Snow and glacial feature identification using Hyperion dataset and machine learning algorithms

Snow and glaciers are important sources of sustainable indicators for the availability of natural resources like freshwater, energy, minerals, forest, and agricultural products. Glaciers are disappearing, and snow cover areas are decreasing in the Himalayas due to increasing global temperatures and other related sensations. This study has been done using Hyperion imagery to identify different types of seasonal snow and glacier features in the north-western Himalayan states of India. Minimum noise fraction (MNF), Pixel Purity Index (PPI), and n-Dimensional (n-D) visualizer were applied to the atmospherically corrected Hyperion image. The spectral unmixing algorithm was applied to collect the endmembers or pure pixels from the Hyperion image. These endmembers were used for image classification. Different snow and glacier facies such as clean snow, ice mixed debris, blue ice, refreezing ice, dirty snow, dirty glacier ice, firn, moraine, glacier ice, and water body were found out using advanced pixel-based classification techniques. The field visits were conducted two times in a year from 2017 to 2019 in different parts of Himachal Pradesh, India. Sentinel-2 satellite data were used for accuracy assessment of the classification maps. Three advanced classifier methods artificial neural network (ANN), support vector machines (SVM), and random forest (RF) were used for the classification of Hyperion image. The overall accuracy for the classification algorithms ANN, SVM, and RF is 81.14%, 87.27%, and 90.98% respectively. These classification methods prove to be beneficial for classification of snow and ice properties for remote locations which are difficult to access and have rough weather conditions. However, appropriate field investigations and analyses using multiple satellite dataset are useful to increase the reliability of the results obtained from image classification.