B08_Attachment 17 - SAMPLE_Large Review_ISAP Eval draft revised Appx D PSPAP Monitoring Plan_103020.pdf

ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 1 of 40

ISAP evaluation of the Pallid Sturgeon Population Assessment Program monitoring plan – revised Appendix D of the Missouri River Recovery

Program Science and Adaptive Management Plan

Independent Scientific Advisory Panel

30 October 2020

This report responds to a charge1 from the Missouri River Recovery Program (MRRP) to the Independent Scientific Advisory panel (ISAP) to review the Pallid Sturgeon Population Assessment Program (PSPAP) monitoring plan. The ISAP2 appreciates the efforts by the Corps during this past summer to assemble in a single document a review-version of a monitoring plan that continues in development. The ISAP recognizes that work toward monitoring the status and trend of pallid sturgeon in the upper and lower Missouri River has a long and uneven history, that there is institutional desire to take advantage of past data collection in informing current and future management decisions, and that future data collection might be constrained by funding limitations. Against that background, the ISAP lauds the Corps for advances in its pallid sturgeon population modeling efforts and acknowledges the Corp’s intent to bring statistical power to its monitoring deliverables. The ISAP understands that pallid sturgeon monitoring needs to service explicit information needs that are articulated in the 2018 Biological Opinion (for operation of the Missouri River main stem reservoir system et al.) and the intent of the 2018 Science and Adaptive Management Plan (SAMP, under the MRRP).

The revised PSPAP – presented as Appendix D of the SAMP provides the foundation for Missouri River pallid sturgeon monitoring and is intended to meet programmatic information needs. Specifically "The PSPAP is designed to meet many needs that include tracking progress in meeting MRRP sub-objectives, estimating metrics associated with evaluating incidental take specified in the 2018 BiOp, providing inputs for the IPSPM, developing relationships between annual flow variation and reproductive success and recruitment to age-1, and providing flexibility given uncertain future funding (e.g., potential for reduced efforts) and unpredictable environmental conditions (e.g., high flows can limit river access)" (Appendix D, page 5). The sub-objectives for the MRRP program are to increase pallid sturgeon recruitment to age-1 and maintain or increase the numbers of pallid sturgeon as an interim measure until sufficient and

1 See Appendix A of this report.

2 This evaluation by the ISAP was authored by Steve Bartell (chair), Steve Chipps, ad hoc panelist Barry Noon, Chris Guy, Dennis Murphy, and Melinda Daniels.

ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 2 of 40 sustained natural recruitment occurs. Metrics for the program are identified in the sub-objectives:

1) catch rate of age-0 and age-1 pallid sturgeon, 2) abundance of age-0 and age-1, derived from modeling, and 3) survival of hatchery and naturally reproduced pallid sturgeon to age-1. Those metrics are to be used to determine program progress in the evaluation phase of the adaptive management process and to justify adjustments in the management of pallid sturgeon.

Establishing whether the programmatic sub-objectives are being met is the fundamental responsibility of the MRRP monitoring and assessment effort, which includes project effectiveness monitoring, which is not the subject of this review.

In this report, the ISAP responds to five (compound) “questions” framed by the Corps and vetted through the Missouri River Recovery Implementation Committee’s (MRRIC’s) Fish Working Group. It should be understood that had the questions been posed in an interactive, in-person MRRIC forum, those queries would have initiated discussion that would range widely into issues bearing on approaches to data collection, sampling design considerations, alternative uses of gear and tools, and timing of field operations, all towards implementing effective, efficient, and accountable pallid sturgeon monitoring – in detail beyond the scope of this review. Here the ISAP answers the questions and presents observations that clarify the answers or provides relevant explanatory information. However, the responses here are not exhaustive per se; the present exchange cannot be viewed as finished product. In developing an omnibus monitoring scheme for pallid sturgeon that can inform the breadth of management decisions under the MRRP that lie ahead, there are more questions to ask and certainly more to say in response.

Understood through that lens, the PSPAP monitoring plan – Appendix D from the SAMP – is not complete, nor with agency responses to review comments offered in this report will it be “finished.” Rather the PSPAP document should be viewed as evergreen, subject to continuous refinement. Pallid sturgeon monitoring under the MRRP should be viewed as adaptive, just as is the management agenda targeting pallid sturgeon. Programmatic monitoring will be subject to continuous amendment and should be subject to periodic review, perhaps linked to review of the Adaptive Management and Compliance Report or less-frequent programmatic review.

The following report sections reflect the ISAP’s detailed evaluation of Appendix D and supporting materials identified in the task charge (Appendix A of this report). The review was guided by charge questions provided to the ISAP by the MRRIC (in Appendix A and recounted in section headers below). The ISAP offers a number of recommendations that the Corps might consider as the PSPAP moves forward through implementation, and closes the report with a coda, setting realistic expectations for pallid sturgeon management and the ability of resource managers to detect pallid sturgeon responses to program actions.

ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 3 of 40

Responses to the review questions

1. A component of PSPAP is age-0 sturgeon monitoring as well as monitoring of recent recruits (age-1 to age-3 pallid sturgeon). The primary purpose of this component is to gain reliable estimates of pallid sturgeon reproductive success and recruitment and track changes over time which can then be related to river conditions (e.g., natural annual flow variations).

a. Will the spatial extent, temporal scale, and intensity of age-0 sampling in the currently proposed sampling scheme provide sufficient and reliable data that can be used to assess reproductive success, parameterize population models, evaluate effect of natural flows events, and contribute to assessing performance of targeted management actions?

Given the current uncertainty in the estimates of CPUE and occupancy rate, it is challenging to

1) assess reproductive success with regard to targets, 2) estimate population model parameters,

3) determine the effects of natural flows events on reproductive output, and 4) assess the performance of targeted management actions. The ISAP recognizes the difficulty in sampling age-0 and age-1 pallid sturgeon; the comments in this review are intended to help the Corps meet realistic expectations for the sampling and analyses as they relate to program objectives.

Monitoring can be viewed as the repeated sampling of a population over a span of time (Hankin et al. 2019). The span of time and frequency of sampling is dictated, in part, by the ecology and life history of the target species. For pallid sturgeon, a long-lived species with delayed age at first reproduction, the study duration is likely to exceed a decade or more in order to understand whether the population is responding positively to management actions (White 2019). At its foundation, the proposed monitoring plan for pallid sturgeon represents status and trend monitoring (Reynolds et al. 2016). Status is often defined in terms of the value of the monitoring state variable (herein population abundance) at a particular point in time, and trend is defined as a change in status over a period of time.

Drawing from the discussion in Hankin et al. (2019), the design of a monitoring program has two interrelated components: 1) the sample design used to select the sample units on each sampling occasion and 2) the temporal schedule (frequency and duration) for surveying the sample units.

These are often referred to as the membership design component and the revisit design component, respectively. One way to understand membership design is to envision a list of all the primary sample units within the sample frame. For pallid sturgeon monitoring, this would be a list of the all the river bends, the primary sample units, within the range of the species in the Missouri River (the sample frame). The sample of units to survey would then be based on some probabilistic selection of n units from the list. The revisit design can be complex; it could involve many ways of conducting surveys in time and space. Designs refer to “panels” where a panel

ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 4 of 40 represents a group of survey units all sampled during the same time period (McDonald 2013).

Panel designs range from always revisit where the same set of n sample units are revisited on each sampling occasion, to never revisit where no sample unit is ever revisited. The sturgeon monitoring plan is an always-revisit design where only a single panel is formed and the same set of sample units is surveyed on each occasion (each year). Many environmental sampling designs use an equal probability of selection from the membership list and an always-revisit survey schedule.

In general, monitoring programs require two key actions: 1) estimating the current state of the system (status) and 2), estimating how system state is changing over time (trend). Appendix D has multiple references to “status and trend monitoring,” but includes minimal discussion of how the trend analyses will be conducted. These two components, status and trend, are directly tied to issues of sample design.

Design issues relevant to estimating the current state of the system focus on 1) the number of sample units required for a precise and unbiased estimate of current state; and 2) how the sample units are arranged in space. In the current design, the sample frame for monitoring is hierarchically structured according to river segments (n=11), bends within segments, and macrohabitats nested within bends. Bends are the primary sample unit. Within each sample unit, surveys are conducted to estimate the occurrence of age-0 fish using occupancy models, and the abundance of age-1 and older fish using mark-recapture methods. Decisions regarding sample sizes require an a priori statement of desired precision on the estimates of occupancy rate (age-0 fish) and population abundance (age-1 and greater).

Appendix D (page 34) states that “… the precision needed in the population estimates needs to be determined through discussions with the agencies.” Questions about the sample size requirements cannot be completely answered until precision targets are set. Issue two requires a determination of the required spatial allocation of sample units to achieve the defined precision targets. If there is pronounced spatial heterogeneity in the value of the state variables (occupancy and abundance), then a stratified design will usually achieve precision targets with smaller sample sizes. Stratification may partially address the issue of low sample size. This question and Appendix D (page 149) make reference to the possibility of a stratified design, but this may not have been implemented to date.

Addressing the adequacy of the spatial extent and temporal scale of the survey design can be usefully informed by a priori estimates of the components of variance analysis. Appendix D does not include that analysis. For example, if catch rate (C) = CPUE is the state variable, then

,( , ) s t s tC s t segment timeµ ε= + + +

ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 5 of 40 and information on the following variance terms is required:

~ (0, )

~ (0, ) ~ (0, ) s segment t time s t residual segment time σ σ ε σ

Question 1a is difficult to address because the scale of the experimental unit is not clearly described. Scale is important because it influences the sample size and how data are analyzed (are pooled) to address the sub-objectives, flow events, and targeted management actions. For example, Appendix D states (lines 1121 and 1122) that "yearly changes in abundance of young sturgeon will be measured using CPUE or changes in occupancy rates." Are the yearly changes at the macrohabitat scale, segment scale, or management unit pooled to represent the upper river and lower river spatial extents? Macrohabitat appears as the scale for the experimental unit for CPUE from the data presented in Table 15. If this assumption is true and those data are representative of the variation in CPUE for age-0 pallid sturgeon (i.e., coefficient of variation = 248%), then it is doubtful that changes – assuming standard parametric statistical methods are used to describe changes – will be detected given the uncertainty of the CPUE estimates.

