"A method for representing spatially correlated observation errors for wind data"

The International Symposium on Data Assimilation Online (ISDA-Online) 8 Jan,2021 "Data Assimilation Methodology"
International Symposium on Data Assimilation - Online

The objective of this research project is to drastically improve the spatial density of the observations exploited in the numerical prediction systems ARPEGE and AROME. This requires accurately representing the observation error correlations. To this end, we use a technique coming from the field of oceanography, based on the solution of a diffusion equation, which we decide to apply on unstructured meshes. A first study dealing with scalar data makes use of the finite element method. It provides a way to represent horizontal error correlations for scalar data, such as brightness temperature from satellite data. Experimental validation was achieved using measurements from the infrared imager MSG/SEVIRI, which are assimilated both in ARPEGE and AROME. We propose to extend this method to the representation of wind error correlations. These are vectorial data, meaning that every observation location is associated with two values, zonal and meridional. We focus specifically on scatterometer measurements, that are available every 25km but only assimilated every 100km. First, the wind field is decomposed into one divergent component and and rotational component. Then, the scalar correlation operator is applied to each component. Finally, the wind field is reconstructed while maintaining the symmetry and the positivity of the correlation operator. Experiments show agreement with the analytical results. In the future, all types of observations will be considered, whether or not they are conventional, and we will extend the method to the three dimensional case to address the specific case of radars for instance.