Incremental typhoon-position-oriented sensitivity analysis for targeted observation

Several techniques have been previously developed to identify the regions where additional observations are required to reduce errors in typhoon track forecasts. However, contemporary methods detect the sensitivity of an "ad hoc" metric relevant to the typhoon motion such as total energy norm or wind speed integrated over some prescribed region with vertically-uniform weighting rather than typhoon location itself. Therefore, the refinement of objective method without ambiguity is one of the major issues on adaptive observation research. In this study, a new sensitivity analysis is proposed in which a typhoon-location itself is taken as a metric. This sensitivity is interpreted as the slope of regression line (or its approximate) between typhoon-location and state variables based on ensemble run as shown in the Fig. 1. In addition to its clear objectivity, this technique has several advantages over other methods such as applicability to longer time scale prediction and finite-amplitude perturbations even when the distribution of location has bimodal peaks. Here, an incremental approach is employed to detect the synoptic sensitivity features as in an adjoint-based data assimilation method because the degree-of-freedom in the base model far exceeds the executable number of ensemble simulations. This Incremental Typhoon-position-Oriented Sensitivity Analysis (ITO-SAn) is validated for the case of Typhoon Shanshan (2006). The sensitivity field of typhoon central latitude at 48 hour after the initial time (0000Z on 15 September 2006) with respect to vorticity field is characterized by the swirling pair pattern centered at the initial typhoon location (Fig. 2). The signals in ITO-SAn are maximized in the middle troposphere and far stronger than that with respect to divergence field. Further experiments ensure that the sensitivity signal is a quantitative measure to evaluate the changes in the ensemble-mean typhoon location by adding increments. Although the horizontal distribution is similar to that obtained from the method of adjoint-derived sensitivity steering vector (ADSSV) at this verification time, ITO-SAn is likely to work well in terms of appropriate weighting in the vertical coordinate and suppression of spurious signals in ADSSV.