This paper demonstrates the efficacy of data-driven localization mappings forassimilating satellite-like observations in a dynamical system of intermediatecomplexity. In particular, a sparse network of synthetic brightness temperaturemeasurements is simulated using an idealized radiative transfer model andassimilated to the monsoon-Hadley multicloud model, a nonlinear stochasticmodel containing several thousands of model coordinates. A serial ensembleKalman filter is implemented in which the empirical correlation statistics areimproved using localization maps obtained from a supervised learning algorithm.The impact of the localization mappings is assessed in perfect model observingsystem simulation experiments (OSSEs) as well as in the presence of modelerrors resulting from the misspecification of key convective closureparameters. In perfect model OSSEs, the localization mappings that use adjacentcorrelations to improve the correlation estimated from small ensemble sizesproduce robust accurate analysis estimates. In the presence of model error, thefilter skills of the localization maps trained on perfect and imperfect modeldata are comparable.
展开▼
