ABSTRACT
Title of Document: QUANTIFICATION OF ERROR IN AVHRR 
NDVI DATA
Jyoteshwar R. Nagol, Ph.D., 2011
Directed By: Dr. Stephen Prince
Department of Geography
Several influential Earth system science studies in the last three decades were 
based on Normalized Difference Vegetation Index (NDVI) data from Advanced Very 
High Resolution Radiometer (AVHRR)  series of instruments.  Although AVHRR 
NDVI data are known to have significant uncertainties resulting from incomplete 
atmospheric correction, orbital drift, sensor degradation, etc., none of these studies 
account for them. This is primarily because of unavailability of comprehensive and 
location-specific quantitative uncertainty estimates.
The first part of this dissertation investigated the extent of uncertainty due to 
inadequate atmospheric correction in the widely used AVHRR NDVI datasets. This 
was accomplished by comparison with atmospherically corrected AVHRR data at 
AErosol RObotic NETwork (AERONET) sunphotometer sites in 1999. Of the 
datasets included in this study, Long Term Data Record (LTDR) was found to have 
least errors (precision=0.02 to 0.037 for clear and average atmospheric conditions) 
followed by Pathfinder AVHRR Land (PAL) (precision=0.0606 to 0.0418), and Top 
of Atmosphere (TOA) (precision=0.0613 to 0.0684).  
` Although the use of field data is the most direct type of validation and is used 
extensively by the remote sensing community, it results in a single uncertainty 
estimate and does not account for spatial heterogeneity and the impact of spatial and 
temporal aggregation. These shortcomings were addressed by using Moderate 
Resolution Imaging Spectrometer (MODIS) data to estimate uncertainty in AVHRR 
NDVI data. However, before AVHRR data could be compared with MODIS data, the 
nonstationarity introduced by inter-annual variations in AVHRR NDVI data due to 
orbital drift had to be removed. This was accomplished by using a Bidirectional 
Reflectance Distribution Function (BRDF) correction technique originally developed 
for MODIS data. 
The results from the evaluation of AVHRR data using MODIS showed that in 
many regions minimal spatial aggregation will improve the precision of AVHRR 
NDVI data significantly. However temporal aggregation improved the precision of 
the data to a limited extent only.
The research presented in this dissertation indicated that the NDVI change of 
~0.03 to ~0.08 NDVI units in 10 to 20 years, frequently reported in recent literature, 
can be significant in some cases. However, unless spatially explicit uncertainty 
metrics are quantified for the specific spatiotemporal aggregation schemes used by 
these studies, the significance of observed differences between sites and temporal 
trends in NDVI will remain unknown.
QUANTIFICATION OF ERROR IN  AVHRR NDVI DATA
By
Jyoteshwar R. Nagol
Dissertation submitted to the Faculty of the Graduate School of the 
University of Maryland, College Park, in partial fulfillment
of the requirements for the degree of
Advisory Committee:
Dr. Stephen D. Prince, Chair
Dr. Eric Vermote,
Dr. Shunlin Liang
Dr. Rachel T. Pinker
Doctor of Philosophy
2011
? Copyright by
Jyoteshwar R. Nagol
2011
Preface
Parts of the material presented in this dissertation have already been published 
in peer-reviewed journals. The results from chapter 2 have been published in Remote 
Sensing of Environment, while material presented in chapter 3 is being revised for 
publication in the Canadian Journal of Remote Sensing. The manuscript presenting 
results from chapter 4 is under preparation for publication in Remote Sensing of 
Environment.
All the three papers have Dr. Vermote and Dr. Prince as second and third 
author respectively. I was responsible for all of the analysis presented in these papers 
as well as the manuscript development. Dr. Vermote and Dr. Prince provided detailed 
guidance in development of the research and in the writing process.
ii
Acknowledgements
I am particularly grateful to my Advisor, Dr. Stephen Prince and co-advisor, 
Dr Eric Vermote. Dr. Prince helped in the development of the vision and context for 
my dissertation, while Dr. Vermote provided detailed guidance and tools for actual 
implementation of my ideas. Both of them have also contributed immensely in my 
development as a researcher. I am also thankful to other members of my advisory 
committee ? Dr. Shunlin Liang and Dr. Chengquan Huang.  
I deeply appreciate all the support and encouragement provided by my friends 
and colleagues like Khaldoun Rishmawi, Nancy Thomas, and Ritvik Sahajpal. I am 
also thankful to NASA?s Earth and Space Science Fellowship Program for providing 
financial support for this research.
iii
Table of Contents
ABSTRACT...................................................................................................................I
PREFACE......................................................................................................................II
ACKNOWLEDGEMENTS.........................................................................................III
TABLE OF CONTENTS.............................................................................................IV
LIST OF TABLES.......................................................................................................VI
LIST OF FIGURES....................................................................................................VII
CHAPTER 1: INTRODUCTION..................................................................................1
1.1. Overview.............................................................................................................1
1.2. Background.........................................................................................................3
1.2.1. Normalized Difference Vegetation Index (NDVI).......................................3
1.2.2. AVHRR Sensor............................................................................................4
1.2.3. Critical Aspects of Data from AVHRR Instruments....................................7
1.2.3.1. Sensor Degradation...............................................................................7
1.2.3.2. Orbital Drift..........................................................................................9
1.2.3.3. Atmospheric Correction......................................................................11
1.2.4. Publicly available processed AVHRR NDVI datasets...............................13
1.2.4.1. Long-Term Data Record (LTDR)........................................................16
1.2.4.2. Global Inventory Modeling and Mapping studies (GIMMS).............17
1.2.5. Uncertainty in Remotely Sensed Quantitative Data .................................18
1.2.6. Past Studies on uncertainty in AVHRR NDVI data...................................20
1.2.7. Precision Assumed for AVHRR NDVI data..............................................21
1.2.8. Statistical Framework for Estimation of Uncertainty in Remotely Sensed 
Data......................................................................................................................22
1.3. Research Goals and Dissertation Structure.......................................................25
CHAPTER 2: INVESTIGATION OF ERRORS DUE TO ATMOSPHERIC 
COMPOSITION..........................................................................................................27
2.1. Introduction.......................................................................................................27
2.2. Data and Methods.............................................................................................27
2.2.1. Approach....................................................................................................27
2.2.2. Reference Data...........................................................................................29
2.2.3 Simulated AVHRR time series....................................................................32
2.3. Results and Discussion ....................................................................................32
2.4. Conclusions ......................................................................................................38
CHAPTER 3: INVESTIGATION OF IMPACT OF ORBITAL DRIFT ON LONG-
TERM AVHRR NDVI TIME SERIES DATA.............................................................40
3.1. Introduction.......................................................................................................40
3.2. Data and methods..............................................................................................40
3.2.1. Impact of Orbital Drift on AVHRR NDVI Time Series.............................40
 3.2.2. Effectiveness of MODIS BRDF parameters for BRDF Correction of 
AVHRR NDVI Time Series Data........................................................................42
3.2.2.1. Study Sites..........................................................................................44
3.2.2.2. The MODIS BRDF Correction Technique.........................................44
iv
3.2.2.3. Modeled AVHRR TOC reflectance data.............................................45
3.3. Results and Discussion.....................................................................................47
3.3.1. Impact of Orbital Drift on AVHRR NDVI Time Series.............................47
3.3.2. Effectiveness of MODIS BRDF parameters for BRDF correction of 
AVHRR NDVI time series data...........................................................................50
3.4. Conclusions.......................................................................................................54
CHAPTER 4: QUANTIFICATION OF OVERALL UNCERTAINTY AND IMPACT 
OF SPATIAL AND TEMPORAL AGGREGATION ..................................................56
4.1. Introduction.......................................................................................................56
4.2. Data and methods..............................................................................................57
4.2.1. AVHRR data...............................................................................................59
4.2.2. Reference data (MODIS Aqua data)..........................................................60
4.2.3. Impact of Spatial and Temporal Aggregation on AVHRR NDVI 
Uncertainty (Random Error)................................................................................60
4.2.4. Estimation of Location Specific Uncertainty and Evaluation of the 
Significance of Interannual Trends......................................................................61
4.3. Results...............................................................................................................64
4.3.1. Impact of Spatial and Temporal Aggregation on AVHRR NDVI Uncertainty 
(Random Error)........................................................................................................64
4.4. Discussion.........................................................................................................71
4.5. Conclusions.......................................................................................................72
CHAPTER 5: OVERVIEW AND DISCUSSION.......................................................74
5.1. Overview of the Results....................................................................................74
5.2. Global Uncertainty Maps..................................................................................79
5.3. Impact of Uncertainty in AVHRR-NDVI Data on Inter-annual Trends........83
5.4. Implications of the results.................................................................................84
5.5. Future Directions..............................................................................................87
BIBLIOGRAPHY........................................................................................................90
v
List of Tables
Table 1-1. Summary of dataset characteristics for some AVHRR NDVI datasets ... 15
Table 1-2: Flow of the research activities within the dissertation structure ???. 25
Table 2-1: Systematic-bias and uncertainty for red channel (CH1), near infrared 
channel (CH2), and NDVI for TOA, PAL, and LTDR data. Of the 580 data points 168 
(29%) were clear, 360 (62%) were average, and 52 (9%) were hazy ????. 34
Table 2-2: Systematic-bias, and Uncertainty for NDVI from simulated TOA, PAL, 
and LTDR datasets. For the forest site; 20 (4%), 376 (75%), and 105 (21%) of 501 
data points had clear, average, and hazy atmospheres respectively. For the savanna 
site; 40 (9%), 317 (71%), and 89 (20%) of 446 data points had clear, average, and 
hazy atmosphere respectively. For the semi-arid site; 384 (52%), 348 (47%), and 8 
(1%) of 740 data points had clear, average, and hazy atmosphere respectively ?? 35
Table 2-3: Impact of temporal maximum value compositing (MVC) on Systematic-
bias, and Uncertainty for simulated TOA NDVI time series data ???????. 36
Table 3-1: Vegetation structural and optical information used to parameterize the 
flight model for forests at two sites depicted in Fig. 3-1 ??????????... 46
Table 3-2: Yearly rate of change in modeled AVHRR NDVI data for 5 years of 
NOAA-14 (Fig. 3-3). These trends were calculated by fitting linear regression lines to 
yearly average NDVI ????????????????????????. 48
Table 4-1: Locations and land cover of the sites in Australia, Sahel region of Africa, 
and Northern South America where further study of impact of spatiotemporal 
aggregation on AVHRR NDVI uncertainty and its implication for study of inter-
annual NDVI time series trends was carried out ?????????????... 59
Table 5-1: Minimum Valid NDVI change for various degrees of NDVI uncertainty 
for 10 and 20 year time period ????????????????????... 84
vi
List of Figures
Fig. 1-1: NOAA AVHRR Global Area Coverage (GAC) data are generated by 
averaging four out of every five samples (LAC pixels) along every third scan-line 
?????????????????????????????????... 6
Fig. 1-2: AVHRR NOAA-14 spectral response function ??????????.. 12
Fig.  1-3: Statistical framework used for calculation of random-error.. This analysis 
used non-composited daily NDVI data from 2003 to 2006 ?????????... 23
Fig. 2-1: Schematic view of the data processing for reference data ??????.. 30
Fig. 2-2: Spatial distribution of AERONET sites used to calculate reference data. 
After masking for cloud, cloud shadow, and view zenith angle greater than 42?, only 
580 data points from 48 AERONET sites remained for use in calculation of the 
reference data. The number of data points is indicated by the diameter of the circle 
(see inset) ????????????????????????????... 31
Fig. 3-1: Aerial views of sites used for BRDF correction. Site 1 has 50% vegetation 
cover and is located in south-western Australia (116.167?E, 32.1944?S). Site 2, with 
95% vegetation cover, is located in north-eastern Australia (145.8?E, 17.4?S) ?? 43
Fig. 3-2: The effects of forest vegetation cover on the relationship between solar 
zenith angle (SZA) and top of the canopy (TOC) NDVI ??????????.. 47
Fig. 3-3: Impact of orbital drift on AVHRR data modeled for TOA and TOC NDVI 
data for first 5 years of NOAA-14 satellite?s operational life (1995 to 1999) with 
various degrees of vegetation cover at 0?, 30?, and 60? latitudes. These time series 
were produced using FLIGHT canopy radiative transfer model and 6S atmospheric 
radiative transfer program. The simulation assumes evergreen vegetation cover and 
AOT = 0.15.  Note: The high frequency variation observed in SZA, which is due to 
inclusion of data from multiple orbits, causes a similar high frequency variation in 
NDVI time series. The thick line represents a 10 day mean NDVI and the thinner 
lines (not always separable from the mean) represent the maximum and minimum 
value of NDVI for the same 10 day period ???????????????? 49
Fig. 3-4: The effect of MODIS BRDF correction technique in removing the impact of 
orbital drift on top of canopy (TOC) NDVI data. On the left are the data without the 
BRDF correction and on the right are the data with the MODIS BRDF correction. 
Red, near infrared and NDVI are shown for 5 years of AVHRR data. The inter-annual 
NDVI trend was estimated by fitting a linear polynomial to yearly NDVI averages .52
Fig. 3-5: This figure shows the impact of different levels of correction on AVHRR 
NDVI time series data. It can be observed that atmospheric correction of LTDR 
NDVI data removes negative trends introduced by atmosphere and exposes positive 
trends due to surface BRDF. It also shows the effectiveness of MODIS BRDF 
vii
correction technique in removing the positive trends introduced by surface BRDF. 
The inter-annual NDVI trend was estimated by fitting a linear polynomial to yearly 
NDVI averages ??????????????????????????... 53
Fig. 4-1: The three regions (Australia, Sahel, and northern South-America) and the 
locations of the sites (Table 4-1) where further study of the impact of spatial and 
temporal aggregation on uncertainty in AVHRR NDVI data was conducted ??... 58
Fig. 4-2:  Impact of spatial aggregation on uncertainty in LTDR AVHRR NDVI data. 
The spatial aggregation was performed in the spectral domain before calculation of 
NDVI ??????????????????????????????. 64
Fig. 4-3: Impact of various types of monthly temporal compositing techniques on 
uncertainty in LTDR AVHRR NDVI data ???????????????? 66
Fig. 4-4: Impact of the temporal extent of aggregation on uncertainty in LTDR 
AVHRR NDVI data ????????????????????????... 67
Fig. 4-5:  The LTDR AVHRR NDVI uncertainty at the 10 study sites (Table 4-1) . 68
Fig. 4-6: Total NDVI change observed in the 18 year time period of 1982 to 1999. 
The red lines represent the uncertainty ?????????????????.. 70
Fig. 5-1: Summary of results from Chapter 2. Uncertainty in NDVI for TOA, PAL, 
and LTDR datasets (Table 2-1) ????????????????????.. 75
Fig. 5-2: Examples of orbital drift related spurious inter-annual NDVI trends. The 
green, blue, and black colored data represent MODIS Aqua NDVI data, AVHRR 
NDVI data, and the difference between these two data (AVHRR NDVI - MODIS 
NDVI) respectively. The thick red line represents the spurious inter-annual trend. The 
magnitude of the spurious NDVI trend was determined by estimating slope of the 
linear curve fitted to the LTDR AVHRR NDVI and MODIS-Aqua data differences. 
Site-1 (115.05E, 34.25N) is located in vegetated hilly regions of northeastern 
Himalayan range. Site-2 (101.188E, 28.04N) has mixed agricultural and urban/built-
up land cover and is located in Eastern Henan province of China ???????.78
Fig. 5-3: Global uncertainty map of 3x3 pixel spatial aggregates (averages) for LTDR 
AVHRR NDVI data. Deserts, urban areas, wetlands, and the pixels with less than 30 data 
points were excluded from the analysis ????????????????????. 80
Fig. 5-4: Land cover map derived from MODIS data (MOD12C1) for year 2004 with 
IGBP classification schemes ?????????????????????.. 80
Fig. 5-5: Mean uncertainty of each land cover type. Deserts, urban areas, wetlands, 
and the pixels with less than 30 data points were excluded from the analysis ??.. 82
Fig. 5-6: Bioclimatic subzones in the Arctic slope of Alaska investigated by Jia et al., 
(2003) ??????????????????????????????. 86
viii
Chapter 1: Introduction
1.1. Overview
The  Advanced  Very  High  Resolution  Radiometer  (AVHRR)  is  a 
meteorological sensor system onboard the National Oceanic and Atmospheric 
Administration (NOAA) series of polar-orbiting satellites (Kidwell 1998). The 
AVHRR sensor collects data in the visible, near infrared and thermal infrared wave 
lengths. After its capability for measuring Earth surface properties was demonstrated 
(Townshend and Tucker 1981; Tucker et al. 1984), these data have been used 
extensively for Earth system science research. The most useful features of the 
AVHRR dataset are: its temporal extent of nearly three decades; appropriate 
resolution for global and regional scale monitoring (finest resolution of local area 
coverage 1km2, and global area coverage 4.4km2); and the extension of the record by 
next generation sensors, like Visible Infrared Imager/radiometer Suite (VIIRS) 
(Tucker et al. 2005; van Leeuwen et al. 1999).
Many Earth surface vegetation studies make use of the visible and near-
infrared channels of the AVHRR in the form of vegetation indices, chiefly the 
Normalized Difference Vegetation Index (NDVI). Many other AVHRR derived data 
products also use NDVI as a primary input, e.g., global land cover maps (DeFries et 
al. 1995; DeFries et al. 1999), Net Primary Production (NPP) (Prince and Goward 
1995),  burned  area  product  (Barbosa  et  al.  1999),  Fraction  of  Absorbed 
Photosynthetically Active Radiation (fPAR)  (Myneni et al. 1997b), Leaf Area Index 
1
(LAI) (Myneni et al. 1997b), land surface temperature (Jin 2004), and air temperature 
(Prihodko and Goward 1997).
Several influential Earth system science studies in the last three decades were 
based on NDVI data from AVHRR series of instruments. For example Myneni et al., 
(1997) and Tucker et al., (2001) used AVHRR NDVI data to show an increase in plant 
growth and lengthening of the growing season in northern high latitudes. Tucker et 
al., (1991) used these data to estimate the spatial extent of the Sahara Desert and its 
inter-annual variation.  Slayback et al., (2003) found consistent greening in global 
latitudinal bands from 35? to 75? N for years 1982-1999. There are many more such 
examples of AVHRR NDVI based studies, which are considered significant 
contributions and are cited widely by the scientific community (Anyamba and 
Eastman 1996; Bounoua et al. 2000; Cao et al. 2004; Fang et al. 2003; Hicke et al. 
2002; Ichii et al. 2002; Jia et al. 2003; Lambin and Ehrlich 1997; Lee et al. 2002; Liu 
et al. 1994; Lucht et al. 2002; Myneni et al. 1997a; Paruelo et al. 2004; Piao et al. 
2004; Piao et al. 2003; Runnstrom 2000; Slayback et al. 2003; Stockli and Vidale 
2004; Tucker et al. 2001; Vicente-Serrano and Heredia-Laclaustra 2004; Yu et al. 
2003; Zhou et al. 2003; Zhou et al. 2001).
Although, AVHRR NDVI data are known to have significant uncertainties 
resulting from shortcomings like incomplete atmospheric correction, orbital drift, 
sensor degradation (El Saleous et al. 2000), none of these studies account for them. 
