Forecasting Spring Frost Frequency Based on Time Series Decomposition Model and Machine Learning

Spring frost can significantly impact agriculture. Predicting the occurrence frequency of spring frosts is crucial for reducing agricultural economic losses and ensuring food security. To reduce noise interference and improve prediction accuracy, we propose a time series decomposition model to extract the trend features of spring frost frequencies in the Northern Hemisphere, the primary agriculture planting region in the world, from 1948 to 2016. Based on the extracted trend features, we utilized random forest and deep neural network models to forecast future occurrence frequencies of spring frost in the Northern Hemisphere. The results show that spring frost frequencies will continue to increase in some Northern Hemisphere regions in the future.