Optimization of the calculation of the photodissociation rates in the stratosphere in the BASCOE module of the IFS-COMPO

The Integrated Forecasting System (IFS-COMPO) at ECMWF is the global model used by the Copernicus Atmosphere Monitoring Service (CAMS) to provide forecasts and analyses of atmospheric composition, including greenhouse gases, aerosols and reactive gases. Within CAMS, the default configuration of IFS-COMPO relies on the CB05 chemistry scheme for the troposphere and the Belgian Assimilation System for Chemical ObsErvations (BASCOE) module for the stratospheric chemistry.

Currently, the photodissociation rates (Js) in IFS-COMPO can be computed with two methods: Joffline (computationally efficient but inaccurate) and Jonline (more accurate but computationally very expensive). In this work, we implement and evaluate an optimized method for the calculations of the Js in the stratosphere that is as accurate as Jonline and as computationally efficient as Joffline. This method (Jvint) computes the Js using Jonline but on a coarser stratospheric vertical grid (L23) compared to the native vertical grid and will be available in the future cycle 49R1.

We compare three IFS-COMPO simulations using the Jvint, Jonline and Joffline methods with the BASCOE Reanalysis of Aura MLS version 3 (BRAM3) for ozone and with observations from the TROPOspheric Monitoring Instrument (TROPOMI) for nitrous dioxide (NO2). For ozone, the Jvint method significantly reduces the bias with respect to BRAM3 compared to the Joffline method in the upper stratosphere. In the mid-lower stratosphere, however, the Jvint configuration delivers slightly larger bias with respect to BRAM3 when compared to the Joffline method, especially over the Arctic. This bias is consistent with the results from the Jonline method and with the bias reduction in the upper stratosphere. For NO2, the Jvint method also reduces the bias with respect to TROPOMI stratospheric columns compared to the Joffline method at all latitudes.

Concerning the computational cost, it increases in the Jvint method compared to the Joffline method by 5% in December, 16% in July and 12% in November, while the Jonline method is two to three times as computationally expensive compared to Joffline. In addition, the Jvint method improves the seasonal variations of the ozone and NO2 columns compared to the Joffline method, and the differences between the Jvint and Jonline methods are insignificant for most of the stratospheric regions.

We demonstrate the capability of the Jvint method to deliver improved stratospheric columns for ozone and NO2 compared to the Joffline method and significantly reduce the computational cost compared to the Jonline method. The Jvint configuration also includes the time-dependence of the solar flux, ensuring a physically consistent representation of the stratospheric photochemistry.