Fusion-based framework for meteorological drought modeling using remotely sensed datasets under climate change...

Area of Science:

• Hydrology and Climate Science
• Artificial Intelligence in Environmental Modeling
• Remote Sensing Applications

Background:

• Severe drought events necessitate improved prediction and monitoring for effective drought readiness and mitigation.
• Existing meteorological drought modeling approaches require enhancement for accuracy and reliability, especially under changing climate conditions.
• Fusion-based frameworks offer potential for integrating diverse data sources to improve drought assessment.

Purpose of the Study:

• To develop and evaluate a fusion-based framework for meteorological drought modeling using remotely sensed and ground-based data.
• To compare the performance of individual artificial intelligence (IAI) models and advanced fusion models for drought prediction.
• To project future drought conditions under different climate change scenarios.

Main Methods:

• Utilized high-resolution remotely sensed precipitation datasets (PERSIANN-CDR, CHIRPS, ERA5, GPCC) to estimate non-parametric Standardized Precipitation Index (nSPI).
• Employed K-means clustering for station classification and developed four Individual Artificial Intelligence (IAI) models (ANFIS, GMDH, MLP, GRNN) for drought modeling.
• Implemented advanced fusion methods, Multi-Model Super Ensemble (MMSE) and Random Forest (RF), to combine IAI model outputs for enhanced drought projection under RCP4.5 and RCP8.5 scenarios.

Main Results:

• The Random Forest (RF) fusion model exhibited the best performance in the Gavkhooni basin, Iran, with the lowest estimation error (RMSE of 0.391) and highest coefficient of determination (R² of 0.810).
• Drought modeling accuracy was enhanced through clustering stations and fusing results from multiple AI models.
• Analysis indicated that the 12-month timescale drought had a more severe impact on the basin compared to other timescales.

Conclusions:

• The proposed fusion-based framework effectively integrates remote sensing data and AI for accurate meteorological drought modeling and projection.
• The Random Forest model is a highly effective tool for combining multiple drought prediction models, improving overall accuracy.
• Findings highlight the critical need for drought preparedness, particularly concerning long-term drought impacts on the Gavkhooni basin.