Spatio-temporal Prediction Of Rainfall In Semi-arid Regions Based On Deep Learning

In recent years,with the frequent occurrence of various meteorological disasters,the state and society attach increasing importance to meteorological forecast.Timely and accurate weather forecast not only can make the public better plan their work and life,but also can effectively achieve the goal of disaster prevention and reduction,which is of great significance to social stability.Rainfall forecast is a hot topic in the field of weather forecast.Due to the extremely complex weather system,the traditional rainfall forecast technology has a large error,so it is necessary to improve the forecast accuracy by manual correction.Therefore,how to improve the accuracy of rainfall forecast is an important research direction.At the same time,semi-arid areas in China are faced with such problems as scarce rainfall,scarce water resources and fragile ecological environment.Therefore,it is of great practical significance to carry out research on rainfall forecast in semi-arid areas,which is conducive to the future development of the region.In order to improve the accuracy of rainfall forecast in semi-arid areas,this paper presents a rainfall spatiotemporal prediction model based on extreme gradient boosting(XGBoost),graph sample and aggregate(Graph SAGE)and long-short term memory networks(LSTM).Then,based on 21 weather stations which belong to Lanzhou,Dingxi,Baiyin and Linxia Hui Autonomous Prefecture,the rainfall forecast for Lanzhou which is located in semi-arid areas was made.The model structure presented in this paper is a cascade model.Firstly,the XGBoost binary classification model is used to judge whether the samples are rain samples or not.For the sample model predicted as no rain,it is directly determined as no rain.For the rain samples,the Graph SAGE?LSTM model is used to extract the temporal and spatial correlation to predict the rainfall grade.In order to verify the effectiveness of the model,five benchmark models,XGBoost,LSTM,CNN,CNN?LSTM and Conv LSTM,were selected and compared.The experimental results show that the spatial-temporal model with cascade structure presented in this paper shows the best forecast results in the accuracy,precision,recall,and F1 scores,which can effectively improve the rainfall forecast effect.