Non-contact discharge estimation based on integral length scale and Convolutional Neural Networks

Accurate discharge estimation is critical for evaluating sediment transport processes and supporting hydraulic infrastructure planning. However, conventional contact-based methods, such as Acoustic Doppler Current Profilers (ADCP) and electromagnetic flow meters, are often expensive, labor-intensive, and pose operational risks under extreme flow conditions. To overcome these limitations, this study proposes a two-dimensional, non-contact flow measurement framework that integrates Convolutional Neural Networks (CNN) with Large-Scale Particle Image Velocimetry (LSPIV). The CNN is employed to enhance the accuracy and noise resilience of surface velocity reconstruction. From the resulting velocity field, the turbulent Integral Length Scale (ILS) is derived via spatial autocorrelation analysis of velocity fluctuations. This physically meaningful parameter is then used to establish an empirical relationship with flow depth, enabling indirect bathymetric estimation. This study is among the first to combine CNN-enhanced LSPIV with turbulence-based ILS metrics for indirect bathymetry estimation, offering a novel framework for non-contact hydraulic monitoring. The coupled CNN-LSPIV and ILS approach provides a robust and efficient means for surface velocity and water depth estimation without direct field contact. Laboratory flume experiments were conducted under various velocities and bed geometries, including smooth, laterally composite, and longitudinally composite beds. The proposed method consistently reconstructed high-quality velocity fields with a mean error of 5.6% and demonstrated a strong linear correlation (R² up to 0.84) between ILS and measured water depth. Discharge estimates showed an overall error of approximately 17% compared to reference measurements. These results confirm the method’s reliability and its potential as a practical tool for non-contact hydraulic monitoring under diverse flow and channel conditions.