This is primarily because of unavailability of comprehensive and location specific 
quantitative uncertainty estimates. 
2
Most previous investigations of uncertainty in AVHRR NDVI data (Cihlar et 
al. 1998; Eklundh 1995; Fensholt et al. 2006a; Fensholt et al. 2006b; Gallo et al. 
2005; Goward et al. 1993; Nagol et al. 2009) have focused on individual sources of 
error and did not account for the impact of spatial and temporal aggregation. The 
current work was primarily motivated by this deficiency. The overall guiding 
principle of this research was to increase the value of AVHRR NDVI data for Earth 
science research. 
1.2. Background
1.2.1. Normalized Difference Vegetation Index (NDVI)
    ---  Eq: 1-1
NDVI (Eq:1-1) was first suggested by Rouse et al., (1973) and is one of the 
earliest and most popular vegetation indices. It attempts to decrease the atmospheric 
and surface Bidirectional Reflectance Distribution Function (BRDF) effects by 
normalizing the difference between red and NIR by total radiation. (Rouse et al. 1973)
NDVI has been shown to have a linear relationship with fPAR, from which 
important Earth surface biophysical variables like LAI and NPP can be derived 
(Myneni et al. 1997b; Prince and Goward 1995; Tucker 1979). This correlation is 
strongest when a vegetation canopy is neither dense nor sparse. When vegetation 
cover is too sparse, the background signal (e.g., soil color/moisture) can change 
3
NDVI significantly (Huete and Jackson 1987). However, when vegetation cover is 
too dense, NDVI starts to saturate at LAI ? 2 and no longer correlates well with one 
widely-used aspect of the surface vegetation condition, that is LAI (Hatfield et al. 
1985).
Variations in atmospheric composition (in particular aerosols and water vapor) 
affect the satellite based NDVI measurements significantly. Presence of atmospheric 
aerosols increases red reflectance, which in turn decreases NDVI. In AVHRR sensors, 
the broad spectral response functions the presence of atmospheric water vapor also 
decrease NDVI by causing a decrease in NIR reflectance (El Saleous et al. 2000). 
Presence of sub-pixel or thin cirrus clouds are difficult to detect and can also 
significantly  contaminate  NDVI  measurements  of  vegetation  and  lead  to 
misinterpretations. Surface and atmospheric BRDF is another source of error in 
NDVI signal (El Saleous et al. 2000; Kaufmann et al. 2000). For these reasons, use of 
satellite based NDVI data requires greater caution than is generally the case.
1.2.2. AVHRR Sensor
The AVHRR was designed in the mid 1970s for meteorological purposes. 
From 1978 onwards three versions of AVHRR instrument have been flown on the 
NOAA series of polar-orbiting satellite systems (Kidwell 1998). The first version of 
AVHRR instrument, which had a four band configuration, was carried on the TIROS-
N satellite platform (launched October 1978). This was subsequently improved to a 
five-channel instrument (AVHRR/2) that was initially carried on NOAA-7 (launched 
June 1981). The latest version is a six channel instrument (AVHRR/3), which was 
4
first carried on the NOAA-15 satellite platform launched in May 1998 (Robel 2007). 
Of these three versions, data from the AVHRR/2 and AVHRR/3 instruments are 
extensively used by the Earth system science community.
In the AVHRR/2 configuration, the AVHRR instruments collect data in a red 
(0.58?0.68 ?m), a NIR (0.725?1.10 ?m), a mid-infrared (3.55?3.93 ?m) and two 
thermal (10.5?11.3 ?m & 11.5?12.5 ?m) spectral channels. Starting with the 
AVHRR/3 the sensor operation alternated between bands centered on 1.6 ?m (channel 
3a) during the day and on 3.75 ?m (channel 3b) at night, unfortunately disrupting 
continuity of the channel 3 data, but the historical configuration returned in May 2003 
when NOAA-16 again operated channel 3b continuously (Robel 2007).
In addition, the AVHRR/3 instruments allow improved low energy/light 
detection by incorporation of dual gain sensor characteristics for visible and NIR 
channels. The first half of the dynamic range of an AVHRR/3 channel represents 0 to 
25% of the maximum detectible radiance at sensor, while the remaining 75% is 
represented by the second half of the dynamic range (Heidinger et al. 2002). 
5
Fig. 1-1: NOAA AVHRR Global Area Coverage (GAC) data are generated by 
averaging four out of every five samples (LAC pixels) along every third scan-line.
The spatial resolution of the AVHRR data is approximately 1.1 km (~0.01? 
Local Area Coverage (LAC) data at nadir. Due to onboard resource limitations, the 
processor on board the satellite samples the real-time AVHRR data to produce the 
reduced resolution Global Area Coverage (GAC) data. GAC data are generated by 
averaging four out of every five samples (LAC pixels) along every third scan-line 
(Fig. 1-1). Although, the resultant data are treated as 4 km resolution, the actual 
spatial resolution of a GAC data pixel is 1.1x4.4 km with a gap of 3.3 km between the 
scan lines (Kidwell 1998). The 10-bit radiometric precision of the AVHRR data is 
retained.
6
1.2.3. Critical Aspects of Data from AVHRR Instruments
For long-term monitoring of land surface biophysical variables, it is essential 
that the data are well calibrated and exhibit a consistent dynamic range throughout the 
time series. However, AVHRR instruments lack onboard calibration for both red and 
NIR channels, the data is inherently difficult to correct for atmospheric interference 
(El Saleous et al. 2000), and the systematic changes in sun-sensor geometry due to 
satellite orbital drift (temporal precession of the equatorial crossing-time) add 
additional difficulty for the data users.
1.2.3.1. Sensor Degradation
The sensor calibration process relates radiometer response of the instrument to 
incoming radiation flux. The AVHRR instruments are thoroughly calibrated only 
before satellite launch. Past studies have shown that the calibration coefficients of 
reflectance channels change abruptly after launch and then continue changing slowly 
throughout the lifetime of satellites. These changes have generally been attributed to 
changes in sensor response as well as distortion of the spectral response function due 
to sensor degradation. The lack of onboard calibration to monitor these changes 
makes post-launch vicarious calibration mandatory and needs to be continuously 
updated (Abel et al. 1993; Gorman and McGregor 1994; Gutman 1999; Heidinger et 
al. 2002; Holben et al. 1990; Rao and Chen 1995; Rao and Chen 1996; Teillet et al. 
1990; Vermote and Kaufman 1995).
Typically, the AVHRR reflectance data are calibrated by observing a time 
series of AVHRR data over radiometrically stable targets of known reflectance. For 
7
example, many scientists have used observations over desert sites (Heidinger et al. 
2002; Rao and Chen 1995; Rao and Chen 1996). Kaufman and Holben (1993) utilized 
atmospheric scattering and sun-glint over the ocean in addition to desert reflectance 
measurements for calibration of AVHRR reflectance data. Loeb (1997) utilized 
observations over snow/ice covered Polar Regions. Vermote and Kaufman (1995) 
used ocean observations to track the degradation of channel 1 and observations of 
clouds to monitor the changes in the channel 1/channel 2 ratio. (Kaufman and Holben 1993; Loeb 1997; Vermote and Kaufman 1995)
However, the accuracy of these vicarious calibration techniques is limited by 
the  necessary  assumptions  employed,  which  include  assumptions  about  the 
atmospheric conditions, surface BRDF and natural seasonal variability of the so-
called stable target. The calibration accuracy achieved by these approaches is not 
better than 5% (Vermote and Saleous 2006a). 
A much improved calibration technique is presented by El Saleous, (2000). 
This method estimates the offsets for calibration equations of AVHRR reflectance 
channels by averaging the deep space measurements available in each scan line, while 
the gains of the calibration equations are estimated using the in-flight calibration 
method, presented in Vermote and Kaufman (1995). Additional, ancillary datasets 
including detailed atmospheric composition information, wind speed, and reliable 
cloud detection algorithms are used to improve the performance of in-flight 
calibration method. This method has a demonstrated calibration accuracy of 1% (El 
Saleous et al. 2000; Pedelty et al. 2007; Vermote and Saleous 2006a).
The presence of dual-gain reflectance channels on the current series of 
AVHRR sensors (AVHRR/3 instruments) complicates the calibration processes 
8
considerably. Heidinger et al., (2002) and Vermote et al., (2006) have presented 
calibration methods that use data from Moderate Resolution Imaging Spectrometer 
(MODIS) instruments for calibration of the AVHRR dual-gain reflectance channels 
(accuracy = 1%).
1.2.3.2. Orbital Drift
TIROS-N satellites which carry the AVHRR instruments are deployed on sun 
synchronous orbits. However, these sensor platforms suffer temporal recession of the 
equatorial  crossing-time  as  each  platform  ages  (Privette  et al.  1995). This 
phenomenon is commonly known as orbital drift. Solar zenith angle (SZA) 
differences due to orbital drift are greatest near lower latitudes and steadily decrease 
with increase in latitude. For example, by year 1999, SZA at equator for the NOAA-
14 AVHRR instrument deviated more than 35? from the angle at launch sun-sensor 
geometry. However, at 60? latitude, owing to increase in the number of satellite 
overpasses per day at high latitudes the change in SZA was only ~5? (Privette et al. 
1995).
Surface BRDF and variation of atmospheric path length due to change in SZA 
makes NDVI sensitive to the orbital drift (El Saleous et al. 2000; Los et al. 2005; 
Sellers et al. 1996; Tanre et al. 1992). At the top of the canopy (TOC), NDVI 
increases with increase in solar zenith angle (Deering et al. 1992). This is due to 
increase in the fraction of solar radiation intercepted by the vegetation canopy. Past 
studies have shown that sensitivity of TOC NDVI to changes in SZA decreases with 
increase in LAI (Kaufmann et al. 2000).
9
Conversely, when considering the top of the atmosphere (TOA), an increase in 
SZA leads to a decrease in NDVI. This is due to the increase in atmospheric path 
length, which leads to an increase in path radiance in red and a decrease in NIR due to 
atmospheric absorption (Holben 1986; Privette et al. 1995; Tanre et al. 1992). The 
sensitivity of TOA NDVI to orbital drift increases with increase in LAI. This is 
because increase in LAI leads to reduced red reflectance, while the atmospheric 
contribution to the at-sensor red channel measurement (path radiance) increases with 
increase of SZA. Because NDVI is highly sensitive to red reflectance, orbital drift 
alone can cause considerable decrease in NDVI estimates. This atmospheric 
component of the impact of orbital drift on AVHRR-NDVI time series data is 
commonly observed at dense tropical forest sites (Tucker et al. 2005). Because, the 
impact of orbital drift on NDVI is systematic, it limits the investigation of temporal 
and spatial trends in land surface NDVI.
A number of empirical and statistical techniques have been used to deal with 
the impact of orbital drift. For example, Sellers et al., (1996) uses a simple empirical 
procedure, where straightforward empirical relationships between AVHRR NDVI and 
SZA were developed for dense green vegetation and bare soil land cover. These two 
relationships were then used to normalize the data to at nadir illumination condition 
by assuming a linear relationship between SZA correction and NDVI magnitude 
(Sellers et al. 1996). Jiang et al., (2008) dealt with the impact of orbital drift in Global 
Vegetation Index (GVI-2) dataset by adjusting the NDVI values to same level as 
NDVI data from the year of satellite launch. Pinzon et al., (2005) used an adaptive 
empirical mode decomposition (EMD) method to statistically remove the impact of 
10
solar zenith angle drift in GIMMS NDVI dataset (Tucker et al. 2005). Several past 
studies have proposed orbital drift correction by systematic BRDF normalization 
using semi-empirical kernel models (Latifovic et al. 2003; Los et al. 2005; Privette et 
al. 1997; Roujean et al. 1992). Most of these studies have derived BRDF kernel 
parameters from AVHRR data themselves, even though most of the AVHRR datasets 
have incomplete atmospheric correction and inaccurate cloud masks. To overcome 
this shortcoming, Bacour et al., (2006) used POLDER data to parameterize BRDF 
kernels, but this method needed an a priori knowledge of existing land cover. (Bacour et al. 2006)
1.2.3.3. Atmospheric Correction
Atmospheric scattering and absorption substantially affect the AVHRR 
reflectance channels.  The red and NIR channels are spectrally broad and include 
ozone and water vapor absorption bands. The channel 1 (red) encompasses an ozone 
absorption band and a weak water vapor absorption band. The channel 2 (NIR) 
contains a strong water vapor absorption band extending from 0.930 to 1.000 ?m 
(Fig. 1-2) (Tanre et al. 1992; van Leeuwen et al. 1999). Because of this, the visible 
and near-infrared radiation reaching the satellite is differentially absorbed by 
atmospheric gases. Molecular backscatter by atmospheric aerosols introduces an 
additional directional reflectance (path radiance), which has high temporal as well as 
spatial variability (El Saleous et al. 2000; Tanre et al. 1992).
11
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0.4 0.6 0.8 1 1.2
Wave Length (?m)
Sp
ec
tra
l R
es
po
ns
e
Channel 1 (Red)
Channel 2 (NIR)
Fig. 1-2: AVHRR NOAA-14 spectral response function
Water vapor absorption can cause a 10-30% reduction of reflectance in NIR 
channel, while ozone absorption can reduce red channel reflectance by 5-15%, while 
atmospheric molecular scattering (Raleigh scattering) and aerosol effects can increase 
red channel reflectance as much as the reflectance of vegetated surfaces. During 
dense haze, path radiance can completely obscure surface properties. In terms of 
NDVI, even moderate haze can decrease the true NDVI of a densely vegetated 
surface by as much as ~ 0.1 (Tanre et al. 1992). 
Unlike MODIS, AVHRR instruments do not have additional spectral channels 
that allow derivation of information about atmospheric composition to be used for 
atmospheric correction. However, correction for Raleigh scattering, and ozone 
absorption can be performed easily and has been included in processing streams of 
12
many publicly available AVHRR NDVI datasets (James and Kalluri 1994; Pedelty et 
al. 2007). Correction for water vapor absorption, which is crucial for obtaining 
reliable  NDVI  measurements  over  sparsely  vegetated  areas,  has  also  been 
successfully implemented by El Saleous, (2000). The aerosol effect which is highly 
variable in both space and time domains is the most challenging factor to correct. 
Temporal compositing techniques, like maximum NDVI, are widely used to minimize 
the effect of atmospheric spectral attenuation (Cihlar et al. 1994; Holben 1986), but 
this technique is of limited use in daily or short compositing periods (El Saleous et al. 
2000).
1.2.4. Publicly available processed AVHRR NDVI datasets
Most vegetation studies do not use NDVI data calculated from at-sensor 
measures (TOA) from individual overpasses, but use processed and composited 
datasets. Some examples of GAC data based global datasets are LTDR (Long Term 
Data Record) (Pedelty et al. 2007), PAL (Pathfinder AVHRR Land) (James and 
Kalluri 1994), GIMMS (Global Inventory Modeling and Mapping studies) (Tucker et 
al. 2005), GVI (Global Vegetation Index) (Gutman et al. 1995; Kogan and Zhu 2001), 
GEOCOMP by Canadian Centre for Remote Sensing (CCRS) and a dataset produced 
at Joint Research Center (JRC) in Italy (Malingreau and Belward 1994).
There are also LAC datasets at 1km resolution. For example, the global 1km 
resolution NDVI produced by USGS (Eidenshink and Faundeen 1994; Teillet et al. 
2000), a regional 1km dataset for conterminous USA and Alaska (Eidenshink 2006; 
13
Eidenshink 1992), a similar dataset produced by CCRS for the Northern USA 
(Latifovic et al. 2005), and pan European 1km AVHRR NDVI data  produced by JRC 
in Italy (Swinnen et al. 2007). There are similar projects in many other regions such 
as South-Africa, UK, Sweden, and Australia..
Some of these datasets are partially corrected for atmospheric perturbation and 
may omit extreme view zenith angles. Most of these datasets also use some kind of 
spatiotemporal aggregation, such as temporal maximum value compositing, and 
spatiotemporal  smoothing.  These  corrections  and  aggregation  techniques  are 
generally acknowledged to decrease uncertainty in the satellite measurements of 
surface properties (Cihlar et al. 1998; Holben 1986). Table 1-1 presents a summary of 
dataset characteristics for some commonly used AVHRR NDVI datasets. Of these 
datasets, GIMMS dataset is currently the most popular while the relatively new LTDR 
dataset has the most comprehensive processing to date. These two datasets are 
discussed in more detail below.
14
Table 1-1. Summary of dataset characteristics for some AVHRR NDVI datasets
15
Name of dataset LTDR  V3 PAL GIMMS-G GVI USGS USGS global
Spatial Extent Global Global Global Global
Conterminous 
USA and 
Alaska
Global
Spatial Resolution 0.05 degree (5km) 8 km 8 km 16 km 1km 1, 2, 4, 8, 16 km
                      Temporal
                    Compositing period Daily
10 day & 
monthly bimonthly
weekly & 
monthly
weekly & 
biweekly 10day
Availability 1981- 2006 1981-2000 1981-2006 1985-2000 1989-present
06/1992 to 09/1993 
and 
02/1995 to 05/1996
Calibration
(Vermote and 
Kaufman 1995; 
Vermote and 
Saleous 2006b)
(Rao and 
Chen 1996)
(Los 1998; Rao 
and Chen 1996)
(Rao and Chen 
1996)
(Heidinger et al. 
2002; Teillet 
and Holben 
1994; Vermote 
and Kaufman 
1995)
(Teillet et al. 1990; 
Vermote and 
Kaufman 1995)
Cloud Screening
Red thresholds 
from MODIS 
global surface 
reflectance 
database
CLAVR
(Stowe et al. 
1995)
Temperature 
thrusholds for 
channel 5
No
CLAVR
(Stowe et al. 
1995) no
Raleigh correction Yes Yes No No Yes Yes
Ozone correction Yes Yes No No Yes Yes
Water Vapor 
correction Yes No No No Yes No
Atmospheric 
Aerosol correction
Expected in next 
version No
No
(corrected for 
volcanic aerosols)
(Vermote and 
Saleous 1997)
No No No
Data Input GAC (Global Area Coverage ) GAC GAC GAC
LAC (Local 
Area 
Coverage )
LAC
Orbital Drift 
Yes 
BRDF correction
(Vermote et. 2009)
No
Yes
Statistical time-
series adjustment
(Pinzon et al. 
2002)
No No No
Key Citation (Pedelty et al. 2007)
(James and 
Kalluri 
1994)
(Tucker et al. 
2005)
(Kidwell 1994, 
1997)
(Eidenshink 
2006; 
Eidenshink 
1992)
(Eidenshink and 
Faundeen 1994; 
Teillet et al. 2000)
 
1.2.4.1. Long-Term Data Record (LTDR)
LTDR is a consistent, long term dataset based on the AVHRR and MODIS 
data. LTDR is processed from AVHRR GAC data. The product includes daily NDVI 
as well as surface reflectance and radiance data at a spatial resolution of 0.05? (~5km 
at equator), which matches the Climate Modeling Grid (CMG).
The LTDR processing includes vicarious calibration of the visible and NIR 
channels using a cloud/ocean technique, which has been shown to have 1% error 
(Vermote and Kaufman 1995; Vermote and Saleous 2006b). The Earth location of 
each sensor measurement is mapped using inverse navigation technique (Rosborough 
et al. 1994). Unlike, forward navigation method, this technique does not lead to data 
gaps and geophysical bias (El Saleous et al. 2000). The atmospheric correction 
includes removal of the effects of Raleigh scattering, ozone, and water vapor. 
