Comparison of Neural Networks and Support Vector Machines for the Mass Balance Ablation Observation of Glaciers in Baltoro Region
Keywords:
Feed Forward Neural Networks, Glacier mass balance, Support Vector Machines, Geographical Information System, LandsatAbstract
Glaciers are melting rapidly across poles due to climate changes. Apart from poles, land-based glaciers are also melting dramatically. It is crucial to monitor and reduce the rapid melting of glaciers. However, on-site monitoring of glaciers is challenging, costly, time-consuming, and entails subjectivity. Therefore, remote sensing techniques are indispensable to provide an automated and cost-effective solution in order to enhance the land-based monitoring of glaciers. This work presents a remote sensing technique that is based on neural networks (NN) and support vector machines (SVM) used for monitoring Baltoro glaciers, situated in the Karakurum range. The classifier classifies land, cloud, greenery, and the glacier mass. The data set used is the Landsat imagery from 1976, 1991, 1994, 2000, 2010, 2013 and 2016 including June and July. We observe that SVM achieves an accuracy of 99.97 %, while ANN attains an accuracy of 98.87 %. The results show that the glaciers have experienced a significant ablation in their masses with the temperature rise around the globe.