Key Takeaways & Executive Findings
- •• • CNN-BiLSTM-Attention achieved the best performance for water temperature prediction, with R² exceeding 0.95 and RMSE below 0.5°C, making it suitable for stable, low-complexity forecasting scenarios. • • TCN-Transformer outperformed other models for pH, dissolved oxygen, and turbidity, achieving R² improvements of 8-12% over SVR and XGBoost, and reducing RMSE by up to 20% for turbidity, demonstrating its capability in capturing nonlinear and long-range dependencies. • • Combining IQR outlier removal with multiple imputation strategies reduced prediction error (MAE) by 15-25% across all models, highlighting the critical role of data quality enhancement in water quality forecasting. • • Bayesian optimization consistently improved model performance, with average R² gains of 0.03-0.05 compared to default hyperparameters, ensuring robust and reliable predictions for operational deployment.
Abstract
This study develops a multivariate time-series forecasting model for water quality in the Lhasa River, focusing on four key indicators: water temperature, pH, dissolved oxygen, and turbidity. Data preprocessing integrated multiple missing-value imputation strategies and interquartile range (IQR) outlier removal. Boxplots and relative standard deviation (RSD) assessed data distribution and dispersion, while autocorrelation and Pearson correlation analyses revealed periodic patterns and inter-variable relationships. Four representative algorithms—Support Vector Regression (SVR), Extreme Gradient Boosting (XGBoost), CNN-BiLSTM-Attention, and TCN-Transformer—were optimized via Bayesian hyperparameter tuning. Model performance was evaluated using MAE, MSE, RMSE, and R². The study systematically compared the effects of different missing-value handling methods, both independently and combined with IQR outlier removal. Results indicate that CNN-BiLSTM-Attention excels in water temperature prediction, suitable for relatively stable and simple patterns. In contrast, TCN-Transformer demonstrates superior performance for pH, dissolved oxygen, and turbidity, which exhibit strong nonlinearity and long-term dependencies, effectively capturing temporal dependencies and coupling relationships. The findings provide a viable technical route and theoretical reference for river water quality monitoring and intelligent early-warning systems.
1. Introduction
Routine water quality monitoring in high-altitude rivers such as the Lhasa River generates extensive time-series data, yet raw datasets are often plagued by missing values and outliers, undermining the reliability of subsequent predictive models. Conventional statistical methods, including linear interpolation and mean substitution, fail to capture the complex, non-linear dynamics of aquatic systems, while traditional machine learning models like SVR and XGBoost struggle with long-term temporal dependencies. This study addresses these bottlenecks by systematically evaluating multiple imputation strategies and IQR-based outlier removal, coupled with advanced deep learning architectures, to enhance both data quality and prediction accuracy.
The experimental protocol compares four models—SVR, XGBoost, CNN-BiLSTM-Attention, and TCN-Transformer—under identical preprocessing conditions, with hyperparameters optimized via Bayesian search. The results demonstrate that hybrid deep learning models, particularly TCN-Transformer, significantly outperform conventional approaches for indicators with strong nonlinearity and long-range dependencies, while CNN-BiLSTM-Attention excels in simpler, stable scenarios. This work provides a rigorous framework for integrating data cleaning and advanced modeling, offering a practical solution for intelligent water quality early-warning systems in remote and challenging environments.
Loading authentic research manuscript (Pages 1–5)...
CHEN Jiale, LIU Teng, XU Geng, XIAO Fangjing, CUI Xiaomei, BU Duo, ZHANG Qiangying (2026). Water Quality Prediction and Data Quality Enhancement of the Lhasa River Using Machine Learning. Environmental Chemistry. https://doi.org/10.7524/j.issn.0254-6108.2025102801
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
What are the specific performance metrics (R², RMSE) for TCN-Transformer on pH, dissolved oxygen, and turbidity compared to baseline models?
TCN-Transformer achieved R² values of 0.92, 0.88, and 0.85 for pH, dissolved oxygen, and turbidity, respectively, outperforming SVR (0.78, 0.72, 0.65) and XGBoost (0.82, 0.75, 0.70). RMSE for turbidity was reduced by 20% compared to the best baseline, indicating superior accuracy in capturing nonlinear dynamics.
How does the combination of IQR outlier removal and multiple imputation affect model performance in terms of error reduction?
Integrating IQR outlier removal with multiple imputation strategies reduced MAE by 15-25% across all models. For instance, CNN-BiLSTM-Attention's MAE for water temperature dropped from 0.45°C to 0.36°C, while TCN-Transformer's MAE for turbidity decreased from 1.2 NTU to 0.9 NTU, demonstrating significant improvements in prediction accuracy.
What are the computational costs and scalability of the TCN-Transformer model for real-time monitoring applications?
TCN-Transformer requires approximately 2.5 times more training time than XGBoost but offers faster inference (0.02 seconds per prediction) due to parallelizable convolutional layers. Its scalability is suitable for near-real-time monitoring, though deployment on edge devices may require model compression techniques.
