SEARCH

SEARCH BY CITATION

Keywords:

  • echolocation;
  • feeding buzz;
  • GEEs;
  • habitat preference;
  • modelling;
  • multi-scale;
  • PODs;
  • Tursiops

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. Data accessibility
  10. References
  11. Supporting Information
  1. Understanding which environmental factors drive foraging preferences is critical for the development of effective management measures, but resource use patterns may emerge from processes that occur at different spatial and temporal scales. Direct observations of foraging are also especially challenging in marine predators, but passive acoustic techniques provide opportunities to study the behaviour of echolocating species over a range of scales.
  2. We used an extensive passive acoustic data set to investigate the distribution and temporal dynamics of foraging in bottlenose dolphins using the Moray Firth (Scotland, UK). Echolocation buzzes were identified with a mixture model of detected echolocation inter-click intervals and used as a proxy of foraging activity. A robust modelling approach accounting for autocorrelation in the data was then used to evaluate which environmental factors were associated with the observed dynamics at two different spatial and temporal scales.
  3. At a broad scale, foraging varied seasonally and was also affected by seabed slope and shelf-sea fronts. At a finer scale, we identified variation in seasonal use and local interactions with tidal processes. Foraging was best predicted at a daily scale, accounting for site specificity in the shape of the estimated relationships.
  4. This study demonstrates how passive acoustic data can be used to understand foraging ecology in echolocating species and provides a robust analytical procedure for describing spatio-temporal patterns. Associations between foraging and environmental characteristics varied according to spatial and temporal scale, highlighting the need for a multi-scale approach. Our results indicate that dolphins respond to coarser scale temporal dynamics, but have a detailed understanding of finer-scale spatial distribution of resources.

Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. Data accessibility
  10. References
  11. Supporting Information

The environment in which animals move is often heterogeneous, and resources are patchily distributed (Fauchald 1999). Foraging activity is therefore expected to be concentrated in specific areas or times, where individuals can maximize their energy intake (Sydeman et al. 2006; Anderson et al. 2010). Knowing where and when animals preferentially forage and which environmental factors drive these preferences can improve our understanding of processes that affect habitat use and movements (Guisan & Zimmermann 2000; Redfern et al. 2006), informing conservation and management efforts (Markowitz et al. 2004; Bailey & Thompson 2009).

Disruption of behaviour by human activities can have consequences at a population level (New et al. 2013). Disrupted activities (Duchesne, Côté & Barrette 2000; Lusseau 2003) may jeopardize energy balance and the health of the individuals involved, potentially mediating changes in population dynamics through effects on vital rates (Kight & Swaddle 2007; New et al. 2013). Effective policy and conservation measures are required to manage sublethal effects of disturbance, particularly during vulnerable life stages (New et al. 2013), such as energy acquisition through foraging (Lusseau 2003).

There is often uncertainty about the scale at which environmental features may be associated with the occurrence of foraging activity, both in space and time (Pinaud & Weimerskirch 2007; Bailleul et al. 2008). Animals' perception of their dynamic environment is largely unknown, and prey are likely to be distributed in hierarchical patches (Fauchald 1999; Redfern et al. 2006). Studying foraging distribution at the wrong resolution can lead to unreliable predictions (Doniol-Valcroze et al. 2007), to the underestimation or overestimation of the quantified relationships with the environment (Johnson et al. 2002; Ciarniello et al. 2007) and inappropriate protection measures (Yorio 2009). A multi-scale approach is generally preferable to avoid such problems (Levin 1992; Mehlum et al. 1999; de Knegt et al. 2011).

Observing predator foraging activity is challenging, particularly in wide-ranging marine predators (Redfern et al. 2006). It can be even more difficult to evaluate the distribution of prey resources (Torres, Read & Halpin 2008). To overcome this, it is generally assumed that areas of greatest usage by predators reflect higher quality habitat (Bestley et al. 2008; Bailey & Thompson 2010). Telemetry devices directly sample the underwater behaviour of marine species, providing an indication of foraging activity (e.g. Jonsen, Flemming & Myers 2005) and feeding success (e.g. Bestley et al. 2008), but their deployment is limited by cost, logistics and ethical issues. Passive acoustic techniques offer effective alternatives for sampling the behaviour of highly vocal species such as odontocetes (Van Parijs et al. 2009; Marques et al. 2013).

The present study focuses on a small isolated population of approximately 195 (95% density interval: 162–253) bottlenose dolphins (Tursiops truncatus), which ranges over the eastern coast of Scotland (Cheney et al. 2013). Most individuals spend at least part of their lives in the inner Moray Firth (Fig. 1), which was designated as a Special Area of Conservation (SAC), under the European Habitats Directive (92/43/EEC). The Moray Firth also supports a complex mix of traditional and emerging marine industries, many of which have the potential to cause disturbance to foraging dolphins (e.g. Bailey et al. 2010b). Previous visual studies in this area indicate that these dolphins forage in discrete patches in their habitat (Hastie et al. 2004) and, at least in some of these patches, seem to use tidal fronts to aid in prey capture (Bailey & Thompson 2010).

image

Figure 1. Map of the study area and locations of the POD deployments. The four main sampling sites are circled. Contains Ordnance Survey data © Crown copyright and database right 2011.

Download figure to PowerPoint

Here, we use a long-term data set from passive acoustic monitoring devices deployed across the Moray Firth to infer patterns of foraging activity in time and space, and to determine scale dependency in dolphin responses to the environment.

Materials and methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. Data accessibility
  10. References
  11. Supporting Information

Inferring Foraging Activity in Echolocating Species

Autonomous acoustic data loggers have been increasingly used over the past decade to monitor odontocete occurrence from detections of echolocation clicks (e.g. Carlström 2005; Philpott et al. 2007; Bailey et al. 2010a; Leeney, Carslake & Elwen 2011). Richardson et al. (1995) categorized bouts of echolocation clicks (or click trains) into two general types: (i) regular orientation clicks with longer inter-click intervals (i.e. the interval between two consecutive click detections, hereafter ‘ICI’), used to scan the environment and (ii) discrimination clicks, with shorter and decreasing ICIs, used to obtain detailed information as they locate and attempt to capture prey (Thomas, Moss & Vater 2004). Trains of discrimination clicks (or feeding buzzes) are associated with foraging in many dolphin, porpoise, beaked whale and bat species (Miller, Johnson & Tyack 2004; Thomas, Moss & Vater 2004; Carlström 2005; Madsen et al. 2005; Leeney, Carslake & Elwen 2011). If feeding buzzes can be recognized using their ICI characteristics, the large data sets available from passive acoustic monitoring studies could be used to infer patterns of foraging activity across a range of scales.

