UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Assessing forest disturbances for carbon modeling : building the bridge between activity data and carbon… Mascorro, Vanessa S. 2014

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

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata

Download

Media
24-ubc_2015_february_mascorro_vanessa.pdf [ 3.47MB ]
Metadata
JSON: 24-1.0165575.json
JSON-LD: 24-1.0165575-ld.json
RDF/XML (Pretty): 24-1.0165575-rdf.xml
RDF/JSON: 24-1.0165575-rdf.json
Turtle: 24-1.0165575-turtle.txt
N-Triples: 24-1.0165575-rdf-ntriples.txt
Original Record: 24-1.0165575-source.json
Full Text
24-1.0165575-fulltext.txt
Citation
24-1.0165575.ris

Full Text

ASSESSING FOREST DISTURBANCES FOR CARBON MODELING: BUILDING THE BRIDGE BETWEEN ACTIVITY DATA AND CARBON BUDGET MODELING   by  Vanessa S. Mascorro  B.Sc., University of Guadalajara, 2003  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  MASTER OF SCIENCE in THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Forestry)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) December 2014  © Vanessa S. Mascorro, 2014 ii  Abstract  Detailed observations of natural and anthropogenic disturbances that alter the forest structure and the distribution of carbon are essential to estimate changes in forest carbon sinks and sources.  Remote sensing is one of the primary sources to provide observations of land cover and land-cover change for carbon studies and other ecological applications due to its ability to monitor the Earth’s surface on a regular and continuous basis. However, observations of change are often not attributed directly to an underlying disturbance type and are not well validated, especially in tropical areas.  The overall objectives of this thesis are to 1) assess forest disturbances (natural and anthropogenic) and derive activity data for carbon budget modeling, and 2) estimate the impact of different activity data on the terrestrial carbon balance for REDD+ in Mexican tropical forests.   To do so, a novel Multi-Source, Multi-Scale Disturbance (MS-D) assessment method was developed to: 1) characterize natural and anthropogenic forest disturbances; 2) obtain land-cover change observations; and 3) attribute land-cover changes to their most likely disturbance driver.  Spatially-explicit layers of major disturbance types were generated in annual time steps for carbon modeling across the Yucatan Peninsula from 2005 to 2010. Using geospatial techniques and regression-tree analysis the MS-D approach successfully attributed 86% of land-cover changes derived from the MODIS satellite imagery to their underlying disturbance cause, creating synergies between remote-sensing products, forest inventory and ancillary datasets.   iii  Four remote-sensing products derived from Landsat and MODIS satellites were then compiled, providing inputs of activity data for carbon modeling with the CBM-CFS3. Two map sequences were generated for each product, with and without attributing land-cover changes to disturbance type with the MS-D approach. Annual carbon fluxes were simulated to compare the impact of: 1) spatial resolution, 2) temporal resolution, and 3) attribution/non-attribution of land-cover changes by disturbance type on carbon flux estimates. The results clearly demonstrated that different choices of satellite imagery and attribution of changes to disturbance types change the estimated carbon balance. This study provides an integral cost-effective approach to derive activity data for carbon modeling, and support policy and decision-making for forest monitoring and REDD+. iv  Preface  This thesis is based on two scientific papers currently in review for which I am the lead author. I was responsible for the primary research development, data collection and data analysis, interpretation of the results and writing of both manuscripts. Overall project oversight, advice on methodology and editorial assistance were provided by Dr. Nicholas Coops. Dr. Werner Kurz was involved in the project development and provided considerable guidance in forest carbon principles and carbon budget modeling, as well as editorial assistance in both chapters. Marcela Olguin provided project oversight on intensive monitoring sites and forest carbon modeling in Mexican ecosystems, including editorial assistance for Chapter 2. Dr. Sarah Gergel provided guidance and editorial comments in landscape ecology and forest disturbances. Dr. Peter Marshall provided guidance and editorial assistance in stand dynamics and forest inventory. Potential publications arising from this thesis include:   Chapter 2: Mascorro, V. S., Coops, N.C., Kurz, W.A., Olguin, M. (2014). Attributing Changes in Land Cover Using Independent Disturbance Datasets: A Case Study Of The Yucatan Peninsula, Mexico. Regional Environmental Change.   Chapter 3: Mascorro, V. S., Coops, N.C., Kurz, W.A., (2014). Choices of satellite imagery changes forest carbon sinks to sources. v  Table of Contents  Abstract .......................................................................................................................................... ii Preface ........................................................................................................................................... iv Table of Contents ...........................................................................................................................v List of Tables ............................................................................................................................... vii List of Figures ............................................................................................................................. viii Acknowledgements ........................................................................................................................x Chapter 1: INTRODUCTION ......................................................................................................1 1.1 Forest Disturbances ......................................................................................................... 1 1.2 Natural and Human-induced Disturbances Shaping Mexican Landscapes .................... 2 1.2.1 Anthropogenic Disturbances ....................................................................................... 3 1.2.2 Natural Disturbances ................................................................................................... 6 1.3 Forest Carbon Monitoring for REDD+ ........................................................................... 8 1.3.1 Forest Carbon Pools .................................................................................................... 9 1.3.2 Carbon Budget Modeling .......................................................................................... 11 1.4 Monitoring Forests from Space with Remote Sensing ................................................. 13 1.5 Research Objectives ...................................................................................................... 18 Chapter 2: ATTRIBUTING CHANGES IN LAND COVER USING INDEPENDENT DISTURBANCE DATASETS:  A CASE STUDY OF THE YUCATAN PENINSULA, MEXICO. ......................................................................................................................................20 2.1 Introduction ................................................................................................................... 20 2.2 Materials and Methods .................................................................................................. 24 2.2.1 Study Area ................................................................................................................ 24 2.2.2 Forest Disturbance Assessment ................................................................................ 25 2.2.2.1 Forest Disturbance Ancillary Data .................................................................... 25 2.2.2.2 National Forest Inventory ................................................................................. 29 2.2.3 Remote Sensing Observations of Land-cover Change ............................................. 31 2.2.4 Attributing Land-cover Changes to Underlying Disturbance Types ........................ 32 vi  2.3 Results ........................................................................................................................... 35 2.3.1 Forest Disturbance Assessment ................................................................................ 35 2.3.1.1 Fires................................................................................................................... 35 2.3.1.2 Hurricanes ......................................................................................................... 36 2.3.1.3 Agricultural Activities ...................................................................................... 37 2.3.1.4 Settlement Expansion........................................................................................ 39 2.3.1.5 Ground-plot Observations of Forest-cover Change .......................................... 39 2.3.2 Remote Sensing Observations of Land-cover Change ............................................. 42 2.3.2.1 Attributing Land-cover Changes to Underlying Disturbance Types ................ 44 2.4 Discussion and Conclusions ......................................................................................... 49 Chapter 3: CHOICE OF SATELLITE IMAGERY AND ATTRIBUTION OF CHANGES BY DISTURBANCE TYPE CHANGES FOREST CARBON SINKS AND SOURCES .....54 3.1 Introduction ................................................................................................................... 54 3.2 Methods and Data ......................................................................................................... 58 3.2.1 Study Area ................................................................................................................ 58 3.2.2 Remote Sensing Data ................................................................................................ 60 3.2.3 Accuracy Assessment of Forest Cover Change Estimates ....................................... 62 3.2.4 Disturbance Attribution ............................................................................................ 63 3.2.5 Pre-Processing Data .................................................................................................. 64 3.2.6 Carbon Budget Modeling .......................................................................................... 65 3.3 Results ........................................................................................................................... 66 3.3.1 Forest Change Detection ........................................................................................... 66 3.3.2 Forest Change Accuracy Assessment ....................................................................... 70 3.3.3 The Impact of Activity Data on Estimates of Carbon Fluxes ................................... 71 3.4 Discussion and Conclusions ......................................................................................... 74 Chapter 4: CONCLUSIONS .......................................................................................................80 4.1 Key Findings ................................................................................................................. 81 4.2 Future Research ............................................................................................................ 83 References .....................................................................................................................................88  vii  List of Tables  Table 1-1 Summary of Studies that Describe Anthropogenic Disturbances in Mexican Landscapes ...................................................................................................................................... 5 Table 1-2 Summary of Studies that Describe Natural Disturbances in Mexican Landscapes........ 7 Table 1-3 Forest monitoring and disturbance assessment studies with remote sensing ............... 16 Table 2-1 Land cover change matrix from 2005 to 2010 based on the MODIS land-cover classification. ................................................................................................................................ 43 Table 3-1 Remote sensing products used as activity data inputs for carbon modeling ................ 62 Table 3-2 Error matrix of Landsat maps ....................................................................................... 70  viii  List of Figures  Figure 2-1 Study area: the Yucatan Peninsula states and land cover types. ................................. 24 Figure 2-2 The Yucatan Peninsula municipalities and spatial stratification units ........................ 26 Figure 2-3 Tracks of tropical hurricanes that crossed the Yucatan Peninsula from 2005 – 2010 28 Figure 2-4 Flowchart of the MSD-D (Multi-Scale, Multi-Source Disturbance assessment). The approach is based on a three step implementation: 1) map type, extent, and location of forest disturbances, 2) derive land-cover change observations, and 3) attribute land-cover changes to their most likely disturbance driver. ............................................................................................. 34 Figure 2-5a Fires registered throughout the Yucatan Peninsula from 2005 – 2010. .................... 35 Figure 2-6 Hurricanes that hit the Yucatan Peninsula from 2005 to 2010. .................................. 37 Figure 2-7 a-b Evidence of Agricultural practices within the Peninsula from 2005 – 2010. ....... 38 Figure 2-8 Settlement impact across the Yucatan Peninsula estimated from INFyS (2004-2009) ground-plot data. ........................................................................................................................... 39 Figure 2-9 Mean basal area (m2 ha-1) obtained from the first and subsequent  NFI cycle, and the difference ...................................................................................................................................... 40 Figure 2-10 Forest cover change across the YP (upper graph) expressed by loss/gain in mean basal area (m2 ha-1), and by spatial unit (bottom graph). *Spatial Units:TrDF: Tropical dry forest, TrHF: Tropical humid forest; C: Campeche Q: Quintana Roo, Y: Yucatan ................................ 41 Figure 2-11 Regression tree analysis showing the reduction in basal area (y) explained by major disturbance types. .......................................................................................................................... 44 Figure 2-12 Estimated harvest areas resulting from the spatial interpolation of the reduction in basal area over the two NFI measurement periods ....................................................................... 45 ix  Figure 2-13 Decision tree analysis - the attribution of disturbance type to the pixels with land-cover change. ................................................................................................................................ 46 Figure 2-14 MODIS Land cover change pixels (a) and final classification of the pixel as disturbance type (b)....................................................................................................................... 47 Figure 2-15 Land-cover change attributed by disturbance type (a), the percentage (b) and the total area changed by class in the period (c). ................................................................................ 48 Figure 3-1 Study area: The Yucatan Peninsula, Mexico. ............................................................. 59 Figure 3-2 Land-cover change maps derived from the different remote sensing products. ......... 67 Figure 3-3 Land-cover change maps derived from the different remote sensing products (thematic). ..................................................................................................................................... 68 Figure 3-4 Percentage of the total area change detected in the study area from 2002-2010 ........ 69 Figure 3-5 Annual area changed by disturbance type derived from Landsat (VCT, Hansen and INEGI) and MODIS remote sensing land-cover change products. .............................................. 69 Figure 3-6 Annual carbon fluxes estimated with different sources of activity data ..................... 71 Figure 3-7 Cumulative difference of carbon fluxes, estimated with different sources of activity data ................................................................................................................................................ 72    x  Acknowledgements  This research was undertaken as part of the “Integrated Modeling and Assessment of North American Forest Carbon Dynamics and Climate Change Mitigation Options” project funded by the Secretariat of the Commission for Environmental Cooperation (CEC). Special thanks to Dr. Karen Richardson; without her continuous support this research would not have been possible. I thank the following institutions for their support for the project, making this research feasible: the Canadian Forest Service, the Mexican National Forestry Commission (CONAFOR), the United States Forest Service (USFS), the Mexico-Norway Project (MNP), the Mexican National Commission for the Knowledge and Use of Biodiversity (CONABIO), the NASA Goddard Space Flight Center, and the North American Land Change Monitoring System (NALCMS).  I am especially grateful to Dr. Richard Birdsey and the US Agency International Development for their generous support of my scientific and professional development, giving me the invaluable opportunity to attend and participate in two international technical workshops. I thank Marcela Olguin from the MNP for her valuable insights, feedback and support throughout the project. I am grateful to Max Fellows and Scott Morken from the Canadian Forest Service for providing the software (CBM-CFS3, Recliner), the computer programming knowledge and assistance on how to run the software for the carbon budget simulations.  Thanks to Shannon Franks and Jeffrey Masek from NASA Goddard Space Flight Center who generously processed and provided the remotely-sensed imagery of the vegetation change tracker (VCT).   Additionally, I acknowledge the following government officials who kindly provided information and data: Rene Colditz from CONABIO, Vanessa Maldonado from MNP, Alejandro xi  Torres from the National Water Commission (CONAGUA), and Carlos Zermeño, Carlos Godinez, Carmen Meneses, Enrique Serrano, Rafael Flores, Miriam Vargas, Rafael Morales and Miriam Melendez from CONAFOR.   I am deeply grateful to my supervisor Dr. Nicholas Coops who has supported me throughout my thesis with his time, patience, and efficient management system; always available to provide me guidance when needed. I offer my enduring gratitude to Dr. Werner Kurz, whose example, insight, encouragement and advice, inspired me and guided me to expand my knowledge and vision. I want to thank my committee members Dr. Sarah Gergel and Dr. Peter L. Marshall for enriching my manuscripts with their ideas and insight. I also recognize and thank my fellow IRSS’s lab mates for the stimulating discussions, their knowledge, company and support.    xii          To my best friend and beloved Alvaro Madero, for lighting up my life with colors, and sharing this dream with me; and to my parents, the Maderos, and the Harabors, for their immense love and support.   1  Chapter 1: INTRODUCTION  1.1 Forest Disturbances Disturbances can be defined as events that shape forests over time and space, by altering their structure, composition and functioning, including changes to the landscape dynamics and regeneration processes (Turner, 1989; Foster et al., 1998; Dale et al., 2001; Franklin et al., 2007; Lorenz & Lal 2010).  These events can originate from natural causes, or be the consequence of human activities, also known as anthropogenic disturbances. Some disturbance events can be a combination of both, such as fire ignition which can be driven by anthropogenic causes, and then spread by natural forces (Dale et al., 2001).    The study of natural and anthropogenic disturbances of forested ecosystem has received increasing attention as they can influence both ecological and environmental processes in a positive or negative way, at various spatial scales (Turner, 1989; Linke et al., 2007),  and impact biological diversity conservation (Naeem et al., 2009), forest succession (Turner, 1989), and global climate change (Turner, 2010). Moreover, it has been widely acknowledged that forest disturbances play a key role in the global carbon cycle ( Randerson et al., 2002; Bosworth et al., 2008; Kurz et al., 2009; Lorenz & Lal 2010). They are one of the main drivers that alter the transfers of forest carbon stocks among biomass and dead organic matter carbon pools, contributing to GHG emissions to the atmosphere and subsequent removals from the atmosphere when forests regrow.    2  A number of attributes need to be considered to disentangle the complexity of disturbances, including their type, timing or frequency, extent or spatial distribution and severity.  According to Linke et al. (2007), the timing of a disturbance refers to the seasonality and frequency of occurrence (the return interval over time).  The size or extent is used to define the area impacted by the event and the form that the disturbance takes (Turner, 2010), whereas the severity of a disturbance describes the degree of the damage and alterations caused to the ecosystem (Linke et al., 2007).   Depending on their scale and magnitude, disturbances can be classified as either major (or stand-replacing), and minor (or non-stand replacing).  Stand-replacing disturbances (e.g. wildfires), typically result in quick and severe alterations to land cover by killing all the trees in the stand, disrupting the stand development on a large-scale (Franklin et al., 2007), and returning it to the initiation stage with significant impacts on the carbon cycle and the ecosystem patterns (Lorenz & Lal 2010). These disturbance types have major implications for the composition, health, and distribution of the species in the forest (Turner, 2010). Alternatively, non-stand replacing disturbances (e.g., low intensity fires, insects and diseases) are less intensive, kill individual or small groups of trees, result in lower tree mortality rates and reduced impacts on the ecosystem (Lorenz & Lal 2010).    