Discussion


The model simulation results showed that aggregations of passively transported larvae coming down the Williamson River were dispersed over a larger area in 2008, after the flooding of Tulana, than in prior years when those aggregates were constrained by the Williamson River channel from spawning grounds to the mouth of the river at Upper Klamath Lake. In 2009, after Goose Bay was flooded, the larvae dispersed over an even larger area. Two of the three agencies involved in this study sampled larvae in each year from 2006 to 2009, and among these 4 years, the larval catches in both gear types (larval trawls and pop nets) were greatest in 2006 and least in 2009 (figs. 12–14; table 3). The USGS collected data from plankton nets only in 2008 and 2009; catches at most sites were lower in 2009 than in 2008. These results are broadly consistent with passive transport of the larvae as demonstrated by the modeling exercise. With the restoration of the Delta, the larval aggregate dispersed through the newly flooded areas, and remnants of the aggregate entered Upper Klamath Lake through multiple pathways. This resulted in lower simulated densities of fish at net locations along the shorelines of Upper Klamath and Agency Lakes. 


Evidence for the simulated changes in the number of larvae entering Upper Klamath Lake at the mouth of the Williamson River between 2006 and 2009 can be found in the pop net data collected along the Upper Klamath Lake side of the levees that formed the southern boundary of Goose Bay. Pop net data were collected along these levees from 2006 to 2009 (fig. 2). At these pop net sites, there was a decrease in the catch of larval suckers in 2009 after the restoration of Goose Bay was complete (Erdman and Hendrixson, 2010). This was the first year that the annual mean catch per unit effort (CPUE) at these pop net sites was less than in nets in the restored wetlands, which included Tulana and Goose Bay (Erdman and Hendrixson, 2010). The lower CPUE at the pop net sites in Upper Klamath Lake is consistent with the model simulation results showing fewer larvae entering Upper Klamath Lake at or north of the Williamson River mouth, then being transported southeastward along the Upper Klamath Lake side of the levees.


There was a large decrease in the mean CPUE in pop nets and larval trawls between 2006 and 2007, however, which was not a result of a change in the landscape, and only a small decrease between 2007 and 2008, when a large change in the landscape occurred with the flooding of Tulana (table 3). This was likely because the annual mean CPUE catch statistics reflect not only the changes in the landscape as the Delta underwent restoration, but also the interannual differences in the number and species composition of larvae. Different total densities and different relative numbers of LRS and 
SNS/KLS sucker larvae were observed in the drift at the Modoc Point Road bridge during the 4 years of this study (table 3). In 2006 and 2008, the second (and third, in 2008) pulses of larvae (dominated by SNS/KLS) were comparable in size to the first pulses dominated by LRS larvae (fig. 3); the difference in the density of the two species in the net catches at the Modoc Point Road bridge was not statistically significant in 2006, and in 2007 and 2009 the density of LRS was greater (Wilcoxon test, p<0.05). In 2008, however, the density of SNS/KLS was greater (Wilcoxon test, p<0.05). These drift data provide the upstream boundary condition for the model simulations, and because no species-specific differences in transport or mortality are included in the model, the simulated densities downstream of all species combined are influenced by the individual species in proportion to their occurrence in the drift. However, the statistics of species distribution in the various gear types tell a different story. Pop nets and larval trawl nets consistently captured more 
SNS/KLS larvae than LRS larvae, even in years when the drift of LRS from spawning sites was comparable to or greater than the drift of SNS/KLS. Therefore, the large decrease in mean CPUE at these sites between 2006 and 2007, which was not simulated by the model, could be due to the drift being dominated by LRS in 2007, and LRS not being captured as effectively by the gear. In 2008, the drift was dominated by SNS/KLS, and so the greater dispersal of larvae due to the opening of Tulana could have been compensated to some extent by the greater efficiency of the gear in capturing the dominant species. In contrast to the pop net data, in 2008, 81 percent of the suckers caught in the plankton nets and identified to species were LRS (Burdick and others, 2009; table 3). The reverse distribution was seen in 2009, when 90 percent of the identified suckers were SNS/KLS (Burdick and Brown, 2010; table 3). The plankton nets captured large numbers of LRS in 2008, but the mean density of this species in plankton nets decreased by two orders of magnitude in 2009, even though more LRS came down the Williamson River in the drift in 2009 and the number of SNS/KLS captured in plankton nets remained roughly the same (table 3). These sites are located primarily in Tulana and in Upper Klamath and Agency Lakes, and north of the Williamson River mouth (fig. 2). The large decrease in LRS catches at these sites between 2008 and 2009 suggests that LRS were transported north through Tulana in 2008 but not in 2009, consistent with model simulation results, and indicates that LRS may be better described by the assumptions of passive transport than SNS/KLS. 


