Multi-temporal vegetation canopy water content retrieval and interpretation using artificial neural networks for the continental USA

An inversion of linked radiative transfer models (RTM) through artificial neural networks (ANN) was applied to MODIS data to retrieve vegetation canopy water content (CWC). The estimates were calibrated and validated using water retrievals from AVIRIS data from study sites located around the United States that included a wide range of environmental conditions. The ANN algorithm showed good performance across different vegetation types, with high correlations and consistent determination coefficients. The approach outperformed a multiple linear regression approach used to independently retrieve the same variable. The calibrated algorithm was then applied at the MODIS 500 m scale to follow changes in CWC for the year 2005 across the continental United States, subdivided into three vegetation types (grassland, shrubland, and forest). The ANN estimates of CWC correlated well with rainfall, indicating a strong ecological response. The high correlations suggest that the inversion of RTM through an ANN provide a realistic basis for multi-temporal assessments of CWC over wide areas for continental and global studies.