1.2 Natural and Human-induced Disturbances Shaping Mexican Landscapes  Mexico has approximately 64.8 million hectares (ha) of forest (FRA FAO, 2010), of which around 49% is temperate, 46% tropical and subtropical and 5% other forest types (INFyS, 2012). Human activities are the major force that shape Mexican ecosystems (Ramírez-Marcial  et al., 3  2001; Read & Lawrence, 2003; Urquiza-Haas et al., 2007; Arredondo-León et al., 2008; Endara Agramont et al., 2012). Direct consequences of human-induced disturbances include deforestation, fragmentation, degradation (Arredondo-León et al., 2008) and changes in species composition and land use. This can lead to decreases in biomass and regeneration, and therefore have a negative impact on the net carbon balance (Ramírez-Marcial  et al., 2001; Urquiza-Haas et al., 2007; Endara Agramont et al., 2012). Below I summarize the most common disturbance types occurring within the area.  1.2.1 Anthropogenic Disturbances Over the broad diversity of ecosystems in Mexico, anthropogenic disturbances have been one of the main drivers that shaped the landscape over time (Ramírez-Marcial et al., 2001; Urquiza-Haas et al., 2007; Endara Agramont et al., 2012) (Table 1-1).  These activities mainly involved urban development, selective and clear-cut logging, shifting cultivation (known as milpa system) and extensive livestock grazing. The direct consequences of these activities have been deforestation, fragmentation, degradation of habitats and alterations to landscape dynamics (Ramírez-Marcial et al., 2001; Urquiza-Haas et al., 2007; Endara Agramont et al., 2012).    In coniferous and deciduous forests in central Mexico, illegal logging and other extractive activities are one of the main causes of disturbance, resulting in large open gaps in the forest canopy that have a negative impact for natural regeneration and conservation (Endara Agramont et al., 2012).  Dickinson et al., (2000) documented that selective logging for wood extraction activities also affected the tropical dry forests in the southern part of Mexico, reducing tree numbers, altering the understory vegetation, and promoting changes in the species composition 4  during succession. They concluded that the gaps created by these activities tended to be larger than gaps originating from natural disturbances, impacting the forest understory and soil substrate, and leaving the forest stands more susceptible to pest and diseases. This process may also result in a drastic change in the regeneration process and the floristic composition in old-growth stands.  Ramírez-Marcial et al. (2001) found most forest logging in the southern mountain rainforest was driven by shifting agriculture where the forest was converted into “milpa” cultivation systems, and livestock grazing. They found that grazing and trampling reduced the soil moisture, increased the probability of fire during the dry season, and reduced the ability of the forest to regenerate due to higher seedling mortality. Read & Lawrence (2003) reinforced these findings; shifting cultivation created large areas of secondary forest, reduced basal area, above ground biomass and tree densities, increased the number of stumps, and led to changes in diversity and species composition. While at fine spatial scales, local species richness and natural regeneration might benefit from the development of forest gaps (Sánchez-Gallén et al., 2010), these gaps also open the way for other disturbance types such as fire (Toledo-Aceves et al., 2009).   5  Table 1-1 Summary of Studies that Describe Anthropogenic Disturbances in Mexican Landscapes  * Location corresponds to: CM: Central Mexico, NWM: Northwestern Mexico, SM: Southeastern Mexico, SEM: Southeast Mexico. * SR: stand-replacing disturbance, NSR: non-stand replacing disturbance, M: mixed impact, stand and non-stand replacing disturbance. * Severity levels correspond to: L: low, M: medium, H: high, L-M: low to medium, M-H: medium to high.   Disturbance Type Location Region ExtentVegetation TypeSR vs NSRSeverity Main findings ReferencesAgriculture SEMSouthern Yucatan Peninsula2 haDry tropical forestSR M-HHigh biomass accumulation during succession mainly influenced by age and not precipitationRead et al. (2003)Agriculture/Forest  plantationsCM & SEMYucatan; Chiapas and Michoacan states6,500 ha Various SR M-HLarge-scale cultivations brings deforestation; small-scale cultivations shows offsets in C emissions from clearing the landSkutsch et al. (2001)Harvesting CMNevado de Toluca National Park. Southwest of the Valle de Toluca51,000 haPine-oak forest typesNSR MGaps opening due to wood extraction increases understory; changes in species composition; grazing and frequent fires reduce regenerationEndara Agramont et al. (2012)Harvesting SEMLos Tuxtlas region - Southeastern of Veracruz899 haTropical rain forestNSR LSmall fragments may promote natural regeneration and species richness following disturbance Sanchez et al. (2010)Harvesting SEMEjido Tres Garantias - Quintana Roo state32,265 haSemi-deciduous tropical rain forestNSR MCanopy openness promotes regeneration but opens gap for other disturbances; existing harvesting procedures may not be enough to sustain regenerationToledo et al. (2009)Harvesting SEMYucatan Peninsula - Quintana Roo and Yucatan States58 haDry tropical forestNSR LSignificant decreases in biomass; some C remains stored;high recovery rates but lower than other tropical forests that retained old-growth treesUrquiza-Haas et al. (2007)Mixed CMSierra Fria region of Sierra Madre Occidental - Aguascalientes state16,300 haPine-oak forestsNSR LShift in species composition; moderate recovery; moderate increases in vegetatation coverChapa-Bezanilla et al. (2008)Mixed SEMSouthern of Quintana Roo state18,000 haDry tropical forestNSR LLogging disturbs the understory vegetation, litter and soil, causing root-sprouting and openning canopy gaps.Dickinson et al. (2000)Mix d SMPueblo Nuevo Solistahuacan - Chiapas state74,000 haMontane rain forestM MChanges in floristic composition; reduction in trees density; high impact in understory vegetationRamıŕez-Marcial et al. (2001)Settlement CMTuxpan River basin in Michoacan state188,750 haPine-oak forestsSR M-HHigh deforestation and land use; trend towards shifting agriculture; shrubland grassland expansion;  scarce forest regenerationArredondo-Leon et al. (2008)6  1.2.2 Natural Disturbances Studies of natural disturbances that affect Mexican ecosystems mainly focus on fires and hurricanes (Table 1-2). In general, more research exists on non-stand-replacing disturbances across the country, where estimates are focused on the effects on local to regional areas. Fulé et al. (1997, 1999) examined fire regimes in pine-oak forests in northern Mexico. They found that frequent, low intensity fires were most common, burning forest understory vegetation, but rarely causing major damage to the overstory and forest canopy. While fires release carbon when burning organic matter, non-stand replacing fires have become a forest management option to limit overall carbon losses by reducing the severity and frequency of large stand-replacing fires (Lorenz & Lal, 2010).  Consequently, large intensive stand-replacing fires occur less often than in the past due to the short fire exclusion lengths, low accumulation of forest fuels, and burning of varying rates at different times (Fulé et al., 1999; Drury, 2006).  In dry tropical forests, Vargas et al. (2008) found that frequent fires decreased soil fertility, reducing forest recovery and therefore carbon accumulation by limiting nutrient provisioning.    Skinner et al. (2008) reconstructed fire regime characteristics in the mixed-conifer forests of the northwest region of Mexico and found an increase in the number of low severity fires over a 209-year period. Their findings suggest that 5-year cycles of warm-wet conditions and low fire activity, tend to promote widespread fires. In southeast tropical montane cloud forests, Asbjornsen et al. (2005) found that almost all vegetation cover was killed due to deep slow-burning ground fires. This was attributed to rapid water drainage from high-elevation karstic soils that are sensitive to extreme droughts. The resulting large severe fires turned the forest into an immediate net source of carbon and non-CO2 greenhouse gases (methane and nitrous oxide). 7  Table 1-2 Summary of Studies that Describe Natural Disturbances in Mexican Landscapes  Disturbance Type Location Region ExtentVegetation TypeSR vs NSRSeverity Main findings ReferencesFire SMChimalapas; Oaxaca state590,993 haTropical mountain cloud forestsSR HWater drainage in karstic soils results in droughts that promote severe fires; huge above and below ground biomass losses due to slow-burning from severe ground firesAsbjornsen et al. (2005)Fire NWMLas Bayas Forestry Reserve of Sierra Madre Occidental - Durango state5000 haMadrean pine-oak forestsNSR L-MLow frequent fires observed; wildfires occur in recurrent dry years and were suggested to promote madrean pines regenerationDrury (2006)Fire NWMLa Michilila Biosphere Reserve of Sierra Madre Occidental - Durango and Zacatecas states 230 haPine-oak forestsNSR LMore low frequent fires; mixed lightning and human-induced ignitions; forests adapted to frequent fire can promote tree growth; therefore C uptakeFulé et al. (1999)Fire NWMSierra Madre Occidental - Durango state200,000 haPine-oak forestsNSR LLow frequent fires observed recurring from 4 to 5 years; fire exclusion promotes wildfires and big loads of wood fuels; low fires promote high regeneration and low oversoty high density; high C uptakeFule et al. (1997)Fire CMManantlán Biosphere Reserve in Jalisco and Colima states3600 haPine-oak forestsM LFire promotes pine dominance and reduce fuel loads; pines regenerate quickly in gaps from wildfiresJardel-Peláez et al. (2006)Fire NWMSierra San Pedro Martir - Baja California state40,655 haMixed needleleaf forestsNSR L-MSuggested 5-year lag between warm/wet conditions will promote wildfires; low frequent fires  found; no wildfiresSkinner et al. (2008)Fire CMMorelos; Mexico and Distrito Federal states4.18 million haTemperate forest & dry tropical forestM M-HVegetation type and slope have a great influence on firest distribution; ENSO years can change the frequeuncy of firesManzo-Delgado et al. (2009)Fire SEMEl Eden Ecological Reserve, Quintana Roo state2,500 haTemperate forest & dry tropical forestM M-HUnfrequent fires promote natural succession, but frequent fires decrease soil fertility and carbon sequestration; local carbon stocks and fluxes depend much in successional agesVargas et al. (2008)Hurricane (Dean) SEMEjido Xhazil Sur -Quintana Roo state25,000 haDry tropical forestNSR MMinor impact in forest composition and species diversity, major impact on the forest structureNavarro-Martínez et al. (2012)Hurricane (Dean) SEM Yucatan state2.68 million haDry medium and low tropical forestsM HMODIS is effective for extent and impact assessment; minor impact detected in pasture and agriculture lands; widespread defoliationRogan et al. (2011)Hurricane (Dean) SEMYucatan Peninsula -  Quintana Roo and Campeche states11,000 haDry tropical forestM MHigh structural damage;  high incidence of sprouting; changes in species composition; low mortality ratesVandecar et al. (2009)Hurricane (Gilbert; Roxanne)SEMQuintana Roo state2 haMedium-statured semi-evergreen forestM MMinor change in species composition; decrease in biomass; increase in deadwood and litter; long-term mortality; reduced forest healthWhigham et al. (2003)8  * Location corresponds to: CM: Central Mexico, NWM: Northwestern Mexico, SM: Southeastern Mexico, SEM: Southeast Mexico. * SR: stand-replacing disturbance, NSR: non-stand replacing disturbance, M: mixed impact, stand and non-stand replacing disturbance. * Severity levels correspond to: L: low, M: medium, H: high, L-M: low to medium, M-H: medium to high.  Hurricanes are another type of natural disturbance that mainly affects tropical oceanic regions due to their proximity to the coast and warm temperatures (Foster et al., 1998) These storms mainly occur in the coastal areas of the Yucatan Peninsula, where the tropical forests are frequently altered by hurricanes of differing magnitude based on the Saffir-Simpson scales (Whigham et al., 2003; Rogan et al., 2011; Vandecar et al., 2011). Hurricanes tend to exhibit long-term impacts expressed by low tree mortality rates, high defoliation of the forest canopy, few changes in species composition, decreases in biomass of living trees, and opening gaps for subsequent fires (Whigham et al., 2003). In forests under management, Navarro-Martínez et al. (2012) found few changes in forest composition and species diversity, but major impacts on the structural characteristics. They found that the understory vegetation was more vulnerable since small trees do not withstand hurricane winds as well as bigger trees.   1.3 Forest Carbon Monitoring for REDD+ Carbon is an essential element for life, providing fiber, food and energy (Bosworth et al., 2008), and it directly impacts the climate system through increasing atmospheric CO2 concentrations (Chapin et al., 2009; Covey & Orefice, 2009; Spalding, 2009). Forest ecosystems play a key role in the global carbon cycle (Bosworth et al., 2008; Chapin et al., 2009; Turner, 2010). They have the potential to reduce the increasing concentrations of atmospheric greenhouse gas by sequestering large amounts of atmospheric carbon through plant photosynthesis storing it in the 9  vegetation and soils for long periods of time (Randerson et al., 2002; Covey & Orefice, 2009; Kurz et al., 2009). According to Pan et al., (2011) estimates, forest ecosystems contain one of the largest carbon sinks in the world. However, carbon emissions from deforestation and forest degradation are the second largest source of carbon dioxide CO2 emissions into the atmosphere after the burning of fossil fuels (IPCC, 2007; Spalding, 2009; Houghton et al., 2012).   As climate change negotiations evolve, the United Nations Framework Convention on Climate Change (UNFCCC) has urged countries to take actions aimed at Reducing Emissions from Deforestation and forest Degradation (REDD). This was extended to include the role of conservation, sustainable forest management, and enhancement of forest carbon stocks (the REDD+ “plus”). The successful development of national and regional strategies intended to reduce greenhouse gas emissions through REDD+ require an improved understanding of forest carbon dynamics and drivers of changes in forest carbon sinks and sources. Hence, countries seeking to reduce their greenhouse gas emissions for climate change mitigation actions are undertaking efforts to develop comprehensive forest monitoring systems for REDD+.  1.3.1 Forest Carbon Pools According to the Intergovernmental Panel on Climate Change (IPCC) Guidelines (IPCC, 2006), terrestrial carbon can be divided into five pools: above-ground and below-ground biomass, that together make up the biomass pool; litter and dead wood, that can also be called the dead organic matter (DOM) pool; and finally, the soil organic carbon pool (SOC), also known as the soil organic matter (SOM) pool. Each will be explained in more detail below.    10  Aboveground biomass includes all the living plants and woody living forms on the land surface, including trees, shrubs, herbs, stems, stumps, branches and foliage (IPCC, 2006). It is in this pool where the main process of CO2 uptake occurs during photosynthesis, mainly inside the plant leaves. Plants absorb energy from the sun and use it to assimilate the atmospheric CO2 and fix it into plant biomass during photosynthesis (Lorenz & Lal, 2010). The belowground biomass pool - comprised of fine roots that are less than 2 mm diameter - stores the carbon in the biomass of the living roots.  Carbon quantification of this pool is often disregarded due to the inherent complexity to distinguish it from the soil organic matter or the litter pool (IPCC, 2006).   The dead organic matter pool is comprised of the litter pool and the dead wood pools. The litter pool contains the detritus of leaves, fruits, flowers, twigs and small branches, excluding larger diameter (10 cm) wood (Brown & Lugo, 1982) in various states of decomposition on the soil organic layer. This includes only litter which is larger than 2 mm diameter from the soil organic matter pool and less than 10 cm diameter proposed for the dead wood pool (IPCC, 2006). Depending on root exudates, microbial metabolism and fragmentation, litter can take from months to years to decay, releasing carbon into the atmosphere and entering into the soil at lower turnover rates (Lorenz & Lal, 2010).  The dead wood carbon pool is comprised of all woody debris not included in the litter pool. This pool encompasses all the dead wood on the forest surface, standing dead trees, stumps, dead logs, coarse woody debris and dead roots, that have a diameter larger than or equal to 10 cm (IPCC, 2006). In mature forests, this pool can contain up to 20% of the aboveground biomass and can be the same size as the litter pool (Lorenz & Lal, 2010).  11  Price & Ashton (2009) define the soil organic matter pool as a variety of minerals and compounds that include fine roots, litter and plant residues. Bacteria and other microbial organisms decompose matter in this pool, releasing most into the atmosphere and turning the remainder it into recalcitrant organic carbon compounds that can drain into deeper layers of the soil. As a result, carbon gets released into the atmosphere during the process. The IPCC (2006) suggests including all live and dead fine roots, and dead organic matter with  diameters less than or equal to 2 mm found at the depth of the first 30 cm of the soil profile in this pool. These limits can vary from one country to another, according to their national specifications.  1.3.2 Carbon Budget Modeling Quantifying and monitoring terrestrial carbon sinks and sources remains a challenge despite increasing efforts to address them. Tracking carbon changes triggered by natural and anthropogenic disturbance events remains a complex and difficult endeavor accompanied by high uncertainties (Spalding, 2009). Estimates obtained for a particular flux strongly depend on the method selected, the measurement time, scale and the definition of the components included in the approach (Chapin et al., 2006).  Carbon models provide a sound basis to simulate terrestrial carbon dynamics and quantify the forest carbon stocks and changes integrating data from different spatial and temporal scales (Birdsey et al., 2013a) Two types of modeling approaches are commonly used for carbon studies, empirical and process-based models. Process-based models typically require exhaustive information to explicitly simulate the physical and biological processes involved in photosynthesis, carbon allocation and respiration and therefore, data are more difficult to obtain, 12  calibrate and compute for large scales (Adams et al., 2013). Empirical models, on the other hand, are typically simpler, as their modeling approach is based on an understanding of the relations and interactions of different mechanisms involved in forest carbon dynamics (Adams et al., 2013).   The carbon budget model of the Canadian forest sector (CBM-CFS3) is an empirical model developed by the carbon accounting team of the Canadian Forest Service to quantify forest carbon emissions and removals at the stand, regional, or national level (Kurz et al., 2009). Compliant with IPCC guidelines, it uses algorithms to quantify transfers among the five terrestrial carbon pools, the atmosphere and the forest product sector (Kull et al., 2011). This model can incorporate information on forest disturbance events allowing the simulation of different mitigation scenarios to explore their effects on the forest carbon pools. Following disturbance, the CBM-CFS3 simulates the yield curve transitions and succession dynamics, forest growth and carbon allocation in the biomass, litter, dead wood and soil pools. Based on the stage of stand development and ecological characteristic, it applies litterfall and decomposition rates and uses yield curves to simulate forest growth, mortality and regeneration (Kull et al., 2011).     Since each disturbance event impacts carbon pools in different ways, generic carbon estimates that do not differentiate by disturbance type and severity, over space and time, may result in incorrect assumptions and errors in accounting of carbon fluxes (Spalding, 2009). This is very important, as anthropogenic and natural disturbances can turn forest ecosystems into sources of CO2 emissions, where the respiration of plants, soil, litter and dead woody debris surpass the net 13  primary productivity (Brown, 2002). Likewise, the time that it takes to recover and become a renewed carbon sink can vary from months to centuries depending on the new plants growth rates and dead organic matter decay rates (Covey & Orefice, 2009).    1.4 Monitoring Forests from Space with Remote Sensing Detailed observations of the land cover and land-cover change are essential for sustainable forest management, carbon accounting and monitoring, environmental and biodiversity conservation studies and other ecological applications (Hansen et al., 2008; Hayes & Cohen 2007; Pflugmacher et al., 2012; Potapov et al., 2008; Wulder & Coops, 2014). Spatially-explicit information is relevant to characterize the effects of natural and anthropogenic disturbances on vegetation change (DeFries et al., 2002; Cohen et al., 2010).  As each disturbance affects the forest differently, the carbon dynamics associated with clear-cut logging, wildfire, agriculture, insect outbreaks and non-stand replacing disturbances, also differ and need to be mapped separately (Masek et al., 2011). While there are many examples of research towards this goal (e.g., Hilker et al., 2009; Masek & Collatz., 2006; Potter et al., 2012), further research is necessary to distinguish the type or cause of forest disturbance for accurate estimations of subsequent carbon emissions into the atmosphere (Schroeder et al., 2011).  