Effective sample size for glacier mass balance

Abstract

Correlograms from multiple time series of point mass balance, measured on White Glacier (Axel Heiberg Island, Canada) and Abramov Glacier (Alai Range, Kirgizia), show that the correlation decreases with the difference in elevation between the points. The correlogram is used to calculate an integral spatial scale or effective sample area which, when divided into the area of the glacier, yields an estimate of the number of degrees of freedom in a stake‐based estimate of the whole‐glacier balance. This number – the ‘effective sample size’– is a small fraction of the number of point measurements; indeed, it is independent of the number of measurements. Estimates of average balance for elevation bands are at one remove from raw stake measurements, but they are amenable to a principal component analysis which confirms that the effective sample size is very small. The small effective sample size means that uncertainty in a typical measurement of whole‐glacier mass balance cannot be much less than the large value implied by the conservative assumption that stakes are perfectly correlated. One way around this difficulty would be to increase the role of prior physical understanding by seeking to model the spatial variability of mass balance. A successful model would need only a few parameters, and would allow for the joint estimation of both magnitude and uncertainty; the uncertainty in mass balance could be derived objectively from the uncertainty in the parameters. This, however, would require good estimates of the variability of mass balance at the elevation‐band scale, which might in turn require that many measurement networks be redesigned.

Key words:

• Mass balance
• accuracy
• sample size
• statistical modelling
• White Glacier
• Abramov Glacier.