Exploring the use of ensemble Kalman filtering for the assimilation of local observations in a continuous single-column model evaluation at the ARM sites
Authors
Roel Neggers — University of Cologne
Peter Baas — Royal Netherlands Meteorological Institute
Category
Atmospheric State & Surface
Description
Composite hodograph of the nocturnal 200-m wind for one month of SCM simulations (July 1999) for the SGP Central Facility site. The full lines indicate various SCM versions; the dashed line indicates averaged observations of the 915-MHz wind profiler. Numbers indicate local time. While still underestimating the amplitude of the LLJ, the TKE scheme clearly performs better than the REF version. In this case, the assimilation of near-surface data does not lead to improved simulations. However, since both model versions still develop their own unique state, this result suggests that the enKF method can be used as an alternative to commonly applied relaxation techniques.
Single-column models (SCMs) have become a widespread and successful tool to evaluate model parameterizations. An SCM simulation consists of the offline integration of all sub-grid physical processes from a general circulation model, using prescribed forcings and boundary conditions. Generally, the purpose is to gain insight in the behavior of the physics and to highlight deficiencies of the parameterizations.
However, the prescribed large-scale forcings always carry uncertainties. This potentially hampers an evaluation of SCM results against observations. Therefore, in SCM modeling, relaxation is often applied towards a “true” state that can be either observations or model products. Relaxation prevents excessive model drift, while still allowing the physics to develop its own unique state. In this way, a valid comparison of SCM results with observations remains possible.
A possible alternative to relaxation is the assimilation of local measurements. As such, in the present study we explore the use of an ensemble Kalman filter (enKF). Here, we study the impact of assimilating 10-m observations of temperature, specific humidity, and both components of the horizontal wind obtained from the ARM archive of SGP measurements. Additionally, we assimilate soil temperature and soil moisture as provided by ERA-interim.
We utilized the SCM version of the Integrated Forecasting System (IFS) of the European Centre for Medium-Range Weather Forecasts (ECMWF). In this study we compare two model versions. One applies a first-order closure formulation of turbulent diffusion (IFS CY31r1, REF); the other employs a TKE-closure (TKE) formulation. As large-scale forcings, we use data from the ARM SGP variational analysis and ERA-interim.
The purpose of our study is twofold. Firstly, we investigate which of the two model versions gives a better representation of momentum mixing. A comparison of one year of SCM simulations with observations from a 915-MHz wind profiler shows that TKE gives a much better representation of the wind, in particular for the nighttime hours. Among others, this is reflected in a better representation of the nocturnal low-level jet.
Secondly, we study the impact of the enKF. During daytime the influence is much stronger than during nighttime. Compared to independent atmospheric sounding data, the assimilation of near-surface observations gives a considerable reduction of the rms errors of temperature and relative humidity.