Ancillary data for water vapor and ozone correction were acquired from the NOAA 
Center for Environmental Prediction (NCEP) and Total Ozone Mapping Spectrometer 
(TOMS) respectively. An aerosol correction for this dataset is still being implemented 
(Pedelty et al. 2007) and not yet available in this version (LTDR version 3).
The LTDR version 3 data are normalized to a uniform sun-sensor geometry 
using the BRDF correction technique developed for MODIS data (Vermote et al. 
2009a). This BRDF correction technique will be discussed in detail in chapter 3 of 
this dissertation. This version of the dataset also uses an improved cloud masking 
technique based on monthly thresholds for BRDF corrected visible channel. The 
global maps of visible channel thresholds are calculated from monthly mean and 
standard deviation of BRDF corrected (Vermote et al. 2009a) MODIS Terra and Aqua 
16
data at CMG resolution of 0.05 degrees (Vermote et al. 2009b). The dataset is 
available in non-composited daily format with detailed ancillary data. This allows the 
data users to choose the appropriate compositing technique suitable to their 
application. The characteristics of some commonly used compositing methods are 
presented in chapter 4.
1.2.4.2. Global Inventory Modeling and Mapping studies (GIMMS)
GIMMS-g (Global Inventory Modeling and Mapping Studies) NDVI data are 
currently the most commonly used AVHRR NDVI data for time series analysis. Like 
LTDR, GIMMS data are also derived from GAC data. The GIMMS data processing 
includes calibration, corrections for orbital drift, corrections for volcanic aerosols, 
and spatiotemporal smoothing for removal of non-vegetation signal to produce 
bimonthly Maximum Value Composite (MVC) NDVI data. The data are currently 
(2010) available for years 1981 to 2006 (Tucker et al. 2005).
The GIMMS dataset also uses inverse navigation to map Earth location to 
each sensor measurement (El Saleous et al. 2000). Each composite image is further 
checked for navigation accuracy manually by comparing with a reference continental 
coastline map. Images with navigation errors greater than one pixel are investigated 
further, reprocessed and manually registered to the reference data.
The red and NIR channel data from NOAA-7 to 14 are calibrated using 
coefficients from Vermote and Kaufman (1995). The resulting NDVI fields are further 
adjusted using desert calibration technique from  Los (1998). The data are then 
corrected for the effects of solar zenith angle drift using a statistical method (EMD) 
17
described in Pinzon (2005). However, the GIMMS NDVI data derived from NOAA-
16 uses preflight coefficients for calibration of red and NIR channels to produce 
bimonthly NDVI composites. The EMD method is then used directly to ensure zero 
inter-annual slopes at desert sites and also to adjust for the impact of solar zenith 
angle drift. The NOAA-16 NDVI data are then inter-calibrated to NOAA-14 data 
using SPOT Vegetation NDVI time series data (Tucker et al. 2005). The dynamic 
range of the entire NDVI dataset is further adjusted using non-linear regression 
equations to  make it compatible with NDVI data from the MODIS and SPOT 
VEGETATION instruments (Tucker et al. 2005).  (Pinzon et al. 2005)
The GIMMS dataset does not employ any atmospheric correction  except 
during the El Chichon and Mt. Pinatubo volcanic stratospheric aerosol periods 
(Vermote and Saleous 1997) and relies solely on bimonthly maximum value 
compositing technique to reduce the impact of atmospheric variations. Cloud 
screening is performed by a channel 5 thermal mask of 0o C for all continents and a 
10o C mask for Africa. 
1.2.5. Uncertainty in Remotely Sensed Quantitative Data 
The outcome of any quantitative measurement depends not only on the item 
being measured, but also on the measuring system, the procedure, the operator, the 
environment and many other sources (Bell 1999). Thus, all measurements have 
associated uncertainties which reflect incompleteness in knowledge of the quantity 
being measured and have probabilistic basis. 
18
Uncertainty in remotely sensed data is assessed by the validation process. The 
accepted definition of validation in Earth observation context is: assessment of the 
quality of the data by independent means (Justice et al. 2000). The satellite-based 
remotely sensed data products are usually validated by comparison with similar 
products derived from independent sources. These independent products are derived 
from a combination of in situ data and imagery from airborne and satellite based 
sensors. 
For the validation process to be valid, the accuracy of the independent data 
product (reference data) has to be traceable to some known standard or, at minimum, 
it should be stable, consistent, and considerably more accurate than the signal that is 
being evaluated (Justice et al. 2000; Morisette et al. 2002). 
To facilitate this activity an extensive network of validation sites called Earth 
Observation System (EOS) Land Validation Core Sites has been established at both 
regional and global scale (Morisette et al. 2002). This network adds to existing field 
programs like the LTER (Long Term Ecological Research) network (Franklin et al. 
1990), science data networks like FLUXNET (Baldocchi et al. 2001), and AERONET 
(AErosol RObotic NETwork) (Holben et al. 2001) and international research efforts 
like the Southern Africa Fire and Atmosphere Research Initiative 2000 (SAFARI 
2000) (Swap et al. 2000), and the Committee on Earth Observing Satellites (CEOS) 
working group on Calibration and Validation (WGCV) (Justice et al. 2000).
The MODIS Land data products validation process makes extensive use of 
data from these networks to derive reference data. One of the primary difficulties in 
using in situ data for validation is the scaling (aggregation) of the point data from 
19
validation sites to the coarser spatial resolution of remotely sensed global or regional 
datasets (Morisette et al. 2002). This difference in scale is resolved by coupling field 
data with airborne data from sources like MODIS Airborne Simulator (MAS), the 
Airborne Visible Infrared Imaging Spectrometer (AVIRIS), and  MODIS Quick 
Airborne Looks (MQUALS) and higher resolution satellite imagery data from 
instruments like IKONOS, Advanced Spaceborne Thermal Emission and Reflection 
Radiometer (ASTER), and Landsat (Morisette et al. 2002).
1.2.6. Past Studies on uncertainty in AVHRR NDVI data
AVHRR, unlike MODIS has not been subjected to such a rigorous validation 
process. Past studies of uncertainty in AVHRR reflectance and NDVI datasets have 
used various methodologies such as comparison with improved data from the same 
(Goward et al. 1993) or other sensors, validation using in situ data (Fensholt and 
Sandholt 2005), analysis of variation of the measurements for assumed stable targets 
(Cihlar et al. 1998) and use of spatial statistical methods such as semivariograms 
(Eklundh 1995). The AVHRR data have often been evaluated by comparison with 
better processed data from other sensors, for example Fensholt et al. (2006) who 
evaluated GIMMS and PAL AVHRR NDVI datasets by comparing them to data from 
SPOT, and Gallo et al. (2005) who compared AVHRR NDVI data with MODIS and 
investigated the feasibility of attaining continuity of NDVI products through future 
sensors systems. 
Most of the past investigations of uncertainty in AVHRR reflectance and 
NDVI data have not been comprehensive either in scope or scale. They have mostly 
20
focused on specific aspects of the uncertainty and their results are relevant to specific 
sites or regions only. For example, the study presented by Stellmes et al. (2010) 
evaluated NDVI trends in AVHRR data by comparing them to trends in Landsat data. 
The relevance of results from this study is limited to parts of Mediterranean. Fensholt 
et al., (2005) evaluated vegetation indices from AVHRR and MODIS instruments 
using 2 years of in situ data in Senegal. Fensholt et al., (2006) investigated how well 
AVHRR NDVI data from GIMMS and PAL datasets correlate with SPOT data in 
African continent. Gallo et al., (2005) similarly investigated the strength of 
correlation between AVHRR and MODIS data in the conterminous United States. 
Numerous other examples also exist in recent literature (Brown et al. 2006; Cihlar et 
al. 1998; Eklundh 1995; Fensholt et al. 2009; Goward et al. 1993; Ji et al. 2008; 
Kangas et al. 2001; Ouaidrari et al. 2003; Pouliot et al. 2010).  Because  many 
influential studies in the field of Earth system science have been based on AVHRR 
NDVI, a thorough understanding of uncertainty is long overdue. (Fensholt et al. 2006a; Fensholt and Sandholt 2005; Gallo et al. 2005; Stellmes et al. 2010)
1.2.7. Precision Assumed for AVHRR NDVI data
The approximate precision of NDVI of terrestrial vegetation is often, 
implicitly, assumed to be between ~0.03 and ~0.08 NDVI units. For example Jia et 
al., (2003) reported changes in the vegetation of arctic Alaska over 21 years using 
AVHRR data of 0.056?0.0032 to 0.082?0.028 NDVI units. Slayback et al., (2003) 
reported an increase of 0.027 to 0.081 NDVI units for a period of 18 years (1982-99) 
for global latitude bands from 35? to 75? N. Piao et al., (2003) reported an increase of 
similar magnitude (0.0325 NDVI units) in 18 years in China. There are many other 
studies using AVHRR NDVI (Piao et al. 2006; Young and Harris 2005), where 
21
changes much smaller than 0.03 units are treated as significant. At this time, the 
accuracy of the AVHRR data implied in these studies has not been validated and the 
significance of the differences is not stated explicitly. Thus, it is possible that some 
studies have assumed greater accuracy than is justified.
1.2.8.  Statistical Framework for Estimation of  Uncertainty in 
Remotely Sensed Data
The National Polar-orbiting Operational Environmental Satellite System 
(NPOESS) project has defined a statistical framework which represents uncertainty in 
two components: (1) systematic bias; and (2) random-error. Of these two, the random-
error component represents relative error which is pertinent for inter-annual studies, 
while the information about systematic bias becomes relevant only when the data are 
being used in conjunction with other datasets. This study will primarily focus on 
random-error component of uncertainty (Fig. 1-3). To facilitate discussion the terms 
?random-error? and ?uncertainty? will be used interchangeably in rest of this 
dissertation.
22
Fig.  1-3: Statistical framework used for calculation of random-error.. This analysis 
used non-composited daily NDVI data from 2003 to 2006.
23
Analysis presented in Fig. 1-3 illustrates a statistical framework for estimation 
of the uncertainty (random-error) using growing season NDVI data from an 
agricultural site in southwestern Australia. Uncertainty here is calculated as the 
standard deviation of the data relative to the systematic-bias line. The systematic-bias 
line can be estimated using two methods: (1) by fitting a first order polynomial to the 
two dimensional scatter plot comparing AVHRR NDVI data to the reference data; (2) 
when the slope of the polynomial is one or the R-squared of the polynomial fit is too 
low the bias can be directly estimated as the mean of deviations of the AVHRR NDVI 
data from the reference data. The uncertainty of 0.04 estimated for LTDR AVHRR 
NDVI data at the example site in Fig. 1-3, implies that, at this site an NDVI value of 
0.6 will actually have 68% probability of ranging between 0.56 and 0.64 (i.e. 
0.6?0.04) and 95% probability of ranging from 0.52 to 0.68 (i.e. 0.06?0.08).
When evaluating satellite data using in situ measurements the presence of 
systematic-bias is not necessarily an evidence of error. Rather it could be a result of 
fundamental differences in sensor design (e.g., spectral response function, radiometric 
resolution, sensor calibration, etc.), atmospheric interference in satellite data, and 
shortcomings in the process of scaling up the point data to the coarser satellite 
resolution (Cohen and Justice 1999; Morisette et al. 2002). Random-error on the other 
hand can be attributed to measurement error, variations in atmospheric composition, 
residual cloud contamination, geolocation error, and surface and atmospheric BRDF 
effects in satellite data.
When evaluating a satellite based remotely sensed dataset using a higher quality 
satellite based dataset, the systematic-bias is a result of differences in sensor design 
24
and deployment, such as spatial resolution, spectral response function, sensor 
calibration, radiometric resolution, sun-sensor geometry, and differences in post-
processing algorithms (atmospheric correction, BRDF correction, and vicarious 
calibration). The random-error can be attributed to the geolocation error, and variation 
of pixel size in relation to viewing geometry, and the complex interactions of the 
differences between the sensor systems. Clearly, for this method of evaluation to be 
valid, the reference dataset has to have significantly lower error.
1.3. Research Goals and Dissertation Structure
Steps Research Chapters
Part 1  Investigation of errors due to atmospheric variation Chapter 2
Part 2  Investigation of impact of orbital drift Chapter 3
Part 3
 Quantification of overall uncertainty and impact of spatial 
and temporal aggregation Chapter 4
Table 1-2: Flow of the research activities within the dissertation structure
The overall goal of this research was to estimate uncertainty of AVHRR NDVI 
data and the impact of various degrees of spatial and temporal aggregation. This was 
achieved in 3 steps (Table 1-2). The first part investigated the extent of uncertainty 
due to inadequate atmospheric correction in the three widely-used, publicly-available, 
AVHRR NDVI datasets. The second part investigated the interaction between surface 
and atmospheric BRDF and its contribution to the sensitivity of AVHRR-NDVI time 
series data to orbital drift. A new approach for surface BRDF correction was 
proposed. The third part focused on estimation of spatially-explicit, comprehensive 
uncertainty metrics for AVHRR NDVI data and the impact of various degrees and 
types of temporal and spatial aggregation. This part primarily focused on the LTDR 
25
dataset. The overall guiding principle of the investigation was to provide detailed 
information about the precision of AVHRR NDVI data and the corrections that can 
increase the value of the data for Earth science research. 
 This dissertation consists of 5 chapters. The Chapters 2, 3, and 4 are 
presented in the self-contained format of journal articles (one of which is published 
and one in review) and present results from parts 1, 2, and 3 of this dissertation 
respectively (Table 1-2). Chapter 5 summaries the results and discusses their 
implications on Earth system science studies.
26
Chapter 2: Investigation of Errors Due to Atmospheric 
Composition
2.1. Introduction
Most of the available AVHRR NDVI datasets have incomplete atmospheric 
correction, resulting in considerable, but largely unknown, uncertainty in the 
significance of differences in NDVI and other short wave observations from AVHRR 
instruments. The purpose of this study was to gain a better understanding of the 
impact of this incomplete or absent atmospheric correction in widely-used, publicly 
available processed AVHRR-NDVI long-term datasets. 
Three datasets were evaluated: TOA; PAL; and an early version of the new 
Long Term Data Record (LTDR) (http://ltdr.nascom.nasa.gov/ltdr/ltdr.html) (Pedelty 
et al. 2007). The other publicly available datasets like GIMMS and GVI have no 
atmospheric correction and rely solely on temporal compositing to reduce the effects 
of changes in atmospheric properties. Uncertainty was assessed by comparing the 
processed AVHRR datasets to atmospherically corrected AVHRR data at AERONET 
sites and analyzing simulated AVHRR NDVI data. 
2.2. Data and Methods
2.2.1. Approach
The extent of uncertainty due to inadequate atmospheric correction in the 
widely used AVHRR NDVI datasets was investigated by comparison with data to 
27
which  comprehensive  atmospheric  correction  has  been  applied.  Atmospheric 
compositions derived from AERONET sunphotometer data were used for space and 
time-specific atmospheric correction. These corrected data are referred to as reference 
data. The statistical framework presented in chapter 1 (section 1.2.8) was used to 
estimate uncertainty metrics.
AERONET is an aerosol monitoring system and part of the extensive remote 
sensing validation networks established in recent years over a wide variety of land 
cover types and latitudinal zones. AERONET is a world-wide network of automatic 
sun and sky scanning spectral radiometers which measure every 15 min. The data 
from these instruments are processed to derive atmospheric aerosol optical properties, 
precipitable water, and aerosol size distribution (Holben et al. 2001). In 1999, there 
were about 100 operational sites and by 2006, the network had grown to more than 
300. 
This study evaluated TOA (no atmospheric correction) (Pedelty et al. 2007), 
PAL, and LTDR datasets. The PAL processing stream includes correction for Raleigh 
scattering and ozone, while the early version of LTDR that was tested here is 
corrected for Rayleigh scattering, ozone, and atmospheric water vapor (Pedelty et al. 
2007). PAL data were resampled from 8km to 5km resolution using nearest neighbor 
method to make them comparable to LTDR and TOA data with 5km (0.05? ) 
resolution. The year 1999 was used for this investigation because non-composited, 
individual, satellite overpass data were available for all three datasets used (TOA, 
PAL, and LTDR) and there were an adequate number of operational AERONET sites.
28
Although the data from year 1999 provided a useful overview of the error 
patterns in AVHRR NDVI data record, it represented only one year and most of the 
data were in North America and in semi-arid and desert areas elsewhere, owing to the 
smaller network of AERONET stations at that time. Moreover, aggressive cloud 
masking derived from AERONET led to very few data being available to investigate 
the impacts of temporal compositing. To overcome these shortcomings, further 
analysis was performed on simulated AVHRR data for representative sites in tropical 
forest (Belterra, Brazil), savanna (Skukuza, South Africa), and semi-arid (Sevilleta, 
Arizona, USA).. This was done by using atmospheric conditions and cloud cover 
frequency from AERONET, NDVI phenology derived from MODIS, and sun-target-
sensor geometry from the full length (1981-2000) of the AVHRR data record.
2.2.2. Reference Data
Comprehensive atmospheric correction was carried out on TOA AVHRR data 
for 9 pixel matrices centered on each available AERONET site (Fig. 2-1) using the 6S 
(Vermote et al. 1997) atmospheric radiative transfer program. Aerosol optical 
thickness, aerosol size distribution, and precipitable water data from AERONET as 
well as ozone concentration data from TOMS (Total Ozone Mapping Spectrometer) 
(McPeters and Labow 1996) were utilized in this atmospheric correction process. The 
TOMS instruments measure ultraviolet (UV) radiation backscattered by Earth?s 
atmosphere and surface at wavelengths between 310 - 380 nm. UV wavelengths 
shorter than 340 nm are used to estimate ozone concentration in the atmospheric 
column.
29
 
Fig. 2-1: Schematic view of the data processing for reference data
Only AVHRR pixels with view zenith angle less than 42? were used. The data 
were masked for cloud and cloud shadow using quality flags from LTDR (El Saleous 
et al. 2000; Stowe et al. 1995). Further cloud masking was performed using an 
aggressive AERONET derived cloud mask. After application of 6S to remove 
atmospheric effects from the reference data, the NDVI and surface reflectance were 
used to estimate the magnitude of the errors of the publicly-available, processed, 
AVHRR NDVI datasets. After masking for cloud and water, approximately 580 data 
points (sites x days) remained from 48 AERONET sites (Fig. 2-2). Of these data 
points 29% were clear (AOT < 0.05), 62% were average (AOT = 0.05 to 0.25), and 
9% were hazy (AOT > 0.25). Most of these data points were in North America (60%) 
or in semi-arid (30%) and desert (17%) areas elsewhere (Fig. 2-2).
30
TOA calibrated 
AVHRR Ch 1 
and 2 data
Atmospheric 
data from
AERONET and 
TOMS
Atmospheric 
Correction
(6s)
Reference data
(At Surface NDVI)
Comparison with 
three AVHRR 
NDVI data sets 
Fig. 2-2: Spatial distribution of AERONET sites used to calculate reference data. After masking for cloud, cloud shadow, 
and view zenith angle greater than 42?, only 580 data points from 48 AERONET sites remained for use in calculation of 
the reference data. The number of data points is indicated by the diameter of the circle (see inset).
