude of the NINO4 index during the two strongest EP El Niño events. Besides the fact that this index classified the strong EP El Niño as CP El Niño events, the index showed similar magnitudes for the strongest EP El Niño 16 events and other identified events. The IEMI and the EMI show a dip below the one standard deviation threshold line in the middle of what looks like an identified event in 1994/95, 2002/03 and 2004/05. The various indices only agree on very strong events (e.g. strong EP El Niño in 1997/98 or strong CP El Niño in 2009/10), but fail to identify weaker or unusual events (see Table 3.1). This is the case when the indices pick up signals from both types of events and group them together as one type: NINO1+2 and the index from the combined regression EOF analysis only identify the strongest EP El Niño events. NINO3.4 and NINO4 are biased by signals of both types of events. For example, NINO3.4 classifies the 2009/10 event as an EP El Niño event, despite the fact that the peak warming was centred near the International Dateline. In summary, the NINO3.4 index seems to well represent any warming that occurs in the equatorial Pacific, but it cannot distinguish very well between the two types. NINO4 index classifies all years except 1990/91 as CP El Niño years. The events in 1987/88 and 1991/92 are identified by NINO3, NINO3.4 and NINO4 but in both cases NINO3.4 reaches the highest values, which is not the case for the events in 1982/83 and 1997/98. Further, the event in 1987/88 was also identified by the index obtained from the combined regression EOF. This index also reached the threshold for four months during the event in 1991/92 and hence almost identified this year as a CP El Niño event. It is hence unclear what ENSO type the SSTA during 1987/88 and 1991/92 should be labelled. It is also noticeable that since 1990 most of the occurring El Niño events appear to be of the CP El Niño type (Kao and Yu, 2009). Table 3.2 summarizes which years are identified if a different classification method is applied. Only NINO3, NINO3.4 and NINO4 are evaluated to classify events. An event is identified according to the index that reaches one standard deviation for at least five consecutive months and shows higher values compared to the other indices. For example, all three indices reach the threshold of one standard deviation for five consecutive months during the event in 1982/83 but NINO3 reaches the highest values, which confirms that the SSTA peaked close to the South American coast. This event is hence classified as an EP El Niño. This classification method only classifies 1982/83 and 1997/98 as EP El Niño events. The ambiguity of the two events in 1987/88 and 1991/92 is confirmed by this clas- 17 sification method, these time periods reach highest values in the NINO3.4 region which means that their peak centre of warming falls between the typical peak cen- tres of EP and CP El Niño events. Only the events in 1994/95, 2002/03, 2004/05 and 2009/10 reach their highest values in the NINO4 region and are classified as CP El Niño events. The event in 2002/03 is clearly identified by the NINO4 index. This method of classification shows compared to the IEMI or EMI, no dip in the indices in the middle of an event. The event in 1990/91 was too weak to be picked up by any of these indices. 18 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 −3 −2 −1 0 1 2 3 4 S t a n d a r d  d e v i a t i o n   NINO3 NINO1+2 regEOF EP NINO3.4 Figure 3.4: Times series of different EP El Niño indices. RegEOF EP is the time-series of the combined regression EOF analysis, describing the EP El Niño. The vertical black line marks the one standard deviation threshold. An EP El Niño event is identified if an index reaches the threshold for at least five consecutive months. 19 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 −4 −3 −2 −1 0 1 2 3 S t a n d a r d  d e v i a t i o n   IEMI EMI NINO4 NINO3.4 regEOF CP Figure 3.5: Times series of different CP El Niño indices. RegEOF CP is the time-series of the combined regression EOF analysis, describing the CP El Niño. The vertical black line marks the one standard deviation threshold. A CP El Niño event is identified if an index reaches the threshold for at least five consecutive months. 20 Table 3.1: Identified years of EP and CP El Niño events. An event is classified if the indices reach the one standrad deviation threshold for at least five consecutive months. ENSO indices 82/83 87/88 90/91 91/92 94/95 97/98 02/03 04/05 09/10 EP El Niño Indices: NINO1+2 X X NINO3 X X X X EOF EP El Niño X X NINO3.4 X X X X X CP El Niño Indices: NINO4 X X X X X X X X EMI X X X* (X*) X* X IEMI X X* (X*) X* X EOF CP El Niño X X X X * Years when the IEMI and the EMI show a substantial drop below the threshold. Table 3.2: Years of EP and CP El Niño events, classified by only NINO3, NINO4 and NINO3.4. An event is identified according to the index that reaches the highest value above one standard deviation for at least five consecutive months. ENSO indices 82/83 87/88 90/91 91/92 94/95 97/98 02/03 04/05 09/10 NINO3 X X NINO3.4 X X NINO4 X X X X 21 Li et al. (2010) suggested using NINO3 and IEMI to identify EP- and CP El Niño events. While NINO3 seems to be suitable to identify EP El Niño events, the IEMI seems to have possible limitations. The CP El Niño event in 2002/03 would, for example, technically speaking not reach the definition of five consecu- tive months above one standard deviation (see Figure 3.5). In order to understand possible problems of the IEMI (and EMI), the event in 2002/03 is analysed further in section 3.4.3. 3.4.3 Case study of the Central Pacific El Niño event in 2002/03 The IEMI, as well as the EMI, fall below the one standard deviation threshold dur- ing the middle of several identified CP El Niño events (see Figure 3.5). During the event in 2002/03 the IEMI drops below the threshold from October until February (see Figure 3.6). As a consequence, this time period does not reach the threshold of five consecutive months and can strictly speaking not be identified as a CP El Niño event. All the other events that also show drops in the indices still reach the threshold for five consecutive months. The averaged SSTA from November until February 2002/03 show that positive SSTA were present in the central equatorial Pacific (see Figure 3.8) and resembled the CP El Niño pattern, which was identified by the combined regression EOF. The existence of this warming is further confirmed by the other indices (NINO3, NINO3.4 and NINO4) which record positive SSTA during this time in the equato- rial region without a dip (see Figure 3.6). The IEMI, as well as the EMI, were not able to identify this event properly. The dip in the IEMI (and EMI) during the 2002/03 CP El Niño event is related to the way the index is calculated. The IEMI incorporates averaged SSTA of three different regions across the equatorial Pacific (A, B and C, see Figure 3.2). To reach high values and hence to be identified as a CP El Niño event, region B and C must show colder SST than region A. If these conditions are met, the IEMI will be positive. During the 2002/03 event the SST also warmed up in region B, even though the peak warming was located in the centre of the equatorial Pacific (see Figures 3.8 and 3.7), which biased the index computation. If the region B shows positive SSTA, as it does during the event in 2002/03, this leads to lower IEMI 22 Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul −1 −0.5 0 0.5 1 1.5 st an da rd  d ev ia tio n Different El Nino indices during 02/03 CP El Nino event   IEMI NINO3 NINO4 NINO3.4 Figure 3.6: Monthly values of different indices during the CP El Niño event from July 2002 until July 2003. The IEMI dips below the one standard deviation threshold in the middle of the event, while the other indices raise above the threshold. Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 SS TA  in  d eg re e C SSTA of boxes A, B and C during 02/03 CP El Nino event   Box A Box B Box C Figure 3.7: Monthly SSTA averaged over the three boxes A, B and C during the CP El Niño event in 2002/03. These boxes are used to compute the IEMI. Box B shows substantial warming. 23 Figure 3.8: SSTA averaged over NDJF during the CP El Niño event in 2002/03. The peak warming lies in the centre of the equatorial Pacific but warming extends to the South American coast, into region B. values. The dip below the threshold during the event in 2002/03 is hence related to the way the index is calculated and also to the different weights that are given to the regions. This analysis shows that the CP El Niño classification with the IEMI is sensi- tive to the parameters and assumptions made by Li et al. (2010). If, for example, the weighting of the different regions of the IEMI equation is slightly altered, the dip can be reduced. Ashok et al. (2007) weighted regions B and C equally in the EMI equation. If this procedure is followed in the computation for the IEMI, and the weight of region B is set equal to the weight of region C, the drop can be substan- tially reduced, to the extent that the index does not fall below the threshold during October until February of 2002/03. However, as a consequence of the alteration, the time period from October 1991 until March 1992 would reach the threshold of one standard deviation and would hence be classified as a CP El Niño as well. This additional identification further points out that it is unclear which type of ENSO event the SSTA during 1991/92 resemble the best. This different results, which are obtained when the IEMI is slightly altered, show that the IEMI (as well as the EMI) are not robust enough. If the dataset, the climatology or the threshold are changed, different events are identified. 3.4.4 Extended analysis of identified Central Pacific and Eastern Pacific El Niño events The index computation and comparison in Section 3.4.2 shows that some events are clearly identified as one ENSO type by multiple indices, while other events are identified as both types at the same time. Designation of the events of 1987/88, 24 90/91 and 91/92 is especially difficult. Therefore, it is worthwhile to further ex- amine the characteristics of all the nine identified events individually to see if the events were classified consistently. Following the classification scheme of NINO3 and IEMI, as suggested by Li et al. (2010), four of them are EP El Niño (1982/83, 1987/88, 1991/92, 1997/98) and four are CP El Niño events (1990/91, 1994/95, 2004/05, 2009/10). The CP El Niño event during 2002/03 is included despite the fact that the IEMI did not identify it, as the SSTA composite in Section 3.4.3 con- firmed the existence of an event. All nine events were scrutinized regarding their SSTA evolution and their pre- cipitation and wind anomaly patterns, once with a composite perspective, and once with a meridional perspective of each individual event. The composite analysis is based on grouping the events as shown in Table 3.3 and aids in revealing key dif- ferences between the two types of events. The meridional analysis examines each of the nine events and hence extracts differences within the different events of one type. Moreover, the meridional analysis helps to see if the grouped events show a coherent pattern regarding the key variables and if classifying them into binary classes is justified. Table 3.3: Selected EP El Niño and CP El Niño event years for the composite analysis EP El Niño CP El Niño 82/83 90/91 87/88 94/95 91/92 02/03 97/98 04/05 09/10 Composite perspective of the two different event types In order to get a general impression of the seasonal evolution of SSTA of the CP El Niño and the EP El Niño events, it is useful to graph seasonal composites. The intensity of SSTA is higher during the average EP El Niño than during the average CP El Niño event. Further, SSTA develop earlier and decay later during 25 Figure 3.9: Composite of seasonal SST from the summer (top) before the event until the spring (bottom) after the event for the EP El Niño (left) and CP El Niño (right). SSTA are in ◦C. the averaged EP El Niño event: The average March, April, May (MAM) composite during EP El Niño events still displays substantial SSTA, whereas SSTA decay faster during the average CP El Niño event. Kao and Yu (2009) found that the average duration of an EP El Niño is fifteen months, whereas the average CP El Niño lasts for only eight months. Both types show their mature phase, despite the overall length, during the boreal winter (December, January, February (DJF)) (see Figure 3.9). There is a clear difference in the location of the peak warming centre as well as the propagation pattern of anomalies. During EP El Niño events the SSTA first build up close to the South American west coast in the NINO1+2 region and then propagate westward into NINO3, NINO3.4 and slightly into the NINO4 region. The centre of positive SSTA stays connected to the coast and centres east of 150◦W. SSTA decay last in the NINO1+2 region. Negative SSTA stay to the west of the peak warming centre and propagate slightly toward the east, north of the warming centre (see left panels in Figure 3.9). Analysis of the SSTA evolution of each individual EP El Niño event revealed that the events in 1987/88 started to build up 26 19 82   1 98 3   120E 180 120W 70W J J A S O N D J F M A M J 19 87   1 98 8   120E 180 120W 70W J J A S O N D J F M A M J 19 91   1 99 2   120E 180 120W 70W J J A S O N D J F M A M J 19 97   1 99 8   120E 180 120W 70W J J A S O N D J F M A M J −1 0 1 2 −1 0 1 2 −1 0 1 2 −1 0 1 2 Figure 3.10: Evolution of equatorial SSTA of EP El Niño events latitudinally averaged over 5◦N-5◦S from June before until June after the event. and peaked earlier than the other events. The event in 1991/92 showed the opposite behaviour: it peaked later than the other events (see Figure 3.10). During the average CP El Niño event SSTA appear first in the centre of the equatorial Pacific, between the dateline and 150◦W and off the coast of Mexico. They show very little propagation toward South America but more so off the equa- tor to form a horseshoe pattern. Kao and Yu (2009) found that the subsurface anomalies also show very little propagation. The cold tongue off the coast of South America is clearly detectable at the onset of the event and at the end, but vanishes during the peak anomalies from September until February (see right panels in Fig- ure 3.9). Individual evolution analysis of SSTA of CP El Niño events shows that all events except the event in 1990/91 showed positive SSTA in region B during the peak phase. Region B shows weaker positive anomalies than the peak anomalies in the centre of the equatorial Pacific but they are not negative anomalies, which means that these events do not show the tripole pattern of positive SSTA flanked by negative SSTA on each side as described by Ashok et al. (2007). All events 27 19 90   1 99 1   120E 180 120W 70W JJ A SO N D JF M A M J −1 −0.5 0 0.5 1 1.5 19 94   1 99 5   120E 180 120W 70W JJ A SO N D JF M A M J −1 −0.5 0 0.5 1 1.5 20 02   2 00 3   120E 180 120W 70W JJ A SO N D JF M A M J −1 −0.5 0 0.5 1 1.5 20 04   2 00 5   120E 180 120W 70W JJ A SO N D JF M A M J −1 −0.5 0 0.5 1 1.5 20 09   2 01 0   120E 180 120W 70W JJ A SO N D JF M A M J −1 −0.5 0 0.5 1 1.5 Figure 3.11: Evolution of equatorial SSTA of CP El Niño events latitudinally averaged over 5◦N-5◦S from June before until June after the event. except the weak event in 1990/91 follow mostly the symmetric-decaying pattern, described by Yu and Kim (2010), with symmetric build up and decay of SSTA. It further shows simultaneous warming in the central and eastern equatorial Pa- cific. The decay is usually followed by cooling in the eastern Pacific after the peak warming (see Figure 3.11). It needs to be noted that positive SSTA of strong EP El Niño events reach as far west as the dateline. In turn, positive SSTA of strong CP El Niño events also extend to the South American coast. The key difference lies in the peak warming centre. The previous analysis established that CP El Niño and EP El Niño show sub- stantially different SSTA patterns and evolution. Since the peak warming centre of the two types occurs at zonally different locations, the atmospheric feedbacks are expected to be different as well. Figure 3.12 and 3.13 show the composites of precipitation and wind anomalies of each type. The composites are based on months that showed peak anomalies prior identified by examining the evolution of 28 Figure 3.12: Equatorial precipitation anomalies composite from November until February for EP- (top) and CP (bottom) El Niño events. Precipi- tation anomalies are shown in mm/day. the anomalies over the duration of the individual events (results not shown). The precipitation composites (DJF) show that the centre of precipitation anoma- lies is shifted to the west of 150◦W and more locally constrained during CP El Niño events compared to the location and the zonal spread of the peak anomalies during an EP El Niño event (see Figure 3.12). This is in accordance with the shift in lo- cation of the maximum SSTA of the two types and confirms that the precipitation anomalies are affected by SSTA (Philander, 1990). Clearly visible is also the nega- tive (positive) precipitation anomalies over the western (eastern) equatorial Pacific in the EP El Niño composite, which is only detectable to a lesser extend or even reversed in the CP El Niño composite. This leads, for example, to opposite precip- itation anomalies over Ecuador and Peru