Assimilation of the SCATSAR-SWI with SURFEX: Impact of local observation errors in Austria

TL;DR

This study integrates satellite-derived soil moisture data into a surface model to improve soil moisture estimation and assesses its impact on weather prediction accuracy in Austria, highlighting the importance of local error characterization.

Contribution

It develops a data assimilation approach using local error estimates for soil moisture data and evaluates its effect on weather forecasts in Austria.

Findings

Slightly positive to neutral impact on atmospheric forecasts.

Degradation of soil moisture analysis accuracy with small observation errors.

Use of local error variances influences assimilation performance.

Abstract

The proper determination of soil moisture on different scales is important for applications in a variety of fields. We aim to develop a high-level soil moisture product with high temporal and spatial resolution by assimilating the multilayer soil moisture product SCATSAR-SWI (Scatterometer Synthetic Aperture Radar Soil Water Index) into the surface model SURFEX. In addition, we probe the impact of the findings on the Numerical Weather Prediction (NWP) in Austria. The data assimilation system consists of the NWP model AROME and the SURFEX Offline Data Assimilation, which provide atmospheric forcing and soil moisture fields as mutual input. To address the known sensitivity of the employed simplified Extended Kalman Filter to the specification of errors, we compute the observation error variances of the SCATSAR-SWI locally using Triple Collocation Analysis and implement them into the assimilation system. The verification of the forecasted 2 m temperature and relative humidity against measurements of Austrian weather stations shows that the actual impact of the local error approach on the atmospheric forecast is slightly positive to neutral compared to the standard error approach, depending on the time of the year. The direct verification of the soil moisture analysis against a gridded water balance product reveals a degradation of the unbiased root mean square error for small observation errors.