Sensitivity of ECMWF’s seasonal forecast model CY49R2-NEMO4-SI3 to CO2 and anthropogenic aerosol forcings: Experimental design and impact on climate trends

Abstract. Detection and attribution studies routinely rely on climate model experiments with modified anthropogenic forcings, such as those produced within the Coupled Model Intercomparison Project (CMIP). However, recent work has shown that free-running CMIP models fail to reproduce important aspects of observed decadal trends which contain a forced component. Despite their advantage over CMIP models due to reduced mean-state biases, higher predictive skill, better representation of trends due to frequent re-initialization and large ensemble sizes, seasonal forecasting systems have not previously been used within a formal attribution framework.

Here we introduce a novel set of counterfactual seasonal hindcasts conducted with ECMWF’s coupled seasonal forecasting system using cycle 49R2 coupled to NEMO4 ocean and SI3 sea ice models (CY49R2-NEMO4-SI3). Unlike conventional hindcasts, these simulations are initialized under alternative climate-change scenarios by modifying both the atmospheric forcing and the ocean and sea-ice initial conditions. We derive an observation-based estimate of the anthropogenically forced ocean-temperature signal and either amplify or remove this signal from the three-dimensional ocean initial conditions while preserving spatial gradients. Retrospective forecasts for 1993–2023 are performed using a control configuration together with enhanced and reduced forcing experiments. They use up-to-date ocean initial conditions (ICs) and model versions closely aligned with the next operational seasonal forecasting system SEAS6 configuration.

In the enhanced forcing experiment, post-1993 CO2 increases and the anthropogenic ocean warming signal are doubled annually relative to the control. In the reduced forcing experiment, CO2 concentrations are fixed at 1993 levels and the post-1993 anthropogenic ocean warming signal is removed from the ICs. Additional sensitivity experiments aim to isolate the role of aerosol forcing.

The counterfactual hindcasts produce substantially altered long-term temperature trends while largely retaining seasonal prediction skill, interannual variability and model drift, demonstrating that the imposed perturbations are dynamically consistent with the forecasting system. Enhanced-forcing experiments strengthen several observed climate trends that are underestimated in the control simulation, including trends in top-of-atmosphere radiative fluxes and aspects of the atmospheric circulation over the tropical Pacific. However, strengthening the observed tropical Pacific zonal sea-surface-temperature gradient produces only a weak atmospheric response, leading to insufficient enhancement of equatorial easterlies and Walker-circulation trends. As a result, the coupled atmosphere–ocean system fails to sustain the imposed SST-gradient anomaly through Bjerknes feedbacks, suggesting a fundamental limitation of the model. The impact of amplified aerosol forcing is comparatively weak, likely reflecting the omission of indirect aerosol effects in the model version used.

These results demonstrate that counterfactual seasonal hindcasts provide a practical new framework for dynamical attribution studies while also offering a powerful approach for diagnosing model deficiencies in the simulated response to anthropogenic climate forcing.