Coupling machine learning and weather forecast to predict farmland flood disaster: A case study in Yangtze River basin

Abstract

Accurate water level prediction is the premise of farmland waterlogging prediction. A simple water level prediction model (FDPRE) based on four machine learning (ML) algorithms and weather forecasts were developed. The model can not only predict two key driving factors of waterlogging, rainfall and node water level but also estimate disaster losses. The results showed that the random forest and Multiple perception model (R2 ranged from 0.7180 to 0.9803 and 0.5717 to 0.9965) performed best. In the case of flooding lasting for one day, the economic loss of waterlogging under the 100 mm rainfall scenario (23.53 million dollars) was much higher than that under the 50 mm rainfall (12.69 million dollars). Under the two rainfall scenarios, the yield reduction rate in the lower reaches of the Sihu basin was higher than that in the upper reaches. The method of coupling ML and weather forecasts can well predict farmland waterlogging.