Interchangeability of mountain permafrost probability models, northwest Canada

Abstract Spatial models of mountain permafrost probability based on measurements of the basal temperature of snow (BTS) and ground‐truthing were developed for three study areas located >200 km apart. The interchangeability of these locally derived empirical‐statistical models was examined by predicting BTS values from the equations developed at the other two sites, and using logistic regression to relate these modelled values to the local ground‐truthed data. Equations for two of the areas were effectively interchangeable, producing permafrost extent predictions within 2% of each other, and permafrost probabilities within ±0.1 for more than 85% of the cells. Predictions were much less similar when their equations were applied to the third area and when its equation was applied to them. Model interchangeability appears to depend on where a site lies on a continuum from elevation controlled (infinite ratio of the standardised coefficient of elevation to that of potential incoming solar radiation in the BTS equation), to radiation dominated (ratio much less than unity of the same variables). Extensive ground‐truthing is essential, helping to constrain the logistic regression and hence the overall permafrost percentage. Additional investigations are needed to relate these ratios to regional climate in order to allow the spatial interpolation of permafrost probability models across the North American Cordillera. Copyright © 2008 John Wiley & Sons, Ltd.