Assimilating ESA-CCI Land Surface Temperature into the ORCHIDEE Land Surface Model: Insights from a multi-site study across Europe

Abstract. Land surface temperature (LST) plays an essential role in water and energy exchanges between Earth’s surface and atmosphere. Recent advancements in high-quality satellite-derived LST data and land data assimilation systems present a unique opportunity to bridge the gap between global observational data and land surface models (LSMs) to better constrain the water/energy budgets in a changing climate. In this vein, this study focuses on the assimilation of the ESA-CCI LST product into the ORCHIDEE LSM (the continental part of the IPSL Earth System Model) with the aim of optimizing key parameters to improve the simulations of LST and surface energy fluxes. We use the land data assimilation system for the ORCHIDEE model (ORCHIDAS) to conduct a series of synthetic twin data assimilation experiments accounting for real data availability and uncertainty from ESA-CCI LST to find an optimal strategy for assimilating LST. Here, we test different strategies of assimilation, notably investigating: i) two optimization methods (random-search and gradient-based) and ii) different ways to assimilate LST using the only raw data and/or different characteristics of the LST diurnal cycle (e.g. mean daily, daily amplitude, maximum and minimum temperatures, morning and afternoon gradients). Upon identifying the optimal approach, we use it to assimilate ESA-CCI LST data across 34 sites in Europe provided by the WarmWinter database. Our results demonstrate the effectiveness of assimilating 3-hourly CCI-LST data over a single year in 2018, improving the accuracy of simulated LST and fluxes. This improvement, assessed against CCI-LST and in situ observations, reaches up to 60 % reduction in root mean square deviation, with a median reduction of 20 % over the entire validation period (2009–2020). Furthermore, we evaluate the effectiveness of optimized parameters for application at larger scales by using the median of optimized parameters per vegetation type across sites. Notably, the performances for both LST and fluxes exhibit not only consistent stability over the years, comparable to using site-specific parameters, but also indicate a significant improvement in the modeled fluxes. Future works will be focused on refining the utilization of the observation uncertainties provided by the ESA-CCI LST product (e.g. decomposed uncertainties and spatio-temporal variability) in the assimilation process.