Interactive effects of ozone pollution and atmospheric drought on crop yields: A spatially explicit assessment.

Surface ozone pollution and drought are escalating threats to agroecosystem sustainability, yet their combined impacts on crop production remain poorly quantified, especially under future climate scenarios. Here, we apply an interpretable machine learning framework to assess the nonlinear interactions between ozone exposure, vapor pressure deficit (VPD), and other climate variables on U.S. soybean yields from 1981 to 2021. Using two severe drought years (1988 and 2012) as historical analogs reflecting pre- and post-ozone regulation and cultivar improvement contexts, we simulated soybean yield losses for representative future extreme drought years under two Shared Socioeconomic Pathways (SSP126 in 2073 and SSP585 in 2093). Results show that although VPD and ozone individually contribute modestly to yield variation, their interaction markedly amplifies losses in drought-prone regions such as Kansas. Irrigated areas, particularly in Nebraska and the Lower Mississippi River Basin, experienced substantially lower losses, highlighting the role of water management in buffering compound stress effects. Future projections indicate soybean yield reductions of up to 37.5 % under SSP585 compared to 17.1 % under SSP126, corresponding to estimated economic damages of $13 billion and $5.93 billion, respectively. Spatial analyses indicate that the Midwest and Northern Plains-including Minnesota, Iowa, Illinois, and North Dakota-may be future hotspots of soybean yield loss under intensified ozone and drought stress. These findings underscore the importance of incorporating compound stressor interactions into agricultural risk assessments, and offer spatially explicit guidance for region-specific adaptation to sustain crop productivity and ecosystem services under accelerating climate change.