IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 44, NO. 7, JULY 2006 1787
Evaluation of the Consistency of Long-Term
NDVI Time Series Derived From AVHRR,
SPOT-Vegetation, SeaWiFS, MODIS,
and Landsat ETM+ Sensors
Molly E. Brown, Jorge E. Pinzón, Kamel Didan, Jeffrey T. Morisette, and Compton J. Tucker
Abstract—This paper evaluates the consistency of the Nor-
malized Difference Vegetation Index (NDVI) records derived
from Advanced Very High Resolution Radiometer (AVHRR),
SPOT-Vegetation, SeaWiFS, Moderate Resolution Imaging Spec-
troradiometer, and Landsat ETM+. We used independently
derived NDVI from atmospherically corrected ETM+ data at
13 Earth Observation System Land Validation core sites, eight
locations of drought, and globally aggregated one-degree data
from the four coarse resolution sensors to assess the NDVI records
agreement. The objectives of this paper are to: 1) compare the
absolute and relative differences of the vegetation signal across
these sensors from a user perspective, and, to a lesser degree, 2)
evaluate the possibility of merging the AVHRR historical data
record with that of the more modern sensors in order to pro-
vide historical perspective on current vegetation activities. The
statistical and correlation analyses demonstrate that due to the
similarity in their overall variance, it is not necessary to choose
between the longer time series of AVHRR and the higher quality
of the more modern sensors. The long-term AVHRR-NDVI record
provides a critical historical perspective on vegetation activities
necessary for global change research and, thus, should be the basis
of an intercalibrated, sensor-independent NDVI data record. This
paper suggests that continuity is achievable given the similarity
between these datasets.
Index Terms—Advanced Very High Resolution Radiometer
(AVHRR), Moderate Resolution Imaging Spectroradiometer
(MODIS), SPOT, vegetation.
I. INTRODUCTION
VARIOUS remote-sensing-based studies have revealedcompelling spectral relationships between the red and
near-infrared (NIR) part of the spectrum to green vegeta-
tion [1]. Due to vegetation pigment absorption (chlorophyll,
proto-chlorophyll), the reflected red energy decreases, while
the reflected NIR energy increases as a result of the strong scat-
tering processes of healthy leaves within the canopy. Directly
using the amount of reflected red and/or NIR radiation to study
the biophysical characteristics of vegetation is very inadequate,
Manuscript received November 1, 2004; revised August 12, 2005. This work
was supported in part by USAID’s Famine Early Warning System Network.
M. E. Brown and J. E. Pinzón are with the SSAI/NASA Goddard Space Flight
Center, Greenbelt, MD 20771 USA (e-mail: molly.brown@gsfc.nasa.gov).
J. T. Morisette and C. J. Tucker are with the NASA Goddard Space Flight
Center, Greenbelt, MD 20771 USA.
K. Didan is with the TBRS Laboratory, Soil, Water and Environmental Sci-
ences, University of Arizona, Tucson, AZ 85721 USA.
Digital Object Identifier 10.1109/TGRS.2005.860205
for reasons rooted in the intricate radiative energy interaction
at the canopy level, background, and atmospheric impacts on
the signal and the nonuniqueness of the signatures. When,
however, two or more bands are combined into vegetation
index (VI), the vegetation signal is boosted and the information
become more useful [2]. Vegetation indices can then be used
as surrogate measures of vegetation activity [3], [4]. The most
widely used form of VI, the Normalized Difference Vegetation
Index (NDVI), was introduced by Deering in 1978 [5] and
Tucker in 1979 [3] and is the ratio of the difference of the NIR
and red band divided by their sum. The NDVIs properties help
mitigate a large part of the variations that result from the overall
remote-sensing system (radiometric, spectral, calibration,
noise, viewing geometry, and changing atmospheric condi-
tions). Some land-surface types are not robustly represented
by NDVI, such as snow, ice, and nonvegetated surfaces, where
atmospheric variations and sensor characteristics dominate [6].
NDVI is often used as a monitoring tool for the vegetation
health and dynamics, enabling easy temporal and spatial com-
parisons [7]. In order to make effective use of NDVI data, issues
related to the remote-sensing system need to be addressed. The
most serious are clouds, which render any observation useless
by obstructing the target, and, to a lesser degree, the effects of
the Bidirectional Reflectance Distribution Function (BRDF). To
overcome these issues, maximum value compositing was devel-
oped as an operational approach to producing cloud free consis-
tent NDVI maps. Multiple daily images are processed to create
a representative, cloud-free image with the least atmospheric
attenuation and viewing geometry effects [8]. The maximum
value compositing (MVC) technique is the most widely used
method and is based on maximizing the NDVI signal over a
preset period of time. While MVC helps screen for clouds, it
was found to also favor extreme viewing geometry (large solar
zenith angles and large view angles in the forward scatter direc-
tion) [9] and, to a lesser extent, cloud shadow. Several studies
attempted to address these issues with modest and mixed results
[10]–[12].
On a global basis, several factors can influence differences
in NDVI across sensors. Impacts from BRDF are well docu-
mented [10], [13]. Additionally, the Spectral Response Func-
tions (SRFs) for the different sensors can lead to systematic
differences in NDVI [14]. Each sensor has its own instanta-
neous field of view, swath width, and orbiting geometry. Ad-
0196-2892/$20.00 © 2006 IEEE