Using artificial neural network to better evaluate surface turbulent heat fluxes in weather and climate numerical models

<p><span>The </span><span>s</span><span>urface turbulent fluxes, namely sensible and latent heat fluxes, are keys </span><span>factors</span> <span>governing the boundary layer processes. </span><span>Therefore, </span><span>their </span><span>correct </span><span>representation</span><span> in numerical models is </span><span>crucial</span><span> for accurate weather forecasts and climate </span><span>projections</span><span>. However, the </span><span>formulation</span><span> of these fluxes in </span><span>such</span><span> models is the second source of uncertainty, </span><span>leading to incorrect surface-atmosphere interactions in the simulations</span><span>. Model evaluation is essential to draw development perspectives. Existing method</span><span>s</span><span> mostly consist of direct comparison between observed and modelled fluxes, blending other sources of errors such as incoherent grid-scale </span><span>representation (soil and vegetation types)</span><span> and inaccurate environmental forcing (radiative fluxes, temperature, </span><span>moisture</span><span> and wind speed). Thus, </span><span>q</span><span>uantifying errors solely due to fluxes formulation </span><span>is</span><span> still challenging. This study, within the framework of the French project MOSAI (Model and Observation for Surface-Atmosphere Interactions), aims at </span><span>proposing</span><span> a novel evaluation approach to </span><span>better</span><span> identify the weakness of numerical models </span><span>in </span><span>surface turbulent fluxes formulation. </span><span>T</span><span>he concept is to freeze the errors due to other sources </span><span>by</span> <span>using</span><span> a </span><span>machine-learning</span><span> model, notably </span><span>a multi-layer perceptron, trained to estimate the fluxes from variables describing the conditions in the surface layer. Hourly data collected over several years at three operational instrumented sites of ACTRIS-France research infrastructure (SIRTA in Paris, M&#233;t&#233;o-pole in Toulouse and P2OA in Lannemezan), are used. Then, after adaptation to the outputs of numerical models involved in </span><span>the </span><span>MOSAI project (RegIPSL, LMDZ, AROME and ARPEGE), the trained-perception will be applied to assess their surface turbulent fluxes.</span></p>