Short-term Precipitation Prediction Based on a Neural Network Method from Radar Observations

Yan JI

Abstract


Based on the CINRAD_SA Doppler radar data and the rain gauge data of regional stations in Wuhan, the nowcasting of the radar echoes and rain intensity was conducted using artificial neural networks (ANN). First of all, the reflectivity values extracted from the raw data were interpolated to the three-dimensional rectangular lattice grid of 1km*1km in horizontal direction at the height of 1.5 and 3 km. The nearest 25 grid points above the stations were chosen as the input layer of the neural networks. The results show that the correlation coefficient R in the radar-rainfall estimation is more than 0.6, and the RMSE is less than 5 (0.1mm/6min) in most sites. The ANN extrapolation experiments indicate that the accuracy rate of 36mins forecast is higher than 50% at the threshold of 5 dBz. With the extension of forecast lead time, the accuracy rate decreases and drops to 45% for the 1 h forecasts.

Keywords


Nowcasting, Precipitation, Artificial Neural Networks, Radar data


DOI
10.12783/dtcse/aiie2017/18223

Full Text:

PDF

Refbacks

  • There are currently no refbacks.