Biomass burning emission factors (BB EFs) exhibit significant variations across the savanna biome due to environmental conditions and fuel composition. Although we understand some factors contributing to this variability, quantifying their impact for integration into BB emission models has been limited by the absence of in situ EF measurements, especially concerning temporal dynamics. Traditional measurement methods are costly, time-consuming, and potentially biased toward either flaming (in airborne sampling) or smoldering (ground sampling) combustion products. To expand the spatiotemporal coverage of in situ measurements, we aimed to develop a versatile approach applicable to both prescribed and non-prescribed burns.
We conducted four measurement campaigns in South Africa's Kruger National Park (KNP) and three in Brazil's Estação Ecológica Serra Geral do Tocantins (EESGT) to create a drone (UAS)-based sampling system and a measurement strategy for savanna landscape fires. In KNP, we compared UAS bag samples with continuous emission measurements from a passing fire front using a measurement mast. This confirmed the UAS's suitability for measuring fire-averaged trace gas EFs and hinted at the potential use of lightweight particle sensors for aerosol EF measurements, which led to laboratory fire experiments for aerosol equipment calibration.
Beyond EF measurements, we utilized prescribed burning campaigns in KNP and EESGT to develop a method for mapping fuel load and burned area with a Micasense rededge multispectral camera. Subsequently, after developing and calibrating the EF measurement methodology, we captured the spatiotemporal variability of EFs and fuel conditions across the savanna biome. We targeted frequently burning vegetation types across a moisture availability gradient, conducting campaigns in both early dry season (EDS) and late dry season (LDS) for each vegetation type. Over four years, we measured EFs of CO, CO2, CH4, and NO in Botswana, Australia (xeric savannas), Mozambique, and Zambia (mesic savannas). We also assessed pre- and post-fire fuel parameters.
While some aspects of our approach could have been improved, such as site representativeness due to the challenges of finding safe, representative areas in the natural landscape, our extensive dataset on savanna fire EFs represents a significant step towards dynamic EF integration into global models. We identified factors impacting EFs, such as fractional tree cover and timing of measurements, and related them to fuel and meteorological conditions using satellite data.
To transform this EF measurement dataset into a modeling framework for global assessments, we experimented with various regression methods, including machine learning techniques. These models, informed by satellite data for spatiotemporal extrapolation, substantially enhanced the estimation of gas EFs across the savanna biome. They indicate a redistribution with higher EFs of reduced gases in mesic savannas and lower EFs in xeric savannas. These models will serve as the foundation for a spatiotemporal assessment of EFs in savannas, to be utilized in GFED5.