Statistical summer mass-balance forecast model with application to Brúarjökull glacier, South East Iceland

DARRI EYTHORSSON,ANDRI GUNNARSSON,SIGURDUR M. GARDARSSON,BIRGIR HRAFNKELSSON

doi:10.1017/jog.2018.22

DARRI EYTHORSSON, ANDRI GUNNARSSON + Show 2 more

Open Access

https://doi.org/10.1017/jog.2018.22

Copy DOI

Abstract

ABSTRACTForecasting of glacier mass balance is important for optimal management of hydrological resources, especially where glacial meltwater constitutes a significant portion of stream flow, as is the case for many rivers in Iceland. In this study, a method was developed and applied to forecast the summer mass balance of Brúarjökull glacier in southeast Iceland. In the present study, many variables measured in the basin were evaluated, including glaciological snow accumulation data, various climate indices and meteorological measurements including temperature, humidity and radiation. The most relevant single predictor variables were selected using correlation analysis. The selected variables were used to define a set of potential multivariate linear regression models that were optimized by selecting an ensemble of plausible models showing good fit to calibration data. A mass-balance estimate was calculated as a uniform average across ensemble predictions. The method was evaluated using fivefold cross-validation and the statistical metrics Nash–Sutcliffe efficiency, the ratio of the root mean square error to the std dev. and percent bias. The results showed that the model produces satisfactory predictions when forced with initial condition data available at the beginning of the summer melt season, between 15 June and 1 July, whereas less reliable predictions are produced for longer lead times.

Highlights

Water storage in snow and ice is an important factor in the hydrological cycle in many regions of high altitudes and latitudes like Iceland, where 11% of the country is covered by glaciers (Bjornsson and Palsson, 2008)
The automatic weather stations (AWS) data were aggregated to average values to represent the initial conditions of the system at four different dates in spring, for the periods beginning on 1 April and ending on 15 May, 1 June, 15 June and 1 July
In Bayesian model averaging (BMA), model uncertainty is evaluated by assigning prior probabilities to all models being considered, whereas in the frequentist model averaging (FMA), no prior probabilities are required and all estimators are determined by the data (Buckland and others, 1997; Raftery and others, 1997; Hoeting and others, 1999)

Summary

Introduction

Water storage in snow and ice is an important factor in the hydrological cycle in many regions of high altitudes and latitudes like Iceland, where 11% of the country is covered by glaciers (Bjornsson and Palsson, 2008). Simulation and prediction of melt behavior provide valuable information for water resources management, e.g. regarding reservoir fill rates, potential power production and load on hydraulic structures. Short-term predictions of a few days improve daily operations and risk analysis, whereas longer term prediction of seasonal melt behavior assists in the systematic optimization of networks of reservoirs and diversions. Prior work in melt modeling of Icelandic glaciers has focused on either empirical (degree day) and physical (energy balance) modeling approaches. Both have shown good performance for simulating glacial mass balance (e.g. De Ruyter de Wildt and others, 2003b; Marshall and others, 2005; Carenzo and others, 2009; Engelhardt and others, 2014). The vast potential of remotesensing data has been increasingly applied to snowmelt estimation in basins where little information is available (Kalra and others, 2013; Qiu and others, 2014; Drolon and others, 2016)

Objectives

Methods

Results

Conclusion