Mathieu Woillez

Multifrequency species classification of acoustic-trawl survey data using semi-supervised learning with class discovery

Acoustic surveys often use multifrequency backscatter to estimate fish and plankton abundance. Direct samples are used to validate species classification of acoustic backscatter, but samples may be sparse or unavailable. A generalized Gaussian mixture model was developed to classify multifrequency acoustic backscatter when not all species classes are known. The classification, based on semi-supervised learning with class discovery, was applied to data collected in the eastern Bering Sea during summers 2004, 2007, and 2008. Walleye pollock, euphausiids, and two other major classes occurring in the upper water column were identified.

Interannual Changes in Biomass Affect the Spatial Aggregations of Anchovy and Sardine as Evidenced by Geostatistical and Spatial Indicators

Geostatistical techniques were applied and a series of spatial indicators were calculated (occupation, aggregation, location, dispersion, spatial autocorrelation and overlap) to characterize the spatial distributions of European anchovy and sardine during summer. Two ecosystems were compared for this purpose, both located in the Mediterranean Sea: the Strait of Sicily (upwelling area) and the North Aegean Sea (continental shelf area, influenced by freshwater). Although the biomass of anchovy and sardine presented high interannual variability in both areas, the location of the centres of gravity and the main spatial patches of their populations were very similar between years. The size of the patches representing the dominant part of the abundance (80%) was mostly ecosystem- and species-specific. Occupation (area of presence) appears to be shaped by the extent of suitable habitats in each ecosystem whereas aggregation patterns (how the populations are distributed within the area of presence) were species-specific and related to levels of population biomass. In the upwelling area, both species showed consistently higher occupation values compared to the continental shelf area. Certain characteristics of the spatial distribution of sardine (e.g. spreading area, overlapping with anchovy) differed substantially between the two ecosystems. Principal component analysis of geostatistical and spatial indicators revealed that biomass was significantly related to a suite of, rather than single, spatial indicators. At the spatial scale of our study, strong correlations emerged between biomass and the first principal component axis with highly positive loadings for occupation, aggregation and patchiness, independently of species and ecosystem. Overlapping between anchovy and sardine increased with the increase of sardine biomass but decreased with the increase of anchovy. This contrasting pattern was attributed to the location of the respective major patches combined with the specific occupation patterns of the two species. The potential use of spatial indices as auxiliary stock monitoring indicators is discussed.

A Geostatistical Definition of Hotspots for Fish Spatial Distributions

Research surveys at sea are undertaken yearly to monitor the distribution and abundance of fish stocks. In the survey data, a small number of high fish concentration values are often encountered, which denote hotspots of interest. But statistically, they are responsible for important uncertainty in the estimation. Thus understanding their spatial predictability given their surroundings is expected to reduce such uncertainty. Indicator variograms and cross-variograms allow to understand the spatial relationship between values above a cutoff and the rest of the distribution under that cutoff. Using these tools, a “top” cutoff can be evidenced above which values are spatially uncorrelated with their lower surroundings. Spatially, the geometric set corresponding to the top cutoff corresponds to biological hotspots, inside which high concentrations are contained. The hotspot areas were mapped using a multivariate kriging model, considering indicators in different years as covariates. The case study considered here is the series of acoustic surveys Pelgas performed in the Bay of Biscay to estimate anchovy and other pelagic fish species. The data represent tonnes of fish by square nautical mile along transects regularly spaced. Top cutoffs were estimated in each year. The areas of such anchovy hotspots are then mapped by co-kriging using all information across the time series. The geostatistical tools were adapted for estimating hotspot habitat maps and their variability, which are key information for the spatial management of fish stocks. Tools used here are generic and will apply in many engineering fields where predicting high concentration values spatially is of interest.

Indicator-Based Geostatistical Models For Mapping Fish Survey Data

Marine research survey data on fish stocks often show a small proportion of very high-density values, as for many environmental data. This makes the estimation of second-order statistics, such as the variance and the variogram, non-robust. The high fish density values are generated by fish aggregative behaviour, which may vary greatly at small scale in time and space. The high values are thus imprecisely known, both in their spatial occurrence and order of magnitude. To map such data, three indicator-based geostatistical methods were considered, the top-cut model, min–max autocorrelation factors (MAF) of indicators, and multiple indicator kriging. In the top-cut and MAF approaches, the variable is decomposed into components and the most continuous ones (those corresponding to the low and medium values) are used to guide the mapping. The methods are proposed as alternatives to ordinary kriging when the variogram is difficult to estimate. The methods are detailed and applied on a spatial data set of anchovy densities derived from a typical fish stock acoustic survey performed in the Bay of Biscay, which show a few high-density values distributed in small spatial patches and also as solitary events. The model performances are analyzed by cross-validating the data and comparing the kriged maps. Results are compared to ordinary kriging as a base case. The top-cut model had the best cross-validation performance. The indicator-based models allowed mapping high-value areas with small spatial extent, in contrast to ordinary kriging. Practical guidelines for implementing the indicator-based methods are provided.