Data Collection

T-PODs (www.chelonia.co.uk) are autonomous loggers that consist of a hydrophone, an analogue processor and a digital timing/logging system. The devices scan six programmable channels sequentially and, with sufficient difference in detection between two band-pass filters, a click is logged to a resolution of 10 μs. We used a target frequency of 50 kHz and reference frequency of 70 kHz for all deployments, and only considered click trains that were classified by version 8.24 of the dedicated T-POD analysis software as having a high likelihood of being produced by a cetacean (‘Cet-Hi'and ‘Cet-Lo’). Further details can be found in Bailey et al. 2010a.

C-PODs (Chelonia Ltd., Mousehole, Cornwall, UK) are the digital successor of the T-POD and use digital waveform characterization to detect cetacean echolocation clicks. In addition to logging the time of detection, other click features are also recorded and used by the click-train classifier (within version 2.030 of the dedicated C-POD analysis software; Chelonia Ltd, Mousehole, Cornwall, UK) to identify clicks. Clicks were classified on the likelihood of them being generated by cetaceans (‘CetHi’, ‘CetMod’, ‘CetLow’). Further details can be found in Brookes, Bailey & Thompson (In press).

Between 2005 and 2012, PODs were deployed at over 90 locations within the Moray Firth. Location, duration and type of deployment varied according to the objectives of other studies, but data collection covered most of the SAC by 2009. It is currently impossible to distinguish different delphinid species using either T-PODs or C-PODS. We therefore only considered data from 43 locations within the bottlenose dolphins' core range (Fig. 1), where encounters with other delphinids are rare (Reid, Evans & Northridge 2003). The final data set included more than 20 000 days of passive acoustic data (Table 1).

Table 1. Summary of POD deployment information for the data that were used in the present study
YearNo. of sites sampledT-PODC-POD
No. of deploymentsMean duration of deployment (days)No. of deploymentsMean duration of deployment (days)
200527510
2006621690
2007731640
2008113284737
20093954515372
2010234803579
201120031111
201240493

Data Process

For T-POD data, only Cet-Hi or Cet-Lo trains were used in the analysis, as previous work confirmed that these most closely matched visual sightings (Philpott et al. 2007; Bailey et al. 2010a). As no similar assessment exists for C-PODs, a conservative approach was taken and only CetHi and CetMod clicks were included in analyses. The time series of cetacean clicks was exported, and the ICI was then calculated for every logged click. ICIs ranged from 20 μs to > 10 min. These data were multimodal, suggesting that several processes, potentially reflecting different biological functions, could be identified in the ICI time series. Specifically, ICIs could be qualitatively categorized as belonging to one of three groups with distinct peaks. The first peak (hereafter ‘buzz ICIs’) was similar to ICIs associated with foraging in previous studies (Carlström 2005; Madsen et al. 2005). The second peak (hereafter ‘regular ICIs’) corresponded to the expected ICI for regular click trains (Au 1993). The third peak (hereafter ‘long ICIs’) was assumed to correspond to pauses between different click trains.

Previous studies have used an ICI threshold of 10 ms to classify feeding buzzes (Carlström 2005; Leeney, Carslake & Elwen 2011). Here, we fitted Gaussian mixture models to the time series of ICIs to characterize the mixture distribution and assign each ICI to one of the observed processes. Because of the large size of the C-POD data set (Table 1), the analysis was conducted on 12 separate subsets of data, each containing approximately 106 ICIs. The T-POD data set could be analysed altogether. Data were normalized by natural log-transforming the ICIs. The package mixtools within the statistical software R (R Development Core Team 2012) was then used to fit the models via an expectation maximization (EM) algorithm. We specified the number of component distributions k to be equal to three. At times, when the proportion of buzz ICIs to regular ICIs was low, the model would estimate two of the components within the regular ICI range. On these occasions, we investigated whether specifying four components identified a distribution centred on the buzz ICIs and used the Akaike information criterion (AIC) to compare models with three or four components. The results of the mixture models were then used to classify each detected ICI as belonging to one of the three processes.

Data were summarized by hour. For each hour, at each location, we noted whether any ICI and any buzz ICI were detected. As our aim was to explore buzzing activity, we considered only those hourly records where at least one ICI had been detected. Year, Julian date and hour of the record, the POD type (T- or C-) and location were also retained for each hourly record.

Environmental Data

Environmental covariates were extracted for each POD location. Depth was extracted from Marine Digimap, at a resolution of 500 m (SeaZone Hydrospatial Bathymetry). Slope was computed using the library SDMTools in R. Hourly tidal level, speed and direction were obtained using POLPRED (NERC National Oceanography Centre, Liverpool, UK). Sea surface temperature (SST) was taken from NASA's AVHRR Pathfinder data set, and surface chlorophyll-a (Chl-a) from the MODIS sensor on the NASA's Aqua satellite. These remote sensing data were provided as weekly composite maps (1 km resolution) by NERC Earth Observation Data Acquisition and Analysis Service (NEODAAS, Plymouth, UK), and variability in weekly SST and Chl-a was computed using the library SDMTools. Composite ocean front maps from AVHRR SST data were also provided by NEODAAS and processed to obtain measures of the density of fronts in each 1-km cell (front density), the distance to the nearest front (frontal distance), and whether a cell was on the low- or high-value side of the closest major front (front side) (Miller 2009). Finally, we estimated the strength of tidal mixing for each hourly record using the formula log10(H/U3), where H is the depth and U is the speed of the tidal current (Simpson & Hunter 1974).

