Reliable predictions of forest ecosystem functioning require flawless climate forcings

International audience; Hightlights:• Predictions of physiological process depends on climate model, species and region.• Predictions were improved after correction for the three models considered.• Processes simulated exhibited large variability at the plot scale.• This variability faded out at larger scales, owing to an aggregation effect.• Process predictions were more variable during the driest years.Abstract: Climate change affects various aspects of ecosystem functioning, especially photosynthesis, respiration and carbon storage. We need accurate modelling approaches (impact models) to simulate forest functioning and vitality in a warmer world so that forest models can estimate multiple changes in ecosystem service provisions (e. g., productivity and carbon storage) and test management strategies to promote forest resilience.Here, we aimed to quantify the bias in these models, addressing three questions: (1) Do the predictions of impact models vary when forcing them with different climate models, and how do the predictions differ under climate model vs. observational climate forcings? (2) Does the climate impact simulation variability caused by climate forcings fade out at large spatial scales? (3) How does using simulated climate data affect process-based model predictions in stressful drought events? To answer these questions, we present historical results for 1960-2010 from the CASTANEA ecophysiological forest model and use the data from three climate models. Our analysis focuses on monospecific stands of European beech (Fagus sylvatica), temperate deciduous oaks (Quercus robur and Q. petraea), Scots pine (Pinus sylvestris) and spruce (Picea abies) in French forests.We show that prediction of photosynthesis, respiration and wood growth highly depends on the climate model used and species and region considered. Predictions were improved after a monthly mean bias or monthly quantile mapping correction for the three models considered. The processes simulated by the impact model exhibited large variability under different climate forcings at the plot scale (i.e., a few hectares). This variability faded out at larger scales (i.e., an ecological region, 100 km(2)), owing to an aggregation effect. Moreover, process predictions obtained under different climate forcings were more variable during the driest years. These results highlight the necessity of quantifying the bias correction effect on process predictions before predicting flux dynamics with a process-based model.