2.4 | Statistical Analysis
We developed occupancy and detection models using Program MARK accounting for imperfect detection (MacKenzie et al., 2002). There are four main assumptions for occupancy and detection modeling (MacKenzie et al., 2018). First, we assume populations are closed; our sampling on a day and at the three sites associated with a tributary (nine ichthyoplankton tows) all occurred within an hour to meet the assumption of no births, deaths, immigration, or emigration during a sampling event. The next assumption states occupancy and detection probabilities are constant among sites or that heterogeneity in these parameters is accounted for using covariates; we used spatial and environmental covariates to account for spatial and temporal heterogeneity of both parameters. The third assumption is detection histories are independent among sites; we accounted for this assumption through temporal replication where we sampled every 10 d and larvae were < 7 d old (Camacho et al., 2023; M. Weber, unpublished data), ensuring we did not repetitively sample the same larval across multiple sampling events. Finally, the model assumes no false positive detections; we selected taxa that had low ambiguity in visual identification and genetically identified a subset of bigheaded carp larvae to verify identification accuracy (see Camacho et al., 2023 for more details).
We developed encounter histories for each larval taxon and sample date as presence/absence (0 or 1) at each site upstream, downstream, or at the confluence of the major tributary. For example, an encounter history could be comprised of bigheaded carp only collected at the Des Moines River confluence and not at upstream or downstream locations (encounter history 010) or upstream and downstream but not within the confluence (encounter history 101). Occupancy modeling allows the use of covariates to improve model estimates and evaluate occupancy and detection under a range of environmental conditions. We assessed the effect of water volume filtered through the ichthyoplankton net as an effect of sampling effort on detection probability. We also assessed the effects of Julian date, mean and CV of weekly water temperature (°C), and discharge (m3/s) on site occupancy. We evaluated collinearity of covariates with a Pearson’s product-moment correlation analysis (all r < 0.6) before including them in occupancy and detection models. To ensure normality, we centered and scaled all covariates using z-scores prior to analysis. We assessed quadratic effects of Julian date and weekly river temperature on occupancy probability, as we sampled prior to and after hypothesized peak spawning periods and linear models would be inappropriate. We assessed all other covariates as linear effects. We assessed detection models by first comparing species and habitat models before adding additional effects of environmental covariates. Next, we retained the most supported detection model to evaluate species and habitat effects on occupancy. Finally, we assessed potential effects of environmental covariates on occupancy while retaining the most supported species and habitat effects. We ranked models based on Akaike’s Information Criterion corrected for sample size (ΔAICc ) and AICc model weight (wi ; Burnham et al., 2011). We considered models ranked ≤ 2 ΔAICc from the top model as competitive in analyses. Finally, we calculated detection probabilities under increasing sampling size (cumulative detection probability) using the most supported detection model (e.g., Kelly et al., 2021).
3 | RESULTS
We collected bigheaded carp larvae in pools 18 to 20 of the Mississippi River in 53 of 592 sample site-years (naïve Ψ = 0.09). Native larvae were captured more frequently, with freshwater drum (n = 223 of 592 site-years; naïve Ψ = 0.38) captured most frequently followed by shad (n = 141 of 592 site-years; naïve Ψ = 0.24) and percids (n = 132 of 592 site-years; naïve Ψ = 0.22). Thalweg and channel border habitats accounted for the most bigheaded carp (77%) and drum (78%) collections while shad and percids had similar overall collections among habitats (29 to 36%). We captured the most bigheaded carp between Julian days 138 to 240 when water temperatures ranged from 17.6 to 26.6°C and during the highest weekly discharge conditions (mean = 4,124 m3/s, SD = 1,057). We captured percids the earliest (day 113) during the narrowest time (range = 113 to 228, mean = 160, SD = 32), coldest temperatures (mean = 19.9°C, SD = 5), and second highest weekly discharge (mean = 3,629 m3/s, SD = 1,251). We captured freshwater drum the latest (day 256) across the widest range of days (range = 122 to 256, mean = 189, SD = 33) associated with highest weekly temperatures (mean = 23.4 °C, SD = 2.9) and lower discharge (mean = 3,321 m3/s, SD = 1,061). Finally, we collected shad on a similar time frame as freshwater drum (range = 120 to 256, mean = 174, SD = 29), with similar water temperature to bigheaded carp (range = 15.1 to 27.7°C, mean = 22.2 °C, SD = 3), under similar discharge conditions to percids (mean = 3,587 m3/s, SD = 1,100).