Modeling and forecasting rainfed wheat yield based on meteorological variables using combined Artificial intelligence methods

Message:
Article Type:
Research/Original Article (دارای رتبه معتبر)
Abstract:
The wheat has a vital role in the country food security and its yield estimation can be useful in the reginal and macro decisions. To predict the crop yield, artificial intelligence models are known as one of the most suitable tools. Therefore, in the present research a wide range of machine learning models including artificial neural networks (ANN), random forest (RF), support vector regression (SVR) and multivariate adaptive regression spline (MARS) were evaluated in West Azerbaijan where is the most important region in Iran in wheat production. Based on the effective variables on the rainfed wheat yield, including total precipitation during the growing season, number of days prior the first and end rainfall events more than 10mm, relative humidity, evaporation, average sunshine hours and number of frost days were utilized as input variables. The gamma test (GT) method was used in combination with the genetic algorithm (GA) to optimize the number of input variables. Using the gamma test and genetic algorithm, 8 variables were selected as inputs for the models input. Additionally, the models were combined with Ensemble empirical mode decomposition (EEMD) to improve prediction accuracy. The results of this study show The EEMD-MARS model provided the most accurate results, with error evaluation criteria RMSE= 0.112(ton.ha-1), MAE= 0.088(ton.ha-1), NSE= 0.945, and SI= 0.101 for the test stage. Also, in this model, 14 different functions were extracted for estimating rainfed wheat yield. The EEMD-SVR model also had good performance whit comparing other models with error criteria RMSE= 0.132(ton.ha-1), MAE= 0.080(ton.ha-1), NSE= 0.923 in the test stage.
Language:
Persian
Published:
Iranian Journal of Irrigation & Drainage, Volume:19 Issue: 2, May and June 2025
Pages:
243 to 261
https://www.magiran.com/p2870807