Machine Learning-Based Reconstructions of Historical Daily and Monthly Runoff for the Laurentian Great Lakes

High-quality regional streamflow datasets are necessary to support local water resources planning and management in the Great Lakes basin. However, the region's observed streamflow records are limited by the availability of surface water gauges, which go in and out of service over time. In this paper, we present a continuous reconstruction of daily streamflow from 1951-2013 at 656 gauged locations throughout the Great Lakes basin. This dataset is created using a novel regional Long Short-Term Memory (LSTM) model that integrates local climate data, physical catchment characteristics, and runoff observations from nearby gauged sites. We also use this model to estimate monthly runoff into Lakes Superior, Michigan-Huron, St. Clair, Erie, and Ontario. The daily reconstruction product will equip water managers with information to support local water resources analyses, while the monthly runoff product can help improve historical water balance estimation across the Great Lakes and provide critical context for lake level shifts under a changing climate.