Modelling phenology in European beech forests as waves of photo-thermal responses

Plant phenology, which refers to the timing of cyclic or recurring eco-physiological events in plants, provides key information about the seasonal dynamics of vegetation and ecosystem processes. While traditional phenological surveys imply the visual detection of easily observable events like flowering, bud break and defoliation, phenological patterns may be unclear due to combined effects or compensatory processes of driving variables (e.g., temperature and photoperiod). Monitoring the continuous plant development during a season is essential for interpreting temporal changes between observed phenological phases and the environmental impacts on the plants' seasonal dynamic. This is especially true for deciduous broadleaf forests, like European beech, where the canopy is sensitive to intra-annual climatic changes and extreme events like spring frosts. In this view, remotely sensed vegetation indices (VIs) observations, thanks to their high temporal resolution, provide a reference data source for investigating phenological moments (phenometrics) and trends. However, there are still two major drawbacks to be solved: (i) phenometrics are merely mathematical moments of the VI annual curve and do not represent any ecophysiological phase, (ii) the VI annual profile allows to detect the canopy growth processes of a plant community but does not reveal the dormancy-related processes occurring when the dominant vegetation cover is senescing. To fill this research gap, this work proposes a process-based model, named swell (simulated waves of energy light and life), able to simulate the complete VI intra-annual profile of beech forests based on their photo-thermal responses during the dormancy and the growing season. To this aim, the EU-forest dataset was used as spatial information of the European beech forests (data from 16 ecoregions), the MODIS NDVI (2010-2023) as reference data for swell calibration (4426 pixels) and independent evaluation (6672 pixels), and the E-OBS Copernicus dataset as weather input source. The rationale of swell is that each phenophase starts upon the completion of the previous one and progresses as a function of phenophase-specific photothermal requirements, using temperature and daylength as climatic cues. The swell simulations agreed with MODIS NDVI profiles (RMSE < 0.10, PBIAS < 5%, Pearson r > 0.8) across time and ecoregions, obtaining similar performances in calibration and evaluation and comparable performances with a fine-fitting statistical method fitted yearly (i.e., Elmore). Aggregating NDVI simulations by latitude and elevation bands allowed exploring patterns of beech phenology across Europe, whose dynamics can be inspected to reveal the vegetation response of this species in response to largely variable photothermal conditions.