Search

Seasonal dynamics of snow cover is an essential area of research for hydrological modelling and water resource management. With the increased availability of remote sensing data, the timely information of the spatiotemporal distribution of snow cover is feasible at regular intervals. The primary objective of this study is to assess the effect of pansharpening in the accuracy of snow cover change detection in mountainous regions using freely available Landsat-8 multispectral data. In mountainous regions at the medium resolution, the changes at the mountain ridges are seldom identified. The incorporation of pansharpening in the change detection framework facilitates an improvement in the snow cover change detection at the ridges. For pansharpening, the PanNet architecture based on convolutional neural networks was adopted. A study area around Dhundi in the state of Himachal Pradesh in India was selected for the analysis. The experiments were carried out using a subset of Landsat-8 multispectral data acquired in the autumn and the winter seasons of 2017 and 2018, respectively. An improvement of 0.184 and 0.267 in the kappa coefficient was observed for the overall changes in the snow cover and at the ridges, respectively, based on the results from the proposed approach. [ABSTRACT FROM AUTHOR]

Spectral indexes (SI) are widely used for land cover characterization and also in several physical models for the study of land surface processes. For example, the normalized differenced vegetation index (NDVI) is used in the characterization of soil moisture along with shortwave infrared reflectance. However, for hyperspectral imagery (HSI) comprising many bands within a single spectrum, it is significant to identify the optimal bands for the development of SI. In this paper, we study the potential of band selection in specific bandwidths for the determination of SI. The proposed methodology includes two strategies for development of SI: direct SI determined by the best band within specific spectrums and fused SI determined by fusion of two best bands within specific spectrums. The experiments are conducted using three datasets, two corresponding to snow-covered areas, studied using the normalized differenced snow index (NDSI) and one comprising the agricultural area, studied using NDVI. The developed SI are evaluated through a comparison with the supervised classification maps from the corresponding HSI. A kappa coefficient of 0.693, 0.726 and 0.803 was observed between the results obtained from histogram slicing of SI with respect to the classification maps for the three datasets, respectively. [ABSTRACT FROM AUTHOR]