Robust estimation of daily photosynthesis from instantaneous observations through machine-learning integration of radiation and environmental drivers

Remote sensing (RS) facilitates large-scale estimation of vegetation carbon and water fluxes, yet temporal mismatches persist between its instantaneous observations and the daily flux mean or sum required for ecological modeling. Traditional upscaling methods typically convert instantaneous flux observations to daily values through assuming that diurnal flux patterns are mainly driven by solar radiation, failing to capture real dynamics induced by other environmental factors (e.g., temperature and moisture). This introduces substantial errors, particularly in daily carbon flux estimation. To address this issue, we focus on gross primary production (GPP) and develop a new conversion factor model that integrates solar radiation and other key environmental drivers, enabling robust upscaling from instantaneous to daily scales. Using the FLUXNET2015 dataset, the conversion factor ${\gamma }_{env}$, defined as the ratio of instantaneous to daily GPP, was modeled using random forest, with vapor pressure deficit, soil water content, air temperature, and shortwave radiation as predictors. SHapley Additive exPlanations (SHAP) analysis was used to evaluate predictors’ contribution and response mechanisms. Results show that the proposed model outperformed traditional upscaling methods in daily GPP estimation, improving R² by up to 39% and reducing RMSE by up to 82%. Validation across diverse ecosystems, environmental stress levels, and drought conditions further confirmed its superior generalizability over conventional methods. Critically, ${\gamma }_{env}$ retained high accuracy when driven by ERA5-Land reanalysis data instead of site-level tower measurements and reliably upscaled satellite-based instantaneous GPP snapshots to daily estimates, demonstrating scalability for large-scale applications. Moreover, ${\gamma }_{env}$ effectively captured complex diurnal dynamics of vegetation photosynthesis under environmental stress, and through SHAP, revealed the growing role of water or temperature-related drivers in regulating GPP diurnal patterns as stress intensified. Overall, this study presents a structurally simple yet ecologically grounded solution to the temporal mismatch in RS-based GPP estimation, and offers valuable insights for upscaling other ecosystem fluxes.