Structure and validation of RICEPEST, a production situation-driven, crop growth model simulating rice yield response to multiple pest injuries for tropical Asia

RICEPEST, a model simulating yield losses due to several rice pests (sheath blight, brown spot, sheath rot, bacterial leaf blight, stem borers, brown plant hopper, defoliating insects, and weeds) under a range of specific production situations of tropical Asia was developed. The model was assessed, using: (1) combined data sets generated by a series of test-experiments conducted in different sites of the Philippines, India, and China; and (2) one additional, independent, validation-experiment where a wide range of production situations and injury profiles were manipulated at a single site. Model evaluation was based on the analysis of two output variables: grain yield and relative yield loss. The paper reports results of qualitative and quantitative methods used to assess RICEPEST. Qualitative evaluation involved visual examination of graphs where deviations (simulated minus observed) are plotted against simulated values, and displaying an area of acceptance. This method showed that, in general, RICEPEST accounted well for the yield reducing effects of rice pests. Two areas for potential improvement of RICEPEST were however, identified: the simulation of damage caused by dead hearts in water-stressed environments, and the simulation of damage caused by weeds. Quantitative evaluations made use of equivalence- and χ2-tests. The equivalence tests rejected (P≤0.05) the hypothesis of difference between simulated and observed yield and relative yield loss larger than a preset tolerance in both test- and validation-experiments. Conversely, the χ2-tests did not reject the hypothesis of difference in categorised simulated and observed yields and relative yield losses (P≤0.05) in both test- and validation-experiments. RICEPEST proved to simulate adequately yield losses and can be used as a tool to set research priorities for rice crop protection in tropical Asia.