Site-specific model simulations are more difficult to verify than generalized, large-scale patterns. Density simulations of larval transport each spring during 2006–09 predicted that the proportion of larvae flowing into Upper Klamath Lake at the mouth of the Williamson River would decrease markedly in response to the breaching of levees around Tulana and Goose Bay in late 2007 and 2008. Observations at one plankton net site offshore of the Williamson River mouth indicated that catches decreased between 2008 and 2009 (fig. 12; peak catches at site 25981 were 14.5 and 0.2 fish per cubic meter in 2008 and 2009, respectively). Larval catches at the larval trawl Williamson River mouth site (U6) decreased markedly in 2008 and 2009 relative to 2006, which was a year notable for high larval trawl catches among the 14-year period of 1995–2008, but not relative to 2007, which was a year of only moderate larval trawl catches over the same period (Simon and others, 2009; fig. 13). The relative increase in densities along the Goose Bay shoreline at site U5 relative to site U6 that the model predicted for 2009 was not observed. Peak larval catches were 43.8, 32.5, 9.7, and 3.0 fish per cubic meter at U5, and 81.5, 13.9, 24.2, and 35.6 fish per cubic meter at U6 from 2006 to 2009, respectively (fig. 13). Additionally, model simulation results indicated that very few larvae would be detected in Agency Lake prior to 2008, but the Agency Lake catches were greatest in 2006, just as they were in Upper Klamath Lake (fig. 14).


Density simulations showed that larvae at concentrations of between 20 and 50 fish per cubic meter at the Modoc Point Road bridge would have concentrations less than 1 fish per cubic meter when dispersed passively to sites located more than a few kilometers away from the Williamson River channel. Given the low simulated densities in areas of the Delta far from the Williamson River channel and in Upper Klamath and Agency Lakes, larval catches might be expected to be dominated by zero catches, and correlations with simulated densities might be expected to have little or no significance. This is because larval fish are not expected to be uniformly distributed within a 100–200 m area around a sample site, but are expected to be clustered instead. Further, because each net sampled only a fraction of the simulated volume used to predict density at a site, we should have expected numerous zero catches. Rank correlation coefficients, however, were almost uniformly positive and often significant. Most significant rank correlations were about 0.30–0.60, suggesting that although the model predicts the general pattern of distribution, usually less than one-half of the variation in site densities was explained.


Correlations between simulated densities and larval catches based on size class showed that correlation coefficients were highest and most often significant for small larvae (10–13 mm) at plankton net and pop net sites, and for medium larvae (> 13 mm–16 mm) at plankton net and larval trawl sites. The correlations for large larvae (>16 mm–19 mm) were highest and most consistently positive at larval trawl sites, although correlations for large larvae were significant only in 2009. Although not significant, most negative correlations were for large larvae at pop net sites. Small larvae made up 87 percent of all larvae at plankton net sites, 36.2 percent at pop net sites, and 34.1 percent at larval trawl sites. Further, at the larval trawl site closest to the larval source at the mouth of the Williamson River (U6), the small size class made up 59.4 percent of the catch and was similar to the plankton net samples. 


These patterns reflect site locations because most sites closest to the river source where the smallest larvae dominated were pop net sites, and most sites farthest from the source where larger larvae were most likely to be found were larval trawl sites (fig. 2). The patterns also partly reflect the size of gear openings, because the smallest openings are in plankton nets. The gear type (or the physical setting associated with the gear type, such as vegetation presence or water depth) may have influenced the total catch density as well as the proportion of each size class caught. For example, the plankton net site 25535 was located close to the pop net site B, and simulated densities at these two sites were comparable, but the catch densities differed markedly in both 2008 and 2009 (fig. 12). Another comparison between sites 25979 and A1 (fig. 14) shows lower catches in plankton nets than in larval trawls in Agency Lake where the model simulated similar densities. Therefore, although all nets were assumed to be maximally efficient for the smallest larvae, a better understanding of size efficiency of nets is needed to rigorously partition spatial patterns from gear-specific patterns. Additionally, each net sampled different parts of the water column, so in addition to spatial and gear differences, ontogenetic shifts in larval sucker behavior also could account for some of the variation. The youngest larvae feed in the water column, and over time a gradual transition to benthic feeding occurs. Although the approximate 50 percent surface-to-50 percent benthic diet transition occurs at a length of about 20 mm, individuals vary greatly and benthic foods can make up 25 percent or more of the diet in a 15 mm larva (Markle and Clauson, 2006), which indicates that gradual and variable vertical transition in orientation is a likely cause for some of the differences observed in catches. 


