HYPERSPECTRAL MICROWAVE SATELLITE OBSERVATIONS: INFORMATION CONTENT AND CHANNEL SELECTION FOR NUMERICAL WEATHER PREDICTION DATA ASSIMILATION APPLICATIONS

N. Kristiansen¹, C. Harlow¹

¹Met Office,

Microwave satellite temperature and humidity sounding observations, such as those from ATMS and AMSU-A/MHS, are the most impactful observation types used in data assimilation (DA) for numerical weather prediction (NWP). Assimilating these data greatly improves the accuracy of weather forecasts.

Recently, hyperspectral microwave instruments have been developed, offering hundreds to thousands of channels in contrast to around 20 channels on current state-of-the art microwave satellite sensors. Theoretical studies have shown that increasing the spectral resolution increases the vertical resolution of the information allowing for better vertical description of atmospheric temperature and water vapour profiles.

We examine the impact of hyperspectral microwave observations in DA, using satellite observations from the Hyperspectral Microwave Sounder (HyMS) developed by Spire. This highly compact sensor has over 1600 channels across the strong oxygen absorption lines between 50-58 GHz with a narrow bandwidth of 4.8 MHz, and more than 200 channels in the water vapour band between 183-191 GHz.

We use the Degrees of Freedom for Signal (DFS) methodology to estimate the information content of simulated HyMS observations obtained with radiative transfer modelling. DFS is a linear technique that quantifies the gain in information provided by each observation. It measures the effective number of independent quantities whose uncertainty has been improved by the measurement, i.e. it seeks to minimise the error in the NWP analysis.

We rank the channels based on their DFS value to select the most informative channels for assimilation, as only a few hundred channels can be practically assimilated in operational NWP applications.

We find that from the DFS analyses for channels in the oxygen absorption band, the main impact from high spectral resolution is primarily at the upper levels of the atmosphere. This means that the full spectral resolution HyMS observations provide information on atmospheric temperature mostly at the uppermost levels of the NWP model, where the observations have largest sensitivities to temperature, and where the background errors of the NWP model are largest. Assessments of alternative spectral resolutions indicate that more information on tropospheric temperatures can be obtained with larger bandwidth configurations of 40-100 MHz. These resolutions can be obtained by combining or averaging several neighbouring channels from the full resolution HyMS observations, reducing the noise levels of each channel resulting in greater DA impact at lower atmospheric levels.

Further work includes trailing the assimilation of various sets of selected channels including the full spectral resolution and higher bandwidth selections. We will use the operational UK Met Office DA system to evaluate the impact of the HyMS observations in combination with other currently assimilated observation data, as well as evaluating potential Radio Frequency Interference issues.