¹Biology Department, Dalhousie University, Halifax, Nova Scotia, Canada

16.1. Introduction

The Census of Marine Life's overarching goal is to assess and explain the diversity, distribution, and abundance of marine organisms throughout the world's oceans. By stimulating exploration and research in all ocean habitats it has accumulated an unprecedented wealth of new information on the patterns and processes of marine biodiversity on a global scale. Three questions are guiding this research effort. What did live in the oceans? What does live in the oceans? What will live in the oceans? The Future of Marine Animal Populations (FMAP) Project ultimately aims to answer that third question through the analysis and synthesis of available data, and the modeling of patterns and trends in marine biodiversity. This entails all levels of biodiversity, from individuals, to populations, communities, and ecosystems (Box 16.1). 

Despite the ultimate focus on future prediction, the synthetic analyses undertaken within the FMAP project inform all three aspects of the Census, past, present, and future. The rationale is that without a solid understanding of past and present trends, it is impossible to make sound future projections. Likewise, our research efforts encompass different levels of organization, from the movements of individual animals through space and time, to broad macro-ecological patterns of abundance and diversity. Hence, an improved understanding of processes at the level of an individual animal may help inform the interpretation of larger-scale patterns.

Our main analytical tools are meta-analytic models, used to combine and understand species abundance and distribution trends, including both historical and recent data. Models that are effective for synthesis also have potential for prediction, and have been used by others to project potential future effects of fishing and climate change, for example Botkin et al. ( 2007). Moreover, modeling can help define the limits of knowledge: what is known and how firmly, what may be unknown but knowable, and what is likely to remain unknown in the foreseeable future.

FMAP grew out of a workshop held at Dalhousie University in Halifax, Nova Scotia, Canada, in June 2002. Representatives of other Census projects, including the History of Marine Animal Populations (HMAP) project, the field projects, and the Ocean Biogeographic Information System (OBIS), participated and provided guidance in the design of this project. FMAP was originally envisioned and led by Ransom A. Myers, Killam Chair of Ocean Studies at Dalhousie University. His leadership carried the project until his sudden passing in 2007. Two additional FMAP centers were established in 2003 at the University of Iceland with Gunnar Stefansson, and the University of Tokyo with Hiroyuki Matsuda. Since 2007 the project has been co-led by the authors of this chapter.

FMAP's mission has been to describe and synthesize globally changing patterns of species abundance, distribution, and diversity, and to model the effects of fishing, climate change, and other key variables on those patterns. This work has been performed across ocean realms and with an emphasis on understanding past changes and predicting future patterns. The project benefitted throughout from close collaboration with statisticians and mathematical modelers, which enabled the proper processing and analysis of large datasets. FMAP has collaborated with other Census projects to varying degrees, most consistently with HMAP, Tagging of Pacific Predators (TOPP), and OBIS (see Chapters 1, 15, and 17), as well as various deep-sea projects.

This chapter does not intend to provide an exhaustive overview of the research activities within FMAP (see www.fmap.ca for individual projects and publications). Instead, we aim to highlight key areas of interest and discuss major advances that have been made. It is structured along three major research topics, aiming to cover the major research themes of the Census (distribution, abundance, and diversity of marine life): (1) marine biodiversity patterns and their drivers, (2) long-term trends in animal abundance and diversity, (3) distribution and movements of individual animals. In the concluding section we aim to provide some insight into what is unknown, and what is currently unknowable, particularly with respect to predicting the future of marine biodiversity.

16.2.1. Previous Work

Before the Census, mapping of the ocean with respect to our knowledge of fundamental patterns of abundance and diversity was limited. The first global study was published in 1999, presenting a pattern of planktonic foraminiferan diversity derived from the analysis of a large sediment core database (Rutherford et al. 1999). Another study highlighted global hot spots of endemism and species richness for corals and associated organisms (Roberts et al. 2002). Several authors had investigated latitudinal gradients for particular species groups (Hillebrand 2004). Yet compared with our understanding of life on land, synthetic knowledge on marine biodiversity was sparse. It became clear from these early studies, however, that some of the patterns were uniquely different from those seen on land, where biodiversity is generally highest in the tropics (Gaston 2000).

16.2.2. Large Marine Predators

FMAP studies have mainly focused on large pelagic predators such as tuna and billfish, whales, and sharks, for which global data were available. These species groups were found to peak in diversity in the subtropics, often between 20–30 degrees latitude north or south. Although a similar distribution pattern was first described for Foraminifera (Rutherford et al. 1999), we were able to show that this is a more general pattern that applies across very different species groups (Worm et al. 2003, 2005). Furthermore, it became clear that this biodiversity pattern is not static, but dynamically changing on both short and long time scales.

