Shabanov, NV, Huang, D, Yang, WZ, Tan, B, Knyazikhin, Y, Myneni, RB, Ahl, DE, Gower, ST, Huete, AR, Aragao, LEOC, Shimabukuro, YE (2005). Analysis and optimization of the MODIS leaf area index algorithm retrievals over broadleaf forests. IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, 43(8), 1855-1865.
Broadleaf forest is a major type of Earth's land cover with the highest observable vegetation density. Retrievals of biophysical parameters, such as leaf area index (LAI), of broadleaf forests at global scale constitute a major challenge to modern remote sensing techniques in view of low sensitivity (saturation) of surface reflectances to such parameters over dense vegetation. The goal of the performed research is to demonstrate physical principles of LAI retrievals over broadleaf forests with the Moderate Resolution Imaging Spectroradiometer (MODIS) LAI algorithm and to establish a basis for algorithm refinement. To sample natural variability in biophysical parameters of broadleaf forests, we selected MODIS data subsets covering deciduous broadleaf forests of the eastern part of North America. and evergreen broadleaf forests of Amazonia. The analysis of an annual course of the Terra MODIS Collection 4 LAI product over broadleaf forests indicated a low portion of best quality main radiative transfer-based algorithm retrievals and dominance of low-reliable backup algorithm retrievals during the growing season. We found that this retrieval anomaly was due to an inconsistency between simulated and MODIS surface reflectances. LAI retrievals over dense vegetation are mostly performed over a compact location in the spectral space of saturated surface reflectances, which need to be accurately modeled. New simulations were performed with the stochastic radiative transfer model., which poses high numerical accuracy at the condition of saturation. Separate sets of parameters of the LAI algorithm were generated for deciduous and evergreen broadleaf forests to account for the differences in the corresponding surface reflectance properties. The optimized algorithm closely captures physics of seasonal variations in surface reflectances and delivers a majority of LAI retrievals during a phenological cycle, consistent with field measurements. The analysis of the optimized retrievals indicates that the precision of MODIS surface reflectances, the natural variability, and mixture of species set a limit to improvements of the accuracy of LAI retrievals over broadleaf forests.