Error in the inference of fire history from grasstrees

Abstract

Listen

Translate article

Abstract A method for constructing fire histories has recently been proposed for fire‐prone southern Australia based on the pattern of coloured bands in the remnant leaf‐bases on stems of grasstrees (Xanthorrhoea species; Xanthorrhoeaceae). In the absence of alternative high‐resolution techniques extending into the premodern period, this method has been utilized to construct fire histories for forest, woodland and shrubland ecosystems, principally in south‐western Australia. However, the technique has not been validated against known fire histories spanning more than one fire interval. Here we compare fire records from 100 grasstrees with a 30‐year record (1973–2002) of fire data derived from satellite imagery in a region of sandplain shrubland vegetation near Eneabba in south‐western Australia. Fires occurred in eight of the 30 years of the satellite record, with sampled grasstrees burning between zero and four times. The grasstree and satellite records agreed in terms of the overall incidence of fires experienced over the 30‐year period, with the grasstree record matching the satellite record significantly better than chance. However, comparison of the grasstree and satellite records found substantial error in the rate of both false positives and false negatives. Grasstrees failed to identify fire in 83% of fire occurrences identified by the satellite record, down to 53% if an error of ±2 years in the attribution of year of fire was allowed. A similar proportion of grasstree fire incidents were not matched in the satellite record (false positives). The rate of false positives increased with time before present, suggesting a temporal bias in the grasstree record. It is clear that the grasstree record does reflect fire history to a degree, but that it contains at least as many false as true fire records and may tend towards over‐reporting the incidence of fire in the past.