Using tree-ring records to calibrate peak detection in fire reconstructions based on sedimentary charcoal records

Abstract

We compared fire episodes over the past 150 years reconstructed using charcoal particles retrieved from well-dated sediment deposits from two small lakes in the eastern Canadian boreal forest, with dendrochronological reconstructions of fire events from the corresponding watersheds. Fire scars and age structure of living trees highlighted three fire events (ad 1890, 1941, and 1989). To explore the ability to detect these fire events based on sedimentary charcoal records, we explored the influence of two user-determined parameters of a widely used peak-detection algorithm (the CharAnalysis software): (1) the temporal resolution used to interpolate charcoal series and (2) the width of the smoothing window used to model background noise. The signal-to-noise index (SNI) is often used to evaluate the ability to detect peaks in sedimentary charcoal records, which can be related to fire events. SNI values >3 identify records appropriate for peak detection. Selecting standard settings in paleoecological studies (median temporal resolution of the entire sequence and 500- to 1000-year window width) yielded higher global SNI values but failed to detect most recent fire events. Instead, selecting a shorter reference period (the past ~150 years) to determine the temporal resolution to interpolate the charcoal series and a narrower smoothing window (100 years) best matched the tree-ring data despite lower SNI values (often <3.0). However, Holocene fire history differed markedly when reconstructed using different smoothing window widths (100–150 years vs >300 years). Consequently, we suggest using the smallest window width yielding a SNI >3. Practitioners must not necessarily focus on obtaining the highest possible SNI, usually related to wide smoothing windows. We also suggest that fire history reconstructions should focus on core sections presenting fairly constant sedimentation rates. Alternatively, sediments could be subsampled after age–depth models have been obtained.

Fires of the Last Millennium Led to Landscapes Dominated by Early Successional Species in Québec’s Clay Belt Boreal Forest, Canada

“…1910. The link between the recent fires that occurred in the forest bordering the two peatlands and the deposition of charcoal in organic sediments is therefore well-established, suggesting that the same process characterized all of the Holocene period [56], at least in the case of large and severe fires. The fires occurring in our study area (4000 km 2 ) and in the 'Plaine du lac Matagami' during the last two centuries are considered natural because this region was sparsely inhabited until recently.…”

Coniferization of the mixed‐wood boreal forests under warm climate

“…To remove the bias induced by different sedimentation rates and taphonomic processes involved in the sequestration of charcoal in the sediments, the CHAR series from each lake were interpolated to a constant temporal resolution of 20 years, which corresponded to the approximate median sample resolution of the lakes (Table ). The CHAR series were analyzed using CHARANALYSIS 1.1 software (Higuera et al , ) (available at https://github.com/phiguera/CharAnalysis) to identify local fire events according to the updated method (Text S1; Brossier et al , ). Estimated fire dates were used to reconstruct the fire‐return intervals (FRIs) for each lake.…”

Fire Regime in Marginal Jack Pine Populations at Their Southern Limit of Distribution, Riding Mountain National Park, Central Canada

“…Lake Katherine is also a small closed basin, presumably ground water fed, with no significant water courses in or out. Unconsolidated sediment was extracted with a Glew gravity corer [65] down to 40 cm depth and dated with 13 210 Pb samples using a constant rate of supply model [66,67]. The procedure followed for 210 Pb dating was that outlined by MyCore Scientific Inc., (Dunrobin, ON, Canada).…”

“…C peak series were identified by subtracting C background from C interpolated . The C peak series were further decomposed into a C noise and C fire , with the latter representing in theory significant peaks associated with local fires [66,76]. The Gaussian mixture model was used to determine C noise distribution and a locally defined threshold was used to identify significant C peak .…”

“…The minCountP value was set to 0.99 in all runs given the short FRIs characterizing the recent RMNP fire history. The final determination of the peak analysis parameters were based on maximization of the signal to noise (SNI) index (typically >3) and goodness of fit (GOF; p < 0.05) [66,77].…”