during the two different event types. Kao and Yu (2009) state that the precipitation pattern of the CP El Niño event seems to be influenced by the movement of the ITCZ. The zonal surface wind composites (DJF) are in accordance with the SSTA pattern during the two types (see Figure 3.13). The EP El Niño composite shows the peak westerly anomalies to the east of the dateline. During the average CP El Niño event, the zonal wind anomalies occur shifted to the west compared to the EP El Niño. Westerly anomalies are confined to the dateline and less pronounced than during EP El Niño events. It also shows that easterly anomalies are more pro- 29 Figure 3.13: Equatorial zonal wind anomalies composite from December un- til February for EP (top)- and CP (bottom) El Niño events. Wind anomalies are shown in m/s. nounced over the eastern equatorial Pacific in the CP El Niño composite than in the EP El Niño composite. Figure 3.14 shows zonal and meridional wind values dur- ing DJF of the two event types. This figure confirms the main differences between the wind pattern of the two event types: the greater eastward extent of anomalous westerlies during the EP El Niño (top panel) and the dominance of south-easterly trade winds during the winter of a CP El Niño (bottom panel). These trade winds weaken during an EP El Niño event to a greater extent. This difference in wind pattern is linked to where the peak SSTA are located. The development of positive SSTA in the central equatorial Pacific is related to the anomalous eastward currents, induced by anomalous westerlies, and the anomalous easterlies over the eastern equatorial Pacific, which suppress warming in this region. Precipitation, wind and SST anomalies show a coherent picture and hence en- force the fact that ENSO events are coupled ocean-atmosphere phenomena (Phi- lander, 1990). The EP El Niño shows wind anomalies with a greater zonal extent reinforcing that this event type is a whole basin coupled phenomena. The zonal wind anomalies of the CP El Niño event are more locally constrained indicating that this event type reflects more local coupling. Kug et al. (2009) further emphasizes this by evaluating the sea level anoma- 30 120E 180 120W 70W 20S Eq 20N 120E 180 120W 70W 20S Eq 20N Figure 3.14: Winter wind composites (DJF) for EP (top) an CP (bottom) El Niño events. lies during the two types. They confirmed the commonly known sea-saw of the thermocline during the EP El Niño induced by anomalous westerlies, which leads to increased (decreased) sea level over the eastern (western) Pacific. This zonal asymmetry of sea level and heat content induces a discharge of accumulated heat to regions north and south of the equator, which can eventually initiate a La Niña event, following the recharge oscillator paradigm (Jin, 1997). During the CP El Niño they found the maximum sea level anomaly centred with the maximum SSTA and at the edge of the maximum wind anomalies. This reinforces the fact that the SSTA are mostly caused by surface forcing during a CP El Niño, as the weak ther- mocline variability during a CP El Niño event does not lead to a pronounced sea level gradient across the equator as compared to during an EP El Niño event (Kug et al., 2009). Kao and Yu (2009) support that the evolution of the CP El Niño is not related to thermocline variations. They further add that this ENSO type does not perform on a cycle/oscillation basis. Anomalous easterlies over the eastern Pacific during CP El Niño events (see Figure 3.13) can further suppress SST warming by Ekman transport induced coastal upwelling (shoaling of thermocline) but also by enhanced evaporation (Kug et al., 2009). This explains why SSTA close to the South American coast do not experience positive peak anomalies during a CP El Niño event. These characteristics also suggest why the CP El Niño event does not show the 31 same discharge process as the EP El Niño event. Kug et al. (2009) investigated the build up and decay of the sea level anomalies over both types and found that the CP El Niño does not fall into negative heat content anomalies, mainly because the overall heat content is lower during an average CP El Niño event but also due to the general pattern of SSTA and wind anomalies. As a consequence, due to the lower intensity, the SSTA decay quicker and the build up of a cold event following a warm event in the central Pacific is hindered. The physical mechanism behind the CP El Niño, however, is not fully understood yet (Yu and Kim, 2010). The La Niña event occurring after the strong CP El Niño event in 2009/10 forms an exception to this. Kim et al. (2011) confirm this unusual feature and attribute it to eastward moving negative subsurface anomalies. They further sug- gest that the anomalous warm Indian Ocean, as well as stronger than usual Rossby waves play possible roles. Meridional averaged perspective of each individual event To test for consistency within the categorized events, meridional averages of the examined variables (SSTA, precipitation and wind) are plotted for each of the nine events individually (see Figure 3.15). Figure 3.15 reaffirms several key character- istics between the CP El Niño and the EP El Niño, previously identified with the composite analysis, but it also reveals interesting variations within the events of each type and therefore questions the binary grouping. SST, precipitation and wind anomalies during averaged CP El Niño events show a westward shift compared to EP El Niño events (see rightmost graphs in Figure 3.15). Further, the anomalies are generally less pronounced: EP El Niño events reach averaged SSTA of about 2◦ C, whereas the averaged CP El Niño events only reach SSTA of about 1◦ C. The same relation is shown in the averaged wind anomalies; EP El Niño events reach 3 m/s whereas CP El Niño events only reach averaged peak anomaly of around 2 m/s. Surprisingly, the averaged peak anomalies of precipitation lie in a similar range, between 5 and 6 mm/day. The EP El Niño events in 1982/83 and 1997/98 show the greatest anomalies in all examined variables. The weakest events of both types (1987/88, 1991/92 and 1990/91) show abnormal behaviour in all variables compared to the average event. 32 The events in 1987/88 and 1991/92 were classified by EP- and CP El Niño indices. The NINO3.4 index reached in both cases the highest values, higher val- ues than the NINO3 or NINO4 (see Figure 3.4). This means that these two events reached peak warming outside the range that is associated with the EP El Niño event. They both experience peak warming between the dateline and 120◦W and represent more the SSTA pattern of a CP El Niño event than an EP El Niño event (see top left graph in Figure 3.15). The event in 1987/88 also shows peak precipi- tation and wind anomalies that resemble the individual CP El Niño events. These two events therefore exhibit characteristics that fall between the two ENSO types. The event in 1990/91 was only identified by the EMI and the IEMI. The SSTA, precipitation and wind anomalies are very weak and show a different pattern than the other CP El Niño events, which confirms that this event should probably not be grouped with the other CP El Niño events. 33 120E 180 120W 70W −1 −0.5 0 0.5 1 1.5 2 2.5 3 3.5 4 S S T A  [ ° C ]   EP82/83 EP87/88 EP91/92 EP97/98 EP avg 120E 180 120W 70W −1 −0.5 0 0.5 1 1.5 2 2.5 3 3.5 4 S S T A  [ ° C ]   CP90/91 CP94/95 CP02/03 CP04/05 CP09/10 CP avg 120E 180 120W 70W −1 −0.5 0 0.5 1 1.5 2 2.5 3 3.5 4 S S T A  [ ° C ]   EP avg CP avg 120E 180 120W 70W −6 −4 −2 0 2 4 6 8 10 P r e c i p i t a t i o n  a n o m a l i e s  [ m m / d a y ]   EP82/83 EP87/88 EP91/92 EP97/98 EP avg 120E 180 120W 70W −6 −4 −2 0 2 4 6 8 10 P r e c i p i t a t i o n  a n o m a l i e s  [ m m / d a y ]   CP90/91 CP94/95 CP02/03 CP04/05 CP09/10 CP avg 120E 180 120W 70W −6 −4 −2 0 2 4 6 8 10 P r e c i p i t a t i o n  a n o m a l i e s  [ m m / d a y ]   EP avg CP avg 120E 180 120W 70W −2 −1 0 1 2 3 4 5 6 U  w i n d  a n o m a l i e s  [ m / s ]   EP82/83 EP87/88 EP91/92 EP97/98 EP avg 120E 180 120W 70W −2 −1 0 1 2 3 4 5 6 U  w i n d  a n o m a l i e s  [ m / s ]   CP90/91 CP94/95 CP02/03 CP04/05 CP09/10 CP avg 120E 180 120W 70W −2 −1 0 1 2 3 4 5 6 U  w i n d  a n o m a l i e s  [ m / s ]   EP avg CP avg Figure 3.15: Comparison of latitudinally averaged equatorial SSTA (NDJF, top), precipitation (DJF, middle) and wind (SON, bottom) of individual EP (left) and CP (middle) El Niño events. The right most figures show averages. The time periods of the different variables are based on their peak months. 34 3.5 Discussion The objective of this chapter was to analyze the SST variation in the equatorial Pacific over the last thirty years in order to detect the distinct characteristics of the two contrasting types of ENSO. Further, this thirty year study period was utilized to evaluate the different indices that are commonly used to classify the two events. The combined regression EOF analysis adapted from Kao and Yu (2009) suc- cessfully separated the spatial patterns of the two event types and thus confirmed the existence of two ENSO types. The composite analysis further underlined the different characteristics in SST evolution, wind and precipitation patterns. How- ever, the index computation led to inconsistent results. Labelling the years as EP or CP El Niño event resulted to be less straightforward since the indices did not agree with each other or showed substantial limitations. NOAA (2003) recommended to use the NINO3.4 index to classify El Niño events but several papers, including this analysis, demonstrate that this index is not appropriate, since it cannot separate CP and EP El Niño events. This index is in that sense not robust and it can only capture general temperature variation in the equatorial Pacific. The NINO3.4 index grouped the strong 2009/10 CP El Niño event with other EP El Niño events. NINO4, which is used to identify CP El Niño, also fails to differentiate between CP- and EP El Niño types: It captures signals from all EP El Niño events along with signals from the CP El Niño phenomenon. NINO3 properly captures the warming of 1982/83 and 1997/98 as EP El Niño events but also identifies the weaker/mixed events in 1987/88 and 1991/92 as EP El Niño events. The time-series of the combined regression EOF analysis of each dominant modes (EOF EP El Niño and EOF CP El Niño) did not identify all the events correctly either. Spatial patterns vary slightly from one event to the other. As a consequence, weaker/mixed events do not follow the dominant pattern, which are largely influenced by the really strong events of each type, namely the 1982/83, 1997/98 and 2009/10 events. In addition, this method of classifying, which follows the method from Kao and Yu (2009), is too complicated on a daily basis and thus not applicable for forecasting. Li et al. (2010) recommended to use the IEMI, which is supposed to be a supe- 35 rior variant of the EMI, to classify CP El Niño events. This analysis suggests that these two indices have limitations that can lead to misclassification. Since the IEMI and the EMI are based on SSTA of three different regions in the equatorial Pacific, the index can easily be biased, due to the way the index is computed, if one of these regions experiences unusual characteristics of a CP El Niño event. Expected characteristics are negative SSTA in region B and C and positive SSTA in region A. If SST of these three regions do not show warming and cooling in respect to each other for a certain amount of time, the index does not identify this time period as an event. This explains how the indices can dip below the set threshold in the middle of an occurring event (see Section 3.4.3). This demonstrates that the IEMI is highly calibrated to the data set and climatology applied. It is not as robust and does not tolerate abnormal SSTA pattern in the region B. Three out of five identi- fied CP El Niño events show simultaneous warming in the region B, which leads to a dip in the IEMI (and EMI) and hence to misclassification, as shown for the CP El Niño event in 2002/03. This further refutes the statement about the superiority of the IEMI over the EMI made by Li et al. (2010). The IEMI and the EMI are not robust enough to capture abnormal CP El Niño events. Alteration of the IEMI equation (changing the weight of region B) could lessen the extend of the dip but this only reinforces that the IEMI is a highly calibrated index and is depended on the assumptions made by Li et al. (2010). Yu and Kim (2010) classified the evolution of simultaneous warming in the eastern and central equatorial Pacific as symmetric-decaying, because the build up and decaying of SSTA occurs symmetric around the peak phase. They further suggest that the simultaneous warming in the eastern Pacific as well as the decaying of SSTA after the event may be related to the state of the equatorial thermocline. They confirm that the CP El Niño phenomeon is more a result of local ocean- atmosphere coupling and not as strongly linked to the thermocline variation as the EP El Niño event is. The way a CP El Niño terminates, however, might be linked to the state of the thermocline. In a recent study, Yu and Kim (2011) further add, that the triggers and the terminations of CP El Niño events are possibly linked to extra tropical sea level pressure variations. In conclusion, stronger events are easier to classify into binary ENSO types than weaker events. The meridional analysis of the individual events identified 36 events in 1987/88, 90/91, and 91/92 as weaker or mixed events (similar to what Kug et al. (2009) suggested). These events did not fully develop, as the wind and SSTA pattern showed. The events in 1987/88 and 91/92 had their peak warming as a result in the NINO3.4 region and are therefore classified as mixed events. The event in 1990/91 showed very weak and in part abnormal anomalies throughout the different examined variables. This fact, together with the fact that it was only identified by the EMI and IEMI weakens its status as a CP El Niño event. In regard to the teleconnection analysis in Chapter 4, the mixed and weak events identified here will not be selected for the further teleconnection analysis, as they could weaken possibly different extra-tropical teleconnection signals of each type of El Niño. 37 Chapter 4 Teleconnection of Central Pacific and Eastern Pacific El Niño events 4.1 Introduction and background The weather in North America is known to be influenced by large-scale climate oscillations, such as ENSO, via teleconnections (Kiladis and Diaz, 1989). Tele- connections are weather anomalies that are related to a climate phenomenon that occurs at a different place than where the anomalies are experienced (Flannigan and Wotton, 2001). Weather anomalies that are substantially different than nor- mal can impact societies and economies on various levels (Vecchi and Wittenberg, 2010). It is hence from great importance to understand and predict teleconnections in western North America related to ENSO events. Teleconnections are a result of altered interactions between the equatorial Pa- cific and the overlying atmosphere: Anomalous warming of SST, a key charac- teristic of an ENSO event, results in shifted convection and precipitation patterns. This in turn affects the latent heat release into the atmosphere and forces the atmo- sphere to adjust. This adjustment triggers a chain of reaction and leads to weather anomalies around the globe (Hoerling and Kumar, 2000). ENSO has been shown 38 to strongly influence the precipitation pattern of the tropical regions, due to the close proximity (Diaz et al., 2001) but it also influences the Asian monsoon system (Allan et al., 1996) or the weather in New Zealand (Diaz et al., 2001) to name a few. Along with the change in SST and energy exchange during an ENSO event, upper air pressure patterns adjust, which is reflected in the PNA teleconnections (Shabbar, 2006). This eventually influences the position of the jet stream and therefore the trajectories of storms. Weather patterns along the west coast of North America are affected by this change in storm tracks and are, therefore, also greatly influenced by ENSO (Hoerling and Kumar, 2000). Typical ENSO teleconnection analysis is based on observational data or on cli- mate model simulations. Identifying and clarifying the source of weather anoma- lies related with ENSO events is an ongoing challenge. With the recognition of a new type of El Nino (see Chapter 3), a new perspective has been added to this challenge and the teleconnection analysis needs to be reassessed. Several studies that investigated CP El Nino events found that this new type might cause different teleconnection signals than the canonical EP El Niño events (Ashok et al., 2007; Weng et al., 2009). Weng et al. (2009) examined the anomalous winter temperature and precipita- tion conditions in the Pacific rim during recent CP and EP El Niño events. Rather than the broad pattern of dry north and wet southern portions of western north America noted by Dettinger et al. (1998), Weng et al. (2009) associate this pattern only with CP El Niño events, while for EP El Niño events the majority of western north America remains wet. They further state that the temperature anomalies related to the two different types of ENSO show a different pattern. Whereas during a CP El Niño the tem- perature anomalies show a vertical split with the peak positive anomalies in Alaska and Greenland and negative anomalies in eastern Canada and U.S., the EP El Niño temperature pattern shows a horizontal split, with positive anomalies over large parts of the U.S. and Canada and negative anomalies from Alaska to northeastern Canada. Ashok et al. (2007) also found that some regions, including the west coast of the United States, experience opposite weather anomalies during the CP El Niño than during the conventional El Niño. 39 Weng et al. (2009) identified the main cause for the different teleconnections in the tripole pattern of SSTA present during CP El Niño events, which leads to different shaped boomerangs of ocean-atmosphere conditions and differences in the Walker circulation (Ashok et al., 2007). 