Comparison of statistical models for determination of rice field phosphorus critical level

Abstract

The method used in the determination of rice field phosphorus nutrient availability classes is a graphical method of Cate and Nelson (1965) and a modified analysis of variance model (Nelson and Anderson, 1977). On the other hand, there are other models commonly used by researchers in determining the nutrients critical levels that are linear plateau, quadratic, and logarithmic models. Different models could produce different nutrient critical levels and economic benefits. This study aims to determine the best model for determination of rice field phosphorus critical level. The research using multilocation experimental data of P fertilization on rice in Java, Sumatera and Lombok. Each trial consisted of four to five levels of P fertilizer treatment. Measured response is dry grain weight (ku/ha). Soil properties measured was total P (25% HCl) nutrient content before the experiment. The design used was a randomized complete block design with three replications. Results showed that sequentially the AIC value of the models are Cate-Nelson <Linear (Ln-X) <Linear (X) <Linear- Plateau (X) = Quadratic (Ln-X) <Linear-Plateau (Ln-X) <Quadratic (X). However, only the model of Linear-Plateau (Ln-X) was significant (P-value < 0.05), while the other models are not significant (P-values> 0.05). Thus the best model is the model of Linear-Plateau with the logarithmic transformation of independent variables X (Ln-X). Critical threshold of 25% HCl- P is obtained at 34 mg P2O5/100g.