Addressing Biases in Ice Jam Observations by Integrating Multi‐Source Data in a Forested Fluvial Landscape, Southern Quebec

ABSTRACT Exhaustive long‐term and large‐scale ice jam records are scarce in most cold river environments. Many discrete events occur in small, sparsely populated river systems and are poorly represented in open‐source databases. These observation biases are transferred to predictive models of ice jams and the collective understanding of their formation mechanisms. This study addresses these observation biases by using land use as a proxy for ice jam observation probability and by combining direct human observations with dendrochronological records of ice jam activity. The probability of observing an ice jam directly by a witness or indirectly by a tree‐ring dated tree scar increases with the density of urban and forest cover, respectively. The annual probability of occurrence for ice jams calculated from direct observational or dendrochronological records alone correlated poorly with geomorphological factors known to cause ice jams. Correcting the observational biases in individual records with their respective land use densities improved the correlation with the Ice jam Predisposition Index (IJPI), a spatial predictor of the probability of occurrence for ice jams. Correcting the observation bias with land use and combining multisource data (direct and dendrochronological observations) further improved the correlation between ice jams and the IJPI. A multi‐source approach thus partly overcomes the observation bias of individual records. This work highlights the potential impacts of observation biases in direct and indirect (dendrochronological) ice jam records and shows that bias‐corrected, multisource ice jam records could benefit the calibration and validation of ice jam prediction model.