A supplement to nearest-neighbour method for avalanche forecasting

Abstract The nearest-neighbour method is widely used independently as well as in combination with other methods for avalanche forecasting. Though, the nearest-neighbour method has proven fairly helpful, forecasters may often face difficulty with its output in determining the number of real neighbours of the present day. Values of geometrical distances, defined by distance metric used, help only in ranking the past days in terms of their nearness with present day. A situation, where sufficient real neighbours of the present day do not exist in sample database, usual practice of considering 10 nearest-neighbours for decision making may lead to unrealistic conclusions. A method is offered as a supplement to the popular nearest-neighbour method for avalanche forecasting. The method is applied on the same database and parameter vector space as being used for nearest-neighbour method. Forecaster defines the ranges of various parameters around the parametric values of the present day. A geometrical closed volume is generated in the parametric vector space according to the specified ranges, and the set of past days falling within this volume is the output. The output is an explicit evidence of uniqueness (or commonness) of the present day within the specified ranges. This output, when analysed in the backdrop of nearest-neighbour method output on the same data, helps forecasters in fine-tuning the decision. The proposed method was tested for its potential along Chowkibal–Tangdhar (CT) road axis, a stretch of about 20 km with 17 prominent avalanche sites, in Pir Panjal ranges of Indian Western Himalaya. With the prediction of avalanche days with mean probability 0.80 and standard deviation 0.38 along CT axis, model holds promise as a potential supplement tool for avalanche forecasting.