A Machine Learning Model For Radar Quantitative Precipitation Estimation In Beijing, China

Radar quantitative precipitation estimation (QPE) has been mostly addressed by using parametric algorithms based on empirical relations, such as Z − R relations and K
DP
− R relations. These relations are difficult to represent the space-time variability of hydrometeor microphysics. In recent years, machine learning has been widely used in radar rainfall estimation and has gained great improvement. In this paper, we proposed a machine learning method based on the Multi-layer Perceptron (MLP) model, using reflectivity at 4 heights to estimate rainfall. This MLP model is compared with three different Z − R relations and all these methods are quantitatively evaluated and cross-compared using disdrometer observations. Results show that the MLP model can reduce the underestimation of Z − R relations and improve the accuracy of radar QPE, especially when the rain rate is larger than 25 mm/h.