The assimilation of surface observations in mountainous terrain in the WRFDA system

Assimilating surface observations in convective scale data assimilation (DA) systems is not straightforward, since these observations may be affected by small-scale effects not represented in the model, and the model itself might not be able to accurately represent the features of the atmosphere close to the surface. These issues are particularly evident in mountainous terrain. In variational DA systems, such as the Weather Research and Forecasting, Data Assimilation (WRFDA) suite, the available background error (BE) models produce BE variances and covariances that vary smoothly over long distances. Therefore, for instance, assimilating a valley-floor surface observation typically leads to large analysis increments even at nearby mountain tops, which are physically unwarranted and cause high levels of gravity-wave noise. 

Such problems can be partially mitigated in WRFDA by modeling the BE using the Alpha Control Variable Transform (Alpha CVT). 

Like other BE models in WRFDA, this technique derives BE statistics from an ensemble of differences between forecasts with different initial and identical valid times (NMC method), and it makes use of a control variable transform (CVT). Differently from other BE models in WRFDA, it computes analysis increments as a linear combination of the NMC ensemble members.

In this work we consider simulations with a grid spacing of 3.5 km over a domain encompassing the European Alps. We first use pseudo-observation tests to show how different BE specifications in WRFDA affect the assimilation of surface observations of temperature, specific humidity, pressure and horizontal winds components in complex terrain.

We then present real-case assimilation experiments with a limited set of surface observations. Considering the consistency between the variances of innovations and the assigned observation and background errors, we demonstrate the positive impact of the Alpha CVT.