LANDSCAPE CARBON MEASUREMENT IN SYSTEMS OF TREES OUTSIDE OF FORESTS: THE CASE OF AGROFORESTRY SYSTEMS IN RURAL SAVANNAS OF SENEGAL By Moussa Dieng A DISSERTATION Submitted to Michigan State University in partial f ulfillment of the requirements for the d egree of Forestry - D octor of P hilosophy 2015 ABSTRACT LANDSCAPE CARBON MEASUREMENT IN SYSTEMS OF TREES OUTSIDE OF FORESTS: THE CASE OF AGROFORESTRY SYSTEMS IN RURAL SAVANNAS OF SENEGAL By Moussa Dieng Monitoring forest cover changes and c arbon content at r egional and global level by using remote sensing has advanced significantly for closed humid tropical forests, and several methods have been developed. On the other hand monitoring of cover and carbon for semi - arid savannas and woodlands has been under - studied, and more methods are needed. Further, for systems of Trees Outside of Forests in semi - arid regions , monitoring of cover and carbon has not advanced at all, and new methods are needed. In this dissertation , I develop ed methods for remote sensing - assisted carbon measurement and monitoring in semi - arid landscapes of Trees Outside of Forests (TOF) which include: - Plantation and Agroforestry systems - Other trees on farms: planted trees - Other trees on farms: remnant savanna trees In this study, remote sensing satellite techniques were used to estimate carbon at a landscape level in savanna systems. Remote sensing tree crown projection area (CPA) was used as a proxy to predict tree diameter at breast height (DBH). A relationship was established between remote sensing based crown projection area (CPA) and the field - based DBH. Both simple linear and non - linear regression analysis were applied to the Sokone and Karang sites in Senegal. The linear function presents a higher coefficient of determination (R 2 ) in both sites with respectively R 2 = 0.71 and R 2 = 0.79 for Sokone and Karang sites. The non - linear model shows R - squared values ranged between 0.61 and 0.77 for Sokone and Karang sites respectively. The regression equations derived fr om the relationship between remote sensing - based crown projection area (CPA) and the field - based DBH are used to predict the DBH of all trees within the study area knowing their crown projection area from remote sensing. A general allometric equation that uses DBH as a parameter to calculate biomass and carbon per tree was used in this study. My findings show that: (1) A model that uses remote - sensing assisted landscape - scale carbon stock measurement has promise; (2) The relationship between CPA detected f rom remote sensing and allometric scaling is something that can be refined but seems to be a workable approach and these refinements would include an improved relationship model using non - linear relationships, developing a local allometric equation using d estructive sampling, and specific parameters for the savanna or tree type/species and explore the use of automated detection. This study's findings will be useful for the Senegalese government and others with savanna systems. With 1,043,000 ha of savanna systems and trees outside of forests (TOF), my findings could be an important step for integrating TOF into the natural resource management scheme for carbon stock estimation and the reduc tion of greenhouse gas emissions in the forestry sector. The use of remote sensing will lower the costs of field sampling based methods in highly patchy woodland and TOF landscapes and increase the opportunity for small holders and communities of small holders to be engaged in carbon mitigation projects. My findings show t hat, with a minimum training, they will be able to do the tree measurements in their own farms. Copyright by MOUSSA DIENG 2015 v This dissertation is dedicated to the loving memory of my mother Khady Faye who did not live to see the accomplishment that built the foundation and reap the fruits of her sacrifice. This work is most deservedly dedicated to Fatou, my "super - woman" wife, for her love, support, patience, endurance and tolerance. Finally I also dedicate this work to my two ki ds Mouhamadou and Khadidiatou, who often times vi ACKNOWLEDGMENTS AL HAMDU LI L LAH I RABBIL ALAMEEN . Thanks to the Almighty Allah for the incommensurable Graces He favor ed me and provid ed me the ability to complete successfully this dissertation . M any thanks also to my beloved Khadimu Rassul who always inspired me . A special thanks goes to my family for their sacrifice and everything they did to provide me a good educatio n. I am eternally grateful to my father Daouda Dieng and my late mom Khady Faye, my brothers Dame, Babacar, Djiby, my sister Mari è me and my aunt Sokhna Faye "Meer Faya". The completion of this work would not have been possible without support, guidance an d mentoring of my faculty advisor and the chair of my committee, Professor David L . S kole. Professor S kole gave me the opportunity of pursuing a PhD at Michigan State University and provided all resources I needed through my doctoral program. I thank so ve ry much Dr. S kole for his patience, his understanding, his direct and constant support. It was a good privilege to be part of the GOES (Global Observatory for Ecosystem Services) team and benefit from his long experience. Thank you to the GOES colleagues Jay Samek, Walter Chomentowski, Oscar Castañeda, Mike Smalligan, Dan Zelenak, Uy Duc Pham, Hartanto Sanjaya, Linda Zihan. I would also like to thank my committee members Professor Maureen McDonough, Dr. Larry Leefers and Dr. Leo Zulu for their time and a ttention for reviewing my work. Their advices, constructive and valuable suggestions for improving my thesis as well as their commitment, patience and encouragement throughout the duration of this study are deeply appreciated. Many thanks to Dr. Cheikh Mbo w, for his constant support and continuing encouragement since the beginning . vii I am also grateful to the Department of Forestry and the College of Agriculture and Natural Resources for their important support. I would like to thank Dr. Ibra Sene , El Had ji Moussa Gning and Mbaye Dione for their friendship and support, Prof. Abdou Salam Sall, Prof . Amadou Tahirou Diaw Director of LERG team and Prof . Bienvenu Sambou Director of Institut des Sciences de l'Environnement . I am also thankful to my field assist ants Ibrahima Sow and Mamadou Thiam, the drivers Abdoulaye Diouf and Pap a Tine who assisted me during my field work. Thank you to the communities in the different villages of my study area for their hospitality, time, and collaboration which made my field work successful. I deeply appreciate the support of the local government representatives especially the former Sous Pr é fet of Toubacouta district , M. Mamadou Djiba and M. Abdoulaye Ndiaye, M. Pape Seydou Dianko and M . Aly Tall respectively President of the former Rural Communities of Keur Samba Gueye, Toubacouta and Nioro Alassane Tall. I am very thank ful to M. O u mar Sonko technical officer of the Water and Forest Service who introduced me to the local communities. As i t is very difficult to acknowledge a ll those who directly or indirectly supported and helped me , I say a big thank you to those many people for the many ways in which they support me and whose names I have not been able to mention. Thank you for the Senegalese community in East Lansing for t he unforgettable and nice time we spent together. Last but not least, I am deeply grateful for and thankful to the love, patience, tolerance, sacrifice of my exceptional wife Fatou and my two kids Mouhamadou and Khadidiatou during the challenging years of my doctoral program. While being a wife and a mother, Fatou is also pursuing her PhD and is preparing her dissertation defense . Thank you for everything and all my best wishes. viii T ABLE OF C ONTENTS L IST OF TABLES ................................ ................................ ................................ ......................... xi LIST OF FIGURES ................................ ................................ ................................ ..................... xiii KEY TO ABBREVIATIONS ................................ ................................ ................................ ...... xvi CHAPTER 1 . GENERAL INTRODUCTION ................................ ................................ .............. 1 1.1. BACKGROUND AND RESEARCH CONTEXT ................................ .............................. 1 1.2. RATIONALE AND RESEARCH OBJECTIVE ................................ ................................ . 4 1.3. FOCAL POINTS FOR THIS STUDY ................................ ................................ ................. 7 1.3.1. Climate Change Mitigation Potentials for Agro - Forestry ................................ ............. 7 1.3. 2. Advantages of Agro - forestry Carbon Sequestration beyond Climate Change Mitigation ................................ ................................ ................................ ................................ 8 1.3.3. Carbon Sequestration as Part of a Value Chain or Production Strategy ....................... 9 11 1.3.5. Remote Sensing ................................ ................................ ................................ ........... 12 1.3.6. Linking Remote Sensing to Ground Measurement ................................ ..................... 14 1.3.7. Community Outreach and Engagement for Joint Measurements ................................ 15 1.4. DEFINITION OF CONCEPTS ................................ ................................ .......................... 21 1.4.1. Biomass ................................ ................................ ................................ ....................... 21 1.4.2. Cropland ................................ ................................ ................................ ...................... 22 1.4.3. Cr own Projection Area ................................ ................................ ................................ 22 1.4.4. Forest ................................ ................................ ................................ ........................... 24 1.4.5. Savanna ................................ ................................ ................................ ........................ 24 1.4.6. Trees Ou tside of Forest (TOF) ................................ ................................ .................... 24 1.4.7. Rural Community ................................ ................................ ................................ ........ 24 1.4.8. Arrondissement ................................ ................................ ................................ ............ 25 1.4.9. Sous Préfet ................................ ................................ ................................ ................... 25 1.5. GEOGRAPHY OF THE STUDY AREA ................................ ................................ .......... 25 1.6. ORGANIZATION OF THE DISSERTATION ................................ ................................ . 30 REFERENCES ................................ ................................ ................................ .......................... 31 CHAPTER 2 . MODEL DEVELOPMENT TO PREDICT TREE DBH FROM CROWN PROJECTION AREA USING REMOTE SENSING ................................ ................................ .. 36 2.1. INTRODUCTION ................................ ................................ ................................ .............. 36 2.2. MATERIALS AND METHODS ................................ ................................ ....................... 40 2.2.1. The Study Sites ................................ ................................ ................................ ............ 40 2.2.2. Field and remote sensing data collected ................................ ................................ ...... 42 2.2.3. Method ................................ ................................ ................................ ......................... 47 2.2.4. Stratification of the study area ................................ ................................ ..................... 51 2.2.5. Selection of sampled plots ................................ ................................ ........................... 53 2.2.6. Data analysis ................................ ................................ ................................ ................ 58 ix 2.2.6.1. Manual delineation of trees .............. .................................................................... 59 2.2.6.2. Correlation Analysis ............................................................................................ 59 2.2.6.3. Regression Analysis and validation of the models ........................................................... 59 2.3. RESULTS ................................ ................................ ................................ ........................... 60 2.3.1. Relationship between field measured DBH and field measured CPA ........................ 61 2.3.1.1. Model development..............................................................................................61 2.3.1.2. Model validation......................................................................................... ..........6 3 2.3.2. Relationship between field measured CPA and remote sensing measured CPA ........ 64 2.3.3. A Model to predict DBH from remote sensing CPA ................................ ................... 67 2.3.3.1.Model development.............................................................................................. . 67 2.3.3.2. Results comparison with CBP study in Western Kenya.......................................73 2.3.4. Regression model for Remote sensing CPA and field measured DBH ....................... 74 2.4. DISCUSSIONS AND CONCLUSION ................................ ................................ .............. 76 APPENDIX ................................ ................................ ................................ ............................... 81 REFERENCES ................................ ................................ ................................ .......................... 85 CHAPTER 3 . THE ROLE OF REMOTE SENSING IN MEASURING BIOMASS AND CARBON AT THE LANDSCAPE LEVEL IN SAVANNA SYSTEMS ................................ .... 88 3.1. INTRODUCTION ................................ ................................ ................................ .............. 88 3.2. MATERIALS AND METHODS ................................ ................................ ....................... 92 3.2.1. Study Area ................................ ................................ ................................ ................... 92 3.2.2. Data used ................................ ................................ ................................ ..................... 93 3.2.3. Data Analysis ................................ ................................ ................................ ............... 94 3.2.3.1. Remote Sensing Analysis and manual delineation of trees..................................94 3.2.3.2. Regression Analysis..............................................................................................95 3.3. RESULTS ................................ ................................ ................................ ........................... 96 3.3.1. Tree density at landscape level ................................ ................................ .................... 96 3.3.2. Carbon stock and carbon density at landscape level ................................ ................... 97 3.3.3. Relat ive contribution of trees by DBH size to the total carbon ................................ ... 99 3.3.4. Comparison of remote sensing - based biomass and carbon estimation with field data ................................ ................................ ................................ ................................ ............. 102 3.4. DISCUSSIONS AND CONCLUSION : ................................ ................................ ........... 105 REFERENCES ................................ ................................ ................................ ........................ 107 CHAPTER 4 . LANDSCAPE BIOMASS ESTIMATION AND CARBON MAPPING .......... 110 4.1. INTRODUCTION ................................ ................................ ................................ ............ 110 4.2. MATERIALS AND METHODS ................................ ................................ ..................... 113 4.2.1. Study Area ................................ ................................ ................................ ................. 113 4.2.2. Data used ................................ ................................ ................................ ................... 113 4.3. RESULTS ................................ ................................ ................................ ......................... 113 4.3.1. Landscape application of the model ................................ ................................ .......... 113 4.3.2. Carbon density per block in the study area ................................ ................................ 115 4.3.3. Landscape carbon mapping of TOF ................................ ................................ .......... 116 4.3.4. Comparison with other studies ................................ ................................ .................. 124 4.4. DISCUSSIONS AND CONCLUSION ................................ ................................ ............ 127 AP P ENDICES....................................... ...................................................................................129 x Appendix B: TECHNICAL CONTRIBUTIONS THAT CAN BE MADE BY COMMUNITY PARTICIPATION: TOWARD COMMUNITY BASED CARBON MEASUREMENTS ... 130 REFERENCES.. ................................................... .. ...................................................................... 155 Appendix C: EVALUATION OF FARMERS 'TRAINING WORKSHOP AND FIELD WORK EXPERIENCE ................................ ................................ ................................ ........... 157 REFERENCES ................................ ................................ ................................ ........................ 160 CHAPTER 5 . GENERAL CONCLUSION ................................ ................................ ............... 162 5.1. SUMMARIES AND RESEARCH QUESTION REVISED ................................ ............ 162 5.2. LIMITS AND RECOMMENDATIONS FOR FURTHER RESEARCH ....................... 164 5 .3. IMPLICATIONS AND THE PLACE OF THIS STUDY IN THE GLOBAL CLIMATE CHANGE AND SUSTAINABLE DEVELOPMENT AGENDAS ................................ ........ 165 xi LIST OF TABLES Table 2.1 Variables measured in th e field and equipment used ................................ ................... 43 Table 2.2 Spectral, spatial and temporal characteristics of Worldview 2 ................................ ..... 47 Table 2.3 Stratification for Sokone zone ................................ ................................ ...................... 53 Table 2.4 Stratification for Karang zone ................................ ................................ ....................... 53 Table 2.5 Number of sampled plots for each stratum in the Sokone site ................................ ..... 56 Table 2.6 Number of plots for each stratum in the Karang site ................................ .................... 56 Table 2.7 Number of plots and their density for cashew plantations ................................ ............ 58 Table 2.8 Rule of Thumb for interpreting the size of a correlation coefficient ............................ 61 Table 2.9 Regression model and validation statistics for the relationship between Remote sensing CPA and field measured DBH for all sampled trees in Sokone site ................................ ............. 75 Table 2.10 Regression model and validation statistics for the relationship of Remote sensing CPA and field measured DBH for all sampled trees in Karang site ................................ ............. 75 Table 3.1 Tree density by block in the study area ................................ ................................ ........ 97 Table 3.2 Carbon stock and Carbon density by block in the study area ................................ ....... 98 Table 3.3 Relative contribution of trees (tons of C) by DBH size to the total carbon in the Sokone site. ................................ ................................ ................................ ................................ ................ 99 Table 3.4 Relative contribution of trees (tons of C) by DBH size to the total carbon in the Karang site. ................................ ................................ ................................ ................................ .............. 101 Table 3.5 Comparison between remote sensing b ased and field based carbon stock and carbon densities ................................ ................................ ................................ ................................ ....... 104 Table 3.6 Carbon stock and carbon densities in the six blocks of the Sokone and Karang sites where sampled plots were located ................................ ................................ .............................. 105 Table 4.1 Remote sensing stratification of the eighteen blocks in Sokone and Karang sites ..... 114 T able B.1. Demographic characteristics of selected villages ................ .................................... 134 Table B.2. Comparison between communities' biomass estimates and researcher's estimates .. 151 xii Table B.3.Willingness and conditions for communities to participate in carbon measurement . 152 xiii LIST OF FIGURES Figure 1.1 Comparison between Landsat 5 and Worldview Image taken the same day (April 16, 2010) in the Sokone study area ................................ ................................ ................................ ....... 6 Figure 1.2 Above and be low ground biomass of a tree ................................ ................................ 22 Figure 1.3 Crown perimeter and crown projection area of a tree (adapted from Gschwantner et. al , 2009) ................................ ................................ ................................ ................................ ........ 23 Figure 1.4 Location of the study area ................................ ................................ ........................... 26 Figure 1.5 Selected photos of the study area ................................ ................................ ................ 27 Figure 1.6 Changes in land use and their environmental and economic consequences ............... 29 Figure 2.1 Area of cashew production i n Senegal and the study area ................................ .......... 41 Figure 2.2 Selected blocks in the Sokone (a) and Karang (b) sites from the Worldview 2 satellite image in 432 band combination ................................ ................................ ................................ .... 45 Figure 2.3 Flow diagram of biomass estimation and carbon mapping method ............................ 49 Figure 2.4 Block SOK 7 and the different strata ................................ ................................ .......... 52 Figure 2.5 Ground sample frame ................................ ................................ ................................ .. 55 Figure 2.6 Example of the l ayout of sampled plots and measured trees in the Sokone site ......... 57 Figure 2.7 Correlation between field measured DBH and field measured CPA for sampled trees of all types (N= 377) in the Sokone site ................................ ................................ ....................... 62 Figure 2.8 Scatter plot of model validation measured vs. predicted DBH ................................ ... 64 Figure 2.9 Relationship between Remote sensing and field measured CPA of all sampled trees (N=377) in the Sokone site ................................ ................................ ................................ ........... 65 Figure 2.10 Relationship between Remote sensing and field measured CPA for all sampled trees (N=315) in Karang ................................ ................................ ................................ ........................ 66 Figure 2.11 Relationship between Remote sensing and field measured CPA in both sites (N=692) ................................ ................................ ................................ ................................ ....................... 67 Figure 2.12 Model - based DBH using Remote sensing CPA i n Sokone (Linear model) .............. 68 xiv Figure 2.13 Model - based DBH from Remote sensing CPA for all sampled trees in the Sokone site (Non - linear model) ................................ ................................ ................................ ................ 69 Figure 2.14 Model - based DBH from Remote sensing CPA for all sampled trees in the Karang site (linear model) ................................ ................................ ................................ ......................... 70 Figure 2.15 Model - based DBH from Remote sensing CPA for all sampled trees in the Karang site (Non - linear model) ................................ ................................ ................................ ................. 71 Figure 2.16 Relationship between field based DBH and Remote sensing based CPA of all sampled trees in both Sokone and Karang sites (linear model) ................................ .................... 72 Figure 2.17 Relationship between field based DBH and Remote sensing based CPA of all sampled trees in both Sokone and Karang sites (Non - linear model) ................................ ............ 72 Figure 2.18 Relationship between DBH and Crown area in Western Kenya for 855 trees (ICRAF, 2010) ................................ ................................ ................................ ................................ ............. 74 Figure 2.19 The effects of tree disturbances on tree crown shape ................................ ................ 78 Figure A.1. Automated detection in the Sokone landscape ......................................................... 82 Figure A.2. View of automated detection in cluster vegetation .................................................. 83 Figure A.3. View of automated detection in individual big trees ................................................ 84 Figure 3.1 Percentage contribution to total carbon by DBH classes in the Sokone site ............. 100 Figure 3.2 Percentage contribution to total carbon by DBH classes in the Karang site ............. 102 Figure 4.1 Carbon density (t C/ha) per block in the Sokone site ................................ ................ 115 Figure 4.2 Carbon density (t C/ha) per block in the Karang site ................................ ................ 116 Figure 4.3 Carbon in the Sokone site ................................ ................................ .......................... 117 Figure 4.4 Carbon map for block 1 (SOK 1) in the Sokone site ................................ ................. 118 Figure 4.5 Carbon map for Block 7 (SOK 7) in the Sokone site ................................ ............... 119 Figure 4.6 Carbon stock in the Karang site ................................ ................................ ................. 120 Figure 4.7 A portion of a cashew plantation carbon map in the Karang site .............................. 121 Figure 4.8 Carbon map of individual TOF in agricultural land in block 9 (KAR 9) in the Karang site ................................ ................................ ................................ ................................ ............... 122 xv Figure 4.9 Carbon map in a low density block (KAR 2) in the Karang site ............................... 123 Figure 4.10 Carbon map in a low density block (KAR 9) in the Karang site ............................. 124 Figure 4.11 Carbon map in the Sokone site based on Bacc ini et al. (2012) ............................... 126 Figure 4.12 Carbon map in the Sokone site based on our model ................................ ............... 127 Figure B.1. Research activities and their objectives ....................... ........................................... 136 Figure B.2. Village meetings ...................................................................................................... 137 Figure B.3. Training workshop chaired by the State Representative (Sous - Pré fet) ..................1 38 Figure B.4. Field crew in the field measuring trees and recording data .................................... 140 Figure B.5. Characteristics of the participants based on their ethnic groups ............................. 142 Figure B .6. Age strata of participants to the training workshop ................................................ 142 Figure B.7 Percentage of participants based on gender ............................................................. 143 Figure B.8. Level of Educatio n .................................................................................................. 144 Figure B.9. Self - reported knowledge of farmers on their ability to perform tree measurements ........................................................... .......................................................................................... 145 Figure B.10. Participants rating on how much they do think they can locate a plot ................ .. 146 Figure B.11. Participants rating on how much they do think they can identify trees .............. ... 146 Figure B.12. Participants rating on how much they do think they can measure tree DBH ...... .. 147 Figure B.13. Assessment of communities' experience in tree measurement ............................ ,, 149 Figure B .14. Percentage of participants finishing task in a timely manner .............................. .. 