04-148
8/17/06
1:28 PM
Page 1027
A 16-year Time Series of 1 km AVHRR
Satellite Data of the Conterminous
United States and Alaska
Jeff Eidenshink
Abstract
The U.S. Geological Survey (USGS) has developed a 16-year
time series of vegetation condition information for the conterminous United States and Alaska using 1 km Advanced
Very High Resolution Radiometer (AVHRR) data. The AVHRR
data have been processed using consistent methods that
account for radiometric variability due to calibration uncertainty, the effects of the atmosphere on surface radiometric
measurements obtained from wide field-of-view observations,
and the geometric registration accuracy. The conterminous
United States and Alaska data sets have an atmospheric
correction for water vapor, ozone, and Rayleigh scattering
and include a cloud mask derived using the Clouds from
AVHRR (CLAVR) algorithm. In comparison with other AVHRR
time series data sets, the conterminous United States and
Alaska data are processed using similar techniques. The
primary difference is that the conterminous United States
and Alaska data are at 1 km resolution, while others are
at 8 km resolution. The time series consists of weekly and
biweekly maximum normalized difference vegetation index
(NDVI) composites.
Introduction
Since 1989, the U.S. Geological Survey (USGS) National
Center for Earth Resources Observation and Science (EROS)
has produced a 1 km Advanced Very High Resolution
Radiometer (AVHRR) satellite data set of the conterminous
United States and Alaska (Eidenshink, 1992). The data sets
are comprised of a time series of weekly and bi-weekly
vegetation condition or greenness composites based on the
maximum normalized difference vegetation index (NDVI).
The data sets measure the annual and interannual patterns
of vegetation growth and condition over landscapes and
ecosystems in the United States and Alaska. The greenness
data are used to monitor the effects of drought; quantify
agricultural production; monitor the health of forests, shrublands, and grasslands; forecsast fire danger; map land-cover
change; characterize the effects of climate change; and
provide the basis for many other applications.
Two characteristics make the conterminous United
States and Alaska data set valuable: the length of the time
series and the quality of the processing methods. Of these,
the quality of the data processing is the most important. The
quality of the information (i.e., NDVI) derived from the AVHRR
data is determined by the use of consistent, scientificallyvalidated processing methods that account for radiometric
variability due to calibration uncertainty, the effects of the
atmosphere on surface radiometric measurements obtained
from wide field-of-view observations, and the geometric
registration accuracy.
Once quality measurements are achieved, then the
length of the time-series measurements becomes important.
The time-series measurements detect subtle changes in
annual and interannual greenness conditions that can be
used to characterize and quantify biophysical characteristics
of vegetation. The changes in greenness relate to changes in
productivity that can be linked to climate and anthropogenic
effects (Myeni et al., 1997).
There are other AVHRR time-series data sets available for
the United States and other parts of the world. Two of the
more prominent AVHRR time series are the Pathfinder AVHRR
Land (PAL) and the Global Inventory Mapping and Monitoring System (GIMMS) data sets. The PAL data set, produced as
part of the National Oceanic and Atmospheric Administration (NOAA)/NASA Pathfinder AVHRR Land program, contains
global and continental monthly and 10-day composites of
channels 1, 2, 4, and 5 and the NDVI. The PAL data set begins
with 4 km global area coverage (GAC) data, processed to
an 8 km resolution (ftp://disc1.gsfc.nasa.gov/data/avhrr/
Readme.pal). The data are derived from the AVHRR on the
“afternoon” NOAA operational meteorological satellites
(NOAA-7, -9, -11, -14) and cover the time period from 1981 to
2001. The Pathfinder program, in conjunction with the USGS,
also produced a time series of global land 1 km AVHRR 10-day
composites for 1992 to 1996 (Eidenshink and Faundeen,
1994). At that time, the global land 1 km AVHRR data were
processed using techniques similar to those used for the
conterminous United States and Alaska data sets.
The GIMMS data set is available from July 1981 through
present. The GIMMS data set is produced from 4 km GAC
data and is processed to an 8 km resolution. The GIMMS
normalized difference vegetation index has been corrected
for residual sensor degradation and sensor intercalibration
differences, distortions caused by persistent cloud cover
globally, solar zenith angle and viewing angle effects due
to satellite drift, and volcanic aerosols. This data set is
improved over the 8 km PAL based on corrections for calibration, view geometry, volcanic aerosols, and other effects
not related to actual vegetation change, in particular, NOAA-9
descending node data from September 1994 to January 1995,
volcanic stratospheric aerosol correction for 1982 to 1984
Photogrammetric Engineering & Remote Sensing
Vol. 72, No. 9, September 2006, pp. 1027–1035.
U.S. Geological Survey, National Center for Earth
Observation and Science, Sioux Falls, SD 57198
([email protected]).
