The Great Lakes Runoff Intercomparison Project Phase 1: Lake Michigan (GRIP-M)

- Five models were assessed for simulating runoff to Lake Michigan.
- The range in model skill was less than expected when simulating basin-wide runoff.
- Two newcomers to lake-basin modeling (AFINCH and NWS) performed especially well.
- Assimilation of discharge observations improves historical simulation of runoff.
- Since 1998, less than average precipitation has become runoff to Lake Michigan.

SummaryWe assembled and applied five models (one of which included three different configurations) to the Lake Michigan basin to improve our understanding of how differences in model skill at simulating total runoff to Lake Michigan relate to model structure, calibration protocol, model complexity, and assimilation (i.e. replacement of simulated discharge with discharge observations into historical simulations), and evaluate historical changes in runoff to Lake Michigan. We found that the performance among these models when simulating total runoff to the lake varied relatively little, despite variability in model structure, spatial representation, input data, and calibration protocol. Relatively simple empirical, assimilative models, including the National Oceanic and Atmospheric Administration (NOAA) Great Lakes Environmental Research Laboratory (GLERL) area ratio-based model (ARM) and the United States Geological Survey (USGS) Analysis of Flows in Networks of CHannels (AFINCH) model, represent efficient and effective approaches to propagating discharge observations into basin-wide (including gaged and ungaged areas) runoff estimates, and may offer an opportunity to improve predictive models for simulating runoff to the Great Lakes. Additionally, the intercomparison revealed that the median of the simulations from non-assimilative models agrees well with assimilative models, suggesting that using a combination of different methodologies may be an appropriate approach for estimating runoff into the Great Lakes. We then applied one assimilative model (ARM) to the Lake Michigan basin and found that there was persistent reduction in the amount of precipitation that becomes runoff following 1998, corresponding to a period of persistent low Lake Michigan water levels. The study was conducted as a first phase of the Great Lakes Runoff Intercomparison Project, a regional binational collaboration that aims to systematically and rigorously assess a variety of models currently used (or that could readily be adapted) to simulate basin-scale runoff to the North American Laurentian Great Lakes.