Species richness patterns for tuna (Thunnini), billfish (Istiophoridae), and swordfish (Xiphiidae) were derived from a global Japanese longline-fishing dataset (Fig. 16.1). Pelagic longlines are the most widespread fishing gear in the open ocean, and are primarily used to target tuna and billfish. The Japanese data represents the world's largest longline fleet and the only globally consistent data source reporting species composition, catch and effort for all tuna, billfish, and swordfish. Statistical rarefaction techniques were used to standardize for differences in fishing effort and to estimate species richness (the expected number of species standardized per 50 randomly sampled individuals) for each 5° × 5° cell in which the fishery operated.

Figure 16.1 Tuna and billfish species richness over time. Maps depict the number of expected species per 50 individuals as calculated from pelagic longlining catch and effort data using rarefaction techniques. After data from Worm et al. ( 2005).

As seen in Figure 16.1, species richness of tuna and billfish displayed a global pattern with large hot spots of diversity in all oceans in the 1960s. These hot spots faded over time, indicating declining species richness, a pattern most clearly seen in the Atlantic and Indian Oceans. Declining species richness coincided with 5- to 10-fold increases in total fisheries catch of tuna and billfish in all oceans, which may have led to regional depletion of vulnerable species (Worm et al. 2005). In the Pacific, however, initial losses of diversity began to reverse in 1977, coinciding with a large-scale climate regime shift, whereas the Pacific Decadal Oscillation changed from a cool to a warm phase. Climatic drivers were also found to be important on an annual scale. Short-term (year-to-year) variation in species richness showed a remarkable synchrony with the El Niño Southern Oscillation (ENSO) index, with increasing temperatures leading to basin-wide increases in species richness (Worm et al. 2005). This may be explained by warming of sub-optimal temperature habitats. ENSO-related decreases in diversity were seen in the tropical Eastern Pacific, a region that suffers from greatly reduced productivity and associated mass mortality of marine life during El Niño events. A subsequent study showed that seasonal variation in sea surface temperature is driving the taxonomic richness patterns for deep-water cetaceans (whales and dolphins) as well (Whitehead et al. 2008).

For tuna and billfish, as well as cetaceans and Foraminifera, mean sea surface temperature (SST) clearly emerged as the strongest single predictor of diversity, showing a positive correlation over most of the observed temperature range (5–25 °C), but a negative trend above that (Fig. 16.2). This decline of diversity at high temperatures was most pronounced in the western Pacific “warm pool”, which has the highest equatorial SST (warmer than 30 °C), and weakest in the tropical Atlantic, which has the lowest equatorial SST (lower than 27°C). The relation between tuna and billfish diversity and SST could also be independently reconstructed from an analysis of individual species temperature preferences (Boyce et al. 2008).

Figure 16.2 Temperature effects on diversity. Shown are the empirical relationships between sea surface temperature (SST) and species richness for deep-water cetaceans (blue line), planktonic foraminiferans (green line), and tuna and billfish (red line). After data from Worm & Lotze (2009).

Another factor that explained significant variation in tuna and billfish species richness on a global scale was the steepness of horizontal temperature gradients. Sharp temperature gradients are found around frontal zones and eddies that are typically associated with mesoscale oceanographic variability. Fronts and eddies often attract large numbers of species, likely because they concentrate food supply, enhance local production, and increase habitat heterogeneity (Oschlies & Garçon 1998; Hyrenbach et al. 2000). They may also form important landmarks along transoceanic migration routes (Polovina et al. 2001). Finally, dissolved oxygen concentrations were positively correlated with diversity. This likely relates to species physiology, as low oxygen levels (less than 2 ml l ^–1) may limit the cardiac function and depth range of many tuna species (Sund et al. 1981). Regions of low oxygen are located west of Central America, Peru, West Africa, and in the Arabian Sea. Despite optimal SST around 25 °C, most of these areas showed conspicuously low diversity.

Knowledge of the relation between SST and diversity for various species groups (Fig. 16.2) allows us to predict how diversity may change as SST changes spatially and temporally with climate variability and climate change. The effects of climate variability, such as ENSO and the Pacific Decadal Oscillation, are discussed above. With respect to long-term climate change, Whitehead et al. (2008) combined Intergovernmental Panel on Climate Change (IPCC) scenarios for observed and projected changes in SST between 1980 and 2050 with an empirically derived relation of SST and deep-water cetacean diversity. For the baseline 1980 dataset, diversity was predicted to be highest at latitudes of about 30°, falling towards the equator, and more precipitously towards the poles. With global warming, these bands of maximal diversity were predicted to move pole-wards. The warming tropical oceans were predicted to decline in diversity, while richness was predicted to increase at latitudes of about 50°–70° in both hemispheres (Whitehead et al. 2008). These general conclusions were recently corroborated by an analysis of 1,066 exploited fish and invertebrate species (Cheung et al. 2009).

