Using hyper-spectral indices to detect soil phosphorus concentration for various land use patterns.

The management of nonpoint source pollution requires accurate information regarding soil phosphorus concentrations for different land use patterns. The use of remotely sensed information provides an important opportunity for such studies, and the previous studies showed that soil phosphorus shows no clear spectral response feature, while the phosphorus concentrations can be indirectly detected from the normalised difference vegetation indices (NDVI). Therefore, this study uses an optimised index in the RED and near-infrared (NIR) wavelengths to estimate total phosphorus and Olsen-P concentrations. The prediction accuracy is not entirely satisfactory with respect to a mixed land use dataset in which the determination coefficient was maintained at approximately 0.6, with particularly poor performance obtained for forest land group. However, the prediction accuracy increases markedly with the separation of samples into broad land use categories, even the R(2) was exceeded 0.8 for tea plantation group. The soil phosphorus prediction effect showed obvious variance for different land use patterns, which was related to vegetation growth conditions and critical soil properties including soil organic matter and mechanical composition.