The increasing frequency of floods driven by global warming and anthropogenic interventions poses a growing challenge. To address this issue, the present study investigates the impacts of future climate change scenarios on flood vulnerability in the Qareh-Sou watershed, Iran. To this end, an integrated approach combining machine learning algorithms, CMIP6 climate models, and remote sensing techniques was employed. Due to its superior accuracy among global models, the MPI-ESM1-2-HR climate model was selected to estimate precipitation trends under the Shared Socioeconomic Pathways (SSP1-2.6, SSP2-4.5, and SSP5-8.5) across two future time horizons: the mid-future (2041–2070) and the far-future (2071–2100). These scenarios encompass key variables, such as greenhouse gas emissions, urbanization, and land use changes, which play a crucial role in modulating flood susceptibility. Flood susceptibility mapping was conducted using Random Forest (RF) and Support Vector Machine (SVM) algorithms, utilizing 320 flood inventory points alongside 16 environmental and climatic conditioning factors. The predictive accuracy of the models was evaluated using the Area Under the Receiver Operating Characteristic Curve (ROC-AUC). The results demonstrated the outstanding performance of the RF model; by achieving an AUC of 94.75% during the validation phase and maintaining an accuracy exceeding 90% across all future time horizons and scenarios, this algorithm exhibited absolute superiority over the SVM model. These findings are vital for effective flood risk management.
Extended Abstract
Introduction
In recent years, the frequency and intensity of flood events have escalated globally due to anthropogenic climate change and alterations in land use patterns. These extreme hydrological events pose severe threats to critical infrastructure, ecosystems, and human safety, resulting in substantial economic losses annually. Developing nations, in particular, face heightened vulnerability due to limited resources and adaptive capacity. The Intergovernmental Panel on Climate Change (IPCC) reports that shifting precipitation patterns are intensifying atmospheric systems, leading to more frequent flash floods. Consequently, understanding and modeling basin-scale flood dynamics is essential for effective risk management and preventive action. The advent of the Coupled Model Intercomparison Project Phase 6 (CMIP6) and Shared Socioeconomic Pathways (SSPs) has revolutionized environmental hazard assessment by providing more realistic scenarios of future greenhouse gas emissions and socioeconomic development. However, uncertainties associated with downscaling methods and bias correction necessitate robust multi-index evaluation approaches. Furthermore, integrating Remote Sensing (RS), Geographic Information Systems (GIS), and Machine Learning (ML) algorithms—such as Random Forest (RF) and Support Vector Machine (SVM)—has emerged as a powerful methodology for identifying critical flood-prone areas. Despite these advancements, there remains a knowledge gap in simultaneously assessing the impacts of SSP scenarios and data-driven models in susceptible basins in western Iran. Therefore, this study aims to comprehensively evaluate the effects of climate change on flood susceptibility in the Qareh-Sou River Basin using CMIP6 scenarios (SSP1-2.6, SSP2-4.5, and SSP5-8.5). The primary objective is to quantify and map changing flood hazards under mid-future (2041–2070) and far-future (2071–2100) horizons to provide actionable insights for water resource management and disaster risk reduction. The advent of the Coupled Model Intercomparison Project Phase 6 (CMIP6) and Shared Socioeconomic Pathways (SSPs) has revolutionized environmental hazard assessment by providing more realistic scenarios of future greenhouse gas emissions and socioeconomic development. However, uncertainties associated with downscaling methods and bias correction necessitate robust multi-index evaluation approaches. Furthermore, integrating Remote Sensing (RS), Geographic Information Systems (GIS), and Machine Learning (ML) algorithms—such as Random Forest (RF) and Support Vector Machine (SVM)—has emerged as a powerful methodology for identifying critical flood-prone areas. Despite these advancements, there remains a knowledge gap in simultaneously assessing the impacts of SSP scenarios and data-driven models in susceptible basins in western Iran. Therefore, this study aims to comprehensively evaluate the effects of climate change on flood susceptibility in the Qareh-Sou River Basin using CMIP6 scenarios (SSP1-2.6, SSP2-4.5, and SSP5-8.5). The primary objective is to quantify and map changing flood hazards under mid-future (2041–2070) and far-future (2071–2100) horizons to provide actionable insights for water resource management and disaster risk reduction. Given the complex nature of flood phenomena, the integration of machine learning algorithms, such as Random Forest (RF) and Support Vector Machine (SVM), with remote sensing (RS) data provides an innovative approach for the multidimensional and accurate assessment of flood vulnerability (Chen et al., 2024; Hooshyaripor et al., 2024). This integrated data- and model-based framework enhances the ability to capture nonlinear relationships among the natural and anthropogenic factors influencing flood occurrence and substantially improves predictive accuracy compared with conventional statistical methods (Kazemi et al., 2024; Wang et al., 2025). The implementation of this research is expected to contribute not only to the scientific advancement of flood risk management but also to practical decision-making and policy development. The findings can provide a scientific basis for decision-making by key governmental organizations, including the Ministry of Energy, the National Disaster Management Organization, and the Ministry of Agriculture Jihad, particularly in climate change adaptation planning, the development of flood-resilient infrastructure, and the formulation of comprehensive policies aimed at reducing vulnerability in high-risk areas (Khodaei et al., 2025; Naderi et al., 2025).
