نوع مقاله : مقاله پژوهشی
نویسندگان
1 گروه مهندسی آب، دانشکده کشاورزی، دانشگاه ارومیه ، ارومیه، ایران
2 گروه مهندسی آب، دانشکده کشاورزی، دانشگاه ارومیه، ارومیه، ایران
3 گروه علوم خاک، دانشکده کشاورزی، دانشگاه ارومیه، ارومیه، ایران
4 گروه مهندسی آب، دانشکده کشاورزی، دانشگاه ارومیه، ارومیه، ایران،
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Wheat, as a strategic crop, plays a critical role in the nation's food security. This research aimed to compare different algorithms to increase the accuracy of modeling rainfed wheat yield using artificial neural networks at the field scale. For this purpose, a combination of two datasets, including remote sensing indicators and climatic data for 61 farms, was used as input to the models. The southern regions of the Lake Urmia Basin were selected as the case study area. An artificial neural network (ANN) model, serving as the base model, and metaheuristic algorithms, including the Reptile Search Algorithm (RSA) and the Firefly Algorithm (FFA), were employed to enhance ANN training accuracy. The input data included the Normalized Difference Vegetation Index (NDVI), the Infrared Percentage Vegetation Index (IPVI), and five climatic parameters: total precipitation, mean maximum temperature, mean minimum temperature, average temperature, and mean relative humidity. The results revealed that the developed hybrid models outperformed the standalone ANN model in simulating rainfed wheat yield. Specifically, the ANN-FFA model achieved a coefficient of determination (R²) of 0.758 and a root mean square error (RMSE) of 0.189 ton.ha-1 during the test stage, outperforming the ANN-RSA model, which yielded an R² of 0.748 and an RMSE of 0.190 ton.ha-1. Both algorithms reduced the rainfed wheat yield simulation error index by an average of 0.042 ton.ha-1. As a result, the ANN-FFA hybrid model improved the accuracy of simulating rainfed wheat yield, which necessitates its use and evaluation in other similar areas for a comprehensive understanding.
کلیدواژهها [English]
In the recent years, artificial intelligence (AI) techniques have been increasingly employed in agricultural research due to their ability to address complex and nonlinear problems. Among the most well-known and widely used models is the Artificial Neural Network (ANN), which effectively manages intricate nonlinear relationships and interactions between variables. However, in certain cases, ANN model has demonstrated suboptimal performance in predictive tasks, raising concerns regarding their reliability (Kayhomayoon et al., 2022). To address these limitations, metaheuristic algorithms have been extensively utilized recently years to optimize and enhance the accuracy of standalone models such as ANN, Adaptive Neuro-Fuzzy Inference Systems (ANFIS), and Support Vector Regression (SVR) (Milan et al., 2021). The Reptile Search Algorithm (RSA) and the Firefly Algorithm (FFA) are examples of such metaheuristic methods that have proven their effectiveness in optimization tasks. In parallel with robust prediction models, remote sensing technologies have emerged as cost-effective and accurate alternatives to traditional methods for agricultural yield estimation and modeling, especially for rainfed wheat. In recent years, researchers have increasingly integrated remote sensing indices with meteorological parameters to improve yield prediction accuracy for rainfed crops.
This study aimed to simulate and predict rainfed wheat yield across 61 farms in the counties of Oshnavieh, Piranshahr, and Naqadeh, situated in the southern part of West Azerbaijan Province, Iran, near Lake Urmia. The data covered 2005–2006 to 2024–2025, corresponding to the rainfed wheat growing season from early October to late July, obtained from the West Azerbaijan Agricultural Jihad Organization. The two main datasets were: regional meteorological data and remote sensing vegetation indices. Among the vegetation indices, the Normalized Difference Vegetation Index (NDVI) and the Infrared Percentage Vegetation Index (IPVI) were specifically selected to evaluate vegetation cover in the studied area. These datasets served as input variables for the modeling framework. Vegetation indices were extracted using satellite imagery processed in the Google Earth Engine (GEE) platform. For the period from 2006 to 2012, Landsat 5 imagery was used, while Landsat 8 imagery was employed for the period from 2013 to 2024. Initially, the datasets were integrated and modeled with an Artificial Neural Network (ANN). To improve accuracy, two metaheuristic optimization algorithms -the Reptile Search Algorithm (RSA) and the Firefly Algorithm (FFA)- were applied to optimize the models. Ultimately, two hybrid models, ANN-RSA and ANN-FFA were developed for predicting rainfed wheat yield.
Based on the study findings, the input variables for the model included total precipitation, mean minimum temperature, and average temperature, along with the remote sensing indices NDVI and IPVI, were selected as the model inputs. The incorporation of the Reptile Search Algorithm (RSA) and the Firefly Algorithm (FFA) significantly enhanced the training accuracy of the Artificial Neural Network (ANN). During training, combining ANN with RSA and FFA increased the coefficient of determination (R²) by 13.88% and 17.31%, respectively. In the testing phase, further gains were observed, with R² increasing by 17.24% and 18.80%, respectively. Notably, the Firefly Algorithm, despite its relatively simpler structure compared to RSA, yielded superior predictive performance, achieving a root mean square error (RMSE) of 0.189 ton.ha⁻¹ and an R² value of 0.758. Overall, the ANN-FFA model outperformed other configurations, demonstrating that combining meteorological data with remote sensing indices enhances yield prediction accuracy. This highlights the potential of hybrid machine learning models to support data-driven decisions in precision agriculture, especially under rainfed conditions.
Timely and accurate crop yield prediction is essential for policymakers to effectively mitigate climate-related impacts and maximize agricultural products. One of the key outcomes of this study was the development of a robust predictive model that integrates meteorological parameters and remote sensing indices within an intelligent hybrid machine learning framework for forecasting rainfed wheat yield. The proposed model demonstrated promising predictive capabilities, highlighting the potential of combining artificial intelligence with environmental data for agricultural forecasting. A major advantage of such hybrid models is their independence from the need for detailed physical information about the study area often costly, time-consuming, or practically infeasible to acquire in many regions. As a result, this approach offers a cost-effective and scalable solution for yield estimation, particularly in resource-limited settings. Utilizing readily available meteorological and satellite data reduces the need for extensive field surveys and enhances the feasibility of regional yield predictions, especially in resource-limited areas.
In the present research, A.A.: Writing original draft, Formal analysis, Conceptualization, Data curation, Methodology, Validation, Software, Modeling. V. R.: Writing – review & editing. J.B.: Writing – review & editing. F.A.: Writing – review & editing. M.R: Writing – review & editing.
Data is available on reasonable request from the authors.
The authors would like to thank Urmia University and West Azerbayjan province Jahad Keshavarzi for their kindley supports. Also, the authors would like to thank all participants of the present study.
The study was approved by the Ethics Committee of the Urmia University. The authors avoided data fabrication, falsification, plagiarism, and misconduct. The present manuscript has not simultaneously been sent to another journal.
The author declares no conflict of interest.
)