31
2.2.3 Simulated AVHRR time series
6S radiative transfer program was used to create a 1981 to 2000 simulated 
time series for AERONET sites: Belterra (54.952? W, 2.649? S), Skukuza (31.588? E, 
24.992? S), and Sevilleta (106.885? W, 34.355? N) to represent forest, savanna, and 
semi-arid land cover types respectively. Atmospheric composition and a cloud mask 
were derived from one representative year of AERONET measurements and TOMS. 
Representative phenology was derived from MODIS data. The sun-sensor-target 
geometry was extracted from LTDR data. Solar zenith angle and view zenith angles 
were constrained to less than 75? and 50? respectively.
At-surface reflectance measures derived from MODIS was first converted to 
TOA at-sensor measure using 6S. The calculated TOA reflectance data were then 
corrected for Raleigh scattering and atmospheric absorption using 6S to produce PAL 
type data.  Further introduction of correction for water vapor simulated LTDR data. 
2.3. Results and Discussion 
Comparison of data from red and near infra-red channels of TOA, PAL and 
LTDR datasets with reference data at the AERONET stations (Table 2-1) showed that 
in both channel 1 and 2 LTDR has greatly improved systematic-bias and uncertainty 
when compared to PAL and TOA data. PAL consistently underestimated reflectance 
in both channels, which can be attributed to use of obsolete calibration and lack of 
water vapor correction. The comparison of the NDVI from these datasets with 
reference data shows that LTDR is closest (systematic-bias = -0.024, uncertainty = 
32
0.037 for data points with average atmospheric conditions) followed by PAL 
(systematic-bias = -0.035, uncertainty = 0.068), and TOA (systematic-bias = -0.112, 
uncertainty = 0.0684) (Table 2-1). 
33
Table 2-1: Systematic-bias and uncertainty for red channel (CH1), near infrared channel (CH2), and NDVI for TOA, PAL, 
and LTDR data. Of the 580 data points 168 (29%) were clear, 360 (62%) were average, and 52 (9%) were hazy.
Band DATA Clear (AOT <  0.05) Average (AOT = 0.05 to 0.25) Hazy (AOT > 0.25)
Systematic-bias Uncertainty Systematic-bias Uncertainty Systematic-bias Uncertainty
CH1 ??? -0.0020 0.0219 -0.0060 0.0265 -0.0120 0.0352
CH1 PAL -0.0225 0.0228 -0.0189 0.0214 -0.0121 0.0409
CH1 LTDR 0.0005 0.0079 0.0032 0.0174 0.0152 0.0267
CH2 ??? -0.0358 0.0261 -0.0425 0.0303 -0.0561 0.0286
CH2 PAL -0.0508 0.0386 -0.0577 0.0366 -0.0786 0.0413
CH2 LTDR -0.0002 0.0102 -0.0023 0.0182 0.0004 0.0212
NDVI ??? -0.0791 0.0613 -0.1120 0.0684 -0.1521 0.1059
NDVI PAL -0.0145 0.0606 -0.0352 0.0418 -0.1137 0.0877
NDVI LTDR -0.0064 0.0206 -0.0243 0.0373 -0.0612 0.0780
34
Simulated Land
Clear
 (AOT <  0.05)
Average
 (AOT = 0.05 to 0.25)
Hazy 
(AOT > 0.25)
Data cover Systematic-bias Uncertainty Systematic-bias Uncertainty Systematic-bias Uncertainty
??? semi-arid -0.0616 0.0226 -0.0667 0.0234 -0.1400 0.0629
??? savanna -0.1701 0.0436 -0.1770 0.0510 -0.3047 0.1816
??? forest -0.2469 0.0821 -0.2804 0.0849 -0.5861 0.1880
PAL semi-arid -0.0393 0.0114 -0.0514 0.0155 -0.1323 0.0781
PAL savanna -0.0792 0.0154 -0.1072 0.0426 -0.2523 0.1711
PAL forest -0.0555 0.0263 -0.1242 0.0719 -0.5077 0.2185
LTDR semi-arid -0.0071 0.0610 -0.0109 0.0109 -0.0846 0.0850
LTDR savanna -0.0137 0.0105 -0.0379 0.0356 -0.1661 0.1641
LTDR forest -0.0252 0.0206 -0.0845 0.0611 -0.4291 0.1988
Table 2-2: Systematic-bias, and Uncertainty for NDVI from simulated TOA, PAL, and LTDR datasets. For the forest site; 
20 (4%), 376 (75%), and 105 (21%) of 501 data points had clear, average, and hazy atmospheres respectively. For the 
savanna site; 40 (9%), 317 (71%), and 89 (20%) of 446 data points had clear, average, and hazy atmosphere respectively. 
For the semi-arid site; 384 (52%), 348 (47%), and 8 (1%) of 740 data points had clear, average, and hazy atmosphere 
respectively.
35
Land
cover
Comp
Period
Clear (AOT <  0.05) Average (AOT = 0.05 to 0.25) Hazy (AOT < 0.25)
Systematic
-bias Uncertainty % data 
Systematic
-bias Uncertainty % data 
Systematic
-bias Uncertainty % data 
Semi-arid daily -0.0611 0.0217 52 -0.0643 0.0214 47 -0.1400 0.0629 1
Semi-arid 10day -0.0514 0.0191 55 -0.0541 0.0186 45 - - 0
Semi-arid 15day -0.0485 0.0174 58 -0.0514 0.0178 42 - - 0
Semi-arid monthly -0.0447 0.0157 69 -0.0474 0.0197 31 - - 0
Savanna daily -0.1614 0.0360 9 -0.1727 0.0470 71 -0.3067 0.2130 20
Savanna 10day -0.1516 0.0316 11 -0.1594 0.0420 71 -0.3072 0.2158 18
Savanna 15day -0.1499 0.0328 11 -0.1573 0.0426 72 -0.3026 0.2059 17
Savanna monthly -0.1487 0.0328 16 -0.1528 0.0416 72 -0.2552 0.1716 12
Forest daily -0.2347 0.0702 3 -0.2830 0.0774 72 -0.5241 0.1585 22
Forest 10day -0.2179 0.0663 3 -0.2610 0.0754 75 -0.5223 0.1694 22
Forest 15day -0.2075 0.0415 3 -0.2554 0.0722 77 -0.5428 0.1663 20
Forest monthly -0.1967 0.0302 4 -0.2434 0.0721 74 -0.5036 0.1401 22
Table 2-3: Impact of temporal maximum value compositing (MVC) on Systematic-bias, and Uncertainty for simulated TOA 
NDVI time series data
36
The error budget of simulated time series (Table 2-2) compares well with 
empirically derived error budget in Table 2-1. For the forest and savanna sites most of 
the data points (> 70%) had average atmospheric conditions (AOT = 0.05 to 0.25) and 
about 20% were hazy (AOT>0.25).  However, for semi-arid site more than half data 
points had clear atmospheric conditions (AOT < 0.05) and only 0.5% points were 
hazy. Both the direct validation (Table 2-1) and site specific simulated error budget 
(Table 2-2) show that in hazy conditions all three datasets become very unreliable 
with uncertainty as high as 0.22 in the forest site. These results show uncertainty for 
TOA NDVI data ranged from 0.023 for clear atmospheric conditions at the semi-arid 
site to 0.085 for average atmospheric conditions at the forest site (not considering 
data points with hazy atmospheric conditions). For LTDR NDVI data uncertainty 
varied between 0.0061 for clear atmospheric conditions at semi-arid site to 0.0611 for 
average atmospheric conditions at forest site. A similar decrease in uncertainty was 
observed at the savanna site.
Usually AVHRR time series data are used in temporally composited format. 
Many methods are used for compositing, of which MVC (Maximum Value 
Composite) (Holben 1986) is most widely used. Various lengths of compositing 
periods are used. Table 2-3 shows the impact of simple temporal MVC on bias and 
uncertainty of TOA NDVI data at forest, savanna and semi-arid sites. From Table 2-3 
it can be observed that there is no significant impact of temporal compositing on 
Uncertainty in all three sites in average as well as hazy atmospheric conditions. Most 
of the data points selected for savanna and forest sites with MVC criteria had average 
and hazy atmospheric conditions. It was also observed that various degrees of 
37
compositing did not significantly change the proportion of data points in clear, 
average, and hazy atmosphere categories. This was due to persistent high aerosol 
content in atmosphere which limits the capacity of MVC method to find a clear day. 
This persistence of high aerosols is a common feature of these and similar sites. MVC 
was useful only in sites where high aerosol content persisted for shorter periods than 
the length of the composing period. Cloud frequency and high values of AOT play an 
important role in the effectiveness of MVC.
2.4. Conclusions 
The results presented in the Tables 2-1 and 2-2 show that LTDR has the least 
errors followed by PAL, and TOA. Of the two metrics (systematic-bias, and 
uncertainty (random-error)) used to describe the errors; uncertainty, which represents 
standard deviation of the estimates adjusted for the mean bias is most pertinent for 
inter-annual studies as it represents random component of the error.
Although both PAL and LTDR data have lower error budget in both clear and 
average atmospheric conditions, it was observed that magnitude of bias and 
uncertainty is driven primarily by aerosol content in the atmosphere and its 
interaction with solar illumination angle and cloud frequency. It was also seen that 
temporal  maximum  value  compositing  technique  does  not  cause  significant 
improvement of the error budget in regions experiencing persistently high AOT due 
to, for example, biomass burning. Thus implementation of aerosol correction or at 
least some technique to identify and reject even MVCs that still have high aerosol 
contamination would greatly improve the error budget of the data records.
38
Spatiotemporal aggregation techniques, which are used extensively by NDVI 
data users, can decrease uncertainty but one has to be careful to account for the 
spatiotemporal autocorrelation. For example if a region is experiencing fairly uniform 
atmospheric conditions for extended periods of time, spatiotemporal aggregation will 
not decrease the impact of atmosphere on the data. Non-stationarity introduced in the 
data due to systematic intra-annual and inter-annual changes in sun-sensor-target 
geometry (orbital drift) also has to be considered. The next chapter will focus on the 
impact of orbital drift on long-term AVHRR NDVI time series data.
39
Chapter 3: Investigation of Impact of Orbital Drift on 
Long-Term AVHRR NDVI Time Series Data
3.1. Introduction
The NDVI time series data derived from AVHRR series of instruments suffer 
from significant variations as a result of the orbital drift of the satellites through their 
active lives. Orbital drift causes a systematic change in SZA. The sensitivity of 
AVHRR NDVI data to the SZA change depends on surface BRDF, atmospheric 
composition and the type of atmospheric correction used. The systematic error caused 
by orbital drift hinders the detection and interpretation of temporal Earth-surface 
vegetation dynamics.
The research presented in this chapter had two objectives. The first objective 
was to investigate the contributions of surface and atmospheric bidirectional effect to 
the impact of orbital drift on AVHRR-NDVI time series data. The second was to 
examine the effectiveness of MODIS based BRDF parameters (Vermote et al. 2009a) 
for remediation of the impact of orbital drift on AVHRR NDVI data.
3.2. Data and methods
3.2.1. Impact of Orbital Drift on AVHRR NDVI Time Series
The contribution of surface BRDF and atmosphere to the impact of orbital 
drift on AVHRR-NDVI time series data was investigated using simulated NDVI time 
40
series data with varying degrees of vegetation cover (10%, 50%, and 95%); 
atmospheric composition (TOC; and TOA); and at different latitudes (0?, 30?, and 
60?). These simulated time series data were created using actual sun sensor geometry 
for specific latitudes and 0? longitude. The temporal extent of the simulated data was 
restricted to the first 5 years (1995 to 1999) of NOAA-14 satellite?s operational life. A 
constant AOT (Aerosol Optical Thickness) of 0.15 was assumed and view zenith 
angle (VZA), and SZA were restricted to values below 45?, and 75? respectively. 
Inter-annual and seasonal variations in vegetation and atmospheric conditions in the 
modeled time series were kept constant to facilitate examination of trends introduced 
by surface and atmospheric BRDF. Contributions of atmosphere and surface BRDF to 
the impact of orbital drift on AVHRR NDVI time series data were estimated using the 
6S atmospheric radiative transfer program (Vermote et al. 1997) and the FLIGHT 
(Forest LIGHT) model (North 1996).
FLIGHT is a 3D canopy radiative transfer model which uses Monte Carlo 
simulation of photon transport to simulate light interaction in the forest canopy. The 
spatial discontinuity of the forest canopy is approximated using a hybrid approach. 
Geometric primitives (circle, ellipse, cylinder and cone) are used to define the 
structural elements (crowns and trunks) of the forest. Foliage is approximated within 
the crowns by foliage structural parameters (area density, angular distribution, and 
size) and  optical properties of leaf and branch material. BRDF is estimated by 
simulation of the photon path and the chain of scattering events incurred by a photon 
during its journey from the source to the receiver or to its absorption within a forest 
representation. The photon path simulation is modified to reflect finite sized scatters 
41
(Jupp and Strahler 1991). The model has been validated against field measurements 
and by inter-comparison with other three dimensional radiative transfer models (Pinty 
et al. 2004). Comparisons with ground (PARABOLA) and airborne (ASAS) data 
confirmed the ability of FLIGHT model to estimate BRDF and albedo for even the 
most complex case of open coniferous canopies (North et al. 1996; North 1996), with 
an error of 2% for bidirectional reflectance factors (BRF) and 0.5% for albedo.
 3.2.2. Effectiveness of MODIS BRDF parameters for BRDF 
Correction of AVHRR NDVI Time Series Data
 The effectiveness of the BRDF correction technique developed for MODIS 
data (Vermote et al. 2009a) in removing the impact of surface BRDF on the effects of 
orbital drift on AVHRR NDVI data was examined by applying this technique directly 
to modeled TOC NDVI time series data as well as actual AVHRR NDVI data from 
LTDR dataset (http://ltdr.nascom.nasa.gov/). Two sites, with different degrees of 
vegetation cover (50% and 95%), were used for this purpose (Fig. 3-1). The analysis 
was restricted to the data from first 5 years (1995 to 1999) of NOAA-14 satellite?s 
operational 1 life. The VZA was restricted to less than 45? and solar zenith angle was 
restricted to less than 75?.
42
Fig. 3-1: Aerial views of sites used for BRDF correction. Site 1 has 50% vegetation cover and is located in south-western 
Australia (116.167?E, 32.1944?S). Site 2, with 95% vegetation cover, is located in north-eastern Australia (145.8?E, 
17.4?S).
43
3.2.2.1. Study Sites
Of the two sites used in this study (Fig. 3-1), the first site (Site-1) is located in 
south-western Australia (116.167?E, 32.1944?S) and has 50% vegetation. Data from 
this site were used to illustrate the effectiveness of BRDF correction technique on 
sites expected to experience high impact of orbital drift on TOC NDVI. Site-2 has 
95% vegetation cover, and is located in north-eastern Australia (145.8?E, 17.4?S). 
Data from this site were used to demonstrate the impact of BRDF correction on sites 
with minimal impact of orbital drift on TOC NDVI but prominent contribution of 
atmospheric BRDF. Both of these sites have evergreen vegetation and experience 
fairly low aerosol contamination. The rationale behind using only evergreen sites was 
to facilitate investigation of the impact of orbital drift alone on the inter-annual NDVI 
trends.
3.2.2.2. The MODIS BRDF Correction Technique
The MODIS BRDF correction technique presented by Vermote et al., (2009) 
assumes that the shape of the BRDF varies more slowly than the reflectance and this 
change is linearly related to NDVI. This relationship with NDVI is then used to 
account for seasonal variations in BRDF shape. A relationship between BRDF 
parameters and NDVI for each pixel is then derived from the full MODIS time series 
record. Because this method uses MODIS data for definition of BRDF shape, it is not 
limited by incomplete atmospheric correction and inaccurate cloud masks of AVHRR 
NDVI datasets. 
44
The BRDF shape can be described by two parameters, R (surface scattering) 
and V (volume/canopy scattering). Assumption of a constant BRDF shape for the 
entire year allows an easy inversion of parameters R and V from MODIS data. These 
parameters can then be used for BRDF correction of reflectance the AVHRR time 
series data. MODIS directional reflectance time series data, when corrected using this 
method, showed considerable decrease in high frequency variability. However, 
vegetation structure generally has an annual phenological cycle of greening, growth, 
and senescence, which has a direct effect on the BRDF. Therefore, parameters V and 
R also vary with seasonal change in vegetation greenness (NDVI). Inclusion of this 
variation of V and R parameters in relation to NDVI in the BRDF correction process 
improved the correction of the MODIS time series data. A full description and 
discussion of this method can be found in Vermote et al., (2009). 
3.2.2.3. Modeled AVHRR TOC reflectance data
Modeled AVHRR TOC reflectance data were calculated for these two sites 
using FLIGHT, which is capable of modeling the impact of extreme solar zenith 
angles on directional reflectance of a vegetated surface. For each site, the FLIGHT 
model was first parameterized using approximate forest structural values (percent 
forest cover, mean crown diameter, and mean canopy height) inferred from high 
resolution satellite imagery (Fig. 3-1). The foliage parameters (leaf area density, leaf 
angular distribution, and leaf size, and optical properties of leaf and branch, etc) were 
populated using default values provided by the FLIGHT model (Table 3-1). These 
parameters were then iteratively varied within a reasonable range to arrive at the red 
45
and NIR BRDF shapes, representing the variation of reflectance in relation to VZA, 
similar to the ones produced by BRDF parameters from MODIS data (Vermote et al. 
2009a) at a constant SZA of 45?. The goal was to create a realistic definition of forest 
vegetation structure in FLIGHT model to facilitate the investigation of impact of 
extreme SZA variations present in AVHRR datasets.
FLIGHT Parameters Site-1 Site-2
% Vegetation cover 50 95
Crown diameter/height 1 1
Total canopy height 10 m 55 m
Scene LAI 2 6
Leaf red reflectance 0.07 0.045
Leaf NIR reflectance 0.45 0.5
Leaf red transmittance 0.015 0.02
Leaf NIR transmittance 0.39 0.4
Bark red reflectance 0.23 0.1
Background NIR reflectance 0.21 0.5
Table 3-1: Vegetation structural and optical information used to parameterize the 
flight model for forests at two sites depicted in Fig. 3-1.
This study also utilized AVHRR data from the LTDR version-2 dataset. The 
product includes daily surface reflectance and NDVI products at a resolution of 0.05?. 
This version includes atmospheric corrections for Rayleigh scattering, ozone, and 
water vapor. Aerosol correction is still being implemented (Pedelty et al. 2007).
46
3.3. Results and Discussion
3.3.1. Impact of Orbital Drift on AVHRR NDVI Time Series
The impact of vegetation cover on the relationship between TOC-NDVI and 
SZA presented in Fig. 3-2 indicated that the magnitude of TOC NDVI increased with 
an increase in SZA. It was also observed that the NDVI times-series data at sites with 
95% and 10% vegetation cover showed a lesser impact of SZA change when 
compared to the site with 50% vegetation cover. The rate of change of NDVI per 10? 
SZA was 0.070 for the site with 50% vegetation cover, while at sites with 10%, and 
95% vegetation cover the rate of change of NDVI was only 0.013, and 0.008 NDVI 
units per 10? SZA.