Remotely-sensed data are a crucial source of information to detect patterns in the landscape and identify forest attributes over a range of spatial scales on a regular and continuous basis (Coops et al., 2006; Healey et al., 2005). Selecting the appropriate source of satellite data to characterize disturbances depends on the spatial, temporal, spectral and radiometric attributes of the data and the target of interest (Coops et al., 2006).  At landscape and regional scales, Masek & Collatz 14  (2006) suggest that high spatial resolution data are likely better able to quantify the effects of natural and human-induced disturbances on the forest. However, despite the fact that high spatial resolution imagery provides more spatially-detailed information about landscape pattern, the associated high acquisition costs and time-consuming constraints of data processing, may preclude their use (Coops et al., 2006). Consequently, researchers often have to adapt to work with the most affordable and/or available data, instead of using the most suitable ones (Hansen et al., 2008).  Cost limitations cannot always be solved by developing algorithms. It is necessary to integrate different sources of geospatial data (Hansen et al., 2008). Various studies have diversified the use of satellite imagery and other datasets, at multiple scales and resolutions, to fill gaps and improve the process of monitoring ecosystem condition and changes (Hayes & Cohen, 2007; Potapov et al., 2008; Wulder et al., 2010).   For instance, the Moderate Resolution Imaging Spectroradiometer (MODIS) is one of most commonly-utilized sources for monitoring  vegetation cover and changes (Cohen et al., 2002) with low spatial resolutions (250m to 1km) covering large spatial extents of area (Coops et al., 2006). This sensor was developed to provide data suitable for regional and global applications with high frequency of observations, allowing users to track ecosystem dynamics with a coarse spatial resolution on a daily basis at low cost (Pouliot et al., 2009; Wulder et al., 2010).   MODIS data have been used for a wide range of forest monitoring applications including quantifying global forest cover loss (e.g., Hansen et al., 2010), assessing forest stand-replacing disturbances (such as clear-cuts and wildfire) (e.g., Pouliot et al., 2009), detecting changes in forest cover (e. g., Hayes et al., 2008), and predicting terrestrial carbon sinks and sources (e.g., 15  Potter,  2012). However, the coarse resolution of this sensor requires aggregated data to improve the correspondence between stand-level observations and the satellite imagery (Hayes et al., 2008).  Landsat Thematic Mapper (TM) imagery is one of the foremost sources of remote sensing data for characterizing physical attributes of the forest cover at landscape to regional scales, due to its low cost (freely available), high spatial resolution (30 m), and observation continuity since 1972 (Cohen et al., 2010; Coops et al., 2006). Having repeated observations over such a long term period can provide information on historical vegetation trends to estimate recovery rates and improve understanding of disturbance patterns, current forest structure and degradation processes (Pflugmacher et al., 2012). As well, it can improve the detection of medium and low intensity disturbances that result in subtle changes of forest cover, providing more detailed information about the landscape than MODIS (Cohen et al., 2010). Examples include insect outbreak, forest fire (Pflugmacher et al., 2012) and selective logging (Matricardi et al., 2010).    Landsat imagery has been found to provide accurate and efficient measures of vegetation cover, such as canopy degradation and regeneration (Matricardi et al., 2010), live above-ground biomass, basal area, stand height and fire-induced tree mortality (Pflugmacher et al., 2012). However, the relatively small coverage per scene (185 x 185km), the frequency of observations (every 16 days) along with recurrent cloud coverage, may limit its use in some areas, such as the tropics (Hayes et al., 2008). A summary of a select set of disturbance studies using remotely sensed data is shown in Table 1-3.   16   Table 1-3 Forest monitoring and disturbance assessment studies with remote sensing Study Satellite Sensor Resolution Temporality Type Reference Characterizing 23 years of stand replacement disturbance Landsat MSS, TM 30m 16-day Disturbance Cohen et al., 2002 Detecting trends in forest disturbance and recovery Landsat TM 30m 16-day Disturbance Cohen et al., 2010 Carbon emissions from tropical deforestation NOAA AVHRR 1.1km daily Land-cover change DeFries et al., 2002 Assessment of  fire and burn severity Landsat 7 ETM+ 30m 16-day Disturbance French et al., 2008 Monitoring forest cover and change TERRA & Landsat 4,5 and 7 MODIS, MSS, TM, ETM+ 250m, 500m, 1km, 30m daily, 8-day, 16-day Land-cover change Hansen et al., 2008 Quantification of global gross forest cover loss TERRA & Landsat 7 MODIS ETM+ 250m, 30m daily, 16-day Land-cover change Hansen et al., 2010 Assessing tropical forest cover change with multiple scales TERRA & Landsat 5, 7 MODIS, TM, ETM+ 250m,500m, 30m daily, 8-day,  16-day Land-cover change Hayes & Cohen, 2007 Estimation change in forest cover from multi-year MODIS TERRA MODIS 250m, 500m, 1km daily Land-cover change Hayes et al., 2008 Comparison of Tasseled Cap-based data for forest disturbance detection Landsat 5, 8 TM, ETM+ 30m 17 day Disturbance Healey et al., 2005 New data fusion model to map forest disturbance TERRA & Landsat 5 MODIS, TM 500m, 30m 8-day, 16 day Disturbance Hilker et al., 2009 Estimating forest carbon in a disturbed landscape Landsat TM, ETM+ 30m 16 day Disturbance Masek & Collatz, 2006 17  Study Satellite Sensor Resolution Temporality Type Reference Forest disturbance mapped from a decadal record Landsat TM, ETM+ 30m 16 day Disturbance Masek et al., 2008 Assessment of forest degradation by selective logging and fire Landsat TM, ETM+ 30m 16 day Disturbance Matricardi et al., 2010 Using disturbance history to predict forest structure Landsat TM, ETM+ 30m 16 day Disturbance Pflugmacher et al., 2012 Combining imagery to estimate and map boreal forest cover loss TERRA & Landsat 5, 7 MODIS, TM, ETM+ 250m, 30m 8-day, 16 day Land-cover change Potapov et al., 2008 Terrestrial Ecosystem Carbon Fluxes Predicted from large-scale disturbance modeling TERRA MODIS 250m daily Disturbance Potter et al., 2012 Evaluation of annual forest disturbance using static decision tree TERRA MODIS 250m daily Disturbance Pouliot et al., 2009 Mapping wildfire and clearcut harvest disturbances in boreal forests Landsat TM, ETM+ 30m 16 day Disturbance Schroeder et al., 2011 Multiscale satellite and spatial information in support of large area forest monitoring TERRA & Landsat 5, 7 MODIS, TM, ETM+ 250m,500m, 30m daily, 8-day,  16-day Land-cover change Wulder et al., 2009 Continuous monitoring forest disturbance using all available imagery Landsat 5, 7 TM, ETM+ 30m 16 day Disturbance Zhu et al., 2012    18  1.5 Research Objectives The overall objective of the research presented in this thesis was to support the development of science-based decision support tools to assess forest disturbances (natural and anthropogenic) using remote sensing imagery for carbon budget modeling. The latter to derive spatially-explicit layers of activity data for carbon models to quantify consequent GHG emissions and removals in Mexican tropical forests.  To meet the research objective four questions were posed:  1. Which forest disturbance types, extent and severity can be identified with currently available remote-sensing products, forest inventory and ancillary datasets in Mexico’s Yucatan Peninsula?  2. How can remotely-sensed observations of changes in land cover be characterized by disturbance type to provide activity data as inputs for carbon budget modeling?  3. What remote sensing sources can be used to reduce uncertainties in the characterization of forest disturbance assessment for carbon modeling?  4. What is the impact of contrasting remote sensing products derived with different spatial and temporal resolutions on estimates of forest carbon sinks and sources?   19  Chapter 2 derives an integral methodological approach to map the extent, location and severity of major natural and anthropogenic disturbances. Multiple sources at multiple scales from remote sensing products, forest inventory and historical records of forest disturbances are integrated to map forest disturbances. Spatially-explicit layers of fires, hurricanes, settlement expansion, agricultural activities, and harvesting areas are generated for carbon modeling.  These maps are integrated with land-cover change observations derived from MODIS, attributing the changes to the most likely disturbance driver to derive activity data for carbon modeling.  Chapter 3 examines the impact of activity data on changes in carbon emissions and removals from contrasting remote sensing products. In addition to the activity data generated from MODIS in Chapter 2, three remote sensing products derived from Landsat with different approaches and temporal resolutions are included. Landsat observations of land-cover change are also attributed by disturbance type and used as inputs of activity data to examine their impacts on the carbon estimates. These four data sources, annual forest carbon fluxes are simulated with the CBM-CFS3 model. Variations in three main aspects of the activity data on changes of carbon fluxes are examined and discussed: spatial resolution, temporal resolution and attribution of changes to their underlying disturbance driver.  Finally, Chapter 4 discusses key findings, conclusions and implications of this study, including recommendations for future research.   20  Chapter 2: ATTRIBUTING CHANGES IN LAND COVER USING INDEPENDENT DISTURBANCE DATASETS:  A CASE STUDY OF THE YUCATAN PENINSULA, MEXICO.  2.1 Introduction Forested ecosystems are valuable, not only for ecosystem services provisioning (MEA, 2005) and biodiversity conservation (Naeem et al., 2009), but also for their crucial role to mitigate climate change (Nabuurs et al., 2007; Bosworth et al., 2008; Turner, 2010; Pan et al., 2011). Sustainable management of forested ecosystems is critical for both developed and developing countries to reduce net global carbon emissions and stabilize atmospheric carbon dioxide concentrations to avoid “dangerous anthropogenic interference with the climate system” (UNFCCC, 1992). Moreover, forest ecosystems have the ability to offset increasing atmospheric carbon (C) concentrations by sequestering large amounts of carbon and storing it in the vegetation and soils for long periods of time (Bosworth et al., 2008; Chapin et al., 2009; Kurz et al., 2009; Lorenz & Lal, 2010).   Forest disturbances play a critical role in the terrestrial carbon cycle. They alter the size, composition and distribution of forest carbon pools and transfer carbon into the atmosphere (Dale et al., 2001; Kurz et al., 2009; Spalding, 2009; Lorenz & Lal, 2010). Disturbances also affect forest structure and post-disturbance carbon dynamics. After the burning of fossil fuels (e.g., vehicles, power generation, etc.), permanent conversion of forested land to other land uses together with forest degradation are the second major cause of carbon emissions (DeFries et al., 21  2002; IPCC, 2007; Spalding, 2009; Hansen et al., 2010; Houghton et al., 2012). These activities have both natural and anthropogenic causes.   Based on the amount of vegetation removed, disturbances can be categorized as stand replacing or non-stand replacing (Oliver & Larson 1996). Stand-replacing disturbances, such as wildfires and land clearing for agricultural purposes, generally result in severe alteration or removal of the forest structure, killing or removing all the trees (Franklin et al., 2007). In the case of fire, significant amounts of carbon are transferred into the atmosphere from the combustion of biomass and dead organic matter, bringing the forest back to the initiation stage (Lorenz & Lal 2010). In this stage, the forest is a net carbon source because carbon emissions from dead organic matter and soil carbon pools exceed removals by the regrowing vegetation. Eventually, depending on the degree and speed of regrowth, the net carbon balance becomes positive once again and the forest becomes a carbon sink (see Fig. 1 in Kurz et al., 2013).   Non-stand-replacing or minor disturbances (e.g., low-intensity fires, low-level insect or disease infestations, or selective logging) affect forested landscapes at a finer spatial scale, damaging or killing individual trees or small groups of trees. This reduces the overall forest productivity and growth, and thus carbon uptake; however, it results in a smaller proportion of the ecosystem carbon returned into the atmosphere compared to stand-replacing events, and the existing forest often remains a carbon sink in the years after such disturbances.  Measuring and monitoring forest carbon stocks and stock changes remains a complex endeavor accompanied by high uncertainties (Chapin et al., 2006; Spalding, 2009). Despite increasing 22  efforts to assess gross forest cover loss at different spatial and temporal scales, attributing the forest cover loss to clear underlying causes remains an important challenge for carbon and ecosystem monitoring (DeFries et al., 2007; Hansen et al., 2010; Houghton et al., 2012; Potter et al., 2012). Until these causes and drivers of natural and human-induced forest cover change and the subsequent forest recovery and succession dynamics are understood, we cannot fully quantify specific contributions to the carbon balance (Spalding, 2009; Kurz, 2010; Masek et al., 2011; Schroeder et al., 2011). Knowledge of forest disturbance types is required to estimate impacts on carbon stock changes and the associated greenhouse gas (GHG) emissions (Kurz et al., 2009; Spalding, 2009). For example, land-clearing with or without fire is associated with differences in the amounts, timing and composition of CO2 and non-CO2 GHG emissions.  Consistent methodologies and protocols for modeling and reporting carbon emissions and removals are a key component to developing and implementing regional and national strategies to reduce GHG emissions.  In Mexico, the government is conducting several initiatives to develop science-based decision support models and tools to improve ecosystem carbon reporting and to provide policy makers with information on the consequences of REDD+ strategies (SEMARNAT 2009; LGCC 2012). This includes a monitoring, reporting and verification (MRV) system for REDD+ (Reducing Emissions from Deforestation and Forest Degradation).   To fulfill Mexico’s needs to meet numerous national and international commitments to reduce GHG emissions, quantifying forest disturbances is required by disturbance type. Our knowledge of the extent and impact of different disturbance types has serious gaps that can be addressed using remote sensing land-cover change products, together with other geospatial datasets that can 23  improve our understanding of the processes driving ecosystem condition and change (Coops et al., 2006; Hayes & Cohen, 2007; Hansen et al., 2008; Potapov et al., 2008; Wulder et al., 2010). Furthermore, synergies between these sources are deemed a fundamental cost-effective solution for developing and maintaining REDD+ MRV systems (De Sy et al., 2012; Birdsey et al., 2013a).  In this study, I developed a comprehensive approach, the Multi-Scale, Multi-Source Disturbance (MS-D) assessment, to 1) map major forest disturbances (natural and anthropogenic) for carbon modeling, 2) obtain land-cover change observations, and 3) attribute land-cover changes to their most likely disturbance driver.  Annual spatially-explicit layers were generated for major disturbance types from 2005 to 2010 for carbon modeling across the Yucatan Peninsula, an area of special interest as it is an “early action” region for REDD+. These disturbances include fires, hurricanes, human settlements, harvesting and agricultural activity. Land-cover changes were then derived from Moderate Resolution Imaging Spectroradiometer (MODIS) satellite imagery from 2005 to 2010. Integrating data from remote sensing products, forest inventory data and ancillary datasets through geospatial techniques and regression tree analysis, the MS-D approach then attributed 86% of land-cover changes derived from MODIS to their most likely disturbance cause. Finally, I conclude with recommendations on how these data can be used to quantify changes in carbon dynamics within carbon budget models to support climate change policy and decision making.   24  2.2 Materials and Methods 2.2.1 Study Area The focus of this study is the Yucatan Peninsula (YP) that comprises three Mexican states: Yucatan, Campeche, and Quintana Roo, located in Southeastern Mexico, between the Gulf of Mexico and the Caribbean Sea (Figure 2-1). The topography of the region is characterized by flat limestone areas and karst reliefs, where the dominant soil types are comprised of well-drained rendzinas, shallow rocky lithosols, and in some areas chromic luvisols, pellic vertisols and gleysols (Urquiza-Haas el al., 2007). The region has a mean annual temperature of 25° Celsius, resulting in a tropical subhumid climate with precipitation ranging from 900 to 1400 mm per year with dry winters and wet summers (Vandecar et al., 2011). Accordingly, the vegetation of the region has been mainly classified as tropical forests that include semi-evergreen forest, semi-deciduous, deciduous forest, and mangrove forests (Urquiza-Haas et al., 2007).  Figure 2-1 Study area: the Yucatan Peninsula states and land cover types.  *Derived from INEGI’s Land Use and Vegetation Series IV 25  2.2.2 Forest Disturbance Assessment  2.2.2.1 Forest Disturbance Ancillary Data Several datasets were compiled to map the location and severity of disturbances from 2005 – 2010: harvesting, human settlements, fires, hurricanes and clearing for agricultural activity. These data are available at the national scale for almost all the states of the country, including the three states that comprise the YP and were collected from each of the national agencies in charge of them (see description of the datasets below).   Data were either spatially explicit: where disturbance events contained information about their exact location in space; or spatially referenced: where the time and number of disturbances are recorded but not their detailed spatial location (most of them summarized at the municipality level) (Birdsey et al., 2013a). Non-spatial disturbance data were associated with specific municipality polygons. Yucatan, Campeche and Quintana Roo have 106, 11 and 9 municipalities, respectively (INEGI, 2012).    Two ecoregions of North America Level I (CEC, 1997) were intersected with the three state administrative boundaries of the YP to provide a spatial stratification framework, comprised of five spatial units that will be used for carbon budget analyses (Figure 2-2). Additionally, one spatial dataset of areas under forest management for the period was available from Mexico’s National Forestry Commission. 26   Figure 2-2 The Yucatan Peninsula municipalities and spatial stratification units  Wildfire information in Mexico has been recorded since 1970 for each state. Tabular data from this national database were provided by CONAFOR. Historical records include the number of fires and the number of hectares burned per year aggregated by state. From 2005 onwards, the database includes additional information: spatial coordinates of the fire central ignition point, cause of ignition, number of hectares burned, type of ecosystem affected (temperate, tropical, arid) by municipality.   Additionally, spatial datasets were available from 2005 onwards, containing the spatial coordinates of the central ignition point but not the geographic boundary of the area burned. 27  However, 39% of the fire plots in tabular form (mostly from 2005, 2006 and 2009) did not include geographic coordinates and were not contained in the spatial datasets. These fires accounted for approximately 25% of the total area burned over the period. As it was not possible to link the non-spatial fire events with the spatially-explicit land-cover change information, we limited our analysis to the spatial datasets of the fire points with coordinates. Annual fire maps were then generated by buffering the ignition points with an area equal to the number of hectares burned per fire. We categorized fires as either major (area burned ≥ 1000 ha) or minor (area burned < 1000 ha).  Information on the trajectories of tropical storms from 2005-2010 that crossed the region was retrieved from the National Climatic Data Center (NCDC, 2012; Figure 2-3).  For each track, tabular data on the date, points of landfall, pressure and wind speed for each storm were available. To assess the potential impact of the hurricanes beyond the storm trajectory, we buffered each track according to the disturbance severity associated with the Saffir Simpson category (NHC, 2013). To do so we applied buffer distances derived from Skwira et al.’s (2005) rain-band width studies, with 15 km for major impact hurricanes categories (category IV or V), 10 km for category III and II, and 5 km for the remaining lower-impact storms.  