16.2.3. Other Species Groups

Other groups that were investigated with respect to their diversity patterns were deep-water corals and tropical reef fish. The goal was to gain a better understanding of the effects of human impacts such as fishing and ocean acidification on the distribution, abundance, and diversity of different species groups (reviewed by Tittensor et al. 2009b).

A study on tropical reef fish at fished and unfished sites in three oceans revealed predictable changes in the species–area relation (SAR). The SAR quantifies the relation between species richness and sampling area and is one of the oldest, most recognized patterns in ecology. Fishing consistently depressed the slope of the SAR, with the magnitude of change being proportional to fishing intensity (Tittensor et al. 2007). Changes in species richness, relative abundance, and patch occupancy contributed to this pattern. It was concluded that species-area curves can be sensitive indicators of community-level changes in biodiversity, and may be useful in quantifying the human imprint on reef biodiversity, and potentially elsewhere (Tittensor et al. 2007). This study highlighted how human impacts can affect biodiversity through multiple pathways.

Subsequent work focused on cold-water scleractinian corals, an important habitat-forming group of stony corals commonly found on seamounts (Clark et al. 2006). Despite their widely accepted ecological importance, records of cold-water corals are patchy and simply not available for most of the global ocean. In an FMAP-CenSeam (Global Census of Marine Life on Seamounts) collaboration (see also Chapter ), the probable distribution of these corals was derived from habitat suitability models, that incorporated all the available data on cold-water coral distribution in relation to environmental variables such as depth, temperature, and carbonate availability (Tittensor et al. 2009a). Highly suitable habitat for seamount stony corals was predicted to occur in the North Atlantic, and in a circumglobal strip in the Southern hemisphere between 20° and 50° S and at depths shallower than around 1,500 m (Fig. 16.3). Seamount summits in most other regions appeared less likely to provide suitable habitat, except for small near-surface patches. In these models oxygen and carbonate availability played a decisive role in determining large-scale scleractinian coral distributions on seamounts (Tittensor et al. 2009a). These results raise concerns about the possible consequences of ocean acidification (Orr et al. 2005) and the observed shallowing of oxygen minimum zones in the wake of global climate change (Stramma et al. 2008). Both factors would be predicted to limit the distribution of scleractinian corals, and the fauna associated with them.

Figure 16.3 Habitat suitability for cold water corals on seamounts. Colors indicate relative predicted habitat suitability ranging from high (red) to low (blue) as revealed by maximum entropy habitat suitability modeling (after Tittensor et al. ( 2009b)). The photograph depicts Lophelia pertusa framework with rich associated invertebrate fauna, Hatton Bank, Northeast Atlantic (UK Department for Business Innovation and Skills (formerly DTI) Strategic Environmental Assessment Programme, c/o Bhavani Narayanaswamy).

16.2.4. Total Species Richness

The number of species is the most basic index used to measure biodiversity and one that plays a fundamental role in the quantification of human-related extinctions and impacts. Unfortunately, the total number of species remains poorly known in the oceans. For example, Grassle & Maciolek (1992) famously suggested that the number of (largely unknown) deep-sea benthic species is more than 1 million, but may even exceed 10 million. The only published estimate of the total number of marine species relied on an inventory of European fauna that was scaled up to the global level (Bouchet 2006). A more analytical approach has recently become possible through the Census' Ocean Biogeographical Information System (OBIS) in combination with newly developed modeling approaches (Mora et al. 2008). These modeling methods derive estimates of species richness from “discovery curves” of species sampled over time, and produce confidence limits that allow us to estimate the known and unknown of global species richness. An FMAP pilot project on total marine fish species has estimated that there are approximately 16,000 known species of marine fish, with about another 4,000 awaiting discovery (Mora et al. 2008). These methods are currently being used to estimate the known and unknown of total marine species richness.

