Dynamic assessment and prediction of potato disaster loss risk in Gansu Province, China

• There are large interdecadal fluctuations in the potato disaster loss risk. • Potato production in most counties is classified as medium–low or low risk. • High risk counties are mostly clustered in Qingyang and Longnan. • The barycenter migration trajectory of risk indicators varies noticeably. • The Interpolation-EMD-SVM scheme performs well in disaster loss risk prediction. Meteorological disasters occur frequently, and Gansu Province is a sensitive area for food production. Potatoes are a major crop in this province. As a result, executing risk zoning and risk prediction for potato production is quite important. However, in existing risk assessment and prediction research, the dynamic nature of risks and improving the accuracy of risk prediction are urgent scientific issues that must be addressed. Weighting, spatial econometric analysis, climate diagnosis technology, and machine learning models were used to provide a refined spatiotemporal evolution of potato disaster risk in China's Gansu Province, as well as predict future potato production risk. The findings indicate that there are significant interdecadal fluctuations in the potato disaster loss, which has decreased considerably since 2000. The average yield decrease rate in the 1980s, 1990s, 2000s, and 2010s was -13.9%, -15.4%, -9.1%, and -7.3%, respectively, and the county percentage susceptible to severe yield loss was 26.1%, 39.1%, 22.9%, and 12.9%. Second, most counties' potato production falls within the medium–low or low risk region. Eastern and southern Gansu are particularly vulnerable to catastrophic calamities. High risk counties are primarily clustered in Qingyang and Longnan, whereas low risk counties are concentrated in Wuwei and Gannan. Third, high risk locations have altered, and the migration trajectory of the risk indicator’s barycenter shows significant differences in direction and distance. The comprehensive risk moves in a southeast-west-northern direction, but the distance is short. Overall, disaster losses in most counties are decreasing, and future trends will be similar with previous patterns. The Interpolation-EMD-SVM scheme greatly increases the accuracy of the disaster loss risk prediction. The technology and methods provide a scientific foundation for accurately assessing risk dynamic characteristics, managing regional disaster risks, and preventing and mitigating disasters.