Document Type : Research Paper
Authors
1 Associate Professor, Department of Geomorphology, Faculty of Planning and Environmental Sciences, University of Tabriz, Tabriz. Iran.
2 PhD Student in Geomorphology, Faculty of Planning and Environmental Sciences, University of Tabriz, Tabriz.Iran.
Abstract
Keywords
Main Subjects
Floods are among the most destructive natural hazards, posing severe threats to lives, infrastructure, and ecosystems—particularly in northern Iran’s humid forested catchments. The Kela Rud watershed in Babol County, Mazandaran Province, frequently experiences flash floods triggered by steep slopes, intense rainfall, and extensive land‑use modification. In this context, accurate spatial prediction of flood‑prone areas is critical for sustainable watershed management. This study integrates Geographic Information System (GIS)‑based spatial modeling with two probabilistic approaches—the Shannon Entropy (SE) and Evidential Belief Function (EBF) models—to identify and compare the efficiency of both methods in delineating flood‑hazard zones under complex topographic and hydrological conditions of the Hyrcanian forest region.
The research adopted a quantitative, descriptive–analytical approach based on spatial integration. A combination of remote sensing, topographic, climatic, and soil datasets was used to characterize the watershed. The modeling framework involved the extraction of nine flood‑conditioning factors: elevation, slope, aspect, land use, soil type, distance from rivers, distance from roads, drainage density, and the topographic wetness index (TWI). Elevation and terrain derivatives were extracted from the Shuttle Radar Topography Mission (SRTM) DEM with a 30‑m resolution, while land‑use and vegetation cover were derived from Sentinel‑2 imagery (2023) processed using the SNAP Toolbox and ArcGIS 10.8. Climatic data including annual precipitation (2011–2023) were obtained from synoptic and rain‑gauge stations in Babol, Bandpey, and Sangchal. Soil and geological layers were acquired from the Natural Resources and Watershed Organization (2019). The SE model calculated information weights for each factor using the entropy value of their class‑based flood frequency; EBF, based on the Dempster–Shafer theory, used belief (Bel), disbelief (Dis), and uncertainty (Unc) functions to generate the evidential flood probability surface.
To construct and validate the models, 175 spatial points were identified, including 70 flood‑occurrence points (training samples) and 105 non‑flood points (validation samples). These were derived from historical flood records, field reports, and Google Earth Pro time‑series interpretation over the period 2012–2023. The sample distribution was designed to maintain spatial representativeness across physiographic units—mountain, hill, and terrace zones. The power and precision of the models were evaluated using Receiver Operating Characteristic (ROC) curves and the Area Under the Curve (AUC) index, supported by Root Mean Square Error (RMSE) statistics to measure predictive accuracy and model error. This dual validation process ensured unbiased comparison and cross‑validation performance assessment
Although the research is predominantly quantitative, a mixed‑methods framework was applied by integrating numerical modeling with interpretive spatial analysis. Quantitative components included probabilistic computations and statistical weight assignment in SE and EBF, while qualitative interpretation drew on geomorphological expertise and visual inspection of land‑use dynamics from high‑resolution imagery. This hybrid integration allowed the research to not only predict flood susceptibility statistically but also interpret the environmental logic behind spatial patterns. The combination of empirical evidence and expert reasoning enhanced both analytical depth and contextual reliability.
The EBF model outperformed the SE model in both prediction precision and spatial realism. Statistical validation yielded AUC = 0.83 and RMSE = 0.219 for EBF, compared with AUC = 0.71 and RMSE = 0.293 for SE. Factor‑weight comparison revealed that elevation (0.1983), aspect (0.1517), and slope (0.1423) were the dominant drivers of flood susceptibility, contributing nearly 49 % of model variance. Moderate influences were found for land use, soil, and proximity to roads (≈ 29 %), whereas drainage density, river distance, and TWI accounted for only ≈ 22 %. Spatially, high‑risk zones with elevations of 700–1300 m and slopes of 30–45° were mainly concentrated near Shiadeh, Anjilak, and Lamsukola villages. These areas correspond to steep terrain, compact soils, and deforested agricultural land, confirming the hydrological logic behind model predictions.
The integration of spatial data with probabilistic models successfully delineated flood‑susceptible zones in the Kela Rud watershed. The EBF approach, due to its ability to manage uncertainty and synthesize heterogeneous spatial evidence, demonstrated greater robustness and predictive accuracy than the Shannon Entropy model.
The findings underscore the dominant influence of topographic conditions and land‑use patterns in controlling surface runoff and flood potential. The resulting flood‑hazard maps can inform land‑use planning, check‑dam site selection, and community‑based risk mitigation. For future research, combining EBF with machine‑learning algorithms such as Random Forest or Support Vector Machines is recommended to further enhance model precision and adaptability to data‑scarce environments. Overall, this integrative modeling framework offers a reliable methodology for flood‑risk assessment in forested mountainous basins across northern Iran and similar humid regions worldwide.
)