Although many studies have investigated the relationship between rice (Oryza sativa L.) growth and normalized difference vegetation index (NDVI), few have used it as a practical indicator to determine whether topdressing application is necessary. Therefore, in this study, to verify the usefulness of NDVI as a fertilizer application indicator, changes of topdressing effect on rice depending on the NDVI immediately before application were analyzed using hierarchical Bayesian model. When NDVI was low, topdressing application 45–50 d after transplantation greatly improved yield and quality. However, when NDVI was high, the effect of topdressing was scarce. Since the timing of topdressing application in this study is highly effective in increasing leaf area and number of panicles, which was a positive effect when NDVI was low and both sink and source capacity were insufficient, but when NDVI was high, topdressing application would exceed the optimum leaf area index and could not improve source capacity, so there would be little benefit from topdressing application. Sales of rice was simulated using the posterior distribution of parameters and it was considered worthwhile to apply topdressing when NDVI was <0.75. In addition, weather effects were extracted using the posterior distribution of random effects. As a result, the random effects reflected the effects of weather fairly accurately, suggesting that the effects of other factors could be eliminated. In this study, NDVI was found to be useful as an indicator to determine whether topdressing application is needed. In addition, these analytical methods are useful for getting more information from less data.

中文翻译：

---

分层贝叶斯模型分析NDVI对水稻追施效果的变化

尽管许多研究调查了水稻（Oryza sativa L.) growth and normalized difference vegetation index (NDVI), few have used it as a practical indicator to determine whether topdressing application is necessary. Therefore, in this study, to verify the usefulness of NDVI as a fertilizer application indicator, changes of topdressing effect on rice depending on the NDVI immediately before application were analyzed using hierarchical Bayesian model. When NDVI was low, topdressing application 45–50 d after transplantation greatly improved yield and quality. However, when NDVI was high, the effect of topdressing was scarce. Since the timing of topdressing application in this study is highly effective in increasing leaf area and number of panicles, which was a positive effect when NDVI was low and both sink and source capacity were insufficient, but when NDVI was high, topdressing application would exceed the optimum leaf area index and could not improve source capacity, so there would be little benefit from topdressing application. Sales of rice was simulated using the posterior distribution of parameters and it was considered worthwhile to apply topdressing when NDVI was <0.75. In addition, weather effects were extracted using the posterior distribution of random effects. As a result, the random effects reflected the effects of weather fairly accurately, suggesting that the effects of other factors could be eliminated. In this study, NDVI was found to be useful as an indicator to determine whether topdressing application is needed. In addition, these analytical methods are useful for getting more information from less data.