16.3.1. Previous Work

Underlying the changing patterns of biodiversity or species richness are changes in the abundance and distribution of individual populations. Most previous work has emphasized variability in population abundance in relation to climate, oceanography, or other factors on yearly to decadal (see, for example, Attrill & Power 2002) or evolutionary time scales (Vermeij 2004; Jackson & Erwin 2006). Changes in marine life over the Anthropocene (the past few hundred years; an epoch dominated by human influences) have only recently received focused attention. This has two reasons: first, the ocean has long been seen as a vast frontier, where human activities would not leave a permanent mark; second, empirical monitoring data are mostly available just for the past 20 to 50 years, which prevented longer-term studies from reaching back beyond the twentieth century.

16.3.2. Synthesizing Long-Term Trends

Over the past decade, the Census at large, and HMAP and FMAP in particular, have partly overcome these limitations. Although HMAP has made enormous progress in unraveling detailed historical, archaeological, and paleontological records of past changes in different animal populations and regions (see Chapter 1), FMAP has developed ways of combining and analyzing these data to reveal long-term changes in ocean ecosystems, and uncover their drivers and consequences.

One of FMAP's goals has been to synthesize the long-term trends in the abundance, distribution, and diversity of marine life. This has been pursued for coastal regions over the past centuries and millennia (Lotze & Milewski 2004; Lotze et al. 2005, 2006) and continental shelf and open ocean regions over the past 50 years (Myers & Worm 2003, 2005; Worm et al. 2005, 2009). These studies have shown that human impacts have resulted in sharply reduced abundance of target and some non-target populations, as well as range contractions and local extinctions that precipitated local and regional losses of species diversity.

To synthesize long-term trends in population abundances of large marine animals, we analyzed 256 records from 95 published studies, many of them from HMAP, FMAP, or other Census projects (Lotze & Worm 2009). Trend estimates for marine mammals, birds, reptiles, and fish were derived from archaeological, historical, fisheries, ecological, and genetic studies and revealed an average decline of 89% (range: 11–100%) from historical abundance levels (Lotze & Worm 2009). Remarkably, the magnitude of depletion was relatively consistent across different species groups (Fig. 16.4A) despite considerable variability in data quality, analytical methods, and time span of the records. Diadromous fish such as sturgeon and salmon, sea turtles, pinnipeds, otters, and sirenia showed the strongest declines with more than 95%. On the other hand, conservation efforts in the twentieth century enabled several whale, pinniped, and coastal bird species to recover from a historical low point in abundance (Fig. 16.4A). These recoveries have reduced the level of depletion across all 256 analyzed species to 84% on average.

Figure 16.4 Long-term population declines in large marine animals. Shown are relative changes across (A) species groups, (B) ocean realms, and (C) time period (AD) when exploitation started. There are two measures, the decline to the low point of abundance in the past (open circles) and the decline to today (filled circles), with the difference indicating recovery in population abundance (based on data from Lotze & Worm ( 2009)).

Another important dimension of change is the spatial expansion of exploitation, which began in rivers and along the coasts centuries ago and only in the mid-twentieth century moved towards open oceans and the deep sea. Thus, some of the highest population declines can be found in rivers and coastal habitats, with lesser declines found on continental shelves and the open ocean (Fig. 16.4B). Deep-sea habitats differ from this trend, which may be explained by their extreme vulnerability to exploitation (Roberts 2002). Along with this spatial expansion there has been a temporal acceleration in exploitation due to technological advances. Population declines unfolded over hundreds or thousands of years in many rivers and coastal regions, one to two hundred years on the continental shelves, approximately 50 years in the open ocean, and approximately 10–20 years in the deep sea (Lotze & Worm 2009). As a result, the average magnitude of change is almost independent of when exploitation started (Fig. 16.4C). Interestingly though, recoveries are mostly found in species that have been exploited at least 100 years ago and protected in the early to mid-twentieth century, whereas more recently exploited species do not yet show recovery.

Changes in population abundance and distribution have resulted in changes in species diversity. As was discussed previously, there have been remarkable changes in tuna and billfish species richness in the open ocean over the past 50 years (Fig. 16.1). In the coastal ocean, changes in diversity have occurred in two ways: (1) diversity declines have occurred in large marine animals such as mammals, birds, reptiles, and fish due to global, regional, or ecological extinctions; (2) diversity increases have occurred through the invasion of mostly smaller species including invertebrates, plants, unicellular plankton and bacteria, and viruses (Lotze et al. 2005, 2006). This shift in diversity from large- to small-bodied animals has resulted in a different species composition, with consequences for ecosystem structure, functioning, and services (see below).

16.3.3. Drivers of Long-Term Change

The underlying drivers of observed long-term changes may include natural and anthropogenic drivers, as well as the cumulative effects of multiple factors. To unravel the relative importance of different drivers on population and ecosystem changes, we have used a variety of methods, including meta-analysis of large datasets, experimental manipulations, and ecosystem models.