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
0099-1112/06/7209–1027/$3.00/0
© 2006 American Society for Photogrammetry
and Remote Sensing
S e p t e m b e r 2 0 0 6 1027
04-148
8/17/06
1:28 PM
Page 1028
and 1991 to 1994, and improved NDVI using empirical mode
decomposition/reconstruction (EMD) to minimize the effects
of orbital drift (Pinzon et al., 2004).
One major factor that differentiates the conterminous
United States and Alaska data sets is that they are produced
from 1 km local area coverage (LAC) data and processed to
1 km resolution. The PAL and GIMMS data sets are produced
with 4 km GAC data and processed to 8 km resolution. It is
important to note the difference between LAC and GAC data.
AVHRR data are acquired onboard the satellite at approximately 1 km resolution at nadir. The 1 km data is sampled
onboard the satellite to approximately 4 km to produce the
GAC data. The sampling process is a fundamental difference
between LAC and GAC data that affects the spectral measurements and the respective NDVI data. The difference is that
GAC data set is a sampled average of an array of 1 km pixel
values. The sampling procedure used to produce GAC data is
described by NOAA (Kidwell, 1998 and 2000). The implication of the sampling technique is that the radiometric value
of the GAC pixel can be a combination of clear pixels, cloud
covered pixels, or pixels with sub-pixel cloud contamination.
The mixing clear and cloud contaminated measurements in
a GAC pixel affects the radiometric measurement such that it
is very difficult to discriminate cloud contamination in GAC
data and radiometric measurements of the surface can be a
mixture of surface characteristics and clouds. The LAC data
does not use a sample of measurements. So the only mixing
problem is due to sub-pixel clouds. The cloud mask process
for the 1 km conterminous United States and Alaska data
is described later. Beyond the compositing technique, the
GIMMS data processing uses a temperature threshold to identify
clouds. The PAL processing used the Clouds from AVHRR
(CLAVR) detection algorithm that is based on a series of
temperature and brightness thresholds (Stowe et al., 1991).
AVHRR Processing Methods
Over the 16-year period, the USGS has processed AVHRR data
acquired by four different NOAA polar orbiting environmental
satellites. The AVHRR data are acquired over the conterminous United States from the direct reception system at EROS.
The data coverage of Alaska is obtained from NOAA. The USGS
also has developed the AVHRR data acquisition and processing system (ADAPS) to perform radiometric calibration, atmospheric correction, geometric registration, compositing, and
cloud screening of AVHRR data (http://eros.usgs.gov/products/
landcover/avhrrreadme.txt).
Radiometric Calibration
Radiometric calibration is fundamental to understanding the
quantitative measurements of a sensor. Data from NOAA-11,
-14, -16, and -17 AVHRR sensors have been used to produce
the conterminous United States and Alaska data sets. The
source of the calibration coefficients for each sensor varies
(Table 1), but the basic method used to develop and apply
the coefficients is very similar. The calibration coefficients
and methodology for channel 3a and the thermal channels
3b, 4, and 5 are well documented by NOAA (Kidwell, 1998
and 2000). It is widely recognized that the calibration of
channels 1 and 2, the visible and near-infrared bands, must
include an accounting for sensor degradation (Rao et al.,
1993; Kaufman and Holben, 1993; Brest and Rossow, 1992).
Teillet and Holben (1994) provide a comprehensive
evaluation of the calibration coefficients for NOAA-11 AVHRR
channels 1 and 2, and their report provides the basis for the
derivation of time variant calibration coefficients for NOAA-7,
-9, and -11. They derive the coefficients from the results
obtained from the desert validation approach of Kaufman
and Holben (1993). Teillet and Holben (1994) developed
1028 S e p t e m b e r 2 0 0 6
TABLE 1. A SUMMARY OF THE CALIBRATION SOURCES AND THE VALID
TIMEFRAMES USED TO PRODUCE THE CONTERMINOUS UNITED
STATES AND ALASKA DATA SETS
Satellite
Start Date
End Date
Source
NOAA-11
NOAA-11
09/26/1988
03/27/1989
03/26/1989
present
NOAA-14
NOAA-14
12/30/1994
06/31/1995
06/30/1995
present
NOAA-16
NOAA-16
NOAA-17
09/01/2000
06/25/2003
01/01/2004
06/24/2003
present
present
pre-launch
Teillet and
Holben (1994)
pre-launch
Vermote and
Kaufman (1995)
pre-launch
NOAA
NOAA
the time-dependent coefficients based on two methods: a
polynomial fit and a piecewise linear (PWL) fit of the observations over desert areas. They recommended a PWL fit to
describe the trend of the gain or offset (y) with time (x).
