Identifying historic river ice breakup timing using MODIS and Google Earth Engine in support of operational flood monitoring in Northern Ontario

Abstract River ice breakup and resulting flood risk is a nearly annual concern for communities along the five major rivers draining to the James and Hudson Bay Coasts in Ontario (Moose, Albany, Attawapiskat, Winisk and Severn Rivers). Ice breakup within this region has historically been monitored using flight reconnaissance supplemented by assessment of hydrometric data. More recently, remote sensed imagery have been used to monitor near real-time ice breakup and flood risk. However, the near real-time remotely sensed breakup information was found to have limited utility in the absence of a broader spatial and temporal understanding of breakup progression. The primary purpose of this study was to develop a method for generating a dataset of breakup dates. A secondary objective was to calculate statistics from this dataset that can be used to provide context to operational near real-time imagery analysis and improve understanding of ice processes in the study area. An automated method for detecting river ice breakup dates from 2000 to 2017 using MODIS imagery was developed. This method uses a threshold-based technique that aims to maximize river coverage and minimize effects of cloud obstruction. Image processing was completed in the high-performance Google Earth Engine application which enabled iterative classification and model calibration. The breakup date dataset was used to calculate statistics on breakup timing, duration, annual variability and breakup order. An assessment of patterns within these data is discussed, relationships between breakup timing, duration and highwater years is explored and the operational utility of these statistics is described. The classification compared well with Water Survey of Canada derived breakup dates with mean bias ranging from −2.0 days to 6.7 days and mean absolute error of 3.4 days to 6.9 days across the rivers. Latitude and distance upstream were found to be primary controls on breakup timing with drainage network configuration and reach morphology also having an influence. No relationships between highwater years and calculated breakup statistics were found. It is recommended that future studies use the dataset developed in this study in combination with hydrometric and remotely sensed data to improve prediction of highwater and understanding of breakup processes within these rivers.