EventsOHOW 2026 – The 5th International Symposium on One Health, One World
Published
This submission belongs to the session Urban Safety & Disaster Mitigation (USDM) of the event OHOW 2026 – The 5th International Symposium on One Health, One World
Published date
18 Nov, 2026
Contribution&Funding
Conceptualization,
[Pallavi Chaurasia] ; [Wataru Takeuchi]
Methodology,
[Pallavi Chaurasia]
Validation,
[Pallavi Chaurasia]
Formal analysis,
[Pallavi Chaurasia]
Investigation,
[Pallavi Chaurasia]
Writing - original draft,
[Pallavi Chaurasia]
Writing - review,
[editing] ; [Wataru Takeuchi] ; [Prakhar Misra]
Supervision,
[Wataru Takeuchi] ; [Prakhar Misra]
Funding: --
Citation
Pallavi Chaurasia, Wataru Takeuchi, Prakhar Misra, Scale-Dependent Transferability and Predictor Limitations in Satellite-Based Ground-Level Ozone Estimation across the Indo-Gangetic Plain, in Proceedings of OHOW 2026 – The 5th International Symposium on One Health, One World, Hanoi, 26 November–28 November 2026, MDPI: Basel, Switzerland
Share
Email
Facebook
Twitter
LinkedIn

Scale-Dependent Transferability and Predictor Limitations in Satellite-Based Ground-Level Ozone Estimation across the Indo-Gangetic Plain

image
Prakhar Misra 1
1. Civil Engineering, Indian Institute of Technology Roorkee, Roorkee, 247667, India
2. Civil Engineering, The University of Tokyo, Tokyo, 1538505, Japan
Abstract

Ground-level ozone (GLO) is a secondary air pollutant across the Indo-Gangetic Plain, particularly during pre-monsoon season. Ground-monitoring stations are sparse, leaving areas without observations. In our dataset, 16.8% of monitored pre-monsoon days exceeded the WHO Maximum Daily 8-hour Average (MDA8) ozone limit of 100 µg m⁻³. Satellite observations can improve spatial coverage, but their reliability beyond monitoring sites remains uncertain. This study assessed MDA8 GLO for March-June (2021-2025) using CPCB observations, TROPOMI NO2, HCHO, and total ozone, meteorology, land cover, elevation, population density, and temporal indicators. The missing satellite retrievals were gap-filled without using GLO observations. Six predictor configurations were evaluated to assess the contributions of environmental variables, neighbouring observations, residual information, and station’s own lagged ozone. XGBoost, LightGBM and LSTM models were evaluated using random splits, leave-one-station-out (LOSO) and leave-one-region-out (LORO) validation. The environment-only XGBoost achieved R² = 0.48 under random splitting, but its performance declined at unseen stations (LOSO R² = −0.18) and unseen regions (LORO R² = −0.21). In contrast, incorporating station’s lagged GLO improved performance, with XGBoost reaching LOSO and LORO R² of 0.76 and 0.73, comparable to LSTM (0.77 and 0.74). The model skill improved with spatial aggregation, with spatial R² increasing from −0.03 at monitoring sites to 0.62 at city scale, 0.79 at 0.25° resolution, and 0.83 at state scale. The regional mapping at 0.25 ° resolution showed that 85.8% of pixels fell within the model’s area of applicability. Higher GLO concentrations were around Delhi, peaked during April-May, and were highest in 2022 across most of the IGP. By testing transferability across stations, regions and scales, this study defines where satellite-based GLO estimates are dependable. The findings support regional GLO mapping for exposure assessment and air-quality management strategies. Future work should strengthen ground observations and extend the framework beyond premonsoon season.

Keywords
Ground-level ozone Indo-Gangetic Plain
MDA8
satellite remote sensing
machine learning
spatial transferability
area of applicability
Training-Free Detection of Buried Pipes Using Hyperbola Apex Extraction and PCA-Based Line Detection on Vehicle-Mounted GPR Data
Underground Cavity Detection Using Elastic-Wave Frequency Responses for Road Collapse Risk Reduction: Laboratory Investigation and Full-Scale Evaluation