Modelling Fire Weather Index Series

Abstract

<p> The fire weather index (FWI), useful as a measure of forest fire danger, is calculated from precipitation and other weather variables. In the present environmental study, precipitation, fuel moisture codes, and fire behavior indices were available for a reference site and 4 higher elevation sites around Smithers, British Columbia. The objective of the study was to determine whether the use of local precipitation would lead to a different FWI than obtained from precipitation at the reference site.</p> <p> The features of the series of daily FWI values which needed to be taken into account were: peaks following dry periods, serial correlation, and heteroscedasticity. Two types of models were developed to characterize the record as a smooth component, for the upward and downward movements of the index, and a component of correlated error terms. The first type was a parametric Fourier series in a context of a generalized linear model (GLS) that allowed for serial correlation and heteroscedasticity. The second form was a smoothing cubic spline with a bootstrap procedure for estimation of standard errors and confidence bands. The question, of whether FWI on a particular day differed between a higher elevation station and the reference station, was addressed by adding a station effect to the GLS model and by graphical comparison of the smooth curves with confidence bands for the spline method.</p> <p> The Model-3 for the combined station effect is not able to capture the sharpness of the peak and found insignificant while cubic spline smoothing curves fitted to the bootstrap behave well to capture peaks and troughs in the index but it encounter some difficulties for few lower index values.</p>