• unlimited access with print and download
    $ 37 00
  • read full document, no print or download, expires after 72 hours
    $ 4 99
More info
Unlimited access including download and printing, plus availability for reading and annotating in your in your Udini library.
  • Access to this article in your Udini library for 72 hours from purchase.
  • The article will not be available for download or print.
  • Upgrade to the full version of this document at a reduced price.
  • Your trial access payment is credited when purchasing the full version.
Buy
Continue searching

Comparisons of MODIS vegetation index products with biophysical and flux tower measurements

Dissertation
Author: Natthanich Sirikul
Abstract:
Vegetation indices (VI) play an important role in studies of global climate and biogeochemical cycles, and are also positively related to many biophysical parameters and satellite products, such as leaf area index (LAI), gross primary production (GPP), land surface water index (LSWI) and land surface temperature (LST). In this study we found that VI's had strong relationships with some biophysical products, such as gross primary production, yet were less well correlated with biophysical structural parameters, such as leaf area index. The relationships between MODIS VI's and biophysical field measured LAI showed poor correlation at semi-arid land and broadleaf forest land cover type whereas cropland showed stronger correlations than the other vegetation types. In addition, the relationship between the enhanced vegetation index (EVI)-LAI and normalized difference vegetation index (NDVI)-LAI did not show significant differences. Comparisons of the relationships between the EVI and NDVI with tower-measured GPP from 11 flux towers in North America, showed that MODIS EVI had much stronger relationships with tower-GPP than did NDVI, and EVI was better correlated with the seasonal dynamics of GPP than was NDVI. In addition, there were no significant differences among the 1x1, 3x3 and 7x7 pixel sample sizes. The comparisons of VIs from the 3 MODIS products from which VI's are generated (Standard VI (MOD13)), Nadir Adjusted Surface Reflectance (NBAR (MOD43)), and Surface Reflectance (MOD09)), showed that MODIS NBAR-EVI (MOD43) was best correlated with GPP compared with the other VI products. In addition, the MODIS VI - tower GPP relationships were significantly improved using NBAR-EVI over the more complex canopy structures, such as the broadleaf and needleleaf forests. The relationship of tower-GPP with other MODIS products would be useful in more thorough characterization of some land cover types in which the VI's have encountered problems. The land surface temperature (LST) product were found useful for empirical estimations of GPP in needleleaf forests, but were not useful for the other land cover types, whereas the land surface water index (LSWI) was more sensitive to noise from snowmelt, ground water table levels, and wet soils than to the canopy moisture levels. Also the MODIS EVI was better correlated with LST than was NDVI. Finally, the cross-site comparisons of GPP and multi-products from MODIS showed that the relationships between EVI and GPP were the strongest while LST and GPP was the weakest. EVI may thus be useful in scaling across landscapes, including heterogeneous ones, for regional estimations of GPP, especially if BRDF effects have been taken into account (such as with the NBAR product). Thus, the relationships of EVI-GPP over space and time would potentially provide much useful information for studies of the global carbon cycle.

TABLE OF CONTENTS

LIST OF ILLUSTRATIONS …………………………………………………….….. 7

LIST OF TABLES ……………………………………………………………….….. 14

ABSTRACT ……………………………………………………………………….….. 15

I. INTRODUCTION …………………………………………………………….….. 17 Introduction and Context of the problem ..………………………………...….. 17 Objectives …………..…………………………………………………...…… 18 Overview of dissertation …………………..………………………………...….. 18

II. COMPARISON OF MODIS VEGETATION INDEX PRODUCTS WITH GROUND AND SATELLITE-BASED LAI …………………………... 20 Introduction ………………………………………………………………...….. 20 Methodology ………………………………………………………………...….. 25 A. Study area, Sample design for field measurement, and Satellite data ….. 25 B. Regression Method ……....………………..……………………………… 31 Results and Discussion ……..………………………………………………...….. 32 A. Walnut Gulch Study Site ……....………………………………………… 32 B. BigFoot Study Site ……..………………………………………………… 35 Conclusions………..………………………………………………...…………. 42

III. COMPARISON OF MULTI VEGETATION INDEX PRODUCTS WITH TOWER GPP FLUXES ……………………………...……………...….. 73 Introduction ………………………………………………………………...….. 73 Methodology ………………………………………………………………...….. 85 A. Study area ……………………………………………………………..… 85 B. Satellite data ……..……………………………………………………… 88 C. Tower-Based GPP data ............................................................................. 91 D. Time Series Method and Empirical Relationships .................................. 91 Results and Discussion ………………………………………………..……….. 93 A. Comparisons of MODIS VI’s and tower GPP Time Series vegetation Measurements……………………………………………………..…… 93 B. Comparisons between MODIS EVI-tower GPP and MODIS NDVI -tower GPP at varying sample sizes ........................................................ 97 C. Comparisons of the multi-product VI with MOD13, MOD43 and MOD09 ….…..………………………………………………………….. 103 Conclusions ..…………………………………………………………...……. 108

