EUVIMEDEUVIMED entdecken
Crossref

A physically constrained and interpretable deep learning framework for PM2.5 inversion under sparse monitoring conditions in arid regions

Xinyue Zhang, Wen Ma, Jianli Ding, Pai Peng, Hongru Wang, Qianqian Man, Yanyi Yang, Nan Du

Environmental Pollution · 2026

DOI: 10.1016/j.envpol.2026.128233 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Environmental Pollution, 2026. Verfasst von Xinyue Zhang und 7 weiteren Autor:innen. Beteiligte Einrichtung: College of Geographical and Remote Sensing Sciences, Xinjiang University, Urumqi, 830017, China. Hauptthemen laut MEDLINE: Particulate Matter, Deep Learning, Environmental Monitoring, Air Pollutants. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 45 Literaturangaben. Gefördert von: Xinjiang University.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Monitoring fine particulate matter (PM2.5) in arid regions remains difficult due to sparse ground-based networks and frequent extreme pollution episodes. Conventional inversion models often fail to capture these sudden concentration spikes. To address this gap, we developed ResNet-DFL, a physically constrained and interpretable deep learning framework aimed at satellite-based PM2.5 estimation in data-scarce areas. We applied a distribution-aware learning strategy to reduce long-tail bias and integrated Shapley Additive Explanations (SHAP) with two-dimensional Partial Dependence Plots (2D-PDPs) to quantify key drivers and their nonlinear interactions. Validation using independent sparse-station scenarios showed that ResNet-DFL achieves a spatial R2 of 0.79 and a temporal R2 of 0.68. It consistently outperforms benchmark baseline models like Random Forest and Geographically and Temporally Weighted Regression, particularly during heavy pollution events. Our interpretability analysis shows that relative humidity, temperature, and NDVI are the dominant factors governing local pollution. Furthermore, 2D-PDPs confirm that low temperature combined with high humidity primarily triggers PM2.5 accumulation in this region. We also found that a vertically normalized aerosol optical depth (RAOD) is critical to link column-integrated satellite observations with near-surface PM2.5 by adjusting for boundary layer dynamics. Ultimately, ResNet-DFL provides a reliable tool for PM2.5 estimation in data-scarce arid regions, providing practical data support for local air quality management and targeted pollution control.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Particulate MatterDeep LearningEnvironmental MonitoringAir PollutantsAir PollutionClimate

Stichwörter der Autor:innen

Arid regionsInterpretabilityPM(2.5) estimationPhysically constrained deep learningSparse ground monitoring

Substanzen

Particulate MatterAir Pollutants

Autor:innen und Institutionen

  • College of Geographical and Remote Sensing Sciences, Xinjiang University, Urumqi, 830017, China.
  • College of Geographical and Remote Sensing Sciences, Xinjiang University, Urumqi, 830017, China. Electronic address: wenmm08@xju.edu.cn.
  • Xinjiang Institute of Technology, Aksu, 843099, China.

Förderung

  • Xinjiang University

Zitieren

APA

Zhang, X., Ma, W., Ding, J., Peng, P., Wang, H., Man, Q., Yang, Y., & Du, N. (2026). A physically constrained and interpretable deep learning framework for PM2.5 inversion under sparse monitoring conditions in arid regions. Environmental Pollution. https://doi.org/10.1016/j.envpol.2026.128233

Vancouver

Zhang X, Ma W, Ding J, Peng P, Wang H, Man Q, et al. A physically constrained and interpretable deep learning framework for PM2.5 inversion under sparse monitoring conditions in arid regions. Environmental Pollution. 2026. doi:10.1016/j.envpol.2026.128233

BibTeX
@article{Zhang2026,
  author = {Zhang, Xinyue and Ma, Wen and Ding, Jianli and Peng, Pai and Wang, Hongru and Man, Qianqian and Yang, Yanyi and Du, Nan},
  title = {A physically constrained and interpretable deep learning framework for PM2.5 inversion under sparse monitoring conditions in arid regions},
  journal = {Environmental Pollution},
  year = {2026},
  doi = {10.1016/j.envpol.2026.128233}
}
Kostenlos starten →