Integrating image processing and machine learning for phase specific estimation of Manning roughness coefficient in furrow irrigation

Abstract

Accurate estimation of Manning’s roughness coefficient (n) is vital for hydrological modeling and optimizing furrow irrigation systems, yet traditional methods remain limited due to spatial-temporal variability, the need for expert users, the inherent limitations of hydraulic equations, and labor-intensive measurements. This study introduces a novel algorithm that integrates high-resolution image processing with machine learning techniques to dynamically predict n during advance and storage phases in bare furrows. Three scenarios were evaluated: (i) full-field data (including inflow/outflow rates, advance/recession times, infiltration, slope, soil moisture, furrow images and etc.), (ii) images only, and (iii) images plus selected field data (inflow rate, slope, cross-sectional area, clod size and irrigation events). Manning’s n was computed using the SIPAR_ID model and Manning’s equation in advance and storage phases, yielding ranges of 0.017–0.636 (advance phase) and 0.015–0.317 (storage phase), with respective means of 0.083 and 0.054. The Random Forest algorithm in Scenario (i) achieved near-perfect performance (99% precision, recall, and F1-score), while Scenario (iii) preserved 95–96% accuracy with significantly reduced data inputs. Notably, excluding hydraulic variables (Scenario ii) led to a ~ 50% performance drop, highlighting their importance. This approach offers a robust, cost-effective solution for n estimation, bridging the gap between precision, practicality, and real-time application in sustainable water management and precision agriculture.