Where the City Breathes Carbon: Block-Level Road CO₂ Mapping and Data-Driven EV Charger Placement in Delhi

Cities are usually counted in totals: one number for Delhi’s traffic carbon, one number for its charger shortage.Totals hide the question that actually matters, which is where. This paper maps Delhi’s road CO2 emissions block byblock, and then puts electric-vehicle chargers where the map says they will cut the most carbon. We combine a 500-meter gridded traffic emission dataset (CHETNA-Road) with city-level totals from Climate TRACE, fleet compositionfrom India’s vehicle registry (VAHAN), public charger locations from e-Amrit, and hourly air quality from CPCB,and assemble a panel of 264 city blocks observed across all 365 days of 2021. A gradient-boosted tree model(XGBoost) predicts each block’s daily emissions from 15 geographic, calendar, fleet, air-quality, and infrastructurefeatures, and is graded under six validation schemes, including tests where the model must predict blocks and seasonsit has never seen. The model reproduces the city’s daily cycle closely (temporal R^2 0.97) and generalizes to unseen half of the city at R^2 0.60 to 0.64, a difference we report as an honest measure of what a spatial emissions model can and cannot do. SHAP attributions show that urban structure, not air quality, drives the predictions: the size of the zone and its distance from the city center carry most of the signal, while the daily AQI adds almost none (correlation r = 0.046). The spatial estimates then feed a mixed-integer linear program that ranks all 264 blocks by their carbon-cut potential under a grid intensity of 0.72 kgCO2/kW h. Five well-placed chargers are worth 43 kilotons of CO2 per year; thirty are worth 220. A 2030 scenario, in which the grid decarbonizes to 0.45 kg/kW h and the electric fleet share climbs to roughly half the vehicles while travel grows 4 percent a year, reaches a 31 percent reduction in the road carbon of the blocks that go electric. The whole pipeline, the trained model and a liveinteractive atlas are public: the application runs at https://carbon-atlas-india.vercel.app and the code athttps://github.com/ayanavardhan11-cmd/ai-research