28   Figure 2-3 Tracks of tropical hurricanes that crossed the Yucatan Peninsula from 2005 – 2010 *Derived from the National Climatic Data Center hurricane database records  Agricultural activity maps were generated using data from the Secretariat of Agriculture, Livestock, Rural Development, Fisheries and Food of Mexico known as SIACON (SAGARPA, 2012). This program provides tabular data on the total area of annual cultivated land by state and municipality. Since the database lacked spatially-explicit information, we referenced the area of agriculture to the INEGI municipality polygon. Additional data on agriculture were retrieved from INEGI’s Land Use and Vegetation series III (SIII) and IV(SIV), developed for the years 2003 and 2007, respectively (INEGI, 2003 and 2007). We used these datasets to differentiate permanent cultivation (PC) areas from the rest of the agricultural activities. Two masks were generated locating areas that fell under the status of permanent cultivation (PC) (code “RP”) and 29  assign them a 100% impact. Carbon budget models can be parameterized to represent these areas with a constant 100% impact and simulate no forest regrowth. The first mask was developed by filtering PC areas from SIII in the years of 2005 to 2007; and the second, by filtering areas from SIV from 2008 onwards. Annual maps were then produced for the entire region, subtracting the hectares identified under PC by municipality from: i) the cultivated hectares reported annually by municipality in the SIACON program; and ii) the area of the municipality. The degree of impact was expressed as the percentage of total cultivated area by municipality.  2.2.2.2 National Forest Inventory National Forest Inventory data are one of the primary sources used for forest carbon monitoring (Kurz et al., 2009; Birdsey et al., 2013a). Throughout Mexico, ground plot data are routinely collected by the Mexican National Forest Inventory (INFyS, hereafter NFI) and available from CONAFOR. The NFI has a stratified systematic sampling design that operates on a 5-year re-measurement cycle, with 26,220 permanent sample plots across the country providing information on forest condition and change.   Across the study area 3,931 permanent sample plots were located, from which 298 were not sampled and discarded for inaccessibility reasons. Each plot is 1 ha in size and consists of a primary sampling unit, with four secondary sampling units (sub-sampling plots), of 400 m2 each (INFyS, 2012). More than 150 variables are measured on a five-year cycle, including attributes of the understory and overstory vegetation, the soil, and environmental characteristics of the landscape. Two time periods of NFI measurement data were available for the assessment, the first sample cycle 2004-2009, and the re-measurement period 2009-2012. Forest gains and losses 30  were estimated over the study period using the NFI ground measurements from both sampling periods. Gain or loss estimates were obtained by computing the difference in the mean basal area per hectare for each plot between the 2004-2009 and the 2009-2012 sampling cycles. For more details concerning the variables sampled in the NFI, as well as the sample design and the remeasurement framework, readers are referred to the INFyS report (2012).   Settlement expansion areas were derived from NFI plots from 2004-2009 recording evidence of infrastructure disturbance in any of the following environmental impacts: clearing for roads, electrical transmission wires and human settlement. In the sampling plots, the environmental impact is recorded as: non-perceptible, low, medium and high. To convert these scores into a continuous variable we set an incremental degree of disturbance corresponding to the incremental order of severity given to each level of impact described in the NFI sampling manual (INFyS, 2012).   For the “non-perceptible” category, which defines no visible impact on the forest, we set a 10% degree of disturbance. The remaining 90% degree of impact was divided in the three remaining categories, assigning an equal weight to each increasing level of impact every 30% to reach the 100% (i.e. non-perceptible 10%, low 40%, medium 70% and high 100%). Since each plot may have been affected by more than one of these causes in the same measurement period, we merged them into a single “human settlement” category, averaging the impact score for each plot. A geostatistical ordinary kriging model was then used to interpolate the NFI impact scores to the entire area. Kriging is a robust spatial modeling technique that provides optimal and unbiased estimates for unknown values from sampling data, by weighting the observation points 31  that surround the area of interest based on their proximity (Curran & Atkinson, 1998). Ordinary kriging is a general linear regression model given by the following equation:  ẑ(𝑠0) =∑λiz(𝑠𝑗)ni=1  where  ẑ(s0) is the value at the location s0 to be interpolated, z(si) are the sampled plot values and λi are the weights assigned to each unsampled location (Meng et al., 2013). Each sample point is weighted according to the spatial dependence of the data, assigning more weight to those points that are closer to the estimate. This technique is based on the premise that data that are closer in space are more likely to have similar values than those that are farther apart (Tobler, 1970). This approach allowed a wall-to-wall coverage of settlement areas to be estimated across the study area for carbon modeling.  2.2.3 Remote Sensing Observations of Land-cover Change Remotely sensed land-cover change observations were generated from the North American Land Change Monitoring System (NALCMS) developed as part of a collaborative effort between agencies from Canada, the United States and Mexico coordinated by the Commission for Environmental Cooperation (CEC). The NALCMS provides land-cover information across the different ecosystems in North America, derived with ten-day MODIS composites at 250 m resolution (Latifovic et al., 2010). In the case of Mexico, land-cover classification maps embedded in the NALCMS were developed for 2005 and 2010 forming the basis of the comparison in this study; information on the classification approach is provided by Colditz et al., 32  (2012). Fifteen land-cover classes were identified using multiple classifications (ensemble classifier) with decision trees and ancillary datasets including slope, aspect, temperature, precipitation, aerial photography and high-spatial resolution satellite images. The overall accuracy of the map was assessed with a disjoint set of sample data resulting in 82.5% confidence (Kappa of 0.79). To assess the transition in the land cover classes across the YP, we developed a change matrix which calculates the number of pixels that changed land cover class from 2005 to 2010.   2.2.4 Attributing Land-cover Changes to Underlying Disturbance Types Losses in basal area can be explained by tree mortality as the consequence of natural disturbances (e.g., fires, hurricanes) or as a result of tree removals due to harvesting activities (Healey et al., 2005). To assess the relative importance of fires, hurricanes and harvesting on the reduction in basal area in the NFI plots, a regression tree model was developed. Regression tree analysis is a flexible and robust method increasingly applied in ecological research for a number of reasons including the ability to deal with collinear datasets, to exclude non-significant variables, and to allow for asymmetrical distribution of samples (De’ath & Fabricius, 2000; De’ath, 2002; Melendez et al., 2006; Schwalm et al., 2006). The technique automatically separates the response variable (reduction in basal area) into a series of choices that identifies the contribution of each constraining variable, in this case fires, forest management areas and hurricane severity, as the independent variables. At each node, variables are automatically selected to provide the best partitioning of the data that minimize the sum of squares of the response variable (De’ath, 2002).  33  Forest harvest areas were obtained by using an exclusion method, mapping the degree of reduction in the mean basal area from the NFI plots that were not affected by natural disturbances (i.e. fires or hurricanes) explained in the regression tree. We made the explicit assumption that the remaining basal area loss was due to selective or clear-cut logging. Wall-to-wall coverage of harvest areas was then generated with the ordinary kriging approach previously described estimating the degree of impact of harvesting activities across the YP. Based on the relevance of the most likely variables explaining the basal area loss in the regression tree analysis, a final decision tree was developed attributing the MODIS land-cover change pixels to a disturbance type following that order.  To improve our understanding of the complexity of the different disturbance regimes across the YP, the MS-D approach was developed in this study to characterize forest disturbances from 2005 to 2010 and attribute changes in forest cover to their underlying disturbance drivers (Figure 2-4). Annual maps of major disturbance types were generated first with ground-truth ancillary data. Next, the forest cover change observed in the ground-plot data was derived by computing the basal area loss over the period. A regression tree analysis was then used to identify the most likely disturbance type explaining the basal area loss. Finally, the land-cover changes were derived from MODIS satellite imagery from 2005 to 2010, and using the results from the regression tree analysis attributed the change to the most likely disturbance driver.  34   Figure 2-4 Flowchart of the MSD-D (Multi-Scale, Multi-Source Disturbance assessment). The approach is based on a three step implementation: 1) map type, extent, and location of forest disturbances, 2) derive land-cover change observations, and 3) attribute land-cover changes to their most likely disturbance driver.   35  2.3 Results 2.3.1 Forest Disturbance Assessment 2.3.1.1 Fires An overview of fires that occurred in the YP from 2005 to 2010 is shown in Figure 2-5(a).  In the study period, 97% of the fires recorded burned an area equal or less than 1,000 ha (Figure 2-5(b)). The remaining 3% of wildfires burned a total of 80,905 ha, which corresponds to 63% of the total area burned over the period, burning, on average, 3,852 ha per event. The largest fires occurred in the northeast Tropical humid forests of Quintana Roo, 58% of the total fires registered in the period. On average, these fires burned 200 ha per event accounting for 77% of the total area burned over the region from 2005 to 2010.   Figure 2-5a Fires registered throughout the Yucatan Peninsula from 2005 – 2010.  *Derived from CONAFOR’s national database of fires. 36  The maximum area burned (13,240 ha) was registered in the municipality of Benito Juarez in 2006, at the north of the YP. Across the region, the years with most severe fires were 2006 and 2009, burning 94,735 ha (75% of total ha burned). In contrast, 2007 registered the lowest fire regimes accounting for 2% of the total area burned.   Figure 2-5b Fires registered throughout the Yucatan Peninsula from 2005-2010. *Derived from CONAFOR’s national database of fires.  2.3.1.2 Hurricanes Estimates of hurricane disturbance derived from the NHDC across the region can be seen in Figure 2-6 showing the mapped trajectories of tropical storms that hit the YP between 2005 and 2010. Contrary to fires disturbances, the years 2006 and 2009 were atypical for hurricanes with no tracks recorded in those years. In the north, the impact was dominated by larger and strong category V storms (Wilma and Emily), which struck the YP in 2005, and in the center by hurricane Dean (2007).  37   Figure 2-6 Hurricanes that hit the Yucatan Peninsula from 2005 to 2010. *Derived from the National Climatic Data Center hurricane database records.  2.3.1.3 Agricultural Activities Annual maps of agricultural management lands across the YP are presented in Figure 2-7 (a-b) showing the initial and last year of the study (2005 and 2010). Areas under the permanent cultivation status obtained from INEGI SeriesIII (2003) can be observed in Figure 2-7(a) depicted in red with a 100% degree of impact. These areas are mainly located at the central-west of the YP, accounting for 35,509 ha. In 2007, areas under permanent cultivation increased by 15% to 42,071 ha, shown in Figure 2-7(b). Additional areas extracted from the national agricultural database are represented by total cultivated land by municipality on an annual basis. Major areas of impact can be observed at the north of the YP, in the Yucatan state. Here, 38  transitions in the degree of impact over the period can be seen, showing that some municipalities, like Panaba and Sucila (unique areas depicted in red in 2005) went from 80% of cultivated land, to 3% and 5% respectively in 2010. This was mainly due to a 13% decrease in the amount of cultivated hectares reported in 2010 in the municipality records of the agricultural database.   Figure 2-7 a-b Evidence of Agricultural practices within the Peninsula from 2005 – 2010. *Derived from SAGARPA’s national records of agricultural activities.  39  2.3.1.4 Settlement Expansion Evidence of settlement expansion over the period can be observed in Figure 2-8, showing the gradient of disturbance caused by road openings, electric transmission lines, and human settlement. Major affected areas were consistently predicted close to main developed cities - Merida, Cancun and Chetumal - with few patches of lower impact across the rest of the region.   Figure 2-8 Settlement impact across the Yucatan Peninsula estimated from INFyS (2004-2009) ground-plot data.  2.3.1.5 Ground-plot Observations of Forest-cover Change The amount of mean basal area change from the first measurement period (2004-2009) to the re-measurement (2009-2012) is shown in Figure 2-9(a)-(c). Overall 47% of the plots had an average increase in basal area of 4.16 m2 ha-1 between the two periods.  Decreases in basal area were found in 20% of the plots, with an overall mean reduction of 4.42 m2 ha-1, and a maximum loss 40  of 79.2 m2 ha-1 registered in one plot in the tropical humid forests of Campeche. Clusters of plots with basal area loss can be observed in Figure 2-9(c) and Figure 2-10, mainly at the tip and center of the YP, in the tropical humid forests of Quintana Roo, and some parts at the southwest of Campeche. The rest of the plots were either not sampled or inaccessible in the second period (30%), or presented no change (3%).     Figure 2-9 Mean basal area (m2 ha-1) obtained from the first and subsequent  NFI cycle, and the difference (a) Left upper figure: Mean basal area NFI period1; (b) Right upper figure: Mean basal area NFI re-sampling cycle; (c) Bottom figure: Gain/loss between period 1 and 2. 41    Figure 2-10 Forest cover change across the YP (upper graph) expressed by loss/gain in mean basal area (m2 ha-1), and by spatial unit (bottom graph). *Spatial Units:TrDF: Tropical dry forest, TrHF: Tropical humid forest; C: Campeche Q: Quintana Roo, Y: Yucatan   42  2.3.2 Remote Sensing Observations of Land-cover Change From the land-cover change analysis derived with MODIS thematic maps, fourteen of the fifteen land-cover classes identified in Mexico by Colditz et al. (2012) were found in the YP. Land-cover change from 2005 to 2010 totaled 7,281 ha and is shown in Table 2-1.  The most significant change was observed in the Tropical Evergreen forests with a net loss of 5,318 ha (73% of the total forest area loss) mainly due to anthropogenic disturbances that changed forests into cropland and urban areas. Losses in cropland (5,368 ha) were offset by area gained from other classes (5,487 ha), showing a low net gain of 119 ha (2%) in cropland area. The largest net area gain (5,218 ha, 71.7%) took place in the Urban class, followed by gains in Barren land (10%) and Tropical Grassland (8%).    43  Table 2-1 Land cover change matrix from 2005 to 2010 based on the MODIS land-cover classification.    Columns represent the classes from 2005 NALCMS maps, rows represent the classes from the 2010 NALCMS map. Land cover class names: TeEF (1) Temperate needleleaf evergreen forest, TrEF (3) Tropical broadleaf evergreen forest, TrDF (4) Tropical broadleaf deciduous forest, TeDF (5) Temperate broadleaf deciduous forest, MixedF (6) Mixed forest, TrShrub (7) Tropical shrubland, TeShrub (8) Temperate shrubland, TrGrass (9) Tropical grassland, TeGrass (10) Temperate grassland, WL (14) Wetland, CL (15) Cropland, BL (16) Barren land,  Urban (17) Urban and built-up, H2O (18) Water. Numbers in parentheses next to the classes correspond to the MODIS map class numbers.  * Land-cover change matrix generated from CEC’s MODIS NALCMS classification maps derived by CONABIO for 2005 and 2010 for Mexico. Area (Hectares) TeEF TrEF TrDF TeDF MixedF TrShrub TeShrub TrGrass TeGrass WL CL BL Urban H20 Total - 2010TeEF 75 75TrEF 8379281 163 38 8379481TrDF 81 43955 13 439644TeDF 132381 132381MixedF 6 6TrShrub 13 52225 69 13 522344TeShrub 6 6TrGrass 88 6 32675 75 269 6 6 175 333TeGrass 5 19 69WL 288 13 139638 55 156 81 681 14146CL 3713 394 13 138 19 525 3594219 88 6 359976BL 381 94 48181 113 6 49369Urban 1331 6 6 4219 44956 5519H20 13 26 119 319 25 35275 35956Total - 2005 75 8384800 440113 132394 0 522444 0 32719 0 1040831 3599588 48663 45300 3073389939305.5NET gain/loss 0 -5319 -469 -13 6 -100 6 581 69 575 119 706 5219 -138144  2.3.2.1 Attributing Land-cover Changes to Underlying Disturbance Types Results from the regression tree analysis indicated that the classification of area burned explained most of the variance in the reduction of basal area, with fires reducing an average of 5.38 m2ha-1 (p<0.001) of basal area per plot across the YP, between the two observations periods of the NFI  (Figure 2-11).    Figure 2-11 Regression tree analysis showing the reduction in basal area (y) explained by major disturbance types.   At plots which either did not intersect a fire, or had no detected fire, forest management was the most significant variable followed by hurricanes, with lower categories reducing the basal area by a smaller amount than severe hurricanes of higher categories.  Remaining areas from the regression tree analysis assumed as harvest are shown in Figure 2-12.   45   Figure 2-12 Estimated harvest areas resulting from the spatial interpolation of the reduction in basal area over the two NFI measurement periods  Results of the regression tree analysis, MODIS land-cover change pixels attributed to a disturbance type based on a final decision tree (Figure 2-13):   Where MODIS land cover changed from non-urban in 2005 to urban in 2010 the disturbance was attributed to settlement expansion. Analysis of the locations generally matched estimates of settlement expansion derived with the NFI data, close to the cities of Merida, Cancun, Chetumal and Campeche.  Fires were then attributed second, with any MODIS land-cover change pixel overlaying the burned area surface attributed to fire. 46   Change pixels under forest management, not attributed after the fires, were assigned to harvest.  Hurricanes were assigned to those land-cover change pixels which overlaid hurricane damage greater than level 1, and were not attributed to settlements or fires.  Agriculture was attributed after the hurricanes, to land cover change pixels that were classified as non-cropland, but changed into cropland in 2010, and were yet not attributed. Agricultural areas under permanent cultivation were not considered, because they did not overlap areas of change except for a very small area of 38 ha.  Lastly, any remaining land cover change pixels for which we were not able to find corresponding disturbance data were attributed to harvest (16% of the total change pixels).                 Figure 2-13 Decision tree analysis - the attribution of disturbance type to the pixels with land-cover change. 47  The resulting map of MODIS land cover change and the attribution by disturbance type is shown in Figure 2-14(a) and 14(b).  a)  b)  Figure 2-14 MODIS Land cover change pixels (a) and final classification of the pixel as disturbance type (b) 48  In addition, Figure 2-15(a-c) shows the percentage of each land cover class attributed by (a) disturbance type, (b) the total area changed between 2005 and 2010 by disturbance type, and (c) the total pixels changed by class in the period.  a)  b)  c)   Figure 2-15 Land-cover change attributed by disturbance type (a), the percentage (b) and the total area changed by class in the period (c). 49  2.4  Discussion and Conclusions This project was conducted in the context of a larger initiative aimed at developing a national Monitoring, Reporting and Verification (MRV) system for Mexico. We developed a new approach (MS-D) that allows for the attribution of observed land-cover changes to the underlying causes of disturbance.  Knowledge of the cause of disturbance is critical for the accounting of greenhouse gas emissions and removals because different disturbance types have different impacts on carbon stocks, as well as direct and indirect emissions to the atmosphere (Kurz et al., 2009). The MS-D was successfully used to map major forest disturbance types and attribute them to changes in land cover over the ~137,605 km2 region of the YP, an early action area for REDD+, for the period of 2005-2010.  While the main goal of this research was to derive activity data for carbon accounting, results from this study can also be applied to investigate the relationships between the different disturbances for other ecological applications. For instance, 2005 and 2007 experienced increased hurricane activity over the YP, with severe storms affecting the central and north part of the region. In 2006 and 2009, large wildfires burned around 94,735 ha in these same areas.  Studies from Whigham et al. (2003) found that hurricane disturbances promote a large accumulation of forest fuel and opening gaps for subsequent fires. Further research can investigate the contribution of the hurricanes from 2005 and 2007 into the generation of wildfires in the following years. Additionally, estimates of basal area change show losses that coincide with the affected areas by fires and hurricanes. However, there was no relationship with subsequent changes in agricultural activities, which remained constant through 2009. After 2009 50  the region experienced a drop of 13% of planted area, likely as a result of land abandonment which is part of the milpa cultivation system.  Critical for the MS-D was the availability of NFI data to obtain basal area change. However, the time interval between successive measurements of the NFI (a rolling program of re-measurement in 5 year intervals) resulted in only 2 measurements between 2005 and 2010.  