Assessment of Global Antibiotic Exposure Risk for Crops: Incorporating Soil Adsorption via Machine Learning.

Abstract

The overuse and misuse of antibiotics could significantly increase their accumulation in soils. Consequently, antibiotics possibly enter food chain through crop uptake, posing a threat to global food security. Assessing the exposure risks of antibiotics for crops is crucial for addressing this global issue. In this study, we assessed global antibiotic exposure risk for crops, incorporating a machine learning adsorption model based on 4893 data sets from nine antibiotics. The optimized machine learning adsorption model, using the eXtreme Gradient Boosting algorithm and the class-specific modeling strategy, demonstrated relatively good performance. Notably, we introduced unsaturated soil conditions and considered spatiotemporal variations in soil moisture and temperature for the first time in such a risk assessment. Global distributions of antibiotic exposure risk for crops were predicted for March, June, September, and December. The results indicate that soil moisture significantly influences the exposure risk assessment. Relatively high exposure risk for crops was observed during months with colder local temperatures: generally June for the Southern Hemisphere and December for the Northern Hemisphere. The resulting map highlights high-risk agricultural regions, including southern Canada, western Russia, and southern Australia.