150 xvi KEY TO ABBREVIATIONS AGB Above Ground Biomass AF Ag roforestry A/R Afforestation/ Reforestation CCX Chicago Climate Exchange. CDM Clean De velopment Mechanism C2M Carbon2Markets CPA Crown Projection Area DAC D iameter at 20 cm A bove the root C ollar DBH Diameter at b reast h eight FAO Food and Agriculture Organization FM DBH F ield m easured d iameter at b reast h eight GIS G eographic Information S ystem GPS Global P ositioning S ystem GOES Global Observatory for Ecosystem Services HCS Hypersp herical Color Space ICRAF World Agroforestry Center IPCC Intergovernmental Panel on Climate Change KTGAL Kyoto: Think Global, Act Local L DP Local Development Plan MRV M easureme nt, R eporting and V erification NFI National Forest Inventory xvii PASA Projet Agroforestier Senegal o - Allemand REDD Reducing Emissions from Deforestation and forest Degradation RS CPA Remote Sensing Crown Proj ection Area SPSS Statistical Package for the Social Sciences TOF Trees Outside of Forests UNDP United Nations Development Programme UNFCCC United Nations Framework Convention on Climate Change USDA U.S. Department of Agriculture UTM Universal Tr ansverse Mercator VCS Voluntary Carbon Standard WGS World Geodetic System 1 CHAPTER 1 . GENERAL INTRODUCTION 1.1. B ACKGROUND AND RESEARCH CONTEXT Climate change has been an emerging global environmental problem and policy concern, and is now a key economic dev elopment challenge in developing countries in Sub - Saharan Africa (Mbow et al. 2012 ; Collier et al ., 2008) whose consequences include the adverse effects on livelihoods of farmers . Climate change is the result of increased levels of greenhouse gases in the atmosphere caused in part by fossil fuel combustion and land conversion, the latter is due primarily to tropical deforestation. The trend to - date has been toward significant greenhouse gas emissions with few land based sinks . Trends in land use change hav e been predominantly conversion of high biomass ecosystems to low biomass agricultural systems that have significantly lower carbon stocks. The emissions contribution of land - use conversion is estimated to be 10 to 30% of the current total anthropogenic em issions of CO 2 (IPCC, 2001 Houghton et al ., 1985). In Senegal, the current terrestrial carbon stock of the landscape is affected by both land use change and land degradation. There has been a loss of 292 MtC from 1965 to 2000 (Woomer et al. , 2004). One est imate for southern Senegal indi cates that from 1900 to 2000 there has been a landscape - wide decline of carbon stocks in vegetation and soils of 37% (Liu et al ., 2004). One interesting characteristic of the land use component of the carbon cycle is that l and management can reduce emissions and remove carbon from the atmosphere. This dual source - sink characteristic of land use change makes it an important focal point for land management interventions in ways that are not available with energy strategies. Th us, there has been a growing interest in developing modalities for creating carbon sinks in both forest and non forest landscapes, which include, among other forms of intervention: reforestation, 2 afforestation, assisted natural regeneration, plantation man agement, and agro - forestry and other uses of trees outside of forests. As one of these forms of intervention, agro - forestry combines a multitude of environmental, social and economic benefits to rural land owners in addition to carbon sequestration. The u se of woody perennials on farms in rural Africa can provide high - value products and increased incomes over traditional annual crop production systems. However, it is not widely adopted because the delayed benefits obstacle of agro - forestry may be too signi ficant for farmers to adopt these systems by replacing existing annual cropping systems 1 . For example, t he long period (5 - 8 years) to full tree maturity and production may be too long for farmers to feel secure in removing from production cropland that is already producing annual revenues, even if their existing yields and economic returns are meager. As carbon markets have been growing, there has been an increasing emphasis on activities that enable carbon sequestration in tree systems, mostly with large - s cale commercial plantation systems. Smallholder agro - forestry presents another option for carbon 1 The specific role for carbon payments may not lie in its direct sale value, but as a trigger or catalyst that removes the barriers to adoption. Key among these barriers is the so - called systems usually have no - forestry. Since an agro - forestry system takes years before it begins to produce, farmers are often reluctant to make this long term investment in tree establishment on annual cropland. For exa mple, it might take three to six years for an agro - forestry system before fully realizing benefits compared to the only few months needed to harvest annual crops (Franzel and Scherr, 2002). Studies in the GOES lab at MSU suggest that while carbon income mi ght only be 15 - 20% of total farm product income , that is enough to push some farmers into adoption, if the carbon credits can be accrued from the initial planting year. Similarly, those farmers that do engage in adoption of agro - forestry often finance thei r investment with loans, which immediately places an economic exposure or financial risk on the household. Conceivably carbon finance can be used in the place of capital finance with considerably less (albeit different) risk. Dewees (1995) also suggests th Also, some studies have shown that the time delay between the planting and the harvesting of trees put farmers in rural areas in a position where they are less likely to adopt agroforestry ( Scherr , 1995) . 3 sequestration that can also provide environmental co - benefits (increased soil fertility and lower erosion rates) and at the same time a potential source of inc ome for poor smallholder s through carbon sequestration credits. The model developed under the Michigan State University program, Carbon2Markets, provides a - sequestration for smallholder s who adop t agro - forestry, and can alleviate the delayed benefits problem by providing early value - chain income from carbon sequestration. Thus, carbon sequestration in agro - forestry and other Trees O utside of F orests (TOF) has the potential to provide increased inc omes and provide environmental and agronomic co - benefits such as climate change mitigation . The challenge, however, is that before agro - forestry carbon value chains can be established there must be rigorous and accurate protocols for measuring, reporting and verifying carbon sequestration in agro - forestry systems and other TOF . These protocols and methods must be acceptable to the international climate change mitigation community and at ntation and participation. International standards for carbon measurement are being developed and promoted through international conventions, carbon financial markets, and both regulatory and voluntary regimes and institutions. As such, scientific and te chnical rigor for international carbon measurement is extremely high, and involves exact reporting and verification standards and best practices. This is particularly true for new modalities of carbon measurement. Currently the common practice for carbon measurement in tree - based sequestration systems is based in large - scale plantations or reforestation in large tracts of closed canopy forests. There is l ess c redibility 4 placed in small - scale systems involving agroforestry or small - scale refores tation systems that are managed by local land owners, particularly in savanna and woodland landscapes. 1.2. R ATIONALE AND RESEARCH OBJECTIVE Regional and global monitoring of forest cover changes and carbon content using remote sensing has advanced sign ificantly for closed humid tropical forests, and several methods have been developed. At the meantime, f or semi - arid savannas and woodlands, monitoring of cover and carbon has been under - studied, and more methods are needed. Further, monitoring of cover an d carbon for systems of Trees Outside of Forests in semi - arid regions has not advanced at all, and new methods are needed. Also, m ost remote sensing methods for monitoring cover and carbon stocks have used moderate resolution optical sensors such as Lands at. Most of these methods have been either statistical classification or end - member analyses based on spectral characteristics. Consequently, t hese approaches are inaccurate or inappropriate in landscapes with sparse tree cover; most of the land looks bare to optical spectral signature models. Therefore, w e need a new methodology to monitor Woodlands and TOF. emissions and removals: Measurements of changes in forest or tree cove r over time, deforestation, reforestation, degradation (also known as Activity Data in IPCC) Measurements of stocks and stock changes in carbon content of live biomass (also known as Emission Factors) 5 Where convention in IPCC is stated as: ER = AD * E F These measurements lay the foundation for nearly all carbon accounting of emissions and removals associated projects, interventions, investments, and policy actions . They are also fundamental for both s equestration activities and REDD activities, in bas elines and reference emission levels, whether project based or national in scope. There are o pportunities on better investigating this area of research because n ew technologies are emerging that allow very high spatial resolution earth observations : for example, 10 years ago , remote sensors had a 1 - 2 meter s resolution compared to the .46 cm resolution today. This may create opportunities for monitoring landscapes in which tree cover is sparse or widely spaced and whe re individual trees, as object s , can b e detected. If methods exist, remote sensing allows for large area monitoring at the landscape scale . This could r educ e costs of field sampling based methods in highly patchy woodland and TOF landscapes and increase in opportunity for small holders and c ommunities of small holders to be engaged in carbon mitigation projects. 6 Figure 1 . 1 Comparison between Landsat 5 and Worldview Image taken the same day (April 16, 2010) in the Sokone study area In t his r esearch , I aim at addressing these problems of small - scale TOF systems as potentially important sequestration systems, and the necessary measurement requirements for enabling these systems in the international system of carbon management and regulation. T he major question posed to frame such research is : - Can protocols be developed that allow for landscape scale measurements of trees outside of forests with the accuracy and rigor that is acceptable to the international community? This research ex amines this question through an analysis of potential landscape level carbon measurement, reporting and verification (MRV) methods in rural Senegal . And if we want to sustain these methods, local communities have a role to play. Therefore, I will assess the abil ity of those communities to accomplish tree carbon measurements on their 7 own farms , which is different from community participation in its richest and fullest extent, a full treatise beyond the rather narrow measurement scope of this study . The focus of t his study specifically is to develop methods for remote sensing - assisted carbon measurement and monitoring in semi - arid landscapes of Trees Outside of Forests (TOF) which include: Plantation and Agroforestry systems Other trees on farms: planted trees Othe r trees on farms: remnant savanna trees The method us es a combination of advanced remote sensing and basic allometry in - arid savanna ecosystems of West Africa , especially in Senegal ese semi arid woodland a nd agricultural landscapes . While doing so, I will assess the local communities' skills on mak ing accurate ground measurements of critical tree parameters, such as tree stem diameter, tree height, and tree canopy projected area to be used to compute carbon stock and stock changes , and thereby be active participants in, and co - owners of, the MRV process. 1. 3 . F OCAL POINTS FOR THIS STUDY 1. 3 .1. Climate Change Mitigation Potentials for Agro - Forestry To mitigate climate change , massive forest protection and re forestation have been promoted as one way to stabilize the concentration of carbon dioxide in the atmosphere through A fforestation /R eforestation (A/R) activities in the C lean Development M echanism (CDM) or REDD+ activities. However, while these efforts to reduce large scale deforestation or create large - scale reforestation projects provide new land uses for 8 significant climate mitigation (FAO, 2010) , they also can conflict with the growing need for expansion of the agricultural land base (Mbow, 2012) . Small holder forest agroforestry on the other hand could be a compromise solution to the apparent land use conflict between forest conservation, reforestation and agricultural land use, because it increases the storage of carbon and at the same time provide oppo rtunities for agriculture. Moreover, t he agroforestry option may in fact enhance agricultural production (Unruh, 1993). Currently, there is a growing interest in investing in agroforestry systems for this dual benefits of climate mitigation and livelihoo ds enhancements (ICRAF, 2008) and also as adapt to, adverse impacts of changing climate conditions (Verchot, 2007) . In this context it is possible to view agrofores try as both a mitigation and an adaptation option for smallholder s. Agroforestry practices can be implemented in a variety of forms and range of intensities depending on local conditions to provide a locally relevant approach to sustainable agriculture pro duction and soil fertility management. Agroforestry might become and non - food subsistence needs through the ecosystem s ervices it provide s (Jonsson, 1999) . There is also some evidence suggest ing that the high production levels and economic values of agroforestry value chains may actually facilitate the production of financial capita l beyond subsistence levels alone, thereby aiding in capital accumulation and re - investment at the farm level ( Jonsson, 1999 ) . 1. 3 .2. Advantages of Agro - forestry Carbon Sequestration b eyond Climate Change Mitigation In Africa a critical characteristic of the rural landscape is the preponderance of many smallholder farming systems that are operating on the environmental and economic margins, 9 where any newly introduced practice will be adopted only if it improves food security as well as generate s income on a sustainable basis. Under such conditions, climate mitigation interventions, such as increased carbon stocking with trees on farms will need to not only contribute to improve food production but also generate income (Smith, 2012 ) . In this context, clim ate mitigation becomes a co - benefit of a primarily food - focused enterprise. Hence, there are practices that are keys to sustaining agricultural productivity and they include those that minimize the rate of soil degradation, improve soil fertility, increase crop . I n turn , this will provide better conditions for climate adaptation in many of the poor rural communities that might be most adversely affected by climate change and variabi lity (Neupane, 2001) , and then also provide added benefits for climate mitigation through increased greenhouse gas removals from the atmosphere. Agroforestry readily bundles both mitigation and adaptation str ategies and provides several pathways to a range of environmental and social co - benefits and outcomes, including food security, increased farm income, restoration and maintenance of above - ground and below - ground biomass and biodiversity, reestablishment o f biological corridors between protected forests, maintenance of watershed hydrology, improved soil conservation, availability of timber and fuel wood, and ultimately a reduction of pressure on natural forests outside the farming domain (Pandey, 2002) . 1.3 .3. Carbon Sequestration as Part of a Value Chain or Production Strategy Agrofor estry is a promising option to sustain agricultural productivity and livelihoods of farmers (Syampungani et al ., 2010), and a low - cost method to sequester carbon. In the context of emerging carbon markets and ecosystem payment schemes, and 10 where extreme po verty is a crucial problem in many parts of tropical countries, especially in rural areas, carbon benefits of specific farming systems may be a means to encourage and incentivize sustainable agriculture and natural resource management systems. Some studies have shown the value of increasing tree cover through tree planting by households as a solution to the problem of deforestation (Carle et al ., 2002; Medugu, et al ., 2010) . In a country like Senegal, where farmers represent more than 70% of the population and depend on land to develop their activities, agro - forestry presents a real opportunity with potential pay offs for livelihoods, land productivity and environmental integrity. It has also the added benefit of creating options at the household and communi ty level for climate change mitigation and adaptation. Farmers ordinarily don't include, nor often even comprehend, the carbon dimension of their farming system when considering or engaging new land management practices such as agro - forestry. For example , evidence from the Nhambita Project in Mozambique has Campesina Africa, 2012 from carbon, they do not und erstand what carbon is ( Campesina Africa, 2012). Rather, farmers focus more on the direct economic return of the new alternative and the benefit of improving the land management system; altogether contributing to increase annual crop productivity, sustain ability and cost effectiveness. One challenge is to determine how and if, carbon can be another commodity value chain that provides economic returns to One way of assessing the value of car bon sequestration is to measure carbon using more accurate techniques so that it can be sold and benefit those people whose activities 11 contribute to the sequestration. Also, engaging the communities in carbon measurement activities could lower the cost of carbon sequestration and increase the benefits. It is also important to investigate the role of carbon in agroforestry adoption. 1. 3 .4. A logic model has been developed and brie fly laid out below : 1. In land - scarce areas, a groforestry can provide an alternative to dedicatin g land solely to reforestation to reduce pressure on forests; 2. A groforestry as a carbon sequestration system and a product ion system might provide a means f or both climate change mitigation and an enhanced agricultural production ; 3. One major obstacle i s that d elayed benefits prevent widespread adoption of agroforestry; 4. Carbon value chains may offer a way to overcome these barriers to AF adoption. Howeve r, before agroforestry as a carbon sequestration strategy and value chain will be necessary, it require s that its carbon be quantified and measured accurately , be reportable, and verifiable. Many accepted methods for carbon measurement in programs or proj ects that are based on sequestration (as opposed to avoiding emissions from deforestation) have been defined at a project scale with specified boundaries of the project, often on land parcels with a single owner. When trying to develop methods for rural po or farmers, experience suggests that a landscape approach is needed ( Estrada - Carmona et al ., 2014) , whereby many farmers can be bundled into a single project, even if the parcels are not contiguous. The landscape approach offers new opportunities for a g reater participation of the rural poor in carbon markets . However, the means for landscape - scale measurement need to be developed. Remote sensing offers a means to do landscape scale measurement and 12 monitoring, but most advance d methods in recent years hav e focused on closed forests, rather than open woodlands and savannas or, to be sure, systems of trees outside of forests. Thus, th e current study focuses on the need to develop rural community - focused, landscape - wide measurement methods for systems of tre es outside of forests (cf. agro - forestry and related systems) in the savannas of Africa. This measurement and monitoring approach provides a bundle of methods (landscape wide, trees outside of forests, community - scale). 1. 3 .5. Remote Sensing In recent yea rs there has been substantial progress by the research community in developing ways to detect land cover change in tropical forests with remote sensing analysis . I nitially the focus was on measuring the conversion of tropical forests to non - forest land (Sk ole and Tucker 1993) . However, recent advances have made it possible to increase the variety of disturbances that can be detected for closed tropical forests to include deforestation, degradation, logging, fragmentation, reforestation, and fire (Matricardi et al. 2013, Matricardi et al. 2010; Matricardi et al. 2007; Wang and Cochrane 2005). Thus, there are now methods available to remotely detect a full range of disturbance intensities, from outright clearing to low levels of degradation, over large areas. But these approaches have been applied only in very dense closed tropical forests, and in particular in key regions of the Amazon Basin, Central Africa, and South East Asia. There has been very little advancement of detection methods for two important ot her forms of land cover change: 1) deforestation and degradation of open woodlands such as the cerrado and the chaco ecosystems in South America, woodlands of East Africa, Savannas of West Africa, and other open forest ecosystems in the tropics and sub tro pics, and 2) 13 regeneration systems on managed landscapes where biomass recovery occurs as plantations, orchards, agroforestry, and widely - spaced tree complexes associated with agriculture. Some studies have suggested that disturbance in open forest syst ems is quantitatively as important as in closed forests systems (Erickson et al. 2002; Murphy and Lugo 1995; Serca et al. 1998), especially because disturbances in these systems may have longer term degradation effects with lower probabilities of recovery than in closed forest systems. Observations of reforestation and biomass accumulation in trees on agricultural land is important because these data are needed to understand the global magnitude and capacity for carbon sequestration, and to inform decision makers and policy makers on options for carbon management practices that can remove carbon from the global atmosphere. There is also considerable uncertainty on the current land area in woody perennials on farms in developing countries and the global pote ntial for managing carbon sequestration in tree - based agriculture. Some estimates from international organizations suggest a large amount of carbon sequestration already occurring in these managed landscapes (Verchot and Singh 2009). Although frequently t reated as a secondary priority to monitoring of the much higher biomass forests of the humid tropics, the dry and open forests of the world are more abundant than closed forests and usually are more prone to occupation and disturbance by humans. Although t hese low density forests contain as little as 25% of the total carbon stock of humid tropical closed forests, the global area is more than 30% greater and rates of disturbance are suspected to be equal to or greater than the closed forests, and less likel y to recover lost carbon. Thus , it is vitally important to begin to assess the global magnitude of open forest disturbances. 14 What is often unrecognized is that while forested area is declining in developing countries, tree cover on farms is increasing, as farmers substitute annual cropland for the tree products which were formerly available in local forests. F armers are also increasingly seizing specific market opportunities to sell higher - value tree products (e.g. natural rubber, bio - fuels, bio - chemicals, timber). For example, remote sensing in 64 rural locations in Uganda revealed that between 1960 and 1995, forested area declined 50%, agricultural area increased 23%, and the proportion of agricultural land under tree cover increased 22% (Place 2001). Agr icultural land now accounts for over double the area of forested land in Increasing woody biomass on farms in developing countries is seen as a possible global climate and carbon mitigation option that deserves serious attention. Montagnini and Nair (2004) have estimated that a vigorous program to introduce agro - forestry on farms in tropical Africa and Asia has the potential to sequester 3.5 Mg C ha - 1 yr - 1 . Howe ver, for this to be a successful strategy it will be necessary to have sound detection and monitoring systems in place. 1. 3 .6. Linking Remote Sensing to Ground Measurement A significant challenge for efficient, timely and cost - effective automated monitor ing of A/R/AF is the accurate measurement of woody biomass in the landscape. Detecting and measuring individual tree objects that are sparsely planted on farms, along roadways, or in backyards cannot be done accurately with low or moderate resolution sate llite imagery where a single tree may represent only a small proportion of a single pixel. Although pixel unmixing techniques (Keshava 2003) can provide an estimate of the proportion of different land cover types represented in a single pixel, in general, these techniques are not amenable 15 for accurate estimation of tree size, number of trees or other measurements required for accurate assessment forest cover change in sparsely planted landscapes. Also they do not work well for savanna and TOF landsc a pes. Ho wever, recent technological advancements in satellite image acquisition provide access to hyper - resolution imagery o f the Earth and its land uses. For instance, t he Worldview 2 /3 satellite acquires 4 bands of multispectral data at 2.4m resolution and a pan chromatic band at 60 centimeters or finer . Other sensors now can detect resolutions of objects at 0.5 m in size. Very high - resolution imagery enables the detection of objects in the landscape using remote sensing. Subsequent processing can be used to asso ciate attributes, such as size or texture, with detected objects. Therefore, it may be possible to count trees, estimate size and evaluate change remotely , which is a focal point for this study. If one can detect individual trees and tree crowns, there i s a potential to develop a crown - based allometry for individual trees. This would require the development of methods for ground measurements that could support the remote sensing analysis . Most reforestation projects use small sample plots in a sampling fr ame for forest inventory. This ground measurement protocol must be re - defined for systems of individually planted or growing trees, and then also extended to the woodland or savanna landscape. The landscape approach allows one to inventory both land in for ests (in this case savanna) and land with trees outside of forests (in this case agroforestry on farms). 1. 3 .7. Community Outreach and Engagement for Joint Measurements Although there is considerable concern and interest in involv ing local communities in the various international carbon programs, there is also some uncertainty over whether they can participate in the basic measurement process ( Evans and Guariguata, 2008 ). This 16 has made it difficult to include smallholder systems in the international agend a, and thus forgoes the potential of systems like agroforestry from being a low - cost and local system for carbon and climate mitigation. However, work in Southeast Asia by Samek et al . ( 201 4 ) shows that the measurement process could readily engage local communities, at least as partners who are trained to implement the standard forest inventory measurements on the ground. Ground measurements are expensive and time consuming, especially when done by external project developers or verifiers, and it is not l ikely that local community members can readily participate in , and share the benefits of a project that relies on outsiders doing the measurements of their own lands. The first step in the process of claiming ownership of the benefits of carbon sequestrat ion begins with measurements and reporting. So it is important that community - based abilities must contribute to measurements. Yet, c ommunity level understanding of carbon as a service or commodity, and their willingness to participate can be difficult. Ob stacles, such as land and labor availability may hinder willing participant s , according to Tschakert ( 2004 ) . However at the most fundamental and basic level, there is a need to know how well collaborative measurements with the project specialists and the l ocal people could produce accurate measures - a simple but important issue is whether or not traditional and local communities can make accurate measurements, that can withstand the scrutiny of international protocols, and thereby enable the deployment of the Carbon2Markets model. The G lobal O bservatory for E cosystem S ervice s (GOES) l ab oratory at Michigan State University has been working with communities in Africa, South Asia and Southeast Asia on joint measurement campaigns for assessing carbon stocks i n smallholder plantation 17 and community forest systems for more than five years, mostly in small locally - owned stands of teak ( Tectona grandis ), but also with native species in community forests (Samek et al. ( 2011 ) , Samek et al . ( 2013). In early work in Th ai communities of the Inpang Network, a sufficiency economy organization in Northeast Thailand ( Krongkaew 2003; Chalapati 2008), this project produced a new protocol that engaged collaborative teams of local people and university specialists from both Thai and American universities. This protocol was accepted by the Chicago Climate Exchange ( CCX, 200 9 ) . There are a number of other existing C mitigation projects that include smallholders and agroforestry in developing countries, e.g., the Plan Vivo C mitig ation projects in Tanzania, Mexico, Mozambique, and Uganda (Plan 2010). These projects, however, include other forest C mitigation components than agro - forestry, such as afforestation/reforestation, forest conservation, and avoided deforestation the typi cal complement of REDD+ standard is a set of best practices to ensure that a forest C mitigation project provides equitable distribution of benefits, ensures livelihood n eeds are met, includes local people in the development and management of the project, and supports biodiversity and environmental services (Ruiz - De - Oña - Plaza et al. 2011). Plan Vivo projects emphasize capacity - building, long term C benefits, diversifying l ivelihoods, and protecting biodiversity (Plan Vivo, 2010). The Inpang Carbon Bank activities also follow the same principles. Two main differences between the GOES approach to agroforestry C offsets and the Plan Vivo standard are the use of an Internet - ena bled content management application, which uses GIS and remote sensing data analysis, and a focus on develop ing a new C hicago Climate Exchange (C CX ) market approved agroforestry C offset protocol. 