A neural-network approach to radiometric sensing of land-surface parameters

A biophysically-based land-surface process/radiobrightness (LSP/R) model is integrated with a dynamic learning neural network (DLNN) to retrieve the l and-surface parameters from its radiometric signatures. Predictions from th e LSP/R model are used to train the DLNN and serve as the reference for eva luation of the DLNN retrievals, Both horizontally polarized and vertically polarized brightnesses at 1.4 GHz, 19 GHz, and 37 GHz for an incidence angl e of 53 degrees make up the input nodes of the DLNN. The corresponding outp ut nodes are composed of land-surface parameters, canopy temperature and wa ter content, and soil temperature and moisture (uppermost 5 mm), Under no-noise conditions, the maximum of the root mean-square (RMS) errors between the retrieved parameters of interest and their corresponding refer ence from the LSPIR model is smaller than 2% for a four-channel case with 1 9 GHz and 37 GHz. brightnesses as the inputs of the DLNN. The maximum RMS e rror is reduced to within 0.5% if additional 1.4 GHz brightnesses are used (a six-channel case). This indicates that the DLNN produces negligible erro rs onto its retrievals. For the realization of the problem, two different l evels of noises are added to the input nodes. The noises are assumed to he Gaussian distributed with standard deviations of 1K and 2K, The maximum RMS errors are increased to 9.3% and 10.3% for the 1K-noise and 2K-noise cases , respectively, for the four-channel case. They are reduced to 6.0% and 9.1 % for the 1K-noise and 2K-noise cases, respectively, for the six-channel ca se. This is an implication that 1.4 GHz is a better frequency in probing so il parameters than 19 GHz and 37 GHz, In addition, the promising of the pro posed inversion approach an the radiometric sensing of the land-surface par ameters is demonstrated.