(Uncertainty in CPUE is acknowledged in the Appendix D.) Current methods of population monitoring rely on CPUE, which is confounded by factors that affect catchability, including changes in flow. As a result, CPUE does not provide accurate data for model calibration, and in the lower Missouri River is poorly correlated with trends in stocking of hatchery fish (see section 4.4, page 391). Despite those concerns, CPUE remains an imperfect but primary performance metric for a number of important questions related to abundance of age-0 pallid sturgeon (Tables 63 and 64 in Appendix D).

The term "changes" as used in Appendix D lines 1121 and 1122 is vague and there seems to be no description of the statistical approach(es) necessary to evaluate changes (trends). Assuming "changes" is related to the targets on page 6, how will the metrics be statistically evaluated in relation to the targets? Will changes through time be evaluated using a time-series approach to detect "increasing levels of recruitment over time," as mentioned in the targets for sub-objective 1? Given that "changes" will be difficult to detect because of uncertainty in the CPUE estimates, relationships between CPUE and flow or targeted management actions will also be difficult to discern. The spatial scale for CPUE needs to be considered for the relationships with flow and targeted management actions. For example, would the relationship between CPUE and flow have a single data point for each year or each pallid sturgeon macrohabitat? In the former there would be a sample size of nine (for nine years) and in the latter it would be much higher (a sample for each macrohabitat sample by year). Furthermore, is it realistic to assume that CPUE estimates among neighboring macrohabitats are independent?

ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 6 of 40

Sections D.6.3.1 and D.6.3.2 in Appendix D describe the spatial and temporal aspects of pallid sturgeon age-0 and age-1 sampling. Focusing on areas and habitats with high CPUE values for young Scaphirhynchus would seem to maximize the probability of capturing age-0 pallid sturgeon. The only caveat is that consideration be given to obtaining samples through at least late July (preferably August) in the UMOR. Braaten et al. (2012) reported that by the end of their first growing season, naturally growing age-0 pallid sturgeon in the UMOR vary in size from 120-140 mm. By age-1, the mean size of pallid sturgeon was 187 mm, close to the value of <200 mm reported on line 1615 “…proportion of bends occupied by age-0 sturgeon (<200 mm)… will be used to estimate basin-level occupancy,” noteworthy because all fish <200 mm will be classified as ‘age-0’ as presented in Appendix D. Given that size of fish and date of collection will be recorded, it may be worth considering two categories of age-0 pallid sturgeon (age- 0,<140 mm versus age-0, 141- ~200 mm) for occupancy modeling analyses that explore the influence of covariates on age-0 fish. These size ranges probably differ in the LMOR, but a similar differentiation should be considered (first year growing season vs first year winter survival).

Reproductive success in the PSPAP is assessed (estimated) in part through an occupancy analysis. There are a number of concerns regarding the occupancy analyses for age-0 and age-1 sturgeon. Occupancy has a relatively low information content. Occupancy is primarily used to estimate the spatial distribution of a species – it does not allow direct inference to abundance (but, see caveat below). Occupancy rate is a property of the sample unit, not directly a property of the species. All else being equal, a larger sample unit is more likely to be occupied, because it can support more individual sturgeon. In the sturgeon monitoring, inferences to occupancy rate are estimated at two spatial scales – the segment and the bend. Bends are nested within segments in the design, similar to a two-stage sample design. The primary sample unit is the bend.

However, bends vary in size and presumably in sturgeon abundance within a sample unit, introducing heterogeneity in detection probabilities that cannot be directly modeled because abundance is unknown. The occupancy rate of age-0 fish, therefore, might not provide much useful information on the magnitude of recruitment. Because occupancy rate of age-0 sturgeon is not an estimate of age-0 abundance, it is not clear how estimates of this rate can be used to parameterize an age-based demographic matrix model.

Many of the occupancy rates reported in the recent progress report were very high, even though detection probabilities were low. High occupancy rates could be the result of sampling large areas, increasing the likelihood of capturing at least one individual. In addition, since the number of units where one or more fish were detected is adjusted for detection probability, small detection probabilities can result in inflated (and biased) estimates of occupancy. To help understand how low detection probabilities can lead to inflated (and imprecise) estimates of occupancy (ψ), consider the following. Let s represent the number of bends randomly sampled

ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 7 of 40 and sD the number of bends where 1 or more age-0 sturgeon were detected. Then, a naive estimate of occupancy (and its variance) is

ˆ

(1 )ˆvar

Ds s s ψ ψ ψψ

The above estimate of occupancy assumes perfect detectability (that is, no observation error).

