A new advanced AI system anticipates droughts up to 6 months in advance

Valencia.- A new advanced system that combines international climate models, drought indices, and artificial intelligence (AI) allows for the anticipation of droughts up to six months in advance, which is key for water management and early warning in semi-arid regions such as the Júcar river basin.

Researchers from the Institute of Water and Environmental Engineering (IIAMA) of the Universitat Politècnica de València (UPV) Dariana Ávila Velásquez, Héctor Macián and Manuel Pulido have developed this pioneering system within the framework of the 'WATER4CAST 2.0' project, belonging to the PROMETEO program. The work, which has been recently published in the scientific journal Earth Systems and Environment, presents a pioneering approach that integrates multi-model seasonal climate predictions, widely used drought indices, and artificial intelligence techniques, significantly improving the reliability of current predictions, according to sources from the UPV. The research combines seasonal predictions from four international reference systems (ECMWF-SEAS5, Météo-France System8, DWD-GCF2.1 and CMCC-SPSv3.5), available through the Copernicus Climate Change Service (C3S), with historical ERA5 data, which have been post-processed using AI.

Key Results: High Reliability in Drought Prediction

Based on that information, researchers calculate two of the most widely used drought indices internationally: the Standardized Precipitation Index (SPI) and the Standardized Precipitation-Evapotranspiration Index (SPEI), on different time scales (6, 12, 18 and 24 months). In the case of six-month scale indices, reliability reaches values close to 90% in the same month of the prediction's issuance; three months ahead, the predictive capacity is maintained with values above 60%, while for longer time scales, such as 12, 18 and 24 months, the system retains a useful prediction skill up to six months in advance," according to Dariana Ávila Velásquez, lead author of the article. The methodology has been applied to the Júcar Hydrographic Demarcation, one of the most representative areas of the semi-arid Mediterranean, characterized by recurrent droughts, high pressure on water resources and high agricultural, urban and environmental demand. For Héctor Macián, the results confirm that the system is especially effective in reinforcing early drought warning, a fundamental aspect for anticipating management measures, reducing socioeconomic impacts, and increasing resilience to climate change.

The main novelty: integrating models, indices, and artificial intelligence

The main contribution of the work lies in the joint integration of multi-model seasonal predictions, operational drought indices (SPI and SPEI), and AI techniques, which allow for bias correction and better adaptation of the models to a regional scale. In addition, the team has developed a web-based operational implementation, which demonstrates the real applicability of the system for decision-making in water management, beyond the strictly academic field. "The multi-model approach we have developed significantly improves the robustness of predictions and reduces the uncertainty associated with traditional climate forecasts," explained Manuel Pulido, head of the research group on Hydroeconomic Models at IIAMA. Furthermore, the combination of the SPI and SPEI indices offers a more complete view of the phenomenon, as it not only takes into account precipitation deficits, but also the impact of rising temperatures, a key factor in the current context of climate change.

Methodology Transferable to Other Basins

Pulido has pointed out that the methodology is fully transferable to other basins and drought-prone regions, since one of the policies they have followed is to use data with global coverage and 100% open availability, which opens the door to its application in different climatic contexts and its integration into decision-making support systems for water management. IIAMA researchers highlight that this work demonstrates that seasonal predictions can become a reliable and operational tool for drought management, especially when combining several climate models and different indices. "In a scenario of increasing frequency and intensity of droughts due to climate change, this type of tool is essential to move towards a more anticipatory, efficient, and science-based water and risk management," they conclude.