This paper suggests tree-based model incorporated with Bagging (Bootstrap AGGregatING) as a method to combine prediction models. This methodology makes estimates by tree-based model continuous and gives better predictability than that of conventional tree-based models. Both of simulation study and an application to rice yield prediction models indicate that this is a powerful tool for model fusion.

View full abstract

We predict every 10-year potential rice yield in the next 100 years, using results from the transient CO₂ experiment of the MRI's CGCM and the rice crop simulation model named SIMRIW. In this study, we chose 'Koshihikari' and 'Kirara 395' as the representative cultivars in Japan. The maximum potential yield under various climatic conditions is considered using the current planting day and also after adoption of an optimal planting day. Using the current planting day, the growth duration is predicted to be 16-19 days shorter, resulting in a 15% decrease in the potential yield of 'Koshihikari' for the next 100 years. On the other hand, after adopting the optimal planting day, the potential yield can maintain the current level or show an increase of 10-15%. The potential yield of 'Kirara 395', cultivated in Sapporo and Asahikawa, which are the northernmost places for planting rice, is predicted to show a 20% increase after adopting the current planting day, and a 25% increase after adoption of the optimal one for the next 100 years. Moreover, it is predicted that it would be possible to harvest 'Koshihikari' even in the northern part of the Tohoku and Hokkaido districts after 30-40 years, and that the potential yield can reach more than 8t ha^-1 after 100 years.

View full abstract

Retrieval systems for commercial names of Japanese fertilizers, makers or importers of fertilizers, and notifications of Fertilizer Control Law of Japan was developed on WWW (World Wide Web). The technical issue to construct these systems is discussed in this paper. There are three kinds in our retrieval systems: a) The retrieval system for the commercial names of fertilizer according to the rules of the Fertilizer Control Law, b) The retrieval system for makers or importers of fertilizers or soil conditioners, c) The retrieval system for governmental notifications of the Fertilizer Control Law of Japan. In this study, the difference between the method of Hyper-text type and retrieval system to offer necessary information through Internet was compared. At the same time, the difference of the retrieval speed on CGI Scripts created by both Visual Basic and JPerl was compared. In addition, management and operation of these systems were examined practically. In conclusion, it is necessary to adopt the most appropriate method according to the characteristics of the data to be provided, and the method to easily access the database created by commercial database applications via Internet .

View full abstract