Los and colleagues at the NASA/Goddard Space Flight Center have assembled a global data set by month of the normalized difference vegetation index (NDVI) from 1982-1990, using daily global meteorological satellite observations from the Advanced Very High Resolution Radiometer (AVHRR). These data exist in two forms: a data set by month for each continent at ~7.6 km resolution; and a global data set at 1° latitude by 1° longitude. The 1° by 1° data set is being used to provide boundary conditions for general circulation models (GCMs) and is also being used in various primary production models of global vegetation ( e. g. CASA of Potter et al., NASA Ames). The ~7.6 km resolution data sets are being used for various research projects, such as identifying linked El Niño Southern Oscillation (ENSO)-NDVI anomalies, determination of desert boundaries, and modeling primary production at regional, continental, and global scales. Data have not been processed beyond 1990 due to the Mt. Pinatubo volcanic eruption in mid-1991 which necessitated atmospheric correction over most of the planet. A complete reprocessing of the entire 1981-1995 AVHRR data set is presently underway by Los and colleagues at NASA /Goddard Space Flight Center.
One of the goals of this work is to provide terrestrial time-varying satellite data sets of the NDVI from which the fraction of the intercepted photosynthetically active radiation (fPAR) could be determined. Using a combination of a global land cover classification based on Ruth DeFries' (University of Maryland) data and the AVHRR-derived NDVI time series, the following biophysical variables are generated for each land surface 1° x 1° grid cell: fPAR, leaf area index, albedo, surface roughness, and photosynthesis. Two years of these data (1987 and 1988) have been made available to the public (ISLSCP CD-ROM).
The satellite 1° x 1° data set was intended to be used as a simple interactive biosphere representation which would be coupled to general circulation models and global primary production models. By varying the prescribed surface vegetation classification(s), the effects of human modification of the natural environment can be evaluated as they affect global climate through GCM simulations. For example, Shukla, Nobre, and Sellers have simulated precipitation and temperature changes in the climate of South America by making GCM runs for the Amazon Forest intact and for varying degrees of large-scale Amazon deforestation. Shukla, Nobre, and Sellers have shown that significant drying of South America will result from large-scale deforestation in the Amazon.
Before an accurate representation of the global land surface NDVI could be produced, a significant amount of satellite data processing was necessary. Not only were daily satellite data processed from the NOAA 7, 9, and 11 satellites from 1982-1990, but satellite inter-calibration was necessary, as was cloud filtering, solar zenith angle correction, compensation for the 1982 El Chichón volcanic eruption, and replacement of missing satellite data (see Figure 6.1).
The satellite data were processed to produce NDVI from the first two channels of the AVHRR instruments as (2-1)/(2+1). This index is bounded between -1 and +1 and is directly related to the fraction of absorbed photosynthetically active radiation. Daily satellite data were mapped into an equal area projection and combined into monthly maximum value NDVI composites, thus minimizing scan angle, cloud, and atmospheric effects. Next the ~7.6 km NDVI data were averaged into 1° x 1° grid cells.
Empirical methods were developed to calibrate the AVHRR instruments' channel 1 and channel 2 to a fixed target through time. This is necessary to ensure that the satellite data sets are consistent and are not introducing calibration error into the simulations carried out using these data.
After the satellite data were inter-calibrated from 1982-1990 by month, the data were "sanitized" into a more robust data set, referred to as the "FASIR NDVI" which stands for Fourier Adjusted, Solar zenith angle corrected, Interpolation, and Reconstruction of missing data (see Figure 6.2). This was accomplished on a pixel by pixel basis at the 1° by 1° grid cell scale.
Fourier adjustment was used to remove extreme outliers from the time series. Solar zenith angle correction was necessary to compensate for illumination effects and increased atmospheric path lengths at high latitudes. Interpolation of missing data was necessary to represent the boreal forest and other cold areas in winter. Reconstruction of tropical data was necessary because of the very high cloud frequencies in humid tropical forests.
The cleaned FASIR-NDVI data were used to calculate land surface parameters at 1° by 1°. A linear relationship between NDVI and fPAR was empirically derived from the assumption that grid cells with NDVI values close to the maximum correspond to conditions of close to maximum (fPAR = 0.95) and that close to minimum NDVI values correspond to close to minimum fPAR (fPAR = 0.001). Leaf area index was calculated from fPAR using an exponential relationship between fPAR and LAI for broadleaf vegetation types and a linear relationship for needleleaf venation and shrubs (see Figure 6.3). The LAI was used to calculate roughness length, the amount of green leaves as a fraction of total leaf area, and, by assigning optical properties dependent on vegetation type, the albedo. Because the land surface parameters were derived from one series of instruments, they show realistic spatial, annual and interannual variations of the physical properties of vegetation.
Los and colleagues have produced a global satellite data set from which actual surface conditions can be derived and used in GCMs and ecological simulation models, making possible study of both large-scale human modifications to natural vegetation of the planet and of naturally-occurring interannual variability.