We therefore treated the NFI data as a one-time change. The high density of the NFI plots within the area was essential to provide data for spatial interpolation to produce the wall-to-wall estimates of additional disturbance types (i.e. settlement and harvesting areas). The increased detail of disturbances derived using kriging is important to provide the magnitude of impact by disturbance type across the region. Ideally we would like to have had annual datasets to help refine the estimates of when, where and how severely the events occurred. Spatially-explicit or spatially-referenced datasets characterized by year are preferred to model forest carbon dynamics following disturbance as they allow assessment of inter-annual variability.  Sample data of forest disturbance types by year would be desirable, but are difficult to obtain given the cost, efforts and resources required.   We addressed the limitations of the multi-annual NFI data by compiling ancillary datasets available on an annual basis to fill some of the gaps. Improvements to this approach were possible due to the availability of national annual monitoring data of different types of disturbance, including fire, hurricane and agriculture.  While some of these historical records lacked spatially-explicit data, they were successfully used to identify and map affected areas by year for each disturbance type. In our analysis we aggregated the data to the municipality level 51  and allocated it to either all of the spatial unit or a specific group of spatial units within the YP. This is convenient for countries like Mexico where historical records often lack spatial detail.    With respect to the land cover change analysis, we utilized a simple change detection approach using two classification dates. However, the use of post-classification analysis has inaccuracies associated with image calibration, misclassification and satellite-sensor errors (Dai & Khorram, 1998; Fuller et al., 2003). These inaccuracies can produce changes that did not actually occur, and may not detect real changes that occurred. Despite these possible flaws, an extensive review of literature done by Lu et al. (2004) on major change detection techniques found that post-classification comparisons are one of the most common approaches used for change detection. Moreover, in Mexico, studies from Mas (1999) compared six change analysis methods over the state of Campeche and concluded that post-classification comparison was the most accurate procedure, including the benefit of providing the nature of the changes. While most of the land-cover change pixels were successfully attributed with additional evidence of disturbance types, the 16% of the changed pixels for which we were not able to find corresponding disturbance data, “false positives”, remain to be investigated. Further research on “false positives” can also provide more insights for the attribution of the 25% of fires with missing spatially-explicit locations.  The change detection analysis performed using the MODIS imagery was able to determine that most of the conversion between the two time periods (2005-2010) was due to stand-replacing disturbances. For instance, areas of settlement impact observed in the MODIS changed pixels matched only with the location of major impacts, but not with subtle disturbances identified with 52  the NFI plots. This is due to a number of factors. The first is the resolving capacity of the MODIS imagery itself. MODIS-based products can be useful to provide imagery suitable for regional and global applications with high frequency observations, allowing users to keep track of the ecosystem dynamics with a coarse resolution at low cost (Coops et al., 2006; Wulder et al., 2010).  However, due to its low spatial resolution (250 m) it has limitations detecting forest disturbances occurring at finer scales. This broad spatial resolution along with the 5-year time interval makes the identification of non-stand replacing disturbances (e.g. low fires and selective logging) by year difficult.   Forest disturbance studies can be improved with the availability of remote sensing products derived on an annual basis at finer spatial scales, like Landsat satellite imagery. Landsat Thematic Mapper + and the new Landsat 8 OLI can improve the detection of medium and low severity disturbances that result in subtle changes on the forest cover, providing more detailed information about the landscape (Cohen et al., 2010).  However, the small coverage per scene along with the recurrent cloud coverage may sometimes limit its use (Hayes et al., 2008).   Advances in technology are facilitating the processing and characterization of the Earth’s surface with new super-computing systems. An example is the study of Hansen et al. (2013), who developed a map of forest cover loss for the entire world from 2000 to 2012 with a 30 m spatial resolution based on Landsat data. This new map includes the spatially-explicit characterization of forest loss and gain for the period, with annual loss distribution.  Other examples include forest disturbance products derived with a new vegetation change tracker algorithm (VCT) that efficiently processes Landsat time series stacks to characterize forest cover change (Huang et al., 53  2010). Recently, Mexico acquired full coverage of RapidEye satellite imagery for the years of 2011, 2012 and 2013 with a 5-m spatial resolution that can help to characterize in more detail the current status of the forest cover. These remote sensing products can provide the means to enhance and expand the insights from this study.   The strength of this study is its ability to use synergies between remote sensing products, ground truth data and ancillary datasets, thus taking advantage of the best available information acquired by other researchers/institutions. Mapping disturbances and attributing observed land-cover changes to the appropriate disturbance type with the MS-D approach provides a cost-effective solution to obtain activity data for carbon modeling and other ecological applications. Its flexibility allows a comprehensive integration of spatially-referenced and/or spatially-explicit data on an annual or multi-year basis to characterize forest disturbances by type and attribute changes in land cover to their underlying causes. Moreover, it provides a basis to replicate the study in other “early action” areas of Mexico for ecosystem carbon accounting and monitoring, and the development of an MRV system. This advance also allows other countries undertaking efforts to implement an MRV system for REDD+ (e.g. Ecuador, Peru) to explore the potential use of this approach using input data appropriate for their specific conditions. 54  Chapter 3: CHOICE OF SATELLITE IMAGERY AND ATTRIBUTION OF CHANGES BY DISTURBANCE TYPE CHANGES FOREST CARBON SINKS AND SOURCES  3.1 Introduction Monitoring forest cover and change is essential to quantify the amount of carbon that is stored in the vegetation and soils, and the corresponding greenhouse gas (GHG) emissions (Nabuurs et al., 2007; Bosworth et al., 2008; Turner, 2010). Forest ecosystems can mitigate the impacts of climate change by absorbing significant amounts of atmospheric carbon through plant photosynthesis (Chapin et al., 2009; Pan et al., 2011).  However, degradation of forests and conversion to non-forested lands is the second largest cause of carbon dioxide (CO2) emissions into the atmosphere (IPCC, 2007, Spalding, 2009; Hansen et al., 2010; Houghton et al., 2012) and this has led to initiatives such as Reducing Emissions from Deforestation and Forest Degradation (REDD+) (UNFCCC, 2010).   Natural and anthropogenic disturbances are main drivers that alter the forest structure over time and understanding their impacts is therefore critical for quantifying carbon stock changes and subsequent emissions into the atmosphere (Randerson et al., 2002; Kurz et al., 2009; Lorenz & Lal, 2010; Birdsey et al., 2013b). Disturbances can range from fire, hurricane, insects and diseases, to timber harvesting, settlement expansion and other human activities. Since each disturbance activity impacts the terrestrial carbon cycle in unique ways, annual observations of forest changes by disturbance type are necessary to accurately estimate carbon dynamics in the 55  year of disturbance and the years after the disturbance (Kurz et al., 2009 and 2010; Spalding, 2009).   Given the complexity of measuring and monitoring the terrestrial carbon cycle, a number of approaches have been developed (Houghton et al., 2012). Some techniques utilize data collected from ground-plot measurements and allometric equations that relate the physical attributes of trees to aboveground biomass in order to quantify forest carbon (e.g., Vargas et al., 2008; Jaramillo et al., 2013; Orihuela et al., 2013). Other approaches use carbon budget models (e.g., Dai et al., 2014), remote sensing data (DeFries et al., 2002; Asner et al., 2010; Langer et al., 2012), or a combination of both with field measurements (e.g., Masek & Collatz, 2006; Stinson et al., 2011; Potter et al., 2012; Espirito-Santo et al., 2014). Carbon budget modelling is a well-established approach to estimate carbon fluxes at regional to national-scales by integrating data from different spatial and temporal scales (Birdsey et al., 2013a).  For example, in Canada, the main framework used for the National Forest Carbon Monitoring Accounting and Reporting System (NFCMARS) is the carbon budget model of the Canadian forest sector (CBM-CFS3) (Kurz & Apps, 2006; Stinson et al., 2011).  The CBM-CFS3 is also used as a decision support tool for forests managers to quantify the ecosystem carbon dynamics at the landscape level.  In compliance with the Intergovernmental Panel on Climate Change (IPCC) guidelines, the CBM-CFS3 quantifies carbon transfers among the five terrestrial carbon pools: above and belowground biomass, litter, dead wood and soil organic carbon, including atmospheric releases of CO2 and non-CO2 greenhouse gasses and transfers to the forest product sector (Kurz et al., 2009). The CBM-CFS3 offers the advantage of modeling the specific degree to which tree 56  biomass and dead organic matter are affected by different disturbance types (Kull et al., 2011). Following disturbance, it simulates subsequent forest transitions and successional dynamics and determines mortality of vegetation transferred to litter, coarse woody debris and soil organic carbon pools. Based on the stage of stand development and ecological characteristics, the model incorporates litter fall, decomposition and yield data to account for mortality and regrowth.  Carbon models, such as CBM-CFS3 require detailed spatio-temporal information about forest dynamics, including forest disturbances to accurately simulate carbon exchange and emissions. Remote sensing (RS) is one of the primary land-cover data sources for carbon monitoring due to its ability to monitor the Earth’s surface on a regular and continuous basis, including areas otherwise difficult to access (Coops et al., 2006; Wulder et al., 2010).   Recently, an increasing number of regional and global-scale RS studies and products have been developed specifically for forest carbon monitoring and REDD+ at different spatial and temporal scales (Asner et al., 2010, Saatchi et al., 2011; Baccini et al., 2012; Achard et al., 2014). In a global study, Potter et al. (2012) integrated large-scale Moderate Resolution Imaging Spectroradiometer (MODIS) satellite observations with the Carnegie Ames Stanford Approach (CASA), predicting a total of 0.51 Pg C yr-1 emissions from forest disturbance and biomass burning from 2000 to 2009. At at more regional-scale, Masek & Collatz (2006) integrated the CASA model with forest inventory data and higher spatial resolution satellite imagery to predict C fluxes from 1973 to 1999. They used Landsat time-series analysis to assess the effects of land-clearing disturbances (such as logging, harvesting and urbanization) on ecosystem productivity and quantified carbon emissions from biomass losses, decomposition and decay. 57  Significant variability among carbon estimates is an important issue. Such variability is potentially due to differences inherent to the methods used, the spatial and temporal scales involved, and the definition of the components included (Chapin et al., 2006). A recent study by Achard et al. (2014) estimated carbon emissions from deforestation in tropical countries for two decades (1990-2000, 2000-2010) by combining forest cover change maps derived from Landsat imagery and pan-tropical biomass maps. Contrasting their results with those obtained by Hansen et al. (2013), they found discrepancies that can be explained by different approaches used in the satellite image analysis and their definitions of forest.   A global study for tropical countries was also done by Saatchi et al. (2011), integrating inventory data and satellite light detection and ranging (LiDAR) data to quantify the carbon stored in the living biomass (247 Gt C) circa 2000. In a similar study, Baccini et al. (2012) estimated carbon emissions from pan-tropical deforestation and land-use from 2000 to 2010, combining aboveground biomass with regional deforestation rates derived from LiDAR and MODIS, respectively. Comparing their results with the Forest Resource Assessment (FRA), they found their estimates (228.7 Pg C) were 21% higher (Baccini et al., 2012). Likewise, Mitchard et al. (2014) compared estimates from Saatchi et al. (2011) and Baccini et al. (2012) across the Amazon forest to a unique dataset of ground-plot data, and found that the maps over or under-estimated the forest carbon estimates obtained from the ground data by >25%.  It is critical to understand the potential role and implications of different RS observations for estimating ecosystem carbon dynamics following disturbance, in order to reduce uncertainties in measuring and monitoring carbon (DeFries et al., 2007; De Sy et al., 2012). In REDD+ studies, 58  land-cover change observations are one of the main components (also called activity data) required for Monitoring Reporting and Verification (MRV) systems (GOFC-GOLD, 2010). These activity data can be observations of changes in forest areas or land conversions between categories (Hewson et al., 2013).  This study assesses the impact of using different RS data on terrestrial carbon dynamics. We wanted to examine the impacts of three attributes of RS products on the estimates of GHG emissions and removals: (1) the spatial resolution (30 m vs. 250 m), (2) the temporal resolution (1-yearvs. multi-year observations), and (3) the attribution of forest cover changes to disturbance types. To do so we utilise four contrasting RS datasets - two land-cover change maps and two thematic maps - to provide inputs of disturbance events to the CBM-CFS3 and model subsequent forest carbon emissions and removals.  The RS data were derived from MODIS and Landsat satellite imagery developed using different approaches, and spatial and temporal resolutions.  We derived two spatially-explicit layers for each of the RS products, one with land-cover change area, and the second attributing the observed change to its underlying disturbance cause.  We then provided these inputs to CBM-CFS3 and estimated carbon fluxes from 2002 to 2010 in the tropical dry forests of the Yucatan Peninsula, Mexico. We conclude with a discussion of key findings and implications of different activity data sources for future carbon budget analysis.  3.2 Methods and Data 3.2.1 Study Area Our study focused on a ~3.2 million ha area located in the center of the Yucatan Peninsula (YP), Mexico, covering parts of Yucatan and Campeche States (see Figure 3-1). The region’s 59  vegetation is dominated by secondary lowland dry tropical forests (Read & Lawrence, 2003) of mixed ages that have regenerated after cycles of shifting agriculture and land abandonment. The region has a tropical climate, with rainy summers from June to October and dry winters from November to May. The mean average temperature is 26 C with precipitation levels ranging from 900 to 1400 mm yr-1 (Vandecar et al., 2011). The regional topography is characterized by flat limestone areas and low moderate hills; the average slope is 7% and the mean elevation is 116 m (Dai et al., 2014). The main soil types are well-drained rendzinas and shallow rocky lithosols (Urquiza-Haas et al., 2007).   Figure 3-1 Study area: The Yucatan Peninsula, Mexico. 60  3.2.2 Remote Sensing Data Remote sensing observations are essential to track forest disturbance impacts and changes in the land cover and the forest carbon at a range of spatial scales (Powers et al., 2015). Land-cover changes are those remotely sensed observations of the Earth’s surface that changed from one point in time over two or more time periods because of any disturbance (Houghton et al., 2012), including both natural and human-induced causes.  We compiled four remote sensing disturbance products from Landsat and MODIS imagery derived using a range of approaches to compare their impacts on carbon budget estimates (see Table 3-1).   A spatially-explicit dataset derived using the vegetation change tracker algorithm (VCT) which reconstructs the history of the landscape disturbance on an annual or bi-annual basis, depending on data quality and availability, using Landsat, 30 m,  time series analysis (Huang et al., 2010).  Each pixel of the spatial dataset was labeled according to the year in which the forest cover change was identified, from 1985 to 2010 (represented by 16 classes). From this map, we extracted disturbed areas for the years 2003, 2004, 2005, 2007, 2008 and 2010. Change data were unavailable due to cloud cover in 2002, 2006 and 2009.   Annual land-cover change data from a global forest cover change map derived from Landsat imagery at 30 m spatial resolution were also available from Hansen et al. (2013).  The coverage provided direct estimates of forest cover loss and gain from 2000 to 2012, including detailed information of annual forest loss. Assessments of forest cover loss 61  included pixels that completely lost tree canopy cover from a stand-replacing disturbance. Here, we obtained annual cover loss areas for our study period (2002-2010).   We also retrieved three classification maps from the National Institute of Statistics and Geography of Mexico (in Spanish: INEGI): the vegetation and land-use cartography series developed for 2002, 2007 and 2010 at a scale 1:250,000 with approximately 70 classes (INEGI, SIII, SIV, and SV). The vegetation maps were generated using manual interpretation of Landsat and SPOT satellite images, validated with ground-plot data. In addition, a digital elevation model and ancillary datasets that describe the climate, geology, hydrology and topography of the Mexican ecosystems were used in their construction. These maps are a conventional source of information in Mexico for assessing the status and condition of the forested land base. We used the INEGI maps to generate two land-cover change maps: 1) changes detected from observation one to observation two (2002 – 2007); and 2) changes detected between observation two (2007) and three (INEGI, 2010). To simplify this process, we grouped the INEGI vegetation classes into twelve classes established by the National Commission for Knowledge and Use of Biodiversity (CONABIO) and assessed the change from one class to another.   Finally, we retrieved six classification maps developed by the CONABIO in the North American Land Change Monitoring System (NAL CMS) project using MODIS satellite imagery at 250m spatial resolution (Latifovic et al., 2010). Annual spatially-explicit maps were developed from 2005 to 2010, using monthly composites of MODIS, multiple classifications (ensemble classifier) with decision trees and attributes from auxiliary 62  datasets to characterize fifteen land-cover classes in Mexico (Colditz et al., 2012). These attributes include a digital elevation map, aspect, slope, mean temperature, maximum precipitation, sampling data, and aerial photography at high-spatial resolution. From these maps, we assessed the annual transition from one land-cover class to another for the period 2005 to 2010.   Table 3-1 Remote sensing products used as activity data inputs for carbon modeling with the CBM-CFS3 in the Yucatan Peninsula from 2002 to 2010.   3.2.3 Accuracy Assessment of Forest Cover Change Estimates We performed an independent assessment of the remote sensing disturbance datasets using ground-plot data obtained from the Mexican National Forest Inventory (NFI). The NFI is the primary data source of information for forest and ecosystem studies in Mexico (INFyS, 2012). Data are collected in a two-stage sampling scheme comprised of primary plots of one-hectare size and four sub-sampling sites of 400 m2 each (INFyS, 2012). More than 150 variables are measured on a five-year cycle, including attributes of the understory and overstory vegetation, the soil, and environmental characteristics of the landscape. We obtained forest change data from basal area loss/no-loss per hectare computed by Mascorro et al. (2014) for 647 plots that were located across the study area. We intersected these plots with the MODIS and Landsat-derived Map alias Source Type SatelliteSpatial resolutionTemporalityVCT NASAVegetation Change Tracker mapLANDSAT 30m2003, 2004, 2005, 2007, 2008 & 2010HansenUniversity of MarylandForest cover loss map LANDSAT 30m 2002, 2003… 2010INEGI INEGI Classification map LANDSAT 30m2002-2005,  2005-2007,  2007-2010MODISNALCMS/ CONABIOClassification map MODIS 250m 2005, 2006… 201063  land-cover change areas to identify and compare the number of plots with forest change to the proportion of changed areas detected by each RS product using a standard error matrix. The matrix performs a cross-tabulation between the numbers of observations mapped with RS that agree with what is observed on the ground to assess the level of accuracy of the mapped products (Congalton, 1991; Olofsson et al., 2014).  3.2.4 Disturbance Attribution We attributed the land-cover change estimates obtained from the RS products using a Multi-Scale, Multi-Source Disturbance (MS-D) approach developed by Mascorro et al. (2014).  