Conditions under which CNOP sensitivity is valid for tropical cyclone adaptive observations

To determine whether profound improvements in tropical cyclone (TC) forecasting are achievable by deploying dropwindsondes according to conditional nonlinear optimal perturbation (CNOP) sensitivity, observing system simulated experiments (OSSEs) were conducted on 20 TCs that developed over the western North Pacific during 2010 using Mesoscale Model 5 and its 3DVar assimilation system. Of the 20 cases, 13 showed neutral or improved track forecasts of between 0% and 51.2%. Eliminating initial errors within the CNOP pattern, which are related to either the storm directly or the surrounding regimes indirectly, reduced the subsequent track forecast errors. The remaining 7 TCs showed deterioration in the accuracy of the track forecasts over the 48 h forecast period. Accurate forecasts made without adaptive observations, a low sensitivity of forecast errors to initial errors, or major forecast errors associated with regimes other than the TC, can lead to a decline in the accuracy of TC track forecasts. Following analysis of the potential causes of inaccuracy in the track forecasts, we find that TC cases with profound positive effects on track forecasts often satisfy the following four conditions: (i) an inaccurate initial forecast without additional observation data; (ii) proper sensitivity of the forecast errors to the initial errors; (iii) a large proportion of the forecast errors fall within the verification region; and (iv) the TC system is the dominant regime in the verification region at verification time. Seven TCs satisfied these four conditions, and showed a mean reduction of 28.75% in track forecast errors over periods of 12-48 h. This result suggests that the TC cases satisfying these four conditions often show profound improvements on track forecast by dropwindsondes guided by CNOP sensitivity.