Material and Methods
This study employs an integrated approach combining climate modeling, remote sensing, and machine learning. The study area is the Qareh-Sou River Basin in Kermanshah Province, western Iran. First, 320 historical flood and non-flood points were extracted using Sentinel-1 SAR imagery via Google Earth Engine and validated through field surveys. Sixteen effective environmental and climatic factors (including topography, hydrology, geology, and land use) were identified and processed into 30-meter raster layers. Multicollinearity among these factors was assessed using Variance Inflation Factor (VIF) and Tolerance (TOL) tests to ensure model stability. For climate projections, the MPI-ESM1-2-HR model from the CMIP6 ensemble was selected based on its superior performance in simulating historical precipitation trends compared to observational data from Kermanshah and Ravansar synoptic stations. Bias correction and statistical downscaling were performed using the Distribution Mapping (DM) method. Future precipitation trends were analyzed under three SSP scenarios (SSP1-2.6, SSP2-4.5, and SSP5-8.5) for two time horizons: mid-future (2041–2070) and far-future (2071–2100). Flood susceptibility maps were generated using two machine learning algorithms: Random Forest (RF) and Support Vector Machine (SVM). The dataset was randomly split into training (70%) and validation (30%) subsets. Model performance was evaluated using several statistical metrics, including the Area Under the Receiver Operating Characteristic Curve (ROC-AUC), Positive Predictive Value (PPV), Negative Predictive Value (NPV), Sensitivity, Specificity, and Root Mean Square Error (RMSE).
Results and Discussion
The multicollinearity test confirmed the independence of the 16 input factors (VIF < 10; TOL > 0.1). Among the climate models, MPI-ESM1-2-HR demonstrated the highest accuracy (lowest RMSE and highest R²) after bias correction, as validated by Taylor diagrams and statistical indices. Climate projections indicate a significant increase in annual precipitation volume and extreme rainfall events, particularly in the far-future horizon under the SSP5-8.5 scenario. For instance, precipitation in the Ravansar station is projected to more than double under SSP5-8.5 by 2100. In terms of flood susceptibility modeling, the RF algorithm outperformed SVM significantly. During the validation phase, RF achieved an ROC-AUC of 94.75%, with high sensitivity (91.89%) and specificity (84.62%), whereas SVM yielded a lower AUC of 72.35%. The RF model also maintained exceptional stability across all future scenarios, with AUC values consistently exceeding 90%. In contrast, SVM showed moderate performance with higher RMSE values. Spatial analysis revealed that under the mid-future horizon, the SSP1-2.6 scenario unexpectedly resulted in a wider spatial extent of high-risk areas due to the alignment of precipitation patterns with local topography. However, in the far-future horizon (2071–2100) under the pessimistic SSP5-8.5 scenario, both models predicted a dramatic expansion of high and very high susceptibility classes. Approximately 47–48% of the basin’s total area is projected to fall into critical risk categories, with flood-prone zones expanding from lowland plains into mountainous and foothill regions. This shift is attributed to the intensification of extreme rainfall events interacting with steep slopes and impermeable surfaces.
Conclusion
This study demonstrates that climate change will significantly exacerbate flood risks in the Qareh-Sou River Basin, particularly under high-emission scenarios. The Random Forest model proved to be a superior tool for flood susceptibility mapping compared to SVM, offering higher accuracy and stability in handling complex, non-linear relationships between climatic and environmental variables. The findings highlight a critical transition in flood hazard patterns by the end of the century, where risk extends beyond traditional riverine corridors into upstream areas. These results underscore the urgent need for proactive risk management strategies, including improved urban planning, watershed management, and engineering interventions, to enhance resilience against future climate-induced flooding. Policymakers and stakeholders should prioritize adaptive measures based on these high-resolution susceptibility maps to mitigate potential catastrophic losses.
Acknowledgements
This work is based upon research funded by the Iran National Science Foundation (INSF) under project No. 40402705.