Work

Comparative Assessment of Padma River Bank Erosion Susceptibility Using Ahp, Fuzzy Ahp, Random Forest, Support Vector Machine, and XGBoost in a Gis-Based Framework

Year: 2026
Type: preprint
Abstract: Riverbank erosion in Bangladesh poses serious threat to thousands of people, loss of agricultural land, infrastructure sharing losses every year and one of the most destructive geomorphic processes of Bangladesh. This study fills this gap by designing and testing five susceptibility mapping methods fostering integration between Google Earth Engine (GEE) and ArcGIS namely Analytical Hierarchy Process (AHP), Fuzzy AHP, Selecting Random Forest (RF), Support Vector Machine (SVM) and eXtreme Gradient Boosting (XGBoost). Eleven geomorphic and hydroclimatic conditioning factors encompassed terrain, vegetation, hydraulic and climatic parameters were extracted from GEE. Four epochs of multi-temporal Landsat imagery were used to identify erosion and accretion polygons totaling 168.59 km 2 of erosion and 212.60 km 2 of accretion in the study reach. The 4807 sample points (with 3262 points from the erosion and 1545 points from the non-erosion samples) were split in a ratio of 70:30 to train and validate the machine learning (ML) models. XGBoost achieved superior performance with accuracy of 87.18%, F1-score of 0.909, ROC-AUC of 0.903, and Cohen's Kappa of 0.693, followed by RF (accuracy 85.86%, Kappa 0.659) and SVM (accuracy 80.67%, Kappa 0.561). The top three variables identified by RF were NDVI, flood frequency and water occurrence. The AHP consistency ratio of 0.027 (0.10 < CR < 0.15) indicated that the experts were reliable in assigning weights. Maps produced from each of the five methods showed that MCDM methods create wider distance buffer hazard maps, as does ML, suggesting complementary applications. The GEE–GIS–ML framework can be adopted elsewhere for other large alluvial. rivers in South and Southeast Asia.
Language: English
Open Access status: green
OpenAlex ID: W7236600642
Cites: 12
Cited by:
Related to: 10