A hybrid urban gridded inventory & sensor-based approach for source apportionment (HUGISS)
Abstract
A Hybrid Urban Gridded Inventory and Sensor-based Source Apportionment (HUGISS) model is presented for particulate matter source identification by combining gridded emission inventory (EI) estimates with low-cost sensors (LCS) observations. The HUGISS model integrates Gaussian distance decay and wind-directional weighting of gridded emissions with temporal activity profiles and EI priors within a ridge regression approach to estimate source contributions from spatially distributed LCS observations. Ten LCS units were deployed, including two anchor sensors representing static and dynamic emission environments. Pre-deployment colocation with a reference instrument enabled calibration and validation using statistical and machine learning (ML) models. Calibration results indicated strong linear responses, with complex ML models offering less than 3% additional improvement, confirming the sensors' reliability. Overall, static and dynamic sources were found to contribute, on average, 46-60% and 40-53%, respectively, to ambient PM 2.5 concentrations across the various LCS sites, with residuals remaining within ±5%. Seasonal and diurnal analysis showed dominance of dynamic sources during the (dry, June-October) season (52%) and static sources during the (wet, November-April) season, reflecting emission variations. By combining gridded EI data, temporal activity profiles and spatially distributed LCS measurements, HUGISS enables detailed characterization of localized variability across spatial and temporal scales. Model performance (R 2 w > 0.80; relative SSE values remained low) verified the robustness and physical coherence of the HUGISS approach. Applied in a data-scarce urban setting with limited reference monitoring and source information, HUGISS demonstrates a scalable, transferable and cost-effective approach for spatially and temporally resolved source apportionment using distributed LCS measurements supported by EI information.
Related
- Machine Learning in Contaminated Soil Remediation: From Monitoring, Source Apportionment, and Risk Assessment to Practical Applications.
- From real-time to long-term source apportionment of PM 10 using high-time-resolution measurements of aerosol physical properties: methodology and example application at an urban background site (Aosta, Italy)
- Chemical characterization and source apportionment of PM ₁₀ in Belgrade, Serbia: influence of local and regional anthropogenic and natural sources
- Isotopic composition of aerosol iron from anthropogenic sources: implications for source apportionment of aerosol iron
- Comprehensive analysis of PM10 in the metropolitan area of São Paulo – Part I: Chemical characterisation and source apportionment
BibTeX
@article{sharma2026hybrid,
title = {{A hybrid urban gridded inventory & sensor-based approach for source apportionment (HUGISS)}},
author = {Rajat Sharma and Erwann Rayssac and Andry Razakamanantsoa and Agnès Jullien},
journal = {City and Environment Interactions},
year = {2026},
month = sep,
doi = {10.1016/j.cacint.2026.100473},
url = {https://hal.science/hal-05755739v1},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - A hybrid urban gridded inventory & sensor-based approach for source apportionment (HUGISS)
AU - Rajat Sharma
AU - Erwann Rayssac
AU - Andry Razakamanantsoa
AU - Agnès Jullien
PY - 2026
DA - 2026/09/16
JO - City and Environment Interactions
DO - 10.1016/j.cacint.2026.100473
UR - https://hal.science/hal-05755739v1
AB - A Hybrid Urban Gridded Inventory and Sensor-based Source Apportionment (HUGISS) model is presented for particulate matter source identification by combining gridded emission inventory (EI) estimates with low-cost sensors (LCS) observations. The HUGISS model integrates Gaussian distance decay and wind-directional weighting of gridded emissions with temporal activity profiles and EI priors within a ridge regression approach to estimate source contributions from spatially distributed LCS observations. Ten LCS units were deployed, including two anchor sensors representing static and dynamic emission environments. Pre-deployment colocation with a reference instrument enabled calibration and validation using statistical and machine learning (ML) models. Calibration results indicated strong linear responses, with complex ML models offering less than 3% additional improvement, confirming the sensors' reliability. Overall, static and dynamic sources were found to contribute, on average, 46-60% and 40-53%, respectively, to ambient PM 2.5 concentrations across the various LCS sites, with residuals remaining within ±5%. Seasonal and diurnal analysis showed dominance of dynamic sources during the (dry, June-October) season (52%) and static sources during the (wet, November-April) season, reflecting emission variations. By combining gridded EI data, temporal activity profiles and spatially distributed LCS measurements, HUGISS enables detailed characterization of localized variability across spatial and temporal scales. Model performance (R 2 w > 0.80; relative SSE values remained low) verified the robustness and physical coherence of the HUGISS approach. Applied in a data-scarce urban setting with limited reference monitoring and source information, HUGISS demonstrates a scalable, transferable and cost-effective approach for spatially and temporally resolved source apportionment using distributed LCS measurements supported by EI information.
SN - 2590-2520
ER -