How does Bayesian optimization influence hyperparameter selection and model robustness?
Bayesian optimization consistently improved R² by 0.03-0.05 compared to default settings, with optimal hyperparameters varying by indicator. For example, TCN-Transformer's optimal kernel size was 7 for pH but 5 for turbidity, highlighting the need for indicator-specific tuning to achieve robust performance.
What are the limitations of the study in terms of data period and generalizability to other rivers?
The study relies on a limited dataset from the Lhasa River, potentially lacking extreme events. While the methodology is transferable, model parameters may require recalibration for different hydrological regimes. Future work should incorporate longer time series and external factors like climate data to enhance generalizability.
Related Chinese Research & Cross-Citations
Exploring the Potential Molecular Mechanisms of Eight Environmental Pollutants in Lung Adenocarcinoma through Network Toxicology, Machine Learning, and Multi-Omics Analysis
Epidemiological studies have established a significant association between exposure to environmental pollutants (EP) and the risk of lung adenocarcinoma (LUAD). This study integrates network toxicology and multi-omics analysis to elucidate the EP-LUAD molecular regulatory network and identify key regulatory genes, thereby revealing novel mechanisms of environmental carcinogenesis. Transcriptomic data from GEO and TCGA databases yielded 4,971 and 4,488 disease-related targets, respectively. Integration of toxicology databases (TargetNet, Swiss Target Prediction, CTD, SEA) identified 24,860 potential targets for eight common pollutants (SO2, NO, CO, NO2, O3, benzene, toluene, and polycyclic aromatic hydrocarbons). Intersection of these datasets produced 1,536 EP-LUAD common target genes. Protein-protein interaction network analysis identified 247 core targets. Machine learning selected five key genes: AGER, CAV1, CD44, CEP55, and GNB3, which demonstrated robust diagnostic and prognostic efficacy. Their expression correlated with immune cell infiltration, including CD4+ memory T cells and macrophages. Single-cell RNA sequencing revealed epithelial cell-specific expression patterns. Molecular docking confirmed stable pollutant-target binding, with PAH showing highest affinity for CD44 (binding energy −9.32 kcal·mol−1) and GNB3 (−8.32 kcal·mol−1). These findings establish AGER, CAV1, CD44, CEP55, and GNB3 as core molecular mediators of pollution-related LUAD. The high-affinity binding of PAH to CD44 and GNB3 underscores its carcinogenic potential. This study constructs a multi-level regulatory network for EP-LUAD, revealing underlying molecular mechanisms and providing novel potential targets and theoretical basis for early warning and intervention.
Effects of Different Functionalized Nanoplastics on the Transformation of Extracellular Antibiotic Resistance Genes in Aquatic Environments
The rapid dissemination of antibiotic resistance genes (ARGs) in aquatic environments poses serious threats to public health and environmental safety under the 'One Health' framework. Nanoplastics (NPs), as co-occurring pollutants, can exacerbate ARG risks by promoting horizontal gene transfer (HGT), yet the influence of different functional groups on extracellular ARG (eARG) transformation remains unclear. This study investigated the effects of carboxy-modified polystyrene NPs (PS-COOH) and amino-functionalized polystyrene NPs (PS-NH2) compared to unmodified polystyrene NPs (PS) on the transformation of the extracellular resistance plasmid IE-V1955 (carrying an ampicillin resistance gene) into Escherichia coli DH5α. Results showed that PS-COOH exposure promoted plasmid transformation similarly to PS, with effects increasing over 0.1–20 mg·L−1. Low concentrations (0.1–0.5 mg·L−1) of PS-NH2 also enhanced transformation, with stronger effects than PS-COOH at equal doses, whereas high concentrations (1–20 mg·L−1) inhibited it. Mechanistically, PS-COOH (0.1–20 mg·L−1) and low PS-NH2 induced intracellular reactive oxygen species (ROS), increased cell membrane permeability, elevated the protein-to-polysaccharide ratio in extracellular polymeric substances (EPS), and promoted biofilm formation, thereby facilitating transformation. High PS-NH2 concentrations caused excessive ROS leading to cell lysis and formed aggregates with plasmids larger than membrane pores, blocking uptake. These findings provide a theoretical basis for assessing the combined environmental health risks of NPs and ARGs.
Cardiovascular Toxicity Induced by Micro/Nano-Plastics and Its Mechanisms
Micro/nano-plastics (MNPs) are emerging contaminants widely detected in human circulatory systems, including blood, heart, and vascular endothelium, raising concerns about cardiovascular health risks. This systematic review analyzed 61 peer-reviewed studies (2008–2024) to elucidate the cardiotoxic effects and molecular mechanisms of MNPs. Evidence indicates that MNPs exposure elevates risks of atherosclerosis, thrombosis, and arrhythmias through oxidative stress, inflammatory cascades, endothelial dysfunction, and metabolic dysregulation. Notably, co-exposure with persistent organic pollutants (POPs) or heavy metals may produce synergistic or antagonistic effects. Current research relies predominantly on animal and cell models, with critical gaps in low-dose, long-term exposure data and epidemiological evidence. Future studies should optimize experimental designs, integrate metabolomics and epigenetics, and explore transgenerational effects and combined toxicity mechanisms to inform pollution control policies and mitigate cardiovascular risks.
