UNIVERSIDAD POLITÉCNICA DE MADRID ESCUELA TÉCNICA SUPERIOR DE INGENIEROS AGRÓNOMOS DEPARTAMENTO DE INGENIERÍA RURAL TESIS DOCTORAL:

Transcription

1 UNIVERSIDAD POLITÉCNICA DE MADRID ESCUELA TÉCNICA SUPERIOR DE INGENIEROS AGRÓNOMOS DEPARTAMENTO DE INGENIERÍA RURAL TESIS DOCTORAL: Respuesta de la vegetación a variaciones climáticas en praderas y sistemas adehesados Mediterráneos: Metodología de análisis utilizando datos hiperespectrales y multiespectrales Doctorando: MONICA GARCÍA GARCÍA (Ingeniera Agrónoma) Directora: ANA IGLESIAS PICAZO (Doctora Ingeniera Agrónoma) Co-Dírectora: SUSAN L. USTIN (Doctora en Botánica) Tutora: MARGARITA RUIZ ALTISENT (Doctora Ingeniera Agrónoma) Madrid, Septiembre 2003

2 UNIVERSIDAD POLITÉCNICA DE MADRID ESCUELA TÉCNICA SUPERIOR DE INGENIEROS AGRÓNOMOS DEPARTAMENTO DE INGENIERÍA RURAL DOCTORAL THESIS: Vegetation Reponses To Climatic Variability in a Mediterranean Grassland And Savannah. Methodologies of Analysis Using Hyperspectral and Multispectral Data Ph.D. Candidate: MONICA GARCÍA GARCÍA (Agricultural Engineer) Director: ANA IGLESIAS PICAZO (Doctor in Agricultural Engineering) Co-Director: SUSAN L. USTIN (Doctor in Botany) Advisor: MARGARITA RUIZ ALTISENT (Doctor in Agricultural Engineering) Madrid, September 2003

3 A mis padres Jesús y Pilar, A mi hermana Natalia,

4 AGRADECIMIENTOS Esta tesis no hubiera sido posible sin la colaboración estrecha de las tres personas que me han dirigido, cada una de ellas con un papel irremplazable. Quiero agradecer, a la Dra. Ana Iglesias, investigadora del Departamento de Economía y Ciencias Sociales Agrarias, de la E.T.S.LA su labor de dirección, proporcionándome una visión global de la tesis, y animándome siempre. He aprendido mucho de su conocimiento del clima y la vegetación, así como de su actitud abierta, entusiasmo y capacidad para relacionar distintos temas. A Dña Margarita Ruiz Altisent, Catedática del Departamento de Ingeniería Rural, de la E.T.S.LA, por su saber hacer, apoyo y amistad durante estos años, y por ofrecerme la oportunidad de trabajar en el Laboratorio de Propiedades Físicas. Realmente admiro su lucidez en investigación y aprecio mucho su confianza. A Susan Ustin, profesora del Departamento de Land, Air and Water Resovirces de la Universidad de California, Davis, por el equilibrio entre dirección y libertad en el trabajo, su interés, optimismo y amistad, así como por considerar al alumno como un todo con sus circunstancias. La oportunidad de formar parte del laboratorio CSTARS ha sido estupenda y su apoyo tanto en Davis como en Madrid ha sido tremendo. A Keir Keightley, y a Víctor Gil por su ayuda en el preprocesado de las imágenes AVIRIS, a George Scheer, Luis Ruiz y Constantino Valero por el apoyo informático y de uso de equipos del laboratorio. A esta tesis han contribuido muchas personas de una manera o de otra: con sugerencias después de leer los capítulos, discusiones, con los datos o las referencias: Prof. Ted Hsiao, Prof. Richard Plant, Prof. Alicia Palacios, Prof. Pilar Barreiro, Dr. David Riaño, Antonio Trabucco, Prof.. Lourdes Lleó, Dr. Pablo Zarco-Tejada, Prof. Stephan Jacquemoud, Dr. Femando Valladares, Dra. M* José Mariscal, Mike Whiting, Dr. Lin Li, Quiím Hart, y Dr. Pablo Rosso. Muchísimas gracias a mis compañeros del Laboratorio de Propiedades Físicas de productos agrícolas de la E.T.S.LA, por su amistad, y por su ajmda desde el análisis de datos, programas, discusiones, y papeleos: Belén Diezma, José Bermejo, Natalia Hernández, María Marín, Pablo Gutiérrez, Adolfo Moya, Guillermo, lan Homer y Coral Ortiz. A Rigo Pérez de Alejo por su ayuda con el código IDL. A mis compañeros del CSTARS de la Universidad de California, Davis por su ayuda en distintos momentos de la tesis: Shawn Kefauver, Severíne Tumois, y Muy Lai,. Muchas gracias a todos los miembros del Departamento de Ingeniería Rural, especialmente a José Luis García, Profesor titular de la E.T.S..LA y a Jaime Ortiz Cañavate, catedrático de la E.T.S. de Ingenieros Agrónomos, por su apoyo y actitud abierta hacia nuevas líneas de investigación. A la Fundación "Barrie de la Maza" por proporcionarme una beca para estudiar dos años en Estados Unidos, así como al programa NASA EOS program grant No. NAS

