Machine Learning-Based Gully Erosion Susceptibility Mapping: A Case Study of Eastern India

Area of Science:

• Geosciences
• Environmental Science
• Machine Learning Applications

Background:

• Gully erosion is a significant global land degradation issue.
• Understanding gully erosion susceptibility is crucial for effective land management and disaster prevention.

Purpose of the Study:

• To delineate areas with high gully erosion susceptibility (GES) using advanced machine learning techniques.
• To compare the performance of Random Forest (RF), Gradient Boosted Regression Tree (GBRT), Naïve Bayes Tree (NBT), and Tree Ensemble (TE) for GES modeling.

Main Methods:

• Utilized a gully inventory map (GIM) with 120 gullies for training (70%) and validation (30%).
• Employed fourteen gully conditioning factors (GCFs) and the weight-of-evidence (WofE) model.
• Applied RF, GBRT, NBT, and TE algorithms for GES mapping and validated using AUROC, SCAI, PPV, FDR, accuracy, MAE, and RMSE.

Main Results:

• Approximately 7% of the study basin exhibits high to very high susceptibility to gully erosion.
• All models demonstrated strong predictive capabilities for GES.
• The Random Forest (RF) model achieved the highest performance metrics (AUROC = 0.96, PPV = 1.00, FDR = 0.00, accuracy = 0.87) for the validation dataset.

Conclusions:

• The Random Forest model is highly accurate and suitable for modeling gully erosion susceptibility.
• The findings provide a valuable tool for identifying and managing gully erosion in the study basin and similar environments.