To enhance the performance of precipitation intensity forecasting and spatiotemporal distribution pattern forecasting

this paper describes a deep learning model of error-corrected spatiotemporal Rain-Net (ECST-RainNet). With radar echoes and quantitative precipitation estimation sequences as dual-channel inputs

this new model adopts convolutional neural networks to extract spatial features and a spatiotemporal long-short term memory network to capture temporal characteristics

incorporating an error correcting module to reduce systematic bias. We verify its performance and forecasting of typical precipitation events against the rain gauge measurements from the Liulin experimental watershed in Hebei Province. The results show that in comparison with the rainfall measured at these rain gauges

its quantitative precipitation estimation achieves a correlation coefficient of 0.67 and a root mean square error of 4.77 mm/h

substantially outperforming the dynamic radar reflectivity factor (Z) - surface precipitation intensity (R) relationship method. For one-hour lead-time precipitation forecasts

it generates an estimation error lower than that of the machine learning or deep learning model that uses radar echoes as a single data source. Meanwhile

it improves the forecasting accuracy of the distribution patterns and the precipitation center of cumulative areal rainfalls

showing the performance improved by integrating multiple data sources and its significance in radar precipitation estimation and nowcasting.