18 In order for agroforestry C to be developed as a commodity it must, however, be economically viable; thus most forestry projects have been large - scale enterprises. For smallholder s there are market, institutional, and social barriers that all yield costs. The GOES Lab's work has focused on addressing the social b arriers through capacity building and training local people in the Inpang Network, the institutional barriers through partnering with provincial and national organizations representing the national commitments to the UNFCCC, and the market barriers through technological innovations in managing C offset projects. One piece of t his dissertation, centered in Senegal , focus es only on a narrow slice of the first, social barrier in terms of accuracy assessment. The Carbon2Markets internet - based C offset manageme nt system is truly in a software R&D phase. The proposed advantage of the Carbon2Markets system for smallholder C offsets is in lowering the transaction costs associated with field - level measurements and verification. The fully developed on - line C manageme nt system will - line tools that will register the field boundaries, calculate the baseline C, estimate leakage from on - farm management practices and report future amounts of C sequestrat ion ( ex ante projections). The on - line tools require input data uploaded to the database. Field boundar ies are either delineated using a hand held GPS receiver, uploaded via the Internet, entered as coordinates on - line, or drawn on - screen using hyper - reso lution satellite imagery (1 m or less) and Internet - GIS tools. An important question related to this approach is " Can smallholder agroforestry C offset projects be scaled up? " With an on - line Internet - GIS management application, such as Carbon2Markets, an d the potential to use satellite remote sensing data to directly measure 19 and monitor C sequestration in biomass, the proposition is that it is feasible. Aggregating individual farm parcels is a key function of the Carbon2Markets software application that a llows scaling and lowers scale - up costs. The smallholder teak plantation studies in Thailand demonstrate how farms spread across five provinces can be managed as a single project and scaled up to national levels. Ground - based measurements are essential com ponents of this method and will be a requirement for project validation and verification against rigorous international standards for all projects in the foreseeable future. The coupling of satellite remote sensing measurements with ground measurements of C in biomass means larger areas can be assessed with fewer field data requirements. While progress has been made with the Inpang Carbon Bank project the GOES Lab is cautiously optimistic about realizing agroforestry C offset transactions and the potentia l economic benefits to rural farmers in the near term. Follow - on studies have been developed in other smallholder plantation systems and community forests in Thailand, Laos and Vietnam. This work has expanded the experience of the GOES Lab in the context o f REDD+ forests as well. One could imagine that these new projects in Asia have been developed against a backdrop of views, perceptions and arguments over the pros and cons of REDD+ at the community level. Indeed there is a resonating view among those skep tical about REDD+ implementation that identifies the risk that REDD+ may marginalize local people, hinder their access and rights to forest and forest resources, and negatively impact local livelihoods. These are legitimate concerns, even in light of discu ssions about Larson have reviewed some of these concerns (Ribot and Larson, 2012). These Asian 20 projects and the proposed Senegal site are not focused on analyzing the various forest policies and implementation of these policies in the three countries, which can indeed impact how local people benefit or are marginalized in such act ivities as REDD+. Rather, our research is targeting the practical implementation of engaging communities in the tasks of measuring biomass in forested areas and the collection of field data that could be integrated with satellite remote and GIS analysis i n a Carbon2Markets MRV system. From preliminary results of Asia n projects, the GOES Lab has identified several important elements that affect community - based field measurements with GIS and remote sensing for a REDD+ MRV (Samek et al. 2013). Forest biom etric data collection can be done using simple or hi - tech tools. The diameter at breast height of a tree can be measured devises are now common in University labs and with local agencies in the field. The management of data measurements, however, must be systematic and uniform across biomass plots within a project location. Common print outs of spreadsheets, or data field forms in data loggers are important. Tools to manage all project data are also very important. Ideally, with web - based service becoming more and more common, data management can be developed as a set of tools for REDD+ implementation that include basic description and management, document and file ma nagement, plot level data management and carbon stock calculations, geo - spatial data management and even emissions reporting. Measurement and monitoring for REDD+ can effectively combine local, community data collection with expert analysis using remote sensing and GIS . National - level forest measurements often are conducted under National Forest Inventory (NFI) programs in which permanent sample plots are established and repeat ed measurement s are taken every few 21 years. Unfortunately, not all countries hav e adequate NFI programs in place. Forest monitoring, to assess areas and rates of change, is most often conducted using satellite remote sensing. The combination of g round based plot biomass measurements and satellite remote sensing analyses is powerful fo r measuring and monitoring REDD+ carbon stock and carbon stock changes. This is because, m easurement and monitoring of REDD+ carbon is not dependent on ground only or remote sensing only techniques, but a combination of the two. Integrating community leve l plot measurement data with geo - spatial analysis (GIS and remote sensing) supports REDD+ measurement and monitoring requirements for an MRV system. The combination captures or utilizes the opportunity to integrate community level abilities with professio nal, expert analysis. The limitation of the aforementioned work of the GOES Lab is that all these projects deal with forests or forest stands smallholder plantations, community forests, or REDD forests and do not address the application of the Carbon2 Markets MRV in systems of trees outside of forests nor agroforestry systems in predominantly agricultural landscapes. There is a need to explore the effectiveness, accuracy and precision of joint measurement campaigns in savanna landscapes and agroforestry systems outside forests. 1. 4 . D EFINITION OF CONCEPTS The following terms are used throughout this study, and the definitions given here are used to clarify those terms. 1.4.1. Biomass Biomass is a vegetation attribute that refers to the dry weight of a plant material within a given area. It includes living and dead above and below ground biomass (Figure 22 1 .2 ) as well as soil organic matter (IPCC, 2007) . However, only above and below ground biomass is considered in this study. Figure 1 . 2 Above and below ground biomass of a tree 1.4.2. Cropland There are many components of cropland as classified by the USDA based on the major uses. C ropland includes land planted for crops, crop failure, cropland used only for pasture and idle cropland (USDA/NASS, 2011). However, in this study, the concept is used to refer to only land on which agricultural crops are grown. 1.4.3. Crown Projection Area The crown projection area of a tree is the area of the vertical projec tion of the outermost perimeter of the crown on the horizontal plane (Gschwantner et. al , 2009). We determined the CPA by using the formula for the area of an ellipse which is 23 o Figure 1 . 3 Crown perimeter and crown projection area of a tree (adapted from Gschwantner et. al , 2009) 24 1.4.4. Forest The FAO define forest as a l and with a minimum area of 0 .5 ha occupied by trees with more than 5 m height and more than 10 % tree cover and a gricultural lands are excluded in this definition (FAO, 2010). According to the definition, the potential of trees outside of forests, especially in agricultural lands and more importantly, in the case of agrof orestry based systems and crop lands might be neglected in terms of providing benefits within the framework of the carbon market . In this study, forest is defined as "a land with a minimum area of 0.5 ha and occupied by trees with at least 30% tree cover and can reach a minimum height of 2 meters" as stated by the national definition adopted by Senegal . 1.4.5. Savanna It is defined as g rassland with scattered trees or shrubs . The tropical savannas occur between the tropical dry forests and the deserts. These savannas are warm and have low precipitation that is highly seasonal (Chapin et al ., 2002). 1.4.6. Trees Outside of Forest (TOF) According to the FAO (2002), these categories of trees are growing in areas not belonging to the category of forest, fo rest land, or other wooded land. In this study, the concept of trees outside of forest include s all trees in agricultural land, agroforestry systems, trees growing inside the settlements and along the roads. Regarding the size, only trees that attained a d iameter at breast height (DBH) of 10 cm or more are included in this definition. 1.4.7. Rural Community A Rural Community is the smallest unit of local government in Senegal composed of several villages. However, since 2013, all of the rural communities are now Communes after 25 the new law No. 2013 - 10 of December 28, 2013, which established the General Code of Local Government. 1.4. 8 . Arrondissement It is the next highest level after the Rural Community and is composed by 2 or more Rural Communities. 1.4. 9 . Sous Préfet The Sous - Préfet is the State Representative at the Arrondissement level. 1. 5 . G EOGRAPHY OF THE STUDY AREA The research is located in Senegal, in a semi - arid environment that is dominated by savanna ecosystems, and marginal arable lands. Th e study is carried out in the Fatick Region, located 190 km from the capital of Dakar. 26 Figure 1 . 4 Location of the study area The region is in the southern part of the so - called peanut basin, part of the S udanian zone characterized by savanna landscapes. This region is poorly developed economically, and suffers from low vegetation and crop productivity. Agriculture in Senegal and in this region of Senegal is dominated by peanut production and is now undergo ing severe productivity and production declines. Peanut production has been steadily declining over the past decades for historical, political, economic and environmental reasons. After playing a prominent part in the economic and social history of Senegal (Bonnefond and Couty, 1991) in the 1970s and 1980s, overall peanut production and commodity prices are now declining, leading to a reduction of peanut exports and loss of farmer incomes. Several factors have caused peanut yield to fall and consequently re 27 decline (Tappan et al. 2004) due to the reduction of fertilizer use, reduction of fallow periods (Kelly et al., 1996), and the concomitant shortening of the soil regeneration period (Reinwald, 1997). To this , must be added the rapid population growth of the country and the intensive use of land which has also contributed to agricultural land degradation and soil fertility depletion. Figure 1 . 5 Selected photos o f the study area Photo 1: Cashew plantation intercropped with groundnuts in Kouatine Village Photo 2: TOF (Prosopis africana) with c ashew p lantation on the background Keur Samba Gueye Village Photo 4: Big TOF (Acacia albida) in Senghor village Photo 3: Scattered individual trees in Senghor village's landscape 28 With increased population densities and associated reduction in fallow periods, soil fertility decline is becoming acute, not only in Senegal but throughout the tropics (FAO, 2005). This has been leading to a situation where soil and land degradation are an emerging threat to food security in developing countries (Scherr, 1999), which will at the same time increase the vulnerability of these places to climate change and reduce the range of options that farmers have for adaptation. Land degradation leads to lower water holding capacity, loss of soil organic matter, and increased loss of soil nutrients which leave farmers vulnerable to drought and increased temperatures. With little access to remediation resources such as irrigation, fertilizers, and new drought tolerant varieties of crops, farmers in the tropics and Senegal are likely to be disproportionately and adversely affected by climate change. These aforementioned factors have social and economic consequences for fa rmers and seriously affect their livelihood systems beyond income loss alone e.g. food security, health, nutrition, and sanitation. As land use intensification is reducing or eliminating fallow rotation periods, trees that are used in the agricultural landscape are being removed to expand the land under cultivation or as a source of fuel wood. Wood and charcoal are the main source of primary - million people (Media Club South Africa, 2009). Fir ewood and charcoal are used by 55.5 % and 11% of Senegalese households respectively (Senegal, 2010). The combined influence of shortened fallow period and the reduction of woody perennials are contributing to increased land degradation. With the addition al land 29 degradation and land scarcity, forest land is increasingly being brought under cultivation (Figure 1. 6 ). Like many developing countries, Senegal is facing major problems in the forestry sector and on forest la nds generally from deforestation and forest degradation. Forest cover has been declining over the past twenty years from 9, 348,153 ha in 1990 to 8,473,153 ha in 2010 (FAO, 2010). The rate of forest loss, estimated to be 45,000 ha per year in 2005, has bee n reduced to 40 000 ha per year due to national reforestation initiatives but this loss rate - Shortened fallow rotation period - Reduction of woody perennials Land intensification and land degradation Reduction of NPP Reduction of soil fertility Crop yield decrease Reduction of carbon stock Forest cover decline Greenhouse gas increase Farmers' incomes decrease Figure 1 . 6 Changes in land use and their environmental and economic consequences 30 remain s high and requires government intervention. The main causes of forest loss are clearing for agriculture (38 %), illegal production of charcoal (25%), fire ( 25%), other causes such as mining ac counting for 12% (FAO, 2010). While the factors that we mentioned are still causing negative consequences for both the environment and the social and economic conditions of communities, mitigation strategies exist that m ay contribute to reduce the effects. Carbon sequestration is among these strategies and agro - forestry is also among the mitigation activities that can benefit both the environment and the communities that practice these activities. As a sustainable practic e, it "helps to achieve both mitigation and adaptation objectives while remaining relevant to the livelihoods of the poor smallholder farmers in Africa" (Mbow et al ., 2014). However, measuring the carbon sequestered by these practices is a big challenge, e specially in savanna systems. 1. 6 . O RGANIZATION OF THE DISSERTATION This dissertation is organized in five chapters including the introduction (Chapter 1). In Chapter 2 a model is developed to predict tree DBH from crown projection area using remote sensi ng high resolution satellite imagery. H igh resolution data and ground data are combined to develop this model for trees outside forest s . Chapter 3 focuses on the role of Remote Sensing in measuring biomass and carbon at landscape level in savanna systems . A regression analysis between field measured DBH as dependent variable and remote sensing crown projection area as independent variable is developed. The regression equation is used to predict tree DBH and an allometric equation used to calculate the bioma ss and carbon for all trees . Chapter 4 is dedicated to apply the model at the landscape level and to produce carbon map of the study area. Finally, chapter 5 presents a summary of the results and discusses their implications, limits and recommendations for further research. 31 REFERENCES 32 REFERENCES Bonnefond, P. and Couty, P. (1991). Agricultural Crisis: Past and Future. In D. &. Jammeh, The Political Economy of Senegal under structural Adjustment (p. 232). New York: Praeger Publishers. Brown, S. (1997). Estimating biomass and biomass change of tropical forest. A primer. Rome: FAO Forestry Paper 134. Brown, S. (2002). Measuring, Monitoring and Verification of Carbon Benefits for Forest Based Projects. Philosophical Transactions: Mathematical, Ph ysical and Engineering Sciences, Vol. 360, No. 1797 , 1669 - 1683. Chapin, F., Matson, P., & Mooney, H. (2002). Principles of Terrestrial Ecosystem Ecology. New York: Springer - Verlag. Collier et al. (2008). Climate Change and Africa. Oxford Review of Econom ic Policy, Vol 24 (2) , 337 - 353. Dewees, P.A. (1995). Trees on farms in Malawi: Private investment, public policy, and farmer choice. World Development, 23 (7) , 1085 - 1102. Estrada - Carmona, Natalia; Abigail K. Hart; Fabrice A.J. Declerk; Celia A. Harvey; J effrey, C. Milder. (2014). Integrated landscape management for agriculture, rural livelihoods, and ecosystem conservation: An assessment of experience from Latin America and the Caribbean. Landscape and Urban Planning , 1 - 11. Evans, Kristen. (2008). Partic ipatory monitoring in tropical forest management: a review of tools, concepts and lessons learned. Bogor, Indonesia: Center for International Forestry Research (CIFOR). FAO. (2005). Realizing the economic benefits of agroforestry: experiences, lessons and challenges. In FAO, State of the World's Forests. Rome: FAO. FAO. (2006). Global forest resources assessment 2005. Main Report, FAO Forestry Paper. Rome: FAO. FAO. (2010). Global Forest Resources Assessment, Main Report. Rome: FAO. Franzel, S. a. (2002). I ntroduction. In s. a. Franzel, Trees on the farms; Assessing the Adoption Potential of Agroforestry Practices in Africa (pp. 1 - 11). Wallingford: CABI. Gschwantner, Thomas.; Schadauer, Klemens; Claude Vidal, Adrian Lanz et al. (2009). Common Tree Definition s for National Forest Inventories in Europe. Silva Fennica 43 (2) , 303 - 321. 33 Houghton, R.A. et al. (1985). Net flux of carbon from tropical forests in 1980. Nature, 316 (6029) , 617 - 620. Intergovernment Panel on Climate Change. (2001). Climate Change: The Scientific Basis. Summary of Policy Makers. Cambridge, UK: Cambridge University Press. International, W. (2008). Winrock Terrestrial Sampling Calculator. Retrieved from Winrock: http://www.winrock.org/ecosystems/tools.asp Intergovernment Panel on Climate Change. (2007). Good Practice Guidance for Land Use, Land Use Change and Forestry. Jonsson. K.; Ong. C.K.; Odongo, J.C.W. (1999). Influence of scattered nere and karite trees on microclimate, soil fertility and millet yield in Burkina Faso. Experimental a griculture.35 , 39 - 53. Kelly, V. et al. (1996). Cash crop and Food grain Productivity in Senegal: Historical View, New Survey Evidence, and Policy Implications. MSU International Development Paper, No. 20 . Keshava, N. (2003). A Survey of Spectral Unmixin g Algorithms. Lincoln Laboratory Journal, Vol 14, Nm 1 , 55 - 78. Liu, S., M. Kaire, E. Wood, O. Diallo and L.L. Tieszen. (2004). Impacts of land use and climate change on carbon dynamics in south - central Senegal. Journal of Arid Environment, 59 , 583 - 604. Matricardi, E.; Skole, D.; Cochrane, M.; Pedlowski M. and Chomentowski, W. (2007). Multi temporal Assessment of Selective Logging in the Brazilian Amazon Using Landsat Data. International Journal of Remote Sensing 28 , 63 - 82. Matricardi, E.; Skole, D.; Pe dlowski, M. and Chomentowski, W. (2013). Assessment of forest disturbance by selective logging and forest fires in the Brazilian Amazon using Landsat data. International Journal of Remote Sensing 34 , 1057 - 1086. Matricardi, E.; Skole, D.; Pedlowski, M.; C homentowski, W. and Fernandes, L. (2010). Assessment of tropical forest degradation by selective logging and fire using Landsat imagery. Remote Sensing of Environment 114 , 1117 - 1129. Mbow, Cheikh; David Skole; Moussa Dieng; Christopher Justcie; Dominick Kweshi; Landing Mane; Mohamed El Gamri; Vincent Von Vordzogbe and Hassan Virji. (2012). Challenges and Prospects for REDD+ in Africa: Desk Review of REDD+ Implementation in Africa. Copenhagen: GLP Report No 5. Mbow, Cheikh; Michel M. Verstraete; Bienvenu S ambou; Amadou Tahirou Diaw; henry Neufeldt. (2013). Allometric models for aboveground biomass in dry savannah trees of the Sudan and Sudan - Guinean ecosystems of Southern Senegal . Journal of Forest Research , 1 - 8. 34 Mbow, Cheikh; Peter Smith; David Skole; Lalisa Duguma and Mercedes Bustamante. (2014). Achieving mitigation and adaptation to climate change through sustainable agroforestry practices in Africa. Current Opinion in Environmental Sustainability (6) , 8 - 14. Media Club South Africa. (2009, April 23) . Africa Good News. Retrieved from Africa Good News: http://www.africagoodnews.com/development/environment/380 - senegal - green - charcoal - to - save - forests.html Medugu, Nasiru Idris; M. Rafee Majid; Foziah Johar and I.D. Choji. (2010). The role of afforestation programme in combating desertification in Nigeria. International Journal of Climate Change Strategies and Management, Vol. 2 Iss: 1 , 35 - 47. Montagnini, F. and Nair, P.K.R. (2004). Carbon sequestration: an underexploited environmental benefit of agroforest ry systems. Agroforestry Systems, 61 , 281 - 295. Neupane, Ramji P. and Thapa, Gopal B. (2001). Impact of agroforestry intervention on soil fertility and farm income under the subsistence farming system of the middle hills, Nepal. Agriculture, Ecosystems and Environment (84) , 157 - 167. Pandey, D. (2002). Fuelwood studies in India. Myth and reality. Bogor: CIFOR. Place, Frank; Joe Ssenteza and Keijiro Otsuka. (2001). Customary and Private Land Management in Uganda. In K. O. Place, Land Tenure and Natural Resou rce Management: A Comparative Study of Agrarian Communities in Asia and Africa (pp. 195 - 233). Baltimore and London: The John Hopkins University Press. Plan Vivo (2010). Website: http://www.planvivo.org/ Accessed December 16, 2013 Reinwald, Brigitte. (1997 ). 'Though the Earth Does Not Lie". Agricultural Transitions in siin (Senegal) under Colonia Rule. Paideuma,Bd , 143 - 169. Ribot, Jesse and Anne M. Larson. (2012). Reducing REDD risks: affirmative policy on an uneven playing field. International Journal of the Commons , 233 - 254. Samek, Jay et al. (2014). Integrated community - based participatory carbon measurement and monitoring with satellite remote sensing and GIS in REDD+ MRV Systems. In C. T. Louis Lebel, Climate Risks, Regional Integration, and Sustaina bility in the Mekong Region (p. 285). Bangkok: Stockholm Environment Institute. Scherr, Sarah J. (1999). Soil degradation: A threat to developing - country food security by 2020? Food, Agriculture and the nevironment Discussion Paper (27) . Senegal. (2010). Deuxieme communication nationale du Senegal: Convention Cadre des Nations Unies sur les Changements Climatiques. Dakar: Ministere de l'Environnement et de la Protection de la Nature. 35 Skole, David and Tucker, C. (1993). tropical deforestation and habitat f ragmentation in the Amazon: Satellite data from 1978 to 1988. Science 260 , 1905 - 1910. Symapungani, s. et al. (2010). The Potential of Using Agroforestry as a Win - Win solution to Climate Change Mitigation and Adaptation and Meeting Food Security Challenge s in Southern Africa. Agricultural Journal, 5 (2) , 80 - 88. Tappan, G.G., M. Sall, E.C. Wood , M. Cushing. (2004). Ecoregions and land cover trends in Senegal. Journal of Arid Environments vol 59 (3) , 427 - 462. Tschakert, P. ( 2001). Human dimensions of car bon sequestration: A political ecology approach to soil fertility management and desertification control in the Old Peanut Basin of Senegal. Arid Lands Newsletter, No. 9 . Unruh, J. R. (1993). Carbon storage in agroforestry: an estimate for sub - Saharan Afr ica. Climate Research , 39 - 52. Verchot, Louis V. & Singh V.P. (2009). Carbon Sequestration Opportunities with Smallholder Communities: Forestry, Agriculture and Agroforestry. 4th World Congress on Conservation Agriculture, February 4 - 7. New Delhi: World Co ngress on Conservation Agriculture. Verchot, Louis V., et al. (2007). Climate Change: Linking adaptation and mitigation through agroforestry. Mitigation And Adaptation Strategies For Global Change , 901 - 918. Wang, C; J. Qi and Cochrane, M. (2005). Assessme nt of Tropical Forest Degradation with Canopy Fractional Cover from Landsat ETM+ and Ikonos Imagery. earth Interactions 9 , 1 - 18. Wommer P.L., Tieszen L.L., Tappan, G.G., Toure A., Sall, M. (2004). Land use change and terrestrial carbon stocks in Senegal. Journal of Arid Environments (59) , 625 - 642 36 CHAPTER 2 . MODEL DEVELOPMENT TO PREDICT TREE DBH FROM CROWN PROJECTION AREA USING REMOTE SENSING 2.1. INTRODUCTION Carbon sequestration in agro - forestry has the potential of providing increased income and livelihoo d support, several environmental and agronomic co - benefits, and potential for climate change mitigation. At the same time, it can reduce the barriers to widespread adoption of these high carbon stock agro - forestry systems. For example, in the Nhambita Proj ect in Mozambique, results have shown that systems that combine sequestration and cash crop production have higher benefits compared with sequestration only, although they have less - carbon sequestration potential (Palmer and Silber, 2011). In addition to d irect ecosystem uses carbon appears to be a top - up benefit for farmers as it can provide additional incomes. Accounting for the market value of carbon in these systems cannot be achieved without a proper assessment of their performance for sequestering car bon (i.e. carbon sequestration rates), hence the need for cost effective methods suited to estimate carbon stocking potential of trees in farming landscapes. The international climate change mitigation community recognizes the need for rigorous and accur ate protocols for measuring, reporting and verifying carbon sequestration in agro - community level implementation and participation. International standards for carbon measurement are being developed and promoted through the international conventions, carbon financial markets, and both regulatory and voluntary regimes and institutions. As such, scientific and technical requirement s for international carbon measurement s are extremely hig h, and involves high reporting and verification standards. Currently the common use of carbon measurement in tree - based sequestration systems is in large - scale 37 plantations, or reforestation in large tracts of closed canopy forests. Less certainty and accorded to small - scale systems involving agroforestry or small - scale reforestation systems that are managed by local land owners, particularly in savanna landscapes. There are numerous landscape - wide deforestation and carbon measurement protocols for forests, particularly closed forests (Matricardi et al. , 2012). The Voluntary Carbon Standard (VCS) is also us ed in forest carbon accounting to lead a greenhouse gas reduction program across the world (VCS, 2014) and has developed several methods for forests. However, few methods are available for sparsely wooded systems, such as savannas. The Voluntary Carbon Standard (VCS) protocol for the Kasigau Corridor REDD Project in Kenya (VCS, 2011) is among those methods although it uses plot - level measures and not remote sensing. There has n o t been much progress on developing a method for assessing tree biomass over large areas in these non - forested landscapes. The current methods for landscape - level carbon measurement are particularly challenging and are unlikely to meet scientific and rigorous standards (for example the requirements of the CDM Executive Board) and potential carbon project investors. Therefore, there is a lack of confidence in existing measurement systems with implications that go beyond t he measurement of carbon credits, and create doubts about the project viability. New advances in remote sensing of canopies and individual tree crown detection offer great promise. It is also important to note that most of the forestry methods used to es timate biomass are based on diameter at breast height or 1.3 m from tree - stem base ( DBH ). In this study, we explore the role of crown projection area ( CPA ) as an allometric measure that can be obtained from high resolution r emote s ensing images, and investi gate the possibility of using 38 remote sensing CPA to predict DBH. Thus, the research goal of this study is to determine whether or not protocols can be developed to allow landscape scale measurements of trees outside of forests with the accuracy and precisi on that is acceptable to the international standards. To respond to this question I suggest basic measurement tools for monitoring carbon stocks in systems of trees outside of forests using remote sensing, ground measurements and GIS beca use there is a ne c arbon m easurement and r eporting that requires linking remote sensing to ground measurement. A significant challenge for efficient, timely and cost - effective automated monitoring of A/R/AF is the accurate measurement o f woody aboveground biomass in the landscape. Detecting and measuring individual tree objects that are natural trees or sparsely planted on farms, along roadways, or in backyards cannot be done accurately with low or moderate resolution satellite imagery where a single tree may represent only a small proportion of a single pixel. While pixel unmixing techniques (Keshava 2003) can provide an estimate of the proportion of different land cover types represented by a single pixel, in general, these techniques are not amenable for accurate estimation of tree size, number or other measurements required for accurate assessment of forest cover change in sparsely planted landscapes. However, recent technological advancements in satellite image acquisition provide ac cess to hyper - resolution imagery of the Earth and its land - uses. Very high - resolution imagery enables the detection of objects in the landscape using remote sensing. The Worldview 2 satellite acquires 4 bands of multispectral data at 2.4 m spatial resolut ion and a panchromatic band at 4 6 centimeters. Subsequent processing can be used to associate attributes, such as size or texture , with detected objects. Therefore, it is possible to count trees, estimate size, and evaluate change remotely. 