Evaluation of a Multiple Linear Regression Model for the Prediction of Panicle Number in Rice

Abstract

In Japan, it is required to develop rice production technology that can achieve stable targets of yield and quality, given the high demands of consumers about rice quality, the necessity of enhancing our international competitive power, and a declining trend in the production of first grade rice (Ministry of Agricultaure, Forestry and Fisheries, 2011). Rice yield is determined by yield components such as panicle number, spikelet number, percentage of ripened grains, and grain weight (Kokubun, 2010). Among these components, panicle number is the most important because it significantly affects other yield components as well as rice quality. Panicle number is largely determined by tiller number, which is affected by various factors such as air temperature (Kobayashi et al., 1985), solar radiation (Shimizu et al., 1962; Kitagawa et al., 1988), water temperature (Honjo et al., 1968; Ueki, 1968), and soil nitrogen content (i.e., quantity of ammo–nium in the soil solution and exchangeable ammonium in the soil (Ando et al., 1988). Yet, the interaction of these factors is complicated; thus, tiller number has mainly been controlled on the basis of trial and error experiences of farmers. However, unstable rice yield and quality in recent years indicate limitation of agricultural practice when based on farmer experience. Hence, the development of a prediction model that expresses the relationship between tiller or panicle number and various influential factors is required to support the decision–making process of farmers in agricultural practice. There has been a limited number of studies regarding prediction models on tiller or panicle number, with such studies being focused on factorial analysis. Tanifuji and Thokairin (1985) developed the successive prediction model for tiller and panicle numbers that focused on short–term forecasts. In this model, tiller and panicle numbers were predicted 10 to 20 days in advance by using plant length and tiller number at the time of prediction and daily mean temperature of ordinary year. This model may provide useful information to support decision–making for top–dressing; however, the model does not provide information about required basal fertilizer application, which is the most important practice to control tiller number. In addition, the explanatory variables included in the model were not sufficient to express Evaluation of a Multiple Linear Regression Model for the Prediction of Panicle Number in Rice