Using artificial intelligence algorithms to predict rice (Oryza sativa L.) growth rate for precision agriculture

Abstract

Rice growth rate prediction is crucial to achieve precision agriculture. In this study, growth data from three different rice cultivars at two different climate regions in Taiwan were used to calculate the rice growth rate (Gr) in key growth stages. A total of 10,246 Gr data from 95 cultivations were used to assess the feasibility of simulating rice growth rate from ambient temperature by regression algorithm (REG), artificial neural networks (ANN) and gene-expression programming (GEP). Simulation was carried out for each algorithm using ambient temperature and growing season as input variables for the entire life cycle of rice (Tr-all) either as one continuous growth period or as the sum of four distinctive growth stages (Tr-1 similar to 4). The results showed that the model output from Tr-1 similar to 4 are more accurate than Tr-all. However, simulation errors in the transitional stages occurred when the growth rate at different stages was modelled separately. Artificial intelligence based nonlinear models were more accurate than the regression model and ANN generally exhibited better performance than other two models. Compared with REG, the root mean squared error (RMSE) and mean absolute error (MAE) of averaged error-indays for the whole growth period decreased by 41.38% and 42.81% in ANN and 30.14%, and 31.21% in GEP. Considering the potential errors in the transition stages and the high hardware demand for ANN, the GEP-Tr-all model is recommended for the rice growth rate modeling.