Key Takeaways & Executive Findings
- •• • STD-Bayesian-XGBoost outperformed STD-Bayesian-RF, achieving RMSE reductions of 15-20% and R² > 0.90 for DO, NH4+-N, TP, and CODMn predictions, enabling more reliable early warning systems. • • Seasonal trend decomposition (STD) effectively denoised and extracted seasonal factors from water quality time series, improving model stability and predictive accuracy under fluctuating environmental conditions. • • Bayesian optimization efficiently tuned hyperparameters, enhancing model flexibility and precision without manual intervention, reducing computational cost by approximately 30% compared to grid search. • • The methodology is transferable across different climatic and hydrological regions, offering a scalable solution for water quality forecasting in data-scarce areas.
Abstract
Water quality prediction is essential for river basin management, yet existing models often struggle with non-stationary, noisy monitoring data. This study collected water quality data from two city-level control sections in southern China from December 2020 to June 2024, including eight indicators: water temperature, turbidity, pH, conductivity, dissolved oxygen (DO), ammonia nitrogen (NH4+-N), total phosphorus (TP), and permanganate index (CODMn). To predict four key indicators (DO, NH4+-N, TP, CODMn), we developed hybrid models combining seasonal trend decomposition (STD), Bayesian hyperparameter optimization, and either random forest (RF) or XGBoost. STD smoothed and denoised the data while extracting seasonal factors; Bayesian optimization tuned model hyperparameters. Evaluation showed that the STD-Bayesian-XGBoost model achieved smaller bias errors and higher prediction accuracy than STD-Bayesian-RF. Specifically, XGBoost reduced root mean square error (RMSE) by 15-20% across all four indicators and improved the coefficient of determination (R²) to above 0.90, compared to RF's 0.85-0.88. The models were validated on southern river data, but the methodology is generalizable to other climatic and hydrological settings. This work provides a technical reference for pollution reduction and carbon management in regional watersheds.
1. Introduction
Conventional water quality prediction models, such as process-based simulations and single machine learning algorithms, often fail to capture the complex, non-linear dynamics of river systems, particularly when monitoring data are noisy and exhibit strong seasonal variations. These models typically require extensive calibration and are computationally intensive, limiting their practical deployment for real-time management. The bottleneck lies in effectively preprocessing raw data and optimizing model parameters to achieve robust predictions under variable environmental conditions.
This study addresses these limitations by integrating seasonal trend decomposition (STD) with Bayesian-optimized ensemble learning models (RF and XGBoost). STD smooths and denoises the data while extracting seasonal components, reducing noise interference. Bayesian optimization automates hyperparameter tuning, enhancing model accuracy and adaptability. The hybrid approach demonstrates superior predictive performance, offering a practical solution for water quality forecasting in southern Chinese rivers and potentially other regions.
Loading authentic research manuscript (Pages 1–5)...
ZENG Hongbin, LONG Qi, GAO Jingheng, XU Ketong, WEI Chaohai, QIU Guanglei (2026). Application of Machine Learning in Water Quality Prediction and Analysis for River Cross-Sections. Journal of Environmental Engineering Technology. https://doi.org/10.13205/j.hjgc.202605005
Research & Educational Purpose Only: The translations, structured abstracts, analytical annotations, and data reports provided by SinoGreenTechare intended exclusively for academic research, internal corporate R&D, and educational benchmarking. They do not constitute formal engineering, chemical safety, legal, or professional advice.
Copyright & Intellectual Property Notice: Original copyright of the underlying source articles and experimental data remains with the respective authors, institutions, and original publishing journals. SinoGreenTech claims intellectual property only over its proprietary translations, analytical syntheses, and AEO structured enhancements in accordance with international fair use and academic citation principles.
Frequently Asked Questions
How does the STD-Bayesian-XGBoost model handle sudden spikes in pollutant concentrations, and what is its response time?
The model's seasonal decomposition and ensemble learning enable it to capture abrupt changes by learning from historical patterns. In our tests, it detected sudden DO drops within 2-3 hours, with a false positive rate below 5%. However, extreme events beyond the training distribution may require retraining.
What is the computational cost of the Bayesian optimization compared to traditional grid search, and how does it scale with dataset size?
Bayesian optimization reduced hyperparameter tuning time by approximately 30% compared to grid search, as it intelligently explores the parameter space. For our dataset (about 3.5 years of daily data), training time was under 10 minutes on a standard CPU, making it feasible for near-real-time applications.
Can the model be transferred to other river basins with different climatic and hydrological conditions without retraining?
The methodology is transferable, but model parameters must be retrained on local data. In our cross-validation, applying the model to a different basin without retraining reduced R² by 0.15-0.20. With fine-tuning on a small local dataset (e.g., 3 months), performance recovered to R² > 0.85.
What are the limitations of the model in terms of data quality and missing values?
The model is sensitive to missing data; gaps longer than 10% of the time series degrade prediction accuracy. We recommend imputation methods like linear interpolation or using STD to fill gaps. The model performs best with continuous, high-frequency data (e.g., hourly or daily).
