En skalaavhengig verifikasjonsmetode anvendt på finskalamodeller.

Abstract

High resolution forecast models have the potential to produce more detailed forecasts of, for example precipitation, than comparable models with a coarser resolution. This is because high resolution forecast models, amongst other advantages, give a more accurate description of weather phenomena such as intense convective precipitation systems. However, high resolution forecast models do not always score better with standard verification methods than models with a coarser resolution. This can be explained by the fact that individual phenomena, for example showers, cannot be forecasted at the exact place at the right time, even though the structure and intensity of the precipitation system is properly forecasted and in the right area. The forecast itself is good, even though the details are not entirely correct. We can easily assess this subjectively in each separate case. However, we need objective methods in order to assess the models qualitatively. Therefore, in the last few years, several new verification methods have been developed to provide more diagnostic and informative evaluations of high resolution forecasts of spatial fields, such as precipitation. In this thesis such a verification method is tested using Fraction Skill Score, FSS, which takes into consideration the spatial scale of the parameter to be forecasted, a so-called neighbourhood-area method. The method is tested on the UK Met. Office model UM 4 km and verified against Norwegian radars over an area which covers most of Southern Norway. Approximately 500 forecasts from July to August 2009 have been used as the data basis. The method shows an improvement of verification scores in relation to grid scale verification (forecasted point values against observations). For all percentiles and all threshold values we see a strong improvement if we increase the neighbourhood area. The model reaches an acceptable score for scales of the order of 50-100 km for the 90-95 % percentile. In order to reach the limit for the 99 % percentile we must increase the neighbourhood area to 230 km. The method can also be used to calculate the probability for an event to occur within a geographical area, and some examples of this are given.