The assimilation of surface observations in the European Alpine region

This dissertation presents the development, testing and refinement of an operational Numerical Weather Prediction (NWP) routine based on the Weather Research and Forecasting (WRF) model for a domain encompassing the European Alps. Surface observations from official observational networks and privately owned Automatic Weather Stations (AWS) are assimilated via the 3DVAR algorithm in this operational chain to obtain refined weather forecasts at the convective scale. The setup was developed using technical choices aimed at making the system affordable for a private meteorological service with limited computational resources. The goal is to build a timely NWP system capable of quickly and effectively assimilating observations in near real-time with a beneficial impact on forecasts, while keeping the computational load to a minimum. In a series of preliminary experiments with the developed setup, the assimilation of surface observations had mixed effects on forecast skill compared to forecasts without assimilation. Additionally, the analysis increments and innovations exhibited biases, which is suboptimal from a Data Assimilation (DA) perspective. Two approaches were tested to address these challenges. First, a combination of innovation bias correction and error covariance tuning was evaluated. Second, an alternative background error model was implemented. Both approaches resolved the detrimental effects observed following the assimilation of surface observations in the preliminary experiments, albeit with a slight reduction of the beneficial effect of the assimilation step on the forecasts. Overall, this dissertation highlights the technical challenges associated with assimilating surface observations in a low-cost NWP setup intended for operational use and offers guidance on potential strategies to address them.