Fig. 3-2: The effects of forest vegetation cover on the relationship between solar 
zenith angle (SZA) and top of the canopy (TOC) NDVI
47
The combined impact of surface and atmospheric BRDF on AVHRR-NDVI 
time series is presented in Fig. 3-3 and Table 3-2. At the equator, the modeled TOC 
NDVI data for the site  with 50% vegetation cover (Site-1) showed a positive trend of 
0.0218 NDVI per year (Fig. 3-3 and  Table 3-2), a trend that was greatly weakened at 
TOA to a mere 0.0061 NDVI per year. This was because, unlike surface BRDF, 
atmospheric components introduce a negative trend in NDVI with increase in SZA. 
The same phenomenon was also observed in NDVI time series data at the sites with 
10%, and 95% vegetation cover, where a weak positive trend was removed. (Fig. 3-3; 
Table 3-2).
Latitude
Percent 
Tree Cover
Yearly Rate of 
change in NDVI
TOC TOA
0 95 0.0033 -0.0053
0 50 0.0218 0.0061
0 10 0.0053 0.0000
30 95 0.0014 -0.0031
30 50 0.0148 0.0005
30 10 0.0049 -0.0004
60 95 0.0022 -0.0022
60 50 0.005 -0.0043
60 10 0.0038 -0.0017
Table 3-2: Yearly rate of change in modeled AVHRR NDVI data for 5 years of 
NOAA-14 (Fig. 3-3). These trends were calculated by fitting linear regression lines to 
yearly average NDVI.
48
Fig. 3-3: Impact of orbital drift on AVHRR data modeled for TOA and TOC NDVI 
data for first 5 years of NOAA-14 satellite?s operational life (1995 to 1999) with 
various degrees of vegetation cover at 0?, 30?, and 60? latitudes. These time series 
were produced using FLIGHT canopy radiative transfer model and 6S atmospheric 
radiative transfer program. The simulation assumes evergreen vegetation cover and 
AOT = 0.15.  Note: The high frequency variation observed in SZA, which is due to 
inclusion of data from multiple orbits, causes a similar high frequency variation in 
NDVI time series. The thick line represents a 10 day mean NDVI and the thinner lines 
(not always separable from the mean) represent the maximum and minimum value of 
NDVI for the same 10 day period.
49
At 30? latitude the SZA had a weaker inter-annual trend leading to similarly 
weak inter annual trends in TOC NDVI. For example, at the site with 50% vegetation 
cover, the trend was only 0.0148 NDVI per year when compared to 0.0218 NDVI per 
year at equator. However there was a seasonal component to SZA which created intra-
annual patterns similar to seasonal vegetation dynamics. At 50% vegetation cover, 
this seasonality was fairly prominent even in TOA NDVI data (Fig. 3-3; Table 3-2).
At 60? latitude there were practically no inter-annual trends in SZA, hence 
there were no inter-annual trends in either TOC or TOA NDVI. However there was a 
strong seasonal component to the SZA trend, which led to a similarly prominent, but 
false impression of seasonal vegetation dynamics in TOC NDVI data. Especially at 
beginning and end of the growing season, when SZA was very high, both the surface 
BRDF and the atmospheric impact were extreme and sometimes canceled each other 
out. This phenomenon could be seen in the TOA NDVI time series where both sites 
(50% and 10% vegetation cover) at 60? latitude showed no intra-annual variations, 
although there were strong intra-annual variations in TOC NDVI.
3.3.2. Effectiveness of MODIS BRDF parameters for BRDF 
correction of AVHRR NDVI time series data
Fig.3-4 demonstrates the effectiveness of MODIS BRDF parameters for 
normalization of the modeled AVHRR TOC NDVI to 45? SZA and 0? VZAs. The 
inter-annual trend of 0.02 NDVI units per year in TOC NDVI was reduced 
50
considerably to a trend of 0.005 NDVI units per year for site 1 (~50% vegetation 
cover). The spurious seasonality was also considerably reduced. The TOC NDVI time 
series data for site 2 (~95% forest cover) had a weak inter-annual trend of 0.004 
NDVI units per year which was only slightly affected by BRDF correction. However, 
the large inter-annual variation in near infrared was greatly reduced. The BRDF 
correction technique was further tested directly on AVHRR data from LTDR dataset 
(Fig. 3-5). Although the AVHRR data used for this study did not include correction 
for aerosol, the BRDF normalization improved surface reflectance and NDVI trends 
considerably. In both modeled TOC and actual AVHRR data the high frequency noise 
in red and NIR channels due to VZA variation was also effectively removed (Fig. 3-
5).
51
Fig. 3-4: The effect of MODIS BRDF correction technique in removing the impact of 
orbital drift on top of canopy (TOC) NDVI data. On the left are the data without the 
BRDF correction and on the right are the data with the MODIS BRDF correction. 
Red, near infrared and NDVI are shown for 5 years of AVHRR data. The inter-annual 
NDVI trend was estimated by fitting a linear polynomial to yearly NDVI averages.
52
Fig. 3-5: This figure shows the impact of different levels of correction on AVHRR NDVI time series data. It can be observed that 
atmospheric correction of LTDR NDVI data removes negative trends introduced by atmosphere and exposes positive trends due to 
surface BRDF. It also shows the effectiveness of MODIS BRDF correction technique in removing the positive trends introduced by 
surface BRDF. The inter-annual NDVI trend was estimated by fitting a linear polynomial to yearly NDVI averages.
53
3.4. Conclusions
The analysis presented in this chapter indicates that the extent of vegetation 
cover is a crucial factor affecting sensitivity of TOC NDVI time series data to orbital 
drift. This sensitivity was found to be highest at vegetation cover close to 50%. At 
both 10% and 95% vegetation cover, the impact of SZA change on TOC NDVI was 
negligible (Fig. 3-2, 3-3). For dense tropical forest sites, the atmosphere was found to 
be the dominant source of inter-annual NDVI trends. SZA for sites at high latitudes 
showed prominent seasonality with very high solar zenith angles at the beginning and 
end of the growing season. At these high SZAs, impact of both surface and 
atmospheric BRDF was also extreme, thus severely decreasing the reliability of 
phenological metrics derived from NDVI data.
As noted before, TOC NDVI usually shows a positive relationship with solar 
zenith angle, while the atmosphere imposes a negative relationship in TOA NDVI 
data. It was observed that depending on the SZA, atmospheric composition, and 
vegetation cover of the site being studied, either one of these opposite relationships 
can dominate the other, sometimes even canceling each other out and causing a 
spurious improvement in the quality of the data. Thus the atmospheric correction 
exposed the surface BRDF effect on NDVI time series suppressed by atmospheric 
interference. Clearly with improvements in atmospheric correction, an operational 
BRDF correction technique significantly improves the precision of the AVHRR 
NDVI data. In this study it was found that the BRDF correction technique developed 
for MODIS data (Vermote et al. 2009a) provides a promising method to remove the 
54
contribution of surface BRDF to the impact of orbital drift on NDVI and is worth 
applying to future developments of AVHRR processing such as the new LTDR 
archive. 
55
Chapter 4: Quantification of Overall Uncertainty and 
Impact of Spatial and Temporal Aggregation 
4.1. Introduction
Although previous investigations of uncertainty in the AVHRR NDVI data 
have given some valuable insights (Cihlar et al. 1998; Eklundh 1995; Fensholt et al. 
2006a; Fensholt et al. 2006b; Gallo et al. 2005; Goward et al. 1993; Nagol et al. 
2009), most of them, including the research presented in chapters 2 and 3, do not 
include all sources of uncertainty. AVHRR NDVI data are normally used with some 
kind of temporal as well as spatial aggregation (averaging, maximum value or median 
value compositing, etc) to reduce the impact of various shortcomings in the data 
processing.  These  compositing  methods  are  generally  believed  to  decrease 
uncertainty in the data considerably (Cihlar et al. 1994; Holben 1986), yet there have 
been no studies to quantify the extent of these positive effects. Bearing in mind the 
many influential studies of Earth system science that have been based on AVHRR 
NDVI, a better understanding of uncertainty is of critical importance. The purpose of 
the research presented in this chapter was to estimate uncertainty in AVHRR NDVI 
data and to investigate the impact of various degrees of spatial and temporal 
aggregation.
56
4.2. Data and methods
The uncertainty in AVHRR NDVI data was estimated by comparison with 
coincident data from MODIS Aqua instrument. The relatively low uncertainty in 
MODIS data (0.007 to 0.017 NDVI units) (Vermote et al. 2006) when compared to 
that of AVHRR data (0.0206 to 0.078 NDVI units) (chapter-2) makes this approach of 
estimating uncertainty reasonable.  However, this method does not account for 
residual errors in the MODIS data itself, which can lead to an underestimation of 
uncertainty. The extent of this underestimation is indeterminable, because the spatial 
variability and the impact of spatial and temporal aggregation on the uncertainty in 
MODIS data is not yet determined. One chief advantage of using this framework is 
that, it allows the calculation of location specific estimates of uncertainty and also 
account for impacts of spatial and temporal aggregation. 
The temporal overlap between NOAA-16 and MODIS Aqua instruments (2003-
2006) was used for estimation of uncertainty. This study used NDVI data from three 
regions, Australia, the Sahel region in Africa, and Northern South-America, to sample 
a wide range of vegetation types and atmospheric compositions (Fig. 4-1).
The statistical framework defined in section 1.2.8 (chapter 1) was used to 
estimate uncertainty (random error) in the AVHRR NDVI data. The uncertainty 
statistic was calculated as the standard deviation of the data points relative to the 
systematic bias line. The systematic component (bias) was estimated by fitting a first 
order polynomial to the two dimensional scatter plot comparing AVHRR NDVI data 
to the reference data (MODIS Aqua data). 
57
Australia
Fig. 4-1: The three regions (Australia, Sahel, and northern South-America) and the locations of the sites (Table 4-1) where 
further study of the impact of spatial and temporal aggregation on uncertainty in AVHRR NDVI data was conducted. 
58
Site ID Region Land cover Type long lat
1 Australia Cropland 117.80 -32.65
2 Australia Savanna 143.00 -16.50
3 Australia Open Shrubland 134.65 -25.30
4 Australia Woody Savanna 147.75 -25.10
5 Australia Mixed (Grassland / open Shrubland) 134.20 -17.55
6 Sahel (Africa) Cropland -15.65 14.15
7 Sahel (Africa) Grassland 8.35 13.45
8 Sahel (Africa) Open Shrubland -0.45 15.90
9 Northern South America Mixed (Forest / Bare) -45.40 -10.10
10 Northern South America Mixed (Woody Savanna / Cropland) -59.95 -16.00
Table 4-1: Locations and land cover of the sites in Australia, Sahel region of Africa, 
and Northern South America where further study of impact of spatiotemporal 
aggregation on AVHRR NDVI uncertainty and its implication for study of inter-
annual NDVI time series trends was carried out.
4.2.1. AVHRR data
The AVHRR NDVI data from LTDR version-3 dataset were used in this study 
(http://ltdr.nascom. nasa.gov/ltdr/ltdr.html) (Pedelty et al. 2007). The product includes 
daily NDVI as well as surface reflectance data at a spatial resolution of 0.05?, which 
matches the Climate Modeling Grid (CMG) used for MODIS Terra and Aqua.
The  LTDR  preprocessing  includes  vicarious  calibration,  atmospheric 
correction for Rayleigh scattering, ozone, and water vapor, and an improved cloud 
mask technique (Vermote et al. 2009b). The AVHRR data were normalized to a 
uniform sun-sensor geometry using the BRDF correction technique developed for 
MODIS data (Vermote et al. 2009a). This technique has been demonstrated to be 
59
reasonably effective in removing the impact of orbital drift on AVHRR NDVI data 
(chapter 3).
4.2.2. Reference data (MODIS Aqua data)
The MODIS Aqua based CMG surface reflectance product (MYD09CMG) 
was used as reference data for estimating the uncertainty in LTDR AVHRR NDVI 
data. Data from MODIS Terra instrument were not included because of a more than 4 
hours of difference in overpass time when compared to AVHRR instruments, while 
Aqua platform has a near simultaneous overpass.
The MODIS CMG data product is estimated daily on a 0.05 degree grid using 
MODIS Level-1B bands 1-7 (459-2155nm). The MODIS CMG atmospheric 
correction scheme accounts for the effects of atmospheric gases, aerosols, and thin 
cirrus clouds. The data were further corrected for BRDF effect using the technique 
proposed by Vermote et al., (2009).
4.2.3. Impact of Spatial and Temporal Aggregation on AVHRR NDVI 
Uncertainty (Random Error)
The impact of spatial aggregation on the uncertainty in AVHRR NDVI data 
was investigated by comparison with reference data at various levels of spatial 
aggregation (1x1, 3x3, 5x5, 7x7 pixel matrix averages) for the dominant land cover 
types in the three regions (Australia, Sahel, and northern South-America) (Fig. 4-1). 
Since NDVI is a non-linear transformation, the spatial aggregation was performed in 
60
the spectral domain before calculation of NDVI. A land cover map derived from 
MODIS data (MOD12C1) for year 2004 was used in the analysis. MOD12C1 has a 
resolution of 0.05 degrees and is derived with the same algorithm used in the Global 1 
km Land Cover Type product (MOD12Q1) (Hodges 2002). This product contains 
three classification schemes, of which, IGBP (International Geosphere Biosphere 
Programme) global vegetation classification scheme was used in this study. The map 
gives the dominant land cover type and the sub-grid frequency of the other land cover 
classes.
All the pixels in each dominant land cover type, except the ones on the edge 
of the land cover patches, were included in the calculation of final precision statistic 
for that land cover type. Any spatially aggregated pixels with even one cloud 
contaminated pixel were excluded. Impact of commonly used temporal compositing 
techniques were investigated only for 3x3pixel spatial aggregates. The temporal 
compositing techniques included in this analysis were: 1) monthly maximum value, 
2) monthly median, 3) monthly average, 4) quarterly average, and 5) yearly average. 
4.2.4. Estimation of Location Specific Uncertainty and Evaluation 
of the Significance of Interannual Trends
In previous sections overall uncertainty estimates for different land cover 
types in various regions were addressed. However uncertainty at different locations 
within a land cover type can show significant variations, the impact of sub-pixel 
variability in NDVI was also not accounted for. AVHRR NDVI data from the same 
dataset, depending on the spatial and temporal compositing scheme used, can have 
61
different degrees uncertainty. This makes it necessary for an AVHRR NDVI data user 
to estimate uncertainty for the specific location and aggregation scheme they have 
chosen to use. 
Here ten sites from Australia, northern South America, and Sahel region of 
Africa (Table 4-1; Fig. 4-1) were used to illustrate the method for estimation of 
uncertainy in AVHRR NDVI data for specific locations and aggregation schemes. 
Two aggregation schemes involving spatial aggregation over 0.15?x0.15? area (3x3 
pixels) and 0.45?x0.45? area (9x9 pixels) and temporal aggregation over a month were 
examined. 
Using the uncertainty estimates for AVHRR NDVI data, to evaluate the 
significance of the observed inter-annual trends involves a few additional steps. Of 
the two aggregation schemes examined the one involving spatial aggregation over 
9x9pixel area was used to illustrate these steps. The 18 year period (1982 to 1999) 
was used for the trend analyses. To use the uncertainty estimates for validation of 
observed inter-annual NDVI change in pre NOAA-16 era (year ?2000), it was 
necessary to assume that the LTDR V3 processing sufficiently deals with the 
systematic errors arising from sensor degradation, and orbital drift.
The inter-annual trends in AVHRR NDVI data were calculated using the 
method utilized by Donohue et al., (2009) for calculating inter-annual fPAR trends in 
Australia. This method uses linear (ordinary least squares) regressions on a month-by-
month basis. Trends were estimated for each of the twelve months in the year (e.g., 
trend for January, February, March, etc.). Because these regressions are linear, the 
yearly or seasonal inter-annual trends can be estimated by averaging the monthly 
62
trends. The inter-annual trend calculated using this method is less vulnerable to data 
gaps. (Donohue et al. 2009)
The uncertainty estimate for each monthly inter-annual trend was calculated 
by introducing the NDVI uncertainty into the linear regression model as measurement 
error. The uncertainty in inter-annual trend estimated in this manner is a combination 
of measurement error (uncertainty in AVHRR NDVI data) and coefficient of 
determination (R-squared) of the linear regression model. Due to this the uncertainty 
estimates are sensitive to the length of time series and the data gaps also. The 
uncertainty statistic for the yearly inter-annual NDVI trend was then estimated by 
calculating quadratic mean of the uncertainty in the twelve monthly trends. The goal 
here was to demonstrate the calculation of thresholds for inter-annual NDVI change 
that must be exceeded for the observed trends to be considered significant.
63
4.3. Results
4.3.1. Impact of Spatial and Temporal Aggregation on AVHRR 
NDVI Uncertainty (Random Error)
Fig. 4-2:  Impact of spatial aggregation on uncertainty in LTDR AVHRR NDVI data. 
The spatial aggregation was performed in the spectral domain before calculation of 
NDVI. 
64
Uncertainty in the LTDR AVHRR NDVI data, to which no spatial or temporal 
aggregation was applied, varied from 0.036 in Australian open shrub-land to 0.11 for 
evergreen broadleaf forests in northern South-America (Fig. 4-2).  The investigation 
into the impact of spatial aggregation (Fig. 4-2) revealed that the uncertainty in 
AVHRR NDVI data decreases greatly as more pixels are aggregated. Even a 3x3pixel 
(9 pixels) aggregation considerably reduced the uncertainty in the data for all the land 
cover types in the three regions. For example, just by aggregation over 3x3 pixel 
matrices, the uncertainty in AVHRR NDVI measures in northern South-American 
evergreen broadleaf forests decreased from 0.11 to 0.06. Further aggregation over 7x7 
pixel matrices (49 pixels) further decreased the uncertainty to 0.048 NDVI units. The 
uncertainty for Australian woody savanna showed a similar improvement from 0.068 
to 0.039, and 0.029 at 3x3 and 7x7 pixel aggregation respectively. Uncertainty for 
other land cover types from all three regions, including Sahel, showed similar 
improvements. Most of the improvement was achieved with 3x3 pixel aggregation 
step. Although further spatial aggregation resulted in some reduction in uncertainty, it 
was relatively small (Fig. 4-2). Because of this, all further analyses of errors in 
AVHRR NDVI data were conducted on 3x3 pixel aggregates.
 Investigation into the impact of three different monthly temporal aggregation 
schemes (maximum value, median, and average) conducted on 3x3pixel spatial 
aggregates (Fig.4-3) revealed that all three of these compositing schemes resulted in 
only slight decrease in uncertainty. 
65
Fig. 4-3: Impact of various types of monthly temporal compositing techniques on 
uncertainty in LTDR AVHRR NDVI data.
66
Fig. 4-4: Impact of the temporal extent of aggregation on uncertainty in LTDR 
AVHRR NDVI data
A consistent decrease in uncertainty was observed with increase in temporal 
extent of aggregation (Fig. 4-4). For example, the uncertainty in AVHRR NDVI in 
Australian evergreen broadleaf forests decreased from 0.044 NDVI units for daily 
data to 0.036, 0.031, and 0.020 NDVI units for monthly, quarterly, and yearly 
temporal aggregation respectively. The uncertainty for northern South-American 
woody savanna showed a similar reduction of uncertainty from 0.056 NDVI units for 
67
daily data to 0.054, 0.042, and 0.029 NDVI units for monthly, quarterly, and yearly 
temporal aggregation respectively. Most of the other land cover types in all three 
regions showed similar pattern. However for northern South-American evergreen 
broadleaf forests, even yearly aggregation resulted in only ~10% (0.01 NDVI units) 
reductions in uncertainty. This may be due consistently hazy atmospheric conditions 
and limited availability of cloud free days.
