Application of Multi ML-Model for Multi-Scale Meteorological Drought Monitoring under CMIP6-Climate Scenario in Data-Scarce Semi-Arid Regions

<title>Abstract</title> <p>Meteorological drought forecasting is critically important as one of the fundamental challenges in water resources management, particularly in semi-arid regions such as the Karkheh Hydrosystem in iran. In this study, daily precipitation data from four CMIP6 models under SSP126 and SSP585 climate scenarios for the period 2026–2039 were used to calculate the Standardized Precipitation Index (SPI) at four time-scales; 3, 6, 9, and 12 months. The climate model data were downscaled and bias-corrected using the Quantile Delta Mapping (QDM) algorithm. Subsequently, five machine learning models; including GPR, XGBoost, SVM, MLP, and Random Forest, were trained and evaluated using historical data from 1966 to 2018. The results showed that the GPR model, which achieved the highest coefficient of determination (R² = 0.944) and the lowest error values at the 6-month scale, was identified as the best-performing model, followed by XGBoost and SVM. It was also found that the 6-month scale, exhibiting the highest accuracy across all models, is the optimal time scale for drought forecasting. Analysis of SPI forecasts revealed that as the time scale increases, the percentage of drought decreases and conditions tend toward wetter periods. However, the intensity of extreme events under the pessimistic SSP585 scenario is significantly greater than under the optimistic SSP126 scenario. Overall, the optimistic SSP126 scenario, characterized by lower fluctuations and only mild drought, provides more stable conditions for water resources management. The findings of this study indicate that the GPR model is an efficient and reliable tool for forecasting and designing an early drought warning in the Karkheh Hydrosystem. Additionally, it is suggested to use 6-month timeframe for medium-term planning and a 12-month timeframe for long-term water resource management in this basin.</p>