Remote sensing-based machine learning and deep learning for agriculture and forestry: Yield estimation, land-cover mapping, and plant stress – A systematic review (2020–2025)

The integration of remote sensing with Machine Learning (ML) and Deep Learning (DL) has rapidly expanded in agricultural and forest monitoring, generating a heterogeneous body of literature that requires systematic synthesis. This study aims to identify recent methodological patterns within the sampled 2020–2025 evidence base, evaluate reporting practices, and highlight research gaps in remote sensing–artificial intelligence (AI) applications for agricultural and forestry monitoring. A systematic literature review covering 2020–2025 analyzed 105 research articles. Proportional stratified sampling was applied across four repositories (IEEE Xplore, Scopus, Web of Science, and ScienceDirect), supported by pilot testing and semantic expansion of search terms. Information on algorithms, sensors, applications, and evaluation metrics was extracted using a structured three-phase protocol. Convolutional Neural Networks dominate land-cover mapping and segmentation tasks, while Long Short-Term Memory networks and Transformers are widely used for time-series analysis in applications such as yield estimation and plant stress monitoring. Classical ML approaches such as Random Forest, Gradient Boosted Decision Trees, and Support Vector Machines remain competitive baselines when feature engineering is robust. Multimodal fusion, including optical–synthetic aperture radar (SAR) integration, is used to preserve complementary structural information when optical observations are cloud-constrained, although climate-stratified comparisons remain scarce. In forestry, hyperspectral imaging (HSI)–light detection and ranging (LiDAR) combinations with three-dimensional convolutional neural network (3D-CNN) architectures improve species discrimination. Hybrid crop–AI models integrating biomass or Leaf Area Index (LAI) with crop simulators enhance physiological consistency and early yield prediction, although limitations persist in spatiotemporal validation, uncertainty quantification, and standardized evaluation metrics. Operational deployment requires rigorous spatiotemporal validation schemes, explicit uncertainty estimation, and standardized evaluation protocols. These findings provide methodological guidance for developing scalable remote sensing–AI monitoring systems and highlight priority research directions including multimodal data fusion, hybrid process–AI models, and efficient learning strategies for data-scarce agricultural and forestry environments. • A comprehensive review of 105 studies on the utilization of remote sensing in farming and forestry has been conducted. • The utilization of deep learning models has been demonstrated to yield a high degree of accuracy in the prediction of crop yield. • The implementation of hybrid approaches has been demonstrated to enhance performance in diverse farm landscapes. • Standards, transferability, and large-scale use continue to pose significant challenges.