Abstract:
|
The time it takes for a lightning caused wildfire to be detected after the lightning strike occurs is called the holdover time. In this paper, the distribution of holdover times in Ontario is studied through local likelihood-based density estimation methods developed for interval-censored data. In order to estimate the conditional distribution of holdover time, given an index of fire danger at the time of the lightning stroke, methodology for multivariate and conditional density estimation for interval-censored data is developed. A connection is made with the Nadaraya-Watson local regression estimator, and its natural extension to the interval-censored data context. It is demonstrated that the conditional density for holdover times can be estimated, and that there is likely a relation between holdover time and the fire danger index. This relation may have implications for fire management.
|