The field of agricultural monitoring and modeling is moving towards the development of more universal and generalizable models that can effectively integrate multiple physical models and datasets. This is driven by the need for more accurate and robust predictions of key crop variables, such as carbon and nitrogen fluxes, and the limitations of traditional process-based physical models and data-driven models. Recent work has focused on developing knowledge-guided encoder-decoder frameworks and lightweight multispectral crop-weed segmentation models that can leverage knowledge of underlying processes and integrate multiple modalities, such as RGB, Near-Infrared, and Red-Edge bands. The release of large-scale comprehensive datasets, such as IrrMap, is also providing a rich foundation for irrigation analysis and mapping. Noteworthy papers include:

• Knowledge Guided Encoder-Decoder Framework Integrating Multiple Physical Models for Agricultural Ecosystem Modeling, which demonstrates the effectiveness of a knowledge-guided encoder-decoder model in predicting key crop variables.
• IrrMap: A Large-Scale Comprehensive Dataset for Irrigation Method Mapping, which introduces a large-scale dataset for irrigation method mapping and provides a pipeline for dataset generation and extension.
• Knowledge-Informed Deep Learning for Irrigation Type Mapping from Remote Sensing, which presents a novel Swin-Transformer based approach for irrigation type mapping and achieves significant improvements over baseline models.