On evaluation of precipitation fields with rain station data

Abstract

Nowadays, limited area numerical weather prediction models provide meteorological forecasts with horizontal grid spacing of only a few kilometers and grid spacing will decrease further in the coming years caused by progress in high-performance computing [9, 26]. Precipitation forecasts are of primary interest for both researchers and the public. For example, in flood forecasting systems precipitation is the crucial input parameter, especially in mountainous watersheds. Like the grid spacing of weather prediction models the grid spacing of regional climate models is decreasing. Precipitation forecasts have to be evaluated and errors have to be quantified. The most important evaluation method is comparison of meteorological simulation results with meteorological observations. But, before errors can be quantified two decisions have to be made. First, a set of useful statistics has to be chosen. This shall not be the issue of this paper. The interested reader is referred to, for example, Murphy and Winkler [23], Wilks [31], Wilson [32]. We apply for illustration a small set of simple continuous statistics. Our focus is on the second problem: What is the observational reference? Rain station data is commonly preferred to remote sensing data, in particular radar data, because of the relatively large measurement uncertainties [e.g., 1, 14, 34]. Is it reasonable to compare precipitation forecasts valid for grid boxes with several kilometers in diameter with sparsely distributed rain station data valid for small areas of 1000cm2? This is often done in an operational framework since it can be implemented by simple means. This areato- point evaluation is criticized and it is proposed to perform some upscaling or regionalization of the station data up to forecast grid resolution [29, 12]. Regionalization can be done by some fitting approach yielding a precipitation analysis. For example, a recent analysis of precipitation for the European Alpes by Frei and Haller [17] has a time resolution of 24 h and a spatial grid of about 25km with regionally even lower effective resolution depending on the available surface station network. This type of analysis is useful for model validation at the 100 km-scale [see, e.g., 5, 16, 18], but not at 10 km-scale or even less. Analysis is a smoothing regionalization. This deteriorates application in higher-moment statistics if the network is not dense enough. The statement 'dense enough' critically depends on the applied pixel support (is a pixel value representative for boxes with diameter of 100, 10, or 1 km?) and the analysis scheme. Another regionalization approach is stochastic simulation of precipitation fields with conditioning on the available station data. The idea of this is that the data is respected and the spatial variability is represented more realistically than in the analysis. Then the forecast can be compared with an ensemble of simulated fields. The ensemble mean field is an analysis but the mean higher-moment statistics have not the same value than if the forecast is just compared with the analysis alone. This paper applies regionalization and performs area-to-point or area-toarea comparison in evaluation of daily precipitation forecasts. The forecasts to be evaluated by example are the forecasts of the NWP model ALADIN that is operational at the Austrian national weather service with 10 km grid spacing. ALADIN, the forecast days, and the available station data are introduced in the next section. Section 3 discusses the applied evaluation approaches and subsequent sections discuss the respective results. Finally, some concluding remarks will be given. © 2009 Springer Berlin Heidelberg.