Abstract -- A prototype 4D-variational dat assimilation system was first implemented by using a shallow-water (SW) model and its adjoint counterpart. It had been applied to explore the potential values in predicting typhoon tracks with a cheap simplified dynamical model.
In a series of so-called hindcast experiments applied to a notorious Typhoon Herb (1996), it was noted that the original SW model predicted tracking errors in 24/48/72 hours are 129/295/491km with a sample sizes of 15/13/11, respectively. By defining different cost functions to include known analysis, say 12 hours ago, a series of 4D-Var-like experiments were conducted to obtain ‘new’ model initial conditions. In a statistical sense, the model performance has been boost virtually for all designs, especially for predictions 2 days and longer in advance. The 24-/48-/72-hour forecasterrors are now 119/237/402km for the best experiment set.
Apparently, the implementation of a 4D-Var system onto a simple SW model is affordable and can be very beneficial.