Predicting corn future prices using artificial intelligence models : integrating remote sensing-derived agricultural data and multifaceted market factors

Abstract

Accurate forecasting of corn futures prices is crucial for various stakeholders, including commodity traders, food manufacturers, and policymakers, to make informed decisions. With the advent of artificial intelligence (AI), machine learning (ML), and satellite remote sensing technologies, there is potential for developing sophisticated price forecasting models that incorporate data from regular influence factors (agricultural, macroeconomic, geopolitical, and other commodity factors) and climate-related. This study aims to determine appropriate AI algorithms that can utilize remote sensing satellite images to predict corn yield production, develop an ensemble of AI models integrating remote sensing, geopolitical, agricultural, and economic data for predicting corn future prices, and evaluate and optimize the parameters of the ensemble AI model to maximize the prediction accuracy compared to previous studies. The study employs two main methodologies for remote sensing: the image segmentation method using the DeepLabV3++CBAM+Deep Supervision Model and the land reflectance method. For price prediction, the CNN+LSTM+Self-Attention time series prediction model and the PatchMixer model are utilized. The results demonstrate that using the land reflectance in remote sensing technology to estimate the corn planting area can provide large-scale and continuous surface information with a certain accuracy, and when compared with USDA data, this remote sensing technology of reflectance can monitor and predict the yield well when data coverage is sufficient. Furthermore, the study utilizes remote sensing data and image segmentation technology to assess the impact of natural disasters on corn planting areas in Illinois, revealing a decrease in the corn planting area after the disaster, reflecting the event's impact. The study also importantly explores various factors influencing corn futures price fluctuations and constructs predictive models by incorporating different data sources. The results show that incorporating remote sensing and USDA data significantly improves short-term prediction accuracy, while climate data has a more moderate impact. The PatchMixer model generally outperforms the LSTM+CNN+ Self-Attention model, especially when integrating remote sensing and USDA data, although prediction errors increase as the time horizon lengthens. This study provides a holistic overview of the application and impact of remote sensing technology and various influencing factors on corn price predictions, showcasing how integrating remote sensing and USDA data can significantly enhance the accuracy of corn future price forecasts. It details significant empirical, algorithmic, model development, and interdisciplinary contributions, explores wide-ranging implications for stakeholders, and acknowledges limitations while providing strategic recommendations for future research, suggesting the exploration of advanced predictive algorithms and the inclusion of more diverse data sources.