ABSTRACT
The aim of this study was to map flood susceptibility in the Tajyar-Sarab watershed, compare the results of the SAW and VIKOR models, and assess the sensitivity of the results to changes in criterion weights. Thirteen topographic, hydrological, climatic, and environmental factors were used. Elevation data were extracted from the 30 m resolution SRTM dataset, NDVI was derived from Sentinel-2 imagery, land use data were obtained from the ESA WorldCover product, and precipitation data were extracted from CHIRPS. All layers were converted into a 30 m raster format and standardized within the range of 0–1 according to the direction of influence of each criterion. Criterion weights were determined using Shannon entropy, and flood susceptibility maps were generated using the SAW and VIKOR methods. A combined map was also produced by averaging the continuous outputs of the two models. Model agreement was evaluated using Pearson correlation, and the robustness of the results was assessed by modifying criterion weights by ±10% and ±20%, resulting in 52 scenarios with 500 Monte Carlo simulations. The highest entropy weights were assigned to SPI and STI, indicating their close dependency. The Pearson correlation coefficient between the SAW and VIKOR rankings was 0.949. According to the final combined map, high and very high susceptibility zones covered 0.60 km², accounting for 0.094% of the watershed area, and were mainly located along flow paths. Sensitivity analysis showed that SPI, STI, and geology had the greatest influence on the final results. The mean correlation coefficient between the combined map and the initial map across Monte Carlo simulations was 0.998. The VIKOR model showed lower sensitivity to weight variations compared with SAW, and the spatial pattern of the susceptibility map remained stable. However, this stability should not be interpreted as resistance to observed flood events.
Extended Abstract
Introduction
Flooding is among the most significant natural hazards worldwide, causing substantial annual damage to infrastructure, agriculture, communities, and natural resources (Rentschler, Salhab & Jafino, 2022). an issue particularly critical in mountainous watersheds, where steep slopes and short concentration times promote rapid hydrological responses and flash flooding (Khosravi et al., 2016).

Method
Thirteen flood conditioning factors were used: elevation, slope, curvature, distance from streams, drainage density, TWI, SPI, STI, NDVI, LULC, rainfall, geology, and soil. Terrain-related factors were derived from a hydrologically corrected 30 m SRTM DEM using standard GIS procedures; streams were extracted via the D8 algorithm (flow accumulation > 500), and distance from streams was computed using Euclidean distance. NDVI was derived from Sentinel-2 imagery, LULC from ESA WorldCover, and mean annual precipitation (2014–2023) from CHIRPS; soil and geological units were rasterized and scored based on their hydrological behavior. All layers were projected to WGS 1984 UTM Zone 38N at 30 m resolution, using bilinear resampling for continuous variables and nearest-neighbor for categorical ones. Criteria were classified into five susceptibility classes and standardized to a 0–1 scale (higher values indicating greater susceptibility), with inverse criteria (elevation, distance from streams, NDVI) standardized accordingly. Criterion weights were computed using the Shannon entropy method, treating each pixel as an alternative. Flood susceptibility was then modeled using entropy-weighted SAW and VIKOR (ideal/anti-ideal values of 1 and 0, v=0.5), with VIKOR output transformed as 1−Q for consistency with SAW. Both outputs were classified into five classes, compared via Pearson correlation and pixel-based spatial agreement, and combined into a final susceptibility map by averaging. Model robustness was evaluated through one-at-a-time sensitivity analysis and a 500-run Monte Carlo simulation, using Pearson correlation, MAE, RMSE, and pixel-wise standard deviation as stability indicators.

Result and discussion
Pearson correlation analysis showed weak to moderate relationships among most criteria, with the highest values between SPI and STI (r=0.963) and between rainfall and elevation (r=0.886), indicating some information overlap. Shannon entropy weighting assigned the highest weights to SPI (0.4686), STI (0.3108), geology (0.0839), and TWI (0.0375), reflecting the spatial variability and information content of each criterion. SAW and VIKOR outputs were strongly correlated (r=0.949), showing similar spatial susceptibility patterns, while pixel-based classification agreement reached 99.60%—largely driven by the dominance of the very-low susceptibility class rather than independent validation. In the combined susceptibility map, high and very high susceptibility zones covered about 0.60 km² (0.094% of the watershed), concentrated along flow-convergence and drainage-related areas. Sensitivity analysis identified SPI, STI, and geology as the most influential criteria, and Monte Carlo simulations confirmed the stability of spatial patterns under weight uncertainty, with mean correlations between simulated and baseline outputs of 0.998 for SAW, VIKOR, and the combined map.

Conclusions
The integration of remote sensing data, geomorphometric and environmental variables, Shannon entropy weighting, and SAW and VIKOR models provided a spatial assessment of flood susceptibility in the Tajyar–Sarab watershed. High and very high susceptibility zones were spatially limited (~0.60 km², 0.094% of the study area), mainly along flow-convergence pathways and drainage networks. The strong correlation between SAW and VIKOR outputs (r=0.949) reflected considerable agreement between the two approaches, though this should not be regarded as independent validation, since both models relied on identical criteria and weights. Sensitivity and Monte Carlo analyses confirmed the relative stability of susceptibility patterns under weight uncertainty; however, the absence of reliable independent flood-event data remains a key limitation, and the final product should therefore be interpreted as a relative susceptibility map rather than a hazard, risk, or deterministic prediction map. From a management perspective, the proposed framework offers a preliminary decision-support tool for prioritizing areas for further field investigation, hydrological–hydraulic modeling, and improved land-use and runoff-management strategies.

Funding
This research was conducted with the financial and institutional support of the Vice-Chancellor for Research of the University of Mohaghegh Ardabili.

CRediT authorship contribution statement
All authors contributed equally to all stages of the study, including research design, data collection, statistical analysis, analysis and interpretation of the results, preparation of the original draft, verification of the results, revision, review, and finalization of the manuscript

Declaration of Generative AI and AI-assisted technologies in the writing process
During the preparation of this work, the authors used QuillBot for text paraphrasing. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

Conflict of interest
The authors declare that they have no conflict of interest.

Ethical considerations
The authors adhered to the principles of research ethics throughout the conduct and publication of this study, and all authors confirm compliance with these principles.

Data availability statement
The data used in this study are available from the authors upon reasonable request.
Acknowledgements
The authors would like to thank the anonymous reviewers for their constructive and scientific comments and suggestions.