5 ACKNOWLEDGEMENTS I want to thank the three persons that made possible this thesis: my director Dra. Ana Iglesias Picazo, my co-director: Professor Susan Ustin and my advisor: Professor Margarita Ruiz Altisent, for always believing in me, and for co-operating together to get this thesis completed. I am grateíul to Dra. Ana Iglesias, researcher at the Agricultural School In Madrid, for her direction, helping me to not get lost and giving me a global view of things. She really cheered me up to complete this, and I have leamt so much from her knowledge of climate and vegetation, her open attitude and visión to link topics, and her enthusiasm at work. To Professor Margarita Ruiz Altisent, catedratic of the Rural Engineering Department, for her unconditional support when doing the thesis and in the past years, for giving me the opportunity to work in the Laboratory of Physical Properties. I admire her clear view at research and really appreciate her confidence. To Professor Susan Ustin, from the University of California, Davis, for the balance between guidance and freedom that she offers at work, for her interest, optimism and friendship, and for considering the student together with her circumstances. Also, the opportimity of taking part of the CSTARS lab has been unvaluable for me. Thanks to Keir Keightley, and Víctor Gil for their help in preprocessing AVIRIS images, and George Scheer, Luis Ruiz and Dr. Constantino Valero for computer and equipment support. This thesis is the result of the input of many people that had contributed in one way or another to make it better with suggestions añer reading chapters, discussions, or giving a hand with the data or references: Prof. Ted Hsiao, Prof. Richard Plant, Dra. Alicia Palacios, Prof. Pilar Barreiro, Dr. David Riaño, Antonio Trabucco, Dra. Lourdes Lleó, Dr. Pablo Zarco-Tejada, Prof. Stephan Jacquemoud, Dr. Femando Valladares, Dra. M^ José Mariscal, Mike Whiting, Dr. Lin Li, Quinn Hart, and Dr. Pablo Rosso. Thanks to all my colleagues of the Physical Properties Laboratory in Madrid for theirfriendship,and for helping with anything from data, programs, discussions, and paperwork, especially Belén Diezma, José Bermejo, Natalia Hernández, María Marín, Pablo Gutiérrez, Adolfo Moya, Guillermo, lan Homer and Coral Ortiz. Especial thanks to Rigo Pérez de Alejo for helping with the IDL code. Thanks to my colleagues from the CSTARS lab for their support at different moments when pursuing the thesis, especially: Shawn Kefauver, Severine Tumois, and Muy Lai. I am very grateful to all the members of the Department of Rural Engineering of the Polj^echnique University in Madrid, especially José Luis García, Professor at the E.T.S..I.A and Jaime Ortiz Cañavate, catedratic of the E.T.S.I.A, for supporting me with the thesis and for their open attitude towards new research áreas in the department. To Foundation "Barrie de la Maza" from Spain for providing fínancial support for pursuing gradúate studies in the USA during two years, and to NASA EOS program grant No. NAS

6 TABLE OF CONTENTS RESUMEN DE LA TESIS 1 CHAPTER1. INTRODUCTION 6 1. JUSTMCATIONOF THE RESEARCH 7 2. OBJETIVOS 9 3. OBJECTIVES ESTRUCTURA DE LA TESIS Y COMPONENTES THESIS STRUCTURE AND COMPONENTS 12 CHAPTER 2. BACKGROUND CLMATE The Climate System El Niño-Southem Oscillation (ENSO) ENSO and global seasonal climate ENSO effects on US climate Climate pattems in California Climate forecast VEGETATION Vegetationresponsesto weatherevents Grassland vegetation in California Vegetation responses to water availability in Mediterranean ecosystems Plant adaptations to cope with rainfall variability Water relations and plant phenology Impacts of rainfall variability on vegetation productivity REMOTE SENSING FOR VEGETATION MONITORE^G Imagingspectrometry:theAVIRIS sensor Spectral, radiometric and spatial calibration of the AVIRIS sensor Linear Spectral Unmixing: Theoretical basis Endmember Selection Strengths and weaknesses of linear unmixing 49 CHAPTER 3. THE STUDY SITE THE CENTRAL COAST OF CALIFORNIA THE STUDY SITE: JASFERRIDGE Phenology and productivity of the vegetation types in Jasper Ridge 59

