The optimal density of atmospheric sounder observations in the Met Office NWP system

Large numbers of satellite observations are discarded from the numerical weather prediction (NWP) process because high‐density observations may have a negative impact on the analysis. In current assimilation schemes, the observation error covariance matrix R is usually represented as a diagonal matrix, which assumes there are no correlations in the observation errors and that each observation is an independent piece of information. This is not the case when there are strong error correlations and this can lead to a degraded analysis. The experiments conducted in this study were designed to identify the optimal density and to determine if there were circumstances when exceeding this density might be beneficial to forecast skill. The global optimal separation distance of Advanced TIROS Operational Vertical Sounder (ATOVS) observations was identified by comparing global forecast errors produced using different densities of ATOVS. The global average of the absolute forecast error produced by each different density was found for a 3‐week period from December 2004 to January 2005. The results showed that, when using the Met Office NWP system with a horizontal model resolution of ∼60 km, the lowest global forecast errors were produced when using separation distances of 115–154 km. However, localized regions of the atmosphere containing large gradients such as frontal regions may benefit from thinning distances as small as 40 km and therefore the global optimal separation distance is not necessarily applicable in these circumstances.