Different statistical methods are combined for the comparison of the spatial structures of two data-sets. This is useful for the validation of climate simulation data with respect to observational data of the same spatial and temporal coverage. We assume that simulated data and observed data are both given as time-series at locations such as grid cells or station locations. The spatial structure of such a data-set is understood as the spatial distribution of clusters (obtained from a cluster analysis of the time-series), which contain climatologically similar locations. If the spatial distribution of clusters was identical for the observed and the simulated data, the simulation would describe the spatial structure of the observations perfectly. Differences between these distributions are quantified using the κ-statistic. κ-variants can discriminate between differences which are due to differing cluster frequencies or and those due to differing spatial distributions. We demonstrate the method using simulation data from the statistical regional climate model STAR for Germany.

No abstract

Abstract. We present two case studies that demonstrate how a common evaluation methodology can be used to assess the reliability of regional climate model simulations from different fields of research. In Case I, we focused on the agricultural yield loss risk for maize in Northeastern Brazil during a drought linked to an El-Niño event. In Case II, the present-day regional climatic conditions in Europe for a 10-year period are simulated. To comprehensively evaluate the model results for both kinds of investigations, we developed a general methodology. On its basis, we elaborated and implemented modules to assess the quality of model results using both advanced visualization techniques and statistical algorithms. Besides univariate approaches for individual near-surface parameters, we used multivariate statistics to investigate multiple near-surface parameters of interest together. For the latter case, we defined generalized quality measures to quantify the model's accuracy. Furthermore, we elaborated a diagnosis tool applicable for atmospheric variables to assess the model's accuracy in representing the physical processes above the surface under various aspects. By means of this evaluation approach, it could be demonstrated in Case Study I that the accuracy of the applied regional climate model resides at the same level as that we found for another regional model and a global model. Excessive precipitation during the rainy season in coastal regions could be identified as a major contribution leading to this result. In Case Study II, we also identified the accuracy of the investigated mean characteristics for near-surface temperature and precipitation to be comparable to another regional model. In this case, an artificial modulation of the used initial and boundary data during preprocessing could be identified as the major source of error in the simulation. Altogether, the achieved results for the presented investigations indicate the potential of our methodology to be applied as a common test bed to different fields of research in regional climate modeling.

No abstract

No abstract