7 2.2. Water relations of JasperRidgespecies Remote sensing at Jasper Ridge Biological Preserve 63 CHAPTER4 64 RELATIONS BETWEEN ENSO (EL NIÑO SOUTHERN OSCILLATION) EVENTS WITH INTERANUAL CLIMATIC VARIABILITY AND PRIMARY PRODUCTIVITYIN CALIFORNIA INTRODUCTION OBJECTIVE STUDY REGIÓN AND SITE DATA 67 4.LClimate AVHRR NDYI biweekly composites Land use and land cover (LULC) Natural pasture data METHODS Climatic and oceanic anomalies Small scale analysis of precipitation anomalies: JRBP NDVI temporal series RESULTS AND DISCUSSION Land use and land cover preprocessing General climate description atthe site Climate and sea surface temperature anomalies NDVI temporal series analysis Error analysis Implications for pasture production and prediction CONCLUSIONS 96 CHAPTER5 97 DETECTION OF INTERANNUAL VEGETATION RESPONSES TO CLIMATIC VARIABILITY USING AVIRIS DATA INTRODUCTION OBJECTIVES JASPER RIDGE STUDY SITE AVmiS DATA METHODS MAGE PREPROCESSING Reflectance retrieval and atmospheric calibration

8 6.2. Georeferencing IMAGE ANALYSIS: SPECTRAL MIXTURE MODEL Endmember Selection Error analysis RESULTS AND DISCUSSION Endmember selection Error analysis Unmixing with four endmembers in grassland áreas Differences in mean unmixed fractions Change detection using unmixed fractions CONCLUSIONS 120 CHAPTER6 122 MULTITEMPORAL ANALYSIS OF HYPERSPECTRAL IMAGES (AVIRIS): SEASONAL AND INTERANNUAL SPECTRAL EVOLUTION AND RELATION WITH ENVIRONMENTAL FACTORS IN JASPER RIDGE INTRODUCTION AM) OBJECTIVE DATA METHODS Image preprocessing Cross-calibration of AVIRIS images Georeferencing Narrowband Índices NormalizedDifferenceVegetation Index (ND VI): Simple Ratio Water Index (SRWI) Photochemical Reflectance Index (PRI) Linear Unmixing modeling of AVIRIS images Analysis of possible sources of error in the mixing model Impact of sensor noise and atmospheric absorption Impact of illumination differences and atmospheric effects Impact of variation in the pixel components RESULTS AND DISCUSSION Image preprocessing Cross-calibration of AVIRIS images Narrow band Índices Normalized Difference Vegetation hdex (NDVI) Simple Ratio Water Index (SRWT) 142 m

9 Ménica García García Photochemical Reflectance Index (PRI) Discussion Linear UnmixingModelingofAVIRIS images Analysis of mean fractions Error analysis Analysis of possible sources of error in the mixing model Impact of AVIRIS noise and atmospheric effects Impact of illumination differences and atmospheric effects Impact of variation in the pixel components Discussion CONCLUSIONS 160 CHAPTER7 162 COMPARISON OF VEGETATION STRUCTURE PATTERNS DETECTED WITH REMOTE SENSING: IMPLICATIONS FOR OPTIMUM SPATIAL RESOLUTION INTRODUCTION DATAUSED SPECTRAL VARIABLES USED Gxeen vegetation fractions Normalized Difference Vegetation Index (NDVI) TMImagepreprocessing METHODS Experimental Semivariogram analysis Experimental Semivariogram at Large scale Semivariogram at Small Scale: Degradation of spatial resolution RESULTS Analysis of the spatial coeffícient of variation Experimental semivariograms at Large-Scale Experimental semivariogram atsmall-scale Degradation of spatial resolution DISCUSSION CONCLUSIONS 184 CHAPTER 8: GENERAL CONCLUSIONS Relationship between climatic variability and Net Primary Productivity at large scale Potential of Mixing Models in detecting changes on Net Primary Productivity 187 IV

10 8.3. Multitemporal analyses of hyperspectral data (AVIRIS). Interannual and seasonal evolution ofthe spectral variables Vegetation structure pattems identifíed by remóte sensing. Optimum spatial resolution. 189 CHAPTER 9: CONCLUSIONES GENERALES Relación entre la variabilidad climática y la productividad primaria a gran escala Potencial de los modelos de mezcla para detectar cambios en la productividad primarial Análisis multitemporal de las imágenes hiperespectrales (AVIRIS): evolución estacional e interanual de variables espectrales Comparación de patrones de la estructura vegetal detectados por teledetección: implicaciones para la resolución espacial óptima 194 FURTHER RESEARCH 196 REFERENCES 199