Variation in the duration of nighttime-only drift in 2009 had no noticeable effect on correlations. However, density simulations showed that nighttime-only drift throughout the channel would be manifested at sites close to the channel as nearly zero daytime densities, as larvae left the channel in “pulses” at night only. Yet, the highest larval catch densities were at sites A and B, which were close to the channel, and where catches were made during the day. This could imply that the nighttime-only drift behavior of the larvae does not persist far below the upstream boundary at Modoc Point Road bridge, either because of an ontogenetic shift, or because the transition from well-defined channel to open-channel sides at the Delta boundary makes it difficult or impossible for the fish to sustain the behavior. Prior to Delta restoration, larvae left the drift during the day by moving to areas with lower currents at the channel sides (Cooperman and Markle 2003). After restoration, this may simply have become more difficult where the channel is no longer constrained and water moves freely over the banks. Some other form of behavior also could be implied, however, such as the ability of even the youngest larvae to hold their position in vegetation or other suitable habitat, in which case the assumption of passive transport made in the model would break down for individuals reaching such habitats. 


The field data provided moderate levels of corroboration for density based simulations. However, there are many sources for disconnects in biophysical models such as this (Leis, 2007), and our approach highlights some weaknesses that need to be improved for both the field data and the simulations. 


Although our size-based spatial patterns tended to corroborate simulations, as noted previously, potential differences in gear efficiencies for different sized larvae need to be determined to more rigorously define those patterns. Intra- and inter-gear differences in efficiencies are seldom mentioned in larval dispersal modeling (Leis, 2007), with most researchers simply acknowledging that different gears deployed in different ways in the same place sometimes give different results (Overton and Rulifson, 2007). Estimates of size selection curves (Millar and Fryer, 1999) would be a first step to improving field data. 


The models also predicted very low densities outside the river. If fish are uniformly distributed, these densities are close to a “density detection limit” for the gear types used, and there is likely to be a great deal of imprecision in the larval catches. For example, at one larva per net, the larval trawl detection limit was between 0.14 and 2 fish per cubic meter; the pop net detection limit between 0.37 and 3.0 fish per cubic meter; and the plankton net detection limit between 0.012 and 3.7 fish per cubic meter. Our field data and models also differed in temporal scale. The coarse temporal resolution (once every week to once every 3 weeks) of the larval catches contrasted with the time scale of the models. For example, it was not possible to determine true peak densities at fixed sites, and to compare them directly to the peak densities at the Modoc Point Road bridge when simulations predicted a decrease in the proportion of larvae exiting the Williamson River mouth after levees were breached. Both of these problems typically require more samples or sampling in a different manner, such as adaptive sampling (Thompson, 2002). 


Overall, the model simulations showed that a primary objective of Delta restoration—to provide nursery habitat for larval suckers—was met (David Evans and Associates, 2005). Prior to restoration, the larval plume entered Upper Klamath Lake at a single point on the shoreline corresponding to the mouth of the Williamson River (Erdman and others, 2011). After restoration, the larval plume was simulated to leave the Williamson River channel and spread out on both sides of the Delta, before entering Upper Klamath Lake through several different openings in old levees, including openings into Agency Lake. Once both sides of the Delta were restored, the model simulation results showed that the plume was likely to spread more quickly and at a higher concentration through the Goose Bay side of the Delta under prevailing wind conditions. Because basic validation for the model has been provided by the comparison to larval catch data, it can now be used with some confidence to simulate the transport of the larval plume under varying conditions of flow, lake elevation, and wind (Wood, 2012). The model also can be used to predict how larval transport will change in the future as Delta vegetation matures, if the drag of various vegetation types is accurately incorporated and an accurate map of vegetation in the Delta is made available. Finally, a potentially important use of the model is to guide larval sucker sampling programs. 


Simulations would be improved by a better understanding of larval non-passive behavior. The first important step would be to determine when the ontogenetic shift to non-passive transport occurs in order to understand how long the simulated plume can be expected to represent realistically the dispersal of a cohort. At least two other aspects of non-passive behavior can be identified. First, the model would be improved by a better understanding of how far down the Williamson River channel the nighttime-only drift behavior persists. Second, strong swimming behavior, if completely random, would have the effect of increasing the effective dispersion of the larval plume. Such behavior could be simulated using the advection/diffusion approach that was used for this study by using a larger diffusivity coefficient in the model. Other types of non-passive behavior would potentially require a different approach. For example, the modeling approach used here cannot simulate the behavior if certain types of vegetation or a particular depth range act as “attractors” to the larvae, causing them to avoid passive transport, slow down, and accumulate in certain areas. A different approach, such as particle tracking, could be used to simulate such behavior.


First posted April 2, 2012