Document Type : Original Research Paper
Authors
Department of Surveying and Geoinformatics Engineering, Faculty of Civil, Water and Environmental Engineering, Shahid Beheshti University, Tehran, Iran
Abstract
Background and Objectives: With the growing global population, increasing pressure on natural resources, climate change, and the rising importance of food security, accurate crop yield prediction—particularly for wheat—has become a crucial research topic. In recent years, the integration of remote sensing data, climatic information, and soil properties with machine‑learning algorithms has created new opportunities for estimating crop yields. However, many previous studies have either focused on a single data source or have been conducted at limited spatial and temporal scales. Consequently, a gap still exists in conducting comprehensive and comparative evaluations of models using multi‑source datasets. The aim of this study was to evaluate the capability of different machine-learning algorithms to predict wheat yield using climatic variables and the remotely sensed Normalized Difference Vegetation Index (NDVI), and to assess the generalizability of these models across different spatial and climatic conditions.
Methods: This study was conducted using multi-source data comprising climatic variables and the remotely sensed Normalized Difference Vegetation Index (NDVI). The study covered nine regions in Germany over a 22-year period, thereby enabling the assessment of spatial and temporal variations in wheat yield. The XGBoost, Random Forest, and Ridge Regression algorithms were employed to predict wheat yield. Among the investigated variables, NDVI and vapor pressure deficit (VPD), as an indicator of atmospheric evaporative stress, were identified as important predictors in the model analyses. Model performance was evaluated in terms of predictive accuracy and error using the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). The study was designed not only to analyze the data but also to enable a regional comparison between conventional and advanced predictive models.
Findings: The results showed that the XGBoost algorithm outperformed both Random Forest and Ridge Regression in predicting wheat yield. XGBoost was able to reconstruct spatial and temporal yield patterns with lower error and higher explanatory power. The findings also indicated that NDVI and vapor pressure deficit (VPD), as an indicator of atmospheric evaporative stress, were important predictors of wheat yield patterns. Higher model accuracy was observed in more homogeneous and high‑yielding regions, whereas prediction errors increased in areas with greater environmental heterogeneity. Nevertheless, the overall trend demonstrated that machine‑learning techniques—especially gradient‑boosting models—have strong capabilities in capturing the complex relationships between environmental factors and crop yield.
Conclusion: Overall, the findings of this study demonstrate that integrating remote-sensing and climatic data with advanced machine-learning algorithms can provide an effective tool for predicting wheat yield and supporting decision-making in smart agriculture. The superior performance of XGBoost compared with the other models indicates its strong capability to capture the nonlinear and multidimensional relationships inherent in agricultural data. Nevertheless, this study had several limitations, including the lack of farm-management data, such as sowing dates, cultivar types, irrigation levels, and fertilizer application rates, as well as the absence of soil data with sufficient spatial and temporal resolution for inclusion in the final models. Moreover, the use of more complex models may increase the risk of overfitting when data are limited. Therefore, future studies are recommended to employ more comprehensive datasets, hybrid spatiotemporal models, and advanced approaches such as recurrent neural networks and convolutional neural networks. Such approaches could improve predictive accuracy and model generalizability while enhancing the practical application of yield-prediction models in sustainable agricultural management and food-security planning.
Keywords
Main Subjects
COPYRIGHTS
© 2026 The Author(s). This is an open-access article distributed under the terms and conditions of the Creative Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)
)