11 LIST OF TABLES Table 2.1.Biomass and productivity of some Mediterranean- type ecosystems. Adapted from Rambal (2001) 30 Table 2.2. Nominal AVIRIS Data Characteristics. (Source Green et al., 1998) 44 Table 3.1 Vegetation cover and green leaf área of Jasper Ridge species. Habit: A (Annual), E (Evergreen), DD (Drought deciduous), WD (Winter deciduous). Green leaf área (%) was measured in May -June 1991 at selected sites within the grassland reported as total aboveground plant área Gamón et al., Mean cover (%) in plots were the species occurred within a 311 transect at the chaparral site (Ackerly et al., 2001) 61 Table 4.1. Vegetation types considered in the analysis at the two scales: Central Coast and Jasper Ridge Biological Preserve (JRBP) 69 Table 4.2. Statistics on the área of vegetation polygons from the Large Scale LULC vegetation map used in the Santa Cruz Mountains Coast Range área. The área and number of polygons in the deciduous forest class is very low compared to the other types 70 Table 4.3. Statistics in the área of vegetation polygons from the Small Scale vegetation map used at JRBP 70 Table 4.4. ENSO events by year (grouped by three months) for the years with NDVI -AVHRR data. Source: NOAA. CPC/NCEP (Climate Prediction Center/National Centers for Environmental Protection). Cold periods are designated as C, and warm periods as W, the (-) sign indicates weak events, and the (+) strong events. No sign indicates modérate events 73 Table 4.5. Categorization of years based on winter SST anomalies (greater than mean + standard deviation) from December of the previous year and January of the current year measured at the NIN04 región 78 Table 4.6. Percentile (%) corresponding to precipitation anomalies in El Niño years with respect to the total precipitation distribution ( ). Precipitation anomalies valúes are calculated every three months between January and June 80 Table 4.7. Probability associated with T-test for the monthly precipitation anomalies in JRBP for El Niño and Neutral years. Categorization of years included modérate and strong SST anomalies in October, November and December of the previous year 80 Table 4.8. Probability associated with T-test for the accumulated 3 and 6 months precipitation anomalies in JRBP for El Niño and Neufral years. Categorization of years included modérate and strong SST anomalies in October, November and December of the previous year 81 Table 4.9 Probability associated with T-test for the monthly precipitation anomalies in JRBP for La Niña and Neutral years. Categorization of years included modérate and strong SST anomalies in October, November and December of the previous year 82 Table obability associated with T-test for the accumulated 3 and 6 months precipitation anomalies in JRBP for La Niña and Neutral years. Categorization of years included modérate and strong SST anomalies in October, November and December of the previous year 82 Table 6.1. AVIRIS image dates used in the analysis 124 Table 6.2. Precipitation ranking in Jasper Ridge between scenes. 5 is the highest and 1 is the lowest. In parenthesis are the precipitation levéis in mm 125 vi

12 Table 6.3. Table with data from georeferencing: number of points for polynomic warp and errors. 126 Table 6.4 table with nr of bands/scene used in the spectral mixture analysis 129 Table 7.1. Original spectral data used in the thesis. hi parenthesis is the original variable used as input in the thesis 166 Table 7.2. Analysis performed with each image and variable used (in parenthesis) 168 vil

13 LIST OF FIGURES Figura l.l.(a). Estructura y componentes de la tesis 11 Figure l.l.(b). Structure and thesis components 12 Figure 2.1. Figure of ENSO cycle. Sea surface temperature and tropical rainfall in the equatorial Pacifíc Ocean during normal. El Niño, and La Niña conditions. The sea-surface temperature is shaded: blue-cold and orange-warm. Dark arrows indicate the direction of air movement in the atmosphere: upward arrows are associated with clouds and rainfall and downward-pointing arrows are associated with a general lack of rainfall. (Reproduced form NOAA, Climate Prediction Center) 16 Figure 2.2. El Niño/La Niña cycle influence of world climate (teleconnections at the world scale). Colors represent the extent of impacts: red (warm), brown (dry), green (wet), light blue (cool), dark blue (cool and wet), light green (warm and wet), orange (dry an cool), yellow (dry and warm). SourceNOAA 19 Figure 2.3. a. Precipitation probabilities for March-May associated with El Niño in North America. Source: Intemational Research Institute for Climate Prediction, (IRI) 21 Figure 2.3. b. Temperature probabilities for March-May associated with El Niño in North America. Source: Intemational Research Institute for Climate Prediction, (IRI) 22 Figure 2.3. c. Precipitation probabilities for March-May associated with La Niña in North America. Source: Intemational Research Institute for Climate Prediction, (IRI) 23 Figure 2.3. d. Temperature probabilities for March-May associated with La Niña in North America. Source: Intemational Research Institute for Climate Prediction, (IRI) 24 Figure 2.4. ENSO impacts on Califomia rainfall by climate división. (Source: NOAA) 26 Figure 2.5. Land Use in USA (Source of data: USGS) 29 Figure 2.6. The electromagnetic spectrum and atmospheric windows 37 Figure 2.7. Spectral features of vegetation and dominant factors controlling leaf reflectance 39 Figure 2.8. The AVIRIS (Airborae Visible Infrared Imaging Spectrometer) concept. (Source: NASA/JPL) 45 Figure 2.9. The AVIRIS sensor (Source: NASA/JPL) 45 Figure 3.1. Location of the área of analysis and study site. Includes part of the Central Coast drainage and partof San Joaquín E)rainage 54 Figure 3.2. Non- federal Land Use in Califomia, Source: USDA, National Resources Conservation Service. Natural Resources Inventory Figure 3.3. Agricultural Land Use in Califomia. Census of agriculture Figure 3.4. Área of crop harvested in Califomia in 1964,1982 and Sources: U.S Burean of the Census. Census of Agriculture 1964, USDA ( NASS) Census of Agriculture. 56 Figure 3.5. Vegetation map from Japer Ridge Biological Preserve. Source: Jasper Ridge Biological Preserve. UniversityofStanford (USA) 59 VIH