For example, an analysis of drivers of long-term population changes in 12 estuaries and coastal seas revealed that exploitation (primary factor) and habitat loss (secondary) were by far the most important causes for the depletion and extinction of marine species over historical time scales (Lotze et al. 2006). Pollution, physical disturbance, disease, eutrophication, and introduced predators also contributed to some species declines, although to a lesser extent. However, the reverse was also true: conservation efforts in the twentieth century, especially reduced exploitation, the protection of habitat, and in some cases pollution control, enabled several species to recover from low abundance. In many cases it was not a single factor, but a combination of exploitation, habitat loss, and other factors that caused a population decline or – in reverse – enabled recovery (Lotze et al. 2006). Whereas species invasions and climate change were less dominant drivers of marine biodiversity change in the past, they may increase in importance in the future (Harvell et al. 2002; Harley et al. 2006; Worm & Lotze 2009).

The cumulative and interactive effects of different drivers have also been explored with multi-factorial laboratory experiments. For example, a three-factorial experiment that used rotifers as a model system showed additive effects between exploitation and habitat fragmentation on population declines and synergistic effects if environmental warming was also involved (Mora et al. 2007). Although each of these three factors individually caused populations to decline by similar amounts, all factors combined resulted in up to 50 times faster declines of experimental populations (Fig. 16.5A). These results highlight the importance of multiple human drivers for past and future population and biodiversity changes in the ocean.

Figure 16.5 Multiple drivers of biodiversity change. (A) Experimental manipulations of the cumulative effects of harvesting, immigration (as a measure of habitat fragmentation), and environmental warming on population decline in a model organism (after Mora et al. 2007, with permission). (B) Effects of different socioeconomic and environmental factors on the regional variability of coral reef communities (adapted from Mora ( 2008)).

Many human drivers have their ultimate roots in the social conditions and economic activities of society. Although some individuals, communities, or societies may exercise overexploitation, habitat destruction, and waste dumping, others promote successful stewardship and governance through harvest regulations, pollution controls, and the protection and restoration of species and habitats (Lotze & Glaser 2009; Worm et al. 2009). In one study, Mora ( 2008) analyzed socioeconomic and environmental databases together, to separate the proximate and ultimate drivers of coral reef degradation. Most of the local and regional variability in fishes, corals, and macroalgae was explained by human-related factors such as agricultural land use, coastal development, overfishing, and climate change (Fig. 16.5B). Significant ecological interactions among the different species groups further highlighted the need for a comprehensive management of human influences on coral reef ecosystems (Mora 2008).

16.3.4. Ecosystem Consequences

What are the consequences of long-term population changes on the structure and functioning of marine ecosystems today and in the future? And how will changes in ecosystem structure affect the services marine ecosystems provide for human well-being? These questions have been difficult to tackle. First, ecosystem structure, functioning, and services are not easy to quantify and the relevant data are not readily available. This problem was overcome by compiling and analyzing long-term datasets on ecological, environmental, and socioeconomic changes in marine ecosystems. Second, a multitude of species, environmental drivers, and human impacts may interact in ways that are impossible to unravel from simple trend analyses. We therefore used ecosystem modeling approaches to determine overall changes in food web structure, energy flows, and stability. These modeling techniques may also serve to project future scenarios under changing environmental or human conditions, which we are currently exploring.

Long-term changes in population abundance as well as the loss (extinction) and gain (invasion) of species has changed the structure of many coastal ecosystems (Fig. 16.6A). This can have important effects on ecosystem functioning. Many species fulfill important ecological functions, including the provision of spawning, nursery, and foraging habitat by wetlands, underwater vegetation, and reef-building organisms (Fig. 16.6B). Most of these habitats also play an important role in filtering particles, nutrients, and pollutants, thereby maintaining good water quality. If the functioning of coastal ecosystems is compromised, so are the ecosystem services provided for human well-being (Worm et al. 2006; Lotze & Glaser 2009). For example, overexploitation, habitat loss, and pollution have depleted many fisheries that previously provided food and employment (Fig. 16.6C). The loss of filter functions together with increasing municipal and industrial discharges have posed health risks to people through harmful algal blooms, contaminants, and disease. Finally, the expansion of oxygen-depleted zones, invasive species, and flooding compromises recreation and shoreline safety (Worm et al. 2006). As human impacts spread offshore and expand to other ocean regions (Halpern et al. 2008), the observed changes in coastal oceans may forecast potential future changes in other habitats.

