Main land research work in WP2 - Digital solutions
Motivation: ECMWF wil expand its aiLand land surface machine learning model to provide an alternative prior for both inversions systems and for full AI-system (including atmosphere, based on the AIFS system). This will involve dataset preparation for training and inference runs to assess the capabilities of aiLand to provide consistent carbon simulation for LAI/biomass and CO2 and CH4 fluxes at hourly frequency. In a first step, aiLand will be adapted to provide consistent output trained on and compared with ecLand EO bias-corrected simulations. In a second step, the best adapted initialization and data assimilation strategy for the expanded aiLand will be explored on a GST (Global Stocktake) year (GST-1 and/or GST-2). This step links strongly to the CONCERTO, CORSO, and CHERRI projects. Nature runs connected with surface fluxes produced within this task will be performed to provide data to WP3 and WP4 and for testing synthetic training datasets in aiLand.
Benefits: ecLand will become a complete land energy, water, carbon emulator for both offline and atmospheric coupled simulations at a reduced computational cost, therefore enabling faster experimentak throughput and high resolution/ensemble applications.