39 Mbow et al. ( 2013) used the close relationship between DBH and tree height to build a biomass estimation model, but they did not include the crown diameter in their research because "it was not exploited further at this stage". Additionally, the relatively low density of trees outside forests makes the assessment by conventional methods costly and time - consuming (Singh and Chand, 2012). If one can detect individual trees and tree crowns, there may be a potential to develop a crown - based allometry for individual trees. This requires the development of methods for ground measurements. Most reforestation projects use small sample plots in a sampling frame for forest inventory. This ground measurement protocol must be re - defined for systems of individually planted trees, a nd then also extended to the woodland or savanna landscape. The landscape approach allows one to inventory both land in forests (in this case savannas) and land on trees outside of forests (in this case agroforestry landscapes). Another reason for combini ng remote sensing and ground data is that biomass is mostly determined by classical allometries from measurements of either diameter at breast height or height, and currently none of these measurements can be obtained from remote sensing. However, it may b e possible to use high resolution satellite imagery to measure tree crown projection area and use it as an independent variable to predict tree DBH. In this study, I develop ed a regression model to predict tree diameter at breast height (DBH) from tree c rown projection area (CPA) for trees outside forests in savanna agroforestry systems. This relationship could then be used with remote sensing if it were possible to accurately measure CPA using remote sensing. In my model, field measures of crown projecti on areas of individual trees, are used to determine the relationship between diameter at breast height (DBH) and CPA. After establishing the relationship, the DBH of 40 trees can be predicted from satellite observations of tree crowns and then used as the ind ependent variable in standard allometric equations to estimate biomass and carbon based on DBH. 2.2. MATERIA LS AND METHODS 2.2.1. The Study Sites The study area for this work is in the Southern part of the so - called peanut basin of Senegal, and is part of the Sudanian zone characterized by savanna landscapes. The ground measurements and model development focused on villages in three Rural Communities of southern Senegal: Toubacouta, Keur Samba Gueye, and Nioro Alassane Tall . A Rural Community is the sma llest unit of local government in Senegal composed of several villages 2 . These communities dominated by a largely rural population are some of the most important production zones for Anacardium agro - forestry in Senegal and are also parts of the former Sene galese - German Anacardium Project (PASA). As shown in Figure 2.1 , Fatick, Ziguinchor and Kolda are the three major cashew production zones in the country. 2 Following the recent local elections that passed law No. 2013 - 10 of December 28, 2013, which established the General Code of Local Government, all of the R ural C ommunities became C ommunes . 41 Figure 2 . 1 Area of cashew production in Senegal and t he study area The climate is semi - arid and the annual rainfall ranges from 400 to 1000 mm. The rainy season usually lasts 3 months from July to September (Toubacouta), but can last 5 months (from June to October) in the Rural Community of Keur Samba Gu eye. The annual rainfall here can reach up to 1845 mm, as was the case in 2003 (LDP, 2009). Rainfall can vary significantly from year - to - year in quantity, timing, and duration. 42 The most common tree species that dominate the vegetation in the rural commun ity of Toubacouta are Cordyla pinata, Cola cordifolia, Mangifera indica , and block plantations of planted trees, including eucalyptus and the large areas of Anacardium occidentale . In the southern where is located the Rural Community of Keur Samba Gueye, t he vegetation is dominated by Combretum glutinosum, Cordila pinata, Pterocarpus erinaceus, Adansonia digitata, Zizyphus mauritiana, Diospyros mespiliformis, Prosopis africana and Parkia biglobosa . Regarding the Rural Community of Nioro Alassane Tall, t he vegetation is very sparse due to overexploitation and the forest cover is deteriorating gradually. It is dominated by species such as Cordyla pinata and Adonsonia digitata in its tree layer, Combretum glutinosum and Guiera senegalensis in the shrub laye r and some popular grass for fodder at the herbaceous layer (LDP, 2009). The main types of soil are locally called Dior and Deck . Dior soils, containing > 95% sand and few nutrients (< 2% organic carbon), are usually associated with former dune slopes. Deck soils are hydromorphic, occasionally flooded, contain 85 - 90% sand, and have 0.2% organic carbon (Badiane et al., 2000). 2.2.2. Field and remote sensing data collected Most of the data collected in this study are extensive field data that were colle cted during t wo field campaigns on May and June 2013. I also used remote sensing data from high resolution satellite imagery. In the next section I will discuss the acquisition and use of these data. For the ground data, a field inventory was conducted th roughout the non - forest land cover area of the study sites, which included measurements in plantations of Anacardium 43 occidentale and other trees on agricultural lands. The primary focus of the field work was to collect individual tree allometric tree data to estimate biomass and carbon stocks. All ground data were collected during a field campaign i n May and June 2013 using the following field parameters and associated equipment (Table 2 .1. ) Table 2 . 1 Variables measured in the field and equipment used Variables Description Instrument Plot dimensions Plot area and determination of trees "in" plot Tape measure Coordinates and altitude X and Y location of the centre of plot and each sampled tree Trimble GPS Tre e Diameter at breast height (DBH) Diameter at breast height Diameter tape and data collection sheet Tree Diameter above the root collar Diameter measured at 20 cm above the root collar for dwarf Anacardium Diameter tape and data collection sheet Tree cr own diameter Two perpendicular crown diameters Diameter tape Tree characteristics Genus and species of each individual tree Trimble GPS and data collection sheet Pictures Taking pictures of trees and field activities Nikon Photo Camera The measurement of tree DBH is used to estimate tree biomass using standard allo m etric equations. The DBH (cm) was measured for all sampled trees at 1.30 m above the ground with a standard DBH tape. For dwarf Anacardium trees, the diameter at 20 cm above the root collar (DAC) was measured as an alternative measure for DBH because of the tendency of Anacardium trees to subdivide into several small trunks below the level of breast height. To measure the tree crown projection area (CPA), a method that consists of measuring at least two crown diameters (m) was used for each tree. The first crown diameter measured is the longest crown diameter as determined by visual inspection of the tree. The second crown diameter is measured at the right angle (90°) to the longest crown dia meter. 44 Regarding the remote sensing data, it should be mentioned that h igh resolution satellite remote sensing techniques are very useful in detecting individual tree crowns but "the accurate monitoring of forest carbon has not been fully demonstrated ( Gonzalez et al ., 2010). The method used by Gonzalez et al . (2010) to validate Quickbird crown diameters against field measurements in forest areas showed a significant correlation between Quickbird - estimated and field measured crown diameters . However, the use of high resolution remote sensing techniques in tropical trees outside of forests areas continues to need more study . 45 Figure 2 . 2 Selected blocks in the Sokone (a) and Karang (b) sites from the Worldview 2 satellite image in 432 band combination a) 46 Figure 2.2. (cont'd) b) To estimate carbon i n trees outside forests, I apply a method that combines remote sensing and field data. Th is method requires fine resolution remote sensing data , in particular using Worldview 2 data (Table 2 .2. ). 47 Table 2 . 2 Spectral, spatial and temporal characteristics of Worldview 2 Spectral, spatial and tempo ral characteristics Quantities Number of bands 8 bands Resolution 0. 5 m (Pan) and 2 m (MSS) Period 100 minutes Orbital velocity 200 km/ 10 seconds Swath width 16.4 km at nadir Collection capacity Up to 1 million km 2 / day Average revisit 1.1 days a t 1 m GSD Date collected April 16, 2010 The Worldview 2 Satellite sensor is capable of acquiring image data at 0. 5 meter panchromatic (B&W) and 2 met ers multispectral resolution. It also features a revisit time of less than three days, as well as the ability to precisely locate an object within just th ree meters of its true geographic location. Two Worldview 2 satellite images were acquired for the two zones in overall study area ( around cities of Sokone and Karang) as they cover all villages in the three rural communities of the study: Toubacouta, Keur Samba Gueye and Nioro Alassane Tall. All imagery were acquired during the dry season (April 16, 2010) which helps reduce the problem of cloud cover. The image projection is UTM Zone 28 N, WGS 84 datum. To work at the higher spatial resolution, the panchro matic and multispectral images were merged through a pan - sharpening process using the Hyperspherical Color Space (HCS) resolution merge tool in ERDAS Imagine. The single high resolution color image created through that process is used and displayed in Arc Map to perform the digitizing and the other remote sensing processing and analysis. 2.2.3. Method The method deployed in this part of the study supports a technical investigation of carbon stock measurement in agroforestry systems and trees outside fore sts (TOF) over large landscapes and for multiple farms/ house holds in multiple communities. With satellite 48 imagery of 1 meter spatial resolution or finer it is possible to remotely detect and measure individual tree crowns and crown projected area (CPA). Wi th a relationship developed in this study between remote sensing - based crow n area and ground - measured DBH, I could develop individual tree allometry based on CPA and DBH. This method involve s several steps described in the following flow diagram (Figure 2. 3) and text that follows: 49 FIELD MEASUREMENT Regression Analysis DBH = f(CPA) + E Allometric equation Aboveground carbon stock Landscape Carbon Mapping Tree species Tree crown diameter Tree DBH Tree location (coordinates) REMOTE SENSING World view MSS (2m) World view Pan (. 5 m) Image fusion World view pan - sharpened image (.5 m) Manual delineation of tree CPA Figure 2 . 3 Flow diagram of biomass estimation and carbon mapping method 50 The method is divided into two components: The first component focuses on ground data measurements, where tree CPA and DBH are measured in the field and a correlation is established. With a strong correlation between CPA and DBH, it is then possible to use crown measurements in a schema from a DBH - based allometric equation to estimate aboveground live biomass in individual trees . Such DBH allometric equations are usually available as standard equations , and are also publi shed in the IPCC guidance documents . Thus the aim of t he model developed here is to represent a relationship between CPA, which could be defined using remote sensing as well as field measures, and DBH. If such a model is specified , then it is highly likely possible to use remote sensing as a surrogate for ground measures , which are usually made by foresters and other specialists working in the field but only in a few limited locations in the landscape due to the work effort that is required . The premise is that by substituting remote sensing for field data collection, a much larger area of the landscape can be captured and measured for carbon stocks. If an entire landscape can be measured, then discontinuous households can be bundled into a single project . T he transacti on costs could be lowered enough to allow their inclusion in carbon credit schemes. The second component involves the use of remote sensing to detect and digitize tree crown projected area (CPA) using hyperspectral imagery of 1 meter spatial r esolution or finer and then predict tree DBH using a model that relates CPA to DBH. In this way, I use a remotely sensed CPA to predict tree DBH in order to then apply a standard allometric equation to compute aboveground live biomass from the imagery obs ervations, on a tree - by - tree basis. The CPA is estimated from field measurements by using the formula for the area 51 of an ellipse which is as follows and requires two field or remote sensing measures of diameter: C PA = * 1/2 of the maximum diameter * 1 /2 of the diameter at 90 o Two points regarding this approach are warranted . First, remote sensing data could be g polygon construction tool in ArcGIS 2010, it is also possible to outline the best polygon. This is a subject for a future study. Second, we attempted to use automating the digitizing of crown polygons in the GOES L ab but there has been so me limited succe ss ( Appendix ) . This too could be the next step in this work, and would lead to an ability to process massive amounts of data and cover very large areas at very low costs. 2.2.4. Stratification of the study area It is important to have precise and accurate plot dimensions and locations as they determine which tree is in or out of the sample d plot and can be cross - mapped from field to imagery using detailed GPS coordinates . The strata map ( Figure 2.4 ) is produced to indicate the sp atial variation of tree den sity. The field sample protocol I used requires the determination of the number of plots, their size and shape, their location, and the type of parameters to be measured within each plot . It also requires a stratification scheme so plots can be used as sam ples and scaled up to a larger area . Stratification also narrows the variation in the regression model. 52 Figure 2 . 4 Block SOK 7 and the different strata The stratification use s of satellite image ry to establ ish the sample frame. The field data collection sites are selected within the former Senegalese/German Cashew Development project area (Figure 1. 4 ) . The work wa s done in two specific sites in the villages around the cities of Sokone and Karang within an ar ea measuring 6 km x 6 km (36 km 2 ) which corresponds to two Worldview 2 satellite images (Figure 2.2) . For each 36 km 2 area, ERDAS Imagine software is used to generate a sample grid of 9 blocks (2 km x 2 km). The different strata are determined and the ar ea of each stratum digitized for both Sokone and Karang zone (Tables 2 . 3 . and 2 . 4 . ) 53 Table 2 . 3 Stratification for Sokone zone SOKONE S trata SOK 1 SOK 4 SOK 7 Total Area (ha) % Area (ha) % Area (ha) % Ar ea (ha) % Cashew 96 24 116 29 44 11 256 21 Agricultural land 208 52 245 61 350 88 803 67 Villages 29 7 8 2 6 1 43 4 Permanent pastures 67 17 31 8 0 0 98 8 TOTAL 400 100 400 100 400 100 1200 100 Three blocks ( SOK 1, SOK 4, and SOK 7) are selected fo r Sokone zone and the different strata identified are: cashew plantation, the agricultural lands, the dwellings (villages) and the permanent pastures. The same number of blocks is selected in Karang ( KAR 1, KAR 8 and KAR 9) as showed in the following T abl e 2 . 4: Table 2 . 4 Stratification for Karang zone KARANG Area (ha) and % of strata KAR 1 KAR 8 KAR 9 Total Area (ha) % Area (ha) % Area (ha) % Area (ha) % Cashew 30 8 67 17 23 6 120 10 Agricultural land 17 1 43 221 55 333 83 725 60 Permanent pastures 199 50 112 28 44 11 355 30 TOTAL 400 100 400 100 400 100 1200 100 2.2.5. Selection of sampled plots The boundary o f each stratum is digitized and the area determined using the calculate geometry tool in Arc G IS . T he number of plots to be sampled is determined based on the land cover type and proportionately to the area covered. Three options were possible for the sampling design: a) locate plots randomly, b) locate plots systematically, or c) proceed by strati fi ed sampling of the landscape. 54 As mentioned by Ravindranath and Ostwald ( 2008) , determination of the appropriate number of sample plots is a critical step but "the complexity of sampling methods and procedures may force researchers to adopt a standard n umber of plots based on the visible heterogeneity of vegetation, soil and other conditions" ( Ravindranath and Ostwald , 2008) . I also acknowledge that there is a trade - off between the level of precision in carbon stock assessment and the cost of getting tha t level of precision . For the purposes of this study, I was only able to collect a limited number of sample plots. I evaluated the number of sample plots to reach 10% accuracy using the Winrock sample tool ( Winrock, 2008 ) and in some cases the number of pl ots exceeded 500. My sample is thus under - represented. Having noted this, it should be pointed out that the Winrock sample guide is meant to provide guidance for field sampling in a measurement framework that uses the samples to extrapolate to the entire l andscape. On the other hand, my methods is focused precisely on developing an alternative to sample - based estimation and moves directly to whole landscape estimation by direct measurement from remote sensing. Thus, the critical issue for my model was not t he number of plots, but the relationship between allometric measurement variables. In this instance more samples or heavily stratified samples could provide low variance in regression model. 55 Figure 2 . 5 Ground sample frame Also it should be noted that sample plots that I have established were not aimed at providing samples for extrapolation but a geographically distributed means to collect a lot of individual tree data. At the same time, these samples could b e used for validation. In the Sokone site, because of the high density of cashew plantation s , a minimum of ten (10) plots of 1024 m 2 (32*32) were randomly selected for this stratum for each of the three selected blocks except for block SOK 7 where only 7 p lots were sampled because of the low density in cashew plantation. For the other scattered TOF in agricultural land, a minimum of six (6) plots for each selected block wa s randomly sampled. A total of 50 plots were finally sampled in the Sokone site (Table 2 . 5 . ). 56 Table 2 . 5 Number of sampled plots for each stratum in the Sokone site SOKONE Number of plots in cashew plantation Number of plots in agricultural lands Total SOK 1 10 6 16 SOK 4 11 6 17 SOK 7 7 10 17 Total 28 22 50 For the Karang site, the agricultural lands are dominant and because of the low area covered by cashew plantation, a minimum of five (5) plots were sampled for this stratum except for block KAR 9 because it h as only 6 % of its area covered by cashew. Overall, a total of 54 plots were sampled in the Karang site (Table 2 . 6 . ). Table 2 . 6 Number of plots for each stratum in the Karang site KARANG Number of plots in cashew plantation Numbe r of plots in agricultural lands Total KAR 1 5 14 19 KAR 8 9 14 23 KAR 9 3 9 12 Total 17 37 54 The geographic location s of all measured trees were recorded with a global positioning system device (GPS), as this allows for data entry into a geographi c information system analysis (GIS) and possible return to the site. 57 Figure 2 . 6 Example of the layout of sampled plots and measured trees in the Sokone site Because of the dominance of cashew agroforestry in the study area, I sampled more plots on these plantations and each sample plot having sides of 32 x 32 m for a plot area of 1024 m 2 (0.1 ha). Most protocols recommend a plot size large enough to include at least eight to ten trees within the plot bound aries (Winrock, 2005). With the 1024 m 2 plot area, and the most common 10 x 20 m cashew tree spacing used in the 2 sites of the study area, 60% of the plots have at least 8 trees per plot (Table 2.7. ). 58 Table 2 . 7 Number of plots and their density for cashew plantations # of trees/ plot # of Plots % 3 to 5 8 17.78 6 to 7 10 22.22 8 to 10 12 26.67 11 to 12 6 13.33 13 to 14 3 6.67 15 to 17 3 6.67 18 to 23 3 6.67 Total 45 100 Although rectangular - shap ed plots can be more representative than square or circular plots of the same area (Hairiah et al., 2010), a method used in India (Singh and Chand , 2012) wa s chosen. It is a square - shaped plot which also helps comply with the linear tree planting systems u sed by farmers. The plot sizes are determined depending on the type of strata. A 32 x 32 m (1024 m 2 ) plot is dedicated to block cashew plantation, and a 100 x 100 m (1 ha) for scattered trees within the agricultural lands. I first measured the length or th e breadth and use the diagonal to ensure a true right angle at each of the four corners (Ravindranath Ostwald, 2008) . 2.2.6. Data analysis Data collected during the field work included the diameter at breast height (DBH), and two crown diameters, the lar gest and the one perpendicular to it that are used to calculate the tree crown area. These data were processed using EXCEL and SPSS. Data collected from all trees were used for this analysis as they have been all identified in the satellite image and digit ized using ArcMap. The digitized crown area was to be later used to correlate with the field measured tree DBH and then compared with the tree crown projection area measured from the field to validate the regression model. 59 2.2.6.1 . Manual delineation of t rees To build the regression model, the crowns of all measured trees were manually digitized. A pan - sharpened image is used to improve the view of the tree crown s and display the image in RGB using combination bands 132 and 432 in ArcGIS 10. At a 1:1000 s cale, we were able to clearly identify trees and delineate crown diameters . In some cases where the edge of the tree crown is not easy to determine, we compared the different band combinations 132 and 432 before performing the manual digitizing of crown di ameters using the Editor Construction Tool in ArcGIS 10. The following rules were set before starting the manual digitizing: Use the same scale (1:1000) for the delineation Use the tree points shapefile overlaid in the image as a reference for delineatio n . Digitize all the sampled trees whose field measurements are available 2.2.6.2. Correlation Analysis I used the Pearson product - moment correlation coefficient (r) to determine the direction of the correlation while the coefficient of determination R 2 h elps measure the strength of the correlation between the field - based crown projection area and tree DBH. The correlation is used to measure the strength, direction and the linear relationship between tree DBH and tree CPA. The coefficient of determination (R 2 ) is used to interpret the percentage of the variation in the DBH that is associated with the CPA. 2.2.6.3. Regression Analysis and validation of the models The regression analysis is used to predict one variable, given the other by determining the rel ationship between these two variables. It is used to describe how, given 60 the value of one variable or attribute (x) often called the independent variable, one may find the corresponding value for the other variable or attribute (y), called the dependent va riable. In the case of this study, the tree DBH is the dependent variable and the crown projection area (CPA) the independent variable. The purpose of the regression analysis is to estimate the change in the DBH from a given change in the CPA by finding a formula that represents the relationship between these two variables. The eventual aim is to be able to predict DBH from remote sensing measured CPA by finding an approximate value of the tree DBH from the value of the remote sensing CPA. To develop regre ssion models for the relationship of CPA with DBH, biomass, and carbon a simple linear (y= a+b.x) function was used. Throughout this process, the method of ordinary least squares (OLS) is used to determine the values of the constants a and b and determine the best fit of the straight line which is the position that makes the sum of the squared deviations of the observed Y values from the regression line a minimum (Husch et al ., 2003). For this case, the method is used to determine the regression coefficient that describes the strength and the sign of the relationship between the remote sensing crown projection area as the explanatory variable and the field measured DBH as the dependent variable. 2.3. RESULTS The results of the correlation analysis are inter preted using a classification of the size of a correlation coefficient made by Hinkle et al ., (2003) and presented in the following table. 61 Table 2 . 8 Rule of Thumb for interpreting the size of a correlati on coefficient Size of Correlation Interpretation .90 to 1.00 ( - .90 to - 1.00) Very high positive (negative) correlation .70 to .90 ( - .70 to - .90) High positive (negative) correlation .50 to .70 ( - .50 to - .70) Moderate positive (negative) correlation .3 0 to .50 ( - .30 to - .50) Low positive (negative) correlation .00 to .30 (.00 to - .30) Little if any correlation 2.3.1. Relationship between field measured DBH and field measured CPA 2.3.1.1. Model developme nt The relationship between field measured tree DBH and crown projection area ( CPA) calculated from field data is established to estimate an initial model . To describe this relationship, data from the Sokone site (377 sampled trees) are used for model development and data (317 trees) from the Karang sit e for validation. The relationship between field DBH and CPA is established to observe the prediction of tree DBH from observations of crown projection area ( CPA). In Figure 2. 7 the graph shows the relationship between the field measured Crown Projection Area (CPA) and the field measured DBH for sampled trees of all types in Sokone site. This includes all cashew and all other TOF trees combined. There is a linear trend within the range of the 377 plotted trees that have been used. The pattern of the scatte r plot shows a moderate correlation between the two variables with a coefficient of determination R 2 = 0.56 for the fit of these data. That means that 56% of the variation in the tree DBH can be explained by variations in the value of the CPA based on the model. 62 Figure 2 . 7 Correlation between field measured DBH and field measured CPA for sampled trees of all types (N= 377) in the Sokone site After classifying trees based on the number of stems, the model was applied for single stemmed as well as for multi - stemmed trees. Results of th is relationship between DBH and CPA from field data presented a better coefficient of determination, which is 0.594 for the 314 single stemmed trees and 0.664 for the 63 multip le multi - stemmed trees. This is consistent with findings from Samek et al . ( 2011 ) working on a GOES project in multi - stem - lychee plantations in Southeast Asia, where a crown - based allom e tric model worked better for multi - stem trees. Also , my results presen t a better correlation compared to those from plantation data collected in the Rukinga Ranch in Eastern Kenya by the Wildlife Works Kasigau Corridor REDD project. T he correlation between DBH and crown diameter in Rukinga shows a lower R 2 (0.514) for multi - stemmed trees ( C BP, 2012) However, it is important to mention that the sample for the Rukinga data is dominated by small trees as 98.5 % (843 out of 855 trees) have less than 50 cm DBH. y = 0.2705x + 21.148 R² = 0.56 0 20 40 60 80 100 120 140 160 180 0 100 200 300 400 500 600 Field Measured DBH (cm) Field Measured CPA (m 2 ) 63 When applied to 297 sampled trees of only cashew ( Anacardium occident ale ) in the Sokone site the relationship between the field measured crown projection area (CPA) as the independent variable and field measured tree DBH shows a moderate correlation between the two variables with a coefficient of determination R 2 = 0.57. T hat means that 57% of the variation in the tree DBH can be explained by variations in the value of the CPA based on the model. The same procedure was applied to the non - cashew sampled trees. A total of 80 trees were used in the Sokone site, and a moderate positive correlation between the two variables was found with a coefficient of determination R 2 = 0.62. That means that 62% of the variation in the tree DBH can be explained by variations in the value of the CPA based on the model. 