Coupling computational fluid dynamics with Decision Tree and k-Nearest Neighbours classifications for stagnation and water quality assessment in arid rivers: evidence from the Tigris

Hathal, Mustafa and Sebestyén, Viktor and Anweiler, Stanislaw and Somogyi, Viola (2026) Coupling computational fluid dynamics with Decision Tree and k-Nearest Neighbours classifications for stagnation and water quality assessment in arid rivers: evidence from the Tigris. APPLIED WATER SCIENCE, 16 (7). No. 201.

Preview

Text s13201-026-02883-1.pdf - Published Version Available under License Creative Commons Attribution. Download (3MB) | Preview

Abstract

Urban river systems in arid climates are experiencing extreme hydrological shifts driven by evolving drought patterns and upstream flow regulation. Despite these transformations, the precise physical mechanisms linking reach-scale stagnation to water quality deterioration remain inadequately resolved. In this study, a Comprehensive Water Assessment Framework was developed by integrating k–ε turbulence-based Computational Fluid Dynamics (CFD) with interpretable machine learning (ML) classifiers—Decision Tree and k-Nearest Neighbour. This framework was applied to diagnose flow dynamics in the Tigris River, Baghdad, over the period 2014–2023. It was revealed that flow stagnation responds non-linearly to discharge, with the stagnant fraction expanding to ~ 72% of the reach length during extreme drought, coinciding with critical Water Quality Index (WQI) collapses (51–55). An empirically derived safe flow threshold of ≈ 200 m³/s was identified, below which hydraulic connectivity and self-purification capacity dropped sharply (R² = 0.98). Furthermore, WQI variability was shown to be strongly governed by a minimal set of drivers, including stagnation extent, turbidity, pH, TDS, Ca²⁺, and PO₄³⁻with a dominant stagnation × turbidity interaction highlighted via multivariate response-surface analysis (R² = 0.97). As CFD-derived diagnostics were replicated with ML models at < 2% error, a high-accuracy, computationally efficient tool for real-time management is provided. These findings offer a scalable outline for drought operations, allowing high-risk urban hotspots to be identified and abstraction strategies to be optimized in water-stressed environments.

Actions (login required)

View Item