This function consists of a series of straight-line segments,
joined to each other at points where values can be reliably
estimated or at times designed to coincide with the start
of the northern hemisphere growing season. The PWL is
suitable for operational use because, unlike polynomial fits,
the PWL will not change retroactively when new data are
added to the end of the time series. This approach assumes
that the degradation trend is linear between measurements
of sensor performance of AVHRR channels 1 and 2. The
equations for time-dependent radiometric calibration to
radiance and reflectance for the visible and near-infrared
channels of NOAA-7, -9, and -11 are described by Teillet
and Holben (1994).
NOAA-14 was launched into an afternoon ascending
orbit in December 1994. Shortly after launch, Vermote and
Kaufman (1995) described the post-launch calibration of
AVHRR channels 1 and 2. Their coefficients were derived
from ocean and cloud observations. Their work provided
the first post-launch calibration coefficients for NOAA-14,
which were used by the USGS in the operational processing
of AVHRR NOAA-14 data.
NOAA-16 became the operational system in January 2001.
Initially, pre-launch coefficients were used for calibration. In
June 2003, NOAA began providing monthly updates of the
calibration coefficients. The updates were based on analysis
of NOAA-16 data using the desert calibration approach. The
post-launch calibration coefficients for NOAA-16 can be found
at http://noaasis.noaa.gov/NOAASIS/ml/calibration.html.
In December 2003, the AVHRR sensor on NOAA-16 developed a problem with the scan motor that rendered the data
useless for greenness mapping. In January 2004, the USGS
began to use NOAA-17 data for greenness mapping. Post-launch
calibration coefficients provided by NOAA are being used for
the calibration of NOAA-17 data (http://noaasis.noaa.gov/
NOAASIS/ml/calibration.html).
Atmospheric Correction
The most substantial improvement to the conterminous United
States and Alaska AVHRR data sets has been the application of
an atmospheric correction for ozone, water vapor absorption,
and Rayleigh scattering. In 2001, the entire existing time
series was reprocessed to include the atmospheric correction,
which has been applied routinely since 2001.
Water vapor absorption affects measurements in the
near-infrared band (channel 2) by reducing the reflectance
by 10 to 30 percent, depending on the viewing geometry
(Tanre et al., 1992). Rayleigh scattering and ozone absorption affect measurements in the red band (channel 1) by
increasing the reflectance by 1 to 2 percent, depending on
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
04-148
8/17/06
1:28 PM
Page 1029
the viewing geometry. The combination of these effects
results in significant error in the computation of NDVI and
other vegetation indices. Figure 1 shows the effect of
atmospheric correction on reflectance values for channels 1
and 2 and the resultant NDVI for a sample of data from the
11–24 May 2005, biweekly composite of the conterminous
United States.
The second simulation of the satellite signal in the solar
spectrum (6S) radiative transfer model is used to quantify
the difference between the radiance measured at the satellite and Earth-leaving radiance (Vermote et al., 1997). The
length of the path and direction that the signal travels are
very important. The viewing geometry factors that affect the
path length and direction are the orientation of the sun and
Figure 1. The effect of atmospheric correction on reflectance values for channels 1 and 2 and the
resultant NDVI for a sample of data from the 11–24 May 2005 biweekly composite of the conterminous United States: (a) AVHRR channel 1 top of atmosphere reflectance and surface reflectance
after atmospheric correction, (b) AVHRR channel 2 top of atmosphere reflectance and surface
reflectance after atmospheric correction, and (c) AVHRR NDVI from top of atmospheric reflectance
and surface reflectance after atmospheric correction.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
S e p t e m b e r 2 0 0 6 1029
04-148
8/17/06
1:28 PM
Page 1030
TABLE 2.
THE QUANTIZATION OF THE PARAMETERS
FOR THE 6S RADIATIVE TRANSFER MODEL
Parameter
Solar Zenith ()
Satellite Zenith (Z)
Relative Azimuth (rel)
Terrain Elevation (T)
Quantization Spacing
0, 15, 30, 40, 50, 58, 64, 69,
73, 76, 78, 80 (degrees)
0, 10, 20, 30, 40, 46, 51,
56, 60, 64, 67, 70 (degrees)
0, 11, 23, 37, 53, 67, 79, 90, 100,
110, 120, 130, 140, 150, 160,
168, 174, 180 (degrees)
0, 0.5, 1.1, 1.8, 2.6 (km)
the sensor relative to the target. The viewing geometry is
provided as part of the conterminous United States and
Alaska data sets. Bandpass calculations of 0.005 micrometer spacing or better are recommended for the simulation
(Teillet, 1992). The 6S radiative transfer model includes the
use of parameterized lookup tables (LUTs) and interpolation.
LUTs are generated from the radiative transfer code using
four key parameters: solar zenith, satellite zenith, relative
azimuth, and terrain elevation (Table 2).
