Towards a quality control of precipitation data

Abstract

The priority program 'Quantitative Precipitation Forecast' funded by the German Research Foundation (DFG) has been implemented in April 2004 for a period of six years. Within this program observations from almost 1000 rain gauges, which are not routinely used, have been collecting since 2007. Therefore a quality control of this observed raw data is needed. First of all a conditional probabilistic model for precipitation was developed. Based on the large scale atmospheric circulation or on small scale precipitation observations, the probability of precipitation exceeding a threshold is estimated. The model is applied to daily precipitation sums (1986-1999) of 231 rain gauge stations from the German Weather Service (DWD) network in Nordrhein-Westfalen (NRW) and NCEP reanalysis data. The procedure is based on a generalized linear model using logistic regression (GLM). Relative vorticity, vertical velocity (both in 850 hPa), and relative humidity (in 700 hPa and 850 hPa) are used as predictor variables. Alternatively precipitation observations are also used as predictor variables. The statistical model forecasts are validated by observations in a cross validation modus using the Brier Skill Score (BSS) and relative operating characteristics (ROC) curves. It is shown that the model represents the observations very well.