Bias adjustment of multivariable heat-stress indices: evaluation of CORDEX historical simulations over southeastern South America

Abstract. Regional Climate Models (RCMs) provide high-resolution climate information, particularly relevant in southern South America, where complex topography and land–coast contrasts strongly influence climate variability. However, systematic biases in RCM simulations can limit their use in impact studies, especially those based on climate indices exceeding absolute thresholds, such as heat stress. Bias adjustment (BA) methods are commonly applied to adjust these biases, but their application in a multivariate context and under climate change conditions remains challenging.

This study evaluates multiple BA methods alongside different adjustment strategies over southeastern South America, focusing on two multivariate heat stress indices derived from temperature and humidity variables and with different sensitivity to their input variables: wet-bulb temperature (WBT) and simplified wet-bulb globe temperature (sWBGT).

BA methods were applied to CORDEX-RCM historical simulations using the high-resolution MSWX dataset as reference. Direct adjustment of the thermal indices was compared with indirect approaches, in which temperature and either specific or relative humidity were adjusted prior to index calculation. Univariate and multivariate BA methods, including trend- and non-trend-preserving approaches, were evaluated under a cross-validation framework during 24 austral summers (1982–2005).

For both indices, the lowest biases were found under the direct approach (i.e., adjusting the indices directly), although multivariate methods achieved comparable performance when the indices were adjusted indirectly (i.e., adjusting individual variables prior to index calculation). Within the indirect approaches, univariate methods yielded similar results, with no clear advantage of trend-preserving methods during the historical period. Multivariate methods provided additional improvements in some RCM simulations in which the dependence structure between temperature and humidity was not realistically represented. Notwithstanding this, biases in the derived indices also depended on the choice of humidity input variable, with adjustment based on specific humidity generally providing better results than adjusting relative humidity, especially for the WBT.

Overall, this study provides insight into the suitability of the BA methods for estimating multivariate thermal indices and paves the way for future assessments of heat stress conditions over South America.