The corrections for ozone absorption and Rayleigh
scattering are straightforward (Teillet, 1991). Appropriate
Rayleigh scattering correction must include an adjustment
for atmospheric pressure that can be derived from the
elevation of the target. Recommended reference values for
Rayleigh optical depths for standard pressure and temperature conditions are available (Teillet, 1990; El Saleous
et al., 1994). The local elevation adjustment can be derived
using digital terrain data. The correction for ozone absorption is based on concentration values from actual measurements derived from the Total Ozone Mapping Spectrometer
(TOMS) or other appropriate sensors (El Saleous et al.,
1994). The ozone climatology described by London et al.
(1976) is used for the conterminous United States and
Alaska data sets.
Atmospheric water vapor concentrations are available
from at least two sources. The NASA Goddard Space Flight
Center distributes the TIROS Operational Vertical Sounder
(TOVS) level 3 geophysical parameters derived using the
physical retrieval algorithm designated as the so-called Path
A and Path B schemes (Suskind et al., 1997). The archive
data products consist of 1-degree by 1-degree global fields
of the three-dimensional, temperature-moisture structure
of the atmosphere. The TOVS data set was used to process
the conterminous United States and Alaska historical data
through May 2000. The second source that has been used
operationally since May 2000 is the NOAA National Centers
for Environmental Prediction (NCEP) global one degree FNL
precipitable water vapor data set.
DeFelice et al. (2003) describe the correction for water
vapor. They investigated the effects of the correction method
using the 6S radiative transfer code at 12 randomly chosen
points within two conterminous United States images from
13 June 1996 and 28 November 1996. First, they took nonatmospherically corrected at-satellite-radiance values and
necessary ancillary data and used them in 6S to retrieve
atmospherically corrected reflectance values. Then, they took
the top of the atmosphere reflectance values and used them
as input to 6S. They found that the correction method and the
6S atmospherically corrected reflectance values were within
0.002 and 0.003 for channel 1 and channel 2 respectively
on 13 June 1996, and similarly within 0.003 and 0.002
for channel 1 and channel 2 on 28 November 1996.
The conterminous United States and Alaska data sets
have no correction for stratospheric aerosol, such as the
correction applied to the GIMMS data set (Vermote et al.,
1030 S e p t e m b e r 2 0 0 6
1997). The only major event that produced stratospheric
aerosol during the 16-year period of the data set is the
eruption of Mount Pinatubo in 1991. The stratospheric
aerosol from Mount Pinatubo had the most significant effect
in the equatorial region of the world. The impact in the
northern hemisphere was measurable.
Geometric Registration
The compositing process requires that each daily overpass be
precisely registered to a common map projection to ensure
that each daily 1 km pixel is referenced to the correct ground
location. Past experiments with image registration have
shown that image-to-image registration provides the precision
needed for temporal data sets, and the use of digital image
correlation techniques produced consistent image-to-image
registration results. Most methods for control point matching
use automatic correlation of image segments with ground
control points, and then integrate the adjustments derived
from the correlation with the orbital model (Brush, 1988;
Cracknell and Paithoonwattanakij, 1989; Kloster, 1989; Kelly
and Hood, 1991; Brunel and Marsouin, 1987). An evaluation
of AVHRR image-to-image registration using automated correlation techniques showed an improvement in geometric
accuracy (root mean square error of less than 1.0 pixel)
compared to traditional image-to-map procedures (Kelly and
Hood, 1991).
The geometric registration of an AVHRR observation
is a multi-step process. First, 150 pixels on either side of
a scan are discarded; this eliminates data from viewing
angles larger than 42 degrees off-nadir and reduces geometric distortion as a result of viewing angle. Next, a systematic
correction is applied to the calibrated channel 2 (nearinfrared) data. The near-infrared channel is used because
water bodies that are commonly used as control points
have the most contrast to land in the near-infrared channel.
The systematic correction is based on a satellite platform
model that yields a geometric registration error of approximately three pixels (approximately 3 km). The final step is
to develop the precision correction using several hundred
ground control points (GCPs) selected from the systematically
corrected image and a corresponding precision registered
base image. The precision registered base image is a mosaic
of cloud free AVHRR observations. Each GCP is centered in a
chip that is an image segment 32 lines by 32 samples. The
locations of the GCP chips from the systematic image are
correlated with the chips from the base image using grayscale correlation. A polynomial transformation is derived
based on a least-square fit of a set of GCPs and corresponding
image (pixel) locations.
An observation used in a composite must have a root
mean square error less than 1 pixel (1,000 m). The polynomial transformation derived from the correlation of channel
2 is then used to georeference all of the remaining bands
associated with an observation. The map projection used for
the conterminous United States data set is Lambert
Azimuthal Equal Area. The map projection used for the
Alaska data set is Albers Equal Area.