4.2 Research objectives The majority of teleconnection analyses base the identification of ENSO events on the NINO3.4 index, as suggested by NOAA in 2003 (see Chapter 3), and do not distinguish between the two different types of events. As mentioned above, broad scales studies about the recently recognized CP El Niño have pointed out that this new type might bring along different teleconnections than the canonical EP El Niño (Weng et al., 2007; Larkin and Harrison, 2005a). These studies have further pointed out that not distinguishing between the two types of events can obscure possibly different teleconnection signals of each type of El Niño. Therefore, it is necessary to distinguish between the two different types and analyse the telecon- nections separately. Otherwise, their possibly different signals get mixed together and could potentially cancel each other out (Larkin and Harrison, 2005a,b). Although the average SSTA of a CP El Niño are not as large as during an average EP El Niño, Kug et al. (2009) found that the teleconnections of the CP El Niño might have an equal influence on global weather anomalies as the EP El Niño. They explain this by the warmer background SST in the central Pacific than in the eastern Pacific. SSTA in the central Pacific can cause a comparable magnitude of vertical motion (convection), which influences the overlying atmosphere and induces teleconnections. The goal of this chapter is to assess the different climate signals of western North America to the past CP- and EP El Niño events from 1981-2010, previously identified in Chapter 3. It is also of interest to see, if grouping all ENSO events into one group and not distinguishing between the types leads to dampened and obscured teleconnection signals. The impact on the weather is measured by focusing on three climate variables: surface temperature, precipitation and 500 hPa geopotential height. The telecon- nection analysis chapter also includes further examination to assess the influence 40 of other climate oscillations, such as the AO, on the weather in western North America. 4.3 Methods Climate composites of the two types are analyzed separately, based on the identifi- cation of the events in Table 4.1. The mixed/weak events in 1987/88, 1990/91 and 1991/92 previously identified in Chapter 3, are not included in this analysis. Their characteristics are in between the two types of events or too weak, therefore, their teleconnections would only dampen the overall signal of the CP- and EP El Niño events. Table 4.1: Selected events for the teleconnection analysis EP El Niño CP El Niño 1982/83 1994/95 1997/98 2002/03 2004/05 2009/10 The climate variables were converted from a Lambert Conformal Conic grid to the resolution of a regular lat-long grid. Seasonal winter (DJF) averages for each individual event, and overall EP and CP El Niño winter composites have been calculated for the selected years, similar to the analysis of Weng et al. (2007) and Larkin and Harrison (2005a). All seasonal composites are based on the anomalies of the different climate variables (GPH, temperature and precipitation) in regard to the same base period as in the analysis of the SST in Chapter 3 (1981-2010). The temporal focus lies on the DJF composites, as the teleconnections are felt strongest during the winter months, when the SSTA peak and the subtropical and subpolar jet streams are stronger (Ashok et al., 2007). The spatial focus lies on western North America with an additional locally focused case study of southeast- ern British Columbia (BC). The region in southeastern BC is the same as in Chapter 5 (see Figure 5.1). To assess the influence of other climate oscillations (AO, PDO, PNA and Atlantic Multidecadal Oscillation (AMO)) EOF and correlation analysis was applied. 41 4.4 Results The result section is spilt in five parts: The first three parts analyse the winter composite results of geopotential height, temperature and precipitation for the CP and EP El Niño events for whole North America (see Section 4.4.1, 4.4.2, 4.4.3) and in a fourth part for southeastern BC (see Section 4.4.4). The final part assesses the possible impact that other large scale climate oscillations could have on the weather in western North America (see Section 4.4.5). 4.4.1 Geopotential height All winter composites of the four CP El Niño events show an individual pressure pattern (see top four panels in Figure 4.1), whereas the two EP El Niño composites seem to show a consistent picture (see bottom two panels in Figure 4.1). During the two EP El Niño event, the Aleutian Low deepens and slightly shifts southeastward. Positive pressure anomalies are found over eastern Canada. These results are conform with the findings of Diaz et al. (2001). The CP El Niño DJF panels do not show a distinctively similar pattern for North America, as the two exceptionally strong EP El Niño events do. When the GPH is examined across western North America it can be seen that in contrast to the negative pressure anomalies persistent over western North America in the EP El Niño composites, CP El Niño composites only show slightly negative or even positive pressure anomalies along western North America (see Figure 4.1). The pressure pattern of the CP El Niño event during the winter 2009/10 stands out because of its magnitude and its substantially different pressure pattern com- pared to the other composites. It shows a deepened Aleutian Low similar to the EP El Niño composites. Different, however, is the strong positive anomalies over northern Canada (see right panel in the middle of Figure 4.1). This pattern and possible reasons are further examined in Section 4.4.5. The mean seasonal composite of all EP- and CP El Niño events, show a sub- stantially different pattern (see two top left panels in Figure 4.2). Notably, the difference in the position and intensity of the Aleutian Low is remarkable. How- ever, the mean of the different CP El Niño events has to be interpreted with caution, as indicated by the large standard deviation. The large standard deviation of the CP 42 El Niño average is due to the variability within the different CP El Niño events (see graphs on the left in Figure 4.2). However, if the focus lies on western North America, the standard deviation of both types lies in the range of 10-30 hPa, re- vealing that the pressure pattern signal of the CP and EP El Nino is comparably consistent for this region of North America. If all six ENSO events are evaluated together the mean pressure pattern does not represent well the different influence the two types of ENSO have on the Aleutian Low, as shown when the GPH is analysed separately for each type. That the two event types have a different impact on the Aleutian Low is shown by the increased standard deviation over the North Pacific compared to the standard deviation of the CP and EP El Niño composites (see graphs on the right in Figure 4.2). 43  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   DJF 1994/95 −200 −150 −100 −50 0 50 100 150 200  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   DJF 2002/03 −200 −150 −100 −50 0 50 100 150 200  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   DJF 2004/05 −200 −150 −100 −50 0 50 100 150 200  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   DJF 2009/10 −200 −150 −100 −50 0 50 100 150 200  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   DJF 1982/83 −200 −150 −100 −50 0 50 100 150 200  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   DJF 1997/98 −200 −150 −100 −50 0 50 100 150 200 Figure 4.1: DJF composite of geopotential height anomalies [500 hPa] for individual CP (top four) - and EP (bottom two) El Niño events from 1981-2010. 44  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   CP DJF mean −150 −100 −50 0 50  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   EP DJF mean −150 −100 −50 0 50  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   ENSO DJF mean −150 −100 −50 0 50  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   CP DJF std 0 10 20 30 40 50 60 70 80 90 100  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   EP DJF std 0 10 20 30 40 50 60 70 80 90 100  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   ENSO DJF std 0 10 20 30 40 50 60 70 80 90 100 Figure 4.2: DJF composite of mean geopotential height anomalies [500 hPa] and respective standard deviation for averaged CP- (left), EP- (middle) and all El Niño events together (right) from 1981-2010. 45 4.4.2 2 m air temperature For the temperature teleconnections of the two ENSO types only the averaged DJF composites for all EP and CP El Niño events are shown and not the individual event years (see Figure 4.3). The temperature pattern during the EP El Niño composite shows a meridional temperature split across North America, with positive temperature anomalies over most part of the U.S. and southern Canada and negative temperature anomalies over southern U.S. and northern Canada. This tripole pattern extends toward the west coast: Western North America shows colder temperature in Alaska, warmer temperatures along most of Yukon, BC and colder than normal temperatures in California (see graph in the middle in Figure 4.3). This horizontal split has also been identified by Weng et al. (2009). The standard deviation linked to this pattern is very low over great parts of North America (see middle bottom panel in Figure 4.3). The winter composite during the CP El Niño events displays a different tem- perature anomalies pattern. Similar to results of Weng et al. (2009), positive tem- perature anomalies are found along western North America with peak anomalies in Alaska. There is no meridional split detectable. Another great difference is found outside western North America, in eastern Canada: compared to the EP El Niño composite, the temperature anomalies are of the opposite sign (see top panel on the left in Figure 4.3). The standard deviation related to the CP El Niño composite is greater than the standard deviation related to the EP El Niño composite. However, along western North America the temperature signals seem to be pretty consistent across the different events. The standard deviation shows low values around 1◦ C (see bottom left panel in Figure 4.3). If all ENSO events are grouped together, the temperature signal is dampened, the peak anomalies are less pronounced. As the temperature signals of the CP El Niño are not that strong in western North America, the standard deviation of the ENSO composite increases only slightly in that region. The great difference seems to be in eastern Canada, where the temperature signals are of the opposite sign, which is reflected in the increased standard deviation in this region (see graphs on the right in Figure 4.3). 46  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   CP DJF mean −6 −4 −2 0 2 4 6  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   EP DJF mean −6 −4 −2 0 2 4 6  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   ENSO DJF mean −6 −4 −2 0 2 4 6  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   CP DJF std 1 2 3 4 5 6 7 8 9 10  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   EP DJF std 1 2 3 4 5 6 7 8 9 10  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   ENSO DJF std 1 2 3 4 5 6 7 8 9 10 Figure 4.3: DJF composite of air temperautre anomalies [◦C, 2m] and respective standard deviation for averaged CP- (left), EP- (middle) events and for all El Niño events together (right) from 1981-2010. 47 4.4.3 Precipitation Precipitation teleconnections associated with canonical El Niño events are thought to cause a dipole pattern in western North America: drier in the north and wetter in the south with a pivotal point around 40◦N (Dettinger et al., 1998). Dettinger et al. (1998) added that ENSO events have not only a seesaw impact on precipitation distribution, but also cause an increase in overall precipitation compared to nor- mal situations. The DJF composites of the two types suggest a more complicated teleconnection pattern than that identified by Dettinger et al. (1998) and provide possible support for the results of Weng et al. (2009) (see Figure 4.4). Similar to the temperature results, only averaged EP and CP El Niño composites are shown, not individual events. The winter precipitation composite of the EP El Niño events shows most of the western U.S. coast with increased precipitation and decreased precipitation in the northern parts of western North America. The negative anomalies extend north- ward into Alaska (see top middle panel in Figure 4.4). The wetter conditions are related to the changed track of cyclones as a result of the altered 500 hPa circu- lation pattern (Diaz et al., 2001). Decreased precipitation anomalies are found in southwestern Canada, Shabbar (2006) confirms this. The CP El Niño composite shows drier than usual conditions along the northern parts of western North America and also more pronounced than during EP El Niño events. It further shows wetter conditions in the south of the western U.S. but also to a lesser extent than during EP El Niño events (see bottom left panel in Figure 4.4). Weng et al. (2009) found similar results and attributes the seesaw precipitation of wet south and dry north of the western U.S. more likely to appear during CP than EP El Niño events. High standard deviation values are concentrated along the west coast in both EP and CP El Niño precipitation composites. This indicates considerable variabil- ity and makes it difficult to draw firm conclusions. The standard deviation values increase even more, if all ENSO events are evaluated together. Also the pattern shown in the ENSO average is a clear mixture of both events and hinders a distinc- tion of the possible different signals of the two event types regarding the seesaw effect (see right panels in Figure 4.4). 48  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   CP DJF mean −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   EP DJF mean −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   ENSO DJF mean −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   CP DJF std 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   EP DJF std 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   ENSO DJF std 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Figure 4.4: DJF composite of accumulated precipitation anomalies [kg/m2] and respective standard deviation for aver- aged CP- (left) and EP- (middle) and all El Niño events together (right) from 1981-2010. 