However, if we assume population closure and conduct repeated surveys (k) on at least some of the sites, an estimate of the per-visit detection probability (p) is possible. Given an estimate of p, we can correct our estimate of ψ to adjust for failure to detect age-0 fish in sample units where they were present:

ˆ

(1 ) (1 )ˆvar( ) , where, 1 (1 ) (the probability of 1 detections in k surveys)

D k s sp p s sp p p ψ ψ ψ ψψ

When estimating ψ , a small p* value can greatly inflate the estimate of occupancy. In addition, the estimate of ψ is less precise because the variance of ψ now has two components: 1) the binomial variance associated with the true value of ψ , and 2) variance due to imperfect detection of age-0 sturgeon. This second component can be large if p* is small, leading to imprecise estimates of ψ .

The occupancy values presented in Table 11 are imprecise and might not be overly informative, unless the uncertainty can be reduced. However, it may be that the described analyses are appropriate and that the presentation of the methods is simply lacking in detail. There is some discussion in Appendix D that occupancy data will be supplemented with CPUE data, which might help infer the magnitude of recruitment. As a counterpoint to the preceding concerns, a strong relationship between occupancy and abundance was demonstrated by Gaston et al. (2000).

Similarly, a recent paper based on a study of a freshwater fish assemblage found a relationship between abundance and occupancy (Miranda and Kilgore 2019). Occupancy may be a better monitoring state variable than suggested in the previous discussion.

ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 8 of 40

b. Similarly, catch rates of age-1 to age-3 pallid sturgeon are used to assess recruitment and relate to annual flow variation. Will the spatial extent, temporal scale, and intensity of the proposed age-1 to age-3 sampling allow resource managers to assess recruitment, parameterize population models, relate recruitment to annual flow variations, and contribute to assessing performance of targeted management actions?

Many of the concerns outlined in 1a above for age-0 pallid sturgeon apply to monitoring catch rate and recruitment of age-1 to age-3 pallid sturgeon. Several concerns are based on the preliminary data presented in Tables 11 and 15 for fish >200 mm that are mostly age 1 to age 3.

As with age-0 monitoring results, there appears to be considerable uncertainty in CPUE and occupancy rate estimates. Correspondingly, there would need to be a large effect size to detect a statistical difference in status, trend, or relation to flow variation based on the PSPAP monitoring scheme described in Appendix D. Addressing the adequacy of the spatial extent and temporal scale of the survey design requires a priori estimates of the components of variance analysis, as previously noted for age-0 pallid sturgeon.

Catch of age-1 to age-3 sturgeon prove that recruitment of fish to these ages occurred, but are accompanied by many uncertainties. For example, age-3 fish represent age-0 fish that have survived 3 years, 3 2 1 0 0( )n s s s n= and, thus inferences derived from older fish confound information on survival and the number of age-0 fish ( 0n ) actually recruited into the population 3 years prior. The proposed sampling of age-1 to age-3 pallid sturgeon, as described in Appendix D, provides data for preliminary estimates of recruitment, to estimate model parameters, to relate recruitment to variation in flow, and to evaluate performance of targeted management actions (including IRCs and spawning habitat). However, estimates of these relationships may be highly uncertain if not adjusted for size-dependent differences in catchability.

The primary challenge lies in the intensity of sampling for this rare species throughout a large, complex river system. Given the variability (lack of precision) of data available to date, it appears that the PSPAP will be challenged to provide data of sufficient accuracy and precision to usefully quantify recruitment of age-1 to age-3 pallid sturgeon. There are restrospective analyses we can use to evaluate recruitment variabiliy and factors related to recruitment variability, that incorporate CPUE values of fish that are fully recruited to the gear, by using weighted regression of loge number-at-age vs year-class (year) and evaulating residuals (Maceina and Bettoli 1998).

As an example, for age-1 to age-5 sturgeon collected in a single year, catch-curve regression could be used to evaluate relative year-class strength by comparing residual values (+ or –) to environmental characteristics that were present when fish were produced (i.e. age-0).

ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 9 of 40

c. Can the proposed monitoring scheme separate the effect of flow variations on catchability and reproductive success/recruitment?

Evaluating the effects of flow variation on reproductive success and recruitment depends on 1) monitoring age-0 production and age-1 recruitment across a range of natural flows and 2) estimating a corresponding functional relationship between flows and population metrics for incorporation into the IPSPM. Evaluation of flow events also require that one or more metrics that characterize flow be included in the analysis as covariates on occupancy rate. The flow metrics are not presented in Appendix D. Useful flow metrics would likely include measures such as minimum, maximum, and average discharge; water elevation; minimum, maximum, and average water velocity; and duration of increasing and decreasing discharge.

While question 1c uses the term “catchability,” the question actually refers to catch per unit effort (CPUE) as discussed in Appendix D. Catch per unit effort – the number of individual fish caught per meter trawled – has long been used in fisheries assessments as an index of fish abundance. CPUE is related to population abundance (N) as

CPUE = qN where q is the proportion of fish removed by one unit of sampling effort (the catchability coefficient). Because CPUE is the product of abundance and catchability, trends in CPUE over time can reflect changes in abundance, catchability, or both.

