Agricultural drought is a complex natural hazard involving multiple variables and has garnered increasing attention for its severe threat to food security worldwide. In the context of climate change and the increased occurrence of drought events, it is crucial to monitor drought drivers and progression to plan the subsequent efforts in drought prevention, adaptation, and migration. However, previous studies on agricultural drought often focused on precipitation or evapotranspiration, overlooking other potential drivers related to crop drought stress. Additionally, macro-level analyses of drought-driving mechanisms struggle to reveal the underlying contexts of varying drought intensities. Northern Italy is one of the most important agricultural regions in Europe and is also a hotspot affected by extreme climate events in the world. In the summer of 2022, an extreme drought struck Europe once again, causing significant damage to the agricultural regions of Northern Italy. However, no studies to date have revealed the potential impacts and extent of extreme drought on this crucial agricultural area at a regional scale. Therefore, a comprehensive understanding of agricultural drought still requires further clarification and differentiated driver analysis. This study proposed a novel framework to comprehensively monitor agricultural drought with ensemble machine learning by constructing an integrated agriculture drought index (IADI) with remote sensing-related data including meteorology, soil, geomorphology, and vegetation conditions. Additionally, the Shapley Additive Explanation (SHAP) explainable model was applied to reveal the driving mechanism behind the drought event that occurred in northern Italy in the summer of 2022. Results indicated that the proposed explainable ensemble machine learning model with multi-source remote sensing products could effectively depict the evolution of agricultural drought with spatially continuous maps on an 8day scales. The SHAP analysis demonstrated that the extreme and severe agricultural drought in the summer of 2022 was closely related to meteorological indicators especially precipitation and land surface temperature, which contributed 68.88% to the drought. Moreover, the new findings also highlighted that soil properties affected the agricultural drought with a contribution of 28.3%. Specifically, in the case of moderate and slight drought conditions, higher clay and soil organic carbon (SOC) content contribute to mitigating drought effects, while sandy and silty soils have the opposite effect, and the contributions from soil texture and SOC are more significant than precipitation and land surface temperature. The proposed research framework could effectively contribute to improving the methodology in agricultural drought research, potentially bringing more instructive insights for drought prevention and mitigation.

Droughts cause significant economic damage worldwide. Evaluating their impacts on crop yield and water resources can help mitigate these losses. Using single variables such as precipitation, temperature, the soil moisture condition index (SMCI) and the vegetation condition index (VCI) to estimate drought impacts does not provide sufficient information on these complex conditions. Therefore, this study uses station-based and remote-sensingbased data to develop new composite drought indexes (CDIs), including the principal component analysis drought index (PSDI) and the gradient boosting method drought index (GBMDI). The first dataset includes historical observations of the standardized precipitation index (SPI), standardized precipitation evapotranspiration index (SPEI), and the self-calibrated Palmer drought severity index (SC-PDSI) at the 1-, 3-, 6-, and 12month timescales. The second dataset consists of remote-sensing-based data including the VCI, SMCI, temperature condition index (TCI), and precipitation condition index (PCI). We validated the results of PSDI and GBMDI by comparing them with historical drought events, in-situ drought indices, and annual winter wheat crop yield data from 2003 to 2022 using a regression model. Our temporal analysis revealed extreme to severe drought events during1990s and 2010s. GBMDI typically aligned with actual drought events and exhibited stronger correlations with in-situ drought indices than PSDI. We observed that drought intensity in winter were more severe than in summer. GBMDI was the most effective method, followed by PSDI, for assessing drought impacts on winter wheat yields. Thus, the proposed integrated monitoring framework and indexes offered a valuable and innovative approach to addressing the complexities of agricultural drought, particularly in evaluating its effects.