Predictive modeling, multi-layer evaluation, and intelligent optimization of water quality based on seasonal multivariate data and composite indices using advanced machine learning algorithms: The Mazlaqan River (Markazi Province, Iran)

Abstract

Water quality monitoring and management are critical for safeguarding public health and sustainable environmental management, especially given the complex seasonal and spatial variations in water resources. The objective of the present research is to determine the spatio-temporal dependency between water quality (surface water) and compare the composite indexes, SAR (Sodium Adsorption Ratio), NSFWQI (National Sanitation Foundation Water Quality Index) and IRWQIsc (Iranian Revised Water Quality Index for Surface Water) by multivariate statistics (MANOVA), partial correlation and sensitivity analysis. This study presents a 12-month monitoring of the Mazlaqan River (Markazi Province, Iran), assessing seasonal and spatial variations in surface water quality. Five sampling stations along the river, from upstream to downstream, were selected to capture comprehensive physicochemical and microbial parameters. This approach provides a detailed understanding of water quality dynamics across different seasons and locations. The statistical analyses revealed that Wilks' Lambda was 0.0618 (F = 24.30, p < 0.0001), Pillai's Trace was 1.1758 (F = 9.45), Hotelling-Lawley Trace was 11.327 (F = 51.99), and Roy's Greatest Root was 10.979 (F = 161.02), confirming significant seasonal variations in water quality parameters. The SAR index fluctuated between 2 and 3, rising to about 3 during time periods 20 to 30, while the NSFWQI index decreased from approximately 80 to 74, and dissolved oxygen (DO) dropped from 8 mg/L to 6 mg/L. BOD₅ increased to around 7 mg/L in summer, and COD ranged between 10 and 16 mg/L. The first principal component (PC1) explained more than 60% of the variance in the data. Predictive models demonstrated high accuracy for SAR and reasonable accuracy for NSFWQI and IRWQIsc. SAR values across all samples ranged from 1.77 to 2.95, classified as low risk (S1). NSFWQI values ranged from 73.9 to 83.6 with a mean of about 78.4, categorized as “Good,” while IRWQIsc ranged from 73 to 83, categorized as “Very Good.” Furthermore, a strong negative correlation was observed between DO and microbial contamination Fecal Coliform (MPN/100 mL) with a coefficient of −0.54 (p = 8.81e-05), a high positive correlation between BOD₅ and Fecal Coliform (MPN/100 mL) with a coefficient of 0.68 (p = 9.78e-08), and a significant positive correlation between COD and Fecal Coliform (MPN/100 mL) with a coefficient of 0.61 (p = 3.84e-06). However, the indices are all acceptable to very good all year round (SAR in class S1 with low sodium hazard, NSFWQI in Good, and IRWQIsc in Very Good), which suggests that the system exhibits overall stability. The findings demonstrate the resilience of the water resources and the significance of some of the interventions during the summer to serve as inputs for machine learning-based and metaheuristic optimization models and actively control the water quality.