Statistical adjustment of radar-based daily precipitation to ground data from the Czech territory

The paper summarises the results of testing various multiple regression models applied to the adjustment of radar-based daily precipitation to ground measured data. Resulting model based on the classified regression technique is proposed to be used operationally in the Czech Weather Service.

The test-input data include radar as well as ground precipitation. The radar-based daily precipitation was determined from maximum reflectivity measured by C band Doppler radar Gematronik METEOR360AC. Ground precipitation was measured by more than 600 ground stations giving daily sums. All data come from the period 1996-1998. The first two years of data are used as developmental data set. The set contains the data from the extreme precipitation period in 07/1997 when a large part of the eastern Czech Republic was flooded. The 1998 data are applied to the independent verification of the models. The verification set includes the days in 07/1998 when an organised convective system produced extreme precipitation (maximum 209mm in about 10h) and was followed by a flash flood over a limited area in NNE part of the Czech Republic. The inclusion of large precipitation in the both sets, developmental as well as verification, is important. It makes possible to improve the radar-based precipitation in the range of large precipitation values.

We use the concept of so called classified multiple regression where different regression models correspond to the quantitative precipitation classes. The decision about the application of the concrete regression model follows from the ground based precipitation estimate. To estimate the ground precipitation the one step correction (weighted average) was used after some tests of other interpolation techniques. Predictor set contain variables depending on the pixel coordinates relative to the radar position, on the rough radar precipitation and in some cases also on the interpolated ground precipitation.

The results of independent tests were compared with (a) rough radar-based estimation, (b) ground-based interpolation , (c) regression model used in the Czech Weather Service at present. The improvements expressed by bias and RMSE are large especially for the large precipitation classes (30-50mm, >50mm) in the summer period. In winter the interpolated ground precipitation gives results of the nearly the same quality as the adjusted radar. Tests with various number of ground stations used to provide the first precipitation guess show that the adjusted radar gives better results in the range of 80 stations. Such number of automatic ground stations will be available in the near future in the Czech Republic. The best technigue based on the use of 80 stations is proposed to be used operationally in the Czech Weather Service. It is found suitable for the adjustment of daily radar-based precipitation. Apart from the statistical results the precipitation fields from the flash flood period (22-23/07/1998) gained by different techniques are presented to show the adjustment effect.