Compositing
Compositing is a technique used to merge multiple daily
observations, acquired over a specific period, into a single
image. Key factors to consider are the compositing method
and the length of the compositing period. Compositing
periods of 7, 10, and 14 days have been used most commonly. The choice of the period is usually based on the
length of time necessary to obtain a composite with minimal
cloud contamination or the amount of time necessary to
observe meaningful changes in surface characteristics. Compositing techniques for AVHRR have been investigated to
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
04-148
8/17/06
1:28 PM
Page 1031
identify a technique that provides the best multichannel,
multitemporal AVHRR data sets (Cihlar et al., 1997). The
general objective is to obtain the maximum NDVI observed
at the view angle closest to nadir, while preserving the
integrity of the individual reflectance channels.
We produced weekly and biweekly composites using
the maximum value compositing (MVC) method (Holben,
1986). The MVC approach requires a series of georegistered
daily observations for a selected time period or compositing
period. The NDVI is examined pixel by pixel for each observation during the compositing period to determine the maximum value. On the average, 10 daily passes per week can
be georegistered and used in a conterminous United States
composite. Selection of the maximum NDVI is assumed to
represent the maximum vegetation “greenness,” a measure
of photosynthetic activity and serves to reduce the number
of cloud-contaminated pixels. The only other criterion used
in the compositing is the exclusion of pixels with a solar
zenith angle greater than 80 degrees. Solar zenith angles
greater than 80 degrees occur in northern latitudes in winter,
especially for periods of major orbital drift.
The output of the compositing process is a 14-band data
set with one band containing the maximum NDVI value for
each pixel selected from the daily overpasses. The remaining
13 bands are the data values that are coincident with the
observation value selected as the maximum NDVI value. They
include calibrated channels 1 to 5, satellite viewing geometry data (three bands), atmospherically corrected channels
1 and 2 reflectance, a quality-control band indicating the
origin of the water vapor and ozone values, a band that
contains a pointer to the scene identification number for
each pixel selected from the same daily pass as the maximum NDVI value, and the cloud mask.
Cloud Screening
Although the maximum NDVI compositing process tends to
reduce cloud contamination in a composite, cloud contamination is often present in the weekly and biweekly composites. Several factors contribute to the cloud contamination.
Often persistent seasonal weather and cloud patterns, such
as the monsoon season in the southwest or early spring
conditions in the northern portion of the conterminous
United States, limit the number of cloud-free observations.
The clouds are also more prevalent in weekly rather than
biweekly composites simply because of the number of
observations available. Snow and clouds in Alaska are
particular problems. As was mentioned in the section on
geometric registration, it is not possible to accurately geometrically register an image if too many clouds are present.
Clouds have low NDVI values because the spectral
reflectance of clouds in channel 1 and channel 2 is nearly
equal. However, sub-pixel clouds produce an NDVI that is a
blend of clouds and surface reflectance. Cloud contamination can confound the measurement of vegetation condition.
The worst-case example is sub-pixel clouds over a very
green surface condition where the resulting NDVI is reduced.
A dramatic reduction may be recognizable as cloud contamination. But a subtle reduction appears as vegetation under
stress, such as by drought.
In the past, we provided, upon request, cloud screened
composites of the conterminous United States and Alaska.
The cloud screen was developed using a brightness threshold of channel 1 reflectance. High values were classified as
clouds. Thick, widespread, overcast clouds are relatively
easy to detect. But sub-pixel clouds are much more difficult.
The occurrence of sub-pixel clouds over green surfaces that
often happens during the summer, or the occurrence of
clouds over snow-covered surfaces in the winter, is difficult to detect. There are also problems with this method
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
especially over bright soil surfaces as in the Great Basin
region in the western United States because it is not possible
to discriminate a cloud from bright soils by simply using the
channel 1 data. Therefore, sub-pixel cloud contamination can
be present in the composites even after cloud screening.
In 2003, we began using an adaptation of the clouds
from AVHRR data-Phase 1(CLAVR-1) algorithm developed by
NOAA (Stowe et al., 1999). CLAVR-1 uses a series of tests of
reflectance, temperature, land-cover, and geographic location
to identify clouds. A decision tree process is used to
identify and validate clouds based on brightness and
temperature thresholds. Other masks, such as barren land
and water, are also used. The decision tree process also
includes a means to restore a pixel to “clear” when criteria
are met. Our adaptation of the algorithm does not use the
spatial neighborhood tests because they did not function on
composites where adjacent pixels could be selected from
different observations.
