Mapping of Debris-Covered Glaciers Using Object-Based Machine Learning Technique

Debris-covered glaciers in High Mountain Asia are important indicators of climatic variability. The present study proposed an improved glacier mapping technique based on a TIRS/(RED/SWIR) band ratio that integrates thermal infrared (TIR), visible red, and shortwave infrared (SWIR) reflectance information with slope parameter to map debris-covered areas. The object-based machine learning technique comprising a hybrid feature selection model and a decision tree classifier was adopted to classify the glacierized region. The mapping results stated that the proposed band ratio combined with the slope parameter has better differentiated supraglacial debris from other glacier surfaces in comparison with the existing debris-covered glacier mapping approaches. The resulted debris-covered glacier boundaries were also validated with reference glacier inventories. The supraglacial debris-covered area was mapped with a high user's accuracy of approximate to 98%. In addition, a high overall classification accuracy in the range of 99.59-99.84% was achieved with the proposed technique that overcomes the challenges of the previous noteworthy studies, confirming that this technique is effective in detecting debris-covered glaciers.