Wind-constrained diagnosis of 500 m urban CO2 emission spatial proxies using PAS observations and WRF-Chem simulations: a Hefei case study

Abstract

High-resolution urban CO 2 emission inventories increasingly rely on spatial proxies, such as road networks, built-up areas, and industrial or energy-related information, to allocate emissions to fine grids. However, whether these static proxies are consistent with ground-based CO 2 observations and model–observation discrepancies under time-varying atmospheric transport conditions remains uncertain. Using observations from ten photoacoustic spectroscopy (PAS) sites over a single 24 h period in Hefei, China, we developed a wind-constrained diagnostic framework for examining 500 m urban CO 2 spatial proxies. Anthropogenic CO 2 emissions for WRF-Chem were generated using a mass-conserving MEIC-to-500 m proxy-based allocation. Normalized road-related, built-up/residential, and industrial/energy-related proxy intensity layers were converted into station-hour upwind exposure metrics using WRF-Chem wind fields and evaluated against PAS-observed CO 2 enhancement anomalies and WRF-Chem residual anomalies. The WRF-Chem simulation produced a mean CO 2 concentration of 444.23 ppmv, compared with a PAS mean of 438.98 ppmv, yielding a mean residual of −5.25 ppmv. Road-related and built-up/residential proxy exposures showed weak positive associations with PAS-observed CO 2 enhancement anomalies, whereas the industrial/energy-related proxy showed no stable enhancement relationship. In the full-day analysis, all three proxy categories exhibited negative associations with WRF-Chem residual anomalies, with the strongest residual association for the industrial/energy-related proxy. Hour-level cluster bootstrap and false-discovery-rate correction supported five of the six full-day proxy–CO 2 associations. However, the negative proxy–residual relationships were concentrated under nighttime and low-PBLH conditions and largely disappeared under daytime and high-PBLH conditions. Parameter sensitivity, leave-one-site-out tests, and wind-direction perturbation analyses further showed that the full-day diagnostic patterns were not determined solely by a single kernel configuration, monitoring site, or wind-direction realization. The proposed method should be interpreted as a screening-level diagnostic framework rather than an emission inversion, concentration correction, or independent validation of proxy accuracy. The results indicate that the apparent proxy–residual relationships are strongly modulated by atmospheric transport and mixing conditions, and the conclusions are limited to the single-day case study.