This approach integrates RS data, forest inventory and ancillary datasets to attribute the land-cover change observations to the most likely disturbance type (natural or anthropogenic). Using NFI data and historical records of forest disturbances in tabular format (both considered ground-truth data), Mascorro et al. (2014) derived annual spatially-explicit layers of major forest disturbance types, and obtained the forest change observed in the ground-plots. A regression tree analysis was undertaken to identify which of the constraining variables (forest disturbance types) best explained the observed forest loss in the field. Once the most likely cause of change was identified, remote sensing maps were used to obtain pixels with land-cover change and overlaid these with the spatially-explicit layers of disturbances characterized by type. Finally, the change was attributed to the most likely disturbance cause according to the relevance authority resulting from the regression tree analysis.  We retrieved ancillary forest disturbance layers characterized by Mascorro et al. (2014) and overlaid them with the VCT, Hansen, MODIS and INEGI RS observations. The disturbance 64  datasets retrieved included annual, spatially-explicit, natural and anthropogenic disturbances from 2005 to 2010: fires, hurricanes, and forest management areas. To characterize the permanent conversion of forestland to non-forestland caused by settlement, we used road coverages retrieved from INEGI (INEGI, 2013). In addition, land-cover classes that changed to urban areas in the classification maps were attributed as settlement.  3.2.5 Pre-Processing Data  To prepare the input data and provide the parameters required by the CBM-CFS3 for each of the simulation runs, we used a software tool called “Recliner” developed and provided by the carbon accounting team of the Canadian Forest Service. The tool facilitates the processing of large volumes of spatially-explicit data and assists users with the preparation of required input data for each simulation. We customized Recliner and matched the RS disturbance types with pre-defined disturbance matrices stored in the CBM-CFS3. Disturbance matrices define the impacts on the carbon pools for each disturbance type and specify the amount of carbon that is transferred from the biomass to the dead organic matter pools or is released into the atmosphere following a particular disturbance type (Kurz et al., 2009; Kull et al., 2011). Fire disturbances were matched to “wildfires”, settlement to “deforestation” and harvest to “clear cut with slash burn”. Disturbance matrices also define the amount of carbon transferred from forest ecosystems to the forest product sector by harvesting activities, and the amount of carbon released to the atmosphere as methane (CH4), carbon monoxide (CO) and carbon dioxide (CO2).    The CBM-CFS3 can simulate stand-replacing disturbance by restoring the stand age to the initiation stage, or simulate partial mortality specified in a “generic mortality” disturbance 65  matrix, ranging from 5% to 95% of impact (Kull et al., 2011). Since the CBM-CFS3 does not contain a specific matrix for hurricanes, we matched these events with a generic disturbance matrix of “10% mortality”, estimated by dividing the mean basal area loss explained by hurricanes in Mascorro et al. (2014) studies, over the mean basal area estimated across the Yucatan Peninsula  (1.71 m2ha-1 / 4.42 m2ha-1). Through Recliner, we specified the year of disturbance impact to the corresponding land-cover change detection year. In cases where annual data were not available, we distributed equally the disturbance events among the number of years contained in the period of detection (e.g. VCT disturbances detected in 2007 where assigned in equal proportions to 2006 and 2007, since no observations were available for 2006).  3.2.6 Carbon Budget Modeling  We provided annual disturbance events to the CBM-CFS3 to quantify carbon exchange from tree biomass mortality, plant detritus decay, soil organic carbon and forest regrowth after disturbance. Additional parameters and data required by the CBM-CFS3 for the simulations were provided by the carbon accounting team of the Canadian Forest Service, and researchers from the Mexican Forest Service. These parameters were kept constant in all the simulations.   Ancillary datasets included detailed information on forest inventory, forest type, forest status, ecological boundaries, age-class structure, and yield curves. Stand age and yield curve data, in combination with the ecological characteristics of the site, are used by the CMB-CFS3 to simulate forest growth and carbon accumulation (Kurz et al., 2009). The stage of stand development is also used to simulate the size and dynamics of the litterfall and decomposition rates for dead organic matter and soil carbon. For forest status, we used protected and non-66  protected conservations areas, which the model employs as a classifier to differentiate the conditions on the landscape. We also used as a classifier the ecoregions level-I retrieved from the Commission for Environmental Cooperation (CEC, 1997): tropical humid forests and tropical dry forests. Estimates of carbon sinks and sources were then generated with the CBM-CFS3 for each of the two spatially-explicit layers of the four RS products, i.e. with and without attribution of cover loss to disturbance types. For simulations without attribution to specific disturbance types we assumed that all forest cover changes were caused by harvest with slash-burning.  3.3 Results  3.3.1 Forest Change Detection The spatial variation in the location and extent of changes in forest area detected by the VCT, Hansen, MODIS and INEGI maps can be observed in Figure 3-2 and Figure 3-3. It is evident that the MODIS products detected fewer changes across the landscape than the Landsat products, likely due to the lower spatial resolution of MODIS. As can be seen in Figure 3-2b and 3-2d, landscape changes between the VCT and Hansen products exhibited similar patterns. With a 30 m spatial resolution, these products detected finer disturbance events across the study area. Undisturbed forest areas were primarily located in the center of the region, expanding to the northwest and southeast of the study area. In contrast, the INEGI maps generated with a classification approach detected larger areas of change as land-cover classes transitioned from one to another (Figure 3-3).   67  a) VCT  b)  c) HANSEN  d)  Figure 3-2 Land-cover change maps derived from the different remote sensing products. VCT map: (a) and (b); Hansen map: (c) and (d). Left side maps show annual non-attributed disturbances; right side show attributed disturbances: green: harvest, pink: settlement, blue: hurricane: orange: fire. The area perception may be misleading; the amount of area affected is better captured in Figure 3-4.   68  e) INEGI  f)  g) MODIS  h)  Figure 3-3 Land-cover change maps derived from the different remote sensing products (thematic). INEGI maps: (e) and (f); and the MODIS maps: (g) and (h). Left side maps show annual non-attributed disturbances; right side show attributed disturbances: green: harvest, pink: settlement, blue: hurricane: orange: fire. The area perception may be misleading; the amount of area affected is better captured in Figure 3-4. The cumulative change events detected in the study area by each RS product across the analysis period is shown in Figure 3-4. It is apparent that the Landsat products detected the highest 69  percentage of change in all of the remote sensing approaches. However, the amount detected was still less than that observed in the INEGI thematic maps.   Figure 3-4 Percentage of the total area change detected in the study area from 2002-2010 with Landsat (VCT, Hansen and INEGI) and MODIS satellite imagery.   Comparing the maximum percentage of area change detected by INEGI (13%) versus the minimum detected by MODIS (1%), there is considerable range in uncertainties resulting from the differences in spatial and temporal characteristics inherent in each RS product. Temporal trajectories of the land-cover changes detected in annual time-steps provided additional information on the inter-annual variation of disturbance events as shown in Figure 3-5.  Figure 3-5 Annual area changed by disturbance type derived from Landsat (VCT, Hansen and INEGI) and MODIS remote sensing land-cover change products. 70  3.3.2 Forest Change Accuracy Assessment Table 3-2, summarizes the results of intersecting the NFI plots describing forest cover loss, with the VCT, the Hansen and the INGEI land-cover change products in an error matrix.  Of the 647 NFI plots located in the study area, 119 showed a forest cover loss, and 528 had no evidence of any basal area reduction in the plot. The first row of each table contains the number of mapping units that each RS product detected as forest cover loss compared to what was observed on the ground with the NFI plots (the total sum of the first column). As can be seen in Table 3-2a and 2b, the overall accuracy of the VCT (83%) and Hansen (82%) maps produced using pixel-based image processing approaches was similar (Tables 3-2a and b). The INEGI accuracy was 79% (Table 3-2c). However, we can observe that all three maps presented have a low Producers’s accuracy, detecting only a small fraction of the changes observed in the NFI plots likely due to forest degradation. The largest uncertainties were found in the MODIS maps, as there was no correspondence between the MODIS changed pixels and the NFI plots.  Table 3-2 Error matrix of Landsat maps VCT (a), Hansen (b) and INEGI (c) identification of change and no change areas  compared to the ground-plot data  a) VCTChange No change Sum Producer'sNFI plots Change 18 101 119 15%No change 11 517 528 98%Sum 29 618 647User's 62% 84%Overall accuracy = 83%b) HansenChange No change Sum Producer'sNFI plots Change 27 92 119 23%No change 22 506 528 96%Sum 49 598 647User's 55% 85%Overall accuracy = 82%c) INEGIChange No change Sum Producer'sNFI plots Change 26 93 119 22%No change 44 484 528 92%Sum 70 577 647User's 37% 84%Overall accuracy = 79%71  3.3.3 The Impact of Activity Data on Estimates of Carbon Fluxes We used the CBM-CFS3 model to estimate the effects activity data on emissions and removals during, and following, disturbances. Carbon fluxes calculated from 2002 to 2010 demonstrated how the estimates changed with different activity data derived from the RS products (Figure 3-6). Results show the annual net changes in ecosystem carbon stocks, with negative numbers indicating a decrease in carbon stocks (an emission of carbon to the atmosphere) and positive numbers an increase in carbon stocks (a removal of carbon from the atmosphere). As can be observed in Figure 3-6, the estimates show both differences in the magnitude of changes and differences in trends resulting from the different sources of activity data. The carbon fluxes changed from one year to another, showing shifts from an overall sink to a source in years with high disturbances.   Figure 3-6 Annual carbon fluxes estimated with different sources of activity data in the Yucatan Peninsula from 2002 to 2010; a: attributed, na: no- attributed. 72  The values from these simulations should not be considered as absolute values; what matters here is the trend and difference between the scenarios simulated with the different inputs of activity data and their specific characteristics. There were a number of assumptions about additional ecological modeling parameters that went into the simulations that have been updated over the last year (e.g. age class structure, and growth curves). It is the relative change between the trends and between the methods what really matters in Figure 3-6.  The cumulative difference in the carbon balance estimates over the period 2002-2010 resulting from each RS product  is over 29 million tC with MODIS-derived estimates suggesting a large sink while INEGI-derived estimates suggesting a large source (Figure 3-7).   Figure 3-7 Cumulative difference of carbon fluxes, estimated with different sources of activity data  in the Yucatan Peninsula from 2002 to 2010; a: attributed, na: no- attributed.  73  Since the ecological parameters and additional data inputs used in each of the simulations were kept constant, the observed differences in the estimates reflect only the effects of the spatial and temporal resolutions inherent to each of the RS observations, and the impacts of attributing cover changes to disturbance types with different impacts on carbon stocks.   In the first year of the simulations we observe the overriding trend of the MODIS disturbance detection underestimating cover changes and the associated carbon losses compared to the Landsat RS products. MODIS-based simulations shifted from a source to a sink in 2005 and remained a sink for the remainder of the simulations.  At the beginning of the simulations, carbon estimates resulting from the Landsat sources were similar for 2002 and 2003, and began diverging in 2004. From 2005 to 2007, simulations based on VCT and Hansen predicted increasing carbon stocks. Consistent with the amount of land-cover observations detected by each product (Figure 3-4), the rate of increase in carbon uptake estimated with VCT from 2005-2007 was higher than that simulated with Hansen. In 2007, Hansen removals were 233,138 tC, whereas the VCT estimates were 1,033,097 tC (Figure 3-5). While some disturbance events were detected in 2005-2007, the amount of area affected was not enough to bring the sink back to a source (e.g., estimates in 2009 from Hansen). Fewer disturbances were observed by VCT in this period, which translated to fewer carbon emissions and more carbon accumulation due to forest growth without disturbance. Nevertheless, in 2008 estimates from these two data sources went in opposite directions.   In the Hansen simulation, the forest continued to sequester carbon in 2008 as the amount of affected area decreased by 37% compared to 2007 (Figure 3-6). This was translated by the CBM-74  CFS as less carbon emissions and more living biomass accumulation, sequestering 3 times more carbon than in 2007 (233,138 tC in 2007 vs 728,392 tC in 2008). Conversely, with the VCT, the amount of change area detected in 2008 increased by 63% (17,800 ha) compared to 2007, releasing considerable amounts of carbon (-1,033,837 tC) to the atmosphere from forest disturbance events that killed the vegetation and redistributed the carbon in the dead organic matter pools during that year.  Moreover, Mascorro et al. (2014) found that 2009 was an intense year of disturbances in the Yucatan Peninsula. Accordingly, the impact of the disturbance events detected by Hansen, resulted in significant amounts of carbon emissions into the atmosphere  (-2,774,298 tC) (Figure 3-6). MODIS derived estimates also showed a decrease in carbon content in 2009, but by smaller amounts due to its limited capacity to detect changes in the landscape. However, with the VCT no land-cover change observations were included in 2009 due to limitations of image acquisition and cloud coverage, and therefore the simulations show no decrease in carbon emissions in 2009, despite the fact that the change observed in 2010 was equally distributed to 2009 and 2010. This shows that even missing a single year in the land-cover observations can lead to errors; especially in countries with rapid-regrowth forests, like in the Yucatan Peninsula.  3.4 Discussion and Conclusions This study assessed the implications from RS activity data derived with different spatial and temporal resolutions on the estimates of carbon emissions and removals.  Annual carbon fluxes were simulated with the CBM-CFS3 from 2002 to 2010 over a 180 x 180 km study area of dry tropical forests in the Yucatan Peninsula, Mexico.   75  To develop reliable estimates of carbon dynamics, detailed observations of the drivers of change are required on an annual basis (Kurz, 2010). Our results showed that higher temporal and spatial resolution of the RS products improved the carbon dynamics estimates and captured the annual inter-variability in the forest. Consistent with the large disturbances detected in 2009, carbon fluxes predicted with the Hansen and MODIS products showed a decrease in total ecosystem carbon content. However, due to the lack of disturbance events detected with the VCT in 2009, corresponding decreases in the carbon stocks from the disturbance events were not reflected in the CBM-CFS3 estimates. This occurred despite the fact that the disturbance events observed over the period 2009-2010 were equally distributed in 2009 and 2010. It is likely that some of the 2009 disturbance events were not detected by the VCT in 2010 due to the fast regrowth rate of the tropical forests of the Yucatan Peninsula.  Carbon dynamics simulated with the INEGI products showed the relevance of identifying the cause of change in the landscape. Natural and human-induced disturbances modify the stand ages and succession dynamics of the landscape in a unique way, changing the live biomass and dead organic matter turnover rates (Franklin et al., 2007; Lorenz & Lal, 2010). Overall, estimates from non-attributed changes presented more carbon emissions than the attributed estimates in all the simulations. However, in the INEGI simulations this difference was more evident. In the first period (2002-2007), the carbon emissions from non-attributed disturbances using INEGI were close to those predicted by the attributed disturbances. However, in the second period (2008-2010), estimates dropped from a sink (240,559 tC) to a source (-3,127,501 tC) releasing about 8 million tC into the atmosphere over the period. While this shift also occurred in the attributed simulations due to a 50% increase in land-cover change areas detected over the period, the 76  impact in the attributed simulations was 40% less than the non-attributed releasing around 4.7 million tC. Depending on the disturbance type, the CBM-CFS3 disturbance matrices determine how much biomass gets killed, re-distributing the carbon in the litter and dead organic matter pools accordingly; affecting the carbon changes more adequately.  Moreover, differences between with and without attribution would be more pronounced if they are reported as CO2 equivalent emissions (as required in REDD+ projects).  These differences would be larger because fires cause additional non-CO2 GHG emissions (in the form of CH4 and N2O), with higher global warming potentials than CO2.   Overall, we found that the traditional pixel-based image processing approach using VCT and Hansen was more accurate than thematic maps (i.e., INEGI, MODIS) at detecting disturbances on the landscape (see Table 3-2). While the INEGI simulations of the first period observations (2002-2006) show a similar trend than VCT and Hansen, change observations for the period 2007 to 2010 are much higher than any of the others. This is likely due to the fact that INEGI maps were generated with a classification approach grouping pixels together with similar characteristics. Therefore the larger disturbances observed over the second period, resulted in higher observations likely due to larger areas that transitioned from one class to another (as opposed to single-pixel changes). Some of the predicted changes using thematic maps will result from inaccuracies inherent to the classification method and classification errors (Dai & Khorram, 1998; Fuller et al., 2003). Nevertheless, post-classification analysis has become a popular method for change detection (Lu et al., 2004). By using the thematic maps we were able to better identify settlement expansion embedded in the land-cover classification approach as urban areas. For instance, the highest percentage of disturbance areas attributed as settlement expansion 77  (56%) was identified with MODIS thematic maps. This was followed by INEGI maps, attributing 36% of the changes to settlement expansion areas. For the RS products generated with a traditional pixel-based approach, we found some limitations in attributing the changes to the underlying disturbance cause due to the lack of spatially-explicit ancillary datasets. Further research is required to increase the accuracy of land-cover changes characterized by disturbance type using additional datasets.   The accuracy with which spatial changes have been predicted throughout the study area (Table 3-2) confirmed findings of others (e.g., Masek et al., 2008; Cohen et al., 2010; Hansen et al., 2010) showing that Landsat-based products are more accurate than MODIS at detecting forest disturbances (Potapov et al., 2008). Although Mascorro et al. (2014) were able to characterize major stand-replacing disturbances for the entire Yucatan Peninsula with MODIS, our results showed that this 250 m resolution satellite imagery has limitations for providing activity data for carbon modeling at finer spatial scales. Typically, satellite imagery from MODIS has focused on broad scale studies (e.g., Wulder et al., 2010; Potter et al., 2012). Due to the low spatial resolution of MODIS, the carbon estimates remained a sink during most of the time-steps in the simulation. In contrast, the Landsat-derived predictions with higher spatial resolution showed a more realistic scenario where the carbon fluxes reflect the complex dynamics of the forest.   The carbon balance estimates obtained for this study are based on CBM-CFS3 simulations using preliminary inventory, age distribution and yield data for the study area.  Ongoing research as part of the Mexico/Norway project and research supported by the CEC will refine and improve the CBM-CFS3 input data.  The absolute values of the numerical estimates obtained in this study 78  may thus change in the future, but the general conclusions about the impacts on carbon balance estimates of the spatial and temporal resolution of RS products, and the attribution of land cover change to causes of disturbance will not be affected by future changes to other input data of the model.  