Modelling Approach

We modelled the presence/absence of buzz ICIs in each hourly record where at least one ICI was detected using binary generalized additive models (GAM; Wood 2006). An autocorrelation function (ACF) plot was used to visualize the level of autocorrelation in the GAM residuals, and generalized estimating equations (GEEs; Liang & Zeger 1986) were then used to explicitly model this observed autocorrelation (see also Dormann et al. 2007; Pirotta et al. 2011; Bailey et al. 2013). Under this approach, data points are classified into blocks, and a correlation structure for the residuals is specified within block (Liang & Zeger 1986). We used each day of deployment as the blocking variable, and the quasi-likelihood under the model independence criterion (QIC; Pan 2001) to select between competing correlation structures. We checked multicollinearity between explanatory variables using the generalized variance inflation factor (GVIF). Models were fitted using the library geepack in R, together with the library splines to extend the GEE-generalized linear models to GEE-GAMs. An approximated version of the QIC, known as the QICu (Pan 2001), was used for model selection. We used the QICu to compare different forms in which an explanatory variable could be included in the model. Specifically, we tested each predictor as a linear term, as a B-spline with four degrees of freedom (hereafter ‘d.f.’) with one internal knot positioned at the average value of that variable and as a B-spline with 5 d.f. and two knots at the quartiles. Cyclic splines are currently not available in the library splines, so the use of GEEs to correct for autocorrelation was prioritized, and B-splines were used for circular covariates (Julian date, hour and tide direction).

Environmental data were not available for every data point, for example due to cloud cover in remote sensing data and because POLPRED cannot predict tidal information at locations close to the coast. Therefore, we initially ran a series of five submodels to explore how the occurrence of buzz ICIs varied in relation to different classes of covariates, each fitted to different subsets of data. These were (i) temporal covariates (year, Julian date and hour of the day); (ii) bathymetric covariates (depth and slope); (iii) tidal covariates (tidal speed, level and direction); (iv) remotely sensed covariates (SST, Chl-a, SST slope, Chl-a slope); (v) remotely sensed frontal metrics (front density, front distance and front side). We also always included a categorical variable POD type, as we expected T-PODs to be less efficient at detecting clicks than C-PODs. Model selection was carried out on the submodels, and the best form for each covariate was identified. These results were then combined into an overall model, which was fitted to the subset of the data with no missing values for the retained predictors. In the final model, we also included latitude and longitude (to account for any spatial pattern not represented by the available covariates) and the tidal mixing parameter. Once the final model was selected using QICu, repeated Wald's tests were used to assess the significance of the retained covariates (Hardin & Hilbe 2003). No variable was excluded on the basis of its significance, but only significant relationships were plotted.

Goodness-of-fit of the final model was evaluated using a confusion matrix that compared predicted and observed occurrences of buzz ICIs, expressed as the proportion of true and false presences and absences. The probability cut-off for predicting the presence of a buzz was selected using a receiver operating characteristic (ROC) curve (Zweig & Campbell 1993), and the area under the ROC curve (AUC) provided a measure of model performance (Zweig & Campbell 1993). We also conducted the goodness-of-fit test for binary GEEs proposed by Horton et al. (1999). We tested the final model using cross-validation, repeatedly fitting the model 100 times to a randomly selected half of the data set (the training set) and predicting the response in the other half (the validation set). Finally, 33 confirmed predation events were used as an independent validation data set to assess the predictive capabilities of the model. These visual observations of fish captures were recorded during visual boat based surveys in the inner Moray Firth. Given the values of the covariates in those specific locations and times, the final model was used to predict the associated probability of buzzing activity and the ROC cut-off probability to classify them as either presences or absences.

Site Specificity in Foraging Features

The factors affecting buzzing activity may vary in space. For example, tidal processes might be associated with foraging only in some locations (Bailey & Thompson 2010). Therefore, we investigated whether the relationships between the occurrence of buzz ICIs and environmental covariates varied between the four locations that were sampled almost continuously between 2005 and 2012 (Chanonry Point, Sutors, Lossiemouth and Spey Bay; Fig. 1). The fitted model included the interactions between location and the temporal variables (year, Julian date and hour of the day), and interactions between location and tidal covariates (speed, level and direction). We used the same model selection procedure described previously to identify the best form of the relationships and most parsimonious subset of predictors. A separate model was fitted to assess the local relevance of the tidal mixing parameter, as this was collinear with tidal speed.

Temporal Scale of Foraging Activities

The temporal scale of the analysis could also affect the results. To investigate whether coarser temporal patterns of buzzing activity were present, the data were summarized by day instead of by hour. Julian date and year (as proxies of any temporal pattern) together with POD location (as a proxy of any spatial pattern) and POD type were tested to determine their relationship with the daily occurrence of buzzes. GEE-GAMs were used to provide consistency in the analytical approach, and the month of each deployment was used as the blocking variable. Model selection by QIC and QICu was repeated, and the fit of the model was evaluated through the confusion matrix and the AUC value.

Finally, because daily patterns may vary in different locations, the interactions between the temporal covariates (Julian date and year) and POD location were investigated at the four main sampling sites.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. Data accessibility
  10. References
  11. Supporting Information

Detection data were available almost continuously for four sites in the Moray Firth, starting in summer 2005 (Sutors and Lossiemouth) or 2006 (Chanonry Point and Spey Bay) until the beginning of spring 2012 (see Table S1 in Supporting information). The remaining sites were sampled discontinuously (see Table S1, Supporting information). Overall, out of more than 480 000 h of observation, the data set included 31 948 positive hourly records in which at least one bottlenose dolphin ICI was detected. Of these, more than 23 000 were recorded in the four main sites.

In seven of the twelve C-POD files, Gaussian mixture models with three components led to the identification of the expected distributions, with peaks at values corresponding to the natural logarithm of the buzz ICI, regular ICI and long ICI values in minutes (e.g. Fig. 2). In the remaining five files, two overlapping component distributions were found for the regular ICI peak, because of the low proportion of buzz ICIs present in the files. Nevertheless, the models with four components correctly discriminated this additional distribution (see Fig. S1, Supporting information) and had lower AICs when compared with the corresponding models with three components. The estimated component distributions (see Table S2, Supporting information) were then used to determine which ICIs were buzzes. As a result, buzz ICIs were identified in 14 222 of 31 948 hourly records. The average buzz ICI was 2 ms (from T-PODs) and 4 ms (from C-PODs). The discrepancy between the two instruments was accounted for in the modelling procedure by including POD type as a covariate.

image

Figure 2. Results of the Gaussian mixture model for one C-POD inter-click interval (ICI) time series. The first component distribution (from the left) corresponds to the buzz ICIs, the second to the regular ICIs and the third to the long ICIs. The values on the x-axis are the natural logarithm of the ICIs in minutes.