Catchability can be affected by a number of factors including fish size, environmental conditions, and gear efficiency; it is difficult to determine without reliable estimates of fish abundance (population size). For heuristic purposes, Figure 1 illustrates the potential impacts of size on catch probability based on simulations using an assumed lognormal distribution of fish sizes with a mean size of 500mm. It is important to note that CPUE, when not adjusted for catchability, can lead to uncertainty in fish-abundance estimates. However, under the unlikely scenario that catchability is similar among sample units, CPUE might provide a reliable index of fish abundance (Pierce and Tomcko 2003). Unfortunately, the relationship of CPUE to the true population size N is unknown, because catchability (q) is not estimated for pallid sturgeon < age 3 in the PSPAP. In addition, q is expected to vary with fish size (length) (see Figure 1), sampling gear, and across time and space. Given the sources of variability in “catch rate” described above, the proposed monitoring scheme will be challenged to separate the effects of flow variation from other confounding factors. Moreover, for occupancy estimates, the highest point estimate of detection probability was 0.19 for age-0 sturgeon (< 200 mm); estimated occupancy values were highly variable (Bayesian credible interval 0.008 – 0.69 in Table 13). Similar to CPUE, there

ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 10 of 40

Figure 1 – (a) simulated probability of capture (p) on the y-axis versus fish size (mm) on the x-axis; and (b) the frequency distribution of capture probabilities from the simulation based on a population size of 500 fish was considerable uncertainty in detection probability, implying it may be difficult to conclusively determine the effects of flow variation on detection probability.

2. Recognizing the pressing need for information on the distribution and survival of age-0 pallid sturgeon, several activities have been proposed to compensate for low capture rates, including evaluating use of shovelnose sturgeon as a surrogate species for evaluating reproduction and/or recruitment, increasing stocking of very young pallid sturgeon, improving identification and characterization of high catch areas to increase ability to stratify sampling effectively (e.g., our targeted sampling in June, 2020 in the LMOR produced about 16,000 age-0 sturgeon), and extending sampling into the middle Mississippi River.

ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 11 of 40

a. Are these additions or amendments to the current sampling design(s) likely to enhance the ability to achieve the goals of the monitoring plan?

The use of shovelnose sturgeon as a surrogate species for pallid sturgeon continues to be an important topic in the MRRP and research should continue toward identifying/demonstrating similarities and differences between the species at age 0 and age 1. Surrogate systems use the response of one species to environmental disturbances or management actions to predict the response of another species to similar environmental phenomena. Use of a substitute species (shovelnose sturgeon) when the focal species (pallid sturgeon) is usually too rare to measure is defensible to the extent that similarities can be demonstrated between the life histories and ecologies of the surrogate and focal species. If a defensible argument can be made that the population-level response of these two sturgeon species to the same environmental stressors and management actions is sufficiently similar, then surrogacy can be a defensible assumption.

In general, the described additions to the sampling plan might increase the likelihood of achieving the goals of the monitoring plan. In particular, understanding recruitment failure in pallid sturgeon might prove to be a key factor in understanding recruitment success of shovelnose sturgeon or vice versa. For two closely related species living in the same environment, where one is arguably successful, while the other is not – there must be a critical, autecological difference between the species at some life-stage(s). For that reason, “… evaluating use of shovelnose sturgeon as a surrogate species for evaluating reproduction” requires some clarification. Although shovelnose sturgeon and pallid sturgeon are quite similar in many respects, there must be fundamental differences in their ecology. Continuing research efforts to discover those differences will be useful in answering questions about surrogacy.

One observation that is frequently mentioned in Appendix D with regard to age-0 sturgeons is the association between catches of small shovelnose and pallid sturgeon; that is, age-0 pallid sturgeon frequently are caught in combination with age-0 shovelnose sturgeon. Assuming that detection probabilities of age-0 shovelnose and pallid sturgeon are similar, there are approaches for comparing spatial patterns in the distribution of these species, including a Mantel test (Fortin and Gurevitch 1993) or a two-dimensional Kolmogorov-Smirnov test (Garvey et al. 1998). These approaches rely on arranging data into bivariate spatial coordinates (e.g., UTMs) and can be used

1) to identify whether a single distribution has arisen by random effects or 2) to compare two bivariate distributions (Garvey et al. 1998). In the example in Figure 2, the distribution of Species 1 (triangles) and Species 2 (circles) in panel A reveal no difference in their spatial distribution, whereas their spatial distributions revealed in panel B are different. Knowing whether years with high catches of age-0 shovelnose sturgeon positively correlate with

ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 12 of 40

Figure 2 – Example results from arranging spatial data into bivariate spatial coordinates on pattern detection (Chipps and Garvey 2007) years of high catches of pallid sturgeon age-0 could provide insight into whether mechanisms that result in strong year classes of shovelnose sturgeon also result in recruitment for pallid sturgeon – this is particularly relevant in the lower Missouri River basin and middle Mississippi River.

If the assumption of equal detection among pallid and shovel nose sturgeon is not tenable, occupancy models can be applied to estimate the probability of co-occurrence – that is, the probability that two species occur in the same sample unit (MacKenzie et al. 2018). These methods explicitly estimate detection probabilities for both species, allowing them to differ. In addition, dynamic models can be fit to multi-year survey data to estimate how co-occurrence patterns change over time.

