Satellite Data Assimilation for Numerical Weather Prediction (NWP)

Specifically, I will talk about:

1. The global observing system is heavily biased towards mass observations whereas vertically resolved wind information is very limited. The Aeolus satellite mission with the first wind lidar launched in 2018 made a significant step to overcome this limitation. In line with theoretical arguments, the satellite had the highest impact per satellite in the ECMWF global NWP system and our recent study shows that these improvements even translate into better precipitation forecasts. 
2. Ensemble sensitivity analysis (ESA) has been an important tool for estimating the potential impact of observations in view of planning the global observing system and potential weather control. Traditional ESA, however, is ignoring that error covariances are variable-dependent with longe-range correlations of smooth atmospheric fields (e.g. temperature) and short-range correlations for highly variable fields (humidity and hydrometeors). Based on our joint 1000-member ensemble simulations, we propose a refined approach that avoids this limitation.
3. Regional convection-permitting NWP demands for high-resolution information on atmospheric convection. Cloud-affected visible and infrared satellite radiances can provide such information with global coverage, but these observations have not been used by operational regional data assimilation systems until recently. In idealized simulation of deep convection, we demonstrated the potential impact of visible and infrared satellite radiances and discuss the effects of non-linear observation operators.
4. The first key ingredient for the assimilation of cloud-affected infrared radiances is the mitigation of the non-Gaussian distribution of departures by cloud-dependent error models. While such error models have been around for more than a decade, our work shows that a refined approach leads to a more Gaussian distribution without additional computational cost and without loosing data with a so-called quality control.
5. Another key ingredient for satellite data assimilation in ensemble Kalman filters is vertical localization. Based on our joint 1000-member simulation, we show that localization should be variable-dependent and potentially needs to be multi-model due to distant radiative effects of clouds on the model boundary layer. 
6. A general key ingredient for the assimilation of observations is a reasonable model representation of the observed quantities. Estimating uncertain model parameters together with the state via parameter estimation has been proposed to mitigate model deficiencies a long time ago, but operational success has been very limited. In our recent study, we discuss necessary conditions for the success of parameter estimation and ongoing work intends to improve model parameterizations of the atmospheric boundary layer and the effects of subgrid-scale orography.