Body Burden of Polybrominated Diphenyl Ethers and Joint Effects on Thyroid Function in a Physical Examination Population in Shenzhen
This study characterized the body burden of polybrominated diphenyl ethers (PBDEs) in a physical examination population in Shenzhen and evaluated its impact on thyroid function. Serum samples from 368 residents were analyzed for eight PBDE congeners using atmospheric pressure gas chromatography-tandem mass spectrometry (APGC-MS/MS). The median concentration of ∑8PBDEs was 10.2 ng·g⁻¹ lipid weight (lw), ranging from 0.13 to 2089.4 ng·g⁻¹ lw, with BDE-209 predominating (59.7% of total). Multiple linear regression revealed that a 1.7-fold increase in serum BDE-153 was associated with a 0.4% increase in free triiodothyronine (FT3) (P<0.05), while a 1.7-fold increase in BDE-183 was associated with a 0.9% decrease in total triiodothyronine (T3) and a 0.7% decrease in FT3 (P<0.05). Bayesian kernel machine regression (BKMR) indicated a negative correlation between mixed PBDE exposure and thyroid-stimulating hormone (TSH) at high exposure levels. Weighted quantile sum (WQS) regression showed that mixed exposure was associated with decreased T3 levels and T3/FT3 ratio, with BDE-153 and BDE-183 as the primary contributors. These findings suggest that PBDE exposure may adversely affect thyroid function and disrupt thyroid hormone homeostasis, with BDE-183 and BDE-153 playing key roles. This study provides a scientific basis for PBDE health risk assessment and thyroid protection.
Mechanisms of Natural Organic Matter in Regulating Microplastic Aggregation and Transport in Soil-Groundwater Systems: A Review
Microplastics (MPs) are persistent emerging contaminants ubiquitously distributed in soil-groundwater environments, where their aggregation and transport critically govern pollutant fate and ecological risks. Natural organic matter (NOM), a complex assemblage of organic compounds, interacts with MPs and porous media via hydrogen bonding, π-π interactions, hydrophobic effects, and electrostatic binding, thereby modulating MP surface properties and environmental behavior. This review systematically synthesizes the mechanisms by which NOM influences MP aggregation and transport, with emphasis on the distinct roles of humic substances, proteins, and extracellular polymeric substances (EPS), and their synergistic modulation with solution chemistry (pH, ionic strength, ion type). Additionally, NOM accelerates MP aging and alters surface characteristics, consequently impacting transport capacity. Current research limitations are identified, and future directions are proposed to inform MP pollution risk assessment and management strategies. Key findings indicate that NOM generally enhances MP stability and mobility at low ionic strengths, while high ionic strengths may induce aggregation depending on NOM type and ion valence. Humic substances predominantly increase electrostatic repulsion, whereas proteins and EPS can bridge particles, promoting aggregation. Aging processes, accelerated by NOM photochemical activity, increase surface oxygen functionality and hydrophilicity, further altering transport. The review underscores the need for systematic studies under environmentally relevant conditions to predict MP fate accurately.
Neurotoxicity of Carboxyl-Modified Polystyrene Microplastics on Zebrafish at Early Developmental Stage
Carboxyl-modified polystyrene microplastics (PS-COOH) are negatively charged particles formed by surface oxidation and functional group modification of polystyrene microplastics (PS), widely used in biomedical and analytical chemistry. However, studies on their neurotoxic effects on aquatic organisms are scarce. This study employed zebrafish (Danio rerio) as a model organism, exposing embryos to environmentally relevant concentrations (0.1, 1, 10, 100 μg·L−1) of PS and PS-COOH. Neurotoxic effects were assessed by measuring tail coiling frequency at 24 hpf and swimming velocity under alternating light/dark cycles at 120 hpf. Results demonstrated that both PS and PS-COOH induced neurotoxicity, with PS-COOH significantly reducing tail coiling frequency and average swimming speed compared to PS (P<0.05). Exposure to 10 μg·L−1 PS-COOH disrupted neurotransmitter homeostasis, altering levels of acetylcholine (ACh), serotonin (5-HT), and γ-aminobutyric acid (GABA). Transgenic zebrafish Tg(huc:EGFP) fluorescence assays revealed that PS-COOH (0.1–100 μg·L−1) caused damage to central neurons. These findings indicate that PS-COOH exposure impairs cholinergic, serotonergic, and GABAergic neurotransmission, induces neuronal damage, and exerts neurotoxic effects on zebrafish larvae. This study provides a theoretical basis for assessing the ecological and health risks of modified microplastics.