Robust climate scenarios for sites with sparse observations: a two‐step bias correction approach

Abstract

ABSTRACTObserved and projected climatic changes demand for robust assessments of climate impacts on various environmental and anthropogenic systems. Empirical‐statistical downscaling (ESD) methods coupled to output from climate model projections are promising tools to assess impacts at regional to local scale. ESD methods correct for common model deficiencies in accuracy (e.g. model biases) and scale (e.g. grid vs point scale). However, most ESD methods require long observational time series at the target sites, and this often restricts robust impact assessments to a small number of sites. This paper presents a method to generate robust climate model based scenarios for target sites with short and (or) sparse observational data coverage. The approach is based on the well‐established quantile mapping method and incorporates two major steps: (1) climate model bias correction to the most representative station with long‐term measurements and (2) spatial transfer of bias‐corrected model data to represent target site characteristics. Both steps are carried out using the quantile mapping technique. The resulting output can serve as end user–tailored input for climate impact models. The method allows for multivariate and multi‐model ensemble scenarios and additionally enables to approximately reconstruct data for non‐measured periods. The method's applicability is validated using (1) long‐term weather stations across the topographically and climatologically complex territory of Switzerland and (2) sparse data sets from Swiss permafrost research sites located in challenging conditions at high altitudes. It is shown that the two‐step approach performs well and offers attractive quality, even for extreme target locations. Uncertainties, however, remain and primarily depend on (1) data availability and (2) the considered variable. The two‐step approach itself involves large uncertainties when applied to short reference data sets or spatially heterogeneous variables (e.g. precipitation, wind speed). For temperature, results are promising even when using very short calibration periods.