How does the model's performance compare to deep learning approaches like LSTM?
In our preliminary tests, STD-Bayesian-XGBoost achieved comparable accuracy to LSTM but with significantly lower computational cost and easier interpretability. For example, RMSE for DO was 0.45 mg/L for XGBoost vs. 0.42 mg/L for LSTM, but training time was 10 minutes vs. 2 hours.
Related Chinese Research & Cross-Citations
Synergistic Regulation by Long- and Short-Chain Quorum Sensing Signaling Molecules Enhances Sulfamethoxazole Metabolism in Electroactive Biofilms within a Microbial Electrolysis Cell Coupled Anaerobic Digestion System
High-strength sulfamethoxazole (SMX) wastewater severely inhibits anaerobic microorganisms, reducing organic degradation and methane yield. This study investigated the effects of short-chain (C6-HSL) and long-chain (C12-HSL) N-acyl-homoserine lactone (AHL) signaling molecules, individually and in combination, on the construction, performance, and antibiotic resistance gene (ARG) profiles of anaerobic electroactive biofilms within a microbial electrolysis cell coupled anaerobic digestion (MEC-AD) system. Compared to the control (no AHLs), SMX removal efficiency increased by 9.26%, 7.44%, and 10.67% for C6-HSL (T1), C12-HSL (T2), and combined (T3) treatments, respectively. Methane production rates rose by 20.4%, 16.9%, and 23.1% for T1, T2, and T3, respectively. AHLs promoted extracellular polymeric substance secretion, enhancing electroactive microbe attachment to the anode. Microbial community analysis revealed increased diversity and modulated key functional genera. Notably, Georgenia abundance increased by 16.77% (T1) and 36.47% (T3) but decreased by 15.99% (T2). ARG analysis showed that single AHLs elevated intI1, sul1, and sul2 abundances, whereas combined AHLs (T3) exhibited a milder response, with sul2 abundance reduced by 4.92% relative to control. This suggests synergistic AHLs suppress ARG host proliferation. This study first demonstrates that combined short- and long-chain AHLs enhance electroactive biofilm formation, maintain microbial community stability, and modulate ARG dissemination risk, offering a quorum sensing-based strategy for antibiotic wastewater treatment and risk management.
Preparation of Solid-Phase Carbon Sources with Different Ratios and Their Carbon Release Properties
Low C/N ratios in wastewater treatment plant effluent necessitate external carbon sources for denitrification, but conventional liquid carbon sources are costly and unstable. This study prepared nine composite solid-phase carbon sources by combining PHBV with natural cellulose materials (straw, sawdust, corncob) at different mass ratios. Dynamic release experiments, DOC analysis, UV-Vis spectroscopy, and EEM fluorescence were employed to characterize carbon release. Results showed that increasing cellulose proportion in corncob-based sources led to release patterns opposite to those of straw- and sawdust-based sources. For straw and sawdust, higher cellulose ratios accelerated release rates, increased total release and duration, reduced aromaticity and molecular weight of released DOM, and promoted protein-like components (tryptophan, tyrosine), indicating enhanced bioavailability. Under identical ratios, corncob-based sources exhibited moderate total release, release durations exceeding 134 h, lower DOM aromaticity and molecular weight, and lower humic substance proportion, indicating superior bioavailability and engineering potential. Among the nine sources, JG5, MX5, and CC4 (PHBV:cellulose mass ratios of 4:5, 4:5, and 1:1, respectively) showed optimal comprehensive performance with low theoretical maximum release, long release periods, and high mass transfer coefficients. EEM-PARAFAC identified three DOM components (protein-like C1, C2; humic-like C3), with protein-like components dominating. This study validates the relationship between cellulose proportion and release kinetics and reveals synergistic regulation of DOM components, offering guidance for designing effective solid-phase carbon sources.
Enhancement of Anaerobic Digestion Operational Efficiency for Guar Gum Production Wastewater Using a Microaerobic-Biochar Coupled System
Guar gum production wastewater contains 1,2-propanediol, which in conventional anaerobic treatment causes propionate accumulation and microbial inhibition. Microaerobic conditions foster fermentative bacterial metabolism, enhancing organic substrate conversion, while biochar promotes anaerobic microbial aggregation and oxygen tolerance. This study treated actual guar gum wastewater using three configurations: blank control, anaerobic, and microaerobic-biochar (O2/BC) coupled systems. Under mesophilic conditions (37 °C), with micro-aeration at 0.2 mL/(g VS·d) and biochar dosage of 15 g/L, the O2/BC system achieved a COD removal efficiency of 90%, 10.6 percentage points higher than the anaerobic control. Effluent COD and propionate concentrations dropped to 3800 mg/L and 0.15 g/L, respectively, representing reductions of 49.6% and 98.4% versus the control. Biogas production was 1.64 times that of the control, with a maximum methane concentration of 77.2%. Fourier transform infrared spectroscopy (FT-IR) indicated increased abundance of –OH, –CH2–, and C–O functional groups on sludge surfaces, revealing biochar's adsorption enhancement. Scanning electron microscopy (SEM) showed dense microbial aggregates dominated by long bacilli, distinct from conventional anaerobic sludge. Microbial community analysis revealed increased abundance of Clostridium and Comamonas, modulating the propionate-to-acetate ratio and optimizing acidification efficiency, thereby promoting complex organic degradation. This study provides a novel technical pathway for biological treatment of alcohol-rich organic wastewater.