Figure 16.6 Ecosystem consequences of marine biodiversity change. (A) Structural changes in 12 estuaries and coastal seas measured as the relative abundance and occurrence of species (as percent from historical baseline) and number of species invasions. (B) Changes in ecosystem functions such as habitat provision and water quality control expressed as nutrient loading and eutrophication response. (C) Loss of ecosystem services including the depletion of fisheries, health risks related to harmful algal blooms (HAB), and diminished recreation related to dead zones (data adapted from Lotze et al. (2006); Worm et al. (2006)).

To understand better the ecosystem effects of marine biodiversity change, we used two different modeling approaches, stochastic network models and mass-balance food web models. First, basic food and interaction webs are assembled from data on species occurrence, abundance, feeding links, and other ecological information (Coll et al. 2008). Next, the two different models enabled us to analyze changes in up to 22 food web properties reflecting (among others) species composition, food-chain length, energy transfer between trophic levels, and the linkage density and complexity of the webs. For example, food webs in the Adriatic and Catalan Seas in the Mediterranean were found to be very similar in terms of their structure and functioning, but were more ecologically degraded compared with food webs from the Caribbean, Benguela, and US continental shelf (Coll et al. 2008). Food web properties estimated by both models yielded very similar results, thereby enhancing confidence in our results compared with any single modeling approach.

The network models also allowed us to analyze the robustness of food webs to simulated species loss (Coll et al. 2008). It had previously been shown that the removal of highly connected species in the food web results in a much higher rate of secondary extinctions than randomly deleted or less strongly connected species. In our analyses, removing the commercial species caused intermediate rates of secondary extinctions indicating that commercial species are often well connected in the food web (Fig. 16.7A). Finally, we could show that a larger degree of ecological degradation in the Mediterranean food webs resulted in a diminished robustness to species loss and a higher rate of secondary extinctions (Fig. 16.7B). This suggests that the degradation of marine ecosystems may accelerate the rate of biodiversity loss in the future.

Figure 16.7 Food web modeling. Robustness of food webs to the simulated extinction of (A) species that are most connected, most commercial, random, or least connected in the Adriatic food web in the 1970s, and (B) the most connected species in food webs from different regions and time periods. Diagonal black lines indicate where 50% of species are lost through combined removals and secondary extinctions (i.e. robustness) (adapted after Coll et al. (2008)).

16.4.1. Previous Work

Movement patterns and behavior of individual animals collectively contribute to broader-scale population distribution, species' ranges, and patterns of biodiversity. The development of a statistical toolbox for studying the movements of electronically tagged marine predators and collaboration with animal trackers provides a powerful complement to the global-scale studies of biodiversity and abundance conducted by FMAP. Our aim has been to elucidate the underlying mechanisms that determine animal distributions at the individual scale, and how these contribute to marine predator distribution and biodiversity patterns at broader scales. Much of our previous knowledge on marine animal movement patterns has been inferred from observations of species' departures and arrivals at geographically disparate locations (Carr 1986) or from traditional mark–recapture methods (Hilborn 1990). These studies were necessarily coarse in scale and yielded little insight into the interactions between foraging or migrating animals and their environment. Furthermore, these approaches only revealed movements to locations where observers were present, biasing estimates of movement rates and, more generally, our understanding of marine animal movement patterns.

All this has changed with the introduction of electronic tagging and telemetry technologies that revolutionized our view of animal movement patterns and distribution in the ocean (Block et al. 2001; Birdlife International 2004; James et al. 2005). Satellite and light-based geolocation tracking technologies now allow us to follow marine animals for protracted periods as they make their living in the ocean (see Chapter 15). The veritable explosion of tracking studies using increasingly refined technologies has revealed, for example, extraordinary 64,000-km round-trip migrations by sooty shearwaters (Shaffer et al. 2006), physiological mechanisms underlying niche expansion in salmon sharks (Weng et al. 2005), and previously unknown return migrations in white sharks (Bonfil et al. 2005). Despite these and many other success stories, our ability to track and document animal movements has far out-stripped our ability to conduct sophisticated analyses of rapidly amassing tracking data. This gap proved to be a fertile area of investigation for FMAP.

16.4.2. Statistical Tools and Key Results

The project has played the leading role in developing a state-of-the-art statistical toolbox for electronic tracking data (Jonsen et al. 2003, 2005). This has been a multidisciplinary venture, involving biologists, ecological modelers, and statisticians. Our approach has focused on state-space models (SSMs), which are statistical time-series tools that, in the present context, allow one to estimate true animal positions from error-prone tracking data. State-space models have two components, a process model that describes how animals move from one position to the next and an observation model that relates the true, unobserved positions to the empirical tracking data.

