Assimilation of land‐surface temperature observations in a 2DEnVar surface data assimilation system within AROME‐France

Abstract The assimilation of land‐surface temperature (LST) observations from satellite instruments is of major interest to represent the land‐surface initial state more accurately in numerical weather prediction (NWP) systems. Current activities at Météo‐France in the limited‐area Application of Research to Operations at Mesoscale (AROME‐France) model are dedicated to the assimilation of LST observations, thanks to an ensemble‐based method recently developed for surface data assimilation (DA). The developments consist of assimilating the LST observations into the two‐dimensional ensemble‐based variational (2DEnVar) screen‐level scheme (2DEnVar‐LST configuration), as has already been performed for 2‐m temperature () and relative humidity (). This leads to the extension of the control vector and the representation of cross‐correlations between these three variables in the background‐error covariance matrix of the day. The LST and 2‐m temperature analysis increments produced by this system are then used with equal weights as “pseudo‐observations” in the soil DA system to initialise soil temperature. The evaluation of 2DEnVar‐LST against 2DEnVar (with no assimilation of LST observations) shows a positive impact on forecasts near the surface and in the lower atmosphere up to 700 hPa. This result is due mainly to the representation of relevant cross‐covariances between , , and LST in the flow‐dependent ensemble matrix. The ensemble‐based 2DEnVar system appears to be robust and reliable for the assimilation of new types of observations from satellites.