Estimating the changing risks of low crop yield using non-stationary generalized Pareto distributions

Estimating the changing risk of low crop yield in a changing climate is an important task in various fields of agricultural research. According to the extreme value theory, the probability of extreme events can be approximated using generalized Pareto distributions. In this study, non-stationary generalized Pareto distributions were used to estimate the changing risk of low crop yield. The proposed methods were applied to global yield data for maize, wheat, rice, and soybean collected from 1961 to 2022, as well as local yield data for wheat, rice, and soybean in Japan, from a start date of either 1948 or 1958 and running to 2020. The results illustrated exacerbated trends of low-yield risk in maize crops in Africa; maize, wheat, and rice in Americas; maize and wheat in Western, Central, and Southern Asia; maize and wheat in Europe; and soybean in Japan. Only wheat in Japan showed trends of mitigating the risk of low yields. The proposed models were also validated through simulations. The results showed that the models can generally estimate the changing risks accurately, and the precision depends on the size of the data set. Although there is still room for improvement in the models, the present study demonstrates that it is possible to estimate changes in the risk of low yield using a data-driven approach based on extreme value theory without assumptions about climate and crop physiology.