An assessment of the performance of bias correction techniques for surface ozone burdens simulated by global chemistry-climate models

Abstract

Despite continuous improvement during recent decades, state of the art global chemistry-climate models (CCMs) are still showing biases compared to observational data, illustrating remaining difficulties and challenges in the simulation of atmospheric processes governing ozone production and decay. Therefore, CCM output is frequently bias-corrected in studies seeking to explore changing air quality burdens and associated impacts on human health (e.g., Rieder et al., 2018). Here we assess the strengths and limitations of different bias correction techniques for CCM simulations with focus on maximum daily 8-hour average surface ozone. Ozone fields are chosen as ozone is known as regional pollutant and thus shows smaller spatial heterogeneity in its burden than e.g. particulate matter. Within our comparison a set of different innovative, as well as, common bias correction techniques are applied to output of selected global coupled CCMs contributing hindcast simulations to the Coupled Model Intercomparison Project Phase 6 (CMIP6). For bias correction and evaluation, we utilize gridded observational data for the European and US domains according to Schnell et al. [2014]. The statistical bias-correction techniques applied and compared are quantile mapping, delta-function, relative and mean bias correction. As surface ozone pollution is commonly associated with a strong seasonal cycle, the adjustment techniques are applied to model data on monthly basis, and skill scores for individual bias correction techniques are compared across individual CMIP6 models for both seasonal and annual timescales over the period 1995-2014. Our results highlight large differences among individual bias correction techniques and advocate for the use of more complex correction strategies involving corrections across the spatio-temporal distribution of the ozone field.