Nearest neighbours for avalanche forecasting in Scotland - development, verification and optimisation of a model

This paper describes a nearest neighbours model used operationally for avalanche forecasting in Scotland. Results are presented as contingency tables and in terms of a range of categorical statistics suggesting that the model performs well in comparison with other models, although avalanche events are overforecast. Genetic algorithms are described as a technique for calculating variable weights and shown to provide an effective mechanism for objectively calculating weights. The model is discussed in the context of conventional avalanche forecasting, and it is proposed that the range of visualisations provided to the user can facilitate hypothesis testing. (C) 2003 Elsevier B.V. All rights reserved.