Download figure to PowerPoint

Hourly Data-Full Data Set

We ran a full model with all the available explanatory variables to investigate multicollinearity, but GVIF values generally remained below 2. The GVIF was > 3 for tide speed and the tidal mixing parameter, as the calculation of mixing was based upon tide speed. Therefore, the two variables were evaluated in separate models. The model residuals were temporally autocorrelated, although less than expected (see Fig. S2, Supporting information).

For each of the five submodels (including temporal, bathymetric, tidal, oceanographic and front covariates), we compared the QIC value when it was fitted with an autocorrelation structure of order 1 (AR1) to the QIC of a working independence model. In all cases, the difference in QIC was < 2, indicating that the two versions of each submodel were substantially equivalent. This was confirmed by a qualitative comparison of the estimated model coefficients, which were virtually unaffected by the correlation structure chosen. For parsimony, we used a working independence model, as recommended when unsure about the true nature of autocorrelation (Pan 2001).

The QICu model selection procedure retained hour, year and the Julian date in the temporal submodel, depth and slope in the bathymetric submodel, tide level, tide speed and tide direction in the tidal model, SST in the oceanographic model, and front density and frontal distance in the front model. In all submodels, POD type was also identified as a relevant predictor. The five submodels were then combined into an overall model (running on a reduced data set) that also included latitude and longitude. Model selection with QICu was carried out on the overall model and, as a result, longitude, hour, year, depth, SST, tide level, tide speed and tide direction were excluded. As tide speed was removed by this second round of model selection, the tidal mixing parameter was also included, but this did not reduce the QICu. Finally, despite being retained by the QICu selection, frontal distance (= 0·14) was not significant under the Wald's test. Conversely, POD type, latitude, Julian date, slope and front density were all significantly associated with the occurrence of buzzes (Fig. 3). The probability of buzz occurrence was lower in winter (January–April), but increased in spring and throughout summer and autumn (Fig. 3c). The use of a B-spline (instead of a cyclic spline) meant that the two extremes of the estimated relationship did not smoothly join, but we still obtained an approximate circular pattern (Fig. 3c). Greater seabed slope was associated with a higher occurrence of buzz ICIs (Fig. 3d), and there was a positive relationship between buzzes and front density (Fig. 3e).

image

Figure 3. Results of the model for the presence/absence of buzzing activity in each POD hourly record. The model was a binary generalized additive model with generalized estimating equations. (a) Estimated relationship between the response variable and the POD type. (b) Estimated relationship with the latitude. (c) Estimated relationship with the Julian date. (d) Estimated relationship with the slope gradient. (e) Estimated relationship with the front density. The shaded areas represent the 95% confidence intervals. The rug plot shows the actual data values.

Download figure to PowerPoint

On average, the final model correctly classified 57·1% of observations, and the value of the area under the ROC curve was 0·59. Horton et al. (1999) goodness-of-fit test provided no evidence of lack of fit (Wald's test statistic = 4·35; d.f. = 9; = 0·88), and the cross-validation confirmed that the model was robust. The shape of the estimated relationships (averaged across the 100 models) was consistent (see Fig. S3, Supporting information), and the model's average performance at classifying the validation sets was 56·6%, similar to the model fitted to the full data set. Moreover, the model successfully classified 77% of the 33 observed predation events as ‘presences’. The final model was then compared with a model in which the spatial covariates (slope and latitude) were replaced by POD location, to investigate the possibility that large-scale spatial mechanisms explaining the distribution of buzzing activity had been missed. The QICu suggested that the model with POD location was slightly better than the model with the spatial covariates, but its performance only increased to 58% of correct classifications.

Finally, to visualize the results, the final model was used to predict the probability of detecting buzz ICIs in each POD location. A randomly selected Julian date was used (250), together with the corresponding front maps in the average year (2009). To represent the relative importance of different areas for buzzing activity, predictions were weighted by the differential usage of those locations by dolphins. This was computed as the proportion of positive hours (hours during which at least one bottlenose dolphin ICI was detected) over the total number of hours covered by POD deployments in each location (Fig. 4; see also Fig. S4, Supporting information).

image

Figure 4. Map of the final model predictions in each POD location, weighted by the site usage. The probability of occurrence of buzzing activity is predicted as a function of the model covariates in that location. A randomly selected Julian date was used (250), together with the corresponding front maps in the average year (2009). Site usage was estimated as the proportion of positive hours over the total number of hours covered by POD deployments in each location.

Download figure to PowerPoint

Hourly Data – Four Main Sites

The QIC suggested that a working independence model was preferable over the specification of an AR1 autocorrelation structure. The QICu model selection procedure retained year and Julian date, hour and tide direction, tide speed and level, and the interactions between these variables and POD location. POD type was also found to be relevant. Estimated relationships were all significant under the robust Wald's tests, but the goodness-of-fit of the final models remained virtually unchanged (57·5% of correct classifications; AUC = 0·61). Because of its collinearity with tide speed, the tidal mixing parameter was evaluated in a separate model, and it was retained by model selection as a significant linear term with an interaction with POD location.

Plots for the estimated relationships show two different patterns: (i) some variables showed a markedly different relationship with the response for each location. For example, the seasonal patterns of buzzing activity (see Fig. S5, Supporting information) or the role of the tidal flow direction (Fig. 5). (ii) other variables appeared to have an effect on the occurrence of buzz ICIs only in some locations. Tidal speed and the tidal mixing parameter, for example, were found to have an effect only at Chanonry Point and, to a much lesser extent, at the Sutors. In these locations, faster tidal flows corresponded to increased occurrence of buzzes, and well-mixed waters were preferred over stratified ones (Fig. 6).

image

Figure 5. Estimated relationship between buz-zing activity and tide direction in each of the four main sites. The binary generalized additive model with generalized estimating equations included the interactions between the covariates and the POD location. The shaded areas represent the 95% confidence intervals. The rug plot shows the actual data values.

Download figure to PowerPoint

image

Figure 6. Estimated relationship between buz-zing activity and the tidal mixing parameter in each of the four main sites. The binary generalized additive model with generalized estimating equations included the interactions between the covariates and the POD location. The shaded areas represent the 95% confidence intervals. The rug plot shows the actual data values.

