Machine Learning and Bayesian Method For Monitoring and Forecasting Jayawijaya’s Tropical Glacier Change (Papers Track)
Muhamad Iqbal Januadi Putra (Universitas Siber Asia); Stuart Phinn (University of Queensland)
Abstract
Investigating the historical extent of tropical glaciers, the spatial patterns of glacier change, and forecasting future glacier coverage using machine learning provides critical insights into the cryosphere and the impacts of climate change on these systems. However, studies on the tropical glaciers of Jayawijaya Mountains remain limited. This study examines the historical and future retreat of glaciers on Jayawijaya Mountains - Indonesia using remote sensing and machine learning techniques. By analyzing 37 cloud-free Landsat images spanning 41 years (1980–2021), we mapped glacier cover changes using the Normalized Difference Snow Index (NDSI) and supervised machine learning Minimum Distance classifier. The results reveal a 99.96% reduction in glacier extent, with the remaining glaciers now confined to the Carstensz and East Northwall Firn regions. Bayesian Weight of Evidence (WofE) was employed to analyze the spatial patterns of glacier change in relation to geomorphological conditions. This analysis demonstrates that glacier retreat on Jayawijaya Mountains is strongly associated with elevation, followed by distance from the peak, aspect (downslope direction of a slope), and slope. Furthermore, spatiotemporal forecasting using a neural network-cellular automata (NN-CA) model predicts that the glacier at Carstensz will disappear by 2027, while the East Northwall Firn glacier will vanish by 2031. The model demonstrates high performance, with an accuracy of 0.994, a Kappa coefficient of 0.85, precision of 0.86, recall of 0.84, and an F1 score of 0.85.