49 4.4.4 Teleconnections in southeastern British Columbia In order to focus on a more spatially constrained region, southeastern BC was chosen. This analysis shows that the teleconnection signals can have a clear and consistent signal if the spatial focus is smaller. Analysis of geopotential height, temperature and precipitation of winter com- posites for southeastern BC during ENSO events showed that distinguishing be- tween the two event types leads to more consistent teleconnection signals. In par- ticular, the composites during the CP El Niño events led to consistent signals in southeastern BC. The standard deviation showed relatively low values, compared to the values during EP El Niño composites. During EP El Niño events, southeastern BC tended to experience negative pres- sure and negative precipitation anomalies. More strikingly was the temperature anomaly, an average of +1.6◦ C was recorded during the two events in 1982/83 and 1997/98. During CP El Niño events the signals were different. The pressure analysis showed positive anomalies, the opposite than during EP El Niño events. The temperature and precipitation analysis resulted in similar signs anomalies but to a weaker extend. The average temperature anomaly during the four CP El Niño events reached only +0.75◦ C. 4.4.5 Possible influence of other climate oscillations There are a number of other large scale climate oscillations besides ENSO that can contribute to weather anomalies in North America on varying time scales (AO, AMO, PDO, PNA). As mentioned in Section 4.4.1, the pressure composite during the CP El Niño event in 2009/10 showed a substantially different anomalies pattern than all the other CP El Niño events. To assess the possible importance of different large-scale climate oscillations and also to suggest a reason for the distinct pressure pattern during the CP El Niño event in 2009/10, further analysis was conducted on GPH. The dominant modes of GPH patterns of the entire study period were determined via EOF analysis. The corresponding time-series of the first few dominant modes were then correlated with the time series of the different large-scale climate oscillations, as well as with 50  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 o N   30 o N   45 o N   60 o N   75 o N   −5 0 5 10 x 10−3 −5 −4 −3 −2 −1 0 1 2 3 4 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 x 104 PC  1  s co re s AO index   Figure 4.5: The dominant mode of DJF GPH anomalies (left) explains 26% of the variance. It shows the greatest correlation with the AO index (right). The red line represent a linear fit. the time-series of NINO3 and NINO4. This analysis helped to identify one of the main drivers behind the pressure patterns. The dominant mode returned by the EOF analysis explains 26% of the variance of geopotential height over the whole study period. It displays higher pressure over the Arctic and parts of northern Canada and lower pressure over the North Pacific and most parts of the U.S. (see panel on the left in Figure 4.5). All correlation tests between the different climate oscillation indices as well as NINO3 and NINO4 and the time-series of the dominant EOF mode resulted in significant correlation values. This means that all climate oscillations as well as ENSO events contribute to the pressure pattern over North America. However, the AO index reached the highest correlation coefficient with the time-series of the dominant mode of the EOF analysis (r=-0.69). This shows that the AO plays a key role in influencing the pressure pattern over North America (Thompson and Wallace, 2000). The same analysis was also conducted on only DJF GPH composites (not shown). The AO index still reached the highest correlation coefficient with the time-series of the dominant mode from the EOF analysis. The explained variance by the first mode increased to 30.6%. When the ENSO indices were tested for cor- 51 relation with the dominant DJF EOF mode the correlation values increased com- pared to the correlation outcome with the dominant EOF mode over all months. This reinforces that the ENSO events show their greatest impacts over the winter months, when they are in their mature phase. The AO (Thompson and Wallace, 1998) represents the dominant pressure pat- tern over the Arctic, which is strongly influencing the strength of the mid-latitude jet stream. A positive AO phase and hence low pressure over the Arctic leads to a stronger jet stream. A strong jet stream tends to block southward cold air out- breaks from the Arctic. The opposite is the case during a negative AO phase and high pressure is persistent over the Arctic. During these conditions the jet stream tends to be weaker and allows cold air reaching the mid-latitudes. When the DJF GPH spatial pattern of the most recent CP El Niño event in 2009/10 is correlated with the spatial pattern of the dominant EOF mode of DJF pressure anomalies, it is apparent that the CP El Niño composite resembles strongly the dominant mode, with high pressure anomalies across high latitudes and anoma- lously low pressure at mid-latitudes (Cohen et al., 2010). Indeed, these two pres- sure patterns reach almost perfect correlation (r=-0.87). The dipole pressure pattern during the peak season of the CP El Niño event in 2009/10 was hence strongly in- fluenced by an exceptional strong negative AO state. This strong negative state is apparent in the AO time series and marks the lowest record since 1950 (L’Heureux et al., 2010). According to this record low, weather anomalies were recorded across North America (L’Heureux et al., 2010). Broad regions of Canada showed positive tem- perature anomalies, while the U.S., especially the east coast, experienced colder than usual temperatures (see Figure 4.6). Cohen et al. (2010) obtained similar re- sults. They further found that the AO explained most of the temperature variance compared to the El Niño signal. Possible signals from the CP El Niño were hence overshadowed by the strong negative AO phase. To filter out the influence from the AO, the DJF GPH com- posites were regressed with the DJF AO index. The DJF geopotential height com- posites were then computed again. This approach resulted in a reduced standard deviation for the EP- as well as the CP El Niño events and hence reduced the vari- ability within the events (results not shown). 52  180 oW  165 oW  150 oW  135oW  120oW  105 oW   90 o W   75 o W   60 o W   15 oN   30 oN   45 oN   60 oN   75 oN   CP DJF 09/10 −6 −4 −2 0 2 4 6 8 10 12 Figure 4.6: Temperature [◦C] anomalies during the winter 2009/10. 4.5 Discussion To assess the different teleconnections during the peak time of CP- and EP El Niño events of the past three decades, three climate variables from the NARR data set were analysed. Seasonal winter composites were formed for the two event types individually but also for all ENSO events grouped together. This composite analysis indicates that the teleconnection signals of the two ENSO types can show substantially different characteristics with a large scale ap- proach across western North America. This has been supported with a spatially focused analysis on southeastern BC. The comparison between the analysis of sep- arated teleconnection signals of each ENSO type and the analysis of the telecon- nection signals of all events grouped together as one type, outlined that mixing two different event types into one group can substantially obscure and dampen telecon- nection signals. This was further supported by the increased standard deviation of the composites consisting of all events grouped into one type. The limited number of clearly identifiable EP and CP El Niño events during the last three decades obviates a clear conclusion. As the teleconnections are analysed separately for EP- and CP El Niño, the composites are based on only a couple of events. With a small sample size (CP El Niño: n=4; EP El Niño:n=2) the variance is naturally high. The here obtained results do not result from significance tests, as it is difficult to draw statistically significant results due to the high number of degrees of freedom. Therefore, the results obtained here have to be interpreted with caution. The two strong EP El Niño events in 1982/83 and 1997/98 led to a relatively 53 consistent teleconnection signal across western North America with a low standard deviation. This low variance among the two EP El Niño events could however be misleading. A larger sample size could reveal different patterns and hence increase the variance. It is interesting that within the available samples, the two EP El Niño events were quite different from the CP El Niño years. Mainly the difference in the GPH seems to indicate a substantial difference. The temperature anomalies also showed different signals: western North America experienced neutral or increased temper- ature anomalies during CP El Niño events, whereas during EP El Niño events the temperature field showed a tripole pattern along the coast. The precipitation anal- ysis confirms the results from Weng et al. (2009). The seesaw pattern of a drier north and wetter south of the U.S west coast tends to be more present during CP El Niño years. During EP El Niño years, the entire west coast of the U.S. shows increased precipitation. The seesaw pattern along the west coast of the U.S. how- ever shows when all ENSO events are examined together. The assumption of a seesaw precipitation pattern along the U.S. west coast could potentially be a result of mixing teleconnections of two event types together. Analysing the precipitation pattern separately for each event type revealed a different pattern. The spatial focus on southeastern BC revealed different signals depending on the type of ENSO event. This could, however, also be influenced by chance alone. Diaz et al. (2001) found that ENSO events usually only explain less than 25% of the annual variation and that they can be overshadowed by other strong variabili- ties. There exists a variety of different climate oscillations that affect the weather of western North America. Climate oscillations such as the AO or the PDO can enhance, weaken or overshadow the possible impacts of ENSO events. Weng et al. (2007) adds that this is from particular importance if ENSO events are weak. The detailed pressure analysis in Section 4.4.5 outlined exactly this issue of overshadowing. The dominant modes of winter pressure patterns over North Amer- ica, obtained by an EOF analysis, showed significant correlation with all the tested climate oscillation indices. The strongest correlation, however, was achieved with the AO, meaning that the winter pressure pattern of North America is mostly in- fluenced by the AO. This fact is best observed when studying the winter pressure pattern during the CP El Niño event in 2009/10, which showed almost perfect cor- 54 relation with the dominant pressure pattern present during a strong negative AO state. The AO, therefore, overshadowed any possible signal from this CP El Niño event itself. To extract the influence from the AO, the monthly pressure values were regressed against the AO index. However, even after regressing, the overall standard deviation of the seasonal pressure patterns showed still a great spread. This means that the AO is only one factor that influences and contributes to the weather anomalies in North America. This issue underlines the complexity of weather and its multi-scalar scope of impact factors. As a consequence, it is difficult to identify the exact source of anomalous weather conditions and to only extract signals that are caused by differ- ent types of ENSO events. 55 Chapter 5 Forest fires in North America 5.1 Introduction and background Forest fires are an important natural disturbance in North America (Agee, 1996). Different studies that analyzed the forest fire history of North America reported an increase of forest fire frequency and area burned over the last few decades in different areas of the United States (e.g. Southern Rockies (Litschert et al., 2012)) but also across Canada (Stocks et al., 2002). In Canada, as an example, more than twenty fires occur on average every day, resulting in a total averaged area burned of two million hectares per year (Stocks et al., 2002). The majority of the total area burned is caused by natural forest fires triggered by lightning strokes (Stocks et al., 2002; Flannigan and Wotton, 2001). Large forest fires are scarce but they cause the majority of the total area burned (Stocks et al., 2002). Forest fires can cause important environmental, economic and societal impacts. They shape and naturally regenerate ecosystems (Agee, 1996). At the same time they endanger societies, property and resources as well as economic loss (e.g. sup- pression cost) in general. Annual fire suppression expenditures average US$500 in Canada (Flannigan et al., 2006) and over US$1 billion in the U.S. (Calkin et al., 2005). In addition, by releasing pollutants and producing smoke, forest fires dete- riorate air quality and visibility (Jaffe et al., 2008). They further release terrestrial carbon to the atmosphere and contribute indirectly to global warming (Amiro et al., 2009). In the case of Canada, large forest fires in 2003, 2004 and 2009 released 56 extensive amounts of carbon that turned the forests into a net carbon source (Stocks et al., 2002). For a natural forest fire to occur certain climatic conditions, sufficient fuel and a natural or anthropogenic ignition source are needed (Price and Rind, 1994). Cli- mate is a strong top-down driver of forest fires (Flannigan et al., 2000) and deter- mines hence forest fire frequency (Westerling et al., 2006) and size (Skinner et al., 1999; Stephens, 2005). Climatic conditions are crucial because they determine forest conditions, which strongly relate to how susceptible a forest is to ignition (Trouet et al., 2006). A key factor in forest fire severity is the fuel moisture content, as this vari- able reflects climatic conditions such as temperature and precipitation and these in turn biomass growth (Flannigan and Harrington, 1988; Gedalof et al., 2005). Many studies showed therefore, that the extent of forest fires is influenced by drought con- ditions (Xiao and Zhuang, 2007). The PDSI is a commonly used index in drought as well as forest fire analysis, as it represents current and previous moisture condi- tions (Shabbar and Skinner, 2004). The self calibrated PDSI is used as a proxy to summarize climatic conditions that could lead to forest fire incidents, as this index is based on temperature and precipitation, two key variables for forest fire occur- rence. Dai (2011b) further mentioned that this index is correlated with soil mois- ture, another important factor for forest fire outbreaks. Westerling et al. (2002), for example, predicted area burned in western U.S. with the the PDSI. Meyn et al. (2010) found a strong link between drought conditions, identified with the PDSI, and area burned at several high-elevation sites across British Columbia and Yukon. There are also several studies that link forest fire severity directly with SSTA, mostly by outlining the influence of SSTA on drought patterns. Shabbar et al. (2011), for example, showed in a recent study that global SST and the PDSI are good predictors for Canadian summer forest fire severity. Further, ENSO, PDO and AMO have been identified as among the major drivers of North American forest fire frequency since 1550 (Kitzberger et al., 2007) and recent (1959-1999) summer forest fire severity in Canada (Skinner et al., 2006). 57 5.2 Research objectives With the emergence of a new type of El Niño (see Chapter 3) and possibly different teleconnections (see Chapter 4) it is important to see if this new ENSO type is as- sociated with a different drought pattern in western North America and also if this influences the forest fire pattern. Understanding a possible influence of the CP El Niño is moreover especially important because forest fires are expected to increase due to climate change (Flannigan et al., 2006). Balshi et al. (2009) estimate that the annual area burned in western North America will double by 2041-2050 in regard to the amount burned during 1991-2000. Further, the CP El Niño is thought to have a link with climate change and is projected to occur more frequently in the future (Lee and McPhaden, 2010; Yeh et al., 2009). Therefore, knowing what influence the Central Pacific El Niño (CP El Niño) has on drought and on forest fire patterns is from great importance, as this knowledge can help to better predict fire suscep- tibility. Additionally, it could also prevent societal and environmental impacts and reduce suppression costs. The goal of this study is to analyse forest fire patterns and their drivers across North America for the last three decades. First, hot spots of area burned and fre- quency across North America are examined (see Section 5.4.1). Different studies have analysed the forest fire history of Canada or the western U.S. but there are not many cross-boundary analyses that unify and compare past forest fire occurrences across all of North America. Second, it is assessed how the two different ENSO types influence the drought pattern, represented by the PDSI, over North America (see Section 5.4.2). Finally, a possible connection between the PDSI and forest fire occurrence (frequency and area burned) in three regions across western North America is tested (see Section 5.4.3). The three regions are based on the results of the hot spots analysis in Section 5.4.1. 