From long-term datasets, the variability of the disparities between models and observations can be assessed. However, discrepancies between models and observations may arise from observation uncertainty/representativeness, from atmospheric conditions (e.g, wind, radiative fluxes, air temperature, humidity) resulting from both the forcing datasets and from the parameterizations of boundary-layer and convective processes (e.g., Santanello et al., 2009 ; Sun et al., 2017), from soil conditions (e.g., soil water availability, root depth – e.g., Wang and Dickinson, 2012) and of course from imperfect parameterizations of surface fluxes. The systematic and direct comparison of absolute values cannot disentangle the various contributions embedded in the model-observations departure and makes the evaluation less useful in improving models. Two methods will be tested in MOSAI to go further than point-to-point, time-to-time or case-study comparisons.
The first approach is based on sensitivity studies performed with 3D models or with their corresponding single-column version, either forced by MOSAI EOP observations or coupled with their LSM, and for which an atmospheric forcing will be derived from operational analyses. The development and assessment effort (Hourdin et al. 2020, Cheruy et al, 2020) that has been undertaken during the preparation of the IPSL climate model for the CMIP6 exercise led to better identify and emphasize a critical step in the model development: the adjustment or tuning stage of the free parameters of the parameterizations which, whatever their degree of refinement, are only an idealized and imperfect representation of complex processes (Hourdin et al., 2017). MOSAI project will therefore also target the tuning of the free parameters involved in the parameterization of energy and water surface fluxes.
The second approach relies on the method proposed by Zhou and Wang (2016), which is based on correlation coefficients and sensitivity parameters between surface fluxes, net radiation and other environmental parameters to evaluate ERA-Interim surface fluxes. This approach allowed to clearly identify weaknesses of the reanalysis over specific land-cover types and offered a guidance to improve the model. This method has not yet been generalized to NWP and climate models. MOSAI will apply this method to several models, and will go forward by (i) taking into account the observation uncertainty/representativeness issues (O1) and (ii) using more sophisticated statistical methods (e.g., random forests methodology) than correlation and linear regression to test the dependency of surface fluxes to several variables at the same time, and not only to each variable individually. It will allow to identify specific weaknesses of each model (O2, and useful for O3) at different spatial and time scales.