Variational inversion of simulated TOVS radiances using the adjoint technique

Abstract

AbstractThe adjoint technique allows the variational approach for assimilating various types of observations in meteorology, including raw radiances measured by satellite, to be treated exactly and at a ‘reasonable’ cost.We have developed the tangent linear and adjoint operators of the International TOVS Processing Package radiative transfer equation. These two tools are used for treating two aspects of variational inversion of satellite sounding radiances.First of all, we apply this technique to the computation of the error covariance matrix of the retrieval/analysis of simulated TOVS sounding radiances. Using this matrix, an original method to interpret satellite sounding radiances in terms of temperature and humidity information is then presented. We can thus evaluate the number of independent parameters which are significantly estimated from radiances, given a direct radiative transfer model, observation and background error statistics. It is found that 19 HIRS channels provide six independent pieces of information while four MSU channels provide three independent pieces of information. When grouping HIRS and MSU channels, only seven independent pieces of information are available. It is shown that in cloudy conditions MSU channels bring additional significant information to HIRS channels especially at the surface. We also study the impact of cloudiness on the quality of the retrieved profiles.Numerical feasibility of 1‐D variational inversion using the adjoint of the International TOVS Processing Package radiative transfer model is demonstrated. Impact of the choice of the inner product on the efficiency of the method is studied. We then discuss the quality of the inversion, which turns out to be strongly related to the quality of the atmospheric information present in the sounding radiances and to the background and observation error statistics. the gain brought by MSU channels at the surface in cloudy conditions is discussed.