Building of the Predictive Model for Ripened Brown Rice Yield Using Data Collected in the Production Sites and Analysis of Determinants of Yield Classes

<p>Rice production in Japan faces concerns over yield reduction and poor grain quality owing to unstable weather conditions and the deterioration of soil fertility. In addition, it needs to increase competitiveness against foreign agricultural products and other food products. This situation requires to develop analytical methods to effectively use data regarding agriculture, which is becoming easily available with information and communication technologies. The objectives of this study were 1) to build the predictive model of three classes (low, middle, and high) of ripened brown rice yield using data collected in the production sites, 2) to clarify limiting factors of yield using the predictive model, and 3) to investigate measures to improve yield. We applied support vector machine to build the predictive model using data collected from 2011 to 2016 in a total of 90 paddy fields. The paddy fields where Hinohikari cultivar was cultivated were located in the city of Itoshima, Fukuoka Prefecture. The predictive model built included three explanatory variables: the number of rough rice, daily mean sunshine hours during the period from five days after heading to the late ripening period, and mineral nitrogen supply from the early panicle formation to the late ripening periods. The predictive model achieved the classification accuracy of 77.8%. Particularly, paddy fields categorized in the low yield class had small tiller number at the panicle initiation stage. As the result, small number of rough rice decreased yield. Some paddy fields, where their yield was categorized in the middle class and was categorized in the high class with the number of rough rice greater than 28 thousand m^-2, reduced yield because of shortage of mineral nitrogen supply during the ripening period. The measures to improve yield were as follows: for the paddy fields categorized in low yield class, to ensure appropriate number of rough rice by increasing tiller number during the vegetative growth stage, and for paddy fields where shortage of mineral nitrogen supply during the ripening period reduced yield, to apply topdressing two times to meet the standard nitrogen application rate in the surveyed region (3.6 kg/10a in total). This measure obtained for topdressing indicates that the analysis using survey data collected in rice production sites supported the standard that has been determined by local cultivation tests as well as trial and error in the production sites.</p>