This study had two objectives: (1) to use the results of an individual-based particle-tracking model of larval sucker dispersal through the Williamson River delta and Upper Klamath Lake, Oregon, to interpret field data collected throughout Upper Klamath and Agency Lakes, and (2) to use the model to investigate the retention of sucker larvae in the system as a function of Williamson River flow, wind, and lake elevation. This is a follow-up study to work reported in Wood and others (2014) in which the hydrodynamic model of Upper Klamath Lake was combined with an individual-based, particle-tracking model of larval fish entering the lake from spawning areas in the Williamson River. In the previous study, the performance of the model was evaluated through comparison with field data comprising larval sucker distribution collected in 2009 by The Nature Conservancy, Oregon State University (OSU), and the U.S. Geological Survey, primarily from the (at that time) recently reconnected Williamson River Delta and along the eastern shoreline of Upper Klamath Lake, surrounding the old river mouth. The previous study demonstrated that the validation of the model with field data was moderately successful and that the model was useful for describing the broad patterns of larval dispersal from the river, at least in the areas surrounding the river channel immediately downstream of the spawning areas and along the shoreline where larvae enter the lake.

In this study, field data collected by OSU throughout the main body of Upper Klamath Lake, and not just around the Williamson River Delta, were compared to model simulation results. Because the field data were collected throughout the lake, it was necessary to include in the simulations larvae spawned at eastern shoreline springs that were not included in the earlier studies. A complicating factor was that the OSU collected data throughout the main body of the lake in 2011 and 2012, after the end of several years of larval drift collection in the Williamson River by the U.S. Geological Survey. Those larval drift data provided necessary boundary-condition information for the earlier studies, but there were no measured boundary conditions for larval input into model simulations during the years of this study (2011−12). Therefore, we developed a method to estimate a time series of larval drift in the Williamson River, and of the emergence of larvae from the gravel at the eastern shoreline springs, that captured the approximate timing of the larval pulse of the Lost River sucker (Deltistes luxatus) and shortnose sucker (Chasmistes brevirostris) and the relative magnitude of the pulses by species and spawning location. The method is not able to predict larval drift on any given day, but it can reasonably predict the approximate temporal progression of the larval drift through the season, based on counts of adult suckers returning to spawn. The accuracy in the timing of the larval pulses is not better than about plus or minus 5 days.

Model results and field data were consistent in the basic progression of both catch per unit effort (CPUE) and larval length through time. The model simulation results also duplicated some of the characteristics of the spatial patterns of density in the field data, notably the tendency for high larval densities closer to the eastern and western shorelines. However, the model simulations could not explain high densities in the northern part of the lake or far into Ball Bay, locations that are far from the source of larvae in the Williamson River or eastern shoreline springs (as measured along the predominant transport pathways simulated in the model). This suggests the possibility of unaccounted-for spawning areas in the northern part of the lake and also that the period during which larvae are transported passively by the currents is shorter than the 46 days simulated in the model. Similarly, the progression of larval lengths in the field data is not a simple progression from smaller to larger fish away from sources in the river and springs, as simulated by the particle-tracking model; the smallest fish were caught at different times near the Williamson River, in the northwestern part of the lake, and in the southernmost part of the lake. This again suggests that fish may be spawning at places other than the river and eastern springs, that our understanding of larval transport is incomplete, or both.

The model was used to run 96 numerical “experiments” in which lake elevation, river discharge, and wind forcing were varied systematically in order to investigate the sensitivity of particle retention to each variable, and with particular emphasis on the idea of managing lake elevation to control emigration. The estimates of particle retention cannot be equated directly to retention of fish larvae, primarily because there was no mortality included in the simulations, but the relative comparison of retention and emigration around the matrix of experimental conditions provided several “big picture” results:

- Variables that cannot be controlled—winds and discharge—had the largest effect on retention. For example, at the lowest river discharge (20 cubic meters per second), simulated retention was high regardless of wind or lake elevation, whereas at the highest river discharge (100 cubic meters per second), retention was low regardless of wind or lake elevation.
- When river discharge and wind were held constant, a higher elevation delayed the onset of the most rapid exit of particles by 1 (from the springs) to 4 (from the river) days, but did not determine overall retention. Only under the combination of conditions consisting of low discharge (50 cubic meters per second or less) and strong wind reversals for several days was there a consistent effect of lake elevation on overall retention several weeks into the simulation, and, under those conditions, retention was at the high end of the possible range regardless of lake elevation.
- Under most combinations of conditions tested, after particles had been in the system for several days, the complex interaction between wind, elevation, and river discharge resulted in particle pathways, and therefore retention, being highly variable and unpredictable, at which point controlling lake elevation could not produce a predictable result. Therefore, on the basis of the model predictions, managing lake elevation probably is not a way to reliably provide any particular level of retention.