Fig. 4-5:  The LTDR AVHRR NDVI uncertainty at the 10 study sites (Table 4-1). 
68
The examination of the location specific uncertainty estimates for monthly 
3x3 and 9x9 pixel aggregates for the 10 sites (Fig. 4-5) revealed that sites with sub-
pixel heterogeneity in NDVI benefited more by spatial aggregation. For example the 
uncertainty in AVHRR NDVI data for the mixed (grassland/open shrubland) site in 
Australia decreased from 0.028 to 0.018 NDVI units and the uncertainty for the 
mixed (Forest / Bare) site in northern South America decreased from 0.028 to 0.012 
NDVI units (Fig. 4-5). Similar improvements were observed at mixed (woody 
savanna / cropland) site in northern South America and the woody savanna site in 
Australia. On the other hand, increase in spatial extent of aggregation from 3x3pixel 
area to 9x9 pixel area had little impact on uncertainty estimates at sites which were 
homogeneous, e.g. the cropland, grassland, and open shrubland sites in Sahel and the 
open sbrubland site in Australia ,
The validation of the inter-annual NDVI change for 18 years (1982 to 1999) at 
the 10 sites (Fig. 4-6) revealed that only two sites, the grassland site (0.027 
NDVI/18years) and the open shrubland site (0.028 NDVI/18years) in the Sahel, 
showed an NDVI change significant at 67% confidence interval. Only the cropland 
site in Australia showed an NDVI change (0.07 NDVI/18years) that was significant at 
95% confidence interval. Although the Australian savanna site showed an NDVI 
change of 0.047 NDVI/18 years, it was not significant at even 67% confidence 
interval (Fig. 4-6). This showed that higher NDVI change does not always result in 
higher confidence. 
69
 
Fig. 4-6: Total NDVI change observed in the 18 year time period of 1982 to 1999. 
The red lines represent the uncertainty.
70
4.4. Discussion
The uncertainty estimates for un-aggregated LTDR AVHRR NDVI (0.036 to 
0.11 NDVI units) (Fig. 4-2) were found to be much higher than the inter-annual 
change considered significant by many previous studies (Jia et al. 2003; Piao et al. 
2003; Piao et al. 2006; Slayback et al. 2003; Young and Harris 2005). Most of this 
uncertainty can be ascribed to lack of comprehensive atmospheric correction in 
AVHRR dataset (Pedelty et al. 2007). Because the GAC sampling scheme can select a 
different sub-pixel area on every overpass, high sub-pixel variability in NDVI can 
considerably increase the uncertainty.
It was observed that even limited spatial aggregation of 3x3 pixels (0.15x0.15 
degrees) reduced the uncertainty considerably. However, sites with high sub-pixel 
heterogeneity benefited from further spatial aggregation (Fig. 4-5). The results 
presented in section 4.3.1 (Fig.4-2) showed that degrading the resolution of AVHRR 
data from 0.05 degrees to 0.15 degrees is a reasonable technique for reduction of 
uncertainty in AVHRR NDVI data. For many applications the benefit due to the 
reduction of uncertainty achieved in this way would outweigh the disadvantages due 
to decrease in spatial resolution.
It was also observed that, after 3x3pixel spatial aggregation, temporal 
aggregation to monthly time period resulted in only slight reduction in uncertainty 
(Fig. 4-3). The temporal aggregation of data to which no spatial aggregation was 
applied was also found to reduce uncertainty, but to a limited extent only. Moreover, 
in many parts of the world the temporal aggregation method is severely limited by the 
number of cloud free days available during a compositing period 
71
Of the three monthly temporal compositing techniques tested (maximum 
value, median, and average), the monthly averaging technique produced slightly but 
consistently better results. This is in contrast with the general understanding that 
MVC is the best compositing technique. This can be attributed to better cloud masks, 
improved atmospheric correction, and BRDF correction employed in the LTDR 
dataset when compared to earlier AVHRR NDVI datasets. These improvements in the 
data processing make the ability of MVC technique to decrease the impact of partial 
atmospheric correction and inaccurate cloud masks redundant.
The technique of evaluation of AVHRR NDVI data by comparison with 
MODIS data allows for estimation of uncertainty at any geographic location and 
aggregation scheme (Fig.4-5). This location-specific uncertainty can then be used to 
evaluate the significance of observed inter-annual NDVI trends and anomalies (Fig. 
4.6). This technique accounts for spatial heterogeneity of uncertainty in the AVHRR 
NDVI data as well as the impact of the specific spatial and temporal aggregation 
scheme being used. However, to be able to use the uncertainty estimates derived 
using data from years 2003 to 2006 to evaluate the observed inter-annual NDVI 
trends and anomalies in pre 2000 era, it is necessary to make sure that there is no 
significant sensor degradation, and orbital drift related non-stationarity in the data.
4.5. Conclusions
This  study  empirically  demonstrated  that,  in  many  regions,  minimal 
spatiotemporal aggregation considerably decreases the uncertainty in the LTDR 
AVHRR NDVI data. Aggregation beyond 3x3 pixels was found useful at sites with 
high sub-pixel NDVI heterogeneity. 
72
The framework for estimation of uncertainty presented in this chapter not only 
accounts for the impact of spatial and temporal aggregation but also provides a way to 
calculate location specific error estimates. These error estimates can be used to set 
informed confidence intervals for future global change and regional studies of inter-
annual vegetation dynamics. 
73
Chapter 5: Overview and Discussion
The goal of this dissertation was to systematically investigate and 
quantify the uncertainty in AVHRR NDVI data. The primary sources of uncertainty in 
the NDVI datasets derived from NOAA AVHRR data are: 1) partial atmospheric 
correction; 2) orbital drift; 3) sensor degradation; and 4) geolocation error. The 
second chapter of this dissertation investigated the impact of partial atmospheric 
correction and the third chapter dealt with non-stationarity introduced by orbital drift 
in the AVHRR NDVI time-series data. In the fourth chapter, instead of quantifying 
uncertainty from individual sources and then accounting for the complex interactions 
between each of them, overall uncertainty in AVHRR NDVI data was estimated 
directly by comparison with MODIS data for the years 2002-2006, the overlap period 
between the two sensors (MODIS Aqua and NOAA-16 AVHRR). This technique also 
facilitated the investigation of the impact of spatial and temporal aggregation. Next 
section will give an overview of the results from second, third, and fourth chapters 
followed by a discussion of implications of these results 
5.1. Overview of the Results
The incompleteness or absence of atmospheric correction is a major source of 
uncertainty in the AVHRR NDVI data. The research presented in chapter 2 assessed 
the uncertainty due to this shortcoming in three publicly available, processed AVHRR 
NDVI datasets (LTDR, PAL, and TOA). This assessment revealed that, of the three 
AVHRR NDVI datasets investigated, LTDR dataset has the lowest uncertainty. 
Uncertainty estimates for the LTDR dataset were found to be 0.021, 0.037, and 0.078 
74
NDVI units for clear, average, and hazy atmospheric conditions respectively (Fig. 5-
1; Table 2-1).
Fig. 5-1: Summary of results from Chapter 2. Uncertainty in NDVI for TOA, PAL, 
and LTDR datasets (Table 2-1). 
 
AVHRR NDVI data is nearly always used with some form of temporal 
compositing to decrease the impact of atmospheric interference and residual cloud 
contamination. For example, the GIMMS dataset is available with 15 day maximum 
value compositing and PAL dataset is available with 10 day and monthly maximum 
value compositing. Even LTDR dataset, which provides daily NDVI data, has been 
used with temporal compositing (Alcaraz-Segura et al. 2010). In most cases AVHRR 
NDVI data are further spatially aggregated, sometimes over vast regions. The spatial 
and  temporal  aggregation  is  generally  claimed  to  decrease  the  uncertainty 
75
considerably (Cihlar et al. 1998; Cihlar et al. 1994; Holben 1986). However, due to 
limited spatial and temporal coverage of AERONET data, the analysis presented in 
chapter 2 could not account for these improvements. The spatial heterogeneity of 
uncertainty was also not addressed. One more limitation was that, the network of 
AERONET sites did not sample all land cover types and latitudinal zones (Fig. 2-2).
Another major shortcoming in the AVHRR NDVI data is the non-stationarity 
caused by orbital drift. The orbital drift related non-stationarity in AVHRR NDVI 
time-series, not only  hinders the detection and interpretation of inter-annual 
vegetation dynamics; it also flouts the assumption of normal distribution essential for 
estimation of the uncertainty. The third chapter of this dissertation investigated the 
impact of this shortcoming and the effectiveness of BRDF correction technique 
proposed by Vermote et al., (2009) in dealing with it. 
The results indicated that, for sites with dense as well as sparse vegetation 
cover, atmosphere is the primary source of the orbital-drift-related, inter-annual 
NDVI trends. However, at vegetation cover close to 50%, the surface BRDF effect 
was the dominant source. Investigation into effectiveness of the BRDF correction 
technique (Vermote et al. 2009) indicated it to be a promising method for reducing the 
contribution of surface BRDF to the impact of orbital drift. Some of the positive 
attributes of this method are: 1) It does not require an a priori knowledge of the land 
cover type and vegetation structure; and 2) limited availability of cloud free data, 
does not affect the derivation of BRDF parameters.
However the BRDF correction method does not address the spurious negative inter-
annual NDVI trends caused by systematic increase in atmospheric path length due to 
76
orbital drift. The term ?spurious trend? here refers to the inter-annual trend in AVHRR 
NDVI time-series data caused, not by actual vegetation dynamics on earth surface, 
but by orbital drift or other shortcomings in the sensor system and post-processing. 
The magnitude of this trend can be estimated as the inter-annual trend of AVHRR and 
MODIS-Aqua NDVI differences (AVHRR NDVI ? MODIS-Aqua data). Fig. 5-2 
(site-2) illustrates the presence of the spurious negative trends using data at an 
agricultural site in eastern Henan province of China, as an example. More than 60% 
of the data at this site had AOT greater than 0.25. This constant hazy atmosphere and 
an annual 4? SZA increase due to orbital drift resulted in a spurious negative inter-
annual trend of 0.02 NDVI units for the four year period (2003-2006). The spurious 
negative inter-annual trends in AVHRR NDVI data were also observed in other 
regions like Sub-Sahelian tropical savannas, tropical forests, and in regions where 
occurrence of high AOT coincided with peak growing season.
In some vegetated hilly regions spurious positive NDVI trends were also 
observed. Fig. 5-2 (site-1) illustrates this using a site in north eastern Himalayan 
region. At this site an annual SZA increase of 4? during the four year period of 2003 
to 2006 resulted in a corresponding spurious positive inter-annual trend of 0.052 
NDVI units. Similar prominent trends were also observed in other vegetated hilly 
regions near Himalayan, and Andes mountain ranges.
77
Fig. 5-2: Examples of orbital drift related spurious inter-annual NDVI trends. The 
green, blue, and black colored data represent MODIS Aqua NDVI data, AVHRR 
NDVI data, and the difference between these two data (AVHRR NDVI - MODIS 
NDVI) respectively. The thick red line represents the spurious inter-annual trend. The 
magnitude of the spurious NDVI trend was determined by estimating slope of the 
linear curve fitted to the LTDR AVHRR NDVI and MODIS-Aqua data differences. 
Site-1 (115.05E, 34.25N) is located in vegetated hilly regions of northeastern 
Himalayan range. Site-2 (101.188E, 28.04N) has mixed agricultural and urban/built-
up land cover and is located in Eastern Henan province of China.
78
The results presented in the second and third chapters of this dissertation dealt 
with individual sources of uncertainty. The goal of the research presented in fourth 
chapter was to estimate the overall uncertainty in the AVHRR NDVI data directly by 
comparison with MODIS data. The impact of temporal and spatial aggregation was 
also examined. 
Only LTDR AVHRR NDVI dataset was included in this analysis. GIMMS 
dataset was excluded because its processing includes temporal smoothing, removal of 
non-vegetation NDVI signal, and manipulation of dynamic range making the direct 
comparison of individual NDVI data to the MODIS data unreasonable. The lack of 
BRDF correction in GIMMS data further complicates the situation. 
Investigation into the impact of various degrees of spatial aggregation (Fig. 4-
4 to 4-7) on the uncertainty in LTDR AVHRR NDVI data (section 4.2.1) indicated 
that minimal spatial aggregation (3x3 pixel spatial averaging) decreased the 
uncertainty considerably. Further spatial aggregation was generally found to be of 
little benefit (Fig. 4-5).
5.2. Global Uncertainty Maps
The method of location specific uncertainty estimation discussed in chapter 4 
can be used to create global and regional uncertainty maps for various spatial and 
temporal aggregation schemes. These maps can then be used to evaluate and 
understand the significance of inter-annual trends and anomalies in LTDR AVHRR 
NDVI data. Fig. 5-3 presents an example of an uncertainty map for LTDR AVHRR 
NDVI data with a 3x3 pixel spatial aggregation. This uncertainty map revealed a 
considerable amount of spatial heterogeneity of uncertainty. 
79
Fig. 5-3: Global uncertainty map of 3x3 pixel spatial aggregates (averages) for LTDR 
AVHRR NDVI data. Deserts, urban areas, wetlands, and the pixels with less than 30 data 
points were excluded from the analysis.
Fig. 5-4: Land cover map derived from MODIS data (MOD12C1) for year 2004 with 
IGBP classification schemes
80
A
C
D
E
B
Uncertainty
 The regions with low vegetation cover, like the open shrublands in Australia, 
southern Africa, southern South America, and southern North America showed 
uncertainty close to 0.02 NDVI units (Fig. 5-3; Fig. 5-4). Parts of grasslands in 
central Asia and Sahelian region also showed similarly low uncertainty (Fig. 5-3; Fig. 
5-4). This can be attributed to low sensitivity of NDVI to atmospheric interference  as 
well as BRDF effect, in sparsely vegetated regions (section 3.3.1; Fig. 3-2: Fig. 3-3).
Relatively high uncertainty was observed in regions with denser vegetation 
cover, like croplands, savannas, and forests. This may be due to higher sensitivity to 
atmospheric interference (Fig. 3-2). Uncertainty estimates higher than 0.035 were 
observed for many pixels in regions like the sub-Sahelian savannas, Bolivian tropical 
forests, and croplands in Indian Gangetic plains, North America, and eastern China, 
where high NDVI and high AOT occur simultaneously (Fig. 5-3; Fig. 5-4).  
In some regions (Fig. 5-3: Regions A, C, E) the non-stationarity in the 
AVHRR NDVI time-series data, due to atmospheric BRDF and orbital drift (Fig. 5-2: 
Site-2), was a significant contributor to the high uncertainty (> 0.06 NDVI units). 
Similarly high uncertainty estimates were also observed in some vegetated hilly 
regions (Fig. 5-3: Regions B, D). This can be attributed to surface BRDF related non-
stationarity described in Fig. 5-2 (Site-1).
81
Fig. 5-5: Mean uncertainty of each land cover type. Deserts, urban areas, wetlands, 
and the pixels with less than 30 data points were excluded from the analysis.
The summary, the AVHRR NDVI uncertainty estimates for each land cover 
type (Fig. 5-5; 5-4) revealed that the uncertainty varies from 0.025 NDVI units for 
open shrubland to 0.04 NDVI units for evergreen broadleaf forest. A closer 
examination of the uncertainty map (Fig. 5-3) revealed a considerable variability of 
uncertainty within each land cover type; this variability was observed to be primarily 
caused by a combination of atmospheric conditions, surface vegetation conditions, 
and topography.
One of the chief limitations for location specific uncertainty estimation, that is 
estimation of uncertainty at an individual AVHRR pixel, is the unavailability of 
adequate data for valid application of the statistical framework defined in section 
1.2.8. Due to this it was not possible to estimate uncertainty in many regions like 
large tracts of tropical forests, tundra, and boreal forests (Fig. 5-3). 
82
At many sites considerable seasonal variability in uncertainty was also 
observed. This seasonal variability showed a strong correlation with the magnitude of 
NDVI. One can account for this seasonal variability by estimating uncertainty for 
individual months, seasons, or NDVI bins (e.g. 0 to 0.2, 0.4 to 0.5, and 0.8 to 1.0 
NDVI units, etc.) separately. However such an endeavor is again hindered by 
inadequate data availability. One technique to overcome this problem is to estimate 
uncertainty for whole regions or sub-regions instead of individual pixels. 
5.3. Impact of Uncertainty in AVHRR-NDVI Data on Inter-annual 
Trends
The inter-annual trends in AVHRR NDVI data are generally estimated using 
linear (ordinary least squares) regression (Donohue et al., 2009; Jia et al., 2003; 
Slayback et al., 2003; Piao et al. 2006; Young and Harris 2005). The uncertainty in 
the inter-annual trend estimated using this technique depends not only on the 
magnitude of uncertainty in AVHRR NDVI data (measurement error) but also on the 
coefficient of determination (R-squared). Coefficient of determination is greatly 
affected by the length of the time series. This understanding was used to quantify the 
impact of uncertainty in AVHRR NDVI data on inter-annual NDVI trends for various 
lengths of time-series using a Monte Carlo approach. During this analysis the data 
were assumed stationary and without temporal gaps. A sample of the results, form the 
Monte Carlo simulations, is presented in Table 5-1. These results can be utilized to 
translate the magnitude of NDVI uncertainty to an estimate of the minimum change 
in NDVI that can be considered significant over a given time period. 
83
NDVI 
precision
Time  Series 
Length
Minimum Valid
NDVI Change
(95% Confidence Interval)
0.02 20 Years 0.0310
0.03 20 Years 0.0466
0.04 20 Years 0.0621
0.06 20 Years 0.0931
0.02 10 Years 0.0439
0.03 10 Years 0.0659
0.04 10 Years 0.0879
0.06 10 Years 0.1318
Table 5-1: Minimum Valid NDVI change for various degrees of NDVI uncertainty for 
10 and 20 year time period
For example it can be ascertained that, at 95% confidence interval, the 
uncertainty of 0.021 NDVI units for LTDR NDVI data in clear atmospheric 
conditions as reported in chapter 2 (Fig. 5-1) translates to an uncertainty of 0.046 and 
0.032 NDVI units for the total observed change over a period of 10 and 20 years 
respectively. This implies that when using LTDR NDVI data even under clear 
atmospheric conditions the observed changes in NDVI must be at least 0.046 over 10 
years and 0.032 over 20 years for the change to be significant. For average 
atmospheric conditions the observed changes in NDVI must be at least 0.081 over 10 
years or 0.057 over 20 years.