5.3 Methods For the forest fire analysis a data set from Canada and one from the United States were merged. These data sets itself were also compilations of data bases from various forest fire agencies across the countries. In order to reach an acceptable co- herency between the two different data sets, they were vigorously cleaned accord- 58 ing to the following listed steps. This cleaning, conducted in AcrGIS10, reduced the amount of forest fires incidents from originally around one million to around 800,000 incidents. • Only forest fires from 1981-2010 were included, as this is the time frame when both datasets overlap (deleted circa 140,000). • Forest fires without a start month were assigned the month when the fire was reported or the month when the fire was extinguished, depending on which date was available. Fires that had no date assigned were deleted (less than 1,000) • Forest fires that were located outside the national boundaries of the U.S. or Canada or in the ocean were excluded (less than 1,000). • Forest fires with no or negative area burned were excluded from the analysis (circa 58,000). To reach common units of area burned, acres, which are used in the data set from the United States, were converted into hectares. • Forest fire causes were grouped into human, natural and unknown. Around 800,000 individual fire incidents were then merged into a point shape- file to obtain one file with the complete fire history of North America from 1981- 2010 (see statistics of all fires in Section 5.4.1). For the analysis in Section 5.4.2 and 5.4.3 only fires with a natural cause are included (circa 220,000). As the point data do not allow for a simple visual interpretation of the forest fire pattern, the point data were converted into a smooth continuous surface. This was achieved by using the Kernel Density Analysis in ArcGIS10. This analysis is a special form of a spatial analysis which identifies peak densities across a spatial data set and accentuates them. This analysis is based on the kernel density function (Silverman, 1986), which incorporates and weights points in close neighbourhood around a centre point to evaluate the density of an attribute value that is associated with the analyzed points. The Kernel density function was used to evaluate the attribute value area burned and the frequency of fire incidents. The search radius was in both cases set to 2.5◦, as this is the resolution of the PDSI dataset. Changing the search radius does not greatly affect the density, it only includes a larger number 59 of points, which are located farther away from the centre point. As a consequence, the resulting density map appears smoother and more generalized. The output units of this analysis are squared hectares. To simplify this unit, the density map has been recategorized in three classes: low, medium and high, based on natural breaks in the data (also known as Jenks). This analysis allows to quickly identify hot-spots of area burned and forest fire frequency across North America (see Section 5.4.1). To give a broad overview of a potential ENSO influence on drought, the rela- tionship between DJF SSTA of the equatorial Pacific and anomalies of June, July, August (JJA) PDSI of North America were examined, similar to the methods of Shabbar and Skinner (2004) (see Section 5.4.2). The winter SSTA were selected because this is the period when CP- and EP El Niño events peak. JJA PDSI anoma- lies were selected because these months resulted to be the most fire active months in North America (see Section 5.4.1). The relationship between SSTA and PDSI anomalies is assessed with a Sin- gular Value Decomposition (SVD) analysis 1 and a (spatial) correlation analysis. SVD analysis is used to examine the coupled variability of these two variables and find the dominant modes that explain the majority of the covariance between them. SVD first decomposes the covariance matrix of the two variables into sin- gular values, these singular values are used to calculate the Square Covariance Fraction (SCF) that is explained by a mode. SVD identifies dominant modes, where the two variables experience a strong coupled variation, as well as the time series corresponding to the dominant modes (Bjornsson and Venegas, 1997). Sim- plified, it can be said that SVD resembles a canonical correlation analysis but adds a spatial pattern to the time series. Further, canonical correlation focuses on max- imized correlation, whereas SVD focuses on maximized covariance between vari- ables (Shabbar and Skinner, 2004). This method allows to examine a possible link between SSTA and drought patterns in North America in a simple way. Similar to the limitations of an EOF analysis however, there is a chance that the modes identified by a SVD analysis are only artificial modes created by the data and not real physical modes (Bjornsson and Venegas, 1997). The physical explanation of the coupled patterns can be supported by a correlation analysis of the time-series 1The SVD analysis was applied on anomalies of each variable. 60 resulting from the coupled modes (Shabbar and Skinner, 2004). To further reduce the possibility that the modes identified by the SVD analysis are only artificial, the relationship between SSTA and the PDSI pattern is also analysed via spatial corre- lation analysis and supported by a composite analysis of the summer PDSI values related to ENSO events. The focus of the forest fire analysis along western North America (see Section 5.4.3) lies on three regions (see Figure 5.1). These regions were chosen based on two criteria: • Firstly, the majority of each selected region had to fall into one ecoregion type of the ecoregion classification of the United States Environmental Pro- tection Agency (EPA). This way forest fires from a similar vegetation type are aggregated. Kilgore (1981) states that forest fire severity depends on vegetation type. Hence aggregating forest fires according to ecoregions is more appropriate than aggregating, for example, according to ecologically inexact governmental boundaries (Littell et al., 2009). However, this aggre- gation method ignores forest fire management policies, such as suppression strategies, which are based on governmental entities. • Secondly, the selection had to capture the hot spots of area burned and/or forest fire frequency previously identified in the kernel density analysis in Section 5.4.1. One of the three selected regions falls into the Taiga of Alaska, the second region focuses on Northwestern Forested Mountains in southeast BC, and the third region falls into the North American Desert in the western U.S. (see Figure 5.1). The first and the last region capture the hot spots of the total area burned, whereas the region in southeast BC captures one hot spot of forest fire frequency, as revealed by the two Kernel Density analyses in Section 5.4.1. The fire season lengths vary with latitude, the temporal focus was thus set from June until August in order to capture the main fire active months across the three regions. These three months capture the majority of the total area burned and forest fire frequency of the three regions. To examine the possible link between area burned/forest fire frequency and drought patterns in these three regions, correlation analysis was conducted. The 61 0 600 1,200300 Km Selected regions for forest fire analysis along the west coast « Legend NORTH AMERICAN DESERTS NORTHERN FORESTS NORTHWESTERN FORESTED MOUNTAINS TAIGA TUNDRA Selected regions Source: EPA Ecoregions Level I Figure 5.1: Selected regions along the west coast of North America. The northern most region mainly falls into the Taiga ecoregion. The mid- dle region, in southeast BC, mostly consists of Northwestern Forested Mountains. The third region in western U.S. represents North American Desert. 62 distribution of the values of the area burned is skewed, as only a small num- ber of fires reach large values. The area burned values have, therefore, been log transformed for the correlation analysis to achieve a normal distribution (Xiao and Zhuang, 2007). 5.4 Results The results of this chapter are split according to the three main research questions: The first section summarizes the forest fire history of North America from 1981- 2010 according to the total area burned, the frequency of incidents and their cause (see Section 5.4.1). The second part consists of a SVD, spatial correlation and composite analysis to determine how equatorial SSTA during DJF (peak months of ENSO events) can influence the summer (JJA) drought pattern of North America (see Section 5.4.2). The drought pattern of North America is represented by the self calibrated PDSI. The third part of the results is devoted to a more spatially focused analysis. It examines the natural forest fire patterns of three selected regions along west- ern North America according to their peak months regarding area burned and fre- quency. The summer fire pattern of these months are then linked to the self cali- brated PDSI of the same months (see Section 5.4.3). 5.4.1 Forest fires history of North America from 1981-2010 The forest fire statistics of North America are summarized in Table 5.1. The total annual area burned as well as the total forest fire frequency of all types of fires from 1981-2010 has increased in the United States but decreased or remained static in Canada (see Figure 5.2). These trends are similar if only natural fires are examined (see Figure 5.3). The reported number of forest fire incidents over the last three decades is greater in the United States than in Canada (588,934 vs. 214,700), the total area burned, however, is similar for both countries (see Table 5.1). The mean area burned per fire is, as a consequence, greater in Canada than in the U.S. The distri- bution of the variable area burned is in both cases skewed, which can be seen by looking at the mean and the standard deviation, as well as the frequency distribu- 63 Table 5.1: Forest fire statistics of North America from 1981-2010. This table includes all types of forest fires, despite their cause. Statistics Canada USA Count 214,700 588,934 Total area burned [ha] 66,559,582 62,015,474 Mean area burned [ha] 310 105 Std [ha] 5156 2517 Max. area burned [ha] 624,883 809,717 Location/year of biggest fire NT/1981 AZ/2002 Average annual area burned [ha] 4,285,853 Most fire active months June, July June, July Frequency per cause 53% human 79% human 45% natural 20% natural 2% unknown 1% unknown Area burned per cause 9% human 43% human 90% natural 55% natural 1% unknown 2% unknown tion (not shown here). This indicates that the majority of the forest fires are small incidents and only a few large fires cause most of the area burned. Despite the varying fire seasons across the two countries, June and July result to be the most fire active months with the most incidents reported. Regarding the cause of the fires, the majority of all fire incidents are caused by humans (inten- tionally or unintentionally) (53% Canada, 79% U.S.). The majority of the area burned, however, is caused by natural fires (90% Canada, 55% U.S.) (see Table 5.1). This means that fires caused by humans generally do not reach a great size. This is especially apparent in Canada where 90% of the total area burned is caused by natural fires (lightning). The natural forest fire pattern of North America over the last three decades shows clear hot spots of area burned and forest fire frequency (see Figure 5.4 and 5.5). Hot spots of total area burned are identified in a narrow band that spreads from central Alaska to central Canada and an isolated area in western U.S. (see Figure 5.4). Hot spots of total fire incidents are located in southeastern British 64 1980 1985 1990 1995 2000 2005 2010 2 e + 0 6 4 e + 0 6 6 e + 0 6 Time A r e a  b u r n e d [ h a ] Canada USA 1980 1985 1990 1995 2000 2005 2010 1 0 0 0 0 1 5 0 0 0 2 0 0 0 0 2 5 0 0 0 3 0 0 0 0 Time F i r e  f r e q u e n c y Canada USA Figure 5.2: Annual area burned [ha] (left) and frequency (right) of all types of forest fires over the last three decades, plotted separately for the U.S. and Canada. The straight lines indicate linear trends. 1980 1985 1990 1995 2000 2005 2010 0 e + 0 0 1 e + 0 6 2 e + 0 6 3 e + 0 6 4 e + 0 6 5 e + 0 6 6 e + 0 6 7 e + 0 6 A r e a  b u r n e d [ h a ] Canada USA 1980 1985 1990 1995 2000 2005 2010 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0 6 0 0 0 Time F i r e  f r e q u e n c y Canada USA Figure 5.3: Annual area burned [ha] (left) and frequency (right) of only natu- ral forest fires over the last three decades, plotted separately for the U.S. and Canada. The straight lines indicate linear trends. Columbia and in the western U.S. (see Figure 5.5). Figure 5.4 and 5.5 also display the three regions that were selected for the detailed analysis of forest fire patterns (see black rectangles) in Section 5.4.3. These regions capture the main hot spots of area burned and frequency across western North America while still mostly representing one ecoregion. 65 0 600 1,200300 Km Kernel Density analysis of area burned between 1981 - 2010 « Source: US Federal Fire Occurrence, Canadian National Fire Database Hot spots of area burned by forest fires Selected regions Low Medium High Figure 5.4: Kernel Density analysis of total area burned over the last three decades caused by only natural occurring forest fires. The dark red areas identify hot spots of total area burned. 66 0 600 1,200300 Km Kernel Density analysis of fire frequency between 1981 - 2010 « Source: US Federal Fire Occurrence, Canadian National Fire Database Hot spots offorest fire frequency Selected regions Low Medium High Figure 5.5: Kernel Density analysis of total forest fire frequency over the last three decades caused by only naturally occurring forest fires. The dark red areas identify hot spots of high total forest fire frequency. 67 5.4.2 El Niño types and drought patterns of North America This section assesses the impact of equatorial SSTA, associated with different ENSO types, on drought patterns of North America. In order to do so a SVD analysis was conducted on DJF SSTA and subsequent JJA self calibrated PDSI anomalies over North America. The first four identified modes account for approximately 94% of the total SCF (squared covariance fraction) between the two variables. The first mode explains more than half of the total SCF and represents hence a dominant and highly signif- icant coupling (see Table 5.2).   5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 −0.1 −0.08 −0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.08 0.1   5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 −0.1 −0.08 −0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.08 0.1   5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 −0.1 −0.08 −0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.08 0.1   5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 −0.1 −0.08 −0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.08 0.1 Figure 5.6: Dominant four coupled modes of SVD analysis between DJF SSTA (left) and subsequent JJA PDSI anomalies (right). 68 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 3 st an da rd ize d Am pl itu de 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 −4 −3 −2 −1 0 1 2 st an da rd ize d Am pl itu de Figure 5.7: Standardized amplitude of time series of the first and third SSTA modes of the SVD analysis. Dark (light) shading represents the previ- ously identified EP (CP) El Niño events in Chapter 3. The second and fourth mode of SSTA could resemble transition stages. The interest is, however, focused on the first and third mode. Their pattern of SSTA strongly resemble the EP and CP El Niño anomaly pattern identified in Section 3.4.1 (see left panels in Figure 5.6). This assumption is confirmed by looking at the time series of the first and third SSTA mode (see Figure 5.7). These time series reach high values in some of the EP and CP El Niño years (shading in Figure 5.7 indicates ENSO years). For example, the DJF time series of the first mode shows two peaks in 1982/83 and 1997/98 representing the two strong EP El Niño events. It, however, also shows peaks during 1987/88, 1991/92 and identified CP El Niñno years in 1994/95, 2002/03 and 2009/10. The time series of the third mode shows peaks in 1990/91, 2004/05 and 2009/10. This mixture of EP- and CP El Niño years is discussed in Section 5.5. All four identified modes of SSTA have also been tested for correlation with the pattern representing a possible trend of SSTA over the three decades. The trend analysis identified a significant positive trend in the western equatorial Pacific and no significant trend in the central or eastern equatorial region (not shown). The trend pattern shows significant negative correlation with all four SSTA modes identified by the coupled SVD analysis. The first and third mode are coupled with an opposite drought pattern (PDSI) across western North America: During the first mode, Alaska and most parts of the 69 Table 5.2: Results of SVD analysis between (DJF) SSTA and (JJA) PDSI Mode SCF r significant 1 64% 0.63 yes, α = 1% 2 17% 0.58 yes, α = 1% 3 9% 0.77 yes, α = 1% 4 4% 0.71 yes, α = 1% Total SCF: 94% Canadian west coast experience drier than normal conditions, whereas the western U.S displays more humid conditions. During the third mode, however, most of the western North American coast and in particular the western U.S experience drier than normal conditions, which is the opposite than during the first mode (see right panels in Figure 5.6). The expansion coefficients (time series) of the dominant modes of SSTA and PDSI are tested for their correlation strength. All of the four couplings are highly significant at the α = 1% level. The third and fourth mode reach higher correla- tion values (r=0.77, r= 0.71) than the first and the second mode (r=0.63, r= 0.58), although they explain less of the total SCF (see Table 5.2). This indicates their relative importance. In order to confirm the different PDSI response in North America to the two ENSO types, obtained from the SVD analysis, a spatial correlation analysis was conducted. This analysis correlates DJF values of the two ENSO indices (NINO3 and NINO4), representing the EP- and CP El Niño, with the JJA PDSI values. It confirms the clear wetter (drier) than normal southern (northern) part of western North America correlated with the NINO3 index and hence the EP El Niño events (see left panel in Figure 5.8). Whereas the NINO4 (CP El Niño events) does not correlate with a clear PDSI pattern of western North America. It, however, shows that the southern west coast experiences less moist conditions whereas the north- ern west coast still shows significant drier conditions associated with CP El Niño events (see right panel in Figure 5.8). The PDSI pattern of the correlation with the NINO3 and the leading EOF mode of SSTA are very similar (not shown). They show high positive correlation (wet 70  5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 −0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5   5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 −0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5 Figure 5.8: Correlation values between NINO3 (left) and NINO4 (right) timeseries and JJA PDSI values over North America. Positive correla- tion values indicate moist conditions, whereas negative values indicate drier conditions. The dashed black lines represent the significant areas with p≤0.05. conditions) in the western U.S. and drier conditions in western Canada. The corre- lation between the PDSI and the NINO4 (right panel in Figure 5.8) and the PDSI and the CP El Niño EOF mode (not shown), confirm that the PDSI pattern after CP El Niño events may differ from that after EP El Niño events. After CP El Niño events western Canada also shows dry conditions but the western U.S. does not show as moist conditions than in the correlation pattern with the EP Niño events. The western U.S. experiences hence a significantly different drought pattern during the two ENSO types. The composite PDSI patterns confirm the results of the SVD and the spatial correlation analysis. After EP El Niño events, the PDSI composite shows a clear horizontal split with a wet hot spot in the western U.S. and drier conditions in western Canada (see left panel in Figure 5.9), whereas the signal after CP El Niño events is weaker and not as clear. The western U.S. experiences drier summers compared to summers after EP El Niño events (see right panel in Figure 5.9). This analysis confirms that SSTA in the equatorial Pacific are linked with dif- ferent drought patterns across western North America. Different SSTA patterns can lead to opposite moisture conditions. This different response in the drought pattern could impact the forest fire occurrence in western North America. 