5

IV. SEASONAL COMPARISONS OF MODIS VEGETATION INDEX, LAND SURFACE WATER INDEX AND LAND SURFACE TEMPERATURE DATA WITH TOWER FLUXES ………….………...….. 130 Introduction ………………………………………………………………...…. 130 Methodology ………………………………………………………………...…. 136 A. Study Sites and Flux Tower Data ………..…………………………....… 136 B. Satellite Data ……………………………………………………...…....… 137 C. Phenology Profile Method and Empirical Method ……………...…....… 139 D. Multiple Site relationships …………………………………….....…....… 140 Results and Discussion ……..…………………………………………...………. 140 A. Comparisons of seasonal dynamics of biophysical and radiometric Profiles ….…..……………………………………………..…...………... 140 B. Relationships among MODIS products (VI, LSWI and LST) with GPP fluxes ……..……..……………………………………..…...………. 147 C. Relationships between MODIS VI-LSWI and VI-LST …......…...………. 150 D. Cross-site relationships between MODIS products (VI, LSWI, LST) and GPP fluxes ……...…………..………………………………..…...………. 152 Conclusions ...…………………………………………………………...……. 152

V. CONCLUSIONS …………………………………………………………….…... 169

APPENDIX A: THE COMPARISON OF RELATIONSHIP BETWEEN VI(MOD13A2) GPP(FLUX) AT 1X1, 3X3, AND 7X7 PIXEL SAMPLING SIZES FOR INDIVIDUAL YEAR DATA …………….…… 170

REFERENCES …………………………………...………………..……….….…… 181

6

LIST OF ILLUSTRATIONS

Chapter II

Figure 1: The location of Study area in Walnut Gulch, Arizona. The 2 intensive study areas are Lucky Hills (brush dominated) and Kendall (grass dominated) …. 44

Figure 2: Layout of LAI sampling at Walnut Gulch Weter Shade Experiment scheme 45

Figure 3: Shrub and Grassland were divided into 6 subgroups, each subgroup composed of 3-5 raingauges by using land classification map from USDA. LAI sampling were performed at these raingauge during July 29 th -August 6 th 2004 and August 14 th –August 25 th 2004 ……………………………………….... 46

Figure 4: The location of MODIS 1 km-pixel extracted from MOD13A2. The location of each pixel relied on the location of each raingauge that performed LAI measurement during 2 cycles of MODIS ………………………………..…. 47

Figure 5: The comparison among 3 VI-based from MODIS with Field LAI; (a) and (b) are for shrubland (c) and (d) are for grassland …………………………..…. 48

Figure 6: The graph showed seasonal profile of shrubland (a) at Lucky hill and grassland (b) at Kendall for 2004. The red arrows indicated the Field LAI was conducted during July 29 th -August 6 th 2004 and August 14 th –August 25 th 2004) for both land cover types ………………………………………………..………..…. 49

Figure 7: The comparison between LAI field-based and LAI MODIS with VI from MOD13A2 (a and b) and from MOD43B4 (c and d) for Shrubland …………………………………..…………………..………..…. 50

Figure 8: The comparison between LAI field-based and LAI MODIS with VI from MOD13A2 (a and b) and from MOD43B4 (c adn d) for grassland …….…. 51

Figure 9: BigFoot study sites composed of Konza (tallgrass prairie), Sevilleta (desert grassland), Agro (agricultural cropland), and Harvard forest (mixed broadleaf forest) ……………………………..………..…………………..………..…. 52

Figure 10: The comparison of 3 VI-based with Field LAI for Agro site; (a) EVI-Field LAI for mixed, (b) NDVI-Field LAI for mixed vegetation, (c) EVI-Field LAI for corn, (d) NDVI-Field LAI for corn, (e) EVI-Field LAI for soybean, and (f) NDVI-Field LAI for Soybean ….…….………………..………..…. 53

7

Figure 11: Comparison between 3 modes of EVI and NDVI with Field LAI for Harvard forest (a) and (b), Konza (c) and (d), and Sevilleta (e) and (f) ….. 54

Figure 12: The Field-based LAI were performed for different period of time as shown for each site; Agro (a), Harv (b), Konza (c), and (d) Sevi ……….………….... 55

