Evaluating the effect of survey technique, time and space on model robustness
4.4.1 Comparing visual and acoustic predictive models
The rates of acoustic and visual detections were significantly different. Over twice the number of harbour porpoise pods were detected acoustically than visually in those segments which had both visual and acoustic effort with sea state ≤ 3. When data from higher sea states were included, the ratio of acoustic to visual harbour porpoise detections increased significantly, by 8 times or more. This is consistent with other studies on similar platforms, for example, 8 times as many harbour porpoise pods were detected acoustically than visually in surveys carried out in the Baltic Sea from the IFAW research motor-sailor Song of the Whale (Gillespie et al. 2005). In my analysis, the reduced visual sightings rate was mainly explained by the effect of sea state on porpoise sightings. In some years, sea state explained as much as 18.2% of the deviance (even when restricted just to sea state ≤ 3), and similar to other studies (Palka 1996), this was mainly due to the rapid drop in sightings rates above sea state 1, with few sightings in sea states > 2.
Although to a much lesser degree (maximum explained deviance of 5.9%), high frequency noise levels significantly affected the acoustic detection rates of harbour porpoises. Detection rates of harbour porpoises decreased with increasing noise levels. This reduction in acoustic detection rate with ambient noise levels is similar to that found for other studies of different cetacean species (Gordon et al. 2000). In my study area, the main factors influencing high frequency ambient noise levels included depth, sediment type, tidal state and current speeds, the engine being on or off, and boat speed. Ambient noise levels were shown to increase when the engine was on and with increasing speed, similar to that found by Erbe (2002). The increase in noise levels in shallower water and gravel substrate are likely to result from the reflective properties in these habitats (Urick 1983). Sound, such as that produced by the survey vessel, will be absorbed differently by different substrate types: mud absorbs sound energy so reflects less sound than sand or gravel substrate (Urick 1983). Sound also reverberates in shallow water, but distributed and absorbed over a wider volume in deeper water (Urick 1983). So any noise produced by the vessel either from the engine or flow noise through the water, is likely to be reflected and amplified within shallow, gravelly habitats.
Similarly, noise generated by the environment will also be a source of ambient noise (Urick 1983). The amount of gravel in the sediment was correlated with tidal current: areas with high tidal currents had more gravel in the substrate. Tidal state and speed may therefore have the effect of moving substrate, or creating more wave action, creating a different source of noise in the water column. Certainly Urick (1983) suggests that noise levels are generally higher in areas of tidal currents. Biological sources of high frequency noise include snapping shrimp, which was suggested by Johnson et al. (1947) to be the main source of high frequency ambient noise in shelf waters. Snapping shrimp are crustaceans that create a very loud broad band click (2- 200 kHz) when they snap their claws in order to scare off predators or stun their prey (Au & Banks 1998). They are bottom living species that tend to prefer gravely substrates (Nadia et al. 2005), or habitats with crevices or holes in which the shrimp can burrow. There is some suggestion that they also prefer areas of low current and shallower water (< 60 m), even if their preferred bottom habitat is available in deeper water (Johnson et al. 1947). Since very loud broadband clicks thought to originate from snapping shrimp were heard and detected frequently throughout the survey area, it is possible that the varying levels of high frequency ambient noise in differing habitats may also reflect the habitat preferences of snapping shrimp.
Modelled noise levels showed that the locations at which there were high predicted noise levels often had visual detections not accompanied by acoustic detections. Higher ambient noise levels are likely to mask quieter or more distant harbour porpoise vocalisations, thus reducing the detection range of the hydrophone
(Richardson et al. 1995). Also, if the habitat is acting to amplify vessel noise (e.g. in shallow areas with gravel substrate) then porpoises may respond to the vessel by moving away (Palka & Hammond 2001). Since visual detections of porpoises are made ahead of the vessel but acoustic detections behind the vessel, any movement away from the vessel would result in porpoises that are detected visually not being detected acoustically. The combination of both reduced detection range, and movement of porpoises away from the vessel in environments that amplify vessel noise may explain this higher detection visually than acoustically in noisy habitats. The inclusion of sea state and noise levels within the models should compensate for the variation in detection rates due to each variable. However, despite compensating
for all measured survey effects, different environmental variables described harbour porpoise distribution when based on visual survey data than when based on acoustic survey data. For example, position in the spring-neaps tidal cycle was an important predictor of visually detected harbour porpoise distributions, but excluded from the acoustic models. In this case, porpoises were visually detected at a higher rate during higher tides (i.e. closer to spring tide). This suggests that porpoises may be more visible at this time, perhaps due to more vigorous foraging activity creating stronger visual cues. Certainly, evidence suggests that prey tend to be more aggregated during spring tides and more dispersed during neaps tides (Irons 1998). This possible change in behaviour of porpoises over the tidal cycle appears to be more evident visually than acoustically, though it would be interesting to analyse foraging sounds (rapid clicks or ‘buzzes’) to see if these are recorded more often in certain areas or certain tidal states. However, the most significant variables in both models (maximum tidal current and percentage mud in the sediment) were correlated, with higher mud in the sediment of areas with low tidal currents. Also examining the predictions based on the visual and acoustic models, they were very similar for the core high predicted density areas despite the models being very different. The high density areas were predicted in both models to be the more coastal areas of the Sound of Jura, Firth of Lorne, around the Treshnish Isles and the Small Isles.
Despite the differences in data collection methodologies, and in detection rates, the models were able to reliably predict high-use areas for harbour porpoises in the southern Inner Hebrides. However, the results of the modelling suggest exercising caution when combining data sets, especially without first compensating for survey effects in the visual survey data and the acoustic survey data separately before combining data sets (unlike Bailey 2006 & Gridley 2005).