Smart farming: Big Data and AI redefine agricultural decision-making

While sensing technologies have advanced rapidly, the study identifies data fragmentation as one of the most persistent challenges. Agricultural data are often siloed across platforms, collected under incompatible standards, or locked behind institutional and commercial barriers. This fragmentation undermines the effectiveness of intelligent decision-making models, which depend on continuous, high-quality, and interoperable data inputs.

The next leap in agricultural performance will depend on how effectively big data and artificial intelligence are integrated into end-to-end decision-making across the entire crop production cycle, says a new review paper published in Agronomy.

In a study titled “A Comprehensive Review of Big Data Intelligent Decision-Making Models for Smart Farms,” an international research team systematically examines more than two decades of global research to assess how data governance and intelligent models are transforming smart farming. The paper maps both technological progress and unresolved bottlenecks, offering one of the most detailed frameworks to date for how smart farms can evolve from fragmented digital tools into fully integrated, data-driven production systems.

Data governance becomes the backbone of smart farm intelligence

The review finds that the foundation of smart farming lies not in algorithms alone, but in how agricultural data are collected, managed, secured, and shared. Modern farms now generate vast streams of information from satellites, drones, soil sensors, weather stations, machinery logs, and production records. These data sources span structured, semi-structured, and unstructured formats, often arriving at high frequency and across large spatial scales.

While sensing technologies have advanced rapidly, the study identifies data fragmentation as one of the most persistent challenges. Agricultural data are often siloed across platforms, collected under incompatible standards, or locked behind institutional and commercial barriers. This fragmentation undermines the effectiveness of intelligent decision-making models, which depend on continuous, high-quality, and interoperable data inputs.

The authors highlight a clear shift toward federated data governance architectures as a solution. Rather than centralizing all data in a single repository, federated systems allow data to remain locally controlled while enabling collaborative model training and decision support across regions and institutions. This approach reduces data-sharing risks, lowers storage costs, and addresses privacy and ownership concerns that have slowed adoption in many farming contexts.

Equally important is the evolution of agricultural data storage and management. The review outlines how smart farms increasingly rely on hybrid storage architectures that combine time-series databases for sensor streams, object storage for remote sensing imagery, relational databases for operational records, and in-memory caching for real-time decision support. These layered systems are designed to balance performance, scalability, and cost, ensuring that critical data remain accessible when decisions must be made within seconds or minutes.

Security and trust emerge as central themes. As farms become data-intensive, concerns around confidentiality, integrity, and misuse grow more acute. The study documents rising use of encryption, access control, audit trails, and blockchain-based verification to ensure that data can be shared securely and traced throughout their lifecycle. Without such safeguards, the authors warn, farmers and institutions may resist participating in smart farming ecosystems, limiting their collective benefits.

Intelligent models reshape decisions before, during, and after harvest

The review also provides a detailed assessment of how intelligent decision-making models operate across three critical phases of agricultural production: pre-season planning, in-season management, and post-harvest evaluation. Together, these phases form a closed-loop system in which data continuously inform decisions, actions generate new data, and outcomes feed back into future planning.

In the pre-season phase, decision models focus on optimizing land use and production strategies before crops are planted. These include land suitability assessment, planting plan optimization, sowing schedule determination, and crop variety recommendation. Early approaches relied heavily on expert judgment and linear evaluation methods, which offered transparency but limited adaptability. The study shows how machine learning and crop growth simulation models have increasingly replaced or augmented these methods, enabling more accurate assessments under complex environmental conditions.

Machine learning techniques such as neural networks, random forest models, and transfer learning systems now analyze historical climate data, soil properties, and yield records to identify optimal planting strategies tailored to specific fields. Mechanistic crop growth models add interpretability by simulating physiological processes, helping bridge the gap between data-driven accuracy and agronomic understanding.

During the in-season phase, intelligent decision-making becomes dynamic and time-sensitive. The review highlights how real-time monitoring and control systems now guide fertilization, irrigation, and pest management with minimal human intervention. Drone and satellite imagery, combined with soil and canopy sensors, provide continuous insight into crop health and environmental conditions.

Advanced models such as deep learning architectures and reinforcement learning systems enable variable-rate fertilization and irrigation, adjusting inputs based on crop growth stage, weather forecasts, and resource efficiency targets. Model predictive control frameworks further refine these decisions by optimizing actions over rolling time horizons, allowing farms to respond to uncertainty while minimizing waste.

Pest and disease management represents one of the most technically advanced yet operationally challenging areas. Deep learning models have achieved high accuracy in identifying disease types from imagery, but the study notes that estimating disease severity remains a major bottleneck. Without reliable severity assessment, precision spraying and fully automated control remain difficult to implement at scale.

In the post-harvest phase, intelligent models extend decision-making beyond yield measurement into logistics, evaluation, and sustainability assessment. Spatiotemporal deep learning models now predict yields before harvest, enabling better planning of labor and machinery. Harvest timing models integrate remote sensing and weather data to minimize losses and optimize quality, while intelligent routing algorithms schedule machinery operations to reduce idle time and fuel use.

The review emphasizes that post-harvest evaluation is increasingly multidimensional. Rather than focusing solely on economic output, smart farms are adopting holistic performance assessments that consider ecological impact, resource efficiency, and social outcomes. Hybrid evaluation frameworks combine quantitative data with expert judgment to provide more balanced assessments of sustainability.

Integration, scalability, and ethics define the next phase of smart farming

Despite rapid progress, smart farming remains at a transitional stage. Many advanced models perform well in experimental or regional settings but struggle when applied across diverse climates, soils, and farming systems. Cross-regional transferability remains limited, with models often requiring extensive recalibration to maintain accuracy outside their original context.

Data scarcity also persists, particularly for rare events such as extreme weather, pest outbreaks, or specific growth stages. The lack of high-quality labeled data constrains the robustness of data-driven models and increases the risk of failure under real-world conditions. At the same time, computational demands pose practical barriers, especially in rural areas with limited connectivity and hardware capacity.

To address these challenges, the authors outline a phased development pathway for intelligent decision-making in smart farms. In the short term, priority should be given to standardized data governance frameworks and the development of agricultural foundation models capable of learning from diverse data sources. These models can improve generalization and reduce the need for region-specific retraining.

In the mid-term, the study points to federated learning and human–machine collaboration as key enablers of scalable intelligence. Federated systems allow models to improve across farms without exposing raw data, while human–machine interfaces enhance transparency and usability for farmers. This stage is critical for building trust and ensuring that intelligent systems support, rather than replace, agronomic expertise.

In the long term, the vision extends to real-time, ethical edge AI deployed directly on farm equipment and local devices. Lightweight models running on edge infrastructure could support millisecond-level decisions for irrigation, fertilization, and pest control, even in connectivity-constrained environments. At the same time, ethical governance frameworks must ensure that smart farming advances sustainability, data rights, and social equity rather than reinforcing existing disparities.