Modeling the Start of Season Date of Hungarian Grasslands Using Remote Sensing Data and 10 Process-Based Models

Vegetation phenology, particularly the start of the growing season (SOS) date, is a key indicator of the climate sensitivity of ecosystems, yet its accurate prediction remains challenging. This study investigates the SOS of Hungarian grasslands between 2000 and 2023 using MODIS NDVI data, testing ten process-based models of varying complexity. Model parameters were optimized with the differential evolution algorithm under three calibration strategies: generic (GEN, aiming for a single model setting for the country), grassland-type (GEN GRASS, where grasslands are first categorized and then a type-specific parameterization is sought), and pixel-level (PIX, where model parameterization is performed for each pixel separately). The models with the lowest RMSE values were AGSI and AHSGSI (driven by temperature, vapor pressure deficit, and photoperiod) under PIX (RMSE = 3.3 days), AGSIwSW (driven by temperature, soil water content, and photoperiod) under GEN and GEN GRASS (RMSE = 7.6 and 6.3 days, respectively), and MGDDwPP (driven by temperature and photoperiod) under GEN (RMSE = 7.6 days). Considering the Akaike Information Criteria, the simplest GDD model (driven by temperature only) was the proposed one under PIX, while MGDDwPP was identified as the best model both in GEN and GEN GRASS. Residual analysis revealed relatively strong co-variation between model errors and some basic climate anomalies (most of all spring temperature and soil water content), enabling statistical corrections that reduced bias close to zero across all models. Integrating local climate and soil information into phenology models enhances their accuracy for grassland SOS estimation in Central Europe.