Artificial Intelligence and IoT for Riverine Oil Spill Detection: A Focused Review and Proposed Adaptive Edge Framework

Abstract

Riverine oil spills are more challenging to detect than marine spills due to shallow depths, high turbidity, and rapidly changing hydrodynamics, which degrade the performance of satellite- and SAR-based detection methods. This review examines how artificial intelligence and the Internet of Things can deliver accurate, low-latency detection in freshwater and defines an AI-IoT system as distributed river-edge nodes with RGB/NIR cameras, thermal infrared sensors, UV/fluorometric probes, and turbidity/multispectral units running lightweight deep models on low-power hardware and networked via LPWAN or NB-IoT with optional federated coordination. The novel contribution is a hydrodynamics-aware, adaptive framework that embeds river flow and turbidity, couples explicit constraints on edge compute, energy, and inference latency, and derives multi-sensor fusion logic from comparative synthesis. Performance is organized along accuracy, decision latency, deployment cost, and environmental adaptability. Using a structured narrative review with scoping elements, the research screened 145 records from major databases. It synthesized 47 peer-reviewed studies (2020-2025), harmonized definitions, and applied descriptive synthesis to manage heterogeneous metrics and protocols. Results show that SAR and hyperspectral methods that excel in marine or controlled settings often degrade in narrow, turbid rivers because of clutter and revisit latency. In contrast, hybrid AI-IoT architectures employing compact CNN/Transformer variants at the edge report high accuracy with millisecond-scale inference and moderate power budgets. Limitations include heterogeneous reporting, non-standard datasets, and limited multi-site validation. The framework and synthesis motivate open benchmarks and coordinated river trials to standardize evaluation and accelerate translation.