D7.3: Report on projected changes in global primary productivity and assessment of impact on global and WA fisheries

About this document This report presents projections of changes in global and regional marine primary productivity, with a particular focus on the Canary Current Upwelling System and its implications for West African fisheries. It utilises storm-resolving Earth System Model (SR-ESM) simulations and AI-based reconstructions to assess ecological trends and inform adaptive fisheries management in the context of climate change. Work package in charge WP7 Executive summary Observations indicate a significant sea surface temperature (SST) warming off the northwest African coast (1995-2015), the highest worldwide for the tropical belt. This warming has led to a redistribution of key fish stocks along the coast, posing a risk to food security. An original statistical analysis using Generalised Additive Models (GAMs), built on 20 years of biological acoustic data, was then developed for small pelagic fish schools and micronekton to identify key environmental data that induce changes in their spatial distribution and relative biomasses. All significant GAMs combining SST, upwelling index and Sea surface Chlorophyll-a concentration (SSC) were considered. Superimposed on the long-term warming, we show that part of the year-to-year variability gives rise to Dakar Niños. We demonstrate these impacts on SSC (here used as a proxy of primary productivity) and show that the nextGEMS model (hereafter referred to as IFS-FESOM) can adequately reproduce key features, such as interannual variability. The latitudinal pattern of the warm SST is correctly reproduced, which is attributed to the realistically resolved Senegal-Mauritania upwelling and Frontal Zone by IFS-FESOM. This is a notable strength of the IFS-FESOM compared to CMIP6 and High-ResMIP models, which have difficulties in resolving coastal upwelling. A key aspect was to develop approaches based on IFS-FESOM simulations, which were primarily limited to physical parameters, in order to assess the benefits of resolving finescale dynamics and predicting future changes in marine ecosystems. We first developed a machine learning approach, based on UNet architecture, to predict SSC from physical parameters, including SST, sea surface salinity, and sea surface height. The model was trained and validated using observational and reanalysis data for independent periods. At the global scale, SSC is projected by most climate models to remain relatively stable through mid-century, though with a wide spread, from weak declines to slight increases, impeding their robustness. IFS-FESOM-based SSC reconstruction shows a near-zero trend and low interannual variability. Yet, IFS-FESOM also provides enhanced spatial resolution in coastal and frontal zones, capturing biologically meaningful patterns that are often missed in coarse-resolution models. In the West African region, model projections of SSC showed greater spatial complexity and variability than from large-scale trends. The IFS-FESOM-based SSC reconstruction indicates a near-zero trend in spatially averaged productivity through 2050, aligning with the majority of CMIP6 models, despite a wide inter-model spread. However, the spatial patterns differ significantly: IFS-FESOM projects localised increases in SSC along the coast, particularly in upwelling zones, while CMIP6 models display a wide range of trends. Even if there is no evidence that patterns from IFS-FESOM are more robust in terms of future SSC trends, this highlights the added value of storm-resolving models in capturing fine-scale processes essential to understanding regional ecosystem responses. Projections based on IFS-FESOM simulations suggest limited changes in biological features across the West African upwelling system by mid-century, despite a high level of uncertainty. Our model-based forecast for Sardinella aurita (a small pelagic fish of key socioeconomic importance in West Africa) does not reproduce the northward shift observed over the past two decades, reflecting a divergence between recent historical trends and longer-term climate projections; nevertheless, the projected northern limit of the stock by 2050 remains consistent with recent observations. For fish school biomass, the key target of fisheries, simulations revealed strong sensitivity to environmental drivers, particularly Ekman transport and SSC, and a dramatic decrease, which could be due to a lack of ecosystemic consideration in our approach. Regarding micronekton, the IFS-FESOM-based projections indicate a slight reduction trend in biomass, but with high uncertainty due to limitations in resolving complex trophic and physical processes. These results illustrate both the potential and the current limits of using storm-resolving climate models for forecasting biological responses in complex coastal ecosystems. Nonetheless, IFS-FESOM performs better than CMIP6/HighResMip in reproducing the SST front along NW Africa and associated extreme warming events, showing the evident benefit of high resolution. Nonetheless, IFS-FESOM performs better than CMIP6/HighResMIP in reproducing the SST front along NW Africa and associated extreme warming events, showing the evident benefit of high resolution. Additionally, the deliverable extends beyond its original scope by incorporating a complementary evaluation using the ICON-O/HAMOCC coupled model, which includes ocean biogeochemistry. This work has revealed that a realistic fine-scale spatial pattern emerges in Northwest Africa from the ICON-O/HAMOCC global simulation (10 km) within the complex Canary East border Upwelling system. While nextGEMS models do not directly include biogeochemistry modules, they enable SSC reconstruction via deep learning approaches using physical drivers. This is a new capability that is currently out of reach for most CMIP6 outputs due to spatiotemporal mismatches and limited spatial and temporal resolution. These results illustrate the potential and also some current limitations of using storm-resolving climate models for forecasting biological responses in complex coastal ecosystems. This deliverable has enabled the development of a forecasting framework specifically suitable for West Africa, a data-poor region where such predictive tools remain scarce.