Our experience indicates that the CLAVR-1 adaptation is
superior to the simple threshold approach. Figure 2a shows
a conterminous United States weekly composite for the
period ending 12 August 2003 with clouds (shown in white)
identified using a brightness threshold method. Figure 2b is
the same composite with clouds identified using the adaptation of the CLAVR-1 algorithm. The regions with the most
difference are the desert southwest and the southeast. Figure 3
shows a comparison of the threshold technique and the
CLAVR-1 algorithm for representative regions of the southwest, and Figure 4 is the southeast United States. The
images of the Great Basin region of the western United
States (Figure 3) show clouds detected using CLAVR-1 (Figure
3a) and a brightness threshold method (Figure 3b) and
clouds detected. The brightness threshold method (Figure
3b) classifies bright soil surfaces as cloud, whereas the
CLAVR-1 method does not. The images near the Texas/
Louisiana border on the Gulf of Mexico (Figure 4) show
clouds detected using a brightness threshold method (Figure
4a) and clouds detected using CLAVR (Figure 4b). The CLAVR-1
method detected many more clouds, including a region of
clouds in the eastern portion of the area.
Band 14 of a composite includes the cloud mask
derived from CLAVR-1. The cloud mask has values from 0 to
200. The individual values are keyed according to which
test was ultimately used to determine if the pixel is clear or
cloudy. Values less than 100 are clear sky; values 100 or
greater are clouds.
Data Scaling Characteristics
The products from the conterminous United States and
Alaska data sets are scaled to byte (8-bit) data. Reflectance
values for channels 1, 2, and 3A are converted to byte
data, where the range 0 to 254 represents 0 to 63.5 percent
reflectance. The value 255 corresponds to reflectance greater
than 63.5 percent. Any feature with greater than 63 percent
reflectance is considered to be bright and non-vegetative.
Energy is converted to brightness temperature using
the inverse of Planck’s radiation function. The brightness
temperatures are represented in Kelvin units. A scaling
factor was used to convert the channel 3B, 4, and 5 brightness temperatures to byte data. A scaling factor of 202.5
is subtracted from the brightness temperature value, and
the difference is multiplied by 2 to maintain one half
percent accuracy (i.e., a brightness temperature of 280
becomes 155).
A separate image band is created for the satellite zenith,
solar zenith, and relative azimuth angle for each image
pixel. The solar zenith is computed in degrees, where
90 degrees represents the horizon or terminator, the dividing line between the illuminated and the un-illuminated
S e p t e m b e r 2 0 0 6 1031
04-148
8/17/06
1:28 PM
Page 1032
Figure 2. (a) Conterminous United States weekly composite for the period ending 12 August 2003. The
clouds (shown in white) were identified using a brightness threshold method, and (b) Conterminous
United States weekly composite for the period ending 12 August 2003. The clouds (shown in white)
were identified using an adaptation of the clouds from AVHRR data—Phase 1 (CLAVR-1) algorithm developed by NOAA (Stowe et al., 1999).
1032 S e p t e m b e r 2 0 0 6
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
04-148
8/17/06
1:28 PM
Page 1033
Figure 3. These images of the Great Basin region of the western United States show clouds detected
using CLAVR (a) and brightness threshold method (b). The brightness threshold method (b) classifies
bright soil surfaces as cloud, whereas the CLAVR method does not.
(a)
(b)
Figure 4. These images near the Texas/Louisiana border on the Gulf of Mexico show clouds (in white)
detected using a brightness threshold method (a) and clouds detected using CLAVR-1 (b). The CLAVR-1
method detected many more clouds, including a region of clouds in the eastern portion of the area.
part of the Earth’s surface. The satellite zenith angle is
computed in degrees, where 90 degrees represents nadir.
Therefore, values less than 90 degrees represent view
angles in the back scattered (easterly) direction, and values
greater than 90 represent the forward scatter (westerly)
direction. Note that the effective field of view of the satellite is approximately 55 degrees each side of nadir, but
computed satellite zenith angles can exceed 55 degrees
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
because of the curvature of the Earth. The relative azimuth
angle is computed as the absolute difference between
the solar azimuth and the satellite azimuth angles. The
computed values are in the 0 to 180 range. The relative
azimuth angle is computed instead of separate azimuth
angles because only the absolute difference between the
azimuth angles is required for atmospheric correction
algorithms.
S e p t e m b e r 2 0 0 6 1033
04-148
8/17/06
1:28 PM
Page 1034
Linking the AVHRR Time Series to the Future
Long-term continuity of the moderate resolution satellite
data record is critical for the study of landscape change
and the extension of existing monitoring applications into
the future. The USGS has created contemporaneous data
sets of AVHRR and MODIS vegetation indices from 2001
through 2003. These data sets are being used to characterize and resolve differences in the instrument signals and
derived indices to ensure the continuity of a long-term
record of vegetation state and condition. The comparison
data sets consist of 16-day NDVI at 1 km resolution data
covering the conterminous United States. Early results
produced by Gallo et al. (2004) indicate strong agreement
(r 2 0.92) between the AVHRR and MODIS NDVI over the
data set as a whole, although agreement may be weaker
for specific land-cover types, such as southeastern evergreen needle-leaf forest (r 2 0.65). It is a goal of the USGS
to develop the linkage of AVHRR and MODIS to ensure the
continuity of a long-term record of vegetation state and
condition.
