A new ESMValTool metric using point source observations from AeroNET

ESMValTool is an open-source community-developed diagnostics and performance metrics tool for the evaluation and analysis of Earth System Models (ESMs). Key to model evaluation with ESMValTool is the use of observational data, which must comply with Climate Model Output Rewriter (CMOR) standards. ESMValTool provides methods to generate CMOR compliant datasets, but these are designed to process gridded observational data sets, such as those from satellites, and currently it is more difficult to develop point source datasets. Here we present a new ESMValTool metric for evaluating model output against an observation-based climatology of aerosol optical depth (AOD) from the Aerosol Robotic Network (AeroNET). This metric includes a downloader and formatter to generate CMOR compliant datasets for the observational AOD timeseries from all AeroNET stations. These are collated into a single NetCDF file. A new ESMValTool recipe and diagnostic process and evaluate the model output against the observational AOD dataset at model grid cells co-located with the AeroNET stations. Model output is processed in the recipe using available pre-processers to generate multi-annual seasonal means. The observational AOD timeseries from the AeroNET stations are processed in the diagnostic to generate multi-annual seasonal means, or climatologies. Because the AOD timeseries from the AeroNET stations can be incomplete, filtering criteria are applied to the data from each station to ensure sufficient temporal coverage according to the users requirements. We evaluate AOD at 440mn simulated by CMIP6 historical ensemble members against the AOD climatologies using the new ESMValTool metric. We also demonstrate how changing the filtering criteria can modify the observational climatologies, and thus the evaluation metrics. The new method extends atmospheric composition evaluation in the ESMValTool framework by adding a key aerosol metric. We hope that the techniques used to develop this metric can be applied to other point source observation datasets.