Download figure to PowerPoint

Daily Data

The analysis of the presence or absence of buzzing activity at a daily scale suggested that Julian date and year, POD type and POD location were all significantly associated with the response (Fig. 7). The estimated seasonal variability in buzzing (with greatest probability of buzz occurrence in summer and autumn; Fig. 7b) confirmed the results of the hourly model, and the probability of detecting buzz ICIs peaked in 2005 and 2010. The goodness-of-fit of the final model was substantially improved, correctly classifying 65% of observations with an AUC of 0·72. The model performed even better when interactions between temporal covariates and POD location were introduced. This model was fitted to data from the four main sites and included Julian date and year, their interactions with POD location and POD type. The inclusion of the interaction terms led to an average of 68·9% correct classifications and an AUC of 0·75.

image

Figure 7. Results of the model for the presence/absence of buzzing activity in each POD daily record. The model was a binary generalized additive model with generalized estimating equations. (a) Estimated relationship between the response variable and the POD type. (b) Estimated relationship with the Julian date. (c) Estimated relationship with the year. (d) Estimated relationship with the POD location. The shaded areas represent the 95% confidence intervals.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. Data accessibility
  10. References
  11. Supporting Information

A Model for the Foraging Activity of an Echolocating Marine Predator

Our findings demonstrate that this extensive passive acoustic data set could be used to investigate the distribution and temporal dynamics of echolocation buzzing activity in a marine predator. Foraging buzzes were identified within the time series of echolocation clicks. Subsequent modelling of spatio-temporal dynamics indicated that both static (latitude, slope) and dynamic environmental variables (Julian date, frontal metrics) were significantly related to buzzing activity.

Observed seasonal patterns of buzzing reflected the seasonal trends in Atlantic salmon (Salmo salar) abundance in the study area, as adult salmon migration starts in spring and continues until late autumn (Marine Scotland Science, unpublished data; available at http://www.scotland.gov.uk/Topics/marine/science/Publications/stats/SalmonSeaTroutCatches). Salmon are an important prey resource for these predators (Janik 2000; Bailey & Thompson 2010), and salmon runs into local rivers are likely to affect dolphin movements and foraging (Hastie et al. 2004). Our results also confirm that bottom topography influences bottlenose dolphin foraging efficiency (see Hastie et al. 2004). Steep and narrow channels may act as bottlenecks for prey, and currents resulting from interactions between tidal flows and steep slopes may concentrate prey and facilitate capture. Finally, model selection retained the relationship between buzzing occurrence and fronts. Several studies have underlined the importance of frontal systems for marine top predators (Doniol-Valcroze et al. 2007; Bost et al. 2009), where enhanced productivity and biomass can affect the distribution of fish prey. Physical aggregation of prey through convergence processes can also lead to the formation of predictable prey patches (Worm et al. 2005; Bost et al. 2009).

Predictions from the final model provided insights into the distribution of dolphin foraging activity. In particular, the model confirmed the importance of the Sutors and Chanonry Point as foraging hotspots in the inner Moray Firth (Hastie et al. 2004; Bailey & Thompson 2010) and identified other important patches (Lossiemouth and Cullen Bay) that can guide future research and management effort.

The model correctly predicted 77% of the visual observations of confirmed feeding events, but the confusion matrix indicates that some variability in the data cannot be explained by the variables considered here. These environmental and temporal variables act as indirect proxies of unknown mechanistic processes (mainly the availability of prey) that influence foraging activity, and are therefore imperfect predictors (Guisan & Zimmermann 2000; Redfern et al. 2006). Moreover, none of the covariates in the final model had an hourly resolution, so fine-scale temporal variability in buzzing activity remained unexplained. When POD location was included as a substitute for the retained spatial covariates (latitude, slope), the fit improved only slightly, confirming that the final model was probably missing a dynamic component, rather than additional spatial information.

Scale Dependency in Foraging

The spatial scale of analysis is known to affect the quantification of relationships between species and their environment (Ciarniello et al. 2007; de Knegt et al. 2011). This is especially true in marine systems, where resources are heterogeneously distributed and habitats have a hierarchical structure (Mehlum et al. 1999; Pinaud & Weimerskirch 2007; Bailleul et al. 2008). When the shape of the estimated relationships between buzzing occurrence and environmental covariates was allowed to vary between the four main sites, the modelling results supported this hypothesis. The overall model identified a general temporal variation in buzzing activity across the year, but our site-level analysis showed marked differences between the four sites. Different local features (e.g. variability in prey densities) can influence seasonal patterns of usage (Hastie et al. 2004). Tidal dynamics also influenced foraging behaviour differently across the four sites, as might be expected if site-specific topography leads to local and complex interactions with tidal movements. As a result, relationships with tidal flow direction varied according to different shoreline orientation and morphology. Moreover, favourable foraging opportunities might arise from tides in some areas, but not in others. Strong tidal flows, for example, could aid in prey capture at locations such as Chanonry Point, where flow interacts with particular bathymetric features (Bailey & Thompson 2010). The site specificity of these relationships highlights why these covariates were excluded in the overall model and why previous studies failed to detect similar relationships over large scales (e.g. Hastie et al. 2004).

The temporal scale of analysis also affected the goodness-of-fit of the final model. While marine top predators are often expected to have a detailed knowledge of their environment (Le Boeuf et al. 2000), our predictions of foraging activity were strongest at a daily rather than hourly level. Thus, dolphins appeared to respond to coarser scale temporal dynamics (e.g. the variation of front density or the seasonality of their prey abundance), while demonstrating a finer-scale response to spatial variation in resources. In the best performing model, relationships with environmental covariates were allowed to vary by site and the occurrence of buzzing activity was assessed at a daily scale. This suggests that it is more profitable to concentrate on areas that have, on average, a higher chance of finding food (Iwasa, Higashi & Yamamura 1981) than to develop a detailed understanding of temporal dynamics in prey availability. Alternatively, the system's temporal dynamics may be so uncertain, as reflected by highly unpredictable salmon movements (Klemetsen et al. 2003; Stewart et al. 2009), to make it impossible for the dolphins to predict food availability at finer temporal scales. Finally, the better performance of coarser scale models could also result from our inability to accurately characterize small variations in prey dynamics (Grémillet et al. 2008).

