Model-constrained 3DVar Approach And Its Application In Tropical Cyclone Forecasting

Data assimilation technique plays a very important role in numerical weather prediction (NWP) that is a typical problem of initial and boundary conditions. In the past decads, the 3D- and 4D-Var data assimilation techniques were improved quikly not only in the theoretical researches but also in the NWPs of operational centers. However, there are still some problems which limit the wide use of Variational data assimilation techniques. The open questions include how to define the coVariance matrix of background error, how to handle the boundary conditions, how to reduce the computing cost of 4DVar, how to eliminate the high-frequency disturbance which are incurred by the insert of observation information and so on. In this dissertation, the efforts are focused on improving the Variational data assimilation technique and applying of this method to numerical forecast of tropical cyclone.In the review of the development of the data assimilation technique in Chapter 1, it is pointed out that a disadvantage of 3-dimensional Variational data assimilation (3DVar) technique is its lack of complicated constraints such as the dynamics and physics in a numerical model, which are used in the 4-dimensional Variational data assimilation (4DVar) technique. On the other hand, however, using a numerical model and its adjoint in the 4DVar technique requires a large amount of computer resources, and thus limits its practical applicability.In Chapter 2, a new 3DVar method is proposed by adding a numerical model constraint. This method minimizes the distance between observation and model Variables and time tendency of model Variables, so that the optimized initial conditions not only fit the observations but also satisfy the constraints of full dynamics and physics of the numerical model. The forward and adjoint models used in this method are as same as those in the 4DVar method but are only integrated one time step to calculate the time tendency. Because observations are only used at one time slice and meanwhile model constraints are applied, the method is called the model-constrained 3DVar (MC-3DVar).A set of ideal experiments based on a shallow water equation model indicates that the model constraints used in MC-3DVar can spread the observation information spatially and balance the model Variables.In Chapter 3, the MM5 (National Center for Atmospheric Research/Penn State Mesoscale Model 5) MC-3DVar system is established using the same forward and adjoint models of MM5 4DVar system. In the MC-3DVar system, the forward and adjoint models are only integrated one time step to calculate the time tendency and gradient.Using the MC-3DVar system, AMSU-A retrieved air temperatures are assimilated into 32 tropical cyclone (TC) cases. The results show a significant decrease in track forecast errors. Meanwhile, one case study of assimilating AMSU-A temperature, QuikSCAT sea-level winds, and cloud drift winds gives dramatic track error decreases. The study shows that the assimilation of these data with MC-3DVar improves TC forecasts and more satellite data give better performances.In Chapter 4, an assimilation cycle is employed to improve the initial conditions using Various data at different time. A case study of typhoon Saomai (2006) from 20BST 8th to 08BST 9th Sep. is carried out in this chapter. The assimilated data include the Cloud Drift Wind, QuikSCAT sea level wind, Dropsonde and Bogus sea level pressure. The 12, 24, 36 and 48 h track forecast of Saomai is improved dramatically after assimilating of these data cycle, and the minimum 48 h track forecasting error is reached by combining the tropical cyclone vortex in the assimilated field and the background field of the output of global model.Finally, the conclusions and discussions are given in Chapter 5.