• English 
  • Spanish 

Bias Correction Intercomparison Project: Applying an ensemble of bias - adjustment methods

Conference: 2nd Workshop on Bias Correction in Climate Studies
Year: 2018
Contribution type: Oral
Abstract URL: Book of abstracts
Nikulin, G., Bosshard, T., Wilcke, R., Yang, W., Bärring, L., , , , Dobler, A., Ioannis T, Koutroulis, A., Grillakis, M., Dosio, A., Vrac, M., Vautard, R., Noel, T., Switnek, M.

We present results from a Bias Correction Intercomparison Project (BCIP). The main driver for initiating the BCIP was a need in a number of European Union projects (FP 6 and 7) to provide bias-adjusted simulations for impact modelling together with information about bias-adjustment-related uncertainties and limitations. Within the BCIP two experiments focusing on different climate zones have been designed, namely: one on the mid-latitude climate taking the Euro-CORDEX simulations (50km) and the second on the tropical climate using the CORDEX-Africa simulations (50 km). 11 bias-adjustment approaches (different methods and the same methods with modifications) have been applied to two regional climate model (SMHI-RCA4 and IPSL-INERIS-WRF331F) simulations over Europe driven by the same global model (IPSL-CM5A-MR) under the RCP8.5 scenario. SMHI-RCA4 on average shows a wet bias over Europe in winter and a mixed pattern in summer: wet in central/northern Europe and dry in southern Europe. IPSL-INERIS-WRF331F has a strong wet bias along some coastal areas in summer and also too wet over the continent in winter. Gridded observations for Europe – E-OBS (v. 10) are used as a reference observational dataset for 1981-2010. A number of various statistics describing climatology are taken for evaluation (the calibration period) and analysis (future scenario) starting from basic seasonal means and ending, with a special emphasize, by high-order statistics as extreme events, variability and climate indices. The analysis shows that in general there is no “best” bias-adjustment method, although some methods are better “balanced” across different statistics for the calibration and scenario periods. Additionally, performance of different approaches depends not only on the bias-adjustment methods but also on the input climate simulations (different kinds of biases).