Hydrological modulation of biochemical oxygen demand in sanitation-limited subtropical rivers: a case study from the Caspian Sea basin
Year: 2026
Type: preprint
Abstract: Subtropical rivers with insufficient wastewater infrastructure may experience episodic hypoxia and downstream ecological stress due to monsoon-driven organic loading. The effects of fecal contamination on oxygen consumption under different hydrological situations, however, are little understood. In order to identify hypoxia drivers and forecast biochemical oxygen demand (BOD5) in three subtropical rivers that flow into Iran's southern Caspian Sea basin, we have developed an interpretable machine learning framework that integrates Random Forest modeling with SHapley Additive exPlanations (SHAP). Based on 114 stratified observations from 18 longitudinal sites during the wet and dry seasons, the model showed high predictive accuracy (R2 = 0.89; MAE = 1.05 mg L⁻¹). It also held up well during leave-one-season-out validation (R2 = 0.87–0.91). Under well-oxygenated conditions, fecal coliform concentrations had no effect on BOD₅, but when dissolved oxygen levels were low, they dramatically increased oxygen demand, suggesting hypoxia-conditioned microbial respiration. For every river, this relationship was geographically constant. The single mid-basin station was the source of all exceedances of the 5 mg L⁻¹ BOD₅ standard, according to hotspot analysis. This suggests that focused sanitation efforts at this site could result in disproportionate downstream oxygen recovery. The study shows how hydrological conditions, water quality procedures, and management-relevant decision support may all be connected through interpretable machine learning.
Source: Research Square
Authors: Reza Dahmardeh Behrooz, Sourav Ghosh
Institutions: Zabol University, Bengaluru City University
Language: English
Subfield: Water Science and Technology
Field: Environmental Science
Domain: Physical Sciences
Open Access status: green
Funder: University of Zabol
Award: G1861152458
OpenAlex ID: W7234837655
