Improvement of the agricultural effective rainfall for irrigating rice using the optimal clustering model of rainfall station network

In this study, the clustering method was applied to improve the usage of effective rainfall (ER) for irrigating rice paddy in the region managed by the TaoYuan Irrigation Association (TIA) of Taiwan. A total of 16 rainfall stations and rainfall data from 1981 to 2000 were used. A traditional area-weighted method (Thiessen polygons method) and an optimal clustering model of ER were evaluated and compared. The optimal clustering model of ER comprised self-organizing map (SOM), k-means (KM), and fuzzy c-means (FCM) clustering algorithms. To obtain optimal clustering data of ER, the clustering groups from two to five of SOM, KM, and FCM algorithms were determined using root-mean-squared-error. The results show that three algorithms with group numbers from two to five are adopted for the monthly optimal clustering model in different months. However, for the annual optimal model, 12 sub-models are assessed and then compared. The results show that the SOM clustering with groups three was the optimal model for annual ER. The optimal clustering model of ER provides a new procedure step in preparation of the irrigation scheduling for the TIA, and the amount irrigation water waste can be reduced from 770.1 to 22.3 mm/year. The planned ER using the optimal clustering model significantly improves the irrigation water use efficient in agricultural water management.