Figure 13: The comparison of VI-Field LAI and VI-MODIS LAI (a) EVI and (b) NDVI from MOD13A2; (c) EVI and (d) NDVI from MOD43B4 for Agro site ... 56

Figure 14: The comparison of the relationship between VI-Field LAI and VI-MODIS LAI (a) EVI and (b) NDVI from MOD13A2; (c) EVI and (d) NDVI from MOD43B for Harvard site ….……………….………………..………..…. 57

Figure 15: The comparison of the relationship between VI-Field LAI and VI-MODIS LAI (a) EVI and (b) NDVI from MOD13A2; (c) EVI and (d) NDVI from MOD43B4 for Konza site ….…………….….………………..………..…. 58

Figure 16: The comparison of the relationship between VI-Field LAI and VI-MODIS LAI (a) EVI and (b) NDVI from MOD13A2; (c) EVI and (d) NDVI from MOD43B4 for Sevilleta site ….…….………………..…………….…..…. 59

Figure 17: The relationship of combined 4 sites, each site using the mean value averaged from each sampling period. All VI derived from MOD13A2, each figure represents as (a) EVI –Field LAI, (b) NDVI-Field LAI, (c) EVI-MODIS LAI , and (d) NDVI MODIS LAI . There were not significantly different between EVI and NDVI from MOD13 and between field LAI and MODIS LAI … 60

Figure 18: The relationship of combined 4 sites, each site using the mean value averaged from each sampling period. All VI derived from MOD43B4 each figure represents as (a) EVI –Field LAI, (b) NDVI-Field LAI, (c) EVI-MODIS LAI , and (d) NDVI MODIS LAI . There were not significantly different between EVI and NDVI from MOD13 and between field LAI and MODIS LAI … 61

Chapter III

Figure 1: Study Sites from upper left to upper right: NOBS, Lethbridge, Tonzi, Sky Oaks,Blodgett, Niwot, Willow, MMSF, Michigan, Harvard, and Howland . 109

Figure 2: Time Series of Harvard Forest site from March 2000-December 2004 ……. 110

Figure 3: Time Series of MMFS site from March 2000-December 2003 ……...……. 110

Figure 4: Time Series of Willow Creek site from March 2000-December 2004…….. 110

8

Figure 5: Time Series of Michigan site from March 2000-December 2002…….......... 110

Figure 6: Time Series of Howland site from March 2000-December 2002 ………….. 111

Figure 7: Time Series of NOBS site from March 2000-December 2004 …………….. 111

Figure 8: Time Series of Niwot site from March 2000-December 2002 …….……….. 111

Figure 9: Time Series of Blodgett site from March 2000-December 2002…..……….. 111

Figure 10: Time Series of Lehtbridge site from March 2000-December 2003……….. 112

Figure 11: Time Series of Tonzi site from January 2002- December 2003……….….. 112

Figure 12: Time Series of Sky Oaks site from March 2000- June 2002…………..….. 112

Figure13: All year comparison between EVI-GPP(a) and NDVI-GPP(b) of Harvard using data from 2000-2003 for different sampling sizes. EVI showed stronger relationship than and NDVI and 3 different sampling sizes did not show significantly different ….………………..…………..………..…. 113

Figure14: The relationship of an average 4 years from 2000-2003 of VI and GPP at Harvard Forest showed different correlation as well ….……...………..…. 113

Figure15: All year comparison between EVI-GPP(a) and NDVI-GPP(b) of MMSF using data from 2000-2003 for different sampling sizes. EVI showed stronger relationship than NDVI and 3x3 and 7x7 km pixel sampling size did not show significantly different ….………………..………………..…. 114

Figure16: The relationship of an average 4 years from 2000-2003 of VI and GPP at MMSF showed different correlation as well ….………………..…………. 114

Figure 17: All year comparison between EVI-GPP(a) and NDVI-GPP(b) of Willow using data from 2001-2003 for different sampling sizes. EVI showed stronger relationship than and NDVI and there were no significantly different among 1x1, 3x3 and 7x7 km pixel sampling sizes ….………………..………...…. 115

Figure 18: The relationship of an average 4 years from 2001-2004 of VI and GPP at Willow site showed different correlation as well ….………………..…….. 115

Figure 19: All year comparison between EVI-GPP(a) and NDVI-GPP(b) of Michigan using data from 2000-2002 for different sampling sizes. EVI showed slightly stronger relationship than and NDVI and 1x1, 3x3 and 7x7 km pixel sampling sizes did not showed significantly different ….…. 116 9

Figure 20: The relationship of an average 4 years from 2000-2002 of VI and GPP at Michigan showed different correlation as well ….………………..………. 116