Observed patterns of distribution and resource use emerge from the integration of processes that occur at different spatial and temporal scales (Levin 1992), but these will be limited by the species' locomotion costs. Bottlenose dolphins may be sampling their habitat at a coarse temporal scale because of their ability to shift between dispersed patches at relatively low energetic costs (Williams 1999). While the success of this strategy will vary with fine-scale fluctuations of prey densities, it may offer long-term advantages in a relatively unpredictable ecological landscape. On the other hand, complex social systems, hunting techniques and behaviour can also lead to site-specific foraging specializations in top predators (Bailleul et al. 2008; Bailey & Thompson 2010). Such local adaptations can only emerge from a site-level analysis. Our results further confirm that a multi-scale approach is required to identify the hierarchical patterns of a species' foraging activity (Ciarniello et al. 2007; de Knegt et al. 2011).

Using Passive Acoustic Techniques to Assess Foraging Distribution

This study demonstrates that the extensive passive acoustic monitoring data sets produced over the last decade can provide important insights into the foraging ecology of these vocal wide-ranging predators. While the ICI distributions that discriminated between buzz and regular clicks were similar to those used in previous studies (Carlström 2005; Leeney, Carslake & Elwen 2011), the mixture modelling framework used here provided a less arbitrary tool for classifying the ICIs into separate categories. Nevertheless, there remain limitations in the use of passive acoustic techniques to investigate foraging activity. First, echolocation processes are highly directional and may be missed by fixed detectors such as PODs. This will result in false absences of foraging activity when data are analysed at a fine temporal scale, potentially explaining the poorer predictive power of our hourly models. Detections may also be influenced by local environmental characteristics (e.g. bottom sediment, or tidal flow). While the relatively homogeneous features of the Moray Firth make the data collected over such scales comparable, other studies might be limited by more variable local conditions. However, our analyses only considered the occurrence of buzzes in samples in which echolocations were detected, and overall POD detection abilities should therefore have less influence on modelling results. Finally, high-repetition click trains can also be emitted during social interactions. These are more likely to be missed by the PODs because of their higher directionality and shorter ICIs than feeding buzzes (Herzing 1996), and we expect most detected buzzes to correspond to feeding buzzes.

Conclusions

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. Data accessibility
  10. References
  11. Supporting Information

This robust analytical framework can be used to assess the occurrence and spatio-temporal patterns of foraging in this echolocating marine predator and has potential applications for other species. Foraging was associated with different environmental characteristics at different spatial and temporal scales, confirming that a multi-scale approach is advisable for the investigation of such scale-dependent activity.

Understanding when and where foraging occurs is crucial for the development of effective conservation measures (Markowitz et al. 2004; Bailey & Thompson 2009). In particular, the association of foraging with specific environmental features is required to inform individual-based models used to predict the consequences of exposure to different stressors (New et al. 2013). The identification of the multi-scale processes that influence foraging should also be taken into account when managing anthropogenic disturbances.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. Data accessibility
  10. References
  11. Supporting Information

This work received funding from the MASTS pooling initiative (the Marine Alliance for Science and Technology for Scotland), and their support is gratefully acknowledged. MASTS is funded by the Scottish Funding Council (grant reference HR09011) and contributing institutions. Passive acoustic data were collected during a series of grants and contracts from Scottish Natural Heritage, Scottish Government, the European Union, Department of Energy & Climate Change, COWRIE, Oil & Gas UK and Talisman Energy (UK) Ltd. Satellite data were received and processed by the NERC Earth Observation Data Acquisition and Analysis Service (NEODAAS) at Dundee University and Plymouth Marine Laboratory (www.neodaas.ac.uk). POLPRED model for tidal information was kindly provided by NERC National Oceanography Centre (Liverpool, UK). The authors would also like to thank Dr. Helen Bailey and Ana Candido for their contribution to the collection of some of the data used in the present study. Thanks to Dr. Nick Tregenza for his advice on POD data processing, Dr. Beth Scott for suggesting the use of the tidal mixing parameter, Dr. Steve Palmer for his GIS help and Dr. Marianne Marcoux for invaluable discussions during the early phases of this work. Finally, we thank two anonymous reviewers for their helpful comments.

Data accessibility

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. Data accessibility
  10. References
  11. Supporting Information

The data set used in this paper is available in the Dryad data repository: doi:10.5061/dryad.83 h07 (Pirotta et al. 2013)

