Precision crop yield prediction model based on deep learning and multimodal data fusion

Abstract

More precise crop yield prediction methods are essential in advancing precision agriculture, ensuring food security, and promoting sustainable agricultural practices. Deep learning technologies have significantly improved prediction accuracy by allowing the analysis of large-scale agricultural datasets. However, existing approaches still face challenges in effectively integrating multisource data, such as meteorological, soil, and remote sensing information, and optimizing model parameters. To address these issues, this study introduces TCGNet, a novel deep learning architecture that combines a time convolutional network (TCN), convolutional long short-term memory (ConvLSTM), and the grey wolf optimization (GWO) algorithm. The TCN is employed for time series analysis, ConvLSTM handles spatiotemporal data, and GWO enhances model parameter optimization. The TCGNet model overcomes the limitations of traditional methods and improves crop yield predictions by leveraging the strengths of these methods. Extensive experiments on datasets including ERA5, SoilGrids, MODIS, and FAO Statistics show that TCGNet outperforms benchmark models, achieving a mean absolute error of 20.05. This result highlights the effectiveness of TCGNet in capturing complex agricultural data patterns and its potential contributions to the field of precision agriculture. The integration of advanced deep learning and optimization techniques in TCGNet not only advances crop yield prediction but also lays a strong foundation for future research in agricultural intelligence.