Figure 21: All year comparison between EVI-GPP(a) and NDVI-GPP(b) of Howland using data from 2000-2002 for different sampling sizes. EVI clearly showed stronger relationship than and NDVI and 1x1, 3x3 and 7x7 km pixel sizes did not show significantly different ….…...………………..…. 117

Figure 22: The comparisons of the relationship between VI(MOD13A2)-GPP(Flux) For both EVI and NDVI at Howland site did not show significantly different among 1x1, 3x3, and 7x7 pixel sampling sizes ….………………..…...…. 117

Figure 23: All year comparison between EVI-GPP(a) and NDVI-GPP(b) of NOBS using data from 2000-2003 for different sampling sizes. EVI showed stronger relationship than and NDVI and 3x3 and 7x7 showed better correlation than 1x1 km sampling size ….………..…………..………..…. 118

Figure 24: The relationship of an average 4 years from 2000-2003 of VI and GPP at NOBS showed different correlation as well …..…………..………..…. 118

Figure 25: All year comparison between EVI-GPP(a) and NDVI-GPP(b) of Blodgett using data from 2000-2002 for different sampling sizes. Both EVI-GPP and NDVI-GPP showed poor correlation for all sampling sizes ….…...…. 119

Figure 26: The relationship of an average 3 years from 2000-2002 of VI and GPP at Blodgett site, EVI-GPP showed poor correlation while NDVI-GPP showed slightly correlation ….………………...…………..…………..………..…. 119

Figure 27: All year comparison between EVI-GPP(a) and NDVI-GPP(b) of Niwot using data from 2000-2003 for different sampling sizes. EVI showed stronger relationship than and NDVI and 3x3 and 7x7 showed better correlation than 1x1 km sampling size ….………………..………...…..… 120

Figure 28: The relationship of an average 4 years from 2000-2003 of VI and GPP at Niwot showed different correlation as well ….………………..………... 120

Figure 29: All year comparison between EVI-GPP(a) and NDVI-GPP(b) of Lethbridge using data from 2001-2004 for different sampling sizes. EVI and NDVI and 3 different sampling sizes did not significantly different ….…………. 121

Figure 30: The relationship of an average 4 years from 2000-2003 of EVI-GPP and NDVI-GPP at Lethbridge did not showed significantly different correlation as well ….……………………………..…………..………..…. 121

10

Figure 31: All year comparison between EVI-GPP(a) and NDVI-GPP(b) of Tonzi using data from 2002-2003 for different sampling sizes. EVI showed stronger relationship than and NDVI and 1x1, 3x3 and 7x7 did not show significantly different ….…………………………...…………..…………..………..…. 122

Figure 32: The relationship of an average 2 years from 2002-2003 of VI and GPP at Tonzi showed different correlation as well ….…..…………..………… 122

Figure 33: All year comparison between EVI-GPP(a) and NDVI-GPP(b) of Sky Oaks using data from 2000-2001 for different sampling sizes. Both EVI and NDVI showed no correlation at this site for all pixel sampling size .…………… 123

Figure 34: The relationship of an average 2 years from 2000-2001 of VI and GPP at Sky Oaks showed no correlation as well ….………………..……………….…. 123

Figure 35: The comparison of the relationship among 3 MODIS VI based-tower GPP from 3x3 MODIS pixel sampling size. The plot derived from combined year data for Harvard forest (a) EVI and (b) NDVI ; for MMFS (c) EVI and (d) NDVI ….……………………...…………..…………..………..…. 124

Figure 36: The comparison of the relationship among 3 MODIS VI based-tower GPP from 3x3 MODIS pixel sampling size. The plot derived from combined year data for Willow forest (a) EVI and (b) NDVI ; for Michigan (c) EVI and (d) NDVI ….………………………………...………..…………..………..…. 125

Figure 37: The comparison of the relationship among 3 MODIS VI based-tower GPP from 3x3 MODIS pixel sampling size. The plot derived from combined year data for Howland forest (a) EVI and (b) NDVI ; for NOBS (c) EVI and (d) NDVI ….………………………………...………..…………..………..…. 126

Figure 38: The comparison of the relationship among 3 MODIS VI based-tower GPP from 3x3 MODIS pixel sampling size. The plot derived from combined year data for Niwot forest (a) EVI and (b) NDVI ; for Blodgett (c) EVI and (d) NDVI ….………………………………...………..…………..………..…. 127

Figure 39: The comparison of the relationship among 3 MODIS VI based-tower GPP from 3x3 MODIS pixel sampling size. The plot derived from combined year data for Lethbridge (a) EVI and (b) NDVI ; for Tonzi (c) EVI and (d) NDVI ….…………………………...………....…………..………..…. 128

