Modeling anchor ice presence–absence in gravel bed rivers

We studied the meteorological and hydraulic conditions conducive to anchor ice formation in three gravel bed rivers of southern Québec (Beauport, Duberger and York rivers). First, a dataset including the dichotomous response variable presence–absence of anchor ice in a river reach, and ten continuous predictors related to air temperature (freezing degree-hours FDH cumulated over 6, 8, 12, 24 and 48 h) and water temperature (W averaged over 6, 8, 12, 24 and 48 h) was used to develop a logistic regression model of anchor ice presence at the river reach level. The model developed in this study used the variable FDH-12 to correctly predict 80.9% of the anchor ice presence–absence observations. More specifically, it predicted that anchor ice is more likely to form when a threshold FDH-12 value of 111.7 °C-h is exceeded. Second, data on the presence of anchor ice at the parcel level (0.25 m2) was collected at the Beauport and Duberger study reaches in order to determine the effect of local hydraulic conditions on the formation of anchor ice. Covariates common to all parcels within a river reach (air and water temperature) and hydraulic variables specific to each parcel (flow velocity, water depth, Froude and Reynolds numbers) were used as explanatory variables. The results of the stepwise hierarchical logistic regression analysis showed that the Froude number was the first variable to enter the model and that it could not be improved significantly by adding any other variable. The model correctly classified 67.5% of the presence–absence observations of anchor ice on the parcels and indicated that anchor ice was most likely to form at Froude values larger than 0.1.