Climate projections of a multivariate heat stress index: the role of downscaling and bias correction

Ana Casanueva,Tord Kjellstrom,Cornelia Schwierz,Sixto Herrera,Sven Kotlarski,Andreas M Fischer

doi:10.5194/gmd-12-3419-2019

Abstract

Abstract. Along with the higher demand for bias-corrected data for climate impact studies, the number of available data sets has largely increased in recent years. For instance, the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP) constitutes a framework for consistently projecting the impacts of climate change across affected sectors and spatial scales. These data are very attractive for any impact application since they offer worldwide bias-corrected data based on global climate models (GCMs). In a complementary way, the CORDEX initiative has incorporated experiments based on regionally downscaled bias-corrected data by means of debiasing and quantile mapping (QM) methods. In light of this situation, it is challenging to distil the most accurate and useful information for climate services, but at the same time it creates a perfect framework for intercomparison and sensitivity analyses. In the present study, the trend-preserving ISIMIP method and empirical QM are applied to climate model simulations that were carried out at different spatial resolutions (CMIP5 GCM and EURO-CORDEX regional climate models (RCMs), at approximately 150, 50 and 12 km horizontal resolution) in order to assess the role of downscaling and bias correction in a multivariate framework. The analysis is carried out for the wet-bulb globe temperature (WBGT), a heat stress index that is commonly used in the context of working people and labour productivity. WBGT for shaded conditions depends on air temperature and dew-point temperature, which in this work are individually bias corrected prior to the index calculation. Our results show that the added value of RCMs with respect to the driving GCM is limited after bias correction. The two bias correction methods are able to adjust the central part of the WBGT distribution, but some added value of QM is found in WBGT percentiles and in the inter-variable relationships. The evaluation in present climate of such multivariate indices should be performed with caution since biases in the individual variables might compensate, thus leading to better performance for the wrong reason. Climate change projections of WBGT reveal a larger increase in summer mean heat stress for the GCM than for the RCMs, related to the well-known reduced summer warming of the EURO-CORDEX RCMs. These differences are lowered after QM, since this bias correction method modifies the change signals and brings the results for the GCM and RCMs closer to each other. We also highlight the need for large ensembles of simulations to assess the feasibility of the derived projections.

Highlights

In the last years the amount of available climate projection data has largely increased thanks to the development of intercomparison projects (Coupled Model Intercomparison Project – CMIP, Taylor et al, 2011; Inter-Sectoral Impact Model Intercomparison Project – ISIMIP, Warszawski et al, 2014) and other initiatives
Our results show that the added value of Regional climate models (RCMs) with respect to the driving global climate models (GCMs) is limited after bias correction
The performance in terms of mean biases of the two Bias correction (BC) methods for individual variables is good and differences related to the parametric (ISIMIP) or empirical (QM) nature of the method may arise on the tails of the distribution, for which quantile mapping (QM) outperforms ISIMIP

Summary

Introduction

In the last years the amount of available climate projection data has largely increased thanks to the development of intercomparison projects (Coupled Model Intercomparison Project – CMIP, Taylor et al, 2011; Inter-Sectoral Impact Model Intercomparison Project – ISIMIP, Warszawski et al, 2014) and other initiatives There have been many efforts towards the distillation of climate data into usable climate information (Hewitson et al, 2014; Fernández et al, 2018) This is largely hampered by the large envelope of uncertainty, which grows in the subsequent steps in the production of climate information, the so-called “uncertainty cascade” (Wilby and Dessai, 2010). In this work we assess the role of downscaling and bias correction as key elements in the development of climate information For this purpose, we intercompare climate change projections of heat stress in Europe coming from different data sources, at different spatial resolution and corrected with two different bias correction methods, in order to identify the major sources of uncertainty in terms of present and future climate

Objectives

Results

Conclusion