Summary
The USGS has produced a time series of weekly and biweekly
maximum NDVI composites of the conterminous United
States and Alaska using AVHRR satellite data. The USGS
AVHRR data set is a 16-year time series of calibrated, georegistered, and atmospherically corrected weekly maximum
NDVI composites for January 1989 to the present. A consistent set of documented standards has been used to process
the data. The data are designed to be flexible enough for
use in both basic research and operational vegetation monitoring programs. The quality of the processing will enable
the linkage of the AVHRR time series to future observations
from moderate resolution satellite data.
The AVHRR time-series products are available for the
conterminous United States and Alaska. Information on
data availability and other general information can be
found at: http://eros.usgs.gov/greenness/index.html. For
further information contact: Customer Services, U.S.
Geological Survey, EROS, 47914 252nd Street, Sioux Falls,
SD 57198-0001; Telephone 800-252-4547 or 605-594-6151;
TDD: 605-594-6933([email protected]).
References
Brest, C.L., and W.B. Rossow, 1992. Radiometric calibration and
monitoring of NOAA AVHRR data for ISCCP, International
Journal of Remote Sensing, 13(2):235–273.
Brown, J.F., B.C. Reed, M.J. Hayes, D.A. Wilhite, and K. Hubbard,
2002. A prototype drought monitoring system integrating
climate and satellite data, Proceedings of Pecora 15 Symposium,
November, Denver, Colorado, unpaginated CD-ROM.
Brown, J.F., and T. Tadesse, 2003. Integrating growing season
satellite metrics with climate data to map and monitor
drought, Proceedings of the 30th International Symposium on
Remote Sensing of Environment, 10–14 November, Honolulu,
Hawaii.
Brunel, P., and A. Marsouin, 1987. An operational method using
ARGOS orbital elements for navigation of AVHRR imagery,
International Journal of Remote Sensing, 8:569–578.
Brush, R.J.H., 1988. The navigation of AVHRR imagery, International Journal of Remote Sensing, 14:629–634.
Burgan, R.E., R.W. Klaver, and J.M. Klaver, 1998. Fuel models and
fire potential from satellite and surface observations, International Journal of Wildland Fire, 8(3):159–170.
Cihlar, J., H. Ly, Z. Li, J. Chen, H. Pokrant, and F. Huang, 1997.
Multitemporal, multichannel AVHRR data sets for land biosphere studies: artifacts and corrections, Remote Sensing of
Environment, 60:35–57.
1034 S e p t e m b e r 2 0 0 6
Cracknell, A.P., and K. Paithoonwattanakij, 1989. Pixel and subpixel accuracy in geometric correction of AVHRR imagery,
International Journal of Remote Sensing, 10:661–667.
DeFelice, T.P., P. Piriano, D. Lloyd, J. Meyer, and T.T. Baltzer, 2003.
Water vapour correction of the daily 1 km AVHRR global land
dataset: Part 1—Validation and use of the water vapour input
field, International Journal of Remote Sensing, 24(11):2365–2375.
Eidenshink, J.C., 1992. The 1990 conterminous United States AVHRR
data set, Photogrammetric Engineering & Remote Sensing, 58(6):
809–813.
El Saleous, N., E. Vermote, and J.-C. Roger, 1994. Operational
atmospheric correction of AVHRR visible and near infrared
data, Proceedings of IGARRS ‘94: International Geoscience
and Remote Sensing Symposium, 8–12 August, California
Institute of Technology, Pasadena, California and the Institute
of Electrical and Electronics Engineers, Piscataway, New Jersey.
Gallo, K., J. Lei, B. Reed, J. Dwyer, and J. Eidenshink, 2004. Comparison of MODIS and AVHRR 16-day normalized difference
vegetation index composite data, Geophysical Research Letters,
31(7):L07502.
Holben, B.N., 1986. Characteristics of maximum-value composite
images from temporal AVHRR data, International Journal of
Remote Sensing, 7:1417–1434.
Kaufman, Y.J., and B.N. Holben, 1993. Calibration of the AVHRR
visible and near-IR bands by atmospheric scattering, ocean
glint, and desert reflection, International Journal of Remote
Sensing, 14(1):21–52.
Kelly, G.G., and J.J. Hood, 1991. AVHRR conterminous United States
reference data set, Technical Papers, Volume 3, 1991 ACSMASPRS Annual Convention, Baltimore, Maryland, American
Society of Photogrammetry and Remote Sensing, Bethesda,
Maryland, pp. 232–239.