71   150° W  120° W   90° W  15° N  30° N  45° N  60° N  75° N −10 −8 −6 −4 −2 0 2 4 6 8 10    150° W  120° W   90° W  15° N  30° N  45° N  60° N  75° N −10 −8 −6 −4 −2 0 2 4 6 8 10 Figure 5.9: Composites of summer (JJA) PDSI patterns of North America after the two ENSO types (left: EP El Niño (1982/83, 1997/98), right: CP El Niño (1994/95, 2002/03, 2004/05, 2009/10)). 5.4.3 Drought and forest fire patterns in western North America As shown in Section 5.4.2, equatorial SSTA seem to be linked to certain drought patterns in North America. Drought can affect evapotranspiration of forests and hence change the condition of fuel and increase the risk of ignition (Westerling et al., 2006). In order to test if drought, indicated by the PDSI, is related with forest fire behaviour, the forest fire patterns of three regions across western North America are examined for this relationship. Only naturally occurring forest fires in these three regions, named central Alaska, southeastern BC and Western U.S, are analysed (see Figure 5.1). The forest fire statistics of the three regions confirm the results of the hot spot analysis in Section 5.4.1. The forest fire frequency is highest in the western U.S. and southeastern BC (see Table 5.3), whereas the total area burned is greatest in central Alaska (see Table 5.4). The length of the fire season increases further south. In order to capture the most fire active months regarding total area burned and frequency but also to account for the variability across the three regions, JJA was selected. This assumption is justified as total monthly area burned (log transformed) and forest fire frequency are positively correlated (see Table 5.5 and Figures 5.10, 5.11 and 5.12). All regions show correlation values of r≥ 0.60 at the α = 5% level. The 72 Table 5.3: Monthly forest fire frequency of three selected regions along the coast of western North America from 1981-2010. Forest fire frequency per month Months Central Alaska Southeastern BC Western U.S. January 1 3 February 10 4 March 9 14 April 2 41 104 May 237 795 646 June 1,584 2,664 337 July 1,466 6,820 9,634 August 133 9,377 9,881 September 6 1,077 2,875 October 54 395 November 4 20 December 3 2 Table 5.4: Monthly area burned of three selected areas along the coast of western North America from 1981-2010. Area burned per month [ha] Months Central Alaska Southeastern BC Western U.S. January 1436 0.1 February 798 85 March 2 2 April 22 52 1169 May 781,585 173,327 32,265 June 7,140,515 58,278 585,640 July 4,383,236 166,232 3,133,909 August 435,274 154,147 3,654,113 September 10,291 7,275 326,458 October 124 6,484 November 1 525 December 0.2 2 73 highest correlation values are obtained in western U.S (r = 0.69). This summer forest fire season (JJA) was then related to averaged PDSI values of the same time period. Table 5.5: Correlation of monthly forest fire frequency and monthly log area burned of the three selected regions along western North America from 1981-2010. Correlation statistics Selected area r Significant Central Alaska 0.60 yes, α = 5% Southeastern BC 0.67 yes, α = 5% Western U.S. 0.69 yes, α = 5% Peak years of JJA area burned and hence forest fire frequency occurred in 1990 and 2004/05 in Alaska (see Figure 5.10). Southeastern BC recorded record years in 1985, 2003 and 2009. In JJA of 2003 more than 1000 fires were reported and caused a total area burned of more than 50,000 ha (see Figure 5.11). Western U.S. experiences peak values of total JJA area burned and forest fire frequency in 1996, 1999/00 and in 2006. During the summer of 2000 a total area of about 800,000 ha burned due to almost 1500 forest fires (see Figure 5.12). The range of the PDSI values represents the variety in summer moisture along western North America. Negative PDSI values represent drier than normal condi- tions, positive values indicate wetter than normal conditions. The index generally reaches values between -10 and +10. Values equal or lower than -3 are considered severe to extreme drought (Dai, 2011b). The region in Alaska shows the smallest range, the PDSI experiences values ranging from only -2.0 to +1.5, whereas the values in western U.S. reach extremes of -4 to +5 over the whole time period. This indicates that the region in western U.S. experiences greater variety in overall moisture conditions compared to the other two regions further north. The PDSI identifies prolonged drier or wetter than normal time periods in all three regions. This is shown by negative or positive values for several consecutive summers. The region in the western U.S., for example, shows two clearly iden- 74 Ar ea  b u rn ed  [h a] 2 4 6 8 10 12 14 16 1985 1990 1995 2000 2005 2010 Fr eq u en cy  0 50 100 150 200 250 PD SI −2 −1 0 1 1985 1990 1995 2000 2005 2010 1981 1981 area burned is log transformed Figure 5.10: Summer (JJA) total area burned (top), forest fire frequency (middle) and averaged PDSI (bottom) for the region in central Alaska. Dark (light) shading represents the previously identified EP (CP) El Niño events in Chapter 3. tifiable time periods of consecutive drier summers than normal. All the summers from 1986 until 1993 showed negative values except for 1989. Another drier than usual time period started in 2000 with five consecutive summers of negative PDSI values (see bottom panel in Figure 5.12). In regard of the general relationship between the averaged drought index and the summed area burned and forest fire frequency, all three regions represent a negative relationship (see scatter plots in Figure 5.13). The strength of this in- verse relationship, varies strongly across the three regions. Not all correlations are significant (see Table 5.6). Only the region in southeastern BC shows significant relationships between the summer area burned and the PDSI, as well as between the forest fire frequency and the PDSI. In general, however, it can be said that when the PDSI values are low (indicating drier than normal conditions) the total area burned is higher and the forest fire frequency is increased and vice versa. How- 75 Ar ea  b u rn ed  [h a] 2 4 6 8 10 12 14 16 1985 1990 1995 2000 2005 2010 area burned is log transformed Fr eq u en cy  0 500 1000 1500 PD SI −4 −2 0 2 1985 1990 1995 2000 2005 2010 1981 1981 Figure 5.11: Summer total area burned (top), forest fire frequency (middle) and averaged PDSI (bottom) for the region in southeastern BC. Dark (light) shading represents the previously identified EP (CP) El Niño events in Chapter 3. ever, extreme dry conditions do not necessarily lead to peak values in the variable area burned or frequency. The region in Alaska only shows a significant negative relationship between the area burned and the PDSI but not between the frequency and the PDSI. The region in western U.S. shows the opposite: The relationship between the area burned and the PDSI is not significant, whereas the relationship between the frequency of forest fires and the PDSI is significant (see Table 5.6). There is no clear relationship detectable between summers after different ENSO years, PDSI values and forest fire behaviour across the three regions of western North America. The region in western U.S. records substantially wet conditions after the two strong EP El Niño events in 1982/82 and 1997/98 and hence lower forest fire frequency and total area burned, whereas after three out of four CP El Niño events the PDSI indicates medium to severe drought conditions. The forest fire patterns during the summers after CP El Niño events shows inconsistent be- 76 Ar ea  b u rn ed  [h a] 2 4 6 8 10 12 14 16 1985 1990 1995 2000 2005 2010 area burned is log transformed Fr eq u en cy  0 500 1000 1500 PD SI −4 −2 0 2 4 6 1985 1990 1995 2000 2005 2010 1981 1981 Figure 5.12: Summer total area burned (top), forest fire frequency (middle) and averaged PDSI (bottom) for the region in western U.S. Dark (light) shading represents the previously identified EP (CP) El Niño events in Chapter 3. haviour. In central Alaska all CP El Niño events lead to wet conditions, whereas the EP El Niño events coincide with variable conditions. The signals are less con- sistent in southeastern BC: EP and CP El Niño events coincide with wet and dry conditions interchangeably. The forest fire responses are for both regions variable and inconclusive. 5.5 Discussion Key findings of this chapter are the different responses in summer drought pat- terns after EP- and CP El Niño events in western North America. Further, the PDSI shows significant negative relationships with forest fire occurrence and area burned in certain regions across the west coats of North America, especially in southeastern BC. 77 −2 0 2 0 5 10 15 lo g ar ea  b ur ne d JJ A PDSI JJA Box A −5 0 5 2 4 6 8 10 12 lo g ar ea  b ur ne d JJ A PDSI JJA Box BC −5 0 5 10 8 10 12 14 lo g ar ea  b ur ne d JJ A PDSI JJA Box WUS Figure 5.13: Scatter plots of summed JJA area burned and averaged JJA PDSI of Central Alaska (left), southeastern BC (middle) and western U.S. (right). Table 5.6: Correlation statistics of summer area burned/forest fire frequency and summer PDSI values Variables Significant r JJA area burned and JJA PDSI Central Alaska yes, α=5% -0.39 Southeastern BC yes, α=5% -0.55l Western U.S. no, α=5% JJA fire frequency and JJA PDSI Central Alaska no, α=5% Southeastern BC yes, α=5% -0.46 Western U.S. yes, α=5% -0.51 The SVD analysis as well as the spatial correlation and the composite analysis pointed out that the summer PDSI pattern over western North America seems to be different after the two ENSO types. Physical credibility to this lagged relationship between winter equatorial SSTA and summer PDSI values is partially provided by Lau and Nath (1996) via the atmospheric bridge concept. The drought analysis showed that the signals associated with the EP El Niño events are clear and significant. The signals after CP El Niño events, however, are weaker and less significant. Especially striking is the difference in the PDSI pattern of the western U.S., summers after EP El Niño events showed substantial wet conditions, whereas summer after CP El Niño events seemed to be drier. There, however, remains some uncertainty regarding the separation of the two ENSO types in the SVD analysis. Variation in the timeseries, especially within 78 the third mode, which resembles the CP El Niño, questions the separation. The time-series of this mode did not reach peak values during all CP El Niño years, previously identified in Chapter 3. Further, the time-series of the first mode, which represents the EP El Niño pattern, also reached high values during some CP El Niño years. This could indicate that the spatial patterns of the EP and CP El Niño events were not fully separated in the SVD analysis. Further interesting was, that the weaker/mixed EP and CP El Niño events in 1987/88, 1991/92 and 1990/91 (identified in Chapter 3) also reached high values in the timeseries of the first and third mode. The weak CP El Niño event in 1990/91 even reached the highest value in the time series of the third mode. The correlation analysis between the spatial pattern of the trend of SSTA and the four dominant modes revealed significant negative correlation between all of them. This could indicate that all four modes partially captured the overall trend in SSTA along with the ENSO signals. This fact could further influence the question- able separation of the SVD analysis. The spatial correlation analysis between the two ENSO indices and the PDSI, as well as the summer PDSI composites after the two ENSO types, however, added credibility to the SVD results. These analyses could confirm the different response in the drought patterns in western North America associated with the two ENSO types. After EP El Niño events the PDSI showed drier than usual conditions across most of Canada. Of particular interest is that British Columbia experienced drier conditions and western U.S. showed wetter conditions than usual. Shabbar and Skinner (2004), Barlow et al. (2001) and Mo and Schemm (2008) obtained similar wet/dry patterns for Canada and the U.S. respectively. The PDSI pattern obtained with the SSTA pattern resembling the CP El Niño event showed the most strik- ingly different response in western U.S., the PDSI indicated substantially drier conditions, which is the opposite than during EP El Niño events. Shabbar and Skinner (2004) conducted a similar SVD analysis but on global SST and PDSI values of Canada between 1940-2002. Their results are therefore only comparable to a certain degree. They identified the overall warming trend as their first mode in global SST. They relate their second and third modes to the ENSO and PDO pattern. In this SVD analysis the first mode resembles their second (ENSO) mode. The fourth mode obtained in this analysis could resemble 79 their third mode, which is related to the PDO. This remains, however, uncertain as only equatorial SST data of a different time period was input into the analysis. Further, a thirty year time period is too short to identify the PDO pattern. The different response of summer PDSI values after EP and CP El Niño events across western North America could potentially be important for forest fire occur- rence. The different PDSI patterns of the SVD analysis are partially supported by the regional analysis of PDSI values and the forest fire patterns across the three regions of western North America (see Section 5.4.3). It resulted that there is a general relationship between negative PDSI values, which indicate dry conditions, and increased forest fire frequency and area burned. Xiao and Zhuang (2007) found a similar general relationship between drought and increased forest fire activity in Canada and Alaska during 1959-1999. Low PDSI values (dry conditions) lead to drier fuels, which enhance the chances for forest fires (Littell et al., 2009). This relationship was, however, not significant in all three examined regions. Comparing the summer PDSI values after the different ENSO years with the forest fire patterns led to no clear detectable relationship in all three examined regions. There was no clear response of forest fire patterns toward different ENSO events. Large forest fires occurred in summers after EP and CP El Niño events. Southeastern BC is the only region, where summer area burned and forest fire frequency showed both a significant negative correlation with the summer PDSI. This means that dry conditions tend to coincide with high fire frequency and large total area burned and wet conditions tend to coincide with a reduced fire activity. The signal of the different ENSO events is, however, inconsistent. The region in southeastern BC records negative PDSI values during the summer after the EP El Niño event in 1997/98, which agrees with the results of the SVD analysis. This drier conditions coincide with a high fire frequency and area burned during this summer. After the EP El Niño event in 1982/83, however, the PDSI shows positive values and hence a decreased fire activity. Summers after all CP El Niño events showed wet conditions, which coincided most of the time with reduced fire activity. The regional analysis of the western U.S. confirmed the PDSI pattern obtained by the SVD and the spatial correlation analysis. The summer PDSI values after both EP El Niño events recorded substantially wet conditions. The forest fire oc- currence was as a consequence reduced, which was shown in a low total summer 80 area burned and fire frequency. This clear connection to ENSO induced moisture variability and the corresponding response in the forest fire pattern in the west of the U.S. has also been established by several other studies (Swetnam and Betan- court, 1990; Westerling et al., 2003). During summers after CP El Niño events, when the SVD and the correlation analysis predicted a drier pattern, the PDSI values indeed indicated drier situations compared to the summers after EP El Niño events. After two out of four CP El Niño events the PDSI recorded even negative values. The forest fire response is accordingly variable. This inconsistent behaviour might explain why the relation- ship is only significant between summer PDSI and fire frequency but not the area burned. Swetnam and Betancourt (1998); Veblen et al. (2000); Westerling et al. (2003) confirm that the summer forest fire risk (related to forest fire frequency) is correlated with drought occurrence. The PDSI patterns resulting from the SVD and spatial correlation analysis in- dicate no anomalies or drier conditions in central Alaska during EP or CP El Niño events. The regional analysis, however, indicates wet summer