Reproducibility of Charcoal Count Data: Empirical Analysis of an Implicit Assumption in Peak Detection Methodology

Reconstructions of fire histories based on analysis of sub-fossil charcoal from sediment cores often rely on a peak detection algorithm. These algorithms identify statistically significant anomalies in charcoal influx relative to changing background. The CharAnalysis program (Higuera et al., 2009, Ecological Monographs 79, 201-219.) is currently among the most well recognized. Factors such as charcoal source area, considerations relevant to timeseries data, and numerical criteria used for objective discrimination of charcoal peaks have been evaluated in several studies, and methodological refinements have been made. However, a critical assumption has remained untested: that the variability of charcoal concentration across a given sedimentary horizon is less than the threshold values used to determine peak significance.

Here we present analysis of replicated within-level samples from three separate sediment cores. The cores are from differing depositional environments (deep lake, shallow lake, wetland) and encompass distinct sediment types (organic-rich lacustrine mud, carbonate mud, peat). Twenty samples were taken from thirty contiguous levels in each core (n=1,800). The data was resampled to construct "Synthetic cores" designed to test for the impact of count variability and/or sample volume on peak detection.

Comparison of count variability by level to significance thresholds shows that 97% of the within-level samples included count differences large enough to be considered as having less than a 5% probability of originating from the same population. The variation in the replicated data resulted in inconsistent results in iterated peak detection and fire frequency reconstructions.

This uncertainty appears to arise from non-random distribution of particles. Graduated quadrat analysis at volumes between 1.25 - 5.0 cm³, which in this study were limited by the diameter of the sediment cores, demonstrate no reduction in clumping. Current research is evaluating expanding the volume analyzed to identify a generally applicable sampling scale appropriate to support the assumption of random particle distribution necessary for peak detection methodology.