Skip to main content
Dryad

Limited increases in savanna carbon stocks over decades of fire suppression

Cite this dataset

Zhou, Yong et al. (2022). Limited increases in savanna carbon stocks over decades of fire suppression [Dataset]. Dryad. https://doi.org/10.5061/dryad.pg4f4qrr5

Abstract

Savannas cover a fifth of the land surface and contribute a third of terrestrial net primary production, accounting for three quarters of global area burned and over half of global fire-driven carbon emissions. Fire suppression and afforestation have been proposed as tools to increase carbon sequestration in these ecosystems. A robust quantification of whole-ecosystem carbon storage in savannas is lacking, however, especially under altered fire regimes. Here, we provide the first direct estimates of whole-ecosystem carbon response to over 60 years of fire exclusion in a mesic African savanna. We found that fire suppression increased whole-ecosystem carbon storage by only 35.4 ± 12% (mean ± standard error), even though tree cover increased by 78.9 ± 29.3%, corresponding to total gains of 23.0 ± 6.1 Mg C ha-1 at an average ~0.35 ± 0.09 Mg C ha-1 yr-1, more than an order of magnitude lower than previously assumed. Frequently burned savannas had substantial belowground carbon, especially in biomass and deep soils. These belowground reservoirs are not fully considered in afforestation or fire suppression schemes but may mean that the decadal sequestration potential of savannas is negligible, especially weighed against concomitant losses of biodiversity and function.

Methods

Study site: Kruger National Park (latitude: 22°20ʹ - 25°30ʹS; longitude: 31°10ʹ - 32°00ʹE) in South Africa

Experimental design: Kruger maintains one of a handful of long-term burning experiments in tropical savannas. The experimental burning plots (EBPs) were initiated in 1954, making them the longest-running fire ecology research project in African savannas. The EBPs are distributed across four different landscapes of Kruger (i.e., Mopani, Satara, Skukuza, Pretoriuskop) with different dominant tree species, parent materials, and rainfall. Each landscape can be considered as an independent factorial design with four replicates (hereafter, strings). Within each string, there are 12 treatments with the fire return interval of each treatment representing a different combination of frequency and season.  For this study, we selected the Pretoriuskop landscape receiving ~700 m rainfall, which broadly represents African savannas that have the potential to reach full tree cover. Among these 12 treatments, we selected plots burned every year in August (hereafter, annual) to represent an extreme fire regime; plots burned every three years in August (hereafter, triennial) to represent the near-natural fire return interval of African savannas; and plots that have not burned since 1954 (hereafter, unburned) to represent savannas with fire-suppressed status.

Light detection and ranging (LiDAR) data collection: Woody biomass was estimated using LiDAR. We used Riegl VUX-1LR LiDAR unit integrated onto a DJI Matrice 600 PRO unoccupied aerial system (UAS) to collect high-resolution airborne LiDAR data. We carried out the LiDAR survey during the middle of the wet season (i.e., January) of 2020 when vegetation was at full leaf-on stage. The flight altitude was 100 m above ground level, flight speed was 8 m s-1, and the LiDAR scan rate was 78.1 lines/second (see Supplementary Table 4 for other parameter settings). The UAS maintained consistent elevation above the ground by using 30 m × 30 m elevational data from the shuttle radar topography mission (SRTM) to adjust flight altitude in real time during the survey. All treatments within each string were surveyed with transects of identical heading to decrease the probability of introducing confounding variables and remote sensing artifacts created by differing survey methodologies or LiDAR scan directions.

Ground penetrating radar data collection: Coarse lateral root biomass from woody plants was estimated using the ground penetrating radar (GPR). GPR profiles were acquired using the Subsurface Interface Radar (SIR) System-4000 with 1.6 GHz shielded antenna and odometer wheels for position recording (Geophysical Survey Systems Inc., NH, USA). Prior to the survey, the grass layer was carefully removed to avoid any interference in the transmission of electromagnetic energy from antenna to soils. The survey was conducted during the dry season (September to November) of 2018 with soil water content less than 5%. At each 10 m × 10 m plot, GPR profiles were collected on a 20-cm grid. If a tree was present on a scanning line, the rest of the GPR profile was obtained from the opposite direction. The topography across all plots was relatively flat with minimal surface relief (< 5 cm). Extra care was taken to ensure the accurate position of each GPR profile with a guide rope and the difference in the length of GPR profile was less than 1% (i.e., 10 cm) of the supposed distance (i.e., 10 m).

Usage notes

This data is comprised of two components: (1) data and code for light detection and ranging (LiDAR) processing; (2) data and code for ground penetrating radar (GPR) processing. Please refer to the paper for more details on data analysis. 

Funding

Yale University

Harvard University

US Forest Service