摘要：Abstract. Bias adjustment is often a necessity in estimating climateimpacts because impact models usually rely on unbiased climate information, a requirementthat climate model outputs rarely fulfil. Most currently used statistical bias-adjustment methodsadjust each climate variable separately, even though impacts usually depend on multiplepotentially dependent variables. Human heat stress, for instance, depends on temperatureand relative humidity, two variables that are often strongly correlated. Whetherunivariate bias-adjustment methods effectively improve estimates of impacts that dependon multiple drivers is largely unknown, and the lack of long-term impact data prevents adirect comparison between model outputs and observations for many climate-relatedimpacts. Here we use two hazard indicators, heat stress and a simple fire risk indicator,as proxies for more sophisticated impact models. We show that univariate bias-adjustmentmethods such as univariate quantile mapping often cannot effectively reduce biases inmultivariate hazard estimates. In some cases, it even increases biases. These casestypically occur (i) when hazards depend equally strongly on more than one climaticdriver, (ii) when models exhibit biases in the dependence structure of drivers and(iii) when univariate biases are relatively small. Using a perfect model approach, wefurther quantify the uncertainty in bias-adjusted hazard indicators due to internalvariability and show how imperfect bias adjustment can amplify this uncertainty. Bothissues can be addressed successfully with a statistical bias adjustment that corrects themultivariate dependence structure in addition to the marginal distributions of theclimate drivers. Our results suggest that currently many modeled climate impacts areassociated with uncertainties related to the choice of bias adjustment. We conclude thatin cases where impacts depend on multiple dependent climate variables these uncertaintiescan be reduced using statistical bias-adjustment approaches that correct the variables'multivariate dependence structure.
