Landscape fires, often referred to as biomass burning (BB), emit substantial amounts of (greenhouse) gases and aerosols into the atmosphere each year. Frequently burning savannas, mostly in Africa, Australia, and South America are responsible for over 60 % of total BB carbon emissions. Compared to many other sources of emissions, fires have a strong seasonality. Previous research has identified the mitigation potential of prescribed fires in savanna ecosystems; by burning cured fuels early in the dry season when landscape conditions still provide moist buffers against fire spread, fires are in general smaller, patchier, and less intense. While it is widely accepted that burned area (BA) and the total carbon consumed are lower when fires are ignited early in the dry season, little is known about the intraseasonal variability of emission factors (EFs). This is important because potentially, higher EFs in the early dry season (EDS) could offset some of the carbon benefits of EDS burning. Also, a better understanding of EF intraseasonal variability may improve large-scale BB assessments, which to date rely on temporally static EFs. We used a sampling system mounted on an unmanned aerial vehicle (UAV) to sample BB smoke in the EstaçÃ£o Ecológica Serra Geral do Tocantins in the Brazilian states of Tocantins and Bahia. The protected area contains all major Cerrado vegetation types found in Brazil, and EDS burning has been implemented since 2014. Over 800 smoke samples were collected and analysed during the EDS of 2018 and late dry season (LDS) of 2017 and 2018. The samples were analysed using cavity ring-down spectroscopy, and the carbon balance method was used to estimate CO.sub.2, CO, CH.sub.4, and N.sub.2 O EFs. Observed EF averages and standard deviations were 1651 (Â±50) g kg.sup.-1 for CO.sub.2, 57.9 (Â±28.2) g kg.sup.-1 for CO, 0.97 (Â±0.82) g kg.sup.-1 for CH.sub.4, and 0.096 (Â±0.174) g kg.sup.-1 for N.sub.2 O. Averaged over all measured fire prone Cerrado types, the modified combustion efficiency (MCE) was slightly higher in the LDS (0.961 versus 0.956), and the CO and CH.sub.4 were 10 % and 2.3 % lower in the LDS compared to the EDS. However, these differences were not statistically significant using a two-tailed t test with unequal variance at a 90 % significance level. The seasonal effect was larger in more wood-dominated vegetation types. N.sub.2 O EFs showed a more complex seasonal dependency, with opposite intraseasonal trends for savannas that were dominated by grasses versus those with abundant shrubs. We found that the N.sub.2 O EF for the open Cerrado was less than half the EF suggested by literature compilations for savannas. This may indicate a substantial overestimation of the contribution of fires in the N.sub.2 O budget. Overall, our data imply that in this region, seasonal variability in greenhouse gas emission factors may offset only a small fraction of the carbon mitigation gains in fire abatement programmes.