Beyond Bias Correction of Climate Change Projections

Within the framework of the European FP6 project ENSEMBLES (ensembles-based predictions of climate changes and their impacts) the systematic bias in simulated monthly mean temperature and precipitation for an ensemble of thirteen regional climate models (RCMs) was explored (e.g. Christensen et al. GRL 2008 doi:10.1029/2008GL035694). The models were forced with the European Centre for Medium Range Weather Forecasting Reanalysis (ERA40) and compared to a new high resolution gridded observational data set for Europe. It was found that each model has a distinct systematic bias relating both temperature and precipitation bias to the observed mean. By excluding a substantial fraction of the warmest and wettest months, respectively, we found that a derived second-order fit from the remaining months can be used to estimate the values of the excluded months. This demonstrates how the common assumption of bias cancellation (invariance) in climate change projections can have significant limitations when temperatures in the warmest months exceed 4–6 °C above present day conditions. Further analysis on the ENSEMBLES regional climate change simulations adopting this approach for bias correction suggests that model spread can be reduced this way. The implication for global models seems to be that the occurrence of extreme regional warming could be subject to similar systematic biases, whereby the projected warming in the warmest scenarios may exacerbated by some models. Thus the degree of uncertainty in climate change projections based on ensembles of GCMs to a certain extent is an artifact of model formulation. The models simulating the strongest warming are likely to be responding too much. The immediate implications for impact analyses based on climate model output are somewhat obvious; be prepared to watch out for GCM or RCM model outliers even more than before. The next appropriate step would seem to be to inspect climate models more carefully with a view towards their ability to depict the more extreme part of weather; if exaggerating be alerted, this may lead to erroneous conclusions related to a warming signal. Last but not least, the need for a full ensemble approach is called for, an analysis based on a single or few models could be highly susceptible to poor model performance — and be aware that the biases discussed here are not the ones that meets the eye if you compare a climate model with the real world as it looks today, it may be even worse than that!