Abstract
For Canada's boreal forest region, the accurate modelling of the timing of the appearance of aspen leaves is important to forest fire management, as it signifies the end of the spring fire season that occurs after snowmelt. This article compares two methods, a midpoint rule and a conditional expectation method used to estimate the true flush date for interval-censored data from a large set of fire-weather stations in Alberta, Canada. The conditional expectation method uses the interval censored kernel density estimator of Braun et al. (Citation2005). The methods are compared via simulation, where true flush dates were generated from a normal distribution and then converted into intervals by adding and subtracting exponential random variables. The simulation parameters were estimated from the data set and several scenarios were considered. The study reveals that the conditional expectation method is never worse than the midpoint method, and that there is a significant advantage to this method when the intervals are large. An illustration of the methodology applied to the Alberta data set is also provided.
Acknowledgments
The authors gratefully acknowledge the support of GEOIDE (SII Project 51), MITACS, and the Natural Sciences and Engineering Research Council of Canada. We also thank Alberta Sustainable Resource Development for the use of their data and the anonymous referee for their helpful suggestions which led to an improved presentation. All computations in the article were performed using the statistical software package R (R Development Core Team, Citation2007).