Stocking and subsequent sampling of age-0 pallid sturgeon and shovelnose sturgeon could provide insight into sampling efficiency and how to reduce variation in CPUE and occupancy rates. This combined stocking could be used to further elucidate the similarities and differences between species with respect to catchability (detection probability) and drift dynamics, which could be used to improve the PSPAP sampling design and address questions regarding surrogacy. Stocking a known quantity of age-0 pallid sturgeon and subsequent recapture could

ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 13 of 40 also be used to put the number of wild age-0 pallid sturgeon in context. Stocking of age-0 sturgeon has been conducted in the upper Missouri River basin and tied to specific research questions related to drift – the studies provided information on drift dynamics that was subsequently used to propose management actions. Similar efforts could be conducted in the lower basin to facilitate a more efficient and powerful monitoring program (that is, to reduce variation with the least amount of effort) for age-0 pallid sturgeon and shovelnose sturgeon.

Targeted sampling provides an efficient approach to increase the probability of catching age-0 pallid sturgeon. In the context of the PSPAP monitoring plan and status and trend evaluation, amendments to the current sampling design should be focused on reducing variation in the metrics used to address the sub-objectives (see recommendations below). Increased catch of age- 0 pallid sturgeon would provide critical information – particularly for research questions related to age, growth, diets, and genetics. However, it is also important to recognize that increased catch might not necessarily decrease variation in the metrics used to evaluate the pallid sturgeon population sub-objectives.

Targeted sampling also provides critical information to the described Bayesian state-space occupancy model applied to age-0 pallid sturgeon. However, the temporal distribution of random and targeted sampling of bends will take one of two forms (Figure 3). In panel (A), random sampling of bends is initiated and continues until the last 2 weeks of the season, when targeted sampling efforts begin. In panel (B), random sampling of bends is initiated and continues until abundance of age-0 fish increases “significantly.” Then all sampling shifts to targeted bends for 2 weeks, before returning to sampling random bends to end the season. There is concern with (B) in how it might affect CPUE. In essence, the interruption of randomized sampling, because of increased catch rates of age-0 fish, could result in depressed CPUE estimates for randomly selected bends. If so, that would likely lead to hyperdepletion in CPUE estimates, where catch rates are not proportional to abundance as given by

CPUE = qNβ , where β>1 (see Pierce and Tomcko 2003)

It appears that targeted efforts could bias random sampling in years when age-0 pallid sturgeon numbers increase, by shifting effort away from random sampling for two weeks (Figure 3B).

Conducting targeted sampling at the end of randomized sampling in years when abundance of age-0 pallid sturgeon does not increase might help to alleviate some of that bias (Figure 3A). A better, albeit less pragmatic, approach would be to randomly intersperse targeted sampling with randomized sampling; but that sampling protocol could create logistical issues and increase monitoring costs given the spatial distribution of bends in each segment. Nonetheless, it raises another question about the uncertainty in age-0 CPUE and whether gains for occupancy

ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 14 of 40

Figure 3 – Schematic of proposed random and targeted sampling of bend sites modeling (targeted efforts) and increased catch of age-0 fish outweigh the costs of randomized sampling for CPUE estimates. Information on the “timing” of age-0 sturgeon sampling is described in Appendix D at line 1397 – “Age-0/1 sampling will begin in late spring/early summer throughout the geographic range of PSPAP….” and “…the timing of sampling will be determined through water temperature, preliminary trawling data, and/or spawning information gathered by telemetry.” Given the importance of water temperature to spawning, embryo development, and duration of drift, it would be prudent to put the “timing of sampling” for age-0 fish into better context. Pallid sturgeon generally spawn at between 16-19 °C (Elliott et al. 2020).

At these water temperatures, embryos require between 13 to 7 days to hatch (Kappenman et al.

2013); post-hatch, it takes approximately 10 to 8 days at 16-19°C (or ~183 to 156 CTUs) for larvae to settle (Mrnak et al. 2020). It is likely that the timing of spawning varies across the geographic range of pallid sturgeon (Elliot et al. 2020), but it is not clear in Appendix D how this issue is addressed and what water temperature criteria are considered to initiate sampling.

Given size-dependent mortality in young sturgeons, any “peaks” in annual abundance would likely be more pronounced during the meso-larvae life stage (22-80 mm), compared to the meta-larvae stage (> 80-150 mm). This information is important to better clarify what size range of

ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 15 of 40 fish are being considered ‘age-0’ pallid sturgeon (Figure 4). Text in Appendix D references age- 0 pallid sturgeon simply as fish <200 mm (line 1615).

Figure 4 – Early life-stage development in age-0 pallid sturgeon. (Photo: Matt Wagner, SDSU. Illustrations: Snyder 2002)

b. Does the ISAP have other recommendations for handling the challenges of low sample size?