conditions after all CP El Niño events and variable conditions after EP El Niño events. The forest fire response is accordingly variable. Similarly, to the correlation analysis of western U.S., this might explain why only the relationship between the area burned and the PDSI is significant but not between the PDSI and the forest fire frequency. The exploratory regional analysis revealed another interesting relationship be- tween the forest fire activity and the drought index: From great importance seemed to be the change from wet to dry periods and vice versa. If a summer with wetter conditions is followed by a summer with drier conditions, this usually led to am increase in total area burned and frequency. This could be related to increased fine fuel build up/biomass growth during the wet summer, which increases flamma- bility during the subsequent dry summer (Heyerdahl et al., 2002; Swetnam and Betancourt, 1998). There are several limitations to this analysis that could explain the elusive con- clusions. Although the forest fire databases were cleaned and only naturally caused fires, grouped according to ecoregions, were included in this analysis, there are several factors that were not accounted for, which could potentially affect the out- come. The two data bases are compilations of several forest fire agencies across 81 the two countries. As a consequence of these different sources, quality and accu- racy of the compiled data sets varied. The two selected data sets were not complete nor without errors and suffered from reporting bias. They further only included fires that were reported by the agencies and hence excluded forest fires that were not reported due to their remote location (Swetnam and Anderson, 2008; Canadian Forest Service, 2010). Further, the influence of forest compositions, age structure (Hély et al., 2001), as well as the influence of altitude (Dissing and Verbyla, 2003), the succesional processes of forest (Balshi et al., 2009) or land use changes (Agee, 1996) were not included or considered. Hence, the results need to be interpreted with caution. Stocks et al. (2002) reported that the area burned as well as the forest fire incidents in Canada increased between 1959-1997. This analysis however revealed a decreasing trend of total annual area burned and a stagnant forest fire frequency between 1981-2010. This discrepancy could be explained by the different data sets as well as the different time period. Stocks et al. (2002) analysed a forest fire data base that only included fires greater than 200ha, whereas the data set in this study included fires of any size. The trend could also be biased by the lack of large forest fire years since 2000. The increasing trend of annual area burned and forest fire frequency in the U.S. is supported by results of Westerling et al. (2006). They explain the increase in forest fires and area burned by raised temperatures, earlier spring melts and, as a consequence, by longer fire seasons. Forest fires are a complex phenomenon and there are several different factors that contribute to an incident, not just the climatic conditions. Forest fires are more than just the sum of the original combustion triangle (fuel, heat and oxygen (Agee, 1996) and the fire behaviour triangle (fuel, topography and climate (Agee, 1996)). Pyne et al. (1996) adds that drought does not necessarily lead to combustion, hence only using the PDSI as a predictor is limited. For example, forest fire management including suppression, plays a key role as well. Therefore, excluding the above mentioned factors from a forest fire occurrence analysis and only focusing on cli- mate conditions (drought) could potentially hinder finding a clear connection. Different types of ENSO might influence the weather in western North Amer- ica differently (see Chapter 4) but establishing consistent teleconnections signals is difficult due to the small numbers of clearly identifiable events during the short 82 time period for which continent-wide fire data was available. Summers in the west- ern U.S. seem to experience a clear drought signal related to the EP El Niño, which is also reflected in the forest fire pattern. The drought signal related with the CP El Niño appears to be different, the forest fire signal, however, is inconsistent. Further north the ENSO signal is weaker, especially during summer, which might explain the varying patterns of drought and forest fire behaviour in the different examined regions. Hessl et al. (2004) and Heyerdahl et al. (2002) confirm this inconsistent ENSO signals on forest fires in their study. Heyerdahl et al. (2002) suggest that, in the Pacific Northwest, the ENSO signal influences the fire season more likely indirectly by altering the timing of the snow melt and thus the length of the fire period. There are many other factors that influence summer weather and hence forest fire patterns. As a consequence, it is difficult to clearly filter out, whether the two types of ENSO lead to a different forest fire pattern across western North America. 83 Chapter 6 Conclusion The last three decades of SSTA, weather variables and forest fire patterns were analysed in order to investigate a newly discovered El Niño type, assess its tele- connections and its possible link to forest fires in western North America. The following three sections give concluding remarks and recommendations for each Chapter. Identification of CP and EP El Niño events Not every EP El Niño and CP El Niño event is the same. Their SSTA, as well as their wind and precipitation patterns show different locations and magnitudes. Strong EP El Niño events such as 1982/83 or 1997/98 and strong CP El Niño events such as 2009/10 set the boundaries, but there are also events that fall in between the two extremes, so called mixed or weak events (Kug et al., 2009). Most identification methods rely on indices based on SSTA, on their magni- tude and the period of months they last. As the results from the different index computations showed, identifying weak or mixed events is especially ambiguous and indices only show a coherent picture for the really strong events. Therefore, we cannot split the ENSO events into two binary groups but have to account for weak/mixed events. According to the analysis of SSTA, precipitation and wind analysis of the in- dividual events that were identified by the different indices, all the classifications suggested by Kug et al. (2009) can be confirmed, except for the event in 1990/91, 84 for which the SSTA are too small to be identified as a CP El Niño event. This analysis points out that there is a need for forecasting agencies to recog- nize that there are two types of ENSO events. Only using the NINO3.4 index to classify events should be avoided, as it can lead to misclassification. Further, the scientific community should agree on a single method to distinguish between EP El Niño and CP El Niño events. This study identified specific deficiencies of each index. Using indices that are complex and incorporate a lot of computations might thoroughly capture the char- acteristics of events but are not user friendly. For example, the Kao and Yu (2009) index based on the combined regression EOF computations is not applicable on a forecasting level and is not flexible enough to capture slightly abnormal SSTA pat- terns. Using just one index, the NINO3.4, as suggested by NOAA, mixes up signals from different events and does not acknowledge the existence of the two distinct types. The same problem appears with the NINO4 index. Using these indices is however suggested if the goal is to identify general warming across the equatorial Pacific and not to distinguish between the two ENSO types. NINO3.4 and NINO4 are suitable for this purpose as they capture the majority of both warming events. The NINO3 index has proved to be a good and robust index to monitor the canon- ical type of El Niño, it however also identifies mixed events. To monitor the new type of El Niño the IEMI and the EMI identify coherent events, do however also show weaknesses that can lead to misclassification, as shown in the example of the CP El Niño event in 2002/03. In order to avoid weaknesses and ambiguities, it is suggested to use multiple indices and evaluate them in relation to each other, to identify SSTA as a type of ENSO events. As already suggested by other studies, NINO3 and NINO4 seem to capture the different characteristics of EP- and CP El Niño events due to their location across the equatorial Pacific (Kim et al., 2011; Lee and McPhaden, 2010). In order to exclude mixed events, evaluating the NINO3.4 is also suggested. An event should be identified according to the index that reaches the highest value among NINO3, NINO3.4 and NINO4, given that they reach the threshold of one standard deviation for at least five consecutive months. The standard deviation is calculated for each index individually and incorporates all the months. This classification is simple, based directly on SSTA and does not require a complicated 85 Table 6.1: Suggested classification of ENSO events from 1981-2010 for tele- connection analysis EP El Niño CP El Niño Mixed events 1982/83 2009/10 1987/88 1997/98 2004/05 1991/92 1995/95 2002/03 1990/91 computation of an index. Table 6.1 shows the events that are identified as EP El Niño, CP El Niño and mixed/weak events according to the above described classification. Of the nine events identified by various indices two can be classified as EP El Niño events, four as CP El Niño events, two as mixed events and one as a very weak event (in italics). There is no one correct way of classifying ENSO events but a scientific agree- ment on a single classification scheme for CP El Niño and EP El Niño events would simplify comparison of results across studies. It would further prevent miss- and/or double-classification. Identifying ENSO events properly is in particular important because CP and EP El Niño events may have different teleconnections and thus potentially different economic and environmental impacts. Several studies have shown that grouping all ENSO events into one type can obscure possibly different teleconnection signals because they are averaged together (Larkin and Harrison, 2005b; Kao and Yu, 2009; Weng et al., 2009; Yeh et al., 2009). For example, as shown in the composite analysis, some places in the west Pacific and southeast Asia experience different precipitation patterns during EP and CP El Niño events. Additionally, weaker events might dampen clear signals as well. Therefore, clear classification and distinction between the event types could help people to better prepare and plan for the possibly different impacts of EP and CP El Niño events. 86 Teleconnection of CP- and EP El Niño events In this section of the analysis it was assessed whether the teleconnections can differ between the peak time of past EP and CP El Niño events, using western North America as the study area. Analysing teleconnection patterns separate for each ENSO type seems to lead to more consistent signals. If the signals of the two different types are mixed and analysed together the standard deviation increases compared to analysing them separately. Comparing the teleconnection signals of the CP El Niño events with results of other studies is difficult because other teleconnection analyses are based on differ- ent definitions of CP and EP El Niño events or do not distinguish between them at all. This reinforces the aforementioned problem of disagreement over ENSO classification and the need for a common system of classifying events (see Chapter 3). The results obtained here, however, do support the main message of other tele- connection studies based on the two different types of events (Weng et al., 2007, 2009): The CP El Niño events tend to show, although not strong and clear, differ- ent teleconnection signals than conventional EP El Niño events and analysing them separately is recommended in order to not mix different signals together. This is especially true when the spatial focus is small, as shown in the case study of south- eastern BC. However, the conclusion is unclear as the results are biased by a brief study period and, as a consequence of this, by a small sample size of identified events. Most of teleconnection analyses examined study periods since 1980 (e.g. Weng et al. (2009); Ashok et al. (2007)), as this marks the beginning of reliable data records. This fact, along with the reason that this analysis is linked to a forest fire analysis, which is facing even shorter data records, explains the rather short time frame of thirty years of the here presented analysis. Including more decades and hence past El Niño events such as the strong EP El Niño event in 1972/73 could change and improve the statistical conclusions. Further analysis confirmed that other climate oscillations, such as the AO, have also substantial impacts on the weather in western North America. It is hence 87 difficult to isolate the impacts that are caused by the two different types. Further research in this area is required. Forest fires in North America In order to analyse possible links between ENSO types and forest fires, a database of forest fire from 1981 through 2010 for North America has been created and analysed. Further, a relationship between equatorial SSTA and drought patterns over North America has been established. The forest fire pattern and its connec- tion to drought has then been examined across three regions across western North America. This analysis showed that winter CP El Niño event conditions correlate with a different lagged summer drought pattern than the canonical EP El Niño event. Especially the region in western U.S. experiences different moisture conditions associated with the two ENSO types. This was established with a SVD analysis that included winter equatorial SSTA and summer PDSI values but also with a spatial correlation and a composite analysis. Forest fire frequency and area burned correlate across all three regions, this re- lationship could be used to make predictions based on one of the variables. Further, the PDSI seems to be a good variable for predicting forest fire patterns, however, the PDSI cannot account for various other aspects that influence the forest fire pattern. Only southeastern BC showed a significant correlation between summer drought and forest fire frequency as well as area burned. The other two examined regions in Central Alaska and soutwestern U.S. only showed significant correla- tions between area burned or forest fire frequency and the summer PDSI values. The study found no clear link between the two ENSO types and the pattern of forest fire occurrence in the three examined regions across western North America. This result may, however, reflect the limited time frame of the analysis. Analysing forest fire behaviour in more detail and incorporating more factors, as well as analysing longer time periods could improve the understanding of the relationship between forest fire patterns and climate influences. Further, instead of examining area burned and frequency of actual forest fires (as done in this study), different results could be obtained if only forest fire prone conditions, indicated by 88 a forest fire index (e.g. Canadian Forest Fire Weather Index (Van Wagner et al., 1985)) were assessed. This approach could alleviate the possible bias of not in- cluding forest fire management impacts on forest fire patterns. Detailed forest fire analysis depends on reliable forest fire data, which is a limiting factor, especially on a spatially large scale. Therefore, it is suggested to use proxy data to be able to extend farther back in time (e.g. tree ring data) or focus on smaller regions. With a longer study period, the importance of climate oscillations which both high and low periodicity could be assessed better. 