Flood is one of the most destructive and frequent natural disasters, causing extensive human and financial losses worldwide. This research aims to delineate floodplains and assess flood risk in the Qarah-Su River located in Kermanshah Province, Iran. In this study, the hydraulic model HEC-RAS was used in both Steady State and Unsteady (Unsteady) flow conditions to simulate floods with various return periods (2 to 1000 years). To estimate peak flood discharge in an ungauged area (Doab Qaranji), two methods, the Area-Discharge method and the SCS Unit Hydrograph method, were utilized. The geometric data required for the model were extracted from a Digital Elevation Model (DEM) using the RAS Mapper module. Furthermore, satellite imagery from Sentinel-1 (radar data), Sentinel-2, and Landsat-8 on the Google Earth Engine (GEE) platform, along with NDWI and MNDWI indices, were used to extract actual flood extents and compare them with the model results.   

   Drought, as one of the most significant climatic hazards in Iran, has intensified in frequency and severity in recent decades under the influence of global warming and altered precipitation patterns. Accurate understanding of the temporal and spatial behavior of drought and forecasting its future trends plays a fundamental role in sustainable water resources management and national food security. Therefore, the aim of this study was to evaluate meteorological and agricultural droughts in Iran using CMIP6 climate models and machine learning algorithms, and to analyze their future changes under different emission scenarios up to the mid-twenty-first century. In the first step, to select the optimal models from all available CMIP6 models for the climatic variables of precipitation, mean, maximum, and minimum temperature, the multi-criteria decision-making method TOPSIS was employed based on weighting five evaluation indices: MAE, MBE, NRMSE, NSE, and R. The ranking results indicated that the CMIP6 models including NorESM2-MM, TaiESM1, and AWI-CM-1-1-MR exhibited the best performance in simulating precipitation; NorESM2-MM, TaiESM1, and EC-Earth3-CC for mean temperature; EC-Earth3-CC, FIO-ESM-2-0, and MPI-ESM1-2-LR for maximum temperature; and GFDL-ESM4, MRI-ESM2-0, and EC-Earth3-CC for minimum temperature achieved the most accurate simulation results. The output data from these models, after spatial downscaling (Nearest) and bias correction using the quantile mapping method, were employed for calculating drought indices and training predictive models. Subsequently, the LSTM recurrent neural network was trained for simulation and forecasting, with 70% of the data allocated to training and 30% to testing for precipitation and potential evapotra  iration (estimated using the FAO-56 Penman-Monteith method), and 80% to training and 20% to testing for drought indices. Performance evaluation of the LSTM model demonstrated its high capability in reproducing the spatiotemporal behavior of drought indices; the NSE value for SPEI exceeded 0.90 and for SPI exceeded 0.6 at most stations. Model errors in estimating precipitation ranged from 2.5 to 84 mm at Zabol and Anzali stations, and for potential evapotra  iration from 11 mm at Anzali to 39 mm at Abadan, indicating high accuracy and stability of the model in arid and semi-arid regions of the country. Comparison of indices showed that SPEI, due to accounting for temperature and water balance, provides a more accurate representation of actual agricultural drought conditions compared to SPI.