Figure 40: The comparison of the relationship among 3 MODIS VI based-tower GPP from 3x3 MODIS pixel sampling size. The plot derived from combined year data for Sky Oaks (a) EVI and (b) NDVI ….………………..……………. 129

11

Chapter IV

Figure 1: Flux tower uses the eddy covariance method to measure the characteristics of the surface-atmosphere exchanges of water, energy and carbon …………... 156

Figure 2: The annual average seasonal profile of 4 deciduous broadleaf forest sites (a) Harvard forest 4-year annual average (2000-2003), (b) MMSF 4-year annual average (2000-2003), (c) Willow 5-year annual average (2000-2004) (d) Michigan 3-year annual average (2000-2002)….…...……………..…….…. 157

Figure 3: The annual average seasonal profile of 4 evergreen needleleaf forest sites (a) Howland 3-year annual average (2000-2002), (b) NOBS 4-year annual average (2000-2003), (c) Niwot 4-year annual average (2000-2003) (d) Blodgett 3-year annual average (2000-2002)….………………..……………….………..…. 158

Figure 4: The annual average seasonal profile of (a) Lethbridge (grassland) 4-year annual average (2001-2004), (b) Tonzi (savanna ) 2-year annual average (2002-2003), (c) Sky Oaks 2-year annual average (2000-2001)….…….…………..….…. 159

Figure 5: The relationship among 4 MODIS products (EVI, NDVI, LSWI, LST) with tower GPP for 4 deciduous broadleaf forest ….…………………..…….…. 160

Figure 6: The relationship among 4 MODIS products (EVI, NDVI, LSWI, LST) with tower GPP for 4 needleleaf forests ….…………………..…………..…..…. 161

Figure 7: The relationship among 4 MODIS products (EVI, NDVI, LSWI, LST) with tower GPP for Lehtbridge:grassland, Tonzi: savanna and Sky Oaks: shurbland ….…………………………...…………..…………..………..…. 162

Figure 8: The comparison of the relationship between EVI-LSWI and NDVI-LSWI of 4 deciduous broadleaf forest sites: (a) Harvard forest, (b) MMSF, (c) Willow, and (d) Michigan. There were no significantly different between 2 VIs for this land cover type ….…………………..…………..………..…. 163

Figure 9: The comparison of the relationship between EVI-LST and NDVI-LST of 4 deciduous broadleaf forest sites. The correlation of EVI-LST showed slightly higher than NDVI-LST ….………………..…………..….……..…. 164

Figure 10: The comparison of the relationship between EVI-LSWI and NDVI-LSWI of 4 needleleaf forest sites. The correlation between NDVI-LSWI showed slightly stronger than EVI-LSWI for Niwot and Blodgett site ….………... 165

Figure 11: The comparison of the relationship between EVI-LST and NDVI-LST of 4 deciduous broadleaf forest sites. The correlation of EVI-LST 12

showed slightly higher than NDVI-LST for most of sites except for Blodgett site ….…………………………………..…………..………..…. 166

Figure 12: The comparison of the relationship between EVI-LSWI and NDVI-LSWI of Lethbridge (grassland), Tonzi (savanna) and Sky Oaks (shrubland). The correlation of both VI at Lethbridge and Tonzi site were the same but there were no correlation at Sky Oaks site ….……………….……..………..…. 167

Figure 13: The comparison of the relationship between EVI-LST and NDVI-LST of Lethbridge (grassland), Tonzi (savanna) and Sky Oaks (shrubland). There were no correlation for all 3 sites ….………………..………………….…. 168

Figure 14: The relationship of mean value of each product for 11 sites (by combined all data and average to derive the mean value for each point). Each figure represents: (a) EVI-GPP (b) NDVI-GPP (c) LSWI-GPP and (d) LST-GPP 169

13

LIST OF TABLES

Chapter II

Table 1: The 6 subgroups of shurbland composed 28 raingauges at Walnut Gulch …. 62

Table 2: The 6 subgroups of grassland composed 23 raingauges at Walnut Gulch ….. 63

Table 3: The correlation of Field LAI and VI (MOD13Q1/ MOD13A2 / MOD43B4) of Shrubland and Grassland at Walnut Gulch ….………………..………….….. 64

Table 4: The correlation of Field LAI and VI (MOD13A2/ MOD43B4) of Shrubland and Grassland at Walnut Gulch ….………………..………………………….….. 65

Table 5: The correlation of MODIS LAI and VI (MOD13A2/ MOD43B4) of Shrubland and Grassland at Walnut Gulch ….………………..…………………….….. 66

