Agriculture turns to AI: Global trends in land use and crop monitoring

Deep learning approaches, particularly convolutional neural networks (CNNs) and other architectures, were used in 49 papers. These models excel at image-based tasks such as land cover classification, vegetation change detection, and species identification. Their growing prominence since 2020 highlights the shift toward more advanced AI in remote sensing applications.

Artificial intelligence is rapidly transforming agricultural monitoring, but its growth is uneven across applications, according to a new systematic review. The research published in Sustainability highlights how AI-based classification techniques are being applied to land use, vegetation health, crop yields, and soil analysis, while leaving major gaps in critical areas such as drought prediction.

The paper, “A Systematic Review of AI-Based Classifications Used in Agricultural Monitoring in the Context of Achieving the Sustainable Development Goals,” brings together findings from 108 peer-reviewed papers published between 2020 and 2024, analyzing methods, tools, and datasets to assess how AI can help agriculture align with the United Nations Sustainable Development Goals (SDGs).

What agricultural problems can AI classification solve?

The review identifies seven core areas where AI has been applied to agricultural monitoring: land use and crop detection, crop yield monitoring, flood-prone area detection, forest and vegetation monitoring, pest and disease monitoring, soil analysis, and drought prediction.

The findings reveal strong research activity in land use mapping and forest monitoring, with 44 and 41 papers respectively. These applications rely heavily on AI models to classify satellite imagery, track changes in vegetation cover, and detect agricultural practices. Such tools are critical for ensuring sustainable land management and monitoring deforestation, directly supporting SDG targets related to responsible consumption and climate action.

On the other hand, drought prediction is strikingly underrepresented, with no qualifying studies meeting the review’s criteria. This gap is significant given the increasing frequency of droughts worldwide and their devastating impact on food security. Similarly, flood-prone area detection attracted only a handful of studies, despite its relevance to resilience planning and disaster risk reduction.

Applications in crop yield monitoring and soil analysis are growing but remain limited compared to land use and forest studies. Yield-focused research often uses AI to forecast production levels for staple crops, while soil studies apply classification techniques to monitor organic carbon content, macronutrients, and pollution. Pest and disease detection, though represented by fewer studies, shows promise with AI models trained to recognize signs of infestations or plant pathogens from imagery.

Which AI methods dominate agricultural monitoring?

The systematic review provides a clear picture of the methods most frequently deployed. Decision tree-based models dominate the field, appearing in 55 of the analyzed papers. Random Forest and gradient boosting methods like XGBoost are especially popular due to their accuracy, interpretability, and robustness in handling diverse datasets.

Deep learning approaches, particularly convolutional neural networks (CNNs) and other architectures, were used in 49 papers. These models excel at image-based tasks such as land cover classification, vegetation change detection, and species identification. Their growing prominence since 2020 highlights the shift toward more advanced AI in remote sensing applications.

Other methods are less common but still notable:

• Support Vector Machines (SVMs) were employed in 19 studies.
• Hybrid models combining multiple algorithms appeared in 13.
• Recurrent neural networks (RNNs) were used in 7 papers, particularly for time-series data.
• Transformers and attention-based models emerged in 5 studies, signaling a newer trend in applying cutting-edge AI to agricultural data.
• Generative adversarial networks (GANs) were tested in 2 studies, mostly for data augmentation and improved classification.

The analysis suggests that while traditional decision tree methods remain foundational, the future of AI in agriculture is moving toward more complex architectures like deep learning and transformer-based models.

What data and tools power AI-driven agriculture?

Agricultural AI models rely heavily on remote sensing, with satellite imagery used in 78 of the reviewed papers. Sentinel and Landsat missions dominate, cited in 43 and 31 studies respectively. These open-access datasets provide consistent, high-resolution imagery that enables large-scale agricultural monitoring.

Other data sources include UAV (drone) imagery in 18 studies, radar in 14, and LiDAR in 4. While satellite data leads in global applications, UAVs are becoming increasingly important for localized, high-precision monitoring. The review also noted innovative uses of stereo cameras and fused radar-optical datasets to improve accuracy.

The software ecosystem reflects this reliance on geospatial and machine learning tools. Python was the most common platform, used in 45 studies, with frameworks like PyTorch and TensorFlow enabling deep learning experiments. Cloud-based platforms such as Google Earth Engine featured in 23 studies, while ArcGIS and QGIS were used for geospatial analysis in 21 and 10 papers respectively.

Despite this progress, the review highlights geographic disparities in research output. China contributed the largest number of studies, while Eastern Europe remains underrepresented despite its significant agricultural land. This imbalance points to unequal access to resources, expertise, and infrastructure.

Closing the gaps in AI for sustainable agriculture

AI-based classifications are proving essential for advancing sustainability in agriculture, but progress is uneven. Heavy focus on land use and vegetation monitoring contrasts with weak coverage in drought prediction and flood risk assessment, areas with direct implications for climate resilience and food security.

Decision tree ensembles and deep learning models currently dominate the field, but integrating these with emerging tools such as transformers, GANs, and hybrid approaches could improve accuracy and adaptability. Future research should also expand the use of UAVs and radar systems to complement satellite data.

Apart from technical improvements, the paper stresses the need for global collaboration, standardized data practices, and inclusive research efforts to address geographic disparities. Stronger participation from underrepresented regions will be critical to ensuring that AI in agriculture contributes equitably to achieving the Sustainable Development Goals (SDGs).