The use of forecast gradients in 3DVar data assimilation

Zhu, Zhaochen; Yan, Hanjun; Ng, Michael K.

doi:10.1016/j.apm.2019.04.038

Cited by 4 publications

(1 citation statement)

References 38 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Report generation only includes a module of report statistics, which requires the corresponding data calculation results to be produced according to the selected elements, and then displayed in the form of reports. The elements it needs to select include: region, time period, time, station type and forecast model [17][18].…”

Section: Figure 1 Overall Structure Of the Meteorological Data Analys...mentioning

confidence: 99%

Optimal Analysis of Weather Forecast Supporting Convolutional Neural Networks

2020

View full text Add to dashboard Cite

In recent years, with the acceleration of social informatization, people's requirements for weather forecasting have gradually increased. Severe convective weather has attracted the attention of the meteorological department because of its characteristics of strong suddenness and great destructive power. As a forecasting method to prevent severe convective weather, short-term forecasting has important research significance. The purpose of this paper is to study the optimization analysis of meteorological forecasts supported by convolutional neural networks. In this paper, the deep learning method is used to conduct an applied research on the precipitation of short-term and imminent forecasting. Precipitation short-term forecasting is essentially the prediction of future radar echoes from a series of radar echo sequences, which can be regarded as a spatiotemporal sequence prediction problem. Based on the research and summary of commonly used neural networks, this paper refers to ConvLSTM (ConvolutionalLSTM) Structure A ConvGRU model (ConvolutionalGRU) combining Convolution Neural Network (CNN) and GRU (Gated Recurrent Unit) is proposed. Since the structure of GRU is simpler than that of LSTM, the effect is not much different. This model is compared to The ConvLSTM structure has faster training speed and smaller memory requirements. Another work of this paper is to improve the convolutional layer based on VGGNet (VisualGeometryGroupNet), using multiple small convolution kernels to stack instead of large convolution kernels, reducing the number of parameters and improving the feature extraction ability of the network. This model gives full play to the advantages of convolutional neural network and GRU, that is, the spatial feature extraction ability of convolutional structure and the memory ability of GRU to deal with time series problems. Finally, the prediction effects of the model and the optical flow method are compared through experiments to verify the applicability of the model in the short-term precipitation forecasting problem.

show abstract

Section: Figure 1 Overall Structure Of the Meteorological Data Analys...mentioning

confidence: 99%