Comparison of statistical models for predicting cost effective nitrogen rate at rice–wheat cropping systems

Abstract

Due to increased economic and environmental concerns, developing statistical models of crop yield has become one of the most important steps in determination of the cost effective rates (CERs) of nitrogen (N) fertilization. Although quadratic models are commonly used to describe wheat and paddy rice yield response to fertilizer rates in the Taihu Lake region of China, few studies have investigated why this model is selected over others. This study evaluated quadratic, exponential and square root models describing the wheat (Triticum aestivum L.) and rice (Oryza sativa L.) yield response to N fertilizer when determining the CERs, while also considering the environmental costs of N losses. All models fit the data almost equally well when evaluated using the variability and standard error statistics. However, there were marked discrepancies among models when calculating the CER of fertilization and the economic returns form Z-test. The quadratic model had a greater CER value (194 kg N ha–1 for rice and 185 kg N ha–1 for wheat) averaged over all sites than the exponential and square root models. The residuals obtained from the quadratic models were closer to a normal distribution than those of the other two models, indicating a less systematic bias. The mean economic uncertainties resulting from the quadratic model were more dependable than the other two models evaluated. These results show that the quadratic model best describes the rice and wheat yield responses and tends to indicate the optimal rates of fertilization while considering the environmental and economic effects of over fertilization for rice and wheat in the Taihu Lake region.