2.3.1. 2 . Model valida tion To validate the model, the following Equation 1 is used: Field measured DBH = 0.2705 * Field measured CPA + 21.148 Equation 1 The equation is applied to the 315 trees in the Karang site to calculate the predicted values of their DBH from the model sample in the Sokone site. This validation is done by comparing the DBH predicted by the model and the field measured DBH o f the same trees. The predicted DBH were plotted against the field measured DBH (Figure 2. 8 ) which gives a R 2 of 0.58, meaning that 58% of the field measured DBH was explained by the predicted DBH from the linear regression model. 64 Figure 2 . 8 Scatter plot of model validation measured vs. predicted DBH 2.3.2. Relat ionship between field measured CPA and remote sensing measured CPA T o measure the closeness of both data sets, t he relationship between field measured and remote sensing CPA is established by comparing these two CPA data sets. A simple linear regression is applied and the regression was done for all trees in Sokone (Figure 2. 9 ) and Karang (Figure 2. 10 ) and a combination of both sites (Figure 2. 11 ). For the Sokone site, a total of 377 sampled trees were used to explain the relationship between the remote sensing and the field measured crown projection area (CPA) in the scatter plot shown in Figure 2. 9 . The pattern of the scatter plot shows a moderate correlation between the two variables with a coefficient of determination R 2 = 0.688. That means that we ha ve 68% correlation between field measured and remote sensing R² = 0.5894 0 20 40 60 80 100 120 140 160 0 20 40 60 80 100 120 140 160 180 Field measured DBH (cm) Predicted DBH (cm) 65 Figure 2 . 9 Relationship between Remote sensing and field measured CPA of all sampled trees (N=377) in the Sokone site For the 297 Anacardium t rees in the Sokone site, the relationship between the remote sensing and the field measured crown projection area (CPA) shows a moderate correlation between the two variables with a coefficient of determination R 2 = 0.629. That means 62% accuracy between f ield measured and remote sensing CPA based on the model. The other 80 trees ha ve more promising results with a strong correlation. The coefficient of determination is 0.766 suggesting 76% accuracy between field measured and remote sensing CPA and showi ng a better fit of the model for natural trees. For the Karang site, a total of 315 sampled trees were used to explain the relationship between the remote sensing and the field measured crown projection area (CPA) in the scatter plot shown in Figure 2. 1 0 . The pattern of the scatter plot shows a moderate correlation between the two variables with a coefficient of determination R 2 = 0.687. That implies a 68% accuracy between field measured CPA and remote sensing CPA based on the model. y = 0.4424x + 14.251 R² = 0.6881 0 50 100 150 200 250 300 350 0 100 200 300 400 500 600 Field measured CPA (m 2 ) Remote sensing CPA (m 2 ) 66 Figure 2 . 10 Relationship between Remote sensing and field measured CPA for all sampled trees (N=315) in Karang When applied to only Anacardium , a total of 182 trees were used and the relationship between the remote sensing and the field measured crown projection area (CPA) revealed a low positive correlation between the two variables. A coefficient of determination R 2 = 0.368 suggested only 36% accuracy. When the Anacardium were excluded and the model was applied to the ot her 133 non cashew tree species alone, the relationship between the remote sensing and the field measured crown projection area (CPA) represents a 0.736 coefficient of determination suggesting that 73% accuracy between field measured CPA and the remote sen sing CPA based on the model. When data from both sites are combined, the relationship between the remote sensing and the field measured crown projection area (CPA) was also tested for all 692 trees. The scatter plot presented in Figure 2.1 1 shows a modera te correlation between the two variables y = 1.1789x + 12.59 R² = 0.6878 0 100 200 300 400 500 600 0 50 100 150 200 250 300 350 Field measured CPA (m 2 ) Remote sensing CPA (m 2 ) 67 with a coefficient of determination R 2 = 0.671, and therefore 67% accuracy between field measured CPA and remote sensing CPA based on the model. Figure 2 . 11 Relationsh ip between Remote sensing and field measured CPA in both sites (N=692) The application of the model based on tree species also shows significant results in both cashew ( Anacardium occidentale ) and non - cashew trees. Using the simple linear regression, an R - squared of 0.597 was obtained for cashew trees and 0.729 for the non - cashew natural trees. 2.3.3. A Model to predict DBH from remote sensing CPA 2.3.3.1. Model development This model is buil t on the correlation between the remotely sensed crown pro jection area and the field - measured tree DBH . After describing the relationship between field bas ed DBH and field based crown projection area, the field measured CPA is replaced by the remote sensing - based measurements to predict DBH to build the model whi ch is applied f or all trees in both the Sokone and Karang sites. y = 1.299x + 10.196 R² = 0.6715 0 100 200 300 400 500 600 0 50 100 150 200 250 300 350 Field measured CPA (m 2 ) Remote sensing CPA (m 2 ) 68 The simple linear function is used to estimate a model between remote sensing CPA and DBH , using the field data for DBH . The resulting linear model is: DBH = 0.5912 (RS CPA) + 13.408 Equation 2 The model data show s a linear relationship of the two variables for the 377 sampled trees in the Sokone site. As shown in Figure 2.1 2 , the pattern of the scatter plot shows a stro ng correlation between the two variables with a coefficient of determination R 2 = 0.7 60 , which means that 7 6 % of the variation in the tree field measured DBH can be explained by variations in the value of the remote sensing CPA based on the model. Figure 2 . 12 Model - based DBH using Remote sensing CPA in Sokone (Linear model) When applied to the 297 sampled Anacardium trees in the Sokone site, a linear model from remote sensing CPA shows a positive linear rel ationship. The coefficient of determination R 2 is 0.711, which means that 71% of the variation in the tree field measured DBH can be explained by variations in the value of the remote sensing CPA based on the model. y = 0.5912x + 13.408 R² = 0.7605 0 20 40 60 80 100 120 140 160 180 200 0 50 100 150 200 250 300 350 Field measured DBH (cm) Remote sensing CPA (m 2 ) 69 Better results were found when only app lying the model to the other 80 TOF tree species, with a coefficient of determination R 2 = 0.84 suggesting that 8 4 % of the variation in the tree field measured DBH can be explained by variations in the value of the remote sensing CPA based on the model. A non linear model shows a lower coefficient of determination R - squared of 0.61 (Figure 2.1 3 ) compared to the linear model which suggests that the linear function is the best fit of the model. However more work could be done to refine the specific non - line ar model and alter its coefficients. Figure 2 . 13 Model - based DBH from Remote sensing CPA for all sampled trees in the Sokone site (Non - linear model) In the Karang site, the results for all 315 trees show a positive linear model based on the remote sensing CPA. As shown in Figure 2.1 4 , the model shows a strong prediction with a coefficient of determination R 2 = 0.791, which means that 79% of the variation in the tree field measured DBH can be explained b y variations in the value of the remote sensing CPA based on the model. The resulting linear model is: y = 29.102ln(x) - 65.959 R² = 0.6179 - 20 0 20 40 60 80 100 120 140 160 180 0 50 100 150 200 250 300 350 Field measured DBH (cm) Remote sensing CPA (m 2 ) 70 DBH = 0.3514 ( RS CPA) + 24.613 Equation 3 Th e model coefficient of determination is 0.71 when applied to only the Anacardium trees, suggesting that 71% of the variation in the 182 field measured DBH can be predicted by variations in the value of the remote sensing CPA based on the model. When applied to the other 133 non - cashew trees, the coe fficient of determination is 0.664. Figure 2 . 14 Model - based DBH from Remote sensing CPA for all sampled trees in the Karang site (linear model) This linear model shows a better result (R - squared = 0.79) co mpared to the non linear model for which the coefficient of determination is 0.77 (Figure 2.1 5 ) y = 0.3514x + 24.613 R² = 0.7913 0 20 40 60 80 100 120 140 160 0 50 100 150 200 250 300 350 Field Measured DBH (cm) Remote Sensing CPA (m 2 ) 71 Figure 2 . 15 Model - based DBH from Remote sensing CPA for all sampled trees in the Karang site (Non - linear model) The model was also estimated from all the species in both Sokone and Karang sites combined. The results for all 692 trees show a positive linear model between the remote sensing CPA and DBH (Figure 2.1 6 ). However, the coefficient of determination R 2 = 0.723 is slightly lower than those obtained when the correlation is established by site. This suggests a probable impact of site condition and characterization . y = 21.255ln(x) - 34.879 R² = 0.7715 0 20 40 60 80 100 120 140 0 50 100 150 200 250 300 350 Field Measured DBH (cm) Remote Sensing CPA (m 2 ) 72 Figure 2 . 16 Relationship between field based DB H and Remote sensing based CPA of all sampled trees in both Sokone and Karang sites (linear model) Compared to the linear, the non linear function presents also a lower coefficient of determination (R - squared= 0.66); suggesting a better results for the l inear (Figure 2.1 7 ). Figure 2 . 17 Relationship between field based DBH and Remote sensing based CPA of all sampled trees in both Sokone and Karang sites (Non - linear model) y = 0.4332x + 20.544 R² = 0.7233 0 20 40 60 80 100 120 140 160 180 0 50 100 150 200 250 300 350 Field measured DBH (cm) Remote sensing CPA (m 2 ) y = 24.418ln(x) - 47.732 R² = 0.6605 0 20 40 60 80 100 120 140 160 180 0 50 100 150 200 250 300 350 Field measured DBH (cm) Remote sensing CPA (m 2 ) 73 2.3.3. 2 . Results comparison with CB P study in Western Kenya Through the Carbon Benefit Project (CBP) , MSU and ICRAF researchers collected data in western Kenya along the Yala River Basin in three established blocks: lower Yala, middle Yala and upper Yala. The objective of that study was to develop a generic allometric equation to establish a valid basis for scaling up above and below ground biomass carbon at landscape scale. Unlike our data where the DBH ranged between 10 and 167.1 cm, the CBP data are for trees whose diameter at breast height (DBH) > 2.5cm . 834 out of the 855 trees, representing 97.5 % of all sampled trees are less than 40 cm DBH. A n examination of these Kenya results regarding the relationship between DBH and crown has shown a good linear rel ation ship with a R - squared of 0. 80 (Figure 2.1 8 ). It is interesting to note that the Senegal model and the Kenya model are similar. The slope of the prediction is closely the same, but varies in the Y - offset. Given the logic that no tree with zero crown would have any DBH (except p erhaps a dead tree) we could do a little more work to refine the y - intercept of these models so that they were closely similar and then we would have a general model for dry systems. 74 Figure 2 . 18 Relationship between DBH and Crown area in Western Kenya for 855 trees (ICRAF, 2010) 2.3.4. Regression model for Remote sensing CPA and field measured DBH The results obtained by correlating field measured DBH and field measured CPA confirmed the difficulties of measuring tree crown size with very high accuracy in the field as compared to remote sensing. C onsequently, results of correlation between remote sensing CPA and field DBH are more significant with an R - squared of 0.760 and 0.791 for Sokone and Karang site s respectively than those between field measured C PA and field measured DBH (R - squared of 0.564 and 0.589 respectively). For this reason, the remote sensing CPA can be used for correlation with the field measured DBH to measure the closeness of the linear relationship between these two variables. A simple linear regression analysis was developed and applied for each of the two sites of Sokone and Karang. For the Sokone site, the results of the regression analysis appear as follows: y = 0.3797x + 6.1925 R² = 0.7947 0 50 100 150 200 250 0 100 200 300 400 500 600 700 DBH (cm) Crown Area (m 2 ) 75 Table 2 . 9 Regression model and validation statistics for the relationship between Remote sensing CPA and field measured DBH for all sampled trees in Sokone site Regression model Constants Multiple R R Square Adjusted R Square a b DBH = a+b* RS CPA 13.408 0.5912 0.8720 0.7605 0.7598 The regression equation is given in the form of y= 13.408 + 0.5912 (x). The adjusted R squared is = 0.7605, which means that 76% of the DBH is determined by the CPA. Also we notice that R square and adjusted R square have almost the same values, 0.7605 and 0.7598 respectively, showing a minimal shrinkage based on this indicator and indicating that the equation can be generalized. For all sampled trees in the Sokone site only, the regression equa tion is given in the form of: DBH = 0.5912 (RS CPA ) + 13.408 Equation 2 For the Karang site, the results of the regression analysis for all sampled trees appear as foll ows: Table 2 . 10 Regression model and validation statistics for the relationship of Remote sensing CPA and field measured DBH for all sampled trees in Karang site Regression model Constants Multiple R R Squar e Adjusted R Square a b DBH = a+b*CPA 24.613 0.351 0.8895 0.7912 0.7906 The regression equation is given in the form of y= 24.613 + 0.351 (x). The adjusted R squared is = 0.7906, which means that 79% of the DBH is determined by the CPA. Also, 76 bot h R squared and adjusted R squared have the same values (0.79), showing no shrinkage based on this indicator and suggesting that the equation can be generalized. For all sampled trees in the Karang site only, the regression equation is given in the form o f: DBH = 0.351 (RS CPA ) + 24.613 (Equation 3 Results of the DBH estimation model accuracy show a good correlation between the measured and the predicted DBH. For all cases, the coefficient of d etermination (R 2 ) is more than 0.7. Using a general linear model that in cludes tree height, effective crown area, soil type, and age factors, Prieditis et al . (2012) found a strong relationship between predi cted and measured DBH with an R - squared of 0.87 . The highest coefficient compared to our results can be explained by the various parameters integrated in the model. These authors suggested that "models that use field or remotely - sensed measurements of a tree height as a predictor variable can be expect ed to produce a reasonably accurate estimate of DBH (R 2 = 0.792)" which is close to the R 2 of 0.757 and 0.791 that we found for Sokone and Karang respectively, although their model is applied in forest area. However, when the model uses crown dimension meas urements and information about age and soil type, the accuracy of DBH increases with an R 2 = 0.872 (Prieditis et al ., 2012). 2.4. DISCUSSIONS AND CONCLUSION In this chapter a model was developed that combined remote sensing and ground data that can be used to estimate biomass in trees outside of forests (TOF) . The role of CPA (crown projection area) from high resolution satellite images to predict tree DBH was 77 explored , and the relationship between tree crown projection area (CPA) obtained from Remote s ensing techniques and the field measured tree DBH investigated. The crown projection area was measured through remote sensing and used as the independent variable to predict the tree DBH. Also, tree crowns measured by remote sensing were compared with th e crowns of trees measured on the ground. For the relationship between Remote sensing CPA and field measured CPA, the results have shown a good relationship between remote sensing tree crowns and the field measured CPA in the Sokone site . T he R 2 values ar e ranging from 0.62 (only cashew trees) to 0.76 (natural trees) and 0.68 when the model is applied for all trees combined (377) . For the Karang site, the R 2 values are 0.68 for all trees combined (315), 0.36 for only cashew (182) and 0.73 for natural tre es (133). These results suggest that the model works better for natural trees. However, there were some limitations to the methods used that might affect these results. The first important weakness is the period between the image acquisition and the fiel d campaign. The satellite images were acquired in 2010 while the field data used was collected in 2013. This might explain for most trees why the values of the field measured CPA were bigger than those of the remote sensing. For example, by correlating f ield measured CPA as the dependant variable with the remote sensing CPA as the independent variable the following equation is produced for all trees in Sokone: y = 0.4424 x + 14.251 meaning that RS CPA = 0.4424* FM CPA + 14.251 Equation 4 For the Karang site the results are similar and the equation obtained for all sampled trees is: y = 1.1789 x + 12.59 meaning that 78 RS CPA = 1.1789*FM CPA + 12.59 Equation 5 It is also important to mention cases where the remote sensing values are bigger due especially to disturbances including pruning and fire. The following pictures (Figure 2.1 9 ) are an illustration of the effects of these two forms of disturba nces on trees. Figure 2 . 19 The effects of tree disturbances on tree crown shape a) Pruning reduces CPA b) Fire reduces CPA To determine the relationship between the remote sensing CPA and the field m easured DBH, both simple linear and non linear regression analysis are used and applied for the three different cases (in the Sokone site, in the Karang site and for both sites combined) . For the linear function, the coefficients of determination are very high for all cases with an R 2 > 0.70 . The non linear function has R - squared values ranged between 0.61 for the Sokone site to 0.77 for the Karang site. Applied to the combined data of both site, the R 2 is 0.72 . Compared to the non linear, results suggest that the linear model is the best. Among the limitations of the methods, I acknowledge that t he spatial pattern of the Anacardium plantations and their visibility in the satellite image are not as obvious as those of individual and scattered trees. As a result, the identification and delineation of tree crowns using remote sensing is less accurate than for the individual trees. This is confirmed by the 79 coefficient of determination of field measured CPA vs. remote sensing CPA for Anacardium , which is low specially for the Karang site where it falls to 0.36 compared to the other species where it reaches 0.76. Secondly, some individual scattered trees have experienced disturbances that affect the shape and the size of their crown. As mentioned earlier, prun ing and fire are the major causes of these disturbances: trees lost some branches and/or even part of their bark, which in turn reduces the size of the crown. Also, the Anacardium pattern and the distance between them usually create a situation of intermi ngling, i.e. the growth of some trees may be affected by the proximity of another tree, which in turn reduces the crown size and modifies the shape of the tree. Consequently, the relationship between DBH and crown area can be affected because of its sensit ivity to crown competition factor (Anderson et al ., 2000). There is also a significant correlation between field measured DBH and remote sensing crown projection area with coefficients of determination of 0.76 and 0.79 for the Sokone and Karang sites res pectively (Figures 2.1 2 & 2.1 4 ). These values can be compared and are close to those found by Kuyah et al. (2012) in Western Kenya. Using destructive sampling, they found that the relationship between crown area and DBH has shown a coefficient of determina tion R 2 = 0.8 5 for harvested trees. The model is also applied to anacardium species alone in both Sokone and Karang sites and the results show a high R - squared of 0.71 in regards to the correlation between remote sensing CPA and field measured DBH, and 0. 63 and 0.36 respectively for Sokone and Karang site in regards to the relationship between the remote sensing CPA and the field measured CPA. 80 When applied to natural trees, the coefficient of determination is higher (0.8 4 ) in Sokone than in Karang (0.66) for the correlation between remote sensing CPA and field measured DBH, and 0.76 and 0.73 regarding the relationship between the remote sensing CPA and the field measured CPA. In all cases, the strength of the relationship of field measured DBH with remot e sensing CPA is high with R - squared > 0.70 . Also the y - intercepts in all cases have positive values suggesting that the equations are suitable for predicting DBH of trees with a delineated CPA outside the range of sampled trees (Anderson et al ., 2000). Ov erall, results of this study show that this developed model can be used to predict tree DBH using high resolution satellite image especially individual trees outside forests where the detection is easier than for clustered trees, and natural trees better t han managed trees. 81 APPENDIX 82 Appendix A: PRELIMINARY RESULTS WITH AUTOMATED DETECTION OF TREE CROWM PROJECTION AREA (CPA) Figure A.1. Automated detection in the Sokone landscape Some testing with automated individual tree detection and automated CPA delineation has promise, but some error of omission remain. 83 Figure A.2. View of automated detection in cluster vegetatio n In some region, the detection result of cluster vegetation area is not good 84 Figure A.3. View of automated detection in individual big trees ... s ome big trees have lower NDVI, and that cause some loss of detection 85 REFERENCES 86 REFERENCES Anderson et al. (2000). Estimating forest crown area removed by selection cutting: a linked regression - GIS approach based on stump dia meters, Forest Ecology and Management , 137, 171 - 177 Badiane, Aminata Niane; Mamadou Khouma and Modou Sene. (2000). Region de Diourbel: Gestion des Sols, Working Paper 15. Somerset, England: Drylands Research. Gonzalez Patrick, Gregory P. Asner, John J. Bat tles Michael A. Lefsky, Kristen M. Waring, Michael Palace. (2010). Forest carbon densities and uncertainties from Lidar, Quickbird, and field measurments in California. Remote Sensing of Environment (114) , 1561 - 1575. Hairiah, Kurniatun; Sonya Dewi; Fahmud din, Agus; Sandar Velarde; Andree Ekadinata; Subekti Rahayu and Meine Van Noordwijk. (2010). Measuring carbon stocks accross land use systems: A Manual, Bogor, Indonesia. World Agroforestry Centre(ICRAF), 155 pages Hinkle Dennis E., Wiersma, William, Steph en G. Jurs . (2003). Applied Statistics for the Behavioral Sciences , 5th Edition. Boston: Houghton Mifflin . Husch, Bertram; Thomas W. Beers, and John A. Kershaw, Jr. (2003). Forest mensuration. Hoboken, New Jersey: John Wiley & Sons. Keshava, N. (2003). A Survey of Spectral Unmixing Algorithms. Lincoln Laboratory Journal, Vol 14, Nm 1 , 55 - 78. Kuyah, Shem; Catherine Muthur; Ramni Jamnadass; Peter Mwangii; Henry Neufeldt; Johannes Dietz. (2012). Crown area allometries for estimation of aboveground tree bi omass in agricultural landscapes in Western Kenya. Agroforestry Systems , 1 - 11. Matricardi, E.; Skole, D.; Pedlowski, M. and Chomentowski, W. (2013). Assessment of forest disturbance by selective logging and forest fires in the Brazilian Amazon using Lands at data. International Journal of Remote Sensing 34 , 1057 - 1086. Mbow, Cheikh; Michel M. Verstraete; Bienvenu Sambou; Amadou Tahirou Diaw; henry Neufeldt. (2013). Allometric models for aboveground biomass in dry savannah trees of the Sudan and Sudan - Guin ean ecosystems of Southern Senegal . Journal of Forest Research , 1 - 8. Palmer, C. and Silber, T. (2011). Trade - offs between carbon sequestration and rural incomes in N'hambita Community Carbon project, Mozambique. Land Use Policy, 29 (1) , 83 - 93. Priediti s, G; Ingus Smits, Irina Arhipova, Salvis Dagis, Dagnis Dubrovkis. (2012). Allometric Models for Predicting Tree Diameter at Breast Height. 8th WSEAS International Conference on 87 Energy, Environment, Ecosystems and Sustainable Development (pp. 105 - 110). Fa ro, Portugal: WSEAS LLC. Samek, Jay et al. (2014). Integrated community - based participatory carbon measurement and monitoring with satellite remote sensing and GIS in REDD+ MRV Systems. In C. T. Louis Lebel, Climate Risks, Regional Integration, and Sustai nability in the Mekong Region (p. 285). Bangkok: Stockholm Environment Institute. Singh, Kuldeep and Pritam Chand. (December 2012). Above - ground tree outside forest (TOF) phytomass and carbon estimation in the semi - arid region of southern Haryana: A synthe sis approach of remote sensing and field data. Journal of Erath System Science (121) No. 6 , 1469 - 1482. VCS. (2011). Pr Deforestation (REDD) credits under VCS Program Website: http://www.v - c - s.org/sites/v - c - s.or g/files/VCS%20Statement%20First%20REDD%20credits%20FINAL.pdf VCS. (2014). Website: http://www.v - c - s.org/GHG - program/ Accessed 13 June 2014 Winrock International (2008). Winrock Terrestrial Sampling Calculator. Retrieved from Winrock: http://www.winrock.org /ecosystems/tools.asp 88 CHAPTER 3 . THE ROLE OF REMOTE SENSING IN MEASURING BIOMASS AND CARBON AT THE LANDSCAPE LEVEL IN SAVANNA SYSTEMS 3.1. INTRODUCTION In many tropical countries, especially in rural areas, there is a reduction of forest areas due to many facto rs including forest clearing for agricultural and ranching purposes, shifting cultivation, removal of dry savanna vegetation and firewood, and agricultural waste burning (Crutzen and Andreae, 1990). Some studies have suggested that disturbance in open for est systems is quantitatively as important as in closed forests (Erickson et al. 2002; Murphy and Lugo 1995; Serca et al. 1998), especially because disturbances in these systems may have longer term degradation effects with lower probabilities of recovery than in closed forest systems. Observations of reforestation and biomass accumulation in trees on agricultural land is important because these data are needed to understand the global magnitude and capacity for carbon sequestration, and to inform decision and policy makers on options for carbon management practices that can remove carbon from the global atmosphere. There is considerable uncertainty on the current land area in woody perennials on farms in developing countries and the global potential for man aging carbon sequestration in tree - based agriculture. Some estimates from international organizations suggest a large amount of carbon sequestration is already occurring in these managed landscapes (Verchot and Singh 2009). Although frequently treated as a secondary priority to monitoring of the much higher biomass forests of the humid tropics, the dry and open forests of the world are more abundant than closed forests and usually are more prone to occupation and disturbance by humans. While these low den sity forests contain as little as 25% of the total carbon stock of humid tropical closed forests, the global area of low density forests is more than 30% 89 greater. Rates of disturbance are suspected to be equal to or greater than the closed forests, and les s likely to recover lost carbon. Thus it is vitally important to begin to assess the global magnitude of open forest disturbances. Additionally, what is often unrecognized is that while forested area is declining in developing countries, tree cover on far ms is increasing, as farmers substitute annual cropland for the tree products which have formerly been available in local forests. Also farmers are increasingly seizing specific market opportunities to sell higher - value tree products (e.g. natural rubber, bio - fuels, bio - chemicals, timber). For example, remote sensing in 64 rural locations in Uganda revealed that between 1960 and 1995, forested area declined 50%, agricultural area increased 23%, and the proportion of agricultural land under tree cover increa sed 22% (Place 2001). Agricultural land now accounts for over double the area To pursue efforts on climate mitigation, global attention is s hifting not only to tropical forests but also to trees outside of forests especially trees on farms. Increasing woody biomass on farms in developing countries is seen as a possible global climate and carbon mitigation option that deserves serious attention . Montagnini and Nair (2004) have estimated that a vigorous program to introduce agro - forestry on farms in tropical Africa and Asia has the potential to sequester 3.5 Mg C ha - 1 yr - 1. However, for this to be a successful strategy it will be necessary to ha ve sound detection and monitoring systems in place. In the Africa context, and unlike in the developed countries, the farm sizes are small and are declining overtime (Jayne et al ., 2003). Although land use based smallholder carbon systems can provide mult iple benefits, they are facing important constraints associated with 90 small - scale measurement. Consequently, small - scale carbon sink projects occupy a limited place in global programs, and investors also have shown little interest in financing such projects (Boyd et al. 2007). It is necessary to explore the possibility of shifting the premise for measurement and monitoring carbon from individual farm plots to a whole landscape and from individual farmers to communities. However, for the time being, accurate measurement of woody biomass in landscapes is facing a significant challenge for efficient, timely and cost - effective automated monitoring of A/R/AF. Detecting and measuring individual tree objects that are sparsely planted or growing on farms, along roadw ays, or in backyards cannot be done accurately with low or moderate resolution satellite imagery where a single tree may represent only a small proportion of a single pixel. Although pixel unmixing techniques (Keshava 2003) can provide an estimate of the p roportion of different land cover types represented by a single pixel, in general, these techniques are not amenable for accurate estimation of tree size, number or other measurements required for accurate assessment of forest cover change in sparsely plan ted landscapes. However, recent technological advancements in satellite image acquisition provide access to hyper - resolution imagery of the Earth and its land uses. The Worldview 2 satellite acquires 4 bands of multispectral data at 2 m resolution and a p anchromatic band at 5 0 centimeters. Very high - resolution imagery enables the detection of objects in the landscape using remote sensing. Subsequent processing can be used to associate attributes, such as size or texture, with detected objects. Therefore, my work is beginning to demonstrate that it is possible to count trees, estimate size and evaluate change remotely. For example, if one can detect individual trees and tree crowns, there may be a possibility to develop a crown - based allometry for individ ual trees (see previou s chapters). 