89 Bibliography R. Adler, G. Huffman, A. Chang, R. Ferraro, P. Xie, J. Janowiak, B. Rudolf, U. Schneider, S. Curtis, D. Bolvin, et al. The Version 2 Global Precipitation Climatology Project (GPCP) Monthly Precipitation Analysis (1979-Present). Journal of Hydrometeorology, 4(6):1147–1167, 2003. → pages 5 J. K. Agee. Fire ecology of Pacific Northwest forests. Island Pr, 1996. → pages 56, 82 R. Allan, J. Lindesay, and D. Parker. El Niño, southern oscillation, and climatic variability, volume 1. Aubrey Books Intl Ltd, 1996. → pages 39 B. Amiro, A. Cantin, M. Flannigan, and W. De Groot. Future emissions from Canadian boreal forest fires. Canadian Journal of Forest Research, 39(2): 383–395, 2009. → pages 56 K. Ashok, S. K. Behera, S. A. Rao, H. Weng, and T. Yamagata. El Nino Modoki and its possible teleconnection. Journal of Geophysical Research-Oceans, 112 (C11):C11007, NOV 8 2007. PT: J; NR: 89; TC: 64; J9: J GEOPHYS RES-OCEANS; PG: 27; GA: 230CR; UT: ISI:000250853800001. → pages 1, 7, 8, 9, 10, 11, 12, 13, 15, 24, 27, 39, 40, 41, 87 M. Balshi, A. McGuire, P. Duffy, M. Flannigan, J. Walsh, and J. Melillo. Assessing the response of area burned to changing climate in western boreal North America using a Multivariate Adaptive Regression Splines (MARS) approach. Global Change Biology, 15(3):578–600, 2009. → pages 58, 82 M. Barlow, S. Nigam, and E. Berbery. ENSO, Pacific decadal variability, and US summertime precipitation, drought, and stream flow. Journal of Climate, 14(9): 2105–2128, 2001. → pages 79 H. Bjornsson and S. Venegas. A manual for EOF and SVD analyses of climatic data. CCGCR Report, 97(1), 1997. → pages 13, 60 90 D. Calkin, K. Gebert, J. Jones, and R. Neilson. Forest Service large fire area burned and suppression expenditure trends, 1970-2002. Journal of Forestry, 103(4):179–183, 2005. → pages 56 Canadian Forest Service. National Fire Database - Agency Fire Data. Natural Resources Canada, Canadian Forest Service, Northern Forestry Centre, Edmonton, Alberta., 2010. URL http://cwfis.cfs.nrcan.gc.ca/en CA/nfdb. → pages 6, 82 J. Cohen, J. Foster, M. Barlow, K. Saito, and J. Jones. Winter 2009–2010: A case study of an extreme Arctic Oscillation event. Geophys. Res. Lett, 37:L17707, 2010. → pages 52 Commission for Environmental Cooperation. Ecological Regions of North America, 2009. URL http://www.cec.org/naatlas/. → pages 6 S. Cravatte, T. Delcroix, D. Zhang, M. McPhaden, and J. Leloup. Observed freshening and warming of the western Pacific warm pool. Climate dynamics, 33(4):565–589, 2009. → pages 12 A. Dai. Drought under global warming: a review. Wiley Interdisciplinary Reviews: Climate Change, 2(1):45–65, 2011a. → pages 6 A. Dai. Characteristics and trends in various forms of the Palmer Drought Severity Index during 1900–2008. Journal of Geophysical Research, 116 (D12):D12115, 2011b. → pages 6, 57, 74 M. D. Dettinger, D. R. Cayan, H. F. Diaz, and D. M. Meko. North-south precipitation patterns in western North America on interannual-to-decadal timescales. Journal of Climate, 11(12):3095–3111, DEC 1998. PT: J; NR: 40; TC: 127; J9: J CLIMATE; PG: 17; GA: 151ZY; UT: ISI:000077751700002. → pages 39, 48 H. F. Diaz, M. P. Hoerling, and J. K. Eischeid. ENSO variability, teleconnections and climate change. International Journal of Climatology, 21(15):1845–1862, 2001. → pages 39, 42, 48, 54 D. Dissing and D. Verbyla. Spatial patterns of lightning strikes in interior Alaska and their relations to elevation and vegetation. Canadian Journal of Forest Research, 33(5):770–782, 2003. → pages 82 P. A. Duffy, J. E. Walsh, J. M. Graham, D. H. Mann, and T. S. Rupp. Impacts of large-scale atmospheric-ocean variability on Alaskan fire season severity. 91 Ecological Applications, 15(4):1317–1330, AUG 2005. PT: J; NR: 58; TC: 34; J9: ECOL APPL; PG: 14; GA: 950SE; UT: ISI:000230876900020. → pages 2 M. Flannigan and J. Harrington. A study of the relation of meteorological variables to monthly provincial area burned by wildfire in Canada (1953-80). J. Appl. Meteorol, 27:441–452, 1988. → pages 57 M. Flannigan and B. Wotton. Climate, weather, and area burned. Forest fires: behavior and ecological effects, pages 351–373, 2001. → pages 7, 38, 56 M. Flannigan, B. Amiro, K. Logan, B. Stocks, and B. Wotton. Forest fires and climate change in the 21 st century. Mitigation and Adaptation Strategies for Global Change, 11(4):847–859, 2006. → pages 56, 58 M. D. Flannigan, B. J. Stocks, and B. M. Wotton. Climate change and forest fires. Science of the Total Environment, 262(3):221–229, NOV 15 2000. PT: J; NR: 39; TC: 126; J9: SCI TOTAL ENVIR; PG: 9; GA: 369KQ; UT: ISI:000165064800003. → pages 2, 57 Z. Gedalof, D. L. Peterson, and N. J. Mantua. Atmospheric, climatic, and ecological controls on extreme wildfire years in the northwestern United States. Ecological Applications, 15(1):154–174, 2005. → pages 57 C. Hély, M. Flannigan, Y. Bergeron, and D. McRae. Role of vegetation and weather on fire behavior in the Canadian mixedwood boreal forest using two fire behavior prediction systems. Canadian Journal of Forest Research, 31(3): 430–441, 2001. → pages 82 A. E. Hessl, D. McKenzie, and R. Schellhaas. Drought and Pacific Decadal Oscillation linked to fire occurrence in the inland Pacific Northwest. Ecological Applications, 14(2):425–442, APR 2004. PT: J; NR: 74; TC: 58; J9: ECOL APPL; PG: 18; GA: 805UZ; UT: ISI:000220392500010. → pages 2, 83 E. K. Heyerdahl, L. B. Brubaker, and J. K. Agee. Annual and decadal climate forcing of historical fire regimes in the interior Pacific Northwest, USA. Holocene, 12(5):597–604, 2002. PT: J; UT: ISI:000177949100008. → pages 81, 83 M. Hoerling and A. Kumar. Understanding and Predicting Extratropical Teleconnections Related to ENSO. El Niño and the southern oscillation: multiscale variability and global and regional impacts, page 57, 2000. → pages 38, 39 92 D. Jaffe, W. Hafner, D. Chand, A. Westerling, and D. Spracklen. Interannual variations in PM2. 5 due to wildfires in the western United States. Environmental science & technology, 42(8):2812–2818, 2008. → pages 2, 56 F. Jin. An equatorial ocean recharge paradigm for ENSO. Part I: Conceptual model. Journal of the Atmospheric Sciences, 54(7):811–829, 1997. → pages 31 M. Kanamitsu, W. Ebisuzaki, J. Woollen, S. Yang, J. Hnilo, M. Fiorino, and G. Potter. Ncep-doe amip-ii reanalysis (r-2). Bulletin of the American Meteorological Society, 83(11):1631–1644, 2002. → pages 5 H.-Y. Kao and J.-Y. Yu. Contrasting Eastern-Pacific and Central-Pacific Types of ENSO. Journal of Climate, 22(3):615–632, FEB 2009. PT: J; NR: 43; TC: 24; J9: J CLIMATE; PG: 18; GA: 415AZ; UT: ISI:000263908000010. → pages 1, 8, 9, 11, 12, 13, 15, 17, 26, 27, 29, 31, 35, 85, 86 G. Kiladis and H. Diaz. Global climatic anomalies associated with extremes in the Southern Oscillation. Journal of Climate, 2:1069–1090, 1989. → pages 38 B. Kilgore. Fire in ecosystem distribution and structure: western forests and scrublands. The Bark Beetles, Fuels, and Fire Bibliography, page 81, 1981. → pages 61 W. Kim, S. Yeh, J. Kim, J. Kug, and M. Kwon. The unique 2009–2010 El Niño event: A fast phase transition of warm pool El Niño to La Niña. Geophysical Research Letters, 38(15):L15809, 2011. → pages 10, 12, 32, 85 T. Kitzberger, P. M. Brown, E. K. Heyerdahl, T. W. Swetnam, and T. T. Veblen. Contingent Pacific-Atlantic Ocean influence on multicentury wildfire synchrony over western North America. Proceedings of the National Academy of Sciences of the United States of America, 104(2):543–548, JAN 9 2007. PT: J; UT: ISI:000243445400027. → pages 57 J.-S. Kug, F.-F. Jin, and S.-I. An. Two Types of El Niño Events: Cold Tongue El Niño and Warm Pool El Niño. Journal of Climate, 22(6):1499–1515, MAR 2009. PT: J; NR: 68; TC: 16; J9: J CLIMATE; PG: 17; GA: 431ZT; UT: ISI:000265104000012. → pages 8, 10, 30, 31, 32, 37, 40, 84 N. Larkin and D. Harrison. On the definition of El Niño and associated seasonal average US weather anomalies. Geophys. Res. Lett, 32:L13705, 2005a. → pages 1, 9, 10, 11, 40, 41 93 N. K. Larkin and D. E. Harrison. Global seasonal temperature and precipitation anomalies during El Niño autumn and winter. Geophysical Research Letters, 32(16):L16705, AUG 20 2005b. PT: J; NR: 17; TC: 20; J9: GEOPHYS RES LETT; PG: 4; GA: 957WQ; UT: ISI:000231404400002. → pages 1, 8, 9, 11, 40, 86 N. Lau and M. Nath. The role of the” atmospheric bridge” in linking tropical Pacific ENSO events to extratropical SST anomalies. Journal of Climate, 9(9): 2036–2057, 1996. → pages 78 T. Lee and M. J. McPhaden. Increasing intensity of El Niño in the central-equatorial Pacific. Geophysical Research Letters, 37:L14603, JUL 24 2010. PT: J; NR: 18; TC: 1; J9: GEOPHYS RES LETT; PG: 5; GA: 631DI; UT: ISI:000280326400005. → pages 1, 4, 10, 12, 58, 85 M. L’Heureux, A. Butler, B. Jha, A. Kumar, and W. Wang. Unusual extremes in the negative phase of the Arctic Oscillation during 2009. Geophysical Research Letters, 37(10):L10704, 2010. → pages 52 G. Li, B. Ren, C. Yang, and J. Zheng. Indices of El Niño and El Niño Modoki: An improved El Niño Modoki index. Advances in Atmospheric Sciences, 27(5): 1210–1220, 2010. ISSN 0256-1530. → pages 9, 10, 11, 13, 22, 24, 25, 35, 36 S. Litschert, T. Brown, and D. Theobald. Historic and future extent of wildfires in the Southern Rockies Ecoregion, USA. Forest Ecology and Management, 269: 124–133, 2012. → pages 56 J. Littell, D. McKenzie, D. Peterson, and A. Westerling. Climate and wildfire area burned in western US ecoprovinces, 1916-2003. Ecological Applications, 19 (4):1003–1021, 2009. → pages 61, 80 F. Mesinger, G. DiMego, E. Kalnay, K. Mitchell, P. Shafran, W. Ebisuzaki, D. Jovic, J. Woollen, E. Rogers, E. Berbery, et al. North American regional reanalysis. Bulletin of the American Meteorological Society, 87(3):343–360, 2006. ISSN 0003-0007. → pages 5 A. Meyn, S. W. Taylor, M. D. Flannigan, K. Thonicke, and W. Cramer. Relationship between fire, climate oscillations, and drought in British Columbia, Canada, 1920-2000. Global Change Biology, 16(3):977–989, MAR 2010. PT: J; NR: 51; TC: 2; J9: GLOB CHANGE BIOL; PG: 13; GA: 554EY; UT: ISI:000274419500007. → pages 57 K. Mo and J. Schemm. Droughts and persistent wet spells over the United States and Mexico. Journal of Climate, 21(5):980–994, 2008. → pages 79 94 B. Netz, O. Davidson, P. Bosch, R. Dave, and L. Meyer. Climate change 2007: Mitigation. Contribution of Working Group III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Summary for Policymakers. IPCC: Climate change 2007: Mitigation. Contribution of Working Group III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Summary for Policymakers, 2007. → pages 2 NOAA. NOAA gets U.S. consensus for El Niño/La Niña index, definitions. press release, Camp Springs, Md., 30 Sept, 2003. URL http://www.noaanews.noaa.gov/stories/s2095.htm. → pages 9, 35 W. Palmer and E. U. W. Bureau. Meteorological drought. US Dept. of Commerce, Weather Bureau, 1965. → pages 6 S. Philander. El Niño, La Niña, and the southern oscillation, volume 46. Academic Pr, 1990. → pages 7, 15, 29, 30 C. Price and D. Rind. The impact of a 2-X-CO2 climate on lightning-caused fires. Journal of Climate, 7(10):1484–1494, OCT 1994. PT: J; NR: 44; TC: 77; J9: J CLIMATE; PG: 11; GA: PN511; UT: ISI:A1994PN51100003. → pages 57 S. Pyne, P. Andrews, and R. Laven. Introduction to wildland fire. John Wiley & Sons Inc, 1996. → pages 82 E. Rasmusson and T. Carpenter. Variations in tropical sea surface temperature and surface wind fields associated with the Southern Oscillation/El Niño. Mon. Wea. Rev, 110(5):354–384, 1982. → pages 7, 15 N. Rayner, D. Parker, E. Horton, C. Foll, L. Alex, D. Rowell, and A. Kaplan. Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late Nineteenth Century. J. Geophys.Res., 108, 2003. → pages 4 A. Shabbar. The impact of El Nino-Southern Oscillation on the Canadian climate. Advances in Geosciences, 6:149–153, 2006. doi:10.5194/adgeo-6-149-2006. URL http://www.adv-geosci.net/6/149/2006/. → pages 39, 48 A. Shabbar and W. Skinner. Summer drought patterns in Canada and the relationship to global sea surface temperatures. Journal of Climate, 17(14): 2866–2880, JUL 2004. PT: J; NR: 53; TC: 47; J9: J CLIMATE; PG: 15; GA: 839NF; UT: ISI:000222791100011. → pages 57, 60, 61, 79 95 A. Shabbar, W. Skinner, and M. Flannigan. Prediction of seasonal forest fire severity in Canada from large-scale climate patterns. Journal of Applied Meteorology and Climatology, 50(4):785–799, 2011. → pages 57 B. Silverman. Density estimation for statistics and data analysis, volume 26. Chapman & Hall/CRC, 1986. → pages 59 W. R. Skinner, B. J. Stocks, D. L. Martell, B. Bonsal, and A. Shabbar. The association between circulation anomalies in the mid-troposphere and area burned by wildland fire in Canada. Theoretical and Applied Climatology, 63 (1-2):89–105, 1999. PT: J; NR: 36; TC: 73; J9: THEOR APPL CLIMATOL; PG: 17; GA: 216EA; UT: ISI:000081427200009. → pages 57 W. R. Skinner, A. Shabbar, M. D. Flannigan, and K. Logan. Large forest fires in Canada and the relationship to global sea surface temperatures. Journal of Geophysical Research-Atmospheres, 111(D14):D14106, JUL 26 2006. PT: J; NR: 45; TC: 17; J9: J GEOPHYS RES-ATMOS; PG: 14; GA: 069ZR; UT: ISI:000239489700007. → pages 57 S. Stephens. Forest fire causes and extent on United States Forest Service lands. International Journal of Wildland Fire, 14(3):213–222, 2005. → pages 57 B. J. Stocks, J. A. Mason, J. B. Todd, E. M. Bosch, B. M. Wotton, B. D. Amiro, M. D. Flannigan, K. G. Hirsch, K. A. Logan, D. L. Martell, and W. R. Skinner. Large forest fires in Canada, 1959-1997. Journal of Geophysical Research-Atmospheres, 108(D1):8149, DEC 20 2002. PT: J; NR: 35; TC: 39; J9: J GEOPHYS RES-ATMOS; PG: 12; GA: 650GQ; UT: ISI:000181255400002. → pages 56, 57, 82 T. Swetnam and R. Anderson. Fire climatology in the western United States: introduction to special issue. International Journal of Wildland Fire, 17(1):1–7, 2008. → pages 2, 82 T. W. Swetnam and J. L. Betancourt. Fire - Southern Oscillation Relations in the Southwestern United-States. Science, 249(4972):1017–1020, AUG 31 1990. PT: J; NR: 37; TC: 251; J9: SCIENCE; PG: 4; GA: DX004; UT: ISI:A1990DX00400038. → pages 2, 81 T. W. Swetnam and J. L. Betancourt. Mesoscale disturbance and ecological response to decadal climatic variability in the American Southwest. Journal of Climate, 11(12):3128–3147, DEC 1998. PT: J; UT: ISI:000077751700004. → pages 81 96 D. Thompson and J. Wallace. The Arctic Oscillation signature in the wintertime geopotential height and temperature fields. Geophysical Research Letters, 25 (9):1297–1300, 1998. → pages 52 D. Thompson and J. Wallace. Annular Modes in the Extratropical Circulation. Part I: Month-to-Month Variability*. Journal of Climate, 13(5):1000–1016, 2000. → pages 51 K. Trenberth and D. Stepaniak. Indices of El Niño evolution. Journal of Climate, 14(8):1697–1701, 2001. → pages 9, 11, 13 K. E. Trenberth, J. M. Caron, D. P. Stepaniak, and S. Worley. Evolution of El Nino-Southern Oscillation and global atmospheric surface temperatures. Journal of Geophysical Research-Atmospheres, 107(D7-8):4065, APR 2002. PT: J; NR: 64; TC: 55; J9: J GEOPHYS RES-ATMOS; PG: 22; GA: 609FY; UT: ISI:000178894000009. → pages 7 V. Trouet, A. H. Taylor, A. M. Carleton, and C. N. Skinner. Fire-climate interactions in forests of the American Pacific coast. Geophysical Research Letters, 33(18):L18704, SEP 23 2006. PT: J; NR: 28; TC: 13; J9: GEOPHYS RES LETT; PG: 5; GA: 088QV; UT: ISI:000240827100005. → pages 2, 57 C. Van Wagner, T. Pickett, and S. canadien des forêts. Equations and FORTRAN program for the Canadian forest fire weather index system. Canadian Forestry Service, 1985. → pages 89 T. Veblen, T. Kitzberger, and J. Donnegan. Climatic and human influences on fire regimes in ponderosa pine forests in the Colorado Front Range. Ecological Applications, 10(4):1178–1195, 2000. → pages 81 G. Vecchi and A. Wittenberg. El Niño and our future climate: where do we stand? Wiley Interdisciplinary Reviews: Climate Change, 1(2):260–270, 2010. → pages 4, 38 I. Walker, R. Sydneysmith, D. Allen, K. Bodtker, and D. Bonin. British Columbia. From impacts to Adaptation: Canada in a Changing Climate, Chapter 8: 329–386, 2008. → pages 2 C. Wang and P. C. Fiedler. ENSO variability and the eastern tropical Pacific: A review. Progress in Oceanography, 69(2-4):239–266, 2006. PT: J; NR: 147; TC: 32; J9: PROG OCEANOGR; PG: 28; GA: 060FG; UT: ISI:000238787200007. → pages 10 97 H. Weng, K. Ashok, S. K. Behera, S. A. Rao, and T. Yamagata. Impacts of recent El Nino Modoki on dry/wet conditions in the Pacific rim during boreal summer. Climate Dynamics, 29(2-3):113–129, AUG 2007. PT: J; NR: 76; TC: 32; J9: CLIM DYNAM; PG: 17; GA: 178EI; UT: ISI:000247203600001. → pages 1, 9, 10, 11, 40, 41, 54, 87 H. Weng, S. K. Behera, and T. Yamagata. Anomalous winter climate conditions in the Pacific rim during recent El Niño Modoki and El Niño events. Climate Dynamics, 32(5):663–674, APR 2009. PT: J; NR: 49; TC: 5; J9: CLIM DYNAM; PG: 12; GA: 407JF; UT: ISI:000263358700005. → pages 1, 11, 39, 46, 48, 54, 86, 87 A. Westerling, A. Gershunov, D. Cayan, and T. Barnett. Long lead statistical forecasts of area burned in western US wildfires by ecosystem province. International Journal of Wildland Fire, 11(4):257–266, 2002. → pages 57 A. L. Westerling, A. Gershunov, T. J. Brown, D. R. Cayan, and M. D. Dettinger. Climate and wildfire in the western United States. Bulletin of the American Meteorological Society, 84(5):595–+, MAY 2003. PT: J; NR: 21; TC: 70; J9: BULL AMER METEOROL SOC; PG: 12; GA: 684TT; UT: ISI:000183223200018. → pages 81 A. L. Westerling, H. G. Hidalgo, D. R. Cayan, and T. W. Swetnam. Warming and earlier spring increase western US forest wildfire activity. Science, 313(5789): 940–943, AUG 18 2006. PT: J; NR: 39; TC: 390; J9: SCIENCE; PG: 4; GA: 074MR; UT: ISI:000239817000035. → pages 57, 72, 82 J. Xiao and Q. Zhuang. Drought effects on large fire activity in Canadian and Alaskan forests. Environmental Research Letters, 2:044003, 2007. → pages 57, 63, 80 S. Yeh, B. Kirtman, J. Kug, W. Park, and M. Latif. Natural variability of the Central Pacific El Niño event on multi-centennial timescales. Geophysical Research Letters, 38(2):L02704, 2011. → pages 4, 10 S.-W. Yeh, J.-S. Kug, B. Dewitte, M.-H. Kwon, B. P. Kirtman, and F.-F. Jin. El Niño in a changing climate. Nature, 461(7263):511–U70, SEP 24 2009. PT: J; NR: 28; TC: 23; J9: NATURE; PG: 5; GA: 497SP; UT: ISI:000270082900037. → pages 1, 2, 7, 10, 11, 58, 86 J. Yoon, S. Yeh, Y. Kim, J. Kug, and H. Min. Understanding the responses of sea surface temperature to the two different types of El Niño in the western North Pacific. Progress In Oceanography, 2012. → pages 10 98 J. Yu and S. Kim. Three evolution patterns of Central-Pacific El Niño. Geophys. Res. Lett, 37:L08706, 2010. → pages 10, 28, 32, 36 J. Yu and S. Kim. Relationships between extratropical sea level pressure variations and the central Pacific and eastern Pacific types of ENSO. Journal of Climate, 24(3):708–720, 2011. → pages 36 99

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

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"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0072998/manifest

Comment

Related Items