City-level carbon emissions data in Southeast Asia from 2000 to 2020.

City-resolved CO₂ emissions are needed to support urban mitigation planning and to track progress toward climate targets, yet long-term, consistent administrative-unit inventories remain scarce in Southeast Asia. We develop an annual CO₂ emissions dataset for 4,413 GADM admin-3 units across 11 Southeast Asian countries for 2000-2020. We downscale EDGAR admin-2 emission totals using a proxy-based allocation framework that integrates nighttime lights, impervious surface information, and urban-rural settlement distributions to represent economic activity, built form, and consumption-related patterns. To address strong regional heterogeneity and uneven sample sizes, we implement an adaptive dual-model strategy that selects province-specific or pooled ("hybrid") models using cross-validated performance. Among evaluated algorithms, gradient boosting provides the best out-of-fold fit. Final city-level estimates are obtained by normalizing predicted intensities within each admin-2 unit and enforcing a strict mass-balance constraint so that city totals exactly match EDGAR baselines. The resulting panel supports analyses of urban emission dynamics, hotspot identification, and policy-relevant comparison across Southeast Asia.