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. Data accessibility
  10. References
  11. Supporting Information
  • Anderson, T., Hopcraft, J., Eby, S., Ritchie, M., Grace, J. & Olff, H. (2010) Landscape-scale analyses suggest both nutrient and antipredator advantages to Serengeti herbivore hotspots. Ecology, 91, 15191529.
  • Au, W.W.L. (1993) The Sonar of Dolphins. Springer-Verlag, New York, NY.
  • Bailey, H. & Thompson, P. (2009) Using marine mammal habitat modelling to identify priority conservation zones within a marine protected area. Marine Ecology Progress Series, 378, 279287.
  • Bailey, H. & Thompson, P. (2010) Effect of oceanographic features on fine-scale foraging movements of bottlenose dolphins. Marine Ecology Progress Series, 418, 223233.
  • Bailey, H., Clay, G., Coates, E., Lusseau, D., Senior, B. & Thompson, P.M. (2010a) Using T-PODs to assess variations in the occurrence of coastal bottlenose dolphins and harbour porpoises. Aquatic Conservation: Marine and Freshwater Ecosystems, 20, 150158.
  • Bailey, H., Senior, B., Simmons, D., Rusin, J., Picken, G. & Thompson, P.M. (2010b) Assessing underwater noise levels during pile-driving at an offshore windfarm and its potential effects on marine mammals. Marine Pollution Bulletin, 60, 888897.
  • Bailey, H., Corkrey, R., Cheney, B. & Thompson, P.M. (2013) Analyzing temporally correlated dolphin sightings data using generalized estimating equations. Marine Mammal Science, 29, 123141.
  • Bailleul, F., Pinaud, D., Hindell, M., Charrassin, J.-B. & Guinet, C. (2008) Assessment of scale-dependent foraging behaviour in southern elephant seals incorporating the vertical dimension: a development of the First Passage Time method. Journal of Animal Ecology, 77, 948957.
  • Bestley, S., Patterson, T.A., Hindell, M.A. & Gunn, J.S. (2008) Feeding ecology of wild migratory tunas revealed by archival tag records of visceral warming. Journal of Animal Ecology, 77, 12231233.
  • Bost, C.A., Cotté, C., Bailleul, F., Cherel, Y., Charrassin, J.B., Guinet, C. et al. (2009) The importance of oceanographic fronts to marine birds and mammals of the southern oceans. Journal of Marine Systems, 78, 363376.
  • Brookes, K.L., Bailey, H. & Thompson, P.M. (in press) Predictions from harbor porpoise habitat association models are confirmed by long-term passive acoustic monitoring. Journal of the Acoustical Society of America.
  • Carlström, J. (2005) Diel variation in echolocation behavior of wild harbor porpoises. Marine Mammal Science, 21, 112.
  • Cheney, B., Thompson, P.M., Ingram, S.N., Hammond, P.S., Stevick, P.T., Durban, J.W. et al. (2013) Integrating multiple data sources to assess the distribution and abundance of bottlenose dolphins Tursiops truncatus in Scottish waters. Mammal Review, 43, 7188.
  • Ciarniello, L.M., Boyce, M.S., Seip, D.R. & Heard, D.C. (2007) Grizzly bear habitat selection is scale dependent. Ecological Applications, 17, 14241440.
  • Doniol-Valcroze, T., Berteaux, D., Larouche, P. & Sears, R. (2007) Influence of thermal fronts on habitat selection by four rorqual whale species in the Gulf of St. Lawrence. Marine Ecology Progress Series, 335, 207216.
  • Dormann, C.F., McPherson, J.M., Araujo, M.B., Bivand, R., Bolliger, J., Carl, G. et al. (2007) Methods to account for spatial autocorrelation in the analysis of species distributional data: a review. Ecography, 30, 609628.
  • Duchesne, M., Côté, S.D. & Barrette, C. (2000) Responses of woodland caribou to winter ecotourism in the Charlevoix Biosphere Reserve, Canada. Biological Conservation, 96, 311317.
  • Fauchald, P. (1999) Foraging in a hierarchical patch system. The American Naturalist, 153, 603613.
  • Grémillet, D., Lewis, S., Drapeau, L., van Der Lingen, C.D., Huggett, J.A., Coetzee, J.C. et al. (2008) Spatial match–mismatch in the Benguela upwelling zone: should we expect chlorophyll and sea surface temperature to predict marine predator distributions? Journal of Applied Ecology, 45, 610621.
  • Guisan, A. & Zimmermann, N.E. (2000) Predictive habitat distribution models in ecology. Ecological Modelling, 135, 147186.
  • Hardin, J.W. & Hilbe, J.M. (2003) Generalized Estimating Equations, 3rd edn. Chapman & Hall/CRC Press, London.
  • Hastie, G.D., Wilson, B., Wilson, L.J., Parsons, K.M. & Thompson, P.M. (2004) Functional mechanisms underlying cetacean distribution patterns: hotspots for bottlenose dolphins are linked to foraging. Marine Biology, 144, 397403.
  • Herzing, D.L. (1996) Vocalizations and associated underwater behavior of free-ranging Atlantic spotted dolphins, Stenella frontalis and bottlenose dolphins, Tursiops truncatus. Aquatic Mammals, 22, 6179.
  • Horton, N.J., Bebchuk, J.D., Jones, C.L., Lipsitz, S.R., Catalano, P.J., Zahner, G.E. et al. (1999) Goodness-of-fit for GEE: an example with mental health service utilization. Statistics in Medicine, 18, 213222.
  • Iwasa, Y., Higashi, M. & Yamamura, N. (1981) Prey distribution as a factor determining the choice of optimal foraging strategy. American Naturalist, 117, 710723.
  • Janik, V.M. (2000) Food-related bray calls in wild bottlenose dolphins (Tursiops truncatus). Proceedings of the Royal Society of London. Series B: Biological Sciences, 267, 923927.
  • Johnson, C.J., Parker, K.L., Heard, D.C. & Gillingham, M.P. (2002) Movement parameters of ungulates and scale-specific responses to the environment. Journal of Animal Ecology, 71, 225235.
  • Jonsen, I., Flemming, J. & Myers, R.A. (2005) Robust state-space modeling of animal movement data. Ecology, 86, 28742880.
  • Kight, C.R. & Swaddle, J.P. (2007) Associations of anthropogenic activity and disturbance with fitness metrics of eastern bluebirds (Sialia sialis). Biological Conservation, 138, 189197.
  • Klemetsen, A., Amundsen, P.-A., Dempson, J.B., Jonsson, B.N., Jonsson, B., O'Connell, M. et al. (2003) Atlantic salmon Salmo salar L., brown trout Salmo trutta L. and Arctic charr Salvelinus alpinus (L.): a review of aspects of their life histories. Ecology of Freshwater Fish, 12, 159.
  • de Knegt, H.J., van Langevelde, F., Skidmore, A.K., Delsink, A., Slotow, R., Henley, S. et al. (2011) The spatial scaling of habitat selection by African elephants. Journal of Animal Ecology, 80, 270281.
  • Le Boeuf, B., Crocker, D., Costa, D.P., Blackwell, S.B., Webb, P.M. & Houser, D.S. (2000) Foraging ecology of northern elephant seals. Ecological Applications, 70, 353382.
  • Leeney, R.H., Carslake, D. & Elwen, S.H. (2011) Using static acoustic monitoring to describe echolocation behaviour of Heaviside's dolphins (Cephalorhynchus heavisidii) in Namibia. Aquatic Mammals, 37, 151160.
  • Levin, S. (1992) The problem of pattern and scale in ecology: the Robert H. MacArthur Award lecture. Ecology, 73, 19431967.
  • Liang, K.Y. & Zeger, S.L. (1986) Longitudinal data analysis using generalized linear models. Biometrika, 73, 1322.
  • Lusseau, D. (2003) Effects of tour boats on the behavior of bottlenose dolphins: using Markov chains to model anthropogenic impacts. Conservation Biology, 17, 17851793.
  • Madsen, P.T., Johnson, M., Aguilar Soto, N., Zimmer, W.M.X. & Tyack, P.L. (2005) Biosonar performance of foraging beaked whales (Mesoplodon densirostris). Journal of Experimental Biology, 208, 181194.
  • Markowitz, T.M., Harlin, A.D., Würsig, B. & McFadden, C.J. (2004) Dusky dolphin foraging habitat: overlap with aquaculture in New Zealand. Aquatic Conservation: Marine and Freshwater Ecosystems, 14, 133149.
  • Marques, T.A., Thomas, L., Martin, S.W., Mellinger, D.K., Ward, J.A., Moretti, D.J. et al. (2013) Estimating animal population density using passive acoustics. Biological Reviews, 88, 287309.
  • Mehlum, F., Hunt, G.L. Jr., Klusek, Z. & Decker, M. (1999) Scale-dependent correlations between the abundance of Brünnich's guillemots and their prey. Journal of Animal Ecology, 68, 6072.
  • Miller, P. (2009) Composite front maps for improved visibility of dynamic sea-surface features on cloudy SeaWiFS and AVHRR data. Journal of Marine Systems, 78, 327336.
  • Miller, P.J.O., Johnson, M.P. & Tyack, P.L. (2004) Sperm whale behaviour indicates the use of echolocation click buzzes ‘creaks’ in prey capture. Proceedings of the Royal Society of London. Series B: Biological Sciences, 271, 22392247.
  • New, L.F., Harwood, J., Thomas, L., Donovan, C., Clark, J.S., Hastie, G. et al. (2013) Modeling the biological significance of behavioral change in coastal bottlenose dolphins in response to disturbance. Functional Ecology, 27, 314322.
  • Pan, W. (2001) Akaike's information criterion in generalized estimating equations. Biometrics, 57, 120125.
  • Philpott, E., Englund, A., Ingram, S.N. & Rogan, E. (2007) Using T-PODs to investigate the echolocation of coastal bottlenose dolphins. Journal of the Marine Biological Association of the UK, 87, 11.
  • Pinaud, D. & Weimerskirch, H. (2007) At-sea distribution and scale-dependent foraging behaviour of petrels and albatrosses: a comparative study. Journal of Animal Ecology, 76, 919.
  • Pirotta, E., Matthiopoulos, J., MacKenzie, M., Scott-Hayward, L. & Rendell, L. (2011) Modelling sperm whale habitat preference: a novel approach combining transect and follow data. Marine Ecology Progress Series, 436, 257272.
  • Pirotta, E., Thompson, P.M., Miller, P.I., Brookes, K.L., Cheney, B., Barton, T.R. et al. (2013) Data from: Scale-dependent foraging ecology of a marine top predator modelled using passive acoustic data. Dryad Digital Repository. http://dx.doi.org/10.5061//dryad.83h07
  • R Development Core Team. (2012) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL http://www.R-project.org/.
  • Redfern, J.V., Ferguson, M.C., Becker, E.A., Hyrenbach, K.D., Good, C., Barlow, J. et al. (2006) Techniques for cetacean-habitat modeling. Marine Ecology Progress Series, 310, 271295.
  • Reid, J.B., Evans, P.G.H. & Northridge, S.P. (2003) Atlas of Cetacean Distribution in North-west European Waters. Joint Nature Conservation Committee, Peterborough.
  • Richardson, W.J., Greene, C.R.J., Malme, C.I. & Thomson, D.H. (1995) Marine Mammals and Noise. Academic Press, Inc, San Diego, CA.
  • Simpson, J.H. & Hunter, J.R. (1974) Fronts in the Irish Sea. Nature, 250, 404406.
  • Stewart, D.C., Middlemas, S.J., MacKay, S. & Armstrong, J.D. (2009) Over-summering behaviour of Atlantic salmon Salmo salar returning to rivers in the Cromarty Firth, north-east Scotland. Journal of Fish Biology, 74, 13471352.
  • Sydeman, W.J., Brodeur, R.D., Grimes, C.B., Bychkov, A.S. & McKinnell, S. (2006) Marine habitat ‘hotspots’ and their use by migratory species and top predators in the North Pacific Ocean: introduction. Deep Sea Research Part II: Topical Studies in Oceanography, 53, 247249.
  • Thomas, J., Moss, C. & Vater, M. (2004) Echolocation in Bats and Dolphins. University of Chicago Press, Chicago, IL.
  • Torres, L.G., Read, A.J. & Halpin, P. (2008) Fine-scale habitat modeling of a top marine predator: do prey data improve predictive capacity? Ecological Applications, 18, 17021717.
  • Van Parijs, S., Clark, C., Sousa-Lima, R., Parks, S., Rankin, S., Risch, D. et al. (2009) Management and research applications of real-time and archival passive acoustic sensors over varying temporal and spatial scales. Marine Ecology Progress Series, 395, 2136.
  • Williams, T.M. (1999) The evolution of cost efficient swimming in marine mammals: limits to energetic optimization. Philosophical Transactions of the Royal Society B: Biological Sciences, 354, 193201.
  • Wood, S.N. (2006) Generalized Additive Models, an Introduction with R. Chapman & Hall/CRC, London.
  • Worm, B., Sandow, M., Oschlies, A., Lotze, H.K. & Myers, R.A. (2005) Global patterns of predator diversity in the open oceans. Science, 309, 13651369.
  • Yorio, P. (2009) Marine protected areas, spatial scales, and governance: implications for the conservation of breeding seabirds. Conservation Letters, 2, 171178.
  • Zweig, M.H. & Campbell, G. (1993) Receiver-operating characteristic (ROC) plots: a fundamental evaluation tool in clinical medicine. Clinical Chemistry, 39, 561577.