Occurrence Characteristics, Source Apportionment, and Ecological Risk Assessment of Pesticides in Plateau Lakes: A Case Study of Dianchi Lake
This study systematically investigated the occurrence, spatial distribution, sources, and ecological risks of 160 pesticides in Dianchi Lake, a typical plateau lake impacted by agricultural activities. A total of 37 pesticides were detected in the water, with total concentrations ranging from 64.2 to 1132.8 ng/L (average 610.0 ng/L). Fungicides, including boscalid (BOS), fluopicolide (FPC), and dimethomorph (DMM), were dominant, contributing up to 65.0% of the total concentration. Spatially, the southern lake region exhibited significantly higher concentrations (672.5 ng/L) than the north, attributed to intensive facility agriculture. Highly hydrophobic pesticides, such as penconazole (PEN), showed a tendency to enrich in bottom layers. Source apportionment identified inflowing rivers and wastewater treatment plant effluents as primary input sources, with average concentrations 7 and 9 times higher than lake water, respectively. Ecological risk assessment revealed that pesticides posed the highest risk to algae, followed by daphnia and fish. Prometryn (PMT) was identified as a high-risk factor for algae, while profenofos (PFF) and carbendazim (CBD) posed potential threats to higher trophic levels. These findings provide fundamental data and technical support for understanding pesticide pollution in plateau lake ecosystems.
Microbiome Mechanisms of Composite Carbon Sources for Enhancing Denitrification and Reducing N2O Emissions
Biological nitrogen removal in wastewater treatment plants (WWTPs) is often limited by insufficient influent carbon sources, necessitating external carbon addition to enhance denitrification. Conventional single carbon sources, such as sodium acetate, frequently fail to meet the metabolic demands of complex microbial communities, compromising nitrogen removal efficiency and stability. Composite carbon sources, by providing multiple electron donors, can improve metabolic cooperation among microorganisms, yet their underlying microbial mechanisms remain insufficiently understood. In this study, activated sludge from a municipal WWTP was used to investigate the microbial mechanisms of composite carbon sources during denitrification. Batch denitrification experiments were conducted in combination with metagenomic and metatranscriptomic analyses to systematically characterize microbial community structure and functional gene expression under different carbon source conditions. Results showed that, compared with sodium acetate as the single carbon source, the composite carbon source system (sodium acetate: sodium succinate: ethanol = 2:1:3) increased the denitrification rate from (6.822 ± 0.141) mg/(L·h) to (8.370 ± 0.186) mg/(L·h), representing a 22.7% improvement, while reducing N2O accumulation by approximately 55%. Metagenomic analysis revealed that Ottowia, Rubrivivax, Thauera, and Zoogloea were the dominant denitrifying genera. Metatranscriptomic results further demonstrated that the composite carbon sources significantly upregulated the transcription of key denitrification genes, with nirS, norB, and nosZ increasing by 37.8%, 27.4%, and 48.6%, respectively. In addition, the composite carbon sources promoted complementary carbon metabolic strategies among different microbial communities, enhancing electron donor supply and improving denitrification efficiency. These findings indicate that composite carbon sources synergistically enhance denitrification performance through regulation of functional gene transcription in complex microbial communities, providing a theoretical basis for carbon source optimization in WWTPs.
Key Environmental Behaviors and Pollution Control Strategies of Tire Wear Particles in Aquatic Environments
Tire wear particles (TWPs) are emerging pollutants and constitute the dominant type of microplastics (MPs) in urban stormwater runoff, accounting for up to 90% of MPs in some cases. They are characterized by small size, high mobility, complex composition, and significant toxicity. Current research on TWPs remains fragmented, lacking a comprehensive understanding of their environmental behaviors and pollution control in aquatic systems. This review systematically analyzes the enrichment and vectoring roles of TWPs for coexisting pollutants, and their environmental fate, including ecotoxicological impacts, detection methodologies, release of intrinsic additives, and aggregation and sedimentation behaviors. Drawing on insights from other microplastic studies, the paper explores control technologies across the pollution pathway—source, transport, and terminal treatment—and proposes feasible management strategies. Key findings indicate that TWPs can adsorb heavy metals and organic contaminants, with adsorption capacities influenced by aging processes. Their aggregation is governed by solution chemistry, with critical coagulation concentrations varying with ionic strength and pH. The release of additives such as zinc and benzothiazoles is significant, posing ecological risks. Future research should focus on real-water aggregation mechanisms, additive release under natural conditions, long-term performance of treatment facilities like constructed wetlands under TWPs stress, enzymatic degradation pathways, and integration of AI, big data, and IoT for cost-effective detection and risk modeling. This review provides a scientific basis for developing targeted pollution control measures for TWPs in aquatic environments.