Hourly NOx Emission Inventories over China and the Greater Bay Area (2017–2022) Derived from a Deep Learning Surrogate-Based Inversion Framework

This dataset provides hourly NOx emission inventories over China and the Greater Bay Area (GBA) from 2017 to 2022, derived through a deep learning surrogate-based inversion framework. The emissions are organized into four nested domains at horizontal resolutions of 27 km (covering mainland China), 9 km (covering southern China), 3 km, and 1 km (both covering the GBA). The inversion was performed using a Temporal U-Net surrogate model trained to emulate the WRF-CAMx chemical transport model. Hourly NOx emissions were iteratively updated by minimizing the loss between surrogate-predicted NO2 concentrations and ground-based NO2 observations. The dataset is intended to support air quality modeling, emission control policy assessment, atmospheric chemistry research, and downstream applications such as data assimilation and exposure estimation.