Hyperspectral Remote Sensing for Agriculture Land Use and Land Cover Classification

Abstract

Food production is accountable for about 20–30% of anthropogenic greenhouse gas emissions, with the agricultural sector becoming the dominant source of these emissions. Land use information is important for agriculture management, the information about which can be obtained by hyperspectral (HyS) remote sensing. The high spectral information from hyperspectral sensors can help in differentiating various LU/LC classes. In LULC, focus is to be laid in classification of closely resembling classes which is possible only from HyS RS. This requires development of specific algorithms. A review of current algorithms for processing of HyS datasets is carried out in this article. This includes validating various atmospheric correction (AC) models, dimensionality reduction techniques (DR), and classification methods. Results show that FLAASH absolute AC model gave a higher resemblance with the ground spectra with higher correlation for agriculture and built-up classes. Classification is performed using seven per pixel classifiers and one ensemble classifier. Support vector (SVM) and ensemble classifiers for both Hyperion and AVIRIS-NG HyS images have shown higher accuracy with accuracy percentage ranging between 90 and 95%. Accordingly, the case studies for delineation of LU/LC under different scenarios facilitate a feasible and viable overall carbon sequestration.