Merging TRMM And Gauge Precipitation In China Based On Deep Learning

Precipitation is the main influencing factor of runoff changes and the most important meteorological process that affects the regional water cycle.Hydrological modeling and forecasting are important methods for flood prevention and disaster reduction,and their simulation and prediction accuracy are very important.Of all the key factors in flood disaster prediction,the most important is the uncertainty of precipitation space.As a result,precipitation is the most important meteorological input in hydrological forecasting models.Obtaining high-precision precipitation spatial data is very useful for hydrological process analysis,water conservancy project planning and design,water resource allocation and management,flood and drought disaster monitoring,and geological disaster warning.Precipitation input in traditional hydrological simulation usually uses observation data from ground rainfall stations,but its inability to accurately reflect the spatial distribution of precipitation limits its application in hydrological models.This is because the spatial distribution reflected by precipitation information obtained from ground discrete stations is limited,and the distribution of gauges is very sparse.Satellite observations can obtain the spatial distribution information of precipitation,which has a wider coverage area.It can not only provide sufficient precipitation distribution information for regions with insufficient data,but also supplement the shortage of traditional ground precipitation.However,there are obvious systematic errors in satellite precipitation products,and it is difficult to obtain high-precision precipitation information.Therefore,the integration of precipitation information based on satellite-rainfall stations can effectively improve the accuracy of quantitative precipitation estimation.To improve the accuracy of quantitative precipitation estimation,numerous models have been developed for merging satellite and gauge precipitation.However,most established merging methods consider the spatial or temporal correlation between satellite data and rain gauge data separately,and the produced merged precipitation is still limited by low spatial resolution and regional inaccuracy.In this study,a deep fusion model is proposed to merge the TRMM 3B42 V7 satellite data,rain gauge data and thermal infrared images by exploiting their spatial and temporal correlations simultaneously.Specifically,the convolutional neural network(CNN)is combined with a long-short-term memory network(LSTM),where the spatial characteristics of the satellite,rain gauge and thermal infrared data are extracted by CNN,while their time dependence is captured by LSTM.Experiment results on 796 rain gauges in China show that:(1)the proposed CNNLSTM model outperforms the comparative models(CNN,LSTM,Multi-Layer Perception).(2)It can improve the accuracy of the original TRMM data in China(reducing the root mean square error(RMSE)and mean absolute error(MAE)by 17.0% and 14.7%,respectively and increasing the Correlation Coefficient to 0.72),even for different precipitation intensities and gauge sparse regions.(3)A dataset of merged daily precipitation from 2001-2005 with a higher resolution of 0.05° and higher accuracy over China is produced.(4)The original TRMM precipitation,CNN-LSTM fusion precipitation and gauges precipitation data were used as the precipitation input in the SimHyd hydrological model in the Xixian basin of the upper Huaihe River.The runoff simulation results show that the CNN-LSTM precipitation fusion model can effectively improve the original application effect of TRMM precipitation data in hydrological simulation.The fusion precipitation data can provide a more reliable data source for regions lacking data.