Joint characterization of vegetation by satellite observations from visible to microwave wavelengths: A sensitivity analysis

This study presents an evaluation and comparison of visible, near-infrared, passive and active microwave observations for vegetation characterization, on a global basis, for a year, with spatial resolution compatible with cli matological studies. Visible and near-infrared observations along with the Normalized Difference Vegetation Index come from the Advanced Very High Res olution Radiometer. An atlas of monthly mean microwave land surface emissiv ities from 19 to 85 GHz has been calculated from the Special Sensor Microwa ve/Imager for a year, suppressing the atmospheric problems encountered with the use of simple channel combinations. The active microwave measurements are provided by the ERS-1 scatterometer at 5.25 GHz. The capacity to discri minate between vegetation types and to detect the vegetation phenology is a ssessed in the context of a vegetation classification obtained from in situ observations. A clustering technique derived from the Kohonen topological maps is used to merge the three data sets and interpret their relative vari ations. NDVI varies with vegetation density but is not very sensitive in se mi-arid environments and in forested areas. Spurious seasonal cycles and la rge spatial variability in several areas suggest that atmospheric contamina tion and/or solar zenith angle drift still affect the NDVI. Passive and act ive microwave observations are sensitive to overall vegetation structure: t hey respond to absorption, emission, and scattering by vegetation elements, including woody parts. Backscattering coefficients from ERS-1 are not sens itive to atmospheric variations and exhibit good potential for vegetation d iscrimination with similar to 10 dB dynamic range between rain forest to an d grassland. Passive microwave measurements also show some ability to chara cterize vegetation but are less sensitive than active measurements. However , passive observations show sensitivity to the underlying surface wetness t hat enables detection of wetlands even in densely vegetated areas. Merging the data sets using clustering techniques capitalizes on the complementary strengths of the instruments for vegetation discrimination and shows promis ing potential for land cover characterization on a global basis.