Table 6: Regression statistic results of MODIS VI- Field LAI at Agro (Corn) Site..…. 67

Table 7: Regression statistic results of MODIS VI- Field LAI at Agro (Soybean) Site .. 68

Table 8: Regression statistic results of MODIS VI- Field LAI at Harvard Forest Site .. 69

Table 9: Regression statistic results of MODIS VI- Field LAI at Konza Site ………... 70

Table 10: Regression statistic results of MODIS VI- Field LAI at Sevilleta Site ….…. 71

Table 11: Regression results of the mean value of combined four different land cover Types of BigFoot ….…………………………………………..……………. 72

14

ABSTRACT

Vegetation indices (VI ) play an important role in studies of global climate and biogeochemical cycles, and are also positively related to many biophysical parameters and satellite products, such as leaf area index (LAI), gross primary production (GPP), land surface water index (LSWI) and land surface temperature (LST). In this study we found that VI’s had strong relationships with some biophysical products, such as gross primary production, yet were less well correlated with biophysical structural parameters, such as leaf area index. The relationships between MODIS VI’s and biophysical field measured LAI showed poor correlation at semi-arid land and broadleaf forest land cover type whereas cropland showed stronger correlations than the other vegetation types. In addition, the relationship between the enhanced vegetation index (EVI)-LAI and normalized difference vegetation index (NDVI)-LAI did not show significant differences. Comparisons of the relationships between the EVI and NDVI with tower-measured GPP from 11 flux towers in North America, showed that MODIS EVI had much stronger relationships with tower-GPP than did NDVI, and EVI was better correlated with the seasonal dynamics of GPP than was NDVI. In addition, there were no significant differences among the 1x1, 3x3 and 7x7 pixel sample sizes. The comparisons of VIs from the 3 MODIS products from which VI’s are generated (Standard VI (MOD13)), Nadir Adjusted Surface Reflectance (NBAR (MOD43)), and Surface Reflectance (MOD09)), showed that MODIS NBAR-EVI (MOD43) was best correlated with GPP compared with the other VI products. In addition, the MODIS VI – tower GPP relationships were 15

significantly improved using NBAR-EVI over the more complex canopy structures, such as the broadleaf and needleleaf forests. The relationship of tower-GPP with other MODIS products would be useful in more thorough characterization of some land cover types in which the VI’s have encountered problems. The land surface temperature (LST) product were found useful for empirical estimations of GPP in needleleaf forests, but were not useful for the other land cover types, whereas the land surface water index (LSWI) was more sensitive to noise from snowmelt, ground water table levels, and wet soils than to the canopy moisture levels. Also the MODIS EVI was better correlated with LST than was NDVI. Finally, the cross- site comparisons of GPP and multi- products from MODIS showed that the relationships between EVI and GPP were the strongest while LST and GPP was the weakest. EVI may thus be useful in scaling across landscapes, including heterogeneous ones, for regional estimations of GPP, especially if BRDF effects have been taken into account (such as with the NBAR product). Thus, the relationships of EVI-GPP over space and time would potentially provide much useful information for studies of the global carbon cycle. 16

I. INTRODUCTION Introduction and Context of the problem The Moderate Resolution Imaging Spectroradiometer (MODIS), onboard NASA's Earth Observing System Terra platform, is designed for monitoring global vegetation biophysical and radiation budget parameters that influence carbon, water and surface energy fluxes. MODIS has been operating since 2000 and has generated many land data products, including vegetation indices (VI’s). VI’s play an important role in studies of global climate and biogeochemical cycles, and are also positively related to many biophysical products and the other spectral indices, such as leaf area index (LAI), land surface water index (LSWI) and land surface temperature (LST) etc.. In addition, VI’s are utilized in several biogeochemical models that are intended for monitoring seasonal and spatial patterns in photosynthetic activity, or gross primary production (GPP). Since VIs are very precise indices and strongly correlate with many biophysical characteristics. Many studies have investigated in these relationships with biophysical and biogeochemical products by using empirical-based approaches. Nevertheless, uncertainties in the MODIS VI’s are present, resulting from variations in sun-sensor geometries, soil background, and canopy structural differences, as well as uncertainties due to corrections for atmospheric effects and also the chemical, water content, and structure dynamics of vegetation canopies, which vary with seasonal and environmental conditions. Moreover, the relationship between VI’s and biophysical products enable scaling- up from canopy stand levels to large areas and across many land cover types. As satellite 17