Kidwell, K.B., 1998. NOAA Polar Orbiter Data User’s Guide, National
Oceanic and Atmospheric Administration, Washington, D.C.
Kidwell, K.B., 2000. NOAA KLM User’s Guide, National Oceanic
and Atmospheric Administration, Washington, D.C.
Kloster, K., 1989. Using TBUS orbital elements for AVHRR image
gridding, International Journal of Remote Sensing, 10:653–659.
London, J., R.D. Bojkov, S. Oltmans, and J.L. Kelley, 1976. Atlas
of the global distribution of total ozone, July 1957–June 1967,
NCAR Technical Note NCAR/TN/133STR, NCAR, pp. 276.
Loveland, T.R., J.M. Merchant, D.O. Ohlen, and J.F. Brown, 1991.
Development of a land-cover characteristics database for the
conterminous United States, Photogrammetric Engineering &
Remote Sensing, 57:1453–1463.
Myeni, R.B., C.D. Keeling, C.J. Tucker, G. Asrar, and R.R. Nemani,
1997. Increased plant growth in the northern high latitudes
from 1981 to 1991, Nature, 386:698–701.
Pinzon, J.E., M.E. Brown, and C.J. Tucker, 2004. Satellite time
series correction of orbital drift artifacts using empirical
mode decomposition, EMD and Its Applications, Chapter 10,
Part II (N.E. Huang and S.S.P. Shen, editors), World Scientific
Publishers, Singapore.
Rao, C.R.N., J. Chen, F.W. Staylor, P. Abel, Y.J. Kaufman, E. Vermote,
W.R. Rossow, and C. Brest, 1993. Degradation of the visible and
near-infrared channels of the advanced very high resolution
radioimeter on the NOAA-9 spacecraft: Assessment and recommendations for corrections, NOAA Technical Report NESDIS 70,
Department of Commerce, Washington, D.C.
Reed, B.C., and L. Yang, 1997. Seasonal vegetation characteristics
of the United States, Geocarto International, 12(2):65–71.
Stowe, L.L., E.P. McClain, R. Carey, P. Pellegrino, G. Gutman,
P. Davis, C. Long, and S. Hart, 1991. Global distribution of
cloud cover derived from NOAA/AVHRR operational satellite
data, Advance Space Research, 11:51–54.
Stowe, L.L., P.A. Davis, and P.E. McClain, 1999. Scientific basis and
initial evaluation of the CLAVR-1 global clear/cloud classification algorithm for the advanced very resolution radiometer,
Journal of Atmospheric and Ocean Technology, 16: 656–681.
Suskind, J., P. Piraino, L. Rokke, L. Irdell, and A. Mehta, 1997.
Characteristics of the TOVS Pathfinder Path A dataset, Bulletin
of the American Meteorological Society, 78:1449–1472.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
04-148
8/17/06
1:28 PM
Page 1035
Tanre, D., B.N. Holben, and Y.J. Kaufman, 1992. Atmospheric
correction algorithm for NOAA-AVHRR products: theory and
application, IEEE Transactions of Geoscience and Remote
Sensing, 30:231–248.
Teillet, P.M., 1990. Rayleigh optical depth comparisons from
various sources, Applied Optics, 29(13):1897–1900.
Teillet P.M., 1991. Radiometric and atmospheric correction procedures for AVHRR preprocessing in the solar reflective channels,
Proceedings of the 5th Colloquium—Mesures Physiques et
Signatures en Télédétection (Physical Measurements and
Signatures in Remote Sensing), No ESA SP-319, 14–18 January,
Courchevel, France, pp. 101–104.
Teillet, P.M., 1992. An algorithm for the radiometric and atmospheric correction of AVHRR data in the solar reflective
channels, Remote Sensing of Environment, 41:185–195.
Teillet, P.M., and B.N. Holben, 1994. Towards operational radiometric
calibration of NOAA AVHRR imagery in the visible and nearinfrared channels, Canadian Journal of Remote Sensing, 20(1):1–10.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
Vermote, E., and Y.J. Kaufman, 1995. Absolute calibration of
AVHRR visible and near-infrared channels using ocean and
cloud views, International Journal of Remote Sensing, 16(13):
2317–2340.
Vermote, E., N. El Saleous, Y.J. Kuafman, and E. Dutton, 1997.
Data preprocessing: 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., D. Tanré, J.L. Deuzé, M. Herman, and J.J. Morcette,
1997. Second simulation of the satellite signal in the solar
spectrum: An overview, IEEE Transactions on Geoscience and
Remote Sensing, 35:675–686.
(Received 16 December 2004; accepted 10 February 2005;
revised 01 August 2005)
S e p t e m b e r 2 0 0 6 1035