3.1 | Detection Models
Our most supported detection probability model while assessing taxa effects indicated differences in bigheaded carp and freshwater drum detections while shad and percids had similar detection (ΔAICc = 0.00, wi = 0.69; Table 1). The second ranked model with each taxonomic group estimated separately also received support (ΔAICc = 1.62,wi = 0.31; Table 1). Next, we assessed potential variation in detection probability for taxa among habitats. Our most supported model indicated detection of bigheaded carp and freshwater drum was similar between thalweg and channel border habitats combined that had different detection compared to backwater habitats (ΔAICc = 0.00, wi = 0.33; Table 1). In contrast, detection probability was similar among habitats for shad and percids (Table 1). Other model combinations receiving support include thalweg and channel border habitats combined for all taxa (ΔAICc = 1.20, wi = 0.18) and all habitats separate for carp and freshwater drum (ΔAICc = 1.29, wi = 0.17; Table 1). The final step of detection model selection was assessing the effect of water volume filtered (i.e., sampling effort) to the top taxa model. The addition of water volume filtered to all taxa (ΔAICc =0.00,wi = 0.35; Table 1) was the most supported model, whereas water volume filtered applied to only bigheaded carp and shad and percids (ΔAICc = 1.48,wi = 0.17) or only to drum and shad/percids (ΔAICc = 2.00, wi = 0.33; Table 1) were also considered competitive models.
The most supported model estimated bigheaded carp detection probabilities higher in the thalweg and channel border habitat (p= 0.15, 95% CI = 0.09 to 0.22) than in the backwater (p = 0.09, 95% CI = 0.05 to 0.17), but estimates had overlapping confidence intervals (Figure 2). Drum detection probability was also higher in the thalweg and channel border (p = 0.40, 95% CI = 0.35 to 0.45) compared to the backwater (p = 0.20, 95% CI = 0.15 to 0.26) while shad and percid detection probability was similar across all three habitats (p = 0.27, 95% CI = 0.23 to 0.31; Figure 2). Detection of bigheaded carp within each habitat tended to be lower than native fishes (Figure 2). Water volume filtered had a positive effect on detection estimates for all taxa/habitat combinations (β = 0.12, 95% CI = 0.01 to 0.23) where detection probability of bigheaded carp increased from 0.15 to 0.48, detection of freshwater drum increased from 0.34 to 0.77, and detection of shad and percids increased from 0.24 to 0.34 as water volume filtered increased from 30 m3 to 50 m3 (Figure 3).
Differences in detection probabilities among taxa resulted in differences in cumulative detection curves and the number of ichthyoplankton samples needed to detect each taxonomic group. Under the average water volume filtered (30.12 m3), bigheaded carp required the most ichthyoplankton samples (25 backwater or 14 thalweg/channel border) to reach a 90% detection probability, 1.4 to 5 times as many samples as native taxa (Figure 4). Freshwater drum required 5 to 10 samples and shad/percids required 8 samples to reach a 90% probability of detection. By increasing the water volume filtered to 40 m3 we observed a 60% to 108% decrease in the samples required to achieve a 90% detection probability compared to the average sample volume (9 to 15 samples for bigheaded carp, 4 to 8 samples for freshwater drum, and 7 samples for shad and percids (Figure 4).