data is increasingly being used to investigate climate change and forecast ecological change, it becomes important to validate and better characterize satellite data and investigate their relationships with sites with known surface conditions and biophysical quantities. Accordingly, validation of satellite products are difficult, as is characterizing their error and uncertainties. Thus, the improvement of VI’s by minimizing noise and uncertainties would significantly improve the quality and accuracy of biophysical retrievals and render their use more valuable for global carbon cycle and climate change studies. Objectives The main objective of this study is to investigate the empirical relationships between MODIS EVI and MODIS NDVI with several biophysical products, including field-measured canopy properties, tower-based canopy fluxes, and satellite-derived vegetation products. We also investigated and compared the differences in the biophysical relationships derived from three different MODIS products that enable computation of VI’s, namely the standard VI (MOD13), the nadir adjusted surface reflectance product (MOD43), and the surface reflectance (MOD09). Overview of dissertation MODIS VIs include the normalized difference vegetation index (NDVI) and the enhanced vegetation index (EVI). Whereas the NDVI is chlorophyll sensitive, the EVI is more responsive to canopy structural variations, including leaf area index (LAI), canopy type, plant physiognomy, and canopy architecture. Thus this study we compared the 18

relationship between 2 VIs with biophysical parameters by separating into 3 chapters as the following: In chapter II we investigated the relationships between EVI and NDVI, with field measured LAI and also evaluated how nadir-adjusted VI’s improve upon the uncertainties generated from the standard VI’s. In the chapter III, we explored the relationships and interannual variations of MODIS VI’s and tower-GPP fluxes, compared the relationships between EVI and NDVI with tower GPP, and also explored the pixel sampling size variability in VI-GPP relationships in order to find the best spatial scale that coincide with flux tower footprints. Moreover, we made comparisons of VI’s generated from the 3 different VI-products (MOD09, MOD13 and MOD43) to assess the best satellite- tower relationships. For chapter IV, we analyzed the seasonal dynamics and compared the relationships among multiple MODIS products that describe canopy states and processes, including the EVI, NDVI, Land Surface Water Index, and Land Surface Temperature products, with tower flux GPP measurements, and then compared their interrelationships to each other. We also investigated the global, cross-site relationships of the VI’s and biophysical products.

19

II. COMPARISON OF MODIS VEGETATION INDEX PRODUCTS WITH GROUND AND SATELLITE-BASED LAI Introduction Vegetation indices (VIs) play important roles in studies of global climate and biogeochemical cycles, especially for the case of carbon, with about a quarter of atmospheric carbon dioxide potentially fixed as gross primary production by terrestrial vegetation annually (Myneni et al., 1997). Spectral transformations of two or more wavebands (especially visible and near-infrared wavelengths) are combined to formulate spectral vegetation indices, which allow measurement of spatial and temporal variations in terrestrial vegetation photosynthetic activity and canopy structure. VIs are computed without any bias or assumptions regarding land cover class, soil type, or climatic conditions (Huete at al., 2002) and thus provide a precise and continuous measure of seasonal, inter annual dynamics of vegetation structural, phenological, and biophysical parameters. Therefore, VIs are important parameters to various kinds of local, and global scale models, including agricultural and rangeland growth models, general circulation models, and biogeochemical models. They are also utilized in various operational applications such as famine early warning systems, land cover classification, health and epidemiology, drought detection, land degradation, deforestation, change detection and monitoring (http://tbrs.arizona.edu/cdrom/VI_Intro/VI_Introduction.html ). In addition, VIs have also been shown to be well correlated with vegetation parameters such as green cover fraction, biomass, and the two key variables required in primary production and global climate studies namely leaf area index (LAI) and the fraction of absorbed 20

photosynthetically active radiation (fAPAR). (Tucker, 1979; Asrar et al., 1989; Sellers, 1985). Since VIs are very precise indices and strongly correlated with many biophysical characteristics, many studies have investigated their relationships with biophysical parameters. One of the most extensively applied relationships are between VIs and LAI. LAI is a key biophysical variable influencing vegetation photosynthesis, transpiration, and the energy balance of the land surface (Bonan, 1995; Running, 1990). Moreover, LAI is an important parameter in ecosystem process models and carbon and hydrologic cycle models (Gower et al., 1999). Consequently, the relationship of VI-LAI is potentially useful in measurement and monitoring of land surface characteristics especially, in the analysis of large-scale changes and for studying of global phenomena. In early research, Jordan (1969) found a strong correlation between red and near- infrared transmittance ratio and LAI. Also, the ratio vegetation index (RVI) (Pearson and Miller, 1972), normalized difference vegetation indices (NDVI), and the perpendicular vegetation index (PVI) (Richardson and Wiegand, 1977) have been found to be well correlated with various vegetation variables such as green leaf area (Wiegand et al., 1979; Holben et al., 1980; Asrar et al., 1984, 1985; Hatfield et al., 1985; Clevers, 1989), the amount of leafy biomass or LAI (Tucker, 1979; Elvidge and Lyon, 1985), percent ground cover, amount of photosynthetically active tissue (Wiegand et al. 1984), and photosynthetic activity (Choudhury, 1987; Hatfield et al., 1984; Sellers, 1985; 1987). Many studies have established the relationships between LAI and remote sensing data (Badhwar et al., 1986; Peterson et al., 1987; Turner et al., 1999) by relying on empirical 21

