Machine Learning in Environmental Research: Common Pitfalls and Best Practices

Environmental Science & Technology, Journal Year: 2023, Volume and Issue: 57(46), P. 17671 - 17689

Published: June 29, 2023

Machine learning (ML) is increasingly used in environmental research to process large data sets and decipher complex relationships between system variables. However, due the lack of familiarity methodological rigor, inadequate ML studies may lead spurious conclusions. In this study, we synthesized literature analysis with our own experience provided a tutorial-like compilation common pitfalls along best practice guidelines for research. We identified more than 30 key items evidence-based based on 148 highly cited articles exhibit misconceptions terminologies, proper sample size feature size, enrichment selection, randomness assessment, leakage management, splitting, method selection comparison, model optimization evaluation, explainability causality. By analyzing good examples supervised reference modeling paradigms, hope help researchers adopt rigorous preprocessing development standards accurate, robust, practicable uses applications.

Language: Английский

---

Attribution of Air Quality Benefits to Clean Winter Heating Policies in China: Combining Machine Learning with Causal Inference

Environmental Science & Technology, Journal Year: 2023, Volume and Issue: 57(46), P. 17707 - 17717

Published: Feb. 1, 2023

Heating is a major source of air pollution. To improve quality, range clean heating policies were implemented in China over the past decade. Here, we evaluated impacts winter and on quality using novel, observation-based causal inference approach. During 2015-2021, causally increased annual PM2.5, daily maximum 8-h average O3, SO2 by 4.6, 2.5, 2.3 μg m-3, respectively. From 2015 to 2021, PM2.5 Beijing surrounding cities (i.e., "2 + 26" cities) decreased 5.9 m-3 (41.3%), whereas that other northern only 1.2 (12.9%). This demonstrates effectiveness stricter cities. Overall, caused mainland reduce 1.9 from potentially avoiding 23,556 premature deaths 2021.

Language: Английский

---

Contributions of various driving factors to air pollution events: Interpretability analysis from Machine learning perspective

Environment International, Journal Year: 2023, Volume and Issue: 173, P. 107861 - 107861

Published: March 1, 2023

The air quality in China has been improved substantially, however fine particulate matter (PM2.5) still remain at a high level many areas. PM2.5 pollution is complex process that attributed to gaseous precursors, chemical, and meteorological factors. Quantifying the contribution of each variable can facilitate formulation effective policies precisely eliminate pollution. In this study, we first used decision plot map out Random Forest (RF) model for single hourly data set constructed framework analyzing causes using multiple interpretable methods. Permutation importance was qualitatively analyze effect on concentrations. sensitivity secondary inorganic aerosols (SIA): SO42-, NO3- NH4+ verified by Partial dependence (PDP). Shapley Additive Explanation (Shapley) quantify drivers behind ten events. RF accurately predict concentrations, with determination coefficient (R2) 0.94, root mean square error (RMSE) absolute (MAE) 9.4 μg/m3 5.7 μg/m3, respectively. This study revealed order SIA NH4+＞NO3-＞SO42-. Fossil fuel biomass combustion may be contributing factors events Zibo 2021 autumn-winter. contributed 19.9-65.4 among (APs). K, NO3-, EC OC were other main drivers, 8.7 ± 2.7 6.8 7.5 3.6 5.8 2.5 2.0 Lower temperature higher humidity vital promoted formation NO3-. Our provide methodological precise management.

Language: Английский

---

Explainable ensemble machine learning revealing the effect of meteorology and sources on ozone formation in megacity Hangzhou, China

The Science of The Total Environment, Journal Year: 2024, Volume and Issue: 922, P. 171295 - 171295

Published: Feb. 27, 2024

Language: Английский

---

Impacts of meteorology and precursor emission change on O3 variation in Tianjin, China from 2015 to 2021

Journal of Environmental Sciences, Journal Year: 2022, Volume and Issue: 126, P. 506 - 516

Published: March 14, 2022

Language: Английский

---

Enhanced ozone pollution in the summer of 2022 in China: The roles of meteorology and emission variations

Atmospheric Environment, Journal Year: 2023, Volume and Issue: 301, P. 119701 - 119701

Published: March 9, 2023

Language: Английский

---

Machine Learning Models for Inverse Design of the Electrochemical Oxidation Process for Water Purification

Environmental Science & Technology, Journal Year: 2023, Volume and Issue: 57(46), P. 17990 - 18000

Published: May 16, 2023

In this study, a machine learning (ML) framework is developed toward target-oriented inverse design of the electrochemical oxidation (EO) process for water purification. The XGBoost model exhibited best performances prediction reaction rate (k) based on training data set relevant to pollutant characteristics and conditions, indicated by Rext2 0.84 RMSEext 0.79. Based 315 points collected from literature, current density, concentration, gap energy (Egap) were identified be most impactful parameters available EO process. particular, adding conditions as input features allowed provision more information an increase in sample size improve accuracy. feature importance analysis was performed revealing pattern interpretation using Shapley additive explanations (SHAP). ML-based generalized random case tailoring optimum with phenol 2,4-dichlorophenol (2,4-DCP) serving pollutants. resulting predicted k values close experimental verification, accounting relative error lower than 5%. This study provides paradigm shift conventional trial-and-error mode data-driven advancing research development time-saving, labor-effective, environmentally friendly strategy, which makes purification efficient, economic, sustainable context global carbon peaking neutrality.

Language: Английский

---

Understanding the relationship between 2D/3D variables and land surface temperature in plain and mountainous cities: Relative importance and interaction effects

Building and Environment, Journal Year: 2023, Volume and Issue: 245, P. 110959 - 110959

Published: Oct. 20, 2023

Language: Английский

---

The application of machine learning to air pollution research: A bibliometric analysis

Ecotoxicology and Environmental Safety, Journal Year: 2023, Volume and Issue: 257, P. 114911 - 114911

Published: April 15, 2023

Machine learning (ML) is an advanced computer algorithm that simulates the human process to solve problems. With explosion of monitoring data and increasing demand for fast accurate prediction, ML models have been rapidly developed applied in air pollution research. In order explore status applications research, a bibliometric analysis was made based on 2962 articles published from 1990 2021. The number publications increased sharply after 2017, comprising approximately 75% total. Institutions China United States contributed half all with most research being conducted by individual groups rather than global collaborations. Cluster revealed four main topics application ML: chemical characterization pollutants, short-term forecasting, detection improvement optimizing emission control. rapid development algorithms has capability characteristics multiple analyze reactions their driving factors, simulate scenarios. Combined multi-field data, are powerful tool analyzing atmospheric processes evaluating management quality deserve greater attention future.

Language: Английский

---

Trends of source apportioned PM2.5 in Tianjin over 2013–2019: Impacts of Clean Air Actions

Environmental Pollution, Journal Year: 2023, Volume and Issue: 325, P. 121344 - 121344

Published: March 4, 2023

Language: Английский

---