Forecasting rice yield with climate variables in Nueva Ecija and Pampanga using mixed-data sampling regression and artificial neural network methods

• Main Authors: Coronel, Anne Gabrielle L., Diaz, Khrystal May B., Velos, Earl Godfred V.
• Format: text
• Language: English
• Published: Animo Repository 2024
• Subjects: Climatic changes—Philippines; Rice > Yields > Philippines > Nueva Ecija; Rice > Yields > Philippines > Pampanga; Regression analysis; Agriculture—Forecasting; Neural networks (Computer science); Mathematics
• Online Access: https://animorepository.dlsu.edu.ph/etdb_math/41; https://animorepository.dlsu.edu.ph/context/etdb_math/article/1040/viewcontent/2024_Coronel_EtAl_Forecasting_rice_yield_with_climate_variables_in_Nueva_Ecija_and.pdf
• Institution: De La Salle University

Climate change induces significant long-term changes in weather conditions and patterns, causing damage to the agricultural sector. Over the past few years, more scientific evidence of weather-related agricultural damages in the Philippines has accumulated. Thus, this study aimed to determine the best predictive model of rice yield in Nueva Ecija and Pampanga among Mixed-Data Sampling (MIDAS) Regression, Long-Short Term Memory (LSTM) neural networks, and Gated Recurrent Unit (GRU), with the climate variables temperature, precipitation, speed of maximum gust, and air pressure as regressors. To determine the best model per province, RMSE, MAE, MAPE, AIC, and BIC were used for comparisons. Results showed that the Adam-optimized GRU model performed the best in forecasting rice yield data in Nueva Ecija, where temperature is seen to be the most influential predictor among the climate variables. On the other hand, the Yogi-optimized LSTM resulted in the superior forecasting accuracy in Pampanga data. Importance scores were distributed almost equally across all variables, with the speed of maximum gust having more importance than the others by a small margin. These results suggest that ANNs demonstrate superior performance compared to MIDAS models when dealing with mixed-frequency data, resulting in a parsimonious model with better forecasting accuracy. The findings of this study may provide management baselines for governing bodies in creating strategies that target environmental and agricultural development in the Philippines. Keywords: Rice Yield, Climate Change, Mixed-frequency time series, Artificial Neural Network, Central Luzon