14 Ménica García García Figare 4.1. Polygons of vegetation classes at the study site from LULC (USGS) añer buffering by 1 km or 0.5 km in deciduous vegetation. Polygons with área less 2 km2 were eliminated. The location of Jasper Ridge is indicated with an AVIRIS image 67 Figure 4.2. (a) Temperature and Precipitation in Jasper Ridge ( ). Source of data: Jasper Ridge Biological Preserve and NASA/GISS. (b) Temperature and Precipitation in Palo Alto ( ). Soiirce of data: Western Regional Climate Center of the Desert Research Institute 76 Figure 4.3. (a) Annual temperature and precipitation in Jasper Ridge ( ); (b) Monthly time series of precipitation in Jasper Ridge ( ). Source of data: Jasper Ridge Biological Preserve and NASA/GISS 76 Figure 4.4. Annual and seasonal precipitation at JRBP for Seasonal valúes are calculated between June of the previous year till July of that year 77 Figure 4.5. Mean monthly precipitation in JRBP for El Niño, La Niña and Neutral years. Error bars represent one standard deviation 78 Figure 4.6. precipitation anomalies in JRBP after removing diy season data, and SST anomalies for the same months from NIN03.4 región 79 Figure 4.7. Monthly precipitation anomalies in JRBP versas warm SST anomalies with a 3 month lag (NIN03.4). When SST anomalies are greater or equal than 1.5, rainfall anomalies are also positive in 70 % of the cases and the mean valué of these anomalies is Figure 4.8. Differences in mean precipitation anomalies differences between El Niño and Neutral years considering weak, modérate and strong El Niño events from CPC/NCEP classifícation for October to November of the previous year. Leñ panel shows mean differences are shown for April (p=0.13) and right panel for March-May (p=0.11). Number of Niño years=l 1, Number of neutral years=8 81 Figure 4.9. NDVI valúes at the study site for March 1996, April 1997, April 1998, June 1997 and October 1995, atthe same dates of AVIRIS images acquisition at JRBP 83 Figure Temporal NDVI series derived from mean AVHRR data within selected vegetation types between and rainfall. (a) Deciduous forest; (b) Evergreen forest; (c) Mixed forest; (d) Herbaceous rangeland; (e) Mixed rangeland; (f) Shrub and brush rangeland; and (g) Non-forested wetland 85 Figure Integral of NDVI over the year calculated as the sum of NDVI valúes within the year for (a) forest vegetation, and (b) herbaceous and shrub land vegetation 86 Figure Seasonality index calculated as the difference between the extreme NDVI valúes within the year for (a) forest vegetation, and (b) herbaceous and shrub land vegetation 87 Figure Relationship between time series of mean NDVI and annual rainfall (Climatic división 4) considering time lags between O and 11.5 for and calculated for 5 variables of accumulated precipitation: Half month (acc 0.5), 2 months(acc2), 3 months (acc3), 6 months (acc6), and 8 months (acc8) 91 Figure Correlation coeffícient (R) between time series of mean NDVI at herbaceous rangeland sites and precipitation at Climatic División 4. Different time lags between O and 11.5 months are shown in the x for 5 variables of accumulated precipitation: Half month (acc 0.5), 2 months(acc2), 3 months (acc3), 6 months (acc6), and 8 months (acc8) 92 Figure Yield of wild hay in tons/ha at the state level, seasonal precipitation data (mm) between June of one year and July of the next year (División 4), and NDVI integral results for the vegetation type herbaceous rangeland. El Niño years are marked with solid arrows, strong IX