New insights into behavioural ecology of European seabass off the West Coast of France: implications at local and population scales

&NA; From 2010 to 2012, 246 data storage tags were deployed on European seabass in the Iroise Natural Marine Park, a marine protected area (MPA) off west Brittany, France. A return rate of 14.6% associated with long time series of data provided new information on fish ecology (e.g. maximum experienced depth greater than 225 m, temperature range 6.80–21.87°C). Depth and temperature series were used to infer individual migration using an innovative hidden Markov model (HMM) especially developed for seabass geolocation. Reconstructed fish tracks revealed that seabass is a partially migratory species, as individuals exhibited either long‐distance migrations towards the Bay of Biscay or the Celtic Sea, or residency behaviour in the Iroise Sea. Fidelity to summer feeding areas and to winter spawing areas was demonstrated. These results suggest that the population is spatially structured. The Iroise Sea is likely a mixing zone for different stocks or sub‐populations, and may also shelter a resident population. At the population scale, such findings may impact ICES stock assessment and the resulting decisions from EU managers. At the local scale, conservation action could be taken by MPA managers. Besides, this study demonstrates the high potential of archival tags for investigating multi‐year behavioural patterns such as site fidelity to offshore spawning areas.

Coupling spectral analysis and hidden Markov models for the segmentation of behavioural patterns

BackgroundMovement pattern variations are reflective of behavioural switches, likely associated with different life history traits in response to the animals’ abiotic and biotic environment. Detecting these can provide rich information on the underlying processes driving animal movement patterns. However, extracting these signals from movement time series, requires tools that objectively extract, describe and quantify these behaviours. The inference of behavioural modes from movement patterns has been mainly addressed through hidden Markov models. Until now, the metrics implemented in these models did not allow to characterize cyclic patterns directly from the raw time series. To address these challenges, we developed an approach to i) extract new metrics of cyclic behaviours and activity levels from a time-frequency analysis of movement time series, ii) implement the spectral signatures of these cyclic patterns and activity levels into a HMM framework to identify and classify latent behavioural states.ResultsTo illustrate our approach, we applied it to 40 high-resolution European sea bass depth time series. Our results showed that the fish had different activity regimes, which were also associated (or not) with the spectral signature of different environmental cycles. Tidal rhythms were observed when animals tended to be less active and dived shallower. Conversely, animals exhibited a diurnal behaviour when more active and deeper in the water column. The different behaviours were well defined and occurred at similar periods throughout the annual cycle amongst individuals, suggesting these behaviours are likely related to seasonal functional behaviours (e.g. feeding, migrating and spawning).ConclusionsThe innovative aspects of our method lie within the combined use of powerful, but generic, mathematical tools (spectral analysis and hidden Markov Models) to extract complex behaviours from 1-D movement time series. It is fully automated which makes it suitable for analyzing large datasets. HMMs also offer the flexibility to include any additional variable in the segmentation process (e.g. environmental features, location coordinates). Thus, our method could be widely applied in the bio-logging community and contribute to prime issues in movement ecology (e.g. habitat requirements and selection, site fidelity and dispersal) that are crucial to inform mitigation, management and conservation strategies.

Estimating total uncertainty for abundance at age estimates from acoustic‐trawl surveys.

A comprehensive quantitative treatment of acoustic‐trawl survey uncertainty that includes all major components of measurement and sampling error is needed to help specify the level of risk in stock assessment models. Acoustic and trawl sampling errors can be addressed by geostatistical simulations, which account for autocorrelation, non‐independence, and non‐random sampling, while additional sources of uncertainty, such as ship motion and instrument errors, can be incorporated into uncertainty estimates by bootstrapping. A conceptual model is presented in the form of a flow diagram, which identifies sources of uncertainty and recommends analytical methods to quantitatively assess the uncertainty associated with abundance at size estimates. Sequential geostatistical simulations of the spatial distribution of backscatter attributed to walleye pollock from Eastern Bering Sea acoustic‐trawl surveys show that the acoustic sampling error is one of the two largest sources of uncertainty in estimates of total abu...