Predictability Studies Of The Seasonal Reduction Of The Upstream Kuroshio Transport And Its Adaptive Observation

With the Regional Ocean Modeling System(ROMS),predictability of the seasonal reduction of the upstream Kuroshio transport(UKT)and its adaptive observation are studied by utilizing the conditional nonlinear optimal perturbation(CNOP)approach.Main conclusions are listed as follows:Firstly,the upstream Kuroshio and its transport variation are well simulated by ROMS.To investigate the impacts of initial errors on UKT prediction and its growth mechanism,CNOPs are calculated through the nonlinear optimization system built with ROMS adjoint model.The results show that initial errors can significantly affect UKT prediction.The large-amplitudes of CNOPs are located around 128°E horizontally and in the upper 1000 m vertically.At the prediction time,CNOPs develop into eddy-like structures affecting the upstream Kuroshio.Meanwhile,the error-evolution shows two characteristics: westward propagation and fast growth.Further studies indicate that baroclinic instability is main reason causing the fast error-growth.Secondly,the optimal sensitive area(OSen)of adaptive observation for predicting UKT variation is determined using the vertically integrated energy scheme,with eventually choosing total energy(TE)scheme and the sensitive area size as 0.5% of the model domain.Subsequently,sensitive experiments are conducted to evaluate the sensitivity of OSen and further investigate the impacts of spatial patterns and locations of initial errors on UKT prediction.The results show that initial errors in OSen tend to result in worse prediction results.Moreover,initial errors with CNOP-like patterns are more likely to cause larger prediction errors.Therefore,adaptive observation in OSen can improve UKT prediction more effectively.Finally,random and adaptive observation networks with different observation settings are constructed and their effects on improving UKT prediction are evaluated by observation system simulation experiments(OSSEs).The results show that adaptive observation strategy is more effective than random observation strategy.The adaptive observation networks with six or eight observation sites and observation distance of 140 km or 165 km generally have the best performances.These optimal adaptive observation networks can improve UKT prediction by approximately 40%,with relatively higher observation efficiency and smaller prediction benefit deviations.This study reveals the impacts of initial errors on UKT prediction and consturcts the optimal adaptive observation networks with appropriate observation parameters.It is expected that in the future the numerical simulation and forecast of the Kuroshio can benefit from the results provided above.