91 This requires the development of methods for ground measurements. Most reforestation projects use small sample plots in a sampling frame for forest inventory. This ground measurement protocol must be re - defined for system s of individually planted trees, and then also extended to the woodland or savanna landscape. The landscape approach allows one to inventory both land in forests (in this case savanna) and land on trees outside of forests (in this case agroforestry). In recent years there has been substantial progress by the research community developing ways to detect land cover change in tropical forests with remote sensing. Remote sensing initially was focusing on measuring the conversion of tropical forests to non - for est land (Skole and Tucker 1993), but recent advances have made it possible to increase the variety of disturbances that can be detected for closed tropical forests to include deforestation, degradation, logging, fragmentation, reforestation, and fire (Mat ricardi et al. 2013, Matricardi et al. 2010; Matricardi et al. 2007; Wang and Cochrane 2005). Thus, there are now methods available to remotely detect a full range of disturbance intensities, from outright clearing to low levels of degradation, over large areas but these approaches have been successful only in very dense closed tropical forests, and in particular in key regions of the Amazon Basin, Central Africa, and South East Asia. There has been very little advancement of detection methods for two impor tant other forms of land cover change: 1) deforestation and degradation of open woodlands such as the cerrado and the chaco ecosystems in South America, woodlands of East Africa, Savannas of West Africa, and other open forest ecosystems in the tropics and sub tropics, 2) regeneration systems on managed landscapes where biomass recovery occurs as plantations, orchards, agroforestry, 92 and widely - spaced tree complexes associated with agriculture. This chapter is focusing on exploring the opportunities of usin g high resolution satellite images to: - detect individual trees in landscapes in which tree cover is sparse or widely spaced and use remote sensing based crown projection area to predict DBH; - s elect and apply an allometric model that estimates biomass o r carbon stocks in A bove ground live biomass (A GLB ) based on accepted standards (ie. IPCC default allometric equations using DBH as inputs) ; - estimate the model developed in chapter 2 which use s remote - sensing derived allometric parameters to estimate ca rbon stocks per tree ; - a pply the model to inventory carbon stocks across an entire savanna landscape dominated by TOF. 3.2. MATERIALS AND METHODS 3.2.1. Study Area The study area is located in two different sites: Sokone and Karang within an area measu ring 6 km x 6 km (36 km 2 ) of two Worldview 2 satellite images. For each 36 km 2 area, a sample grid of 9 blocks (2 km x 2 km) is generated using the ERDAS Imagine software package. Three blocks of 2 km x 2 km representing an area of 400 hectares have been s elected for sampling and data collection in each of the two sites of Sokone and Karang. Therefore, all of the previous calculations to develop the model of predicting DBH from remote sensing crown projected area have been done using data on this area. Also , for the carbon calculation, data for the six blocks (three in each site) are used at the sample level to apply the model to remote sensing and expand it at the landscape level. It cover s an area of 2400 hectares, representing 1200 hectares for each of th e two sites. For a large area 93 application, we extend the area to eighteen blocks (nine in each site), covering an area of 7200 hectares (3600 hectares for each site) (Figure 2.2.). 3.2.2. Data used R emote sensing data are from high resolution satellite im age from Worldview 2 sensor which provides data in eight multispectral bands including; - the coastal blue (400 - 450 nm) which supports vegetation identification and analysis; - the blue (450 - 510 nm) - the green (510 - 580 nm) - the yellow (585 - 625 nm) which assists in the development of "true color" hue correction for human vision representation; - the red (630 - 690 nm) - the red - edge (705 - 745 nm) which aids in the analysis of vegetation condition, directly related to plant health revealed through c hlorophyll production; - the NIR (770 - 895 nm) - the NIR 2 (860 - 1040 nm) which supports vegetation analysis and biomass studies (Digital Globe, 2009). Four (red, blue, green, near infra red) of these bands are standard colors while the other four (red e dge, coastal, yellow, near infra red 2) are new colors. This sensor has a resolution of 0.5 meter panchromatic (B&W) and 2 meters multispectral . The data were acquired for the month of April 16, 2010 and were geo - registered in the Universal Transverse Mercator (UTM) coordinate system (WGS 84, zone 28). Ap ril is a period during which deciduous trees, which constitute the large part of trees outside forest in the study area, bear adequate foliage suitable to provide good remote sensing information as 94 it makes the identification and the distinction of vegetat ion from other land cover features easier. I digitized the crown s of a ll trees within the eighteen blocks and calculated their area and used it to predict their DBH for carbon estimation. I used the Arc Map Editor tool and created circular polygons aroun d the area of the tree crowns. 3.2.3. Data Analysis This sub - section addresses the question of how to measure carbon for trees outside of forests (TOF). Most techniques that use allometric equations to estimate total above ground biomass are using some co mbination of DBH, total height, and wood density as independent variables. For this study, crown projection area measured through remote sensing is used as the independent variable. Data from the remote sensing analysis of individual trees are calibrated w ith ground data. The regression equations developed in chapter 2 established a predictive relationship between crown and tree DBH. Those equations were used to predict DBH and estimate AGB using standard allometric equations that use DBH as the independent variable. To accomplish this task, remote sensing, correlation and regression analyses are combined. 3.2.3.1. Remote Sensing Analysis and manual delineation of trees The complexity of smallholder agricultural systems in tropical areas compared to indust rial agricultural systems requires fine resolution satellite data to allow tree measurements in agricultural landscape. This is possible with the availability of high resolution multispectral satellite imagery where single pixels (0.5 m) can be smaller tha n individual tree crowns, and even for trees as small as 10 cm in diameter at 1.3 m who often can have crown projection areas > 10 m 2 (Milne et al ., 2013). High resolution images from 95 Worldview 2 were acquired for the study area. The selected images were t aken during the dry season (April 16, 2010) to reduce the effect of cloud cover. The image projection is UTM Zone 28 N, WGS 84 datum. To increase both the spatial, temporal and spectral resolution of the image, the Worldview Panchromatic image (0.5 m) wa s pan - sharpened with Worldview Multispectral bands MSS (2m). Individual tree crowns were manually digitized and the area of each tree crown is measured in the satellite imagery. The digitizing is performed using the polygon construction tool in ArcGIS 2010 and the area of each crown calculated and used to determine the relationship with the tree diameter at breast height (DBH) measured in the field. ERDAS and ArcMap software packages are used to perform these tasks. ERDAS Imagine 9.2 software was used for pre - processing the satellite data while ArcGIS 10 helped create and merge the spatial database. The pan - sharpened images were displayed in ArcGIS 10 to perform the manual digitizing and tree crown delineation. 3.2.3.2. Regression Analysis A regression a nalysis is used to estimate the change in the DBH from a given change in the CPA by finding a formula that represents the relationship between these two variables. This information is used to predict DBH from the given CPA by finding an approximate value o f the tree DBH from the value of the CPA using a simple linear function. The regress ion established between these two parameters is used to estimate tree biomass and carbon on a tree - by - tree basis. With this relationship between CPA and DBH developed from ground measurements, the remote sensing delineated CPA is substituted with the ground measurements to cover the larger landscape. 96 3.3. RESULTS In chapter 2, allometric models between tree crown projection area using a manual digitizing of Worldview 2 data and tree DBH measured during field surveys are developed. The model validation through a comparison between predicted and field measured DBH has shown satisfactory results with coefficients of determination of 0.760 and 0.791 respectively for Sokone and K arang sites. Equations 1 and 2 are used to predict DBH knowing the crown projection area from satellite data and estimated biomass and carbon. To measure the carbon on a tree basis, the IPCC general allometric equation (IPCC, 2006) is used. The carbon dens ity and carbon stock in sampled plots is calculated for both Sokone and Karang sites. 3.3.1. Tree density at landscape level To determine the stand densities for each block, remote sensing techniques have also been used, especially for the digitizing of the area of each stratum. After determining the area of strata for each site, the stand densities for each block were calculated from the number of standing trees per ha. Each block has an area of 400 ha but I excluded the area of pasture in some blocks wh ich explains the difference. In the Sokone site:, block SOK 1 with an area of 333 ha, counts 4814 trees, which gives a stand density of 14.4 trees/ha. It is the second block with the highest stand density after block SOK 4 which has a stand density of 17.5 trees/ ha with an area of 369 ha and counting 6470 trees. The third block with the highest stand density is SOK 5 with 10.9 trees/ ha. All other six blocks in the Sokone site have less than 10 trees/ha (Table 3.1) For the Karang site, the two blocks with the highest stand density are KAR 8 and KAR 5 with respectively 13.1 and 11.3 trees/ha. Al other blocks have less than 10 trees/ha (Table 3.1.). 97 Globally, the Sokone site has the highest stand density compared to the Karang site (Table 3.1.). Table 3 . 1 Tree density by block in the study area BLOCKS Block Area (ha) Number of trees Stand density (trees/ha) SOKONE SOK 1 333 4814 14.4 SOK 2 391 3868 9.8 SOK 3 356 2947 8.2 SOK 4 369 6470 17.5 SOK 5 294 3211 10.9 SOK 6 341 3211 9.4 SOK 7 400 2440 6.1 SOK 8 400 1847 4.6 SOK 9 400 1283 3.2 KARANG KAR 1 201 1839 9.1 KAR 2 203 446 2.1 KAR 3 292 1218 4.1 KAR 4 400 1813 4.5 KAR 5 179 2031 11.3 KAR 6 387 1714 4.4 KAR 7 387 2193 5.6 KAR 8 288 3798 13.1 KAR 9 356 1394 3.9 3.3.2. Carbon stock and carbon density at landscape level The following equations developed in chapter 2 are used to predict tree DBH knowing the crown area based on remote sensing; Equation 3 is used for the Sokone site and Equation 2 for the Karang site. DBH = 0.5912 (RS CPA) + 13.408 Equation 2 DBH = 0.3514 (RS CPA) + 24.613 Equation 3 To calculate aboveground bioma ss, the following Equation 6 is used: ABG = exp ( - 1.996+2.32*ln(DBH )) Equation 6 98 Where: ABG = aboveground biomass in kg DBH = diameter at breast height in cm The below ground biomass is calculated by multiplying the above ground biomass by 0.24 which is the root - shoot (R/S) ratio. All values of above and below ground biomass are summed up to count for the total biomass converted into carbon by multiplying the biomass by 0.47. Carbon stock and carbo n densities in both Sokone and Karang are reported in the following Table 3.2. For the Sokone site, the carbon densities vary between 3.9 (block SOK 9) to 17 tC/ha for block SOK which has the highest density. It is followed by block 5 (SOK 5), block 6 (SOK 6), block 2 (SOK 2), block 4 (SOK 4) and block 3 (SOK 3) with respectively 16.4, 12.7, 12.5, 12.4 and 11.3 tC/ha. In addition to the presence of pasture which reduces the area, these high densities are explained by the high density of cashew plantation in these blocks. Table 3 . 2 Carbon stock and Carbon density by block in the study area BLOCKS Area of Block (ha) Carbon stock (t C) Carbon density (t C/ha) SOKONE SOK 1 333 5672.4 17.0 SOK 2 391 4889.8 12.5 SOK 3 356 4019.9 11.3 SOK 4 369 4588.4 12.4 SOK 5 294 4822.7 16.4 SOK 6 341 4315.1 12.7 SOK 7 400 1885.6 4.7 SOK 8 400 2374.0 5.9 SOK 9 400 1568.6 3.9 KARANG KAR 1 201 1543.2 7.7 KAR 2 203 415.9 2.0 KAR 3 292 773.4 2.6 99 Table 3.2 (cont'd) KAR 4 4 00 2535.9 6.3 KAR 5 179 1446.0 8.1 KAR 6 387 1732.6 4.5 KAR 7 387 1318.8 3.4 KAR 8 288 1997.1 6.9 KAR 9 356 1446.8 4.1 3.3.3. Relative contribution of trees by DBH size to the total carbon The classification of trees based of DBH size is helpful in determining the contribution of tree size to the total biomass and carbon. For the Sokone site, Table 3.3. and Figure 3.2. show the contribution of each DBH class to the total carbon for all nine blocks. Table 3 . 3 Relative contribution of trees (tons of C) by DBH size to the total carbon in the Sokone site. DBH Classes (cm) BLOCKS SOK 1 SOK 2 SOK 3 SOK 4 SOK 5 SOK 6 SOK 7 SOK 8 SOK 9 TOTAL 10 - 19.9 0.5 0.1 0.0 2.8 0.1 0.0 0.4 0.0 0.6 4.4 20 - 29.9 102.3 1 5.8 7.5 199.5 17.2 18.6 55.0 4.8 28.4 449.0 30 - 39.9 351.3 178.8 83.6 453.7 123.4 194.1 174.3 71.7 103.8 1734.7 40 - 49.9 581.4 602.3 423.6 843.2 439.6 458.6 313.1 297.5 123.9 4083.2 50 - 59.9 497.1 648.0 522.9 786.6 639.2 606.3 387.2 325.8 116.9 4530.0 60 - 69.9 651.5 661.1 558.8 551.1 511.0 552.9 164.7 284.5 166.9 4102.3 70 - 79.9 400.6 567.8 527.9 508.2 507.3 344.3 252.4 309.5 126.6 3544.4 80 - 89.9 266.9 346.5 413.3 328.4 351.5 155.9 107.4 219.0 135.5 2324.4 90 - 99.9 673.5 211.4 227.6 254.5 197.6 148.6 88 .3 115.2 130.5 2047.3 100 - 109.9 318.7 109.9 119.2 77.0 127.5 115.1 43.5 82.4 60.8 1054.1 110 - 119.9 83.7 132.5 147.5 115.8 148.6 57.6 55.7 51.3 65.5 858.2 120 - 129.9 281.6 43.6 96.8 107.9 29.6 34.7 36.1 36.9 18.6 685.9 130 - 139.9 196.4 39.7 59.9 87.6 108.7 68.6 35.0 40.8 62.4 699.1 140 and < 1266.8 1332.4 831.2 272.0 1621.5 1559.9 172.66 534.56 428.27 8019.5 TOTAL 5672.4 4889.8 4019.9 4588.4 4822.7 4315.1 1885.6 2374.0 1568.6 34136.7 Percentage 16.6 14.3 11.8 13.4 14.1 12.6 5.5 7.0 4.6 Except for block 7, 8 and 9 (SOK 7, SOK 8, and SOK 9) which have the lowest amount of carbon, all other blocks hold at least 10 %. with the highest percentage for SOK 1 (16.6 %) followed by SOK 2, SOK 5, SOK 4, SOK 6 and SOK 3 with respectively 14.3, 14.1, 13.4, 12. 6 and 11.8 %. 100 Figure 3. 1 shows more details about the percentage of each DBH class to the total carbon for the whole nine blocks. 54 % of carbon is hold by trees with 80 cm DBH or less and 66.8 % of the carbon is hold by trees with less than 100 cm DBH. Al though they r epresent on ly 1.14 % of the total tree density , big trees (140 cm DBH or more) hold almost a quarter (23.5 %) of all carbon. This is explaining by the very large DBH of some trees. In the Sokone site, many trees have more than 200 cm DBH. Fi gure 3 . 1 Percentage contribution to total carbon by DBH classes in the Sokone site Results from the Karang site also confirm this tendency of trees with less than 80 cm DBH to contain most of the biomass ( Table 3.4. and Figure 3. 2 ). The quantity of biomass contained in this range of trees is 70.6 % for the whole nine blocks in the Karang site. 87.4 % of the carbon is hold by trees with less than 100 cm DBH. 0 5 10 15 20 25 0.0 1.3 5.1 12.0 13.3 12.0 10.4 6.8 6.0 3.1 2.5 2.0 2.0 23.5 Contribution to total carbon DBH classes 101 Table 3 . 4 Relative contribution of trees (tons of C) by DBH size to the total carbon in the Karang site. DBH Classes (cm) BLOCKS KAR 1 KAR 2 KAR 3 KAR 4 KAR 5 KAR 6 KAR 7 KAR 8 KAR 9 TOTAL 20 - 29.9 0.8 0.0 1.0 0.4 2.6 1.0 8.3 12.9 1.9 28.9 30 - 39.9 1 31.7 31.4 135.0 69.2 185.9 85.2 289.1 619.2 145.1 1692.0 40 - 49.9 493.5 116.7 291.3 332.9 477.7 360.8 403.0 712.5 207.5 3395.9 50 - 59.9 239.6 53.8 110.2 346.9 255.4 336.7 208.3 138.2 154.0 1843.2 60 - 69.9 115.4 30.3 66.0 292.8 128.2 212.8 127.9 58.0 72.7 1 104.1 70 - 79.9 66.5 25.2 40.3 255.6 101.8 171.0 72.1 65.5 218.8 1016.9 80 - 89.9 34.3 14.1 26.9 136.3 34.5 75.7 11.8 16.7 78.7 429.0 90 - 99.9 71.9 17.6 28.7 156.2 60.9 78.4 44.3 23.4 174.0 655.3 100 - 109.9 40.3 11.9 7.6 130.9 55.5 84.6 36.4 11.9 86.1 46 5.2 110 - 119.9 8.9 4.5 4.5 32.2 14.3 56.6 4.9 11.9 94.7 232.5 120 - 129.9 39.1 22.2 11.0 97.0 22.4 34.3 21.8 5.4 49.8 303.0 130 - 139.9 22.0 22.0 0.0 78.4 0.0 43.8 13.1 0.0 19.9 199.2 140 and < 279.0 66.2 50.9 607.1 106.8 191.7 77.7 321.5 143.6 1844.3 T OTAL 1543.2 415.9 773.4 2535.9 1446.0 1732.6 1318.8 1997.1 1446.8 13209.6 Percentage 16.9 13.6 12.2 16.5 12.3 10.5 6.6 7.0 4.4 L ike in the Sokone site, only three blocks (KAR 7, KAR 8, and KAR 9) have less than 10% of the total carbon for the whole sit e. KAR 1, KAR 4, KAR 2, KAR 5 and KAR 3 hold the highest percentage of carbon with respectively 16.9, 16.5, 13.6, 12.3 and 12.2 % of the total carbon. More details about the percentage of each DBH class to the total carbon for the whole nine blocks are sho wn is Figure 3. 2. Big trees with more than 140 cm DBH contribute to 14% to the total carbon. Like in the Sokone site, this is due to th e very large DBH of some trees. 102 Figure 3 . 2 Percentage contribution to tot al carbon by DBH classes in the Karang site In both Sokone and Karang, there is a variability of trees based on their DBH within the 9 blocks which reveals that the most densely populated of the blocks are those dominated by the cashew ( Anacardium occid entale ) species . 3.3.4. Comparison of remote sensing - based biomass and carbon estimation with field data In this section, d ata from sampled plots within the 6 blocks (3 for each site) is used to do a comparison with the remote sensing model - based data for the same blocks. Biomass and carbon for each tree are calculated using the field measured DBH for both Sokone and Karang sites. T o determine the carbon stock and carbon density for ea ch of the six selected blocks (3 for each site) of the two sites, these d ifferent steps are followed: A - Determine the total area of all sampled plots for each stratum B - Determine the percentage of each stratum for each block C - Calculate the quantity of carbon per tree in each plot D - Calculate the total carbon per plot by summing up a ll carbon/tree within the plot 0.0 5.0 10.0 15.0 20.0 25.0 30.0 0.2 12.8 25.7 14.0 8.4 7.7 3.2 5.0 3.5 1.8 2.3 1.5 14.0 Contribution to total carbon (%) DBH classes 103 E - Calculate the carbon stock in all sampled plots for each stratum by summing up the amount of carbon per plot for all plots of the same stratum F - Calculate the carbon density by stratum of each block by multiplying (E) by the area of that stratum within the block G - Calculate the carbon stock within each block by summing up all carbon stock in all strata within each block H - Calculate the carbon density for each stratum by dividing (F) by the area of each stratum I - Calculate the carb on density of each block by dividing its carbon stock by the corresponding area The calculation of the carbon stock and density estimates for a larger area is done by first extrapolating to the full hectare. An expansion factor determined by calculating t he proportion of a hectare that is occupied by a given plot in the study area (P earson et al., 2007) is used. In both sites, the area of pastures is excluded for the biomass and carbon estimation as this land cover t ype is not part of this study. T he area of dwellings (villages) is included with area of TOF in agricultural lands because individual trees within the villages are easily detectable and digitized to allow an application of the model. Comparison between field based and remote sensing based car bon estimates shows that values of the field based carbon are higher than the model - based carbon for all blocks (Table 3.5.). The difference in carbon density estimation varies between blocks. For KAR 1, KAR 8 and KAR 9 blocks in the Karang site, the diffe rence represents respectively 27.3, 30.6 and 35.1 %. 104 This difference is reflected in the carbon density with lower values for the remote sensing based data. Table 3 . 5 Comparison between remote sensing base d and field based carbon stock and carbon densities BLOCKS Area of Block (ha) Remote sensing based Field - based Carbon stock (tC) Carbon density (tC/ha) Carbon stock (tC) Carbon density (tC/ha) KARANG KAR 1 201 1543.2 7.7 2124 .0 10.6 KAR 8 288 1997 .1 6.9 2876.2 10.0 KAR 9 356 1446.8 4.1 2229.2 6.3 Analysis of field data allows a comparison of carbon density by land cover. A fter estimating carbon stock and carbon density at the plot level and knowing the area of each stratum within the block, the carbon density for each stratum is multiplied by the area to estimate the carbon stock for the whole block. For the Sokone site, the values of carbon stocks are higher for block SOK 1 with 9286.1 tons of carbon and decrease towards the south for block S OK 4 (6207.7 tons of carbon) and block SOK 7 (2950.2 tons of carbon) (Table 3.6.). This decrease is more pronounced for the cashew plantation where the densities vary between 78.6 t C/ha in SOK 1 in the northern part to 34.6 tC/ha for SOK 7 in the southern . The highest carbon densities are within the area of cashew plantation and can reach 78.6 t C/ha for SOK 1. This can be explained by the long tradition of cashew plantation in this area with the Senegalo - German Cashew Project (PASA) back to the 1980s. 105 Table 3 . 6 Carbon stock and carbon densities in the six blocks of the Sokone and Karang sites where sampled plots were located B LOCKS Cashew plantation TOF in agricultural land TOTAL Area (ha) Carbon stock (t C) Carbon density (t C/ha) Area (ha) Carbon stock (t C) Carbon density (t C/ha) Area (ha) Carbon stock (t C) Carbon density (t C/ha) SOKONE SOK 1 96 7548 78.6 237 1738.1 7.3 333 9286.1 27.9 SOK 4 116 5031.2 43.4 253 1176.5 4.7 369 6207.7 16.8 S OK 7 44 1523.1 34.6 356 1427.1 4.0 400 2950.2 7.4 KARANG KAR 1 30 1158 38.6 171 966 5.6 201 2124 10.6 KAR 8 67 1559.6 23.3 221 1316.6 6.0 288 2876.2 10.0 KAR 9 23 707.7 30.8 333 1521.5 4.6 356 2229.2 6.3 For the Karang site, results of the carbon sto ck at the plot level scaled up at the block level are similar to those for the Sokone site regarding the predominance of cashew plantations which explains the higher values of carbon stock for that stratum. Block KAR 8 that has the largest area of cashew h olds 1559.6 t of carbon which represent s 54% more than Block KAR 9 with only 707.7 t C. In the meantime, with the larger area of other TOF in agricultural land (333 ha), Block KAR 9 holds the highest carbon stock for this stratum with 1521.5 t C while Bloc k KAR 1 with an area of 210 ha holds only 966 t C (Table 3.6.). 3.4. DISCUSSIONS AND CONCLUSION : Measuring carbon in trees outside of forests is very challenging especially in savannas systems due to the high spatial variability caused by natural but mos tly human disturbances events (Lyster et al., 2013). Using remote sensing predicted crown projection area (CPA), we estimate biomass and carbon per tree for the whole eighteen (18) blocks of the study area in both Sokone and Karang sites. 106 Direct measurem ents using field inventories have been used to estimate biomass and carbon but present some challenges: it is time consuming, expensive labor intensive and difficult to implement in remote areas (Kuya et al., 2012). Our results have shown that these challe nges can be overcome by using remote sensing techniques. With these techniques, I was able for example, to determine the number of trees for each of the blocks and calculate the stand densities (trees/ha). Using a remote sensing - based model, tree DBH is predicted knowing the crown projection area and run an allometric equation to calculate the carbon on a tree basis. In each of the blocks of the study area, we estimated the carbon stock and the carbon density. The carbon densities vary between 17 t C/ha which is the highest (SOK 1) to 2 t C/ha the lowest (KAR 2). The classification by DBH of all trees in each block is used to determine the contribution of trees by DBH class to the total carbon. Unlike a common assumption that big trees hold the most amoun t of carbon, t he classification by DBH class shows that most of the carbon is on trees with small and medium size DBH : 66.8 % of the carbon for the Sokone site and 87.4 % for the Karang site is concentrated on trees with less than 100 cm DBH. 107 REFERENCES 108 REFERENCES Boyd, Emily; Maria Gutierrez and Manyu Chang. (2007). Small - scale forest carbon projects: Adapting CDM to low - income communities. Global Environment Change 17 (2) , 250 - 259. Crutzen Paul J. and M.O. Andreae . (1990). Biomass burn ing in the tropics: impact on atmospheric chemistry and biogeochemical cycles. Science 250 (4988) , 1669 - 1678. FAO. (2006). Global forest resources assessment 2005. Main Report, FAO Forestry Paper. Rome: FAO. IPCC. (2007). Good Practice Guidance for Land U se, Land Use Change and Forestry. Jayne, T.S., Takashi Yamano; Michel T. Weber; David Tschirley; Rui Benfica; Antony Chapoto; Ballard Zulu. (2003). Smallholder income and land distribution for poverty reduction strategies. Food Policy 28 , 253 - 275. Kesha va, N. (2003). A Survey of Spectral Unmixing Algorithms. Lincoln Laboratory Journal, Vol 14, Nm 1 , 55 - 78. Kuyah, Shem; Johannes Dietz; Catherine Muthuri; Ramni Jamnadass; Peter Mwangi; Richard Coe; Henry Neufeldt. (2012). Allometric equations for estimat ing biomass in agricultural landscapes: I - Aboveground biomass. Agriculture, Ecosystems & Environment, Vol. 158 , 216 - 224. Lyster, Rosemarie; Mackenzie, Catherine and McDermott, Constance. (2013). Law, Tropical Forests and Carbon: The Case of REDD +. Camb ridge: Cambridge University Press. Matricardi, E.; Skole, D.; Cochrane, M.; Pedlowski M. and Chomentowski, W. (2007). Multi temporal Assessment of Selective Logging in the Brazilian Amazon Using Landsat Data. International Journal of Remote Sensing 28 , 63 - 82. Matricardi, E.; Skole, D.; Pedlowski, M. and Chomentowski, W. (2013). Assessment of forest disturbance by selective logging and forest fires in the Brazilian Amazon using Landsat data. International Journal of Remote Sensing 34 , 1057 - 1086. Matricar di, E.; Skole, D.; Pedlowski, M.; Chomentowski, W. and Fernandes, L. (2010). Assessment of tropical forest degradation by selective logging and fire using Landsat imagery. Remote Sensing of Environment 114 , 1117 - 1129. Milne, Eleanor; Henry, Neufeldt; Tod d Rosenstock; Micke Smalligan; Carlos Eduardo Cerri; Daniella Malin. et. al. (2013). Methods for the quantification of GHG emissions at the landscape level for developing countries in smallholder contexts. Environmental Research Letters 8 , 1 - 10. 109 Montagni ni, F. and Nair, P.K.R. (2004). Carbon sequestration: an underexploited environmental benefit of agroforestry systems. Agroforestry Systems, 61 , 281 - 295. Pearson, T.R.H. et al. (2007). Measurement Guidelines for the Sequestration of Forest Carbon. NewTown Square: USDA, Forest Service. Place, Frank; Joe Ssenteza and Keijiro Otsuka. (2001). Customary and Private Land Management in Uganda. In K. O. Place, Land Tenure and Natural Resource Management: A Comparative Study of Agrarian Communities in Asia and Afri ca (pp. 195 - 233). Baltimore and London: The John Hopkins University Press. Skole, David and Tucker, C. (1993). tropical deforestation and habitat fragmentation in the Amazon: Satellite data from 1978 to 1988. Science 260 , 1905 - 1910. Verchot, Louis V. & Singh V.P. (2009). Carbon Sequestration Opportunities with Smallholder Communities: Forestry, Agriculture and Agroforestry. 4th World Congress on Conservation Agriculture, February 4 - 7. New Delhi: World Congress on Conservation Agriculture. Wang, C; J. Qi and Cochrane, M. (2005). Assessment of Tropical Forest Degradation with Canopy Fractional Cover from Landsat ETM+ and Ikonos Imagery. earth Interactions 9 , 1 - 18 . 