Low sample size can be overcome by increasing targeted sampling. At this point in the program, it is important that efforts to capture age-0 and age-1 sturgeon are maximized to the extent possible. However, sampling targeted sites reduces the sample frame (the spatial extent of the riverine ecosystem to which inference can be made). We caution that any sort of judgmental sampling will likely result in biased estimates that undermine statistical inference to the entire sturgeon population (see Thompson 2012). The concern expressed in this recommendation – high variability among sample units (bends) in the abundance of age-0 sturgeon and many zero counts – is valid.

To address the problem, a stratified random sample design could be adopted, where abundance strata are based on the catch data from previous years. Stratified sampling is a probability-based sampling design, where sample units (bends) are first stratified into, for example, ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 16 of 40 high/medium/low abundance strata and a random sample of map-units is selected from each stratum. In addition, the number of randomly selected sample units within each stratum should be based on an optimal allocation of samples across the strata (Thompson 2012, pp 146-148). In an optimal allocation design, the number of samples within each stratum depends on 1) the size of the stratum (i.e., number of sample units within that stratum), and 2) the variability in the catch within each stratum. In this design, a stratum with high abundance of age-0 sturgeon will generally have more sample units selected than a stratum with low abundance. This generally occurs in biological studies involving counts because variability in the count increases with the mean count. A stratified design can yield unbiased, probability-based sample estimates.

To illustrate the probability-based approach, consider the conditions of three abundance strata (H, M, L), sufficient resources to sample n sample units (bends), and no prior information on stratum size (number of bends) or variance in the catch. In this case, one would adopt an equal allocation design

3h nn = where, hn = the sample size in stratum h. However, there is information on the number of bends in each stratum, so a proportional allocation design could be developed h nNn N where hN = the number of bends in stratum h, and N = the total number of bends across all strata. If estimates of the standard deviation in the count ( )hσ in each stratum based on past data are available, an optimal allocation design is possible h h h h h h nNn N σ σ

This optimal design should result in more samples being taken in bends with a greater number of age-0 sturgeon and fewer samples in bends with few age-0 sturgeon.

As discussed previously, two primary variance components affect all monitoring programs – sampling and process variation. Only sampling variation can be reduced by increasing sampling intensity. Unfortunately, process variation is the dominant variance component in most environmental monitoring programs. Decreasing sample size will simply increase the sampling variation component when estimating trend. This is demonstrated by the following equation, ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 17 of 40 which estimates the variance in the average linear trend (slope) across sites (from Urquhart 2012). A decrease in s will increase the variance estimate according to

ˆvar( ) residual time s t t σσ β where s = number of sample units (bends or segments) t = time (year)

Very low catch rate, combined with uncertainty in CPUE and occupancy (detection), presents a challenge to traditional approaches for examining changes in age-0 pallid sturgeon abundance.

While targeted efforts may reduce uncertainty and improve our ability to assess age-0 pallid sturgeon abundance and occupancy, the time-frame for acquiring these data is not known, and could delay attempts to explore responses of age-0 pallid sturgeon to environmental changes and management actions. Furthermore, environmental conditions that predict occupancy or abundance can differ substantially (Dibner et al. 2017), complicating decisions over which metric (CPUE or occupancy) is a better predictor of age-0 pallid sturgeon responses to environmental conditions, management actions, or both.

Exploring the relationship between pallid sturgeon abundance and occurrence can be useful for evaluating the direction and magnitude of potential environmental effects (Figure 5). In the example, each point on the graph (green triangle) represents a different year of monitoring data for a given Segment. The initiation of age-0 sampling as set forth in Appendix D sets a baseline (or starting point) for age-0 pallid sturgeon that is characterized by very few fish (low CPUE) showing up at very few sites (low occupancy) represented by the red circle. Over the following 9 years, positive changes in age-0 pallid sturgeon could follow one of three general trajectories in relation to environmental conditions, management actions, or both. Trajectory (A) shows increased occurrence of age-0 pallid sturgeon, with little change in abundance (CPUE; i.e., very few age-0 pallid sturgeon, but at more sites). Increased occurrence might be expected as a result of system-wide changes (e.g., flows) that enhance age-0 pallid sturgeon abundance (CPUE) and hence, distribution. Trajectory (B) shows an increase in the abundance (CPUE) of age-0 pallid sturgeon, but at very few sites, implying that local conditions, rather than system-wide effects may be associated with increased survival and abundance. In Trajectory (C), both abundance

ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 18 of 40

Figure 5 – Specific-abundance versus occurrence of age-0 pallid sturgeon. Adapted from Amundsen et al (1996)

(CPUE) and occurrence of age-0 pallid sturgeon increase compared to baseline (the management objective). Similarly, patterns in annual estimates of age-0 shovelnose abundance and occurrence could be compared to those for pallid sturgeon.

A heuristic approach that includes information on both abundance (CPUE) and occurrence might be helpful in exploring changes in age-0 pallid sturgeon. The approach is based on the relationship between specific-abundance and frequency of occurrence of age-0 fish. For a given Segment, frequency of occurrence (O) of age-0 pallid sturgeon is calculated as:

O = 𝑁𝑁𝑖𝑖

𝑆𝑆

𝑥𝑥 100 where Ni equals number of bends containing age-0 pallid sturgeon and S equals number of bends containing age-0 sturgeon (pallid sturgeon and/or shovelnose sturgeon).

