The Intelligent Interpretation Method For Improving The Quality Of Conventional Meteorological Elements In Numerical Weather Prediction

Numerical weather prediction is the backbone of modern weather forecast operations and related meteorological fields.At present,a seamless and grid-based fine weather forecast business system based on the numerical weather prediction has been formed.Although the research on numerical weather prediction has achieved many achievements,high-precision weather forecast still faces many challenges due to uncertainty and incompleteness.This paper focuses intelligent interpretation on numerical weather prediction.Methods based on deep learning are proposed to forecast visibility,total cloud cover,precipitation type,temperature,precipitation areas,wind speed,and wind direction.The main work is as follows:1.On the forecast of visibility,total cloud cover,and precipitation,this paper makes a statistical analysis of the observation records of these three elements.The related forecast products are discussed.The forecast model is built and trained to forecast visibility,total cloud cover,and precipitation type based on the convolutional neural network.The forecast results of 5 provincial weather stations in Tianjin show that the proposed intelligent interpretation model can optimize the visibility forecast and total cloud cover forecast in numerical weather prediction.The mean absolute error of the visibility forecast can be dropped to about 6km.The mean absolute error of the total cloud cover forecast can be dropped to about 20%.2.On the forecast of temperature and precipitation area,this paper makes a statistical analysis of their observation records.The related forecast products are discussed.The forecast model is built and trained to forecast temperature and precipitation areas based on the fully convolutional neural network and U-net.The forecast results of 267 regional weather stations in Tianjin show that the proposed intelligent interpretation model can optimize the temperature forecast and precipitation forecast in numerical weather prediction.The mean absolute error of the temperature forecast can be dropped to about 1.3℃.The accuracy of the temperature forecast can be improved to about 80%.The accuracy of the precipitation forecast can be improved to about 90%.The threat score of the precipitation forecast can be improved to about3.On the forecast of wind speed and direction,this paper makes a statistical analysis of their observation records.The related forecast products are discussed.The forecast model is built and trained to forecast wind speed and direction based on the multi-task learning and generative adversarial network.The forecast results of 267 regional weather stations in Tianjin show that the proposed intelligent interpretation model can optimize the forecast of wind speed and direction in numerical weather prediction.The mean absolute error of the wind speed can be dropped to about 1.3m/s.The cosine of the difference between the forecasted wind direction and the observed wind direction can be improved to about 0.7.