Spatio-temporal variability in ecosystem properties and biodiversity in relation to fire in Sebangau National Park, Central Kalimantan, Indonesia, 2003-2019

Data were collected from the Natural Laboratory of Peat-swamp Forest special research zone in Sebangau National Park, Central Kalimantan, Indonesia. A long history of research here has enabled compilation of 16 years of data on ecosystem properties (17 variables) and biodiversity (21 variables). The site is an ombrogenous, non-masting peat-swamp forest, which experienced selective logging over a ~40 year period by timber concessions until 1997 and illegal hand-loggers until 2004. This past illegal logging was associated with creation of numerous small canals, causing peat drainage and making the peatland vulnerable to fire during the dry season. Parts of the forest have therefore burned intermittently, creating a mosaic of predominantly unburned forest interspersed with areas burned at different times. The forest is also subject to indirect impacts of fire within the wider landscape, in particular through haze exposure. This context has thus enabled assessment of direct impacts through spatial comparison of ecosystem properties and biodiversity in areas with different fire history (new burn, old burn and unburned forest), plus temporal assessments of the indirect impacts of fire incidence within the wider landscape on ecosystem properties and biodiversity in unburned forest areas.

Regarding the spatial comparison, data were collected for 27 ecological components across 181 sampling locations between April 2017 and September 2021. These locations represent areas affected by a recent major burn event in 2015 (“new burn”; N=72), those recovering from fires up to and including 2006 (“old burn”; N=27), and from forest areas with no history of fire (“unburned”; N=82). Data on the following variables were collected from each of the three (or in some cases two) different fire treatments: ecosystem properties, including microclimate (temperature), vegetation (canopy/fern/grass cover, density of pitcher plants/large Pandanaceae plants/lianas/seedlings/saplings/trees) and forest structure (tree height, aboveground live biomass); and biodiversity, including species richness and abundance for trees, Odonata (split into Anisoptera and Zygoptera), fruit-feeding butterflies (Lepidoptera: Papilionoidea), amphibians, reptiles and avian-focused soundscape indices (Bioacoustic Index, Acoustic Complexity Index, Acoustic Diversity Index and Normalised Difference Soundscape Index). These spatial datasets comparing new burn, old burn and unburned forest treatments are represented as both effect sizes resulting from a hierarchical mixed effects meta-analysis and proportional changes resulting from a generalized linear mixed-effects model.

Regarding the temporal assessments, ecological time-series datasets were collected over 16-year period data (September 2003-December 2019), thus capturing variations in annual fire regimes typical of the region. All temporal data collection periods included at least two megafire events. The datasets encompassed 578 temporally-replicated surveys across 236 sampling locations for nine ecological components (range: 15-173 survey months per sample location). Time-series data were collected on the following variables: ecosystem properties, including river pH, forest productivity (leaf-fall), tree leaf flush and reproductive phenology, fruit-feeding butterflies (Lepidoptera: Papilionoidea), river fish, ground-dwelling birds and medium-large mammals. For ecosystem properties, these temporal datasets are represented as time-series data aggregated by site and season, and for biodiversity as occupancy detection matrices.

The accuracy of data collection was ensured through comprehensive observer training, including regular refresher training, plus maintaining consistency of data collection personnel throughout data collection periods. All data files were screened for potential outliers, typographical errors and missing data, which were resolved with the help of relevant field team members. Sound file spectrograms were scanned visually to identify potential disruptions that dominated the sound profile and may skew the acoustic indices (e.g., heavy rain), with disrupted files subsequently excluded from the analyses. Species identification protocols for biodiversity data from different taxonomic groups follow those used previously at the site, supported by use of locally-relevant identification guides and consultation with external experts. All analyses conducted were specified within a Bayesian framework, implemented in rstan (hierarchical mixed-effects meta-analysis) and JAGS (all GLMMs and occupancy models) called through R version 4.0.2.