Cornice - Development of a Nearest Neighbours Model Applied in Backcountry Avalanche Forecasting in Scotland

This paper describes the development and refinement of a nearest neighbours avalanche forecasting model in Scotland. Avalanche forecasting in Scotland is carried out in five areas, primarily for mountaineers who are either winter climbing or mountaineering in 'backcountry' locations where no avalanche control is carried out. Forecasting is primarily conventional, but computer-models are also used NXD was introduced in the early nineties and based on our experiences with NXD the model described in this paper, Cornice, was developed and introduced. Cornice has been developed in close consultation with avalanche forecasters and aims to provide a tool which integrates well with conventional avalanche forecasting by encouraging hypothesis testing. In order to facilitate this approach, model development must focus not only on the event we are forecasting, but mechanisms for outputting useful information in a range of ways to the forecaster. In this paper we describe two novel elements of Cornice: the development of automated, objective methods for weighting of parameters through the use of a genetic algorithm and the use of graphical output from forecasts allowing the forecaster to visualise relationships between neighbours in terms of clustering in meteorological/snowpack variables and the identification of spatial clusters of avalanche occurrence in location, altitude and aspect. We present results from the operational use of the model over the past three winters for the Ben Nevis area and examine its integration into a conventional avalanche forecasting framework.