Soil moisture correction to improve mineral content estimates from hyperspectral data.

摘要

Continuum removal spectral absorption depth measurements combined with a Gaussian model predicting soil moisture were successfully used for mineral identification and abundance estimates in hyperspectral data from laboratory, field and airborne images.;In soil spectra, small amounts of water significantly diminish the albedo, causing mineral absorption depths to diminish non-linearly. The decline in albedo and the shape of the soil spectrum was modeled with inverted Gaussian functions over the convex hull boundary points of the soil spectrum in the shortwave infrared region (SWIR, 1.2 mum to 2.4 mum wavelengths). The area of the Gaussian curve, the Soil Moisture Gaussian Model (SMGM), reliably determined the soil surface water content (r2 = 0.9 and greater) for laboratory spectra from soils in two Mediterranean regions. This model's advantage is it does not use water absorption bands, e.g., overtones at 1.4 mum and 1.9 mum that saturate from atmospheric water vapor before reaching the instrument.;Multiple regression models to quantify mineral contents from laboratory samples and spectra of bare soils from airborne hyperspectral images obtained over the highly calcareous soils of Tomelloso, La Mancha, Spain, and high clay content soils of Lemoore, California, USA, improved with inclusion of the SMGM with clay and carbonate band depth predictions. Root Mean Square Error (RMSE) was significantly reduced from 8% to 5% carbonate content in analysis of the Tomelloso image by including landform stratification to create continuous tone and classified carbonate content maps. Including the SMGM and stratifying on sodic areas improved the clay content estimate from 4% to 2% with the Lemoore image. The clay continuous tone and classified maps were also created for clay content.