15 events are indicated by red arrows, and modérate events with black arrows. La Niña events are marked with open arrows. A strong event is shown with a red open arrow and a modérate event with a black empty arrow).source: USDA (ÑASS) 95 Figure 5.1 True color composite in RGB from May, AVIRIS reflectance image of Jasper Ridge Biológica! Preserve, which is delineated by white Une 102 Figure 5.2. True color composite in RGB from April 29,1998 AVIRIS reflectance image in Jasper Ridge Biological Preserve which by delineated white line 103 Figxure 5.3. Flow chart of the analysis of hyperspectral images used in the thesis 104 Figure 5.4. Pixel Purity index figure from 1998 April image showing puré pixels of water (shadow) in red, vegetation in green, soil (yellow), crop dark blue in the bands: 1200 nm (90), 1210 nm (94) and 1620 nm (135) 110 Figure 5.5. Set of endmembers used for unmixing. Green vegetation and dry grass were selected automatically based on Pixel Purity Index (PPI), soil by visual inspection and shade by minimum albedo levéis in the image, corresponding to alake 111 Figure 5.6. Unmixing results using three endmembers: soil (red), green vegetation (green) and shade (blue) in spring at Jasper Ridge. Áreas with higher proportion of red correspond to the grasslands 112 Figure 5.7. Comparison of residuals across spectral bands between observed AVIRIS reflectance and modeled with unmixing in 1996 and 1998 for different vegetation types. Grasses present higher residuals as a consequence of senescence 114 Figure 5.8. Residuals between observed and predicted reflectance suggesting non-linear scattering effects due to increasing canopy levéis in high transmitting foliage and background effects from soil and litter in the grasses 116 Figtire 5.9. Differences in residuals between grasslands and chaparral áreas showing non-modeled reflectance from dry grass and litter at the red-edge and in the SWIR regions 117 Figure Differences between mean unmixing fractions in 1996 and 1998 calculated for each vegetation community. Error bars represent one standard deviation for the fraction distribution within each vegetation polygon and year 119 Figure Signifícant changes in green vegetation (GV), dry grass and soil fractions between 1996 and Black áreas correspond to decreases and gray áreas to increases in the wet year (1998) relative to the dry year (1996) 120 Figure 6.1. Laboratory spectra from Jasper Ridge selected for simulating linear mixtures within the pixel. Wavelenght has been resampled to AVIRIS specfral resolution 130 Figure 6.2.AVIRIS Signal-to -Noise ratios in 1987,1994, and 2000 (Green et al, 1999) 132 Figure 6.3. RMSE of reflectance (%) at pseudoinvariant features after performing Empirical Line cross calibration with reference reflectance (April 1997). Results correspond to 1995 October, 1996 March, 1997 June, and 1998 April 133 Figure 6.4. RMSE of Reflectance at pseudoinvariant features after and before performing Empirical line cross calibration (a) October 1995 and (b) June Figure 6.5. Mean reflectance differences between reference and apparent reflectance at some of the sites used in the cross-calibratio for June 1997 scene. Below each graph of mean differences, spectra corresponding to mean+ standard deviation of thepixels at the site are shown ( orange reference reflectance (apr97) and green empirical line calibration (ecljun97)) 136

