Purpose <p>Green roofs are promoted as nature-based solutions for urban climate mitigation, yet their net carbon sequestration potential across contrasting climates remains poorly constrained.</p> Methods <p>We quantified July–September CO₂ emissions and carbon (C) balances of extensive green roofs during the first season after construction and identified key climatic and design drivers. Roofs were established along a climatic gradient in Russia (Ekaterinburg, Moscow, Rostov-on-Don) using universal (15&#xa0;cm) and local substrates (15 and 25&#xa0;cm), planted with Sedum (on shallow substrates only) and native herbaceous communities (on both substrate depths). All treatments were replicated three times giving a total of 18 roof plots per city. CO₂ was measured twice a month using dynamic closed chambers; substrate temperature and moisture were monitored continuously.</p> Results <p>Mean CO₂ emissions ranged from 3.47 ± 1.51&#xa0;g C m⁻² day⁻¹ in Ekaterinburg to 7.65 ± 5.39&#xa0;g C m⁻² day⁻¹ in Moscow and 7.13 ± 6.53&#xa0;g C m⁻² day⁻¹ in Rostov-on-Don. Seasonal dynamics in Ekaterinburg and Moscow were controlled by temperature (Q₁₀= 2.86–3.44), while moisture dominated in Rostov-on-Don. Substrate type was the main driver, with compost-enriched universal substrates emitting more CO₂ than local substrates; substrate depth had a city-dependent effect, while plant community had no influence on emissions. Across all sites, CO₂ emissions exceeded plant C uptake into aboveground biomass, indicating that most systems were net C sources during the first growing season. Higher biomass on universal substrates did not translate into greater C sequestration due to elevated emissions.</p> Conclusions <p>Thus, climate-specific design is important to optimize green roof carbon performance.</p>