In general, there can be two goals to fitting a SSM to tracking data. The first goal is to estimate the true positions of a tagged animal by accounting for the observation error inherent in the tracking data. This is called state filtering and yields a set of position estimates (and associated uncertainties) that occur over regular time intervals (Fig. 16.8). This kind of filtering is fundamentally different from traditional travel rate filters (McConnell et al. 1992) as all observed locations are modeled by a known probability distribution with implausible observations being down-weighted rather than discarded. The result is that all information contained in the data is used to estimate the true positions.

Figure 16.8 Example of state filtering applied to data for a grey seal tagged on Sable Island, Nova Scotia, with a SMRU (Sea Mammal Research Unit) Argos satellite platform terminal transmitter with a 2-day duty cycle. The white points and lines denote the Argos-observed positions, which contain substantial error (note positions on land and some highly improbable movements). The red points denote the estimated true positions with a 2-day time step. Adapted from Jonsen et al. (2005).

Breed et al. (2006) used the state filtering approach to gain insight into the sexual segregation of seasonal foraging in adult grey seals breeding on Sable Island, Nova Scotia. Grey seals are an important generalist predator in the Scotian Shelf ecosystem, with a population that has experienced exponential growth over the past 35 years. From October to December and February to March, males used areas along the continental shelf break, whereas females used mid-shelf regions. Breed et al. (2006) suggested that this broad-scale segregation may help individuals maximize fitness by reducing intersexual competition during primary foraging periods.

The second goal of fitting a SSM is to estimate biological parameters or behavioral states specified in the process model, thereby allowing inference of unobservable processes that drive the movement and distribution patterns. The ability to construct biologically meaningful models and to estimate their parameters directly from complex, error-prone data is the most compelling facet of the SSM toolbox. For analyses of individual movement datasets, this approach is best achieved when individual datasets are combined meta-analytically (Jonsen et al. 2003). Meta-analysis facilitates synthesis of multiple datasets, enabling inference both within and among datasets, and improves parameter estimation from limited datasets.

State-space models allow researchers to think about questions that have no conventional solution. Jonsen et al. (2006) highlighted this by showing that endangered leatherback turtles migrating throughout the North Atlantic slow down at night, perhaps to feed on macrozooplankton that migrate toward the ocean surface and/or because their navigation abilities are less precise than during the day. One problem with this is that calculating travel speeds becomes difficult when travel distance is small during a single day or night period (no more than 30 km) compared with the uncertainty in the observed positions (up to 250 km). Conventional analyses of these data are not able to reveal the patterns in day versus night travel rates that the SSM analysis can (Fig. 16.9). Furthermore, Jonsen et al. (2006) showed that a Bayesian meta-analytic SSM, a model that estimates day versus night travel rates simultaneously from all datasets, yielded superior estimates for individual turtles and provided the basis for prediction at a population level.

Figure 16.9 Comparison of the ratio of day-to-night travel rates for a conventional ad-hoc calculation of travel rates (A) and a meta-analytic SSM fit to all datasets simultaneously (B). The ratios presented in (A) are means with 95% confidence intervals, and those presented in (B) are posterior modes with 95% credible interval raindrops. The black raindrop is the Bayesian predictive distribution (BPD), which forms the basis for making predictions at a population or other appropriate level. Adapted from Jonsen et al. (2006).

A particularly compelling application of the SSM allows one to infer the (hidden) behavioral state of animals based upon the shifts in movement patterns observed in the tracking data. To derive, for example, the probability of an animal foraging versus migrating, a switching model can be added to the standard SSM (Jonsen et al. 2005). Key to this approach are the underlying assumptions that animal movements can be modeled by a small set of correlated random walks, each corresponding to a unique behavioral state, and that animals typically engage in area-restricted type movements (such as slow travel rates with a high frequency of turning) when searching for and consuming prey (Jonsen et al. 2007). The switching model is used to estimate the probabilities that an animal is in a particular behavioral state, conditional upon the previous behavioral state. This approach offers a powerful tool to infer animal behaviors from remote tracking data and is being used by FMAP collaborators to analyze migration and foraging patterns in Pacific leatherback turtles (Bailey et al. 2008; Shillinger et al. 2008). Similar approaches have been adopted for analyzing foraging behaviors of southern bluefin tuna (Patterson et al. 2009).

