Optimizing the revisiting frequency of remotely sensed thermal observations for continuous estimation of ecosystem evapotranspiration and productivity using Bayesian inference

Understanding the energy, water, and carbon fluxes in dryland ecosystems is essential for maintaining ecosystem functioning and biodiversity. The limited in-situ measurements in drylands pose a significant challenge to the accurate monitoring and modelling of ecosystem dynamics. Satellite remote sensing provides high potential to monitor key surface and carbon variables, such as land surface temperature (LST), evapotranspiration (ET) and gross primary productivity (GPP). Although these data provide valuable insights, their temporal resolution is limited to satellite revisit overpasses, which can limit the continuity of monitoring. To address these gaps, dynamic land surface models serve as effective tools for integrating sparse remote sensing observations with continuous simulations of energy, water, and carbon cycles. The Soil-Vegetation-atmosphere Energy, water, and CO₂ traNsfer (SVEN) model exemplifies this approach, offering high temporal resolution simulations that incorporate satellite-based LST and meteorological in-situ inputs. This study focuses on calibrating and validating the model in southeastern Spain, as the only sub-desertic protected area in Europe. Calibration of SVEN was achieved using a combination of MODIS remote sensing data and in-situ LST measurements from an eddy covariance system, ensuring robust parameterization tailored to local field characteristics. Furthermore, the model was validated with in situ measurements, obtained through an eddy covariance tower. The RMSE values for the land surface temperature, latent heat flux, net radiation, sensible heat flux, gross primary productivity, and soil moisture were 1.99 ºC, 25.97 W m^-2, 52.71 W m^-2, 50.90 W m^-2, 1.44 gCm^-2s^-1 and 1.19 m³m^-3, respectively at half-hourly time scale. Normalized root mean square deviations of the simulated values were 7.84%, 10.81%, 5.67%, 7.81%, 13.09% and 6.59%, respectively. Otherwise, it was observed that until 8 days of revisit frequency, the calibration parameters did not affect the model accuracy considerably, increasing the RMSE of variables by 0.42 to 10.53% at the half-hourly time scale. The model’s accuracy across energy, water, and carbon fluxes highlights its potential as a reliable tool for dryland monitoring, offering insights into processes that are critical for ecological management and climate adaptation strategies. By filling the temporal gap between satellite observations, this work demonstrates the value of dynamic models like SVEN in enhancing our understanding of dryland ecosystems and promoting sustainable management practices in water-limited environments. This publication is supported by the EU COST (European Cooperation in Science and Technology) Action CA22136 “Pan-European Network of Green Deal Agriculture and Forestry Earth Observation Science” (PANGEOS).