The Genetic Algorithm Assimilation System Of Mesoscale Models And Its Assimilation Experiments

Variational four-dimensional data assimilation (4DVAR) has been recognized as one of theeffective methods to improve numerical weather prediction (NWP) initial field. But theoptimization method used in 4DVAR should satisfy continual presumption, while thediscontinuity of a forecast model, will make the presupposition available no more; whereas thepresent decent algorithms of 4DVAR easily sink at the local optimization. The geneticalgorithm(GA)requires weaker demand to mathematical properties of cost function, and keepbalance between resolve quality and compute speed. So in this context, a GA based 4DVAR isapplied to the more complex MM5 model in order to improve NWP initial field.We have do some experiments about data assimilation .firstly the output of the GA-based4DVAR model has been compared with that of scheme with an adjoint 4 DVAR model, in orderto verify the performance of the GA assimilation system; secondly, we take different observeddata in the data assimilation system to study the different effects of different kinds of observeddata, include conventional observed data and non-conventional observed data. thirdly, we takedifferent variables in the data assimilation system in order to study the effects of different modelvariables take on the GA data assimilation system. Fourthly, we change weight functions andscale factors, The results suggests that different weighting and scaling have different influence onthe GA assimilation system.In conclusion, the GA-based 4DVAR is of superiority to some degree, and it is a new dataassimilation method, we believe that along with the press of techniques used in data assimilation,several important developments will take place in NWP.