Impact of intense early 20th century fire activity on boreal forest composition in eastern Canada

Background/Question/Methods

Fire plays an important role for boreal forest succession and time since last fire (TSLF) is seen as a useful covariate to infer the spatial distribution of carbon over a large regional scale. Current methods employed to provide TSLF are either spatially or temporarily limited. Forest composition and structure may provide clues to approximate TSLF. Consequently, we developed a forest succession model for black spruce (Picea mariana) dominated forests in eastern Canada (49°N to 52°N and 79° 30’W to 66°W, area, 217 000 km²) by generalizing the empirical relationships that exist between historical records of fire (1880-2000) with forest inventory data and biophysical variables. Random Forest was used to predict TSLF, at the scale of 2km² cells, which corresponds to the observed median fire size in recent fire records (1970-2000). Landscape cells were classified into TSLF ≤ 120 years (young forest) and TSLF > 120 years and TSLF was then estimated for young forest. The mean time since last fire (MTSLF) at the landscape scale (65km² to 2975km²) was computed from the Weibull distribution of cell-level TSLF for better visualization.  

Results/Conclusions

The model classified the landscape into TSLF ≤ 120 and TSLF > 120 years with an overall accuracy of 88.3% and a value of Cohen’s kappa of 0.72.  The proportions of the oldest age classes and the presence of Abies balsamea were the best predictors of TSLF. The coefficient of determination and the root mean square error in predicted TSLF were 0.87 and 6.9 years with 15% bias.  This bias was reduced by converting each TSLF from 1000 bootstrapped replications into decade class and the class with majority vote was selected. Results show that the decades between 1920 and 1940 were characterised by an intense fire activity, with approximately 25% of the study region burned. Studies have reported a doubling of the burn rate from 1970 to 2000 but the resulting burn rate is lower than the one detected between 1920 and 1940. Over 120 years, the decadal mean burn rate varied between 0.2 and 11.6 % year^-1. These results show the importance of lengthening the historical records of fire history maps in order to provide a better perspective of the actual changes of fire regime.