This paper represents a new initialization program and some results of a number of model runs for cases. The samples of less successful prediction from historical cases were computed once again using the new program. The results are much better than that of previous. Some progress have been arrived by introducting this new program into practical prediction experiments during the rainy season 1986.