Study on Short-term and Imminent Precipitation Prediction Model and Accuracy Evaluation Based on Depth Learning

Abstract

In the field of weather forecasting, short-term precipitation forecast is still a global problem. The complicated mechanism of convective weather system, the medium and small scale weather system brings great difficulties to the forecast, which is difficult to detect, understand and forecast. The application of short-term precipitation forecast is still in the initial stage. At present, the optical flow method based on radar echo is widely used in short-term and imminent precipitation prediction, but the optical flow method has obvious limitations. In order to improve the forecast precision of short-term and imminent precipitation forecast model, this study introduces the method of depth learning to build the forecast model. The application of depth learning to short-term precipitation forecasting is still in its initial stage. We analyze the existing models may exist in the shortcomings of the proposed in-depth learning model combination of different ideas. By combining different network structures (such as CONVLSTM, CONVGRU, etc.), we can get a better combined forecasting model. It has certain application value.