5.4. Implications of the results
The results from the investigation of uncertainty in AVHRR NDVI are of 
considerable significance and suggest that the conclusions of several influential 
studies of inter-annual vegetation dynamics that have treated small differences in 
84
AVHRR NDVI as indicators of significant change should be revisited. For example, 
Tucker et al., (2001) reported an increase of 0.031 and 0.033 NDVI units for a 10 
year period (1982-1991) in the 45?-75? latitudinal zone of North America and Eurasia 
respectively. An increase of 0.036 and 0.026 NDVI units was also reported for the 
same regions during the 8 year time period between 1992 and 1999. Another study by 
Zhou et al., (2001) reported an increase of 0.046 and 0.032 NDVI units over North 
America and Eurasia respectively over a period of 19 years (1981-1999). Both of 
these studies used GIMMS dataset, which has uncertainty similar to TOA data. When 
examined in the context of the uncertainty in AVHRR NDVI reported in chapter 2 
(Fig. 5-1), these NDVI changes cannot be considered significant. There are many 
similar examples where small changes in NDVI are reported as significant (Myneni et 
al. 1997; Slayback et al. 2003; Tucker et al. 1991). 
The spatially explicit uncertainty estimates presented in Fig 5-3 provide a tool 
for location specific examination of the significance of the reported NDVI changes. 
For example, Donohue et al., (2009) reported an increase of 0.0182 fPAR units in 26 
years (1981-2006) in Australia. This was estimated by averaging individual inter-
annual fPAR trends for all the pixels in Australia. The fPAR data used in this study 
was derived from PAL and Commonwealth Scientific and Industrial Research 
Organization (CSIRO) AVHRR NDVI datasets (King et al. 2003: Donohue et al. 
2008). The NDVI data were converted to fPAR using simple linear transformation 
equations. Here the relationship between the fPAR and LTDR AVHRR NDVI data 
was calculated in order to transform fPAR back to NDVI. The result was a change of 
0.012 NDVI units over the 26 years. The average uncertainty in LTDR AVHRR 
85
NDVI data for Australian continent was found to be 0.022 NDVI units. Using the 
technique presented in section 5.3, it was determined that a minimum change of 0.029 
NDVI units over a period of 26 years is required for the change to be considered 
significant. The LTDR dataset has a much advanced processing when compared to the 
dataset used in the Donohue et al. (2009) study, so the uncertainty of 0.022 NDVI 
units is likely to be a substantial underestimation. Even if this underestimated 
uncertainty is considered valid for this dataset the NDVI increase of 0.012 observed 
by Donohue et al. (2009) is not significant.
Fig. 5-6: Bioclimatic subzones in the Arctic slope of Alaska investigated by Jia et al., 
(2003).
As discussed earlier, in cases where inadequate availability of data makes 
estimation of location specific uncertainty impossible, the uncertainty can be 
estimated for a region or a sub-region as a whole. An example of the use of this 
Low Shrub
Erect Dwarf Shrub
Prostrate Dwarf Shrub
The Arctic Slope 
of Alaska
86
method is demonstrated here by estimating uncertainty in peak growing season NDVI 
data for the Erect Dwarf Shrub bioclimatic subzone of Alaskan Arctic slope (Fig. 5-6) 
(Jia et al. 2003). For this exercise uncertainty estimates for LTDR data were assumed 
applicable to GIMMS data used by Jia et al., (2003). The uncertainty in peak growing 
season NDVI data for the region was found to be 0.035 NDVI units. For the same 
bioclimatic subzone, Jia et al., (2003) reported an increase of 0.082 NDVI units over 
a period of 20 years (1981-2001). The uncertainty of 0.035 NDVI units implies that a 
minimum increase of 0.054 NDVI units is required for the change to be valid at 95 % 
confidence interval (Section 5.3). Because the reported increase of 0.082 NDVI units 
in yearly peak NDVI is more than this threshold (0.054) it can be considered 
significant.
The above two examples demonstrate the effectiveness of this method to 
interpret the significance of the inter-annual vegetation dynamics inferred from 
AVHRR NDVI data. This considerably enhances the value of the AVHRR NDVI data.
5.5. Future Directions
The uncertainty map presented in Fig. 5-3 does not cover large tracts of Earth 
surface and also does not account for temporal variability in uncertainty. This was 
primarily due to inadequate availability of cloud and snow free data. The discussion 
in sections 5.2 and 5.4 suggested that this problem can be overcome by estimating 
uncertainty for bioclimatic sub-zones instead of individual pixel locations. The 
bioclimatic sub-zones can be characterized by intersecting aerosol climatology maps 
with land-cover map, eco-region map, or even percent vegetation cover map. This 
approach will also provide adequate data for estimation of uncertainty for individual 
87
NDVI bins or compositing periods to account for seasonal variability in uncertainty. 
A global uncertainty map produced using this approach can be of considerable 
relevance to the Earth system science research community.
The uncertainty in AVHRR NDVI data estimated by direct comparison with 
MODIS Aqua data does not account for the uncertainty MODIS data themselves, 
which can be as high as 0.017 NDVI (Vermote et al. 2006). A better spatial 
characterization of uncertainty in MODIS data and the quantification of impact of 
spatial and temporal aggregation can considerably improve the uncertainty estimates 
for AVHRR data.
Even after BRDF correction and vicarious calibration, evidence of substantial 
non-stationarity was found in AVHRR NDVI data at vegetated areas experiencing 
constant hazy atmosphere and in vegetated hilly regions (Fig. 5-2). Presence of non-
stationarity leads to overestimation of uncertainty and also hampers the detection of 
inter-annual trends. Availability of a global map of degree of non-stationarity in 
AVHRR data could be of substantial value.
Most of the future use of AVHRR NDVI data is expected to be in conjunction 
with data from MODIS, or similar sensors like VIIRS (Gallo et al. 2005; Nagol et al. 
2009; Pedelty et al. 2007; van Leeuwen et al. 1999). In this context not only the 
random component of uncertainty but the systematic bias also becomes relevant. A 
detailed study of this aspect of uncertainty can considerably improve the inter-
calibration between NDVI data from AVHRR and MODIS instruments, as is being 
undertaken by the LTDR project. 
88
Both empirical (Tucker et al. 2005) as well as analytical (van Leeuwen et al. 
2006) equations have been developed for inter-calibration of AVHRR and MODIS 
data. Most of these do not account for spatial variability of the systematic bias caused 
by changes in surface and atmospheric conditions. By comparing AVHRR data 
directly to MODIS, more precise and location specific, inter-calibration equations can 
be developed. This kind of study would enhance the interoperability between AVHRR 
and MODIS datasets.
89
Bibliography
Abel, P., Guenther, B., Galimore, R.N., & Cooper, J.W. (1993). Calibration Results 
for NOAA-11 AVHRR Channel-1 And Channel-2 From Congruent Path 
Aircraft Observations. Journal of Atmospheric and Oceanic Technology, 10, 
493-508
Alcaraz-Segura, D., Liras, E., Tabik, S., Paruelo, J., Cabello, J. (2010). Evaluating the 
Consistency of the 1982-1999 NDVI Trends in the Iberian Peninsula across 
Four Time-series Derived from the AVHRR Sensor: LTDR, GIMMS, FASIR, 
and PAL-II . Sensors, 10, 1291-1314
Anyamba, A., & Eastman, J.R. (1996). Interannual variability of NDVI over Africa 
and its relation to El Nino Southern Oscillation. International Journal of 
Remote Sensing, 17, 2533-2548
Bacour, C., Breon, F.M., & Maignan, F. (2006). Normalization of the directional 
effects  in  NOAA-AVHRR  reflectance  measurements  for  an  improved 
monitoring of vegetation cycles. Remote Sensing of Environment, 102, 402-
413
Baldocchi, D., Falge, E., Gu, L.H., Olson, R., Hollinger, D., Running, S., Anthoni, P., 
Bernhofer, C., Davis, K., Evans, R., Fuentes, J., Goldstein, A., Katul, G., Law, 
B., Lee, X.H., Malhi, Y., Meyers, T., Munger, W., Oechel, W., U, K.T.P., 
Pilegaard, K., Schmid, H.P., Valentini, R., Verma, S., Vesala, T., Wilson, K., & 
Wofsy, S. (2001). FLUXNET: A new tool to study the temporal and spatial 
variability of ecosystem-scale carbon dioxide, water vapor, and energy flux 
densities. Bulletin of the American Meteorological Society, 82, 2415-2434
90
Barbosa, P.M., Gregoire, J.M., & Pereira, J.M.C. (1999). An algorithm for extracting 
burned areas from time series of AVHRR GAC data applied at a continental 
scale. Remote Sensing of Environment, 69, 253-263
Bell, S. (1999). Measurement Good Practice Guide No. 11. A Beginner's Guide to 
Uncertainty of Measurement. Tech. rep., National Physical Laboratory
Bounoua, L., Collatz, G.J., Los, S.O., Sellers, P.J., Dazlich, D.A., Tucker, C.J., & 
Randall, D.A. (2000). Sensitivity of climate to changes in NDVI. Journal of 
Climate, 13, 2277-2292
Brown, M.E., Pinzon, J.E., Didan, K., Morisette, J.T., & Tucker, C.J. (2006). 
Evaluation of the consistency of long-term NDVI time series derived from 
AVHRR, SPOT-Vegetation, SeaWiFS, MODIS, and Landsat ETM+ sensors. 
IEEE Transactions on Geoscience and Remote Sensing, 44, 1787-1793
Cao, M.K., Prince, S.D., Small, J., & Goetz, S.J. (2004). Remotely sensed interannual 
variations and trends in terestrial net primary productivity 1981-2000. 
Ecosystems, 7, 233-242
Cihlar, J., Chen, J.M., Li, Z., Huang, F., Latifovic, R., & Dixon, R. (1998). Can 
interannual land surface signal be discerned in composite AVHRR data? 
Journal of Geophysical Research-Atmospheres, 103, 23163-23172
Cihlar, J., Manak, D., & Diorio, M. (1994). Evaluation of Compositing Algorithms 
for AVHRR Data over Land. IEEE Transactions on Geoscience and Remote 
Sensing, 32, 427-437
91
Cohen, W.B., & Justice, C.O. (1999). Validating MODIS terrestrial ecology products: 
Linking in situ and satellite measurements. Remote Sensing of Environment, 
70, 1-3
Deering, D.W., Middleton, E.M., Irons, J.R., Blad, B.L., Waltershea, E.A., Hays, C.J., 
Walthall, C., Eck, T.F., Ahmad, S.P., & Banerjee, B.P. (1992). Prairie 
Grassland Bidirectional Reflectances Measured by Different Instruments at 
the Fife Site.  Journal of Geophysical Research-Atmospheres, 97, 18887-
18903
DeFries, R., Hansen, M., & Townshend, J. (1995). Global discrimination of land 
cover types from metrics derived from AVHRR pathfinder data.  Remote 
Sensing of Environment, 54, 209-222
DeFries, R.S., Townshend, J.R.G., & Hansen, M.C. (1999). Continuous fields of 
vegetation characteristics at the global scale at 1-km resolution. Journal of 
Geophysical Research-Atmospheres, 104, 16911-16923
Donohue, R.J., McVicar, T.R., & Roderick, M.L. (2009). Climate-related trends in 
Australian vegetation cover as inferred from satellite observations, 1981-2006. 
Global Change Biology, 15, 1025-1039
Eidenshink, J. (2006). A 16-year time series of 1 km AVHRR satellite data of the 
conterminous United States and Alaska. Photogrammetric Engineering and 
Remote Sensing, 72, 1027-1035
Eidenshink, J.C. (1992). The 1990 Conterminous United-States AVHRR Data Set. 
Photogrammetric Engineering and Remote Sensing, 58, 809-813
92
Eidenshink, J.C., & Faundeen, J.L. (1994). The 1km AVHRR Global Land Data Set - 
1st Stages in Implementation. International Journal of Remote Sensing, 15, 
3443-3462
Eklundh, L.R. (1995). Noise Estimation in NOAA AVHRR Maximum-Value 
Composite NDVI Images. International Journal of Remote Sensing, 16, 2955-
2962
El Saleous, N.Z., Vermote, E.F., Justice, C.O., Townshend, J.R.G., Tucker, C.J., & 
Goward, S.N. (2000). Improvements in the global biospheric record from the 
Advanced Very High Resolution Radiometer (AVHRR). International Journal 
of Remote Sensing, 21, 1251-1277
Fang, J.Y., Piao, S., Field, C.B., Pan, Y., Guo, Q.H., Zhou, L.M., Peng, C.H., & Tao, 
S. (2003). Increasing net primary production in China from 1982 to 1999. 
Frontiers in Ecology and the Environment, 1, 293-297
Fensholt, R., Nielsen, T.T., & Stisen, S. (2006a). Evaluation of AVHRR PAL and 
GIMMS 10-day composite NDVI time series products using SPOT-4 
vegetation data for the African continent. International Journal of Remote 
Sensing, 27, 2719-2733
Fensholt, R., Rasmussen, K., Nielsen, T.T., & Mbow, C. (2009). Evaluation of earth 
observation based long term vegetation trends - Intercomparing NDVI time 
series trend analysis consistency of Sahel from AVHRR GIMMS, Terra 
MODIS and SPOT VGT data. Remote Sensing of Environment, 113, 1886-
1898
93
Fensholt, R., & Sandholt, I. (2005). Evaluation of MODIS and NOAA AVHRR 
vegetation indices with in situ measurements in a semi-arid environment. 
International Journal of Remote Sensing, 26, 2561-2594
Fensholt, R., Sandholt, I., & Stisen, S. (2006b). Evaluating MODIS, MERIS, and 
VEGETATION - Vegetation indices using in situ measurements in a semiarid 
environment.  IEEE Transactions on Geoscience and Remote Sensing, 44, 
1774-1786
Franklin, J.F., Bledsoe, C.S., & Callahan, J.T. (1990). Contributions of the Long-Term 
Ecological Research-Program - an Expanded Network of Scientists, Sites, and 
Programs Can Provide Crucial Comparative Analyses. Bioscience, 40, 509-
523
Gallo, K., Li, L., Reed, B., Eidenshink, J., & Dwyer, J. (2005). Multi-platform 
comparisons of MODIS and AVHRR normalized difference vegetation index 
data. Remote Sensing of Environment, 99, 221-231
Gorman, A.J., & McGregor, J. (1994). Some Considerations for Using AVHRR Data 
in Climatological Studies - Instrument Performance. International Journal of 
Remote Sensing, 15, 549-565
Goward, S.N., Dye, D.G., Turner, S., & Yang, J. (1993). Objective Assessment of the 
NOAA Global Vegetation Index Data Product.  International Journal of 
Remote Sensing, 14, 3365-3394
Gutman, G., Tarpley, D., Ignatov, A., & Olson, S. (1995). The Enhanced NOAA 
Global Land Dataset from the Advanced Very High-Resolution Radiometer. 
Bulletin of the American Meteorological Society, 76, 1141-1156
94
Gutman, G.G. (1999). On the use of long-term global data of land reflectances and 
vegetation indices derived from the advanced very high resolution radiometer. 
Journal of Geophysical Research-Atmospheres, 104, 6241-6255
Hatfield, J.L., Kanemasu, E.T., Asrar, G., Jackson, R.D., Pinter, P.J., Reginato, R.J., & 
Idso, S.B. (1985). Leaf-Area Estimates from Spectral Measurements over 
Various Planting Dates of Wheat. International Journal of Remote Sensing, 6, 
167-175
Heidinger, A.K., Cao, C.Y., & Sullivan, J.T. (2002). Using Moderate Resolution 
Imaging Spectrometer (MODIS) to calibrate advanced very high resolution 
radiometer  reflectance  channels.  Journal  of  Geophysical  Research-
Atmospheres, 107
Hicke, J.A., Asner, G.P., Randerson, J.T., Tucker, C., Los, S., Birdsey, R., Jenkins, 
J.C., Field, C., & Holland, E. (2002). Satellite-derived increases in net primary 
productivity across North America, 1982-1998. Geophysical Research Letters, 
29
Hodges, J. (2002). MOD12 Land Cover and Land Cover Dynamics Products User 
Guide. Department of Geography, Boston University 
Holben, B.N. (1986). Characteristics of Maximum-Value Composite Images from 
Temporal AVHRR Data. International Journal of Remote Sensing, 7, 1417-
1434
Holben, B.N., Kaufman, Y.J., & Kendall, J.D. (1990). NOAA-11 AVHRR Visible and 
Near-IR Inflight Calibration. International Journal of Remote Sensing, 11, 
1511-1519
95
Holben, B.N., Tanre, D., Smirnov, A., Eck, T.F., Slutsker, I., Abuhassan, N., 
Newcomb, W.W., Schafer, J.S., Chatenet, B., Lavenu, F., Kaufman, Y.J., 
Castle, J.V., Setzer, A., Markham, B., Clark, D., Frouin, R., Halthore, R., 
Karneli, A., O'Neill, N.T., Pietras, C., Pinker, R.T., Voss, K., & Zibordi, G. 
(2001). An emerging ground-based aerosol climatology: Aerosol optical depth 
from AERONET. Journal of Geophysical Research-Atmospheres, 106, 12067-
12097
Huete, A.R., & Jackson, R.D. (1987). Suitability of Spectral Indexes for Evaluating 
Vegetation  Characteristics  on  Arid  Rangelands.  Remote  Sensing  of 
Environment, 23, 213-&
Ichii, K., Kawabata, A., & Yamaguchi, Y. (2002). Global correlation analysis for 
NDVI and climatic variables and NDVI trends: 1982-1990.  International 
Journal of Remote Sensing, 23, 3873-3878
James, M.E., & Kalluri, S.N.V. (1994). The Pathfinder AVHRR Land Data Set - an 
Improved  Coarse  Resolution  Data  Set  for  Terrestrial  Monitoring. 
International Journal of Remote Sensing, 15, 3347-3363
Ji, L., Gallo, K., Eidenshink, J.C., & Dwyer, J. (2008). Agreement evaluation of 
AVHRR and MODIS 16-day composite NDVI data sets.  International 
Journal of Remote Sensing, 29, 4839-4861
Jia, G.S.J., Epstein, H.E., & Walker, D.A. (2003). Greening of arctic Alaska, 1981-
2001. Geophysical Research Letters, 30
Jin, M. (2004). Analysis of land skin temperature using AVHRR observations. 
Bulletin of the American Meteorological Society, 85, 587-+
96
Jupp, D.L.B., & Strahler, A.H. (1991). A Hotspot Model for Leaf Canopies. Remote 
Sensing of Environment, 38, 193-210
Justice, C., Belward, A., Morisette, J., Lewis, P., Privette, J., & Baret, F. (2000). 
Developments in the 'validation' of satellite sensor products for the study of 
the land surface. International Journal of Remote Sensing, 21, 3383-3390
Kangas, M., Heikinheimo, M., & Laine, V. (2001). Accuracy of NOAA AVHRR-
based surface reflectance over a winter-time boreal surface - comparison with 
aircraft measurements and land-cover information. Theoretical and Applied 
Climatology, 70, 231-244
Kaufman, Y.J., & Holben, B.N. (1993). Calibration of the AVHRR Visible And Near-
IR Bands By Atmospheric Scattering, Ocean Glint And Desert Reflection. 
International Journal of Remote Sensing, 14, 21-52
Kaufmann, R.K., Zhou, L.M., Knyazikhin, Y., Shabanov, N.V., Myneni, R.B., & 
Tucker, C.J. (2000). Effect of orbital drift and sensor changes on the time 
series of AVHRR vegetation index data. IEEE Transactions on Geoscience 
and Remote Sensing, 38, 2584-2597
Kidwell, K.B. (1994). Global Vegetation Index User Guide, U.S. Dept. of Commerce, 
NOAA/NESDIS, Satellite Data Services Division, Nat. Climatic Data Center, 
Asheville, N. C.
