Research on the Application of Agricultural Big Data in Plant Growth Prediction

Abstract

The intelligent transformation of agriculture places plant growth prediction as a critical component for ensuring food security, optimizing resource allocation, and enhancing sustainable productivity. Traditional methods reliant on empirical or simplified mechanistic models struggle with the nonlinearity, high dimensionality, and spatiotemporal heterogeneity inherent in agro-ecological systems. This study investigates the paradigm shift enabled by agricultural big data integrating multi-source, real-time streams from IoT sensors, satellites, UAVs, and farm management systems. We propose a ``Multi-source Data Assimilation and Hybrid Intelligence'' (MDA-HI) framework that synergistically couples process-based crop models with ensemble machine learning algorithms---including Transformer-based architectures and Physics-Informed Neural Networks---within a holistic pipeline encompassing multi-modal data fusion, hybrid modeling, and scalable deployment. Empirical validation across major crops (rice, wheat, maize, tomato) in diverse eco-regions of China (2023--2025) demonstrates significant improvements: the MDA-HI model achieved average RMSE reductions of 42.7% for yield prediction and 38.1% for key phenological stage prediction relative to best-in-class standalone models. A large-scale case study on rice-wheat rotation systems showed that data-driven prescriptions reduced nitrogen fertilizer use by 22.5% and irrigation water by 18.3% while increasing yield by 5.1%. The study further establishes a five-dimensional evaluation system covering accuracy, robustness, interpretability, scalability, and economic benefit. Remaining challenges include edge computing for real-time inference, federated learning for privacy-preserving collaboration, and explainability of complex ``black-box'' models. This research concludes that agricultural big data constitutes a foundational catalyst for predictive, precise, and proactive cognitive agriculture, with profound implications for global food system resilience.

Plant phenotyping relevance

農業ビッグデータを用いて生育・収量・フェノロジーを推定するMDA-HI手法を提案し、複数作物・地域で性能検証しており、植物形質推定手法が研究の中心である。

Code and data availability

The supplied article blocks describe the MDA-HI framework, data collection, and results, but contain no data availability statement, no public dataset deposit, no author code repository or URL, and no trained model release. Only generic platform mentions (Sentinel-2, DJI UAV, APSIM) and the article's own DOI/CC-BY link

No evidence-backed public reproduction asset is currently recorded.