A Review of Image-Based and AI-Driven Water Level Monitoring: Methods, Datasets, and Deployment Considerations

Accurate and continuous water-level monitoring is essential for agriculture, flood management, and water resources planning. However, many rivers and agricultural canals remain ungauged because conventional sensing infrastructure is costly to install and maintain, limiting basin-scale water accounting and conservation policies such as water banking and leasing. Computer vision enables water-level estimation from ground-based cameras, unmanned aerial vehicles (UAVs), and satellites, offering a low-cost, non-contact alternative to conventional sensors. This review synthesizes 93 studies selected from 168 articles retrieved from five databases, following PRISMA 2020 guidelines. We summarize contact-based and non-contact sensing methods and examine image-based approaches, including traditional image processing, classical machine learning (ML), and deep learning (DL), alongside their imaging modalities and datasets. DL methods, particularly object detection, semantic segmentation, and transformer-based models, are more robust than earlier approaches under low light, reflections, and cluttered backgrounds, with several studies reporting field errors of approximately 1–3 cm. However, these results are not directly comparable because studies differ in datasets, reference measurements, and evaluation metrics. Three major research gaps emerge: (1) fragmented datasets and inconsistent evaluation protocols; (2) limited cross-site validation and model generalization; and (3) insufficient long-term field evaluation and robustness to illumination changes, reflections, and camera displacement. Future research should prioritize standardized, publicly available multi-site datasets, common evaluation protocols, cross-site and seasonal testing, and adaptive models that maintain reliability under changing field conditions.