CT-ACE v1.0: Global 0.25° Monthly Anthropogenic CO₂ Emissions Derived from Carbon-Transformer(2019-01 to 2026-04)

CT-ACE v1.0 (Carbon-Transformer Anthropogenic CO2 Emissions) is a global monthly gridded dataset of anthropogenic CO2 emissions at 0.25° spatial resolution covering 2019-01 to 2026-04. Data are provided as GeoTIFF rasters in EPSG:4326 geographic coordinates on a global 1440 × 720 latitude-longitude grid spanning -180° to 180° longitude and -90° to 90° latitude. Raster values are stored as float32. The unit is tonnes grid cell⁻¹ month⁻¹. The dataset is generated by Carbon-Transformer, a Swin-Transformer-based framework that maps satellite tropospheric NO2 observations and auxiliary predictors to anthropogenic CO2 emissions. Model inputs include Sentinel-5P/TROPOMI tropospheric NO2, VIIRS nighttime lights, and ERA5-Land 2-m dew point temperature. The model is pretrained on GRACED daily emissions and fine-tuned on EDGAR monthly inventories. This release contains products generated under two inference configurations. Products from 2019-01 to 2024-12 are produced in an idealized hindcast mode (Y_hist → EDGAR), in which the previous-month emission field is taken from the inventory to isolate the model’s ability to map current predictors to emissions without recursive drift. Products from 2025-01 to 2026-04 are produced in an operational rolling inference mode (Y_hist → Prediction), in which the model feeds back its own previous prediction as the history term, mimicking real-world deployment when the latest inventory is unavailable. Filename conventions are as follows: CT-ACE_CO2_GLOBAL_YYYY-MM_hindcast_0p25deg_v1.tif for hindcast products, and CT-ACE_CO2_GLOBAL_YYYY-MM_rolling_baseYYYYMM_0p25deg_v1.tif for rolling products. In these names, hindcast indicates inventory-forced history, rolling indicates prediction-fed history, baseYYYYMM indicates the inventory month used to initialize the rolling sequence, and 0p25deg denotes the 0.25° spatial grid. This dataset is intended for near-real-time monitoring of global anthropogenic CO2 emissions. Users should account for the mode-dependent production strategy when interpreting temporal continuity and comparing files across periods.