In this study, the sensitivity of the model 1-day forecast error to the initial conditions for an Indian summer monsoon case was applied to localize regions with large analysis uncertainties. Our results show that the model 1-day forecast error is most sensitive to errors in the analyses of the lower troposphere, especially over the northern Bay of Bengal around 90E, 20N. More moisture and wind field observations are required over this region to improve the quality of the analyses in order to ameliorate the model forecast skill.
However, this sensitivity study was performed in ``a posteriori" diagnostic way in this study. For practical problems, large forecast uncertainties can also be identified using the ensemble forecast system (see Kalnay and Toth, 1996). The forecast difference may be obtained by subtracting one member of the ensemble from another, then the localized forecast errors may be applied to the adjoint model in order to obtain the adjoint sensitivity.