Estimation of vegetation water content through GA-PLS modeling of modis reflectance data

This work was aimed at investigating the effectiveness of genetic algorithms (GA) - partial least square (PLS) regression in estimation of canopy equivalent water thickness (EWT) from MODIS image data of the SMEX'04 site. Calibration of MODIS spectral variables (single bands, band ratios and curvatures) against the AVIRIS-derived EWT at a degraded resolution resulted in a high correlation (R
²
=0.77) and an equivalent well correlation was obtained for validation on the MODIS pixels not used in the calibration. The spectral variables used in the GA-PLS model include MODIS band 2 (841-875 nm), band 5 (1230-1250 nm), and band curvatures at MODIS band 4 (545-565 nm), band 1 (620-670 nm) and band 2. A caveat for applying these spectral variables to estimate EWT in other areas is that their applicability may depend on local environmental conditions.