FMAP researchers have used the switching SSM approach to identify potential foraging areas in space and time, a first step in quantifying critical foraging habitat and in developing mechanistic predictions of the potential influence of climate variability and changing distribution of foraging predators. Breed et al. (2009) show that adult male and female grey seals forage in different areas of the Scotian Shelf but both sexes tend to focus on relatively small, intensely used foraging sites (Fig. 16.10A). This switching SSM analysis builds on previous efforts that suggested grey seals foraged over broad areas of the shelf (Breed et al. 2006) by resolving spatial patterns in movement behaviors hidden within the tracking data (Breed et al. 2009). Similar patterns of concentrated foraging activity, albeit at a much broader spatial scale, are emerging in an ongoing analysis of leatherback turtle satellite data (Fig. 16.10B). Figure 16.10B presents only a small, but fairly representative, subset of the results so far. Switching SSM analysis of 36 individual tracks, spanning a period from 1999 to 2006, suggests that foraging activity is concentrated along the slope waters of the Scotian Shelf and around Cape Breton in Northeastern Nova Scotia. Both regions are highly productive and likely support large seasonal aggregations of the leatherback's prey, jellyfish.

Figure 16.10 Foraging patterns of adult male and female grey seals (A) and adult or sub-adult leatherback turtles (B) inferred by a switching SSM. In panel (A), blue points are positions associated with foraging behavior and red points are positions associated with transiting behavior. Color intensity indicates the degree of uncertainty in the behavioral state estimates (less intense is less certain). In panel (B), light blue points are positions associated with foraging behavior, red points are positions associated with transiting behavior, and yellow points are positions with uncertain behavioral state estimates. Image of adult female grey seal leaving Sable Island with an Argos satellite tag and VHF radio tag is courtesy of W.D. Bowen. Image of an Argos satellite-tagged leatherback turtle returning to the water after tagging is courtesy of the Canadian Sea Turtle Network. Adapted from Breed et al. ( 2009) (A) and Jonsen et al. (unpublished data) (B).

What is striking about both of these analyses is the distinct patchiness of foraging activity exhibited by both species. Grey seals are benthic foragers and are predominantly tied to shallow banks on the continental shelf, whereas leatherback turtles forage on jellyfish both in the coastal and pelagic realms. Regardless, both species show distinct and annually predictable preferences for relatively discrete regions. Understanding how these behavioral patterns may change in the future as a result of changes in prey distribution and abundance, and as a result of climate variability can provide valuable insight into mechanisms underlying future change in species biodiversity and abundance. Current work is focusing on the biophysical characterization of species' foraging areas and development of switching SSMs that can directly relate environmental gradients to the behavioral patterns hidden within electronic tracking data.

The FMAP Project has been active from 2002 to 2010 as the modeling component of the Census. Our emphasis has been on data synthesis to reveal broad patterns of diversity, abundance, and distribution of marine animals, with a particular emphasis on large predators. We have developed data analysis and modeling techniques that enabled us to estimate the diversity of known and unknown species in the ocean, to derive long-term trends, short-term dynamics, and spatial patterns of diversity change, and to understand better the distribution and behavior of individual species.

In our communications we have strived to use the knowledge gained to inform society about both the drivers of biodiversity change, as well as current and possible future consequences of biodiversity loss for marine ecosystems and human society. Major results and patterns derived from this project are listed in Box 16.2. In interpreting this body of new knowledge, it is important to realize that these patterns do not play out equally everywhere. A necessary weakness inherent in large-scale data synthesis is that individual differences may be lost in averaging across a large population of animals, different regions, and environmental conditions. It may be these differences, however, existing between one place and another, or one population and another, that hold the key to understanding the ocean's future (Worm et al. 2009). There is no doubt that, given our large and growing influence on the marine environment, future societal decisions will drive the trajectory of change in marine animal populations and the ecosystems in which they are embedded. Within FMAP we can highlight past and current trends, and assume different scenarios as to how these may extend into the future. What will actually happen, however, is unknown, as societal choices and technological change will determine future changes. This process will certainly be influenced by the availability of new scientific information, and its perception among the public and decision-makers. It is our hope that the FMAP Project will continue to provide such information into the foreseeable future. 

This chapter and the research highlighted in it have been supported by the Census of Marine Life, funded by the Sloan Foundation. Additional funding came from the Natural Sciences and Engineering Research Council of Canada, the Lenfest Oceans Program, and the Canadian Department of Fisheries and Oceans. We acknowledge the leadership of the late Ransom A. Myers, former Principal Investigator of FMAP, as well as numerous students and researchers from FMAP and other Census projects that made this work possible. We especially thank FMAP founding members Hiroyki Matsuda and Gunnar Steffansson.