Updates to models of streamflow and water temperature for 2011, 2015, and 2016 in rivers of the Willamette River Basin, Oregon

Abstract

Mechanistic river models capable of simulating hydrodynamics and stream temperature are valuable tools for investigating thermal conditions and their relation to streamflow in river basins where upstream water storage and management decisions have an important influence on river reaches with threatened fish populations. In the Willamette River Basin in northwestern Oregon, a two-dimensional, hydrodynamic water-quality model (CE ‑ QUAL ‑ W2) has been used to investigate the downstream effects of dam operations and other anthropogenic influences on stream temperature. By simulating the managed releases of water and various temperatures from the large Willamette Valley Project dams upstream of the modeling domain, these models can be used to investigate riverine temperature conditions and their relation to streamflow to determine where and when conditions are most challenging for threatened fish populations and how dam operations and flow management can affect and optimize thermal conditions in the river. The original models were initially developed to simulate conditions in spring–autumn of 2001 and 2002. This report documents (1) the upgrade of the river models to CE‑QUAL‑W2 version 4.2 and (2) the update of those models to simulate conditions that occurred from March through October of 2011, 2015, and 2016. These years were selected to represent a range of climatic and hydrologic conditions in the Willamette River Basin, including a “cool, wet” year (2011), a “hot, dry” year (2015), and a “normal” year (2016). Six submodels comprise the modeling system updated in this report; each submodel can be run independently or run with the others as a system. These models include the Coast Fork and Middle Fork Willamette River submodel, which includes the Coast Fork and Middle Fork Willamette Rivers, the Row River, and Fall Creek; the McKenzie River submodel, which includes the South Fork McKenzie River downstream of Cougar Dam and the McKenzie River from its confluence with the South Fork McKenzie River to its mouth; the South Santiam River submodel, which comprises the South Santiam River from Foster Dam to the Santiam River; the North Santiam and Santiam River submodel, which includes the Santiam River and the North Santiam River downstream of Big Cliff Dam; the Upper Willamette River submodel, which includes the Willamette River from Eugene to Salem; and the Middle Willamette River submodel, which includes the Willamette River from Salem to Willamette Falls near Oregon City. The models included in this report were originally developed, calibrated, and documented by other researchers. As part of the model updates described here, some model parameters were adjusted to improve stability and decrease runtime. Boundary conditions including meteorological, hydrologic, and thermal parameters were developed and updated for model years 2011, 2015, and 2016. In many cases, the data sources used to drive the 2001 and 2002 models were no longer available, which required the use of new data sources, the determination of a proxy record, or the development of appropriate estimation techniques. Goodness-of-fit statistics for the updated models show a good model fit, with the models simulating subdaily water temperatures at most comparable locations with a mean absolute error of generally less than 1 °C and often nearing 0.5 °C, depending on the individual submodel, and a reasonably low bias. The subdaily mean error for the South Santiam River submodel produced the highest bias of any of the submodels. Goodness-of-fit statistics indicate that the results may be biased cool (ranging from -0.43 °C in 2016 to -0.80 °C in 2011 for subdaily results), but the only water temperature data available for comparison on the South Santiam River is itself estimated, and those estimates are known to be too high in summer. Depending on future modeling needs, that submodel may warrant further refinement, along with additional data collection to properly define and minimize any model bias.