Kidwell, K.B. (1997). NOAA Global Vegetation Index User?s Guide (Revision), U.S. 
Dept. of Commerce, NOAA/NESDIS, Climate Services Division, Nat. 
Climatic Data Center, Suitland, Md.
97
Kidwell, K.B. (1998). NOAA polar orbiter data user's guide. U.S. Department of 
Commerce, NESDIS. NOAA, National Climatic Data Center, Satellite Data 
Services Division, Washington, D.C.
Kogan, F.N., & Zhu, X. (2001). Evolution of long-term errors in NDVI time series: 
1985-1999.  Calibration and Characterization of Satellite Sensors and 
Accuracy of Derived Physical Parameters, 28, 149-153
Lambin, E.F., & Ehrlich, D. (1997).  Land-cover changes in sub-Saharan Africa 
(1982-1991): Application of a change index based on remotely sensed surface 
temperature and vegetation indices at a continental scale. Remote Sensing of 
Environment, 61, 181-200
Latifovic, R., Cihlar, J., & Chen, J. (2003). A comparison of BRDF models for the 
normalization  of  satellite  optical data  to  a  standard  sun-target-sensor 
geometry. IEEE Transactions on Geoscience and Remote Sensing, 41, 1889-
1898
Latifovic, R., Trishchenko, A.P., Chen, J., Park, W.B., Khlopenkov, K.V., Fernandes, 
R., Pouliot, D., Ungureanu, C., Luo, Y., Wang, S., Davidson, A., & Cihlar, J. 
(2005). Generating historical AVHRR 1 km baseline satellite data records over 
Canada suitable for climate change studies.  Canadian Journal of Remote 
Sensing, 31, 324-346
Lee, R., Yu, F., Price, K.P., Ellis, J., & Shi, P. (2002). Evaluating vegetation 
phenological patterns in Inner Mongolia using NDVI time-series analysis. 
International Journal of Remote Sensing, 23, 2505-2512
98
Liu, W.T.H., Massambani, O., & Nobre, C.A. (1994). Satellite Recorded Vegetation 
Response to Drought in Brazil. International Journal of Climatology, 14, 343-
354
Loeb, N.G. (1997). In-flight calibration of NOAA AVHRR visible and near-IR bands 
over Greenland and Antarctica. International Journal of Remote Sensing, 18, 
477-490
Los, S.O. (1998). Estimation of the ratio of sensor degradation between NOAA 
AVHRR  channels  1  and  2  from  monthly  NDVI  composites.  IEEE 
Transactions on Geoscience and Remote Sensing, 36, 206-213
Los, S.O., North, P.R.J., Grey, W.M.F., & Barnsley, M.J. (2005). A method to convert 
AVHRR Normalized Difference Vegetation Index time series to a standard 
viewing and illumination geometry. Remote Sensing of Environment, 99, 400-
411
Lucht, W., Prentice, I.C., Myneni, R.B., Sitch, S., Friedlingstein, P., Cramer, W., 
Bousquet, P., Buermann, W., & Smith, B. (2002). Climatic control of the high-
latitude vegetation greening trend and Pinatubo effect. Science, 296, 1687-
1689
Malingreau, J.P., & Belward, A.S. (1994). Recent Activities in the European-
Community for the Creation and Analysis of Global AVHRR Data Sets. 
International Journal of Remote Sensing, 15, 3397-3416
McPeters, R.D., & Labow, G.J. (1996). An assessment of the accuracy of 14.5 years 
of Nimbus 7 TOMS Version 7 ozone data by comparison with the Dobson 
network. Geophysical Research Letters, 23, 3695-3698
99
Morisette, J.T., Privette, J.L., & Justice, C.O. (2002). A framework for the validation 
of MODIS Land products. Remote Sensing of Environment, 83, 77-96
Myneni, R.B., Keeling, C.D., Tucker, C.J., Asrar, G., & Nemani, R.R. (1997a). 
Increased plant growth in the northern high latitudes from 1981 to 1991. 
Nature, 386, 698-702
Myneni, R.B., Nemani, R.R., & Running, S.W. (1997b). Estimation of global leaf 
area  index  and  absorbed  par  using  radiative  transfer  models.  IEEE 
Transactions on Geoscience and Remote Sensing, 35, 1380-1393
Myneni, R.B., Tucker, C.J., Asrar, G., & Keeling, C.D. (1998). Interannual variations 
in satellite-sensed vegetation index data from 1981 to 1991.  Journal of 
Geophysical Research-Atmospheres, 103, 6145-6160
Nagol, J.R., Vermote, E.F., & Prince, S.D. (2009). Effects of atmospheric variation on 
AVHRR NDVI data. Remote Sensing of Environment, 113, 392-397
North, P., Plummer, S.E., Deering, D.W., & Leroy, M. (1996). Validation of a BRDF 
model in Boreal forest. Proc. of International Geoscience and Remote Sensing 
Symposium. Lincoln, Nebraska, USA (27? 31 May, 1996).
North, P.R.J. (1996). Three-dimensional forest light interaction model using a Monte 
Carlo method. IEEE Transactions on Geoscience and Remote Sensing, 34, 
946-956
Ouaidrari, H., El Saleous, N., Vermote, E.F., Townshend, J.R., & Goward, S.N. 
(2003). AVHRR Land Pathfinder II (ALP II) data set: evaluation and inter-
comparison with other data sets. International Journal of Remote Sensing, 24, 
135-142
100
Paruelo, J.M., Garbulsky, M.F., Guerschman, J.P., & Jobbagy, E.G. (2004). Two 
decades of Normalized Difference Vegetation Index changes in South 
America: identifying the imprint of global change. International Journal of 
Remote Sensing, 25, 2793-2806
Pedelty, J., Masuoka, E.B., Brown, M.E., Pinzon, J., Tucker, C.J., Vermote, E.F., 
Prince, S.D., Justice, C.O., Nagol, J., Roy, D., Junchang, J., Schaaf, C., 
Jicheng, L., Privette, J.L., & Pinheiro, A. (2007). Generating a long-term land 
data record from the AVHRR and MODIS Instruments.  Geoscience and 
Remote Sensing Symposium, 2007. IGARSS 2007. IEEE International, 1021-
1025
Piao, S.L., Fang, J.Y., Ji, W., Guo, Q.H., Ke, J.H., & Tao, S. (2004). Variation in a 
satellite-based vegetation index in relation to climate in China. Journal of 
Vegetation Science, 15, 219-226
Piao, S.L., Fang, J.Y., Zhou, L.M., Guo, Q.H., Henderson, M., Ji, W., Li, Y., & Tao, S. 
(2003). Interannual variations of monthly and seasonal normalized difference 
vegetation index (NDVI) in China from 1982 to 1999. Journal of Geophysical 
Research-Atmospheres, 108
Piao, S.L., Mohammat, A., Fang, J.Y., Cai, Q., & Feng, J.M. (2006). NDVI-based 
increase in growth of temperate grasslands and its responses to climate 
changes  in  China.  Global  Environmental  Change-Human  and  Policy 
Dimensions, 16, 340-348
Pinty, B., Widlowski, J.L., Taberner, M., Gobron, N., Verstraete, M.M., Disney, M., 
Gascon, F., Gastellu, J.P., Jiang, L., Kuusk, A., Lewis, P., Li, X., Ni-Meister, 
101
W., Nilson, T., North, P., Qin, W., Su, L., Tang, S., Thompson, R., Verhoef, W., 
Wang, H., Wang, J., Yan, G., & Zang, H. (2004). Radiation Transfer Model 
Intercomparison (RAMI) exercise: Results from the second phase. Journal of 
Geophysical Research-Atmospheres, 109
Pinzon, J., Brown, M.E., & Tucker, C.J. (2005). Satellite time series correction of 
orbital drift artifacts using empirical mode decomposition. In EMD and its 
Applieations, N.E.Huang and SS.P, Shen (Eds) (Singapore: World Scientific 
Publishers)
Pouliot, D., Latifovic, R., Fernandes, R., & Olthof, I. (2010).  Evaluation of 
compositing period and AVHRR and MERIS combination for improvement of 
spring  phenology  detection  in  deciduous  forests.  Remote  Sensing  of 
Environment
Prihodko, L., & Goward, S.N. (1997). Estimation of air temperature from remotely 
sensed surface observations. Remote Sensing of Environment, 60, 335-346
Prince, S.D., & Goward, S.N. (1995). Global primary production: A remote sensing 
approach. Journal of Biogeography, 22, 815-835
Privette, J.L., Eck, T.F., & Deering, D.W. (1997). Estimating spectral albedo and 
nadir  reflectance  through  inversion  of  simple  BRDF  models  with 
AVHRR/MODIS-like data. Journal of Geophysical Research-Atmospheres, 
102, 29529-29542
Privette, J.L., Fowler, C., Wick, G.A., Baldwin, D., & Emery, W.J. (1995). Effects of 
Orbital Drift on Advanced Very High-Resolution Radiometer Products - 
102
Normalized Difference Vegetation Index and Sea-Surface Temperature. 
Remote Sensing of Environment, 53, 164-171
Rao, C.R.N., & Chen, J. (1995). INTERSATELLITE CALIBRATION LINKAGES 
FOR  THE  VISIBLE  AND  NEAR-INFARED  CHANNELS  OF  THE 
ADVANCED  VERY  HIGH-RESOLUTION  RADIOMETER  ON  THE 
NOAA-7, NOAA-9, AND NOAA-11 SPACECRAFT. International Journal 
of Remote Sensing, 16, 1931-1942
Rao, C.R.N., & Chen, J.H. (1996). Post-launch calibration of the visible and near-
infrared channels of the advanced very high resolution radiometer on the 
NOAA-14 spacecraft. International Journal of Remote Sensing, 17, 2743-
2747
Robel, J. (2007). NOAA KLM USER'S GUIDE (with NOAA-N,-P SUPPLEMENT). 
U.S. Department of Commerce, NESDIS. NOAA, National Climatic Data 
Center, Satellite Data Services Division, Washington, D.C.
Roujean, J.L., Leroy, M., & Deschamps, P.Y. (1992). A Bidirectional Reflectance 
Model of the Earths Surface for the Correction of Remote-Sensing Data. 
Journal of Geophysical Research-Atmospheres, 97, 20455-20468
Rouse, J.W., Haas, R.H., Jr, Schell, J.A., & Deering, D.W. (1973). Monitoring 
vegetation systems in the Great Plains with ERTS. Third ERTS Symposium, 
NASA, SP-351 309-317
Runnstrom, M.C. (2000). Is northern China winning the battle against desertification? 
Satellite remote sensing as a tool to study biomass trends on the Ordos Plateau 
in semiarid China. Ambio, 29, 468-476
103
Sellers, P.J., Los, S.O., Tucker, C.J., Justice, C.O., Dazlich, D.A., Collatz, G.J., & 
Randall, D.A. (1996). A revised land surface parameterization (SiB2) for 
atmospheric  GCMs  .2.  The  generation  of global  fields  of  terrestrial 
biophysical parameters from satellite data. Journal of Climate, 9, 706-737
Slayback, D.A., Pinzon, J.E., Los, S.O., & Tucker, C.J. (2003). Northern hemisphere 
photosynthetic trends 1982-99. Global Change Biology, 9, 1-15
Stellmes, M., Udelhoven, T., Roder, A., Sonnenschein, R., & Hill, J. (2010). Dryland 
observation at local and regional scale - Comparison of Landsat TM/ETM 
plus and NOAA AVHRR time series. Remote Sensing of Environment, 114, 
2111-2125
Stockli, R., & Vidale, P.L. (2004). European plant phenology and climate as seen in a 
20-year AVHRR land-surface parameter dataset.  International Journal of 
Remote Sensing, 25, 3303-3330
Stowe, L.L., Davis, P., & McClain, E.P. (1995). Evaluating the Clavr (Clouds from 
AVHRR) Phase-I Cloud Cover Experimental Product. Satellite Monitoring of 
the Earth's Surface and Atmosphere, 16, 21-24
Swap, B., Suttles, T., Annegarn, H., Scorgie, Y., Closs, J., Privette, J., & B., C. 
( 2000). Report on SAFARI 2000 outreach activities, intensive field campaign 
planning meeting, and data management workshop. Earth Observer, 12 21?25
Swinnen, E., Eerens, H., Heyns, W., Piccard, I., Viaene, P., & Claes, P. (2007). An 
integrated long time series of 1 km resolution NDVI for Europe from the 
NOAA-AVHRR  and  SPOT-VEGETATION  sensors.  Proceedings  of 
MultiTemp 2007: 4th International workshop on the analysis of multi-
104
temporal remote sensing images?, 18-20 July, 2007, Leuven, Belgium. Eds. 
De Lannoy G. et al., Leuven, Gent: Katholieke Universiteit Leuven, 
Universiteit Gent, ISBN 1-4244-0846-6 (2007).
Tanre, D., Holben, B.N., & Kaufman, Y.J. (1992). ATMOSPHERIC CORRECTION 
ALGORITHM  FOR  NOAA-AVHRR  PRODUCTS  -  THEORY  AND 
APPLICATION. IEEE Transactions on Geoscience and Remote Sensing, 30, 
231-248
Teillet, P.M., El Saleous, N., Hansen, M.C., Eidenshink, J.C., Justice, C.O., & 
Townshend, J.R.G. (2000). An evaluation of the global 1-km AVHRR land 
dataset. International Journal of Remote Sensing, 21, 1987-2021
Teillet, P.M., & Holben, B.N. (1994). Towards Operational Radiometric Calibration 
of NOAA AVHRR Imagery in the Visible and Near-Infrared Channels. 
Canadian Journal of Remote Sensing, 20, 1-10
Teillet, P.M., Slater, P.N., Ding, Y., Santer, R.P., Jackson, R.D., & Moran, M.S. 
(1990). 3 METHODS FOR THE ABSOLUTE CALIBRATION OF THE 
NOAA AVHRR SENSORS IN-FLIGHT. Remote Sensing of Environment, 31, 
105-120
Townshend, J.R.G., & Tucker, C.J. (1981). Utility of AVHRR of NOAA 6 and 7 for 
vegetation mapping. In, Matching Remote Sensing Technologies and their 
Applications (Proceedings of an international Conference London, December 
1981) (p. 97)
Tucker, C.J. (1979). Red and Photographic Infrared Linear Combinations for 
Monitoring Vegetation. Remote Sensing of Environment, 8, 127-150
105
Tucker, C.J., Dregne, H.E., & Newcomb, W.W. (1991). Expansion and Contraction of 
the Sahara Desert from 1980 to 1990. Science, 253, 299-301
Tucker, C.J., Gatlin, J.A., & Schneider, S.R. (1984). Monitoring Vegetation in the 
Nile Delta with NOAA-6 and NOAA-7 AVHRR Imagery. Photogrammetric 
Engineering and Remote Sensing, 50, 53-61
Tucker, C.J., Pinzon, J.E., Brown, M.E., Slayback, D.A., Pak, E.W., Mahoney, R., 
Vermote, E.F., & El Saleous, N. (2005). An extended AVHRR 8-km NDVI 
dataset  compatible  with  MODIS  and  SPOT  vegetation  NDVI  data. 
International Journal of Remote Sensing, 26, 4485-4498
Tucker, C.J., Slayback, D.A., Pinzon, J.E., Los, S.O., Myneni, R.B., & Taylor, M.G. 
(2001). Higher northern latitude normalized difference vegetation index and 
growing  season  trends  from 1982 to 1999.  International  Journal  of 
Biometeorology, 45, 184-190
van Leeuwen, W.J.D., Huete, A.R., & Laing, T.W. (1999). MODIS vegetation index 
compositing approach: A prototype with AVHRR data. Remote Sensing of 
Environment, 69, 264-280
Vermote, E., Justice, C.O., & Breon, F.M. (2009a). Towards a Generalized Approach 
for Correction of the BRDF Effect in MODIS Directional Reflectances. IEEE 
Transactions on Geoscience and Remote Sensing, 47, 898-908
Vermote, E., & Kaufman, Y.J. (1995). Absolute Calibration Of AVHRR Visible And 
Near-Infrared Channels Using Ocean And Cloud Views. International Journal 
of Remote Sensing, 16, 2317-2340
106
Vermote, E., Kotchenova, S., & Roa, J. (2006). MODIS surface radiation data 
products maintenance and refinement.  MODIS Science Team Meeting, 
College Park, Maryland, Octuber 31st - Novemeber 2nd , Department of 
Geography, University of Maryland, NASA/GSFC Code 614.5.
Vermote, E.F., Justice, C.O., Csiszar, I., Eidenshink, J.C., Myneni, R.B., Baret, F., 
Masuoka, E., Wolfe, R., & Devadiga, S. (2009b). A Terrestrial Surface 
Climate Data Record for Global Chage Studies, American Geophysical Union, 
Annual Fall Meeting, December 2009, San Francisco, California
Vermote, E.F., & Kotchenova, S. (2008). Atmospheric correction for the monitoring 
of land surfaces. Journal of Geophysical Research-Atmospheres, 113
Vermote, E.F., & Saleous, N.Z. (1997). Data Pre-Processing: Stratospheric Aerosol 
Perturbing Effect on the Remote Sensing of Vegetation: Correction Method 
for the Composite NDVI after the Pinatubo Eruption.  Remote Sensing 
Reviews, 15, 7-21.
Vermote, E.F., & Saleous, N.Z. (2006a). Calibration of NOAA16 AVHRR over a 
desert site using MODIS data. Remote Sensing of Environment, 105, 214-220
Vermote, E.F., & Saleous, N.Z. (2006b). Operational Atmospheric Correction of 
MODIS Visible to Middle Infrared Land Surface Data in the Case of an 
Infinite Lambertian Target. Book Chapter in ?Earth Science Satellite Remote
Sensing?, J. J.Qu, W. Gao. M. Kafatos, R.E. Murphy, V. V. Salomonson,(Eds), 2006., 
publisher Springer.
107
Vermote, E.F., Tanre, D., Deuze, J.L., Herman, M., & Morcrette, J.J. (1997). Second 
Simulation of the Satellite Signal in the Solar Spectrum, 6S: An overview. 
IEEE Transactions on Geoscience and Remote Sensing, 35, 675-686
Vicente-Serrano, S.M., & Heredia-Laclaustra, A. (2004). NAO influence on NDVI 
trends in the Iberian peninsula (1982-2000). International Journal of Remote 
Sensing, 25, 2871-2879
Young, S.S., & Harris, R. (2005). Changing patterns of global-scale vegetation 
photosynthesis, 1982-1999.  International Journal of Remote Sensing, 26, 
4537-4563
Yu, F.F., Price, K.P., Ellis, J., & Shi, P.J. (2003). Response of seasonal vegetation 
development to climatic variations in eastern central Asia. Remote Sensing of 
Environment, 87, 42-54
Zhou, L., Kaufmann, R.K., Tian, Y., Myneni, R.B., & Tucker, C.J. (2003). Relation 
between interannual variations in satellite measures of northern forest 
greenness and climate between 1982 and 1999.  Journal of Geophysical 
Research-Atmospheres, 108
Zhou, L.M., Tucker, C.J., Kaufmann, R.K., Slayback, D., Shabanov, N.V., & Myneni, 
R.B. (2001). Variations in northern vegetation activity inferred from satellite 
data of vegetation index during 1981 to 1999.  Journal of Geophysical 
Research-Atmospheres, 106, 20069-20083.
108