A comparative geospatial multi-criteria decision-making framework for reliable riverbank erosion susceptibility mapping and sustainable planning

Bangladesh, a low-lying country, faces significant riverbank erosion, particularly along the Jamuna River, causing considerable socioeconomic challenges for vulnerable populations. This research aims to assess eleven hydro-geomorphological conditioning factors, namely slope, elevation, channel migration rate, normalized difference vegetation index, stream power index, drainage density, distance to river, topographic wetness index, curvature, aspect, land use and land cover, conducting multicollinearity, ensuring statistical diagnostics, generating and comparing susceptibility maps by AHP (Analytical Hierarchy Process) and FAHP (Fuzzy AHP) in an MCDM (multi-criteria decision-making) framework, and validating predictive performance of both models against a documented erosion inventory dataset comprising field-verified and satellite imagery-based observations. All the factors illustrated limited multicollinearity, and both models consistently indicated that the mid-channel char land unions of Mechra, Kaoakola, and Gabsara were the most susceptible char, whereas Sirajganj Paurashava, Kalia Haripur, and Saidabad remained relatively less exposed. The FAHP map captured more area in smoother gradients and higher susceptibility classes in the unstable unions. Performance was evaluated using ROC-AUC (receiver operating characteristic-area under the curve), which showed that FAHP is slightly better than AHP, with an AUC above 0.75 and a statistically significant bootstrap confidence interval with the corresponding DeLong test. Confusion metrics also confirmed that FAHP performed marginally better than AHP for this dataset, which gives slightly enhanced recall, accuracy, and F1 score, suggesting more effective identification of actual erosion subject to this study area and the data used. Spatial validation using Global Moran’s I showed significant positive spatial autocorrelation for both AHP (Moran’s I = 0.535, p < 0.001) and FAHP (Moran’s I = 0.525, p < 0.001), while LISA analysis identified spatially clustered high- and low-susceptibility zones. The spatial validation further showed that 100% of the High-High cluster points coincided with High & Very High susceptibility zones for both models, confirming their spatial reliability. In conclusion, although FAHP slightly outperformed AHP in terms of predictive performance and provided a more flexible representation of susceptibility, both models produced satisfactory and reliable results for the study area. Therefore, the proposed framework provides a useful basis for disaster risk reduction, erosion monitoring, and adaptive land-use planning in the morphologically dynamic Jamuna River corridor.