Results from this study may help countries undertaking efforts to reduce their greenhouse gas (GHG) emissions for climate change mitigation to promote better informed management decisions. These results provide an improved understanding of the role of RS disturbance observations on ecosystem carbon dynamics and the range of variability of carbon fluxes following disturbance. Systematic forest monitoring, reporting and verification (MRV) systems are required to aid in the successful development of national and regional strategies for REDD+, and to ensure long-term commitments to preserve forests (Herold & Skutsch, 2011; Birdsey et al., 2013a; Hewson et al., 2013). This can only be achieved by implementing RS observations that will allow monitoring of large areas of land in a regular, consistent, and cost-efficient way. Developing climate change mitigation scenarios, priorities, and initiatives requires further knowledge of the drivers of deforestation and forest degradation that can have a significant impact in the reduction of GHG emissions (Houghton et al., 2012; Hosonuma et al., 2012; Birdsey et al., 2013a and 2013b).   This research documents that increasing the spatial and temporal resolution of RS data acquisition reduces uncertainties in the estimates of carbon emissions.  Similarly, improved estimates of change areas are relevant for biodiversity and habitat conservation, ecosystem services provisioning, and carbon monitoring for REDD+. We conclude that spatial scale 79  represents a serious mapping constraint when attempting to encompass large forest areas for ecosystem carbon quantification. Therefore, carbon monitoring decisions should consider direct trade-offs between the spatial detail of finer resolution products (e.g., Landsat, LiDAR, RapidEye) with more precision and less area encompassed per scene, versus moderate and broad spatial resolution observations derived at broader scales (e.g., MODIS). For instance, the Global Forest Observations Initiative (GFOI) regarded MODIS imagery as “too large to be used for generating REDD+ activity data”, suggesting it could be used in complimentary applications (e.g., monitoring near-real time forest change indicators) (GFOI, 2013). Investing resources into improved activity data, including higher spatial and temporal resolutions, along with attributing forest changes to their underlying disturbance type will likely contribute to improved estimates and reduced uncertainties in GHG emissions estimates.  80  Chapter 4: CONCLUSIONS  The overarching goal of this study was to characterize natural and anthropogenic forest disturbances with the aid of remote sensing imagery to provide activity data for carbon budget modeling in a southern tropical region of Mexico. Monitoring forest disturbances is not only relevant for the sustainable management of the forests, biodiversity conservation (MEA, 2005; Naeem et al., 2009) and ecosystem services provisioning, but also for improving understanding of the critical role forest disturbances play in the terrestrial carbon cycle (Bosworth et al., 2008; Kurz, 2010; Lorenz & Lal, 2010; Spalding, 2009). Detailed observations of natural and human-induced disturbances that alter the forest structure over time are necessary to quantify their impact on the terrestrial carbon cycle. Knowledge of the cause of disturbance is critical for the accounting of greenhouse gas emissions and removals as different disturbance types have different impacts on carbon stocks, as well as on immediate and delayed emissions to the atmosphere (Kurz et al., 2009).  The results of this study contribute to the development of science-based decision support models, data and tools to characterize forest disturbances and quantify their impact on the ecosystem carbon balance. As such, it supports policy and management decisions for climate change mitigation options. The first research question was addressed with a novel approach (MS-D) to characterize forest disturbances and attribute land-cover change observations to their underlying disturbance driver. The second research question examined the impact on forest carbon dynamics of different satellite imagery observations of land-cover changes with, and without attribution to disturbance type. The results contribute an improved understanding of the way forest 81  disturbances affect forest carbon dynamics and clearly demonstrate that different choices of satellite imagery can affect estimates of forest carbon budgets and in this study area and time period converted estimated carbon sinks to sources. While the study was undertaken in the Yucatan Peninsula, an area of special interest selected as an “early action” area for its high biodiversity and great potential for REDD+, it can be replicated in other “early action” areas of Mexico for ecosystem carbon accounting and monitoring. Moreover, it provides the basis for scaling up the development of a national MRV system and provides examples of how best available information from other researchers/institutions can be integrated and assessed in a meaningful and comprehensive way. Results from this research may also benefit other countries undertaking efforts to implement an MRV system for REDD+ (e.g. Ecuador, Colombia, Peru) to replicate this study adapted to their national circumstances (data availability, forest types, etc.).  4.1 Key Findings In Chapter 2 a new approach, the Multi-Source, Multi-Scale Disturbance assessment (MS-D), was developed to:  1) Map forest disturbances (natural and anthropogenic) 2) Derive land-cover change observations 3) Attribute the land-cover changes to their most-likely disturbance driver  Using an integrated approach, the MS-D created synergies between forest inventory data, remote sensing products and ancillary datasets to map the major forest disturbances, using the best available information. Historical records of forest disturbances played a key role in the MS-D approach by allowing the extent, location and severity of fires, hurricanes, and agricultural 82  activities in the tropical forests of the Yucatan Peninsula to be mapped on an annual basis from 2005 to 2010. Critical for the identification of forest change was the availability of field observations through the NFI data, over two sampling periods. Using a regression tree analysis, the MS-D approach assessed the most likely contribution of fires, hurricanes, and forest management over the observed forest-cover loss on the ground-plot data. The regression analysis allowed the attribution of the most likely disturbance type to 86% of the land-cover changed pixels detected by remote sensing. Thirty seven percent of the pixels that changed to the urban class were attributed to settlement impact, followed by 2% to fire (p < 0.001), 3% to forest management (p < 0.001) and 26% to hurricanes (p = 0.001); lastly 18% of pixels were attributed to agriculture. The remaining pixels (14%) were assumed to be harvesting. The MS-D approach provided a cost-effective solution for mapping forest disturbances and attribute observed land-cover changes to the appropriate disturbance type to derive activity data for carbon modeling, forest monitoring and other ecological applications.  In Chapter 3 the impact of activity data derived from different satellite imagery on estimates of forest carbon fluxes was demonstrated. Four remote sensing products - two land-cover change maps and two thematic maps - provided various estimates of activity data and causes of forest disturbance which were used as inputs to the CBM-CFS3 model.  Estimates of forest carbon emissions and removals from 2002 to 2009 were produced. By using the MS-D method, an additional spatially-explicit layer of land-cover change observations was developed for each remote sensing product attributing changes by disturbance type. The specific contribution of three main attributes on carbon dynamics was analyzed: (1) the spatial resolution (30 m vs. 250 83  m), (2) the temporal resolution (annual vs. multi-year observations), and (3) the attribution of forest cover changes to disturbance types.   This study demonstrated that activity data acquired on an annual basis were necessary for the accurate simulation of carbon dynamics and quantification of subsequent carbon emissions and removals following disturbance.  The temporal resolution of the annual observations of forest disturbances improved the carbon dynamics estimates capturing the annual inter-variability in the forest. The higher spatial resolution of Landsat-based products resulted in more accurate (ranging from 79% to 83% of overall accuracy) detection of forest disturbances compared with MODIS-derived products, with pixel-based image maps (VCT and Hansen) being more accurate than thematic maps (INEGI, MODIS).   Estimates generated with the INEGI maps attributed by disturbance type highlighted the importance of identifying the cause of disturbance, resulting in 40% lower carbon emissions (4.7 versus 8 million tC) than the non-attributed estimates in the second period of observations (2008-2010). Although the efforts required to obtain annual Landsat-derived activity data with attribution are higher, the reduction in uncertainty of carbon emission and removal estimates can justify the increased effort.  4.2 Future Research Remote sensing technologies are rapidly changing the way we envision and study the Earth. Synergies between data acquired from different satellites are required to improve the monitoring of the forest ecosystems, forest disturbances and consequent atmospheric carbon emissions and 84  removals. This is relevant for forest monitoring, the sustainable management of our forest resources, the development of REDD+ strategies, and climate change mitigation actions. High resolution remotely sensed data such as Light Detection and Ranging (LiDAR) from aircraft and spaceborne platforms and RapidEye satellite imagery are being tested to improve forest monitoring studies.   Based on findings of this research the following is recommended:  Investigate approaches to improve the accuracy of attribution of land-cover changes to their particular forest disturbance driver.   Integrate Landsat observations with other remote sensing sources, particularly LiDAR or RapidEye.  Increase the spatial and temporal resolution of activity data to reduce uncertainties in areas that have been identified as relevant for biodiversity and habitat conservation, ecosystem services provisioning, and carbon monitoring for REDD+.  Replicate this study in other biomes, particularly in temperate forests that may exhibit different carbon emission and removal responses after disturbance.  Combining data sources provides a cost-effective approach to reduce uncertainties and increases their potential use for forest monitoring and REDD+, often filling the spatial and temporal gaps between them. Strategies to develop a national Monitoring Reporting and Verification system should include a combination of data sources, not only from remote sensing products, but also from ground-based measurements, intensive monitoring sites and ancillary datasets.  85  Results from this study can provide some insight into which of the following considerations (increased spatial resolution, annual observations or attributing changes to corresponding disturbance types ) are the most critical for deriving activity data for carbon budget modeling. Based on the results of this study it is suggested:   1. Spatial resolution. The contrasting difference in results from carbon simulations generated with MODIS and Landsat-derived disturbance observations suggests that increased spatial resolution should be the first priority when deriving activity data for carbon modeling. Carbon simulations derived from Landsat at higher spatial resolution showed a more realistic scenario where carbon fluxes reflect the complex dynamics of the forest. In contrast, the coarse resolution of MODIS-derived products reflected a utopic scenario where the forest constantly sequesters carbon across the landscape, mainly remaining a sink over the observation period. The latter occurred, despite having MODIS observations on an annual basis from 2005-2010 with 86% of the changed pixels attributed by disturbance type.  2. Annual observations. Reliable estimates of carbon dynamics require detailed observations of drivers of change on an annual basis (Kurz, 2010). Carbon fluxes predicted with Hansen showed a decrease in total ecosystem carbon content in 2009. However, due to the lack of disturbance events detected with VCT in 2009, corresponding decreases in the carbon stocks from the 86  disturbance events were not reflected in the CBM-CFS3 estimates, despite the fact that the disturbance events observed in 2010 were equally distributed in 2009 and 2010. This result showed that even missing a single year in the land-cover observations can lead to substantial errors; especially in ecosystems with rapid-regrowth forests, such as the Yucatan Peninsula.  The latter trend was observed in both, the attributed and the non-attributed observations; which made having annual-observations a higher priority than the attribution.  3. Attribution of land-cover changes by disturbance type. Identifying the disturbance driver of land-cover change is essential to accurately quantify their particular impact over the forest-carbon dynamics.  Each disturbance type affects the forest structure and successional dynamics in a unique way, changing the amount of carbon that is stored among the vegetation and soils and released into the atmosphere. While 86% of the MODIS land-cover change pixels were attributed to specific disturbance types; due to its coarse spatial resolution, the difference between attributed and non-attributed observations was not perceptible in the simulation-runs.    When increasing the spatial resolution using Landsat observations, only a small percentage of the land-cover change detected was attributed by disturbance type, assuming most of it as harvesting.  However, in the INEGI products classification-approach, the larger areas of change detected over the second-period observations showed that attributing changes by disturbance type is a worthwhile investment. In the second period, the attributed estimates were  40% less than the non-attributed, releasing 87  around 4.7 million tC. These differences between with and without attribution would be even more pronounced if they are reported as CO2 equivalent emissions (as required in REDD+ projects), because fires cause additional non-CO2 GHG emissions (in the form of CH4 and N2O), with much higher global warming potentials than CO2. 88  References  Achard, F., Beuchle, R., Mayaux, P., Stibig, H., Bodart, C., Brink, A., Carboni, S., Desclee, B., Donnay, F., Eva, H.D., Lupi, A., Rasi, R., Seliger, R. & Simonetti, D. (2014). Determination of tropical deforestation rates and related carbon losses from 1990 to 2010. Global Change Biology, 20(8), 2540-2554. Adams, H.D., Williams, A.P., Xu, C., Rauscher, S.A., Jiang, X. & Mcdowell, N.G. (2013). Empirical and process-based approaches to climate-induced forest mortality models. Frontiers In Plant Science, 4, 438. Arredondo-León, C., Muñoz-Jimenez, J., & García-Romero, A. (2008). Recent changes in landscape-dynamics trends in tropical highlands, central Mexico. Interciencia, 33(8), 569-577. Asbjornsen, H., Velazquez-Rosas, N., García-Soriano, R., & Gallardo-Hernández, C. (2005). Deep ground fires cause massive above- and below-ground biomass losses in tropical montane cloud forests in Oaxaca, Mexico. Journal of Tropical Ecology, 21, 427-434. Asner, G.P., Powell, G.V.N., Mascaro, J., Knapp, D.E., Clark, J.K., Jacobson, J., Kennedy-Bowdoin, T., Balaji, A., Paez-Acosta, G., Victoria, E., Secada, L., Valqui, M. & Hughes, R.F. (2010). High-resolution forest carbon stocks and emissions in the Amazon. Proceedings of the National Academy of Sciences of the United States of America, 107(38), 16738-16742. Baccini, A., Goetz, S.J., Walker, W.S., Laporte, N.T., Sun, M., Sulla-Menashe, D., Hackler, J., Beck, P.S.A., Dubayah, R., Friedl, M.A., Samanta, S. & Houghton, R.A. (2012). Estimated 89  carbon dioxide emissions from tropical deforestation improved by carbon-density maps. Nature Climate Change, 2(3), 182-185. Birdsey, R., Angeles-Perez, G., Kurz, W.A., Lister, A., Olguin, M., Pan, Y., Wayson, C., Wilson, B. & Johnson, K. (2013a). Approaches to monitoring changes in carbon stocks for REDD+. Carbon Management, 4(5), 519-537. Birdsey, R., Pan, Y., & Houghton, R. (2013b) Sustainable landscapes in a world of change: tropical forests, land use and implementation of REDD+: Part II, Carbon Management, 4(6), 567-569 Bosworth, D., Birdsey, R., Joyce, L., & Millar, C. (2008). Climate Change and the Nation’s Forests: Challenges and Opportunities. Journal of Forestry, 214–221. Brown, S., & Lugo, A. E. (1982). The storage and production of organic matter in tropical forests and their role in the global carbon cycle. Biotropica, 14(3), 161-187.  Brown, S. (2002). Measuring carbon in forests: current status and future challenges. Environmental pollution, 116(3), 363–72.  CEC (1997). Ecological regions of North America: Toward a common perspective. Montreal, Quebec, Canada: Commission for Environmental Cooperation, 71 p. Chapa-Bezanilla, D., Sosa-Ramírez, J., & de Alba-Avila, A. (2008). Multitemporal study on forest fragmentation in Sierra Fría, Aguascalientes, México. Madera Y Bosques, 14(1), 37-51. Chapin, F. S., Woodwell, G. M., Randerson, J. T., Rastetter, E. B., Lovett, G. M., Baldocchi, D. D., Clark, D. a., et al. (2006). Reconciling carbon-cycle concepts, terminology, and methods. Ecosystems, 9(7), 1041–1050. 90  Chapin, F. S., McFarland, J., McGuire, A. D., Euskirchen, E. S., Ruess, R. W., & Kielland, K. (2009). The changing global carbon cycle: Linking plant-soil carbon dynamics to global consequences. Journal of Ecology, 97(5), 840-850.  Cohen, W. B., Spies, T. A., Alig, R. J., Oetter, D. R., Maiersperger, T. K., & Fiorella, M. (2002). Characterizing 23 years (1972-95) of stand replacement disturbance in western Oregon forests with Landsat imagery. Ecosystems, 5(2), 122-137.  Cohen, W. B., Yang, Z., & Kennedy, R. (2010). Detecting trends in forest disturbance and recovery using yearly Landsat time series: 2. TimeSync - tools for calibration and validation. Remote Sensing of Environment, 114(12), 2911-2924. Colditz, R. R., López-Saldaña, G., Maeda, P., Espinoza, J. A., Tovar, C. M., Hernández, A. V., Benítez, C. Z., et al. (2012). Generation and analysis of the 2005 land cover map for Mexico using 250m MODIS data. Remote Sensing of Environment, 123, 541–552. Congalton, R. (1991). A review of assessing the accuracy of classifications of remotely sensed data. Remote Sensing of Environment, 37(1), 35-46. Coops N.C., White J.C., Wulder, M.A. (2006).  Identifying and describing forest disturbance and spatial pattern. In M.A. Wulder & S. E. Franklin (Ed), Understanding forest disturbance and spatial pattern: remote sensing and gis approaches, CRC Taylor & Francis, 31-61. Covey, K. & Orefice, J. (2009). The Physiological ecology of carbon science in forests stands. In M. L. Tyrrell, M. S. Ashton, D. Spalding, & B. Gentry (Eds.), Forests and carbon: a synthesis of science, management, and policy for carbon sequestration in forests. Yale Printing & Publishing Services, 223-253. Curran, P., & Atkinson, P. (1998). Geostatistics and remote sensing. Progress in Physical Geography, 22(1), 61-78. 91  Dai, X. & Khorram, S. (1999). Remotely sensed change detection based on artificial neural networks. Photogrammetric Engineering and Remote Sensing, 65(10), 1187-1194. Dai, Z., Birdsey, R.A., Johnson, K.D., Manuel Dupuy, J., Luis Hernandez-Stefanoni, J. & Richardson, K. (2014). Modeling Carbon Stocks in a Secondary Tropical Dry Forest in the Yucatan Peninsula, Mexico. Water Air and Soil Pollution, 225(4), 1925. Dale, V., Joyce, L., Mcnulty, S., Neilson, R., Ayres, M., Flannigan, M., Hanson, P., Irland, L., Lugo, A., Peterson, C., Simberloff, D., Swanson, F., Stocks, B. & Wotton, B. (2001). Climate Change and Forest Disturbances. BioScience, 51(9), 723-734. De Sy, V., Herold, M., Achard, F., Asner, G. P., Held, A., Kellndorfer, J., & Verbesselt, J. (2012). Synergies of multiple remote sensing data sources for REDD+ monitoring. Current Opinion in Environmental Sustainability, 4(6), 696-706.  De’ath, G. (2002). Multivariate regression trees: a new technique for modeling species-environment relationships. Ecology, 83(4) 1105-1117. De'ath, G., & Fabricius, K. (2000). Classification and regression trees: A powerful yet simple technique for ecological data analysis. Ecology, 81(11), 3178-3192. DeFries, R., Houghton, R., Hansen, M., Field, C., Skole, D., & Townshend, J. (2002). Carbon emissions from tropical deforestation and regrowth based on satellite observations for the 1980s and 1990s. Proceedings of the National Academy of Sciences of the United States of America, 99(22), 14256-14261.  DeFries, R., Achard, F., Brown, S., Herold, M., Murdiyarso, D., Schlamadinger, B., & de Souza Jr., C. (2007). Earth observations for estimating greenhouse gas emissions from deforestation in developing countries. Environmental Science & Policy, 10(4), 385-394.  92  Dickinson, M., Whigham, D., & Hermann, S. (2000). Tree regeneration in felling and natural treefall disturbances in a semideciduous tropical forest in Mexico. Forest Ecology and Management, 134(1-3), 137-151. Drury, S. A. (2006). The effects of climate and disturbance on madrean pine-oak forests in Mexico's Sierra Madre Occidental. University of Colorado at Boulder.  ProQuest Dissertations and Theses, 235-235.  Endara Agramont, A.R., Franco Maass, S., Nava Bernal, G., Valdez Hernández, J. I., & Fredericksen Todd, S. (2012). Effect of human disturbance on the structure and regeneration of forests in the Nevado de Toluca National Park, Mexico. Journal of Forestry Research, 23(1), 39-44. Espirito-Santo, F.D.B., Gloor, M., Keller, M., Malhi, Y., Saatchi, S., Nelson, B., Oliveira Junior, R.C., Pereira, C., Lloyd, J., Frolking, S., Palace, M., Shimabukuro, Y.E., Duarte, V., Mendoza, A.M., Lopez-Gonzalez, G., Baker, T.R., Feldpausch, T.R., Brienen, R.J.W., Asner, G.P., Boyd, D.S. & Phillips, O.L. (2014). Size and frequency of natural forest disturbances and the Amazon forest carbon balance. Nature Communications, 5, 3434. Food and Agriculture Organization (FAO), United Nations. (2010). The Global Forest Resources Assessment (FRA) 2010. Rome, Retrieved  January 1, 2013,  from: http://www.fao.org/docrep/013/i1757e/i1757e.pdf Foster, D. R., Knight, D. H., & Franklin, J. F. (1998). Landscape patterns and legacies resulting from large, infrequent forest disturbances. Ecosystems, 1(6), 497–510.  