Skin Temperature Analysis for the Assimilation of Clear-Sky Satellite Radiances

Radiances represent the vast majority in numbers of the data that are currently assimilated in the four dimensional variational data assimilation (4D-Var) system of the European Centre for Medium-Range Weather Forecasts (ECMWF) Integrated Forecasting System (IFS). The assimilation of radiances starts with the computation of brightness temperatures expected from the instruments with a radiative transfer code using outputs from the forecast model. These outputs are atmospheric model profiles (e.g., pressure, temperature, specific humidity) and surface parameters (e.g., skin temperature, surface pressure). To produce an optimal model surface and atmospheric state from which the simulated brightness temperatures are the best possible fit to the observed ones, the 4D-Var should be able to optimise the skin temperature together with all the other atmospheric model physical variables.

The control variable of the atmospheric analysis currently does not include surface variables, and the optimisation of the skin temperature is currently achieved by adding a skin temperature component in the 4D-Var control vector, but defined in observation space, i.e., at each observation time and location. This means that for each field of view, if needed, all the observed radiances from this field of view help adjusting a single value of skin temperature, and the optimal skin temperature value is valid only for this field of view. The adjustment is independent of the skin temperature adjustment from other fields of view. In other words, this is a zero-dimensional variational retrieval of the skin temperature at the observation location and time.

A new approach is proposed here where the skin temperature is analysed in model space. The computation of simulated radiances uses the same two-dimensional skin temperature field for all fields of view by interpolating this two-dimensional field at the observation location and time. As a consequence, the combined radiance measurements contribute to the estimation of the skin temperature field in the analysis. This makes the analysed skin temperature field consistent with all available radiance measurements.

The radiative transfer code used in the IFS to compute the radiance equivalent to observation from the model variables admits two separate spectral bands (microwave and infrared) and two different viewing geometries (geostationary and polar) for the infrared band. Sounders from different spectral bands have a different sensitivity to the skin temperature. For example, microwave sounders can penetrate more deeply into the ground layers than the infrared sounders. For this reason, we propose to have a separate skin temperature field per spectral bands (microwave and infrared). For the infrared, we also decided to separate the two viewing geometries as used in the radiative transfer code. The three separate skin temperature fields are referred to as microwave, geostationary (infrared) and (polar, hyper-spectral) infrared.

Having skin temperature fields defined at the beginning of the assimilation window and valid throughout the 12-hour window is not appropriate as the skin temperature as seen by the satellite instruments changes over time within the window, especially over land. At this stage, we do not know how to evolve in time these skin temperature fields. We have instead chosen to extend the skin temperature control vector with one skin temperature field per hour, i.e., 13 fields over the 12-hour assimilation window length, per instrument type.

This paper describes the technical and scientific developments connected to this new skin temperature analysis. In particular, we focus on the background errors of the new skin temperature fields that are introduced to the control vector. With sensitivity studies, we found some configurations of the new approach for which the skin temperature analysis for the radiance assimilation is similar to the current formulation. These configurations also produce similar analysis fit to observations than the current formulation except for particular channels sensitive to the surface. Yet, the first guess fit to observation is not significantly different between them.

The new approach has a neutral impact on the analysis and a neutral to possitive impact on the forecast scores. This is encouraging as there is room for improvement. Moreover, the analysed skin temperature fields are byproducts that could help future developments of the skin temperature model or future developments of the radiance observation operator.