Abstract
<jats:p>Land surface models (LSMs) are fundamental for simulating coupled carbon, water and energy exchanges in Earth system models, yet their performance remains poorly constrained across Africa’s diverse hydroclimatic gradients. We evaluate the Joint UK Land Environment Simulator (JULES) using eddy covariance observations from 16 flux towers spanning major African ecosystem types. Model skill is assessed for gross primary productivity (GPP), ecosystem respiration (Reco), and evapotranspiration (ET), and we examine how model bias and error vary along gradients of mean annual temperature, aridity index, and precipitation anomalies. JULES reproduced broad seasonal dynamics of carbon and water fluxes across several ecosystems but showed systematic limitations in simulating Reco and interannual variability of GPP. Model performance differed among ecosystem types, with wetlands exhibiting large model-observation mismatches and cropland sites showing GPP overestimation linked to excessive simulated leaf area index. To isolate environmental controls on performance, we applied linear mixed-effects modelling to partition site-level variability from climate-driven effects. Aridity and precipitation anomalies emerged as dominant predictors of model bias and error, with wetter and anomalously humid conditions associated with larger deviations from observations, indicating limitations in representing soil hydrology and vegetation water stress processes. By integrating an expanded African flux tower network, this study provides a comprehensive regional evaluation of JULES. The results identify hydroclimatic regimes where model development should be prioritised and emphasise the importance of strengthening benchmarking across underrepresented African ecosystems to improve model transferability and support reliable climate and land management decision support under environmental change.</jats:p> <jats:p> <jats:bold>Plain Language Summary</jats:bold> : Land surface models are computer programs that help scientists understand how land ecosystems exchange water and carbon with the atmosphere. They are widely used to study climate change and support land and water management. However, most testing of these models has been carried out in regions with extensive data, and their performance across Africa’s varied ecosystems is less well understood. In this study, we evaluated the Joint UK Land Environment Simulator using measurements from 16 eddy-covariance flux towers across Africa. These towers directly measure how much carbon ecosystems absorb through photosynthesis and release through respiration, and how much water returns to the atmosphere through evapotranspiration. We compared model simulations with these observations across a range of climates, from dry to wet regions and from cooler to warmer sites. JULES successfully reproduced seasonal patterns of carbon and water exchange at many locations. However, it struggled to accurately represent ecosystem respiration and year-to-year variation in photosynthesis. Using a statistical approach that accounts for differences between sites, we found that dryness and unusual rainfall conditions were the main causes of model errors. These results highlight the need to improve how this model represents soil water processes and vegetation responses to water availability. </jats:p>