Franklin, J. F., Mitchell, R. J., & Palik, B. (2007). Natural disturbance and stand development principles for ecological forestry. USDA For Serv N Res. US Dept. of Agriculture, Forest Service, Northern Research Station.  93  Fulé, P.Z., & Covington, W. W. (1997). Fire regimes and forest structure in the Sierra Madre Occidental, Durango, Mexico. Acta Botanica Mexicana, 43 - 79. Fulé, P.Z., & Covington, W. W. (1999). Fire Regime Changes in La Michilía Biosphere Reserve, Durango, Mexico, Conservation Biology,  13(3), 640-652. Fuller, R.M., Smith, G.M. & Devereux, B.J. (2003). The characterisation and measurement of land cover change through remote sensing: problems in operational applications? International Journal of Applied Earth Observation and Geoinformation, 4(3), 243-253. GFOI (2013). Integrating remote-sensing and ground-based observations for estimation of emissions and removals of greenhouse gases in forests: Methods and Guidance from the Global Forest Observations Initiative: Pub: Group on Earth Observations, Geneva, Switzerland. GOFC-GOLD (2010). A sourcebook of methods and procedures for monitoring and reporting anthropogenic greenhouse gas emissions and removals caused by deforestation, gains and losses of carbon stocks in forests remaining forests, and forestation. Report version COP16-1. Natural Resources Canada, Alberta, Canada. Hansen, M. C., Roy, D. P., Lindquist, E., Adusei, B., Justice, C. O., & Altstatt, A. (2008). A method for integrating MODIS and Landsat data for systematic monitoring of forest cover and change in the Congo Basin. Remote Sensing of Environment, 112(5), 2495-2513.  Hansen, M. C., Stehman, S. V., & Potapov, P. V. (2010). Quantification of global gross forest cover loss. Proceedings of the National Academy of Sciences of the United States of America, 107(19), 8650-8655.  94  Hansen, M.C., Potapov, P.V., Moore, R., Hancher, M., Turubanova, S.A., Tyukavina, A., Thau, D., Stehman, S.V., Goetz, S.J., Loveland, T.R., Kommareddy, A., Egorov, A., Chini, L., Justice, C.O. & Townshend, J.R.G. (2013). High-resolution global maps of 21st-century forest cover change. Science, 342(6160), 850-853. Hayes, D. J., & Cohen, W. B. (2007). Spatial, spectral and temporal patterns of tropical forest cover change as observed with multiple scales of optical satellite data. Remote Sensing of Environment, 106(1), 1-16.  Hayes, D. J., Cohen, W. B., Sader, S. A., & Irwin, D. E. (2008). Estimating proportional change in forest cover as a continuous variable from multi-year MODIS data. Remote Sensing of Environment, 112(3), 735-749. Healey, S., Cohen, W., Yang, Z., & Krankina, O. (2005). Comparison of tasseled cap-based landsat data structures for use in forest disturbance detection. Remote Sensing of Environment, 97(3), 301-310.  Herold, M. & Skutsch, M. (2011). Monitoring, reporting and verification for national REDD plus programmes: two proposals. Environmental Research Letters, 6(1), 014002. Hewson, J., Steininger, M. & Pesmajoglou, S. (2013). REDD+ Measurement, Reporting and Verification (MRV) Manual. USAID-supported Forest Carbon, Markets and Communities Program. Washington, DC, USA. Hilker, T., Wulder, M. A., Coops, N. C., Linke, J., McDermid, G., Masek, J. G., . . . White, J. C. (2009). A new data fusion model for high spatial- and temporal-resolution mapping of forest disturbance based on landsat and MODIS. Remote Sensing of Environment, 113(8), 1613-1627.  95  Hosonuma, N., Herold, M., De Sy, V., De Fries, R.S., Brockhaus, M., Verchot, L., Angelsen, A. & Romijn, E. (2012). An assessment of deforestation and forest degradation drivers in developing countries. Environmental Research Letters, 7(4), 044009. Houghton, R. A., House, J. I., Pongratz, J., van der Werf, G. R., DeFries, R. S., Hansen, M. C., Le Quéré, C.,  Ramankutty, N. (2012). Carbon emissions from land use and land-cover change. Biogeosciences, 9(12), 5125-5142.  Huang, C., Coward, S.N., Masek, J.G., Thomas, N., Zhu, Z. & Vogelmann, J.E. (2010). An automated approach for reconstructing recent forest disturbance history using dense Landsat time series stacks. Remote Sensing of Environment,114(1), 183-198. INEGI (2003). Conjunto de datos vectoriales de la carta de uso del suelo y vegetación, Escala 1:250,000, Serie IV (CONTINUO NACIONAL). Instituto Nacional de Estadística y Geografía (INEGI).  Aguascalientes, México. INEGI (2007). Conjunto de datos vectoriales de la carta de uso del suelo y vegetación, Escala 1:250,000, Serie IV (CONTINUO NACIONAL). Instituto Nacional de Estadística y Geografía (INEGI).  Aguascalientes, México. INEGI (2010). Conjunto de datos vectoriales de la carta de uso del suelo y vegetación, Escala 1:250,000, Serie V (CONTINUO NACIONAL). Instituto Nacional de Estadística y Geografía (INEGI).  Aguascalientes, México. INEGI (2012). Marco geoestadistico nacional (National geostatistic framework). Instituto Nacional de Estadistica y Geografia (INEGI), Aguascalientes, México. INEGI (2013). Conjunto de Datos Vectoriales de Carreteras y Vialidades Urbanas Edición 1.0. Instituto Nacional de Estadística y Geografía (INEGI).  Aguascalientes, México. 96  INFyS (2012). Inventario Nacional Forestal y de Suelos Informe 2004- 2009 (National Forest and Soils Inventory report). National Forestry Commission (In spanish: Comision Nacional Forestal). México.  Intergovernmental Panel on Climate Change (IPCC). (2006). Guidelines for national greenhouse gas inventories. Institute for Global Environmental Strategies, Hayama. Intergovernmental Panel on Climate Change (IPCC), 2007. Fourth Assessment Report (AR4). Climate Change 2007: Synthesis report. IPCC. Jaramillo, V.J., Kauffman, J.B., Rentería-Rodríguez, L., Cummings, D.L., Ellingso, L.J. (2003). Biomass, carbon, and nitrogen pools in Mexican tropical dry forest landscapes. Ecosystems, 6(7), 609-629 Jardel-Peláez, E.J., Morfín-Ríos, J.E., Vargas-Jaramillo, S., Michel-Fuentes, J.M., Cuevas-Guzmán, R., Castillo-Navarro, F., Balcázar-Medina, O.E.  & Quiñónez, E. (2006). Fire regime and fire effects on western Mexico subtropical montane forest ecosystems. Proceedings of the 3rd International Congress on Fire Ecology & Management. Special Session: Fire Regimes and Fire Effects in Mexican Ecosystems. San Diego, California. Kull, S. J., G. J. Rampley, S. Morken, J. Metsaranta, E. T. Neilson, & W. A. Kurz (2011), Operational-scale Carbon Budget Model of the Canadian Forest Sector (CBM-CFS3) Version 1.2: User’s Guide, Natural Resources Canada, Canadian Forest Service, Northern Forestry Centre, Edmonton, AB. Kurz, W. A., & Apps, M. J. (2006). Developing Canada’s national forest carbon monitoring, accounting and reporting system to meet the reporting requirements of the Kyoto Protocol, Mitigation and Adaptation Strategies for Global Change,11, 33–43. 97  Kurz, W. A. (2010). An ecosystem context for global gross forest cover loss estimates. Proceedings of the National Academy of Sciences of the United States of America, 107(20), 9025-9026.  Kurz, W. A., Dymond, C. C., White, T. M., Stinson, G., Shaw, C. H., Rampley, G. J., Smyth, C., Simpson, B.N., Neilson, E.T., Trofymow, J.A., Metsaranta, J., Apps, M. J. (2009). CBM-CFS3: A model of carbon-dynamics in forestry and land-use change implementing IPCC standards. Ecological Modelling, 220(4), 480-504. Kurz, W.A., Shaw, C.H.,  Boisvenue, C., Stinson, G., Metsaranta, J., Leckie, D., Dyk, A., Smyth, C., Neilson, E.T. (2013) Carbon in Canada’s boreal forest — A synthesis.Environmental Reviews, 21, 260-292.  Langner, A., Samejima, H., Ong, R.C., Titin, J. & Kitayama, K. (2012). Integration of carbon conservation into sustainable forest management using high resolution satellite imagery: A case study in Sabah, Malaysian Borneo. International Journal of Applied Earth Observation and Geoinformation, 18, 305-312. Latifovic, R., Homer, C., Ressl, R., Pouliot, D., Hossain, S.N., Colditz, R.R., Olthof, I., Giri, C. & Victoria, A. (2010). North American land Change Monitoring System (NALCMS). In: Giri, C. (Ed): Remote sensing of land use and land cover: Principles and applications. CRC Press. Ley General de Cambio Climatico (LGCC). (2012). General Law of climate change. Camara de Diputados, Congreso de la Union, Mexico. Linke, J., Betts, M. G., Lavigne, M. B. & Franklin, S. E. (2007). Introduction: Structure, function, and change of forest landscapes. In S. E. Franklin (Ed), Understanding forest 98  disturbance and spatial pattern: remote sensing and gis approaches, CRC Taylor & Francis, 1-29 Lorenz, K. & Lal., R. (2010). Effects of disturbance, succession and management on carbon sequestration. In Carbon sequestration in forest ecosystems. Springer Netherlands, 103-157. Lu, D., Mausel, P., Brondízio, E. & Moran, E. (2004). Change detection techniques, International Journal of Remote Sensing, 25(12), 2365-2401. Manzo-Delgado, L., Sánchez-Colon, S., & Álvarez, R. (2009). Assessment of seasonal forest fire risk using NOAA-AVHRR: A case study in central Mexico. International Journal of Remote Sensing, 30(19), 4991-5013.  Mas, J. (1999). Monitoring land-cover changes: A comparison of change detection techniques. International Journal of Remote Sensing, 20(1), 139-152.  Mascorro, V.S., Coops, N.C, Kurz, W.A., & Olguin, M. (2014). Attributing Changes In Land Cover Using Independent Disturbance Datasets: A Case Study Of The Yucatan Peninsula, Mexico. Regional Environmental Change (in review). Masek, J.G. & Collatz, G.J. (2006). Estimating forest carbon fluxes in a disturbed southeastern landscape: Integration of remote sensing, forest inventory, and biogeochemical modeling. Journal of Geophysical Research-Biogeosciences,111(G1), G01006. Masek, J.G., Huang, C., Wolfe, R., Cohen, W., Hall, F., Kutler, J. & Nelson, P. (2008). North American forest disturbance mapped from a decadal Landsat record. Remote Sensing of Environment, 112(6), 2914-2926. Masek, J. G., Cohen, W. B., Leckie, D., Wulder, M. A., Vargas, R., de Jong, B., Healey, S., Law, B., Birdsey, R., Houghton, R. A., Mildrexler, D., Goward, S., Smith, W. B. (2011). Recent 99  rates of forest harvest and conversion in North America. Journal of Geophysical Research-Biogeosciences, 116. Matricardi, E. A. T., Skole, D. L., Pedlowski, M. A., Chomentowski, W., & Fernandes, L. C. (2010). Assessment of tropical forest degradation by selective logging and fire using landsat imagery. Remote Sensing of Environment, 114(5), 1117-1129.   Melendez, K. V., Jones, D. L., & Feng, A. S. (2006). Classification of communication signals of the little brown bat. Journal of the Acoustical Society of America, 120(2), 1095-1102 Meng, Q., Liu, Z., & Borders, B. E. (2013). Assessment of regression kriging for spatial interpolation – comparisons of seven GIS interpolation methods. Cartography and Geographic Information Science, 40(1), 28-39.   Millennium Ecosystem Assessment (MEA) (2005). Ecosystems and Human Well-being: Synthesis. Island Press, Washington, DC Mitchard, E.T.A., Feldpausch, T.R., Brienen, R.J.W., Lopez-Gonzalez, G., Monteagudo, A., Baker, T.R., Lewis, S.L., Lloyd, J., Quesada, C.A., Gloor, M., Ter Steege, H., Meir, P., Alvarez, E., Araujo-Murakami, A., Aragao, L.E.O.C., Arroyo, L., Aymard, G., Banki, O., Bonal, D., Brown, S., Brown, F.I., Ceron, C.E., Chama Moscoso, V., Chave, J., Comiskey, J.A., Cornejo, F., Corrales Medina, M., Da Costa, L., Costa, F.R.C., Di Fiore, A., Domingues, T.F., Erwin, T.L., Frederickson, T., Higuchi, N., Honorio Coronado, E.N., Killeen, T.J., Laurance, W.F., Levis, C., Magnusson, W.E., Marimon, B.S., Marimon Junior, B.H., Mendoza Polo, I., Mishra, P., Nascimento, M.T., Neill, D., Nunez Vargas, M.P., Palacios, W.A., Parada, A., Pardo Molina, G., Pena-Claros, M., Pitman, N., Peres, C.A., Poorter, L., Prieto, A., Ramírez-Angulo, H., Restrepo Correa, Z., Roopsind, A., Roucoux, K.H., Rudas, A., Salomao, R.P., Schietti, J., Silveira, M., De Souza, P.F., 100  Steininger, M.K., Stropp, J., Terborgh, J., Thomas, R., Toledo, M., Torres-Lezama, A., Van Andel, T.R., Van Der Heijden, G.M.F., Vieira, I.C.G., Vieira, S., Vilanova-Torre, E., Vos, V.A., Wang, O., Zartman, C.E., Malhi, Y. And Phillips, O.L. (2014). Markedly divergent estimates of Amazon forest carbon density from ground plots and satellites. Global Ecology and Biogeography, 23(8), 935-946. Nabuurs, G.J., O. Masera, K. Andrasko, P. Benitez-Ponce, R. Boer, M. Dutschke, E. Elsiddig, J. Ford-Robertson, P. Frumhoff, T. Karjalainen, O. Krankina, W.A. Kurz, M. Matsumoto, W. Oyhantcabal, N.H. Ravindranath, M.J. Sanz Sanchez, X. Zhang. (2007). Forestry. In Climate Change 2007: Mitigation. Contribution of Working Group III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change [B. Metz, O.R. Davidson, P.R. Bosch, R. Dave, L.A. Meyer (eds)], Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA. Naeem S., Bunker, D.E., Hector, A. Loreau, M. Perrings, C. (2009).  Incorporating biodiversity in climate change mitigation initiatives Biodiversity, Ecosystem Functioning & Human Wellbeing. Oxford University Press. National Climatic Data Center (NCDC). (2012). Global set of historical tropical cyclones data. National Climatic Data Center, National Oceanic and Atmospheric Administration (NOAA), United States. National Hurricane Center (NHC). (2013). Saffir-Simpson Hurricane Wind Scale. National Oceanic and Atmospheric Administration (NOAA), United States. Retrieved  Novermber 1, 2012, from: http://www.nhc.noaa.gov/aboutsshws.php  101  Navarro-Martínez, A., Durán-García, R. & Méndez-González, M. (2012). El impacto del huracán Dean sobre la estructura y composición arbórea de un bosque manejado en Quintana Roo, México. Madera y Bosques, 18(1), 57-76 Oliver, C.D. & Larson, B.C. (1996). Forest Stand Dynamics. John Wiley & Sons, New York. 520p. Olofsson, P., Foody, G.M., Herold, M., Stehman, S.V., Woodcock, C.E. & Wulder, M.A. (2014). Good practices for estimating area and assessing accuracy of land change. Remote Sensing of Environment, 148, 42-57. Orihuela-Belmonte, D.E., De Jong, B.H.J., Mendoza-Vega, J., Van Der Wal, J., Paz-Pellat, F., Soto-Pinto, L. & Flamenco-Sandoval, A.(2013). Carbon stocks and accumulation rates in tropical secondary forests at the scale of community, landscape and forest type. Agriculture Ecosystems & Environment, 171, 72-84. Pan, Y., Birdsey, R.A., Fang, J., Houghton, R., Kauppi, P.E., Kurz, W.A., Phillips, O.L., Shvidenko, A., Lewis, S.L., Canadell, J.G., Ciais, P., Jackson, R.B., Pacala, S.W., Mcguire, A.D., Piao, S., Rautiainen, A., Sitch, S. & Hayes, D. (2011). A large and persistent carbon sink in the world's forests. Science, 333(6045), 988-993.  Pflugmacher, D., Cohen, W. B., & Kennedy, R. E. (2012). Using landsat-derived disturbance history (1972-2010) to predict current forest structure. Remote Sensing of Environment, 122, 146-165.  Potapov, P., Hansen, M. C., Stehman, S. V., Loveland, T. R., & Pittman, K. (2008). Combining MODIS and Landsat imagery to estimate and map boreal forest cover loss. Remote Sensing of Environment, 112(9), 3708-3719. 102  Potter, C.  Klooster, S., Genovese, V., Hiatt, C., Boriah, S., Kumar, V., Mithal, V., & Garg, A.  (2012). Terrestrial ecosystem carbon fluxes predicted from MODIS satellite data and large-scale disturbance modeling. International Journal of Geosciences, 3(3), 469-479. Pouliot, D., Latifovic, R., Fernandes, R., & Olthof, I. (2009). Evaluation of annual forest disturbance monitoring using a static decision tree approach and 250 m MODIS data. Remote Sensing of Environment, 113(8), 1749-1759. Powers, R.P., Hermosilla, T., Coops, N.C. & Chen, G. (2015). Remote sensing and object-based techniques for mapping fine-scale industrial disturbances. International Journal of Applied Earth Observation and Geoinformation, 34(0), 51-57. Price, S. & Ashton, M. (2009). Characterizing organic carbon stocks and flows in forest soils. In M. L. Tyrrell, M. S. Ashton, D. Forests and carbon: a synthesis of science, management, and policy for carbon sequestration in forests. Yale Printing & Publishing Services, 11-44. Ramírez-Marcial, N., González-Espinosa, M., & Williams-Linera, G. (2001). Anthropogenic disturbance and tree diversity in montane rain forests in Chiapas, Mexico. Forest Ecology and Management, 154(1-2), 311-326. Randerson, J., Chapin, F., Harden, J., Neff, J. & Harmon, M. (2002). Net ecosystem production: A comprehensive measure of net carbon accumulation by ecosystems. Ecological Applications, 12(4), 937-947. Read, L., & Lawrence, D. (2003). Recovery of biomass following shifting cultivation in dry tropical forests of the Yucatan. Ecological Applications, 13(1), 85-97. Rogan, J., Schneider, L., Christman, Z., Millones, M., Lawrence, D., & Schmook, B. (2011). Hurricane disturbance mapping using MODIS EVI data in the southeastern Yucatan, Mexico. Remote Sensing Letters, 2(3), 259-267. 103  Saatchi, S.S., Harris, N.L., Brown, S., Lefsky, M., Mitchard, E.T.A., Salas, W., Zutta, B.R., Buermann, W., Lewis, S.L., Hagen, S., Petrova, S., White, L., Silman, M. And Morel, A.(2011). Benchmark map of forest carbon stocks in tropical regions across three continents. Proceedings of the National Academy of Sciences of the United States of America, 108(24), 9899-9904. SAGARPA (2012). Servicio de informacion agroalimentaria y de consulta 2010 (SIACON). Secretaria de Agricultura, Ganaderia, Desarrollo Rural, Pesca y Alimentacion (SAGARPA), Mexico. Sánchez-Gallén, I., Javier Álvarez-Sánchez, F., & Benitez-Malvido, J. (2010). Structure of the advanced regeneration community in tropical rain forest fragments of Los Tuxtlas, Mexico. Biological Conservation, 143(9), 2111-2118. Schroeder, T. A., Wulder, M. A., Healey, S. P., & Moisen, G. G. (2011). Mapping wildfire and clear-cut harvest disturbances in boreal forests with Landsat time series data. Remote Sensing of Environment, 115(6), 1421-1433.  Schwalm, C.R., Black, T.A., Amiro, B.D., Arain, M.A., Barr, A.G., Bourque, C.P.-., Dunn, A.L., Flanagan, L.B., Giasson, M., Lafleur, P.M., Margolis, H.A., Mccaughey, J.H., Orchansky, A.L. & Wofsy, S.C., (2006). Photosynthetic light use efficiency of three biomes across an east–west continental-scale transect in Canada. Agricultural and Forest Meteorology, 140(1–4), 269-286 SEMARNAT (2009). Special Program on Climate Change 2009-2012. Secretaria de Medio Ambiente y Recursos Naturales (SEMARNAT), Mexico. Retrieved  September 10, 2012, from: http://www.semarnat.gob.mx/programas/Documents/PECC_DOF.pdf 104  Skinner, C. N., Burk, J. H., Barbour, M. G., Franco-Vizcaino, E., & Stephens, S. L. (2008). Influences of climate on fire regimes in montane forests of north-western mexico. Journal of Biogeography, 35(8), 1436-1451. Skutsch, M., De Los Rios, E., Solis, S., Riegelhaupt, E., Hinojosa, D., Gerfert, S., Gao, Y. & Masera, O. (2011). Jatropha in Mexico: Environmental and Social Impacts of an Incipient Biofuel Program. Ecology and Society, 16(4), 11. Skwira, G., Schroeder, J., & Peterson, R. (2005). Surface observations of landfalling hurricane rainbands. Monthly Weather Review, 133(2), 454-465. Spalding, D. (2009). The role of forests in global carbon budgeting. In M. L. Tyrrell, M. S. Ashton, D. Forests and carbon: a synthesis of science, management, and policy for carbon sequestration in forests. Yale Printing & Publishing Services, 223-253. Stinson, G., Kurz, W.A., Smyth, C.E., Neilson, E.T., Dymond, C.C., Metsaranta, J.M., Boisvenue, C., Rampley, G.J., Li, Q., White, T.M. And Blain, D. (2011). An inventory-based analysis of Canada's managed forest carbon dynamics, 1990 to 2008. Global Change Biology, 17(6), 2227-2244. Tobler, W. (1970). Computer movie simulating urban growth in detroit region. Economic Geography, 46(2), 234-240. Toledo-Aceves, T., Purata-Velarde, S., & Peters, C. M. (2009). Regeneration of commercial tree species in a logged forest in the Selva Maya, Mexico. Forest Ecology and Management, 258(11), 2481-2489. Turner, M. (1989). Landscape ecology - the effect of pattern on process. Annual Review of Ecology and Systematics, 20, 171-197.  105  Turner, M. G. (2010). Disturbance and landscape dynamics in a changing world. Ecology, 91(10), 2833-2849. UNFCCC (1992). United Nations Framework Convention on Climate Change. United Nations. UNFCCC (2010). Outcome of the work of the ad hoc working group on long-term cooperative action under the convention—policy approaches and positive incentives on issues relating to reducing emissions from deforestation and forest degradation in developing countries: and the role of conservation, sustainable management of forests and enhancement of forest carbon stocks in developing countries (REDD+),  UNFCCC COP 16 Cancun. Retrieved September 26, 2012, from: http://unfccc.int/2860.php Urquiza-Haas, T., Dolman, P. M., & Peres, C. A. (2007). Regional scale variation in forest structure and biomass in the Yucatan Peninsula, Mexico: Effects of forest disturbance. Forest Ecology and Management, 247(1-3), 80-90.  Vandecar, K. L., Lawrence, D., Richards, D., Schneider, L., Rogan, J., Schmook, B., & Wilbur, H. (2011). High mortality for rare species following hurricane disturbance in the southern Yucatan. Biotropica, 43(6), 676-684.  Vargas, R., Allen, M. F., & Allen, E. B. (2008). Biomass and carbon accumulation in a fire chronosequence of a seasonally dry tropical forest. Global Change Biology, 14(1), 109-124. Whigham, D. F., Olmsted, I., Cano, E. C., & Curtis, A. B. (2003). Impacts of hurricanes on the forests of Quintana Roo, Yucatan Peninsula, Mexico. In Gomez-Pompa, A Allen, MF Fedick, SL Jimenez-Osornio, JJ (Eds.), The Lowland Maya Area. Wulder, M.A. & Coops, N.C (2014). Satellites: Make Earth observations open access. Nature. 513(7516), 30-31 106  Wulder, M. A., White, J. C., Gillis, M. D., Walsworth, N., Hansen, M. C., & Potapov, P. (2010). Multiscale satellite and spatial information and analysis framework in support of a large-area forest monitoring and inventory update. Environmental Monitoring and Assessment, 170(1-4), 417-433.  

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}]}"
                            data-media="{[{embed.selectedMedia}]}"
                            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.24.1-0165575/manifest

Comment

Related Items