110 CHAPTER 4 . LANDSCAPE BIOMASS ESTIMATION AND CARBON MAPPING 4.1. INTRODUCTION I n many developing countries, especially in rural areas, forest areas are decreasing while in the agricultural landscape trees are still present and in some part in large amounts. This is very common in some countries where forest are converted to other lan d uses with increasing density of human population ( De Foresta et al ., 2013) There are numerous landscape - wide deforestation and carbon measurement protocols for forests, particularly closed forests (Matricardi et al. 2010). However, there are few method s available for sparsely wooded systems, such as savannas, and virtually none for systems of trees outside of forests that can be used across whole landscapes. The current methods for landscape level carbon measurement are particularly problematic and are unlikely to meet scientific and rigorous standards (for example the requirements of the CDM Executive Board) and potential carbon project investors. Therefore, there is a lack of confidence in existing measurement systems with implications that go beyond t he measurement of carbon credits and create doubts about the project viability. New advances in remote sensing of canopies and individual tree crown detection offer great promise Currently, there is a growing interest in investing in agroforestry systems for this dual benefits of climate mitigation and livelihoods enhancements (ICRAF, 2008), and also adapt to, adverse impacts of changing climate conditions (Verch ot, 2007). In this context it is possible to view agroforestry as both a mitigation and adaptation option for small holders. Agroforestry practices can be applied across a variety of forms and range of intensities depending on local conditions to provide a locally relevant approach to sustainable 111 agriculture production and soil fertility management. Agroforestry might become a key od and non - food subsistence needs through its ecosystem services provided (Jonsson et al . , 1999). There is also some evidence to suggest that the high production levels and economic values of agroforestry value chains may actually facilitate the production of financial capital beyond subsistence levels alone, thereby aiding in capital accumulation and re - investment at the farm level. In Africa one of the more critical characteristics of the rural landscape is the preponderance of many small holder farming systems that are operating on the environmental and economic margins, where any newly introduced practice will be adopted only if it improves food security (and incomes) on a sustainable basis. Under such conditions, climate mitigation interventions, such as a increased carbon stocking with trees on farms, will need to demonstrate support for improved food production (and income) as well (Smith, 2012). In this context, climate mitigation becomes a co - benefit of a primarily food - focused enterprise. Hence, p ractices that minimize the rate of soil degradation, improve condi tions for climate adaptation in many of the poor rural communities that might be most adversely affected by climate change and variability (Neupane, 2001), and then also provide added benefits for climate mitigation through increased greenhouse gas remov als from the atmosphere. Agroforestry readily bundles both mitigation and adaptation strategies and provides several pathways to a range of environmental and social co - benefits and outcomes, including food security, increased farm 112 income, restoration and maintenance of above - ground and below - ground biomass and biodiversity, reestablishment of biological corridors between protected forests, maintenance of watershed hydrology, improved soil conservation, a vailability of timber and fuel wood, and ultimately a reduction of pressure on natural forests outside the farming domain (Pandey, 2002) Many accepted methods for carbon measurement in programs or projects that are based on sequestration (cf. as opposed to avoiding emissions from deforestation) have been defined at a project scale with specified boundaries of the project, often on land parcels with a single owner. When trying to develop methods for rural poor farmers, experience suggests that a landscape approach is needed, whereby many farmers can be bundled into a single project, even if the parcels are not contiguous. The landscape approach offers new opportunities for the participation of many more of the rural poor in carbon markets, but the means for landscape - scale measurement need s to be developed. Remote sensing offers a means to do landscape scale measurement and monitoring, but most advances methods in recent years have focused on closed forests, rather than open woodlands and savannas or, to be sure, systems of trees outside of forests. This chapter is dedicated to estimate biomass and carbon at a landscape level using remote sensing techniques and producing carbon maps of the study area. The model applied combines advanced remote sensing and moderate to high correlation between remote sensing crown projection area (CPA) and tree DBH enables us to predict DBH knowing the CPA through remo te sensing. The equation from that correlation is used 113 to scale up the calculation across a larger area in the semi - arid savanna ecosystems in Senegal (West Africa). 4.2. MATERIALS AND METHODS 4.2.1. Study Area The study area is presented in chapter 3: n ine blocks are selected for each of Sokone and Karang sites. Those eightee n blocks are shown in Figure 2.2. 4.2.2. Data used The different data used include: - land use and land cover data. I classified and digitized the area based on the different strat a and determine the area of cashew plantation, the area of pasture and the area occupied by agricultural lands which hold the other TOF. These data is used to calculate the carbon stock and the carbon density. These land use and land cover data are from Wo rldview 2 high resolution satellite image; - the biomass and carbon data are obtained using allometric equation and predicted DBH from remote sensing Crown Projection Area (Chapter 3) . 4.3. RESULTS 4.3.1. Landscape application of the model The stratifica tion of the study area reveals a large area of pasture in both sites; which reduces the total area where the model is applied. For the nine blocks of the Sokone site, 8.76 %, representing 315 ha are permanent pasture and 16.03 % are occupied by cashew plan tations. The area occupied by agricultural lands and where TOF are implanted, represents 75.22 % of the study area with 2708 ha (Table 4.1.). 114 For the Karang site, the 833 ha of permanent pasture represent 23.14 % within the nine blocks. At the same time, the area of cashew represents only 5.71 %. The agricultural lands occupied a large area of 2561 ha, representing 71.13 % study area in this site. Table 4 . 1 Remote sensing stratification of the eighteen block s in Sokone and Karang sites BLOCKS Area Cashew plantation(ha) Area O ther TOF (ha) Area of pasture Total Area (ha) SOKONE SOK 1 96 237 67 400 SOK 2 92 299 9 400 SOK 3 58 298 44 400 SOK 4 116 253 31 400 SOK 5 56 238 106 400 SOK 6 43 298 59 400 SOK 7 44 356 0 400 SOK 8 59 341 0 400 SOK 9 26 374 0 400 TOTAL 57 7 270 8 315 3600 KARANG KAR 1 30 171 199 400 KAR 2 4 199 197 400 KAR 3 15 277 108 400 KAR 4 11 389 0 400 KAR 5 15 164 221 400 KAR 6 18 369 13 400 KAR 7 30 357 13 400 KAR 8 67 221 1 12 400 KAR 9 23 333 44 400 TOTAL 20 6 2561 833 3600 Although the study area is dominated by the scattered individual TOF in terms of area, the area occupied by cashew plantations is denser in terms of population and crown cover. This situation is confi rmed by the results in the nine blocks ( SOK 1, SOK 4 and SOK 115 7 for Sokone and KAR 1, KAR 8 and KAR 9 for Karang) where the sampling is done . These block s are highlighted in Table 4.1. 4.3.2. Carbon density per block in the study area The carbon density fo r each of these blocks is calculated and results are presented in Figure 4.1 . Block SOK 1 in the Sokone site has the highest carbon density with 17 t C/ha followed by blocks SOK 5 (16.4 t C/ha), SOK 6 (12.7 t C/ha), SOK 2 (12.5 t C/ha) and SOK 4 (12.4 t C/ ha). Figure 4 . 1 Carbon density (t C/ha) per block in the Sokone site Except for blocks SOK 7 (4.7 t C/ha), SOK 8 (5.9 t C/ha) and SOK 9 (3.9), the lowest carbon densities are found in the Karang site wit h block KAR 5 having the highest carbon density (8.1 t C/ha) for this site. 0 2 4 6 8 10 12 14 16 18 SOK 1 SOK 2 SOK 3 SOK 4 SOK 5 SOK 6 SOK 7 SOK 8 SOK 9 17 12.5 11.3 12.4 16.4 12.7 4.7 5.9 3.9 BLOCKS Carbon density (t C/ha) 116 Figure 4 . 2 Carbon density (t C/ha) per block in the Karang site 4.3. 3 . Landscape carbon mapping of TOF To develop a carbon map, t he carbon stock is calculated on a tree basis. The DBH of each tree identified and digitized on the image is predicted using Equation 1 for the Sokone site and Equation 2 for the Karang site. After predicting the DBH based on the remote sensing based Crown Projection Area, the IPCC general allometric equation for tropical dry forest is used to calculate biomass and carbon for each tree. The frequency of the different types of trees and their size are determined all across the landscape. T he following carbon map s were computed and show high and low biomass based on the type and size of trees. 0 1 2 3 4 5 6 7 8 9 KAR 1 KAR 2 KAR 3 KAR 4 KAR 5 KAR 6 KAR 7 KAR 8 KAR 9 7.7 2 2.6 6.3 8.1 4.5 3.4 6.9 4.1 Carbon density (t C/ha) BLOCKS 117 Figure 4 . 3 Carbon in the Sokone site A carbon map of all nine blocks in the Sokone site are shown in Figure 4. 3 . M ost of the carbon is concentrated in the cashew plantation. B lock 7 (SOK 7) presented in Figure 4.4. is a good example . 118 Figure 4 . 4 Carbon map for block 1 (SOK 1) in the Sokone site This block is 2x2 km which is an area of 400 ha but the model was applied for only 333 ha as there are 67 ha of pasture. These 333 ha have 5672.4 t of carbon stock which represent a carbon density of 17 t/ha. Most of this carbon is hold by cashew plantation which occupied 96 ha although this represents only 28.8 % of the area studied. Trees with low carbon content represented in green and yellow are dominant. They have between 0.049 and 5.33 t C/tree. These values are lower in block SOK 7 where there is few cashew plantation s . Trees with the lowest 119 carbon content have between 0.09 to 2.94 t C/tree (Figure 4.5). Only 11 % (44 ha) of this block area is occupied by cashew plantation. Figure 4 . 5 Carbon map for Block 7 (SOK 7) in the Sokone site Although there is no pasture in this block and the model is applied to the whole 400 ha, the carbon density is very low compared to block 1 (SOK 1). The carbon stock is reported to be 1885.6 t C representing a carbon density of 4.7 t C/ha. This is explained by the dominance of trees with low carbon content, especially small cashew plantation and other scattered TOF in agricultural land. They mostly hav e between 0.09 to 2.94 t C per tree (Figure 4.5.). 120 Figure 4 . 6 Carbon stock in the Karang site The carbon densities are lower in the Karang site compared to the Sokone site As shown in Figure 4.6 , the highest values are for cashew plantation. Like in the Sokone site, the whole landscape is dom inated by trees with less than 3 t C/per tree (Figure 4.7). The rare trees that have more than 3 t C are located in the agricultural land. (Figure 4.8, Figure 4.9 and Figure 4.10). 121 Figure 4 . 7 A portion o f a cashew plantation carbon map in the Karang site 122 Figure 4 . 8 Carbon map of individual TOF in agricultural land in block 9 (KAR 9) in the Karang site 123 Figure 4 . 9 Carbon map in a low density block (KAR 2) in the Karang site 124 Figure 4 . 10 Carbon map in a low density block (KAR 9) in the Karang site 4.3. 4 . Comparison with other studies Our results sh ow that carbon densities vary depending on whether or not there are cashew plantation s , their density and their age; which explains why values in the Sokone site are higher than those in the Karang site. Carbon densities ranged from 3.9 t C/ha in block 9 ( SOK 9) to 17 t C/ha for block 1 (SOK 1) while the highest carbon density in the Karang site is 8.1 t C/ha for block 5 (KAR 5) and the lowest is block 2 (KAR 2) with only 2 t C/ha. T hese estimates are compared with results from other studies : 125 - the first on e is from Western Kenya where Kuyah et al. (2012) provided results with close agreement with our own. Using a destru ctive sampling method, they estimate the carbon density to be 20.8 t / ha. It is important to acknowledge at this point that field data are more accurate than model - based estimation, which can be a factor that explains the difference between the two results. Another factor is that the two studies are from different sites , although they are all carried out in the tropical zone with some similar ities . - the second study is by Baccini et al. (2012) . Using the exact same area in Senegal , I compare their carbon density to my own results. The two maps of the comparison are presented in Figure 4.11 and Figure 4.12. Baccini et al . (2012) found that the values of aboveground live biomass are between 1 0 and 2 2 megagrams per ha. Converting the biomass into carbon (as 0.47 units C per unit biomass), I obtained values ranged between 4.7 and 1 0 . 3 t C/ ha which are very close to my results in some places of th e study area . However, it should be noted that Baccini et al. used a "single average AGB value for each land cover class" which "introduces pseudo - replication, as multiple pixels containing the same land cover class will be given identical ABG values" (Mi tchard et al ., 2011) no matter the number of trees and the variation in size. Each of the different pixel s represents the sum of the above - ground carbon for all trees in the pixel. That does not help to make a good comparison with our study because our dat a are individual tree bas ed . Also, the pixel size ( 463.3 x 463.3) of the MODIS image used by Baccini et al. covered a large area with high heterogeneity. For example, block 1 (SOK 1) where we found the highest carbon density (17 t C/ha) is the block where Baccini et al . reported more pixels with lower values although it is an area of dense cashew plantation. Even i n some places in my study area, they found no carbon while my results show a high potential of carbon . 126 Figure 4 . 11 Carbon map in the Sokone site based on Baccini et al. (2012) 127 Figure 4 . 12 Carbon map in the Sokone site based on our model 4.4. DISCUSSIONS AND CONCLUSION The model developed in chapter 2 is used to estimate biomass and carbon at landscape level. The carbon calculated using allometric equation on a tree basis is extrapolated on a plot and block level to cover large area. Combining remote sensing techniques and basic allometry in systems of "trees outside of forest" the carbon density in the study area is estimated and carbon maps produc ed. Despite the limitations of the study, o ur results are close to those from other studies (Baccini et al . 2012; Kuyah et al .; 2012). As averages , it should be noted for our results that, 128 the carbon densities are higher in some area such as in block 1 (SOK 1) in the Sokone site where the density is 17 t C/ha. These values are consistent with those of Kuyah et al. (2012) and Baccini et al. (2012) an d this c onsisten cy provides more confidence about our remote sensing model but the values are still low and various factors can be used to explain this situation. Those factors include: - Most of the study area is permanent pasture and this stratum was not surveyed as it was not included by the protocol. As this study is focusing on trees outside of forest, most of the and I classified them as forest; - This study did not also take into account the other carbon pools including understory vegetation, shrubs and grasslands as well as trees with less than 10 cm DBH. - Error of estimation: omission of trees during the digitizing process especially for cashew plantations trees are intermingled ; which make difficult the delimitation of tree crown boundaries . 129 APPENDICES 130 Appendix B : TECHNICAL CONTRIBUTIONS THAT CAN BE MADE BY COMMUNITY PARTICIPATION: TOWARD COMMUNITY BASED CARBON MEASUREMENTS B .1. INTRODUCTION There is a rich body of literature on participatory or community - based monitoring of natural resources (e.g. Evans and Guariguata 2008). Although this literature, and the field that it embraces, are important, this chapter does not involve an analysis of participatory monitori ng. The aim of this chapter is to assess communities' ability to accomplish carbon measurement of trees on their own farms. By asking the basic question as to whether local people can work closely with technical and professional experts to deliver accurat e measurements, that in turn , support accepted rigorous international standards, is not participatory monitoring in its richest, or its fullest extent. As such, we recognize the limitations of this study to draw any broad conclusions about modalities or po tentialities for long term sustained engagement of local communities, which is a full treatise beyond the rather narrow measurement scope of this study. Having noted these limitations, this chapter is cognizant of the fact that there is a subject of curre nt disagreement in the use of advanced technology by local communities , ( Fry , 2011) points this out in his review of community forest monitoring potentials for REDD. Herold and Skutsch (2009), Skutsch et al . (2009), Abrell et al . (2009) and others have ad vocated the use of technologies by communities, such as GPS and GIS mapping while others (Danielson , 2005) remain less convinced. Similarly, there is some question as to whether local people can make measurements that are accurate and precise enough to be acceptable for scientific application (Evans and Guariguata 2008). Fry (2011) provides an organization structure to evaluate community measurement that includes accuracy as one of 131 three elements, along with cost and sustainability and cultural relevance. T here may be problems of accuracy and precision when non - professionals engage in scientific activities (Danielsen et al. 2005). Other studies have focused on the capacity of local communities to assess forest biodiversity and forest disturbance. For example Holck (2008) mentioned the increased focus of international forest policies in involving local communities in forest monitoring and management; this involvement being considered as a strategy to improve biodiversity conservation efforts and local livelih ood s especially in developing countries. Also, the high level of commitment of village managers in monitoring resource extraction and disturbance has been studied by Topp - Jorgensen et al. (2005). Case studies confirm that locally based, participatory mon itoring must be simple to be successful: but how simple can the monitoring systems be and still maintain scientific validity? Danielsen et al. (2005) conclude that there is a major gap in understanding the comparability of data between scientifically and l ocally - collected data. Based on the few comparisons of scientific and local monitoring that have been conducted, the minimum amount of data to be collected by local monitoring programs to generate the same results as scientific methods appears to be high ( Evans and Guariguata 2008). As an example, Larrazàbal et al. (2012) have demonstrated from the Kyoto Think Global Act Local ( KTGAL ) project's example that communities are capable of generating data that can meet the standards of the IPCC methodology. They found the data accurate and reliable as "there were no significant differences in the estimate of mean stock or in the confidence level the communities" (Larrazàbal et al ., 2012). Also, for the time being, the importa nce of National Forest Inventory data to measure and monitor forest carbon is well documented. Getting local communities being familiar 132 with tree data collection methods and techniques could help achieve the goals of National Forest Inventory in lower ing c ost and the data collected could reasonably be used for forest carbon assessment. In the Global Observatory for Ecosystem Services laboratory at Michigan State University and through the Carbon2Markets model, the approach to measurement, reporting and ver ification systems used requires the participation of local communities in conducting the protocols and becoming collaborators in the team and the operation of the system. There has been on - going work with communities in Thailand, Vietnam, and Laos PDR whic h suggests that communities can be engaged as co - measurers using the full suite of rigorous international protocols and standards for measurement, particularly for the ground segments (Samek et al. 201 4 ) but these have been experienced in closed forest, pl antation systems, small - holder wood lots and agro - forestry in Asia but not yet in savanna environments of Africa. This project aims to gather more information and experience in a different measurement setting on how accurate and precise local communities c an measure simple ground parameters and play a partnership role in data collection. This project recognizes that just because local land owners could (or not) make accurate contributory measure ments does not mean they would or should, nor will it provide c omplete understanding of what conditions are necessary for full participatory engagement, as these issues are beyond the scope of this measurement study. Some of the limitations of the aforementioned work of the GOES Lab are that these projects all deal wi th forests or forest stands small holder plantations, community forests, or REDD forests and do not address the application of the Carbon2Markets MRV in systems of trees outside of forests nor agroforestry systems in predominantly agricultural landscap es. There is a need to explore the effectiveness, accuracy 133 and precision of joint measurement campaigns in savanna landscapes and agroforestry systems outside forests. That is the aim of this chapter. B .2. MATERIALS AND METHODS B .2.1. The Study Sites The study is carried out in thirteen Senegalese villages located within 190 km from the capital (Dakar) in the Fatick Region. The study area covered mainly the three rural communities of Toubacouta, Nioro Alassane Tall and Keur Samba Gueye. The thirteen villag es are selected purposively us ing 3 criteria in order to capture a maximum of social, cultural, agricultural and ecological variation that exists in this area. The first criterion is related to the social and cultural side and includes ethnic groups ( Wolof , Serer, Bambara, Mandinka, Diola ). The second criterion based on the agricultural and ecological variation includes the soil and vegetation types and the presence of cashew plantation in the village. The third criterion is related to the village size and the distance from the nearest road. For accessibility purpose by the researchers, the village must be located at less than 5 km of the main road. In the rural community of Toubacouta, the majority of the population is dominated by the Mandinka (50%), Serer s (35%), Wolofs (5%) ethnic groups and other minority ethnic groups including Bambara and Poular (T able B .1. ). In the rural community of Keur Samba Gueye, the population is dominated by Wolofs (60%), Mandinka (20%), and 20% represented by the Serers, Diola s and Poular ethnic groups. 134 Table B .1. Demographic characteristics of selected villages No. Villages Rural Community Dominant Ethnic Group Total Population Total Households 1 Boutilimith Toubacouta Mandinka 192 17 2 Dassilamé Socé Toubacouta Mandinka 1 061 100 3 Diaglé Toubacouta Serer 1 163 166 4 Ndoffane Toubacouta Mandinka 252 31 5 Senghor Nioro A Tall Bambara 391 42 6 Kouatine Nioro A Tall Serer 213 23 7 Niokholokho Nioro A Tall Serer 289 34 8 Ngouye Mary Nioro A Tall Wolof 253 27 9 Keur Bakary Camara Keur S Gueye Bambara 235 24 10 Keur Samba Guèye Keur S Gueye Wolof 1 378 186 11 Keur Samba Nosso Keur S Gueye Wolof 1 079 92 12 Keur Sett Keur S Gueye Poular 435 52 13 Sirmang Keur S Gueye Serer 1 091 98 TOTAL 2260 892 The main ty pes of soil are locally called Dior and Deck. The Dior soil, containing > 95% sand and few nutrients (< 2% organic carbon), are usually occupied by former dune slopes while the Deck soil are hydromorphic, occasionally flooded and contain 85 to 90 % sand an d have 0.2% organic carbon (Badiane et al., 2000). The climate is semi - arid and the annual rainfall range s from 400 to 1000 mm with a rainy season usually lasting 3 months from July to September (Toubacouta) but can last 5 months (from June to October) in the Rural Community of Keur Samba Gueye where the 135 annual rainfall can reach 1845 mm as it was the case in 2003 (LDP, 2009). This rainfall can vary significantly from year - to - year in quantity, spacing, and duration. The main economic activity is the rainfe d agriculture dominated by peanuts (Arachis hypogaea) , millet (Pennisetum typhoides) and cashew (Anacardium occidentale) production. The most common tree species that dominate the vegetation are Cordyla pinata , Cola cordifolia , Mangifera indica , and block plantations including eucalyptus and the large areas of Anacardium occidentale . In the south (Rural Community of Keur Samba Gueye), the vegetation is dominated by Combretum glutinosum, Cordyla pinnata, Pterocarpus erinaceus, Adansonia digitata, Zizyphus m auritiana, Diospyros mespiliformis, Prosopis africana and Parkia biglobosa . B .2.2. Research activities To assess the potential technical contribution of local communities to carbon measurements, different activities were held during the field campaign a nd are presented in the following Figure B .1: 136 Figure B . 1. Research activities and their objectives B .2.2.1.Village meetings A villag e meeting was organized in each of the 13 villages of the study area to inform villagers about the objective s of the study through an introduction of the research theme. They were also given information about the necessity of select ing participants to the training workshop and the modalities of the workshop . 137 Figure B .2 . Village meetings Diaglé Villag e Keur Samba Nosso village At least two people were selected for each village for the training workshop. Efforts were made to ensure a minimum of two women were in cluded in the focus group discussions and at least one woman for the training workshop. The selection was based on the following criteria: - Educational level: I suggested that trainees must have a minimum level of education to be selected as some of the tasks that they will perfor m required a minimum level of education which is to be able to read and write in French, in local language or in Ajami ; - Gender: I recommended that the gender dimension must be taken into account to encourage a high participation of women; - Age range: I also suggested that the different group - aged people must be represented in the training. 138 B .2.2.2. Community training workshop The training workshop was organized in Toubacouta, headquarter s of the Arrondissement which includes all the rural communities of the study. We provided a budget to ensure their transportation. It was chaired by the Sous Prefet (Figure B . 3 ) with the presence of one of the three presidents of the rural communities. Figure B .3. Training workshop chaired by the State Representativ e (Sous - Préfet) Before the start of the workshop, all the participants were interviewed using a questionnaire survey to understand whether or not they can measure trees and if so, what are the means and methods they use. This information was later comp ared with a post assessment after the workshop. This workshop had four main components: 1. a presentation/ discussion on climate change, carbon and the role of trees on mitigating climate change; 2. laying out the different steps of a carbon monitoring proces s which include: o Landscape Mapping and Stratification (determine the boundary) o Establishing a system of biomass plots o Accurately locate permanent sample plots using GPS 139 o Doing tree biometric parameters used to calculate carbon stocks o Recording all these d ata collected and time o Upload data on an online geographic information system o Calculation of carbon stock per plot 3. a presentation to introduce the GOES's MRV system to the participants; 4. an evaluation of the participants and what they ha d learned from the t raining was the fourth component of the workshop. After explaining during the training how to do all these tasks, each participant ha d been asked to give a ranking of how much they do think they can do each task. This step was followed by a field trip with all participants to put into practice what they learned during the theoretical training. After field work, each participant was re - evaluated based on how well he /she It is impo rtant to mention that selected participants from the villages of Keur Sett and Keur Bakary Camara did not participate the training workshop because of social ceremonies . Therefore, they did not participate in the field campaign as well. B .2.2.3. Farmer's f ield experience in biomass data collection Following the training workshop, a field campaign was planned for each group of participants in each village. All the participants were taken to the field to practice what they had learned from the training. This was followed by an evaluation of the performance of each participant. Each of the tasks was monitored to determine whether all the activities were carried out easily by participants or they were observed to be beyond their ability. This evaluation was base d on how they manipulate the equipment (GPS), how much time they 140 spend to accomplish each task and how accurate are their measurements compared to our own measurements for the same trees. Figure B .4. Field crew in the field measuring trees and recording da ta B .2.3. Data Analysis B .2.3.1. Training workshop P rior to the training workshop , a n a - priori assessment of communities' ability to measure trees had been issued. D escriptive statistics is used to analyze the data collected during thi s assessment. These data are compared with the post assessment data to determine Participant manipulating GPS device to collect geographic coordinates Participants measuring tree DBH Participants measuring tree crown diameters Participants recording data on the data collection sheet 141 the extent to which the workshop had contributed to increase their understanding of climate change, carbon, the role of trees in mitigating climate change, their interest in p lanting more trees in their farms, their understanding of the importance of measuring trees. The participants also provided information about whether or not they would participate in tree measurements. They also have been asked the conditions under which t hey will be willing to participate. B .2.3.2. Communities' field experience in carbon measurement and reporting Data on tree biometric parameters collected by local participants were analyzed statistically and compared with the data collected previously by the researcher for the same sampled trees. Also, an evaluation of how they perform each of the tasks is issued using descriptive statistics. It is very common that what people say they know or can do is often contradicted by the reality in the field. In th is phase of this study, participant observation we re used to check against what the participants reported regarding their ability to measure trees during the assessment prior to the workshop. B .3. RESULTS B .3.1. Evaluation of participants to the training work B .3.1.1. Characteristics of the participants to the training workshop Of the 30 participants to the training workshop, 83.3 % were male and 16.6 % were female. The ethnic groups are composed of a majority (40%) of Serer, followed by the Wolof (30%), the Bambara (20%) and 10% of Mandinka (Figure B .5 ). 