relationships between the ground-measured LAI and observed spectral responses (Curran et al., 1992; Peddle et al., 1999). In 1995, Myneni et al. used the empirical method to provide the theoretical interpretation of the relationship between vegetation indices and LAI. In addition, many researchers have applied empirical methods on various vegetation types, e.g., grasslands (Friedl et al., 1994), shrublands (Law & Waring, 1994), agroecosystems (Cohen et al., 2003), conifer forests (Chen & Cihlar, 1996; Cohen et al., 2003) and broadleaf forests (Fassnacht et al., 1997). However, the retrieval of LAI using empirical approaches in establishing relationships between VIs and biophysical parameters, such as LAI, may be fail in the case where there are external influences associated with variable solar and viewing geometries, soil background, chlorophyll concentrations, or moisture conditions are different (Jacquemoud et al., 1995). Some researchers have found uncertainties between VI-LAI relationships, Sellers (1985) found that vegetation indices approach a saturation level asymptotically for a certain range of LAI and respond linearly to fAPAR. Baret and Guyot (1991) investigated the relationships between VI and LAI or photosynthetic active radiation by using the SAIL reflectance model and found the asymptotic trend of the VI- LAI when LAI became greater than 3. Moreover, many studies have concluded that the relationships of VI to LAI/fAPAR are dependent on canopy structure, land cover, leaf angle distribution, vegetation clumping, row orientation, spacing, and the optical properties of the canopy components (Asrar et al., 1992; Baret and Guyot, 1991; Choudhury, 1987; Goward and Huemmrich, 1992; Roujean and Breon, 1995). Different canopy types show drastic variations in canopy structure and reflectance properties, 22

Full document contains 198 pages
Abstract: Vegetation indices (VI) play an important role in studies of global climate and biogeochemical cycles, and are also positively related to many biophysical parameters and satellite products, such as leaf area index (LAI), gross primary production (GPP), land surface water index (LSWI) and land surface temperature (LST). In this study we found that VI's had strong relationships with some biophysical products, such as gross primary production, yet were less well correlated with biophysical structural parameters, such as leaf area index. The relationships between MODIS VI's and biophysical field measured LAI showed poor correlation at semi-arid land and broadleaf forest land cover type whereas cropland showed stronger correlations than the other vegetation types. In addition, the relationship between the enhanced vegetation index (EVI)-LAI and normalized difference vegetation index (NDVI)-LAI did not show significant differences. Comparisons of the relationships between the EVI and NDVI with tower-measured GPP from 11 flux towers in North America, showed that MODIS EVI had much stronger relationships with tower-GPP than did NDVI, and EVI was better correlated with the seasonal dynamics of GPP than was NDVI. In addition, there were no significant differences among the 1x1, 3x3 and 7x7 pixel sample sizes. The comparisons of VIs from the 3 MODIS products from which VI's are generated (Standard VI (MOD13)), Nadir Adjusted Surface Reflectance (NBAR (MOD43)), and Surface Reflectance (MOD09)), showed that MODIS NBAR-EVI (MOD43) was best correlated with GPP compared with the other VI products. In addition, the MODIS VI - tower GPP relationships were significantly improved using NBAR-EVI over the more complex canopy structures, such as the broadleaf and needleleaf forests. The relationship of tower-GPP with other MODIS products would be useful in more thorough characterization of some land cover types in which the VI's have encountered problems. The land surface temperature (LST) product were found useful for empirical estimations of GPP in needleleaf forests, but were not useful for the other land cover types, whereas the land surface water index (LSWI) was more sensitive to noise from snowmelt, ground water table levels, and wet soils than to the canopy moisture levels. Also the MODIS EVI was better correlated with LST than was NDVI. Finally, the cross-site comparisons of GPP and multi-products from MODIS showed that the relationships between EVI and GPP were the strongest while LST and GPP was the weakest. EVI may thus be useful in scaling across landscapes, including heterogeneous ones, for regional estimations of GPP, especially if BRDF effects have been taken into account (such as with the NBAR product). Thus, the relationships of EVI-GPP over space and time would potentially provide much useful information for studies of the global carbon cycle.