Physics-constrained inverse neural estimation (PINE) of NO x emission spatial patterns from TROPOMI NO 2 columns over the North China Plain

Satellite NO 2 columns constrain where nitrogen oxides (NO x ) are emitted, but conventional inversions are costly or depend on existing emission products. We present physics-constrained inverse neural estimation (PINE), applied to NO x as PINE-NO x : a neural encoder maps Sentinel-5P/TROPOMI NO 2 columns to emissions, and a fixed, differentiable two-dimensional transport–chemistry decoder maps them back. We retrieve annual, monthly and overpass-time daily emission patterns over the North China Plain at 0.1° for 2023, with the annual total normalized to the EDGAR inventory; the observations adjust the daily relative pattern around this prior, which still sets the magnitude and large-scale structure. On 54 validation days used only to select the training epoch, the retrieved emissions raise the log-space spatial correlation between simulated and observed columns from 0.332 (EDGAR) to 0.856 (observation reconstruction, not emission accuracy). PINE-NO x places less emission than EDGAR in the strongest-emitting cells, but a synthetic test with known emissions gives a similar decrease, so this shift is primarily a structural artefact, most likely of this decoder, not an observed signal. Relative to MEIC, which the inversion does not use, 13.5 % of July emissions are placed in different 0.5° cells, against 18.5 % in the synthetic test and 20.5 % for EDGAR; January shows no clear gain. The 1 σ uncertainty of annual 0.5° emission shares is at least 17.5 % (structural error excluded), mostly from the prior inventory. In a one-month WRF-Chem simulation, the retrieved emissions bring simulated columns closer to TROPOMI but do not improve agreement with surface NO 2 stations.

Authors

Institutions

Publication Details

Journal
Atmospheric chemistry and physics
Published
2026-11-16
DOI
https://doi.org/10.5194/acp-26-16125-2026
Primary Topic
Atmospheric chemistry and aerosols
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Physics-constrained inverse neural estimation (PINE) of NO x emission spatial patterns from TROPOMI NO 2 columns over the North China Plain

Yubing Pan, Daren Lyu, Yinan Wang
1 links
Atmospheric chemistry and physics
Atmospheric chemistry and aerosols
article

Physics-constrained inverse neural estimation (PINE) of NO x emission spatial patterns from TROPOMI NO 2 columns over the North China Plain

Yubing Pan, Daren Lyu, Yinan Wang
article en

Abstract

Satellite NO 2 columns constrain where nitrogen oxides (NO x ) are emitted, but conventional inversions are costly or depend on existing emission products. We present physics-constrained inverse neural estimation (PINE), applied to NO x as PINE-NO x : a neural encoder maps Sentinel-5P/TROPOMI NO 2 columns to emissions, and a fixed, differentiable two-dimensional transport–chemistry decoder maps them back. We retrieve annual, monthly and overpass-time daily emission patterns over the North China Plain at 0.1° for 2023, with the annual total normalized to the EDGAR inventory; the observations adjust the daily relative pattern around this prior, which still sets the magnitude and large-scale structure. On 54 validation days used only to select the training epoch, the retrieved emissions raise the log-space spatial correlation between simulated and observed columns from 0.332 (EDGAR) to 0.856 (observation reconstruction, not emission accuracy). PINE-NO x places less emission than EDGAR in the strongest-emitting cells, but a synthetic test with known emissions gives a similar decrease, so this shift is primarily a structural artefact, most likely of this decoder, not an observed signal. Relative to MEIC, which the inversion does not use, 13.5 % of July emissions are placed in different 0.5° cells, against 18.5 % in the synthetic test and 20.5 % for EDGAR; January shows no clear gain. The 1 σ uncertainty of annual 0.5° emission shares is at least 17.5 % (structural error excluded), mostly from the prior inventory. In a one-month WRF-Chem simulation, the retrieved emissions bring simulated columns closer to TROPOMI but do not improve agreement with surface NO 2 stations.

Atmospheric chemistry and physicsVol. 26(22)
China Meteorological Administration (CN), Chinese Academy of Sciences (CN), Institute of Atmospheric Physics (CN)
National Natural Science Foundation of China
Climate action
Atmospheric chemistry and aerosols
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon