Estimating rice yield in selected rice growing areas in the Philippines using CERES-Rice Model

This study developed two sets of calibrated CERES-Rice Model, using 90 and 135 kg N/ha (Model 1), and calibration using 135 kg N/ha (Model 2) and was evaluated by simulating the effect of varying nitrogen (N) levels: 90, 135 and 180 kg N/ha. The potential yield of popular cultivars, namely: PSB Rc18, NSIC Rc222 and Mestizo 19, were estimated using the CERES-Rice Model of Decision Support System for Agrotechnology Transfer (DSSAT). Potential yield of other location-specific cultivars, such as: NSIC Rc238, NSIC Rc302, PSB Rc82, NSIC Rc216 and NSIC Rc298 were also estimated. Genetic coefficients were determined using DSSAT � Generalized Likelihood Uncertainty Estimation (GLUE) using 2014 WS experimental datasets from different sites. The calibrated models used the weather variables from the weather stations at CMU [Central Mindanao University], CLSU [Central Luzon State University] and UPLB [University of the Philippines Los Baños] and these were validated using other test experimental locations deriving the same season and year. Model 1 predicted the yield of the UPLB grown cultivars better than Model 2. On the contrary, Model 2 predicted the yield of CLSU cultivars better than Model 1. For CMU grown cultivars, both models (1 and2) produced wide deviations between the simulated grain yield and observed grain yield due to pest occurrence in the area. Assessment of model fitness using d-stat (sensitivity analysis across locations) revealed that PSB Rc18 has the best fitness relative to NSIC Rc222 and Mestizo 19. Across cultivars and sites, Model 2, which is calibrated at higher N-rate (135 kg N/ha) has better predictability than Model 1 based on the accuracy assessment. Moreover, Model 2 is more receptive to increasing N-rate levels than Model 1. For more precise yield estimation,further evaluation of the robustness of the generated models has to be undertaken to lessen the significant variations in phenology and grain yield of some cultivars with poor predictability, especially the hybrid cultivars through generation of more data from wet and dry seasons in various growing seasons. Regarding reliability of t models developed, Model 2 (calibrated at 135 kg N/ha) is further recommended since it is more adequate based on the calculations of statistical errors, and slope and intercept of regression between observed and predicted variables. Hence, Model 2 could be used for further validation and sensitivity analysis.