Hyperspectral remote sensing for prediction of soil properties

This study was conducted to evaluate Hyperion satellite data along with field and laboratory spectral observations for predicting soil properties. The spectral observations of in-situ soils in the study area were taken at 85 random locations and soil samples of each location were scanned under laboratory conditions for recording reflectance spectra. Out of 242 bands of Hyperion satellite data, error free 158 bands were used for the study. Laboratory and field measured reflectance data were resampled to obtain Hyperion comparable 158 bands. Three data sets were used for developing prediction models for eight soil properties; Soil Organic Carbon content (SOC), CaCO₃, Mineralizable Nitrogen (N), available Phosphorous (P), available Potassium (K), sand %, silt % and clay %. Prediction models were developed using stepwise regression approaches. The study showed that the corrected Hyperion spectral pattern was comparable with ground-based spectra. Reflectance and derived spectra were used for models and derivatives were found as better predictors of soil properties than that of reflectance or absorbance, irrespective of the platform. Although the predictability of Hyperion reflectance is lower than that of field and laboratory reflectance, some soil properties such as Silt, SOC and CaCO₃ could still be predicted with reasonably good accuracy (R2 > 0.5) with Hyperion data.