16 Figiire 6.6. Mean differences between reference and apparent reflectance at some of the sites used in the cross-calibration of October 1995 scene. Below each graph of mean differences, spectra corresponding to mean+ standard deviation of the pixels at the site are shown ( orange reference reflectance (apr97) and green empirical Une calibration (ecloct97) 138 Figure 6.7. Standard deviation of reflectance between 5 AVIRIS scenes at the concrete pseudoinvariant site añer cross-calibration 139 Figure 6.8. Maps of NDVI (Normalized difference vegetation index) in (a) March 1996; (b) April 1997; (c) April 1998; (d) October 1995; and (e) June Figure 6.9. Maps of SRWI (Simple Ratio Water Index) in (a) March 1996; (b) April 1997; (c) April 1998; (d) October 1995; and (e) June Figure PRI (Photochemical Reflectance index) mean (red) and range (blue) of valúes within each vegetation type and date (scene) 144 Figure Relationships between NDVI and SRWI for evergreen forest at different dates. In the fourth panel, regresión lines corresponding to the previous data are shown together 145 Figure Library endmembers used in multitemporal spectral mixture analysis from Jasper Ridge spectral library 146 Figure Color composites of unmixing with 3 library endmembers in Jasper Ridge and Palo Alto área. Soil (red), green vegetation (green) and shade (blue). (a) March 1996; (b) April 1997; (c) April 1998; (d) June 1997; and (e) October Figxjre Fractions (%) of spectral mixture analysis using 3 endmembers (tarweed grass, soil butano grassland and shade) on the reflectance data per vegetation type and date, (a) serpentine grassland (b) non-serpentine grassland (c) chaparral (d) open scrubland (e) riparian forest (f) open woodland (g) evergreen forest (h) acquatic vegetation. Error bars represent one standard deviation within each vegetation community 150 Figure Root mean square error (RMSE) of linear spectral unmixing sliced by error levéis (%). (a) March 1996; (b) April 1997; (c) April 1998; (d) June 1997; and (e) October Figure RMSE (Root mean square error) using 3 endmembers (tarweed grass, soil butano grassland and shade) on the reflectance data per vegetation type and date, (a) serpentine grassland (b) non-serpentine grassland (c) chaparral (d) open scrubland (e)riparian (f) open woodland (g) evergreen (h) acquatic vegetation. Error bars represent one standard deviation within each vegetation community 152 Figure Linear mixtures soil and vegetation between O and 100 % spectral library spectra (a) puré spectra mixtures (b) SNR from 1998 degraded spectra mixtures (c) SNR from 1995 degraded spectra mixtures 153 Figure Effects on unmixing resdts of sensor noise corresponding to 1995 and 1997 with increasing fraction of true (simulated) green vegetatio from %. (a) bias in green vegetationfiractions (b) RMSE and shade fraction 154 Figure Effects on unmixing results of variability related to differences between scenes related to illumination, and atmospheric differences with increasing fraction of true (simulated) green vegetatin from 0-100% 155 Figure Impacts on estimated green vegetation fraction when introducing different components from the endmembers at the píxel level. Combinations of two specfra were performed between O % to 100 %. Positive bias indícate overestimations and negative biass underestimations XI

17 Mónica García Garda Figure Impacts on estimated green vegetation fraction when introducing different components from the endmembers at the pixel level.. Combinations of two spectra were performed between O % to 100 %. (a) shade fractions (b) RMSE 158 Figure Relationship between RMSE caused by changes in the spectral components of the pixel and bias in estimation of green vegetation fractions (difference between simulated and predicted fractions (a) underestimations and (b) overestimations 159 Figure 7.1. Flow chart of the analysis in this Chapter 169 Figure 7.2. Location of monitoring sites within IRBP to assess the effect of degrading resolution. 172 Figure 7.3. Comparison of the evolution of the spatial Coefficient of variation for NDVI and NDVI time series in the Coastal Range extent calculated for (a) Deciduous forest; (b) Evergreen forest; (c) Mixed forest; (d) Herbaceous rangeland; (e) Mixed rangeland; (f) Shrub and brush rangeland; and (g) Non-forested wetland 176 Figure 7.5. Experimental semivariograms calculated with AVIRIS and TM images in Jasper Ridge duríng different times of the growing season in different years 181 Figure 7.6. Impact of decreasing resolution on change detection between 1996 and 1998 in spring at the selected sites 182 Figure 7.7. Variation in green vegetation fractions when degrading resolution at the grasslands sites (a) spring 1998 (b) spring 1996 (c) relation for the grasslands samples between resolution and % change in green vegetation detected 182 Xll

18 Ménica García García RESUMEN DE LA TESIS

19 Esta, tesis está, formada ^tor. 9 capítulos. EL Capítulo 1 présenla una introducción dondélsejustífica. la necesidad-de conocer la productimdad.de los pastos naturales y su dependencia de la precipitfición y se- plantean, los objed-vos. d _la íesis EIL zonas Mediterráneas,- la. Yariahüídad climática, especialmente la precipitación es el principal factor limitante para el crecimiento vegetal. lin_ seguimiento, de. la prodikajvidad-.primaria e. las, praderas Mediterráneas^ y una-megor comprensión de sus respuestas a distintas escalas de las variaciones climáticas es importante para realizar-un uso más rarínnal de las mismas y predecir -SIL estabilidad.a.larga plazo. Los datos hiperespectrales tienen el potencial de detectar respuestas no detectadas por sensores de banda ancha y pueden ser usados para escalar a sensores de mapeo global de resolución más grosera. El Capítulo 2 continúa con los antecedentes y revisión bibliográfica sobre el clima, la vegetación, j-s\ uso de-jateledetección. parad seguimiento de-la vegetación, haciendo énfasis en la vegetación de zonas Mediterráneas sujeta a recurrentes periodos de sequía. En. el Capitulo 3 se describe la zona de- Studio. Esta zona. se. encuentra localizada -en el Maciza Costero en las Montañas de Santa Cruz, en Cahfomia (USA) con distintos tipos de vegetación Mediterránea^donde,seihan..iealizadQ los análisis a_gran^escala y comprende, d^ 184 x 280 Km. Incluida en ella, se encuentra la Reserva Biológica de Jasper Ridge (JRBP) con 482 ha, pertenecientaaja-imyersidad-delstanfnrd, donde se han realizado Jos, análisis.a escala localy con datos-hiperespectrales. En la zona de estudio los tipos de vegetación que coexisten y el rango de variadóii ambiental, existente, son representaírvos de los que, se_ encuentran en- otras zonas Mediterráneas, pudiendo servir por tanto de estudio piloto para extrapolar los resultados a otras zonas donde el sensor AVIRIS (Airbome Visible-Infrared Imaging Spectrometer) no se encuentra operativo, como por ejemplo en España. ios capítulos -dd 4.al 7 contienen xada uno un manuscrito jdescribiendo una investigación relacionada con los objetivos planteados en formato de artículo pubhcable. En el Capítulo 4 se pretende predecir la productividad primaria de las praderas naturales en la zona delesíudí0 ea función, de. las variaciones climáticas. Los, patrones, da tiempo están influidos por episodios periódicos que ocurren a escalas mayores de tiempo que las variaciones interanuales y estacionales. -Dichos patrones -periqdi os_de. dima,. como pol-ejemplo d. que ocurre, durante EL Niño, (una de las fases de ENSO-El Niño-Southem Oscillation) influyen en la precipitación de regiones dictantes del origen de. dichos.eventos,-como Caü&mia_o la. cuenca. Mediterránea, Así,- aunque las conexiones entre ENSO y la circulación global en California se han descrito (Schonher

