关键词: 干旱预测;风险评估;长短期记忆网络;支持向量机;机器学习;郑州市;

Abstract: Aiming at the problem that few researches have focused on direct correlation between the factors affecting drought and historical drought records in the current drought risk assessment, a new drought risk level assessment method based on deep learning was proposed based on historical drought text data and meteorological data. This method used long and short-term memory neural networks and support vector machine to build a drought risk level assessment model to evaluate the drought risk level that may occur in the future. On this basis, the validity of the constructed model was tested with the meteorological data from 1951 to 2020 in Zhengzhou City and the textual description records of the disaster situation. The results showed that in 2019 and 2020, the spring droughts in the study area were quite serious, and the prediction accuracy rate was 75%. It is verified that the proposed drought risk assessment method with multi-source data is effective in predicting risk levels.

Key words: drought prediction; risk assessment; long and short-term memory networks; support vector machine; machine learning; Zhengzhou City