Identifying Dynamically Induced Variability in Glacier Mass-Balance Records

Glacier mass-balance (i.e., accumulation vs. ablation) provides a direct indicator of a glacier's relationship with climate. However, mass-balance records contain noise due to internal climate variability (i.e., from stochastic fluctuations in large-scale atmospheric circulation), which can obscure or bias trends in these relatively short timeseries. This presents a challenge in correctly identifying the signature of anthropogenic change. "Dynamical adjustment" is a technique that identifies patterns of variance shared between a climate timeseries of interest (e.g., mass-balance) and independent "predictor" variables associated with large-scale circulation (e.g., Sea Level Pressure, SLP, or Sea Surface Temperature, SST). Extracting the component of variance due to internal variability leaves a residual timeseries for which trends can more confidently be attributed to external forcing. We apply dynamical adjustments based on Partial Least Squares Regression to mass-balance records from South Cascade Glacier in Washington State and Wolverine and Gulkana Glaciers in Alaska, independently analyzing seasonal balance records to assess the dynamical influences on winter accumulation and summer ablation. Seasonally averaged North Pacific SLP and SST fields perform comparably as predictor variables, explaining 50-60% of the variance in winter balance and 30-40% of variance in summer balance for South Cascade and Wolverine Glaciers. Gulkana glacier, located further inland than the other two glaciers, is less closely linked to North Pacific climate variability, with the predictors explaining roughly one-third of variance in its winter and summer balance. We analyze the significance of linear trends in the raw and adjusted mass-balance records, and find that for all three glaciers, a) summer balance shows a statistically significant downward trend that is not substantially altered when dynamically induced variability is removed, and b) winter balance shows no statistically significant trends before or after adjustment, though much of the variance is dynamically induced.