Discussion of “Frequency of ice-jam flooding of Peace-Athabasca Delta”

We present an alternative analysis of the observational record of historical ice-jam floods in the lower Peace River presented in Beltaos (2018) and illustrate the need to examine the history of hydrological change in lakes of the Peace-Athabasca Delta (PAD) to identify causes of recently observed drying. Beltaos (2018) uses an observational record of historical ice-jam floods in the lower Peace River, as compiled by Timoney (2009), to suggest that flood frequency changed in 1968, coincident with the start of hydroelectric regulation of the Peace River by the WAC Bennett Dam (British Columbia), and that this has contributed to drying of perched lakes in the Peace-Athabasca Delta (northern Alberta). On a graph that shows the cumulative frequency of occurrence of large ice-jam floods, Beltaos (2018) forces a quadratic equation through the ‘baseline interval’ (i.e., 1880–1967) before regulation and then develops a separate linear regression to characterize the post-dam interval (1971–2017; see figure 3 in Beltaos 2018). Beltaos (2018) postulates that the decrease in slope identifies that regulation of the Peace River is a contributing factor to drying of perched lakes in the delta. We contend that Beltaos’ use of regression analysis on a time series of episodic ice-jam floods is problematic on many accounts. For example, assumptions of the statistical tests presented are violated because residuals are not independent (as shown in Beltaos (2018), figure 3, which demonstrates that non-flood years tend to follow non-flood years). But even if we excuse this, the use of quadratic regression as shown in figure 3 of Beltaos (2018) is flawed because it implicitly predicts that flood frequency increases over time, for which there is no physical basis nor does it align with independent, paleolimnological reconstructions of flood frequency and magnitude from oxbow lake sediment records in the delta that are proximal to the Peace River (Wolfe et al. 2006). Thus, it is misleading to construe an expectation of accelerating flood frequency during the post-regulation era. Furthermore, inflation of the effective sample size by counting every year as an observation for the purposes of regression analysis, as performed by Beltaos (2018) (i.e., cumulative floods for 1920 = 4, 1921 = 4, 1922 = 4, ..., 1931 = 4 are part of his regression so n = 88), results in excessive confidence in the estimated frequency of floods (and slope of the relation) between 12 major pre-regulation flood events over 88 years. We have attempted to account for some of these shortcomings by applying a more conservative linear regression model, which only includes observed floods as data points to characterize the baseline (i.e., 1880–1967) interval (Fig. 1a). This pre-regulation linear regression model also captures the post-1971 observational ice-jam flood frequency data, indicating that there