Supporting Information

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. Data accessibility
  10. References
  11. Supporting Information
FilenameFormatSizeDescription
fec12146-sup-0001-LaySummary.pdfapplication/PDF201KLay Summary
fec12146-sup-0002-FigureS1.pdfapplication/PDF190KFig. S1. Results of the Gaussian mixture model for one C-POD inter-click interval (ICI) time series (four component distributions).
fec12146-sup-0003-FigureS2.pdfapplication/PDF115KFig. S2. Autocorrelation function plot for the residuals of a binary generalized additive model (GAM).
fec12146-sup-0004-FigureS3.pdfapplication/PDF355KFig. S3. Results of the cross-validation procedure.
fec12146-sup-0005-FigureS4.pdfapplication/PDF278KFig. S4. Map of the final model predictions (non-weighted) in each POD location.
fec12146-sup-0006-FigureS5.pdfapplication/PDF276KFig. S5. Estimated relationship between buzzing activity and Julian date in each of the four main sites.
fec12146-sup-0007-TableS1.pdfapplication/PDF28KTable S1. Geographical coordinates and sampling periods of the PODs deployed in the Moray Firth (2005–2012).
fec12146-sup-0008-TableS2.pdfapplication/PDF19KTable S2. Estimated mean and standard deviations for the three processes identified in the inter-click interval (ICI) time series.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.