20 Mónica García Garda y Nicholsnn, 1989X ^sus impaclos_sohre_el clima iio-están.-claros,-y varios estudios han alcanzadn - conclusiones distintas. Hasta que pimío las anomalías ^n la precipitación en la zona-de jesíudio se pueden, relacionar con, los. eventos ENSO es d primer objetiva de este, capítulo. EL águiente^es estudiar la relación entre estas anomalías y la producción de pastos. Para estimar la producción primaria-de_ la.-vegetación se han utilizado, datos- qiiincenales d _ND.YI (Normahzed DifFerence Vegetation Index) del sensor AVHRR entre 1994 a 1998 comparándolos con datos agregados de prndnrdón d^ hsnn natural an Talifomia Tos e^sntos-ensqvienen caracterizados en-fímciórudela temperatura superficial del mar (SST) en las regiones NIÑO del Pacífico. Los análisis realizados revelan- que_la propagación de los eventos. ENSQ en_el clima d^la-zona no sieniprelqcurre.de lamisma forma. Sin embargo, en general, los años de "El Niño" en la Costa Central de Caüfomia se asocian con eventos, húmedos,-qua sel_desencadenan tres meses antes en el Pacífico, mientras, que los años de "La Niña" muestran patrones menos consistentes. Conforme a los datos disponibles, se observa como en los años.de EUSüño ia-y im.ligero inr.rf;mfintn m la pmdnr.c.ión de Hiomasa y retrasos en el ciclo fenológico, presentando una mayor respuesta la vegetación en los años secos. Las series,.-quincenales de -NDVI de AVHRR son. apropiadas para el análisis de. -tendencias fenológicas a esta escala, siendo la integral ^nual -un -método válido para estimar la productividad anual de las praderas. Xapirecipi tación-puede-llegar a e?íplicar. un.j50 % de la varianza temporal del^ NDVI medio en cada comunidad vegetal. Este porcentaje depende del desfase temporal utilizado. EL Capítulo 5. (García, and Ustin,-2Í)01) consiste-en una aplicación específica, orientada, a la_. evaluación del potencial de los modelos lineales de mezcla en la detección de las respuestas de la vegetación a las-variaciones Lunáticas en primavera^utilizando-datos dd sensor AYIRIS (AicbomC- Visible-Infrared Imaging Spectrometer). La zona de estudio comprende 482 ha de pradera Mediterránea natural.eajasper Ridge.BíologicaL P-reserveL(IISA).-Se-compararon los resultados dedos imágenes AVIRIS adquiridas en primavera de un año húmedo y otro medio respecto a la precipitación. - Los, resultados, riel modelo linp^il dfi mezcla muestran que las diferencias medias, en. las fracciones de mezcla para los tipos de vegetación no eran significativamente distintas entre fechas debido a. la alta, variabilidad- espacial a teavés del. pai^^e- Sin. embaigov diferencias significativas entre comiuiidades vegetales fueron encontradas entre años sobre la base del píxel, subrayando la. importancia de-análisis acecíneos de-la. locaüzación. Esto signüca cpe. una. cobertura multitemporal de imágenes hiperespectrales es clave para comprender la dinámica de la vegetación. En el Capítulo 6 se amplían los objetivos del capítulo 5, realizando un análisis multitemporal, y evaluando el uso del modelo de mezcla con los mismos parámetros en todas las imágenes y píxeles