Similarly, for a Segment, specific-abundance of age-0 pallid sturgeon (SA) is calculated as

ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 19 of 40

SA = ∑𝑁𝑁𝑖𝑖∑𝑇𝑇𝑖𝑖

𝑥𝑥 100 where Ni is the number of age-0 pallid sturgeon and Ti is the total number of age-0 sturgeon (shovelnose and/or pallid sturgeon) at bends with age-0 pallid sturgeon. Rather than assuming constant catchability across macrohabitats and time, it could be assumed that catchability of age- 0 pallid sturgeon and shovelnose sturgeon varies similarly across space and time – a reasonable assumption given that detection probabilities are similar for both species. As an example, consider the following five bends where a total of nine age-0 pallid sturgeon were collected and a total of 514 shovelnose sturgeon were collected. Specific abundance of age-0 pallid sturgeon is then calculated as:

SA = 2+1+1+3+2

265+59+81+3+115

= 9

= 0.0172 = 1.7%

A biplot of SA versus O can be useful for evaluating the direction and magnitude of changes in age-0 pallid sturgeon occurrence (see Figure 5).

3. Another component of PSPAP is population estimation of juvenile and adult pallid sturgeon.

a. Parameterizing the population model requires abundance, survival, and growth for the demographic matrix model and when employed as an individual based model additional information on spatial distribution, size distribution, growth, origin (hatchery, wild/unknown, hybrid), and movement. Is the monitoring plan set up to estimate values needed to characterize abundance, survival, spatial distribution, origin, and movement?

The PSPAP monitoring plan is designed to estimate abundance by fish origin and covers the necessary spatial distribution. Population estimates were calculated for hatchery, hybrid, and wild (unknown) pallid sturgeon by length category and management unit. Similar to other metrics, there is considerable uncertainty in the estimates, particularly in the Central Lowlands management unit (Table 19). Survival estimation is described and has been estimated for hatchery-origin fish in the upper basin (Rotella 2017). It appears a similar approach will be used throughout the basin. Survival estimates from embryo hatch through the first growing season for age-0 pallid sturgeon will continue to be difficult to estimate with any degree of certainty.

Based on information provided in Appendix D, it appears as if the mark/recapture design will provide estimates of most of the key demographic parameters. Table 2 usefully identifies

ISAP Evaluation of draft revised Appendix D PSPAP Monitoring Plan Page 20 of 40 parameters in the IBM and DM that can be informed by PSPAP monitoring activities. Appendix D states that a robust mark/recapture design (following Williams et al. 2002) will be employed.

This model combines open and closed capture/recapture data structures and can provide estimates of time-specific abundance, survival, capture probabilities, and temporary emigration rates. In addition, abundance estimates will be made for three surgeon size classes – <600mm, 600-800mm, and >800mm – with separate estimates for wild, hybrid and hatchery fish (see Appendix D, page 86). Individual estimates by size class seems logical; however, it is not clear why the size of captured fish is not treated as a continuous covariate. Incorporating size as an individual covariate in the robust design could be accommodated to model individual size-dependent heterogeneity in capture probability.

Under unique sampling processes, it is possible that targeted sampling may provide unbiased estimates of key population metrics. For example, Section D.6.2.1 indicates that sampling will emphasize upper Missouri River bends, where the highest catch rates have been historically encountered. Similarly, selection of sampled bends, gear types, and timing of sampling all appear identified in relation to previous success in collecting pallid sturgeon of varying age. At the same time, it should be noted that of 497 sturgeon captured, only 4 were pallid sturgeon. No age-0 pallid sturgeon were collected.

Parameterizing the population model gets at the question of verifying the model components;

that is, is the model structured correctly? Given the current description and understanding of the demographic model and associated inputs, the technical team has done a good job in capturing and explaining life-stage dependent survival probabilities, reproductive events, etc. In some cases, using values estimated from the literature or related studies, and in other cases with ‘first approximations’ reflecting the paucity of data. Furthermore, model developers familiar with the inner workings of the model have conducted simulation modeling for parameter estimates to document model verification. These are complex models with many input parameters. A rigorous “sensitivity analysis” of the models could help reviewers better identify which parameters and supporting data need to be carefully considered as part of PSPAP data collection efforts.

Understanding how response variables, such as population size, respond to sensitive input parameters (or management actions) will be important in shifting the question before planners from “Have we built the model right?” to “Have we built the right model?” For example, Section D.3.3 states that functional responses between a management action and a modeled demographic rate can be used to construct modeling scenarios to project outcomes or identify thresholds in pallid sturgeon response, but there is no mention in the PSPAP of attempts to develop functional responses for specific management actions.

The question also makes reference to an individual-based pallid sturgeon population model. The application of individual-based models (IBMs) has been successful for evaluating growth dynamics of age-0 pallid sturgeon in the Missouri River (Deslauriers et al. 2018) and should be applicable to juvenile and adult fish.