Classification of glacier with supervised approaches using PolSAR data.

Area of Science:

• Glaciology
• Remote Sensing
• Machine Learning

Background:

• Identifying and classifying distinct glacier features (snow, ice, debris cover) from satellite imagery remains a challenge.
• Accurate glacier feature classification using remote sensing data is crucial for environmental and societal applications.

Purpose of the Study:

• To develop a fully polarimetric SAR (PolSAR) deep neural network (DNN) classification approach for extracting alpine glacier features.
• To evaluate the performance of the developed DNN approach against traditional Support Vector Machines (SVM) for glacier feature classification.

Main Methods:

• Utilized a fully polarimetric SAR (PolSAR) deep neural network (DNN) classification model.
• Tested the DNN approach on Siachen and Bara Shigri glaciers.
• Compared classification results and accuracy metrics (Overall Accuracy, Kappa coefficient) with Support Vector Machines (SVM).

Main Results:

• The DNN classification approach achieved high overall accuracy (91.17% for Siachen, 89% for Bara Shigri) and good kappa coefficients (0.88 for Siachen, 0.85 for Bara Shigri).
• GF-DNN classification demonstrated an improvement of over 10% in overall accuracy compared to SVM for both glaciers.
• The study confirmed the effectiveness of DNNs for classifying glaciated terrain features.

Conclusions:

• Deep neural networks offer a powerful and accurate method for classifying alpine glacier features using PolSAR data.
• The developed GF-DNN approach significantly outperforms SVM, highlighting its potential for glaciological research and monitoring.
• This advanced classification technique enhances our ability to analyze and understand glaciated environments.