142 Figure B . 5. Characteristics of the participants based on their ethnic groups The majority (58%) are between 41 and 60 years old with 29% in the 21 to 40 year range and 6% for the ranges of 19 - 20 and 61 and above (Figure B .6 ). The average age of participants is 47 years, with the youngest reported as 19 years old and the oldest as 63 years old. Figure B .6. Age strata of participants to the training workshop 0 5 10 15 20 25 30 35 40 Wolof Serer Bambara Mandinka 30 40 20 10 0 10 20 30 40 50 60 19 - 20 21 - 40 41 - 60 61 and above 6 29 58 6 143 Despite a strong recommendation during the selection for a high representation of women, the percentage of females is very low (16.6 %) compared to the male which is 83.3 % (Figure B .7) . Figure B . 7 Percentage of participants based on gender About 2/3 of the participants (63%) can read and writ e either in French or in Ajami (a practice of writing other language (in this case Wolof) using a modified Arabic script) : 27 % have a primary uncompleted level and 13 % a secondary level while 37 % who declared to never have been at school, can neither r ead nor write (Figure B .8 ) . 83.33 16.66 Male Female 144 Figure B .8. Level of Education B .3.2. Results of a - priori assessment of communities' ability to perform tree measurements: Through a priori survey before starting the training, we asked participants questions about whe ther or not they can measure tree diameter, tree height, tree crown diameter and if so how they do it. From t his prior - assessment of farmers ability to measure tree parameters (DBH, height, crown) , we knew many acknowledged knowing how to measure tree tru nk diameter, tree height or tree crown diameter , although they have always use d local techniques. For questions regarding their ability to measure tree diameter, tree height or tree crown diameter, the majority of participants declared knowing how to perfo rm these tasks : 97 %, 83 % and 93 % for trunk diameter, height , and tree crown diameter , respectively. Also, all the participants said that they c ould locate plot s in the field and identify tree species o n their farms and t hrough out the general landscape ( Figure B .9 ). 0 5 10 15 20 25 30 35 40 Never been at school Ajami Primary uncompleted Secondary 37 23 27 13 145 Figure B .9. Self - reported knowledge of farmers on their ability to perform tree measurements Although there was high number of participants who c ould perform these tasks, their performance was different. Regarding the ir capability of lo cat ing plot s all participants declared knowing how to do it but their confidence level varie d: 10% said they can do it, 23 % can do it well, 7 % can do it very well and the majority (60%) are more confident as they said they can do it s o very well (Figure B .10 ) 0 10 20 30 40 50 60 70 80 90 100 Locate plot Identify tree species Measure tree DBH Meaure Tree Height Measure tree crown diameter 100 100 97 83 93 Yes No 146 Figure B .10. Participants rating on how much they do think they can locate a plot Participants are more confident in their ability to identify trees in their farms and even across the whole landscape. 70 % said they know so very well how to identify tree species, 27 % know it very well and 3 % said they can do it well (Figure B .11 ) Figure B .11. Participants rating on how much they do think they can identify trees 0 10 20 30 40 50 60 I think I can do it I can do it well I can do it very well I can do it so very well 10 23 7 60 0 10 20 30 40 50 60 70 I can do it well I can do it very well I can do it so very well 3 27 70 147 Regarding their confidence i n their ability to measure tree DBH, particip ants were classified into four groups: 10 % think they can just do it, 20 % said they can do it well, 37 % can do it very well and 33 % can do it so very well (Figure B .12 ) Figure B .12. Participants rating on how much they do think they can measure tree DB H Although their confidence about how much they think they can perform tasks is different, all the participants declared knowing how to locate a plot, identify tree species or measure tree DBH. Some of the techniques that they used to do these measureme nts are modern especially when they used a tape while others use traditional techniques including bamboo rod to measure the top of the trees. Some participants are using the measuring rope to measure the diameter and the tops of the trees. The older, with their experience, use the shadow of the trees at certain times to determine both the height and the length of the crown diameter. 0 5 10 15 20 25 30 35 40 I think I can do it I can do it well I can do it very well I can do it so very well 10 20 37 33 148 The results of this prior assessment is compared with the post assessment not only to assess what they really can do but also evaluate the impact of the training workshop and how it contributes to increase their knowledge on how to do tree measurement. B .3.3. Communities' performance in carbon measurement and reporting During the communities' field measurements, participants wer e rated and results of this rating in presented in Figure B .13. Regarding how much they were able to accomplish tasks they were rated from 1 to 4 to determine their performance: we rated 1 when one participant w as not at all able to accomplish the task, 2 when he or she did it with strong guidance, 3 when the task is accomplished with just little guidance and 4 when accomplished without any guidance. Participants have a good familiarity with their environment and they did not have any difficulty locating p lots. The assessment shows that all (100%) the participants were able to locate easily the plots without any guidance. Tree identification is also a task that communities can accomplish: 93 % were able to identify all trees without any difficulty or guida nce while 7 % were able to identify trees but they needed some guidance from other crews for some species. Regarding their ability to measure tree DBH, of the 30 participants, 87% did the measurement correctly without any guidance, 13 % with few guidance. Regarding the measurement of tree crown diameter, all participants were able to accomplish this task without any guidance contrasting with their ability to use GPS to record data where only one participant was able to accomplish this task with just few g uidance, 83 % were not able to use the device. 1 49 Their ability to record data on the data collection sheet is also low as more than half (57%) were not able to accomplish this task. This can be explained by the low level of education of participants suggesti ng the need of recruiting educated people . Figure B .13. Assessment of communities' experience in tree measurement To measure tree attributes (DBH, crown diameters) and record all data including the GPS coordinates, our own experience shows a duration of 4 to 5 minutes for each tree. This time frame has been used to determine the ability of participants to finish tasks in a timely manner. However, we took into account the fact that they might need more time, and we doubled the amount of time and fixed i t to 10 minutes for each tree. I observed field crew teams measuring and recording during data collection and we rated each field crew team depending on how much they finished the tasks in a timely manner with three possible cases: when they finish within 7 or 10 minutes and if they needed more than 10 minutes. Results of the assessment show that of the 30 participants, 43 0 20 40 60 80 100 Locate plot Identify trees Measure tree DBH Measure tree crown diameter Record data using GPS Record data on the data collection sheet 83 57 13 3 7 13 3 17 100 93 87 100 23 He was not able to do it He did it with strong guidance He did it very well with just few guidance He did it so very well without any guidance Percentage of participants 150 % accomplished all the tasks in 7 minutes or less while 40 % needed all 10 minutes to finish and 17 % went beyond the 10 minutes (Figur e B .14) . Figure B .14. Percentage of participants finishing task in a timely manner B .3.4. Reliability of communities' estimates The reliability of the carbon estimates made by communities was tested by comparing them to our own measurements. DBH and cro wn diameters of t he same trees were measured by both the researcher and the communities. Results of comparison between these two data show a slight difference. The biomass estimates of all trees that both we and the communities have measured are summed up. A comparison of the values shows that the difference in biomass estimation in all cases is less than 7 %. In some cases, including in Senghor ( - 2.82 %), Niokholokho ( - 0.60) and Diaglé ( - 1.09), the communities' measurements are lower than our own estimate s. For the remaining cases, our estimates are lower. 0 5 10 15 20 25 30 35 40 45 They did it in more than 10 mns They did it within 10 mns They did it in less than 7 mns 17 40 43 Percentage pf participants 151 Table B .2. Comparison between communities' biomass estimates and researcher's estimates Biomass (sum of all Biomass/tree in Kg in each village) Mean estimates by local communities Mean estimates by researcher % difference of mean Keur Samba Gueye 10164.39 9987.59 1.74 % Keur Samba Nosso 7499.53 7005.87 6.58 % Sirmang 18319.30 17264.53 5.76 % Senghor 4167.35 4285.06 - 2.82 % Niokholokho 9759.02 9817.80 - 0.60 % Ngouye - Kouatine 14497.08 14317.38 1 .24 % Diaglé - Ndoffan 26000.93 26285.04 - 1.09 % TOTAL 90407.6 88963.28 1.59 % B .3.5. Conditions for communities to participate in carbon measurement To identify the conditions for communities to participate in carbon measurements, we asked three major questions to the participants to the workshop and the field campaign. Based on the criteria on which they were selected, what they represent for their respective villages, and their potential ability of influencing the rest of the villagers, we assume that their responses to our questions are valid enough to represent the perceptions of a large community. The three questions that we asked are: 1 - Will you be willing to participate in carbon measurement? 2 - Under what conditions will you be willing to partic ipate? 3 - If you need compensation, what type of compensation would you looking for? The responses to these questions are summarized in the following table B .3 : 152 Table B .3.Willingness and conditions for communities to participate in carbon measurement QUESTIONS RESPONSES Number % Will you be willing to participate in carbon measurement? YES 30 100 NO 0 0 Under what conditions will you be willing to participate? Volunteer 2 6.6 With compensation 28 93.3 If you need compensation, what typ e of compensation would you looking for? Monetary compensation 29 96.6 Community infrastructure 1 3.3 All participants are willing to participate in carbon measurement if they are solicited. However, their participation must be motivated by compensatio n. On the 30 participants only 2 accept to participate without monetary compensation. These two participants considered that the knowledge they gain from participating and the capacity they will have to accomplish the tasks by themselves are sufficient eno ugh. The remaining 28 participants accept to participate in return for a payment of their participation. Additionally to the monetary compensation, most of the respondents would ask for field equipment including tape, GPS and camera. It is important to men tion that a deeper investigation of the cost of community participation is needed because knowing the cost will determine the extent to which the participation of local communities on carbon measurement can constitute a strategy. 153 B .4. DISCUSSIONS AND C ONCLUSION The objectives in this chapter was to explore the effectiveness, accuracy and precision of joint measurement campaigns in savanna landscapes and agroforestry systems outside forests and test the ability of local communities to measure tree parame ters for carbon estimation. The findings show that local communities are able to perform some tasks in the process of measuring tree parameters to estimate carbon, although the majority of participant s wa s illiterate in the sense that they did not attend formal education (defined as French school) and can neither read nor write. Our results show that even people with no formal education level can do some tasks including measuring tree crown dimension (holding the tape does not require a certain level of ed ucation). Some older people, with their higher level of experience, use the shadow of the plant at certain times to estimate the height of a tree. They also suggested that this technique can be used t o determine the tree crown area but I did not test those techniques to assess their reliability and validity. Regarding the data recording, people that have an arabic ( Ajami ) level can perform the task. As shown in Table B .2 , there is not a big statistical difference between our own measurements during the sam e period of the field work and for the same trees and the communities' measurements. This is a confirmation that, with a minimum training and although they might not receive a formal education, local communities can actively participate in forest inventori es and tree measurement to estimate biomass and carbon. 154 Regarding the conditions under which local communities can participate to carbon measurements, most of the respondents considered monetary compensation as a condition of their participation. 155 REFERENCES 156 REFERENCES Abrell, Elan; Kabir ,Bavikatte; Harry Jonas, Ilse Kohler, et al. (2009). Bio - cultural community protocols: A community Approach to Ensuring the Integrity of Environmental Law and Policy. UNEP. Badiane, Aminata Niane; Mamadou Khouma and Modou Sene. (2000). Region de Diourbel: Gestion des Sols, Working Paper 15. Somerset, England: Drylands Research. Danielson, Finn; Neil D. Burgess and Andrew Balmford. (2005). Monitoring matters: examining the potential of locally - based approach es. Biodiversity and Conservation , 2507 - 2542. Evans Kristen and Manuel R. Guariguata. (2008). Participatory monitoring in tropical forest management: e review of tools, concepts and lessons learned. Bogor, Indonesia: Center for International Forestry Rese arch (CIFOR). Environmental Science and Policy 14 , 181 - 187. Herold, Martin and Margaret Skutsch. (2011). Monitoring, reporting and verification for national REDD + programmes: Two proposals. Environment Research Letters 6 , 1 - 11. Holck, M.H. (2008). Participatory forest monitoring: an assessment of accuracy of simple cost - effective methods. Bidoversity and Conservation, vol 17 , 2023 - 2036. Larrazabal, Alejandra; Michael K McCa ll, Tuyeni H Mwampamba and Margaret Skutsch. (2012). The role of community carbon monitoring for REDD+: A Review of experiences. Current Opinion in Environmental Sustainability (4) , 707 - 716. Samek, Jay et al. (2014). Integrated community - based participa tory carbon measurement and monitoring with satellite remote sensing and GIS in REDD+ MRV Systems. In C. T. Louis Lebel, Climate Risks, Regional Integration, and Sustainability in the Mekong Region (p. 285). Bangkok: Stockholm Environment Institute. Skutsc h, M., Van Laake, P.E., Zahabu, E., Bhaskar, S.K., and Pushkin, P. (2009). The value and the feasability of community monitoring of biomass under REDD+. CIFOR. Topp - Jorgensen, Elmer; Michael K. Poulsen; Jens Friis Lund and John F. Massao. (2005). Community - based monitoring of natural resource use and forest quality in montane forests and miombo woodlands of Tanzania. Biodiversity and Conservation (14) , 2653 2677 . 157 Appendix C : EVALUATION OF FARMERS 'TRAINING WORKSHOP AND FIELD WORK EXPERIENCE C .I - TRAI NEE IDENTIFICATION Name Village Gender Age Contact (phone/email) C .II - A - PRIORI SURVEY 1. Do you know how to locate plot? |__| 1. Yes |__| 2. No 2. Do you know how to identify tree species? |__| 1. Yes |__| 2. No 3. Do you know how to meas ure tree trunk diameter? |__| 1. Yes |__| 2. No 4. If Yes, how do you do it? ............................................................... ............................................................... ............................................. 5. Do you know how to measure tree height? |__| 1. Yes |__| 2. No 6. If Yes, how do you do it? ............................................................... ............................................................... .............................. .............. 7. Do you know how to measure tree crown diameter? |__| 1. Yes |__| 2. No 8. If Yes, how do you do it? ............................................................... ............................................................... ...... ....................................... C .II I - WORKSHOP EVALUATION C .III.1. From 1 to 4, rate how much has the training workshop increased your understanding of Climate Change. 4 3 2 1 158 C .III.2. - From 1 to 4 rate how much has the trai ning workshop increased your understanding of Carbon. 4 3 2 1 C .III.3. - From 1 to 4, rate how much has the training workshop increased your understanding of the role of trees in mitigating Climate Change. 4 3 2 1 C .III .4. - From 1 to 4, rate how much has the training workshop increased your interest in planting more trees in your farms. 4 3 2 1 C .III.5. - From 1 to 4, rate how much has the training workshop increased your understanding of the importance of measuring trees. 4 3 2 1 C .III.6. - From 1 to 4, rate how much do you think you can do each task. Locate plot 1 2 3 4 Tree identification 1 2 3 4 Measuring tree DBH 1 2 3 4 Measuring tree height 1 2 3 4 Measuring tree crown diameter 1 2 3 4 Recording data using GPS 1 2 3 4 Recording data on the data collection sheet 1 2 3 4 Finish tasks in a timely manner 1 2 3 4 1: I can do it 2: I can do it well 3: I can do it very well 4: I can do it so very well 159 C . I V - EVALUATION OF FIELD EXERCISE Locate plot 1 2 3 4 Tree identification 1 2 3 4 Measuring tree DBH 1 2 3 4 Measuring tree height 1 2 3 4 Measuring tree crown diameter 1 2 3 4 Recording data using GP S 1 2 3 4 Recording data on the data collection sheet 1 2 3 4 Finish tasks in a timely manner 1 2 3 4 1: The participant were not able to accomplish the task 2: The participa nt performs the task with strong guidance 3: The participant performs the task with few guidance 4: The participant performs the task without guidance 160 REFERENCES 161 REFERENCES Baccini, A.; N. Laporte; S.J. Goetz; M. Sun and D. Huang. (2008). A first map of tropical Africa's above - ground biomass derived from satellite imagery. Environ. Res. Lett 3 , 1 - 9. De Foresta, Hubert; Eduardo Somarriba; August Temu; Désirée Boulanger; Hélène Feuilly and Michelle Gauthier. (2013). Towards the Assessment o f Trees Outside Forests. Rome: Resources Assessment Working Paper, FAO. Jonsson. K.; Ong. C.K.; Odongo, J.C.W. (1999). Influence of scattered nere and karite trees on microclimate, soil fertility and millet yield in Burkina Faso. Experimental agriculture.3 5 , 39 - 53. Kuyah, Shem; Catherine Muthur; Ramni Jamnadass; Peter Mwangii; Henry Neufeldt; Johannes Dietz. (2012). Crown area allometries for estimation of aboveground tree biomass in agricultural landscapes in Western Kenya. Agroforestry Systems , 1 - 11. M atricardi, E.; Skole, D.; Pedlowski, M.; Chomentowski, W. and Fernandes, L. (2010). Assessment of tropical forest degradation by selective logging and fire using Landsat imagery. Remote Sensing of Environment 114 , 1117 - 1129. Mitchard, E.T.A. et al. (2011 - ground biomass derived from satellite imagery. Environmental Research Letters , 1 - 7. Neupane, Ramji P. and Thapa, Gopal B. (2001). Impact of agroforestry intervention on soil fertility and farm income under the subsistence farming system of the middle hills, Nepal. Agriculture, Ecosystems and Environment (84) , 157 - 167. Pandey, D. (2002). Fuelwood studies in India. Myth and reality. Bogor: CIFOR. Verchot, Louis V., et al. (2007). Climate Change: Linking adaptation and mitigation through agroforestry. Mitigation And Adaptation Strategies For Global Change , 901 - 918 . 162 CHAPTER 5 . GENERAL CONCLUSION The objective of this research w as t o e xplore the possibilities of developing a protocol that allows landscape scale measurements of trees outside of forests combining high resolution satellite imagery and ground dat a ; which protocol having the accuracy and the rigor scientifically acceptable and usable by the international community . The findings and limits of the re search along with the recommendations for further research are summarized as follows: 5 .1. S UMMARIES AND RESEARCH QUESTION REVISED The major research question of this study was: - Can protocols be developed that allow for landscape scale measurements of tr ees outside of forests with the accuracy and rigor that is acceptable to the international community? The d eveloped model is applied to measure the carbon stocks in rural farming systems under agro - forestry systems including cashew as well as other indivi dual trees in the agricultural landscape. Our results show that carbon in smallholder farms, especially for individual trees outside forest can be measured with accuracy comparable to what obtained from other studies. A correlation between field measured DBH and remote sensing CPA has shown a strong correlation between the two variables for the linear model with coefficients of determination R 2 > 0.7 0 except when applied to the case of individual TOF (133 trees) in the Karang site where it is 0.66 . By comp aring the linear and the non - linear our results suggest that the linear model is the best one. These results suggest that this developed model can be 163 used to predict tree DBH using high resolution satellite image s especially for individual trees outside fo rests where the detection is easier than for clustered trees. The model is slightly better for natural, unmanaged compared to managed trees. The application of the model to estimate biomass and carbon at the landscape level has shown satisfactory results c ompared to other studies: - the first one from Western Kenya where Kuyah et al. (2012) results are very close to our own carbon estimates. Using a destructive sampling method, they estimate the carbon density to be 20.8 t / ha where our results are betw een 3.9 to 17 t C/ha depending on the blocks. - the second comparison is made with Baccini et al. (2012) u sing the exact same area . The two maps of the comparison presented in Figure 4.11 and Figure 4.12. show some closeness in some areas but in others, there are differences . Based on these different results, the answer to the research question can be summarized as follows: i t is difficult to directly validate the model against field measurement methods because allometry based on DBH has its own error and the sample size for this landscape needs to be several hundred plots. However, our results show that the model and field based methods closely concur. We concluded that a model that uses remote - sensing assisted landscape - scale carbon stock measurement has promise. The relationship between CPA detected from remote sensing and allometric scaling is something that can be refined but seems to be a workable approach. Refinements would include: - Improved relationship model using non - linear relationships - Devel oping a local allometric equation using destructive sampling, and specific parameters for the savanna or tree type/species 164 - Pursue and improve the a utomated detection to extract the crown features to replace the hand digitizing. Regarding the potential o f local communities to make accurate contributions to the carbon measurement process, our results have shown that they have skills and an ability to perform some tasks in the process of measuring tree parameters and estimate carbon. Despite the low percent age of literacy, most of the participants have shown a good performance in measuring tree crown dimension s , tree DBH, recording data and finishing tasks in a timely manner. Their measurements have also shown a high accuracy when compared to our own measure ments made during the same period of the field work and for the same trees. Consequently, local communities have an important role to play on carbon measurement and our results confirmed that they can play that role. 5 .2. L IMITS AND RECOMMENDATIONS FOR FU RTHER RESEARCH This research has some limitations presented as follow: 5 .2.1. Regarding the tree object detection, we faced some difficulties related to the irregular pattern of the crown shape. It was not easy to manually digitize tree crown with minimal error. Also, when trees are intermingled as in the case of cashews, it was very difficult to determine the boundary between tree crowns. 5 .2.2. Also the irregularity of the tree crown's outline does not facilitate the field measurements of CPA, which coul d affect the accuracy of the measurement. 5 .2.3. The satellite image was acquired on April 2010 and the field data collected on May - June 2013. The time lag between image acquisition and field data collection may a ffect the size of tree crown which can be slightly bigger than in the World view image because of the gro wth season in between (Song et al . , 2010). Also, it was not easy to identify pruned trees 165 in the image. Building an algorithm to detect individual tree to replace the manual digitizing could pr oduce more accurate results. 5 .2.4. The carbon estimation derived from the use of a general allometric equation has also some shortcomings in terms of accuracy. Building a specific allometric equation, especially for Anacardium occidentale (cashew) would i ncrease the accuracy of carbon estimates. 5 .2.5. In this study, the formula for the area of an ellipse is used to calculate the field measured CPA of sampled trees. However, the definition of an ellipse in that way may include area not actually occupied by the crown (Anderson et al., 2000), especially if the crown is not circular or has indentations. An overestimation or underestimation of the crown area might occur and influences the relationship between CPA and DBH. 5 .3. I MPLICATIONS AND THE PLACE OF THIS STUDY IN THE GLOBAL CLIMATE CHANGE AND SUSTAINABLE DEVELOPMENT AGENDAS Using a landscape based model to estimate carbon for trees outside of forest can be a useful tool to value the carbon stock in agroforestry and savanna systems. Such model has an impo rtant role to play in monitor ing and managing carbon within these systems. Knowing the importance of carbon sequestration in mitigating climate change and the possibility of estimating carbon stored on TOF and valuing this carbon for the benefits of local communities may contribute to increase biomass carbon stock in trees outside of forest. Also, a landscape approach has an important role to play for development as it offers rural poor good opportunities to participate in carbon markets especially African countries. This also provides significant carbon storage opportunities for the agricultural sector through land use and cultivation practices that promote more trees in farms. This pers pective is even more beneficial that eve climate - smart objectives, 166 including improved food security and rural livelihoods as well as climate change adaptation et al., 2012). I addressed the ability of local communities to measure trees on their own farms. The results presented in Appendix A show that e ngaging local communities in measuring carbon in their farms requires a minimum training. When they can make the tree measurements to estimate the carbon on their own farms, t hey can be more motivated to participate in the carbon markets, reducing the barriers such as the measurement and monitoring costs that prevent developing countries from taking full advantage of the growing carbon markets. The model we developed will contr ibute to address the carbon measurement problem, especially in large areas for systems of trees outside of forests (cf. agro - forestry and related systems) in the savannas of Africa with 9 million km 2 of woodland of which 1.4 million for West Africa (Chidu mayo et al ., 2010) and 1,043,000 ha for Senegal (Mayaux et al ., 2004) . This carbon measurement problem is a serious obstacle in implementing initiatives such as REDD + and other forestry activities that aim at combining tackling climate change and developme nt .