Soil erosion is one of the major ecological and environmental problems affecting watershed security, agricultural productivity, and the sustainable utilization of land resources. Under the combined influence of extreme weather events, land-use change, and human disturbance, the spatiotemporal evolution of soil erosion has become increasingly complex, creating an urgent need for accurate real-time monitoring and dynamic prediction. Traditional soil erosion monitoring methods mainly rely on field surveys, remote sensing interpretation, hydrological modeling, and centralized cloud-based data processing. Although these approaches provide important support for regional erosion assessment, they still suffer from low temporal resolution, delayed response, high communication overhead, limited local adaptability, and insufficient real-time early-warning capability in complex environments. In particular, when massive multi-source sensor data must be processed continuously, centralized architectures often encounter bottlenecks in latency, bandwidth consumption, and system robustness. To address these problems, this study proposes an edge-computing- supported dynamic prediction system for real-time soil erosion monitoring. The system integrates Internet of Things sensing devices, edge nodes, wireless communication networks, and a cloud-edge collaborative architecture to achieve efficient acquisition, transmission, analysis, and feedback of erosion-related data. Multi-dimensional indicators, including rainfall intensity, soil moisture, runoff status, vegetation coverage, slope-related terrain information, and surface sediment changes, are collected in real time through distributed sensing units. Edge nodes perform data cleaning, feature extraction, local fusion, and preliminary prediction near the data source, thereby reducing transmission pressure and improving response efficiency. On this basis, a dynamic prediction model combining temporal sequence analysis, spatial factor fusion, and intelligent learning algorithms is constructed to identify erosion risk levels and predict erosion trends under changing environmental conditions. A visualization and warning module is further designed to display monitoring results, risk distribution, and temporal evolution characteristics intuitively. Experimental results based on 12 monitoring points and 328,000 valid records from measured and simulated scenarios show that the proposed system achieves an average response time of 1.12 s, which is 31.4% lower than that of the traditional centralized processing mode. The proposed prediction model reaches an Accuracy of 95.8% and an F1-score of 95.3% on the test set, showing clear advantages in monitoring timeliness, prediction accuracy, and operational stability. The study provides a feasible technical framework for intelligent soil and water conservation, disaster prevention, and ecological management, and offers practical support for the digital and intelligent development of soil erosion monitoring systems.
