Species sightings data

Kenya has among the highest quality species sightings and catch trend data known for a developing country. Further, Kenyan reef fish species occur throughout much of the Indian Ocean and coral reefs worldwide. We defined reef fish species as shallow-living (<100 m depth), tropical/ subtropical, benthic or benthopelagic fishes that either associate with hard bottom substrates composed of coral, algae, or dead coral reef, or alternatively, that occupy adjacent sandy substrate and seagrass. We reconstructed occurrences of each historically-exploited species using four data sources. These included archaeological records (750–1500), naturalists’ species lists (1759–2003), catch assessment surveys (1995–2013), and underwater surveys (1991–2013). The archaeological records were only from two sites along the Kenyan coast in the extreme north and south, whereas the catch assessments and underwater surveys were conducted more broadly (Fig 1). The spatial resolution of the naturalist’s species lists is unknown.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Map of Kenya showing the locations of various data sources for reef fish species occurrences through time.

https://doi.org/10.1371/journal.pone.0211224.g001

Data limitations

A number of biases and limitations exist when reconstructing species occurrence from disparate data sources, including differences in types of sighting and the reliability of the sighting. For example, the archaeological remains represent only captured species and not the full assemblage of fish. Species may have existed in the past but were not captured and therefore undetected and impossible to assess for extinction risk. Variable levels of preservation and sampling strategies are also affect occurrences in the archaeological record. Other limitations include variation in the temporal resolution and data gaps, primarily in our study from 1500–1991 CE. Unlike archaeological data, species identified in naturalists’ lists provide a single date when a species was observed. Spatial scale and resolution is also variable and the archaeological data here represents two distinct regions, whereas catch assessments and underwater surveys are distributed more broadly but differ in their coverage of habitats and the species assemblages. These limitations can be mitigated by the Bayesian extinction analysis explained below.

A framework for identifying species vulnerable to local extinction

We used a cohesive framework that included two steps. The first step compares across data sources to identify which reef species are absent or rare in current records compared to the past. The second step is to leverage across data sources to determine the probability that those species identified in step 1 may be locally extinct while accounting for uncertainties in disparate data sources.

(2) Estimating the probability of local extinction.

We used the Bayesian extinction analysis to verify if our framework identified historically exploited species absent from modern data sources as species that show a high probability of local extinction. Bayesian extinction analysis addresses whether a species is rarely sighted, or is actually locally extinct. It does this by computing the probability of a species being extant using sightings data from disparate data sources (S1 Datasheet) [36]. Following the approach derived by Lee [36], the analysis accounted for the small sample size, survey effort, and the uncertainty of sightings in different types of data sources. To implement the Bayesian model, we derived estimates for the priors (P(XT), P(E)) and uncertainty around the priors, for detectability for definite sightings (d₁), detectability for uncertain (d₂) sightings, false detectability for uncertain sightings (f₂) and survey effort (s). We present results for the grouper Epinephelus coioides to describe the approach, which infers extinction from sighting records [36].

Two priors were derived for this method: the prior probability of seeing the species, given it is extant given the sighting record P(XT) and a prior belief that the species went extinct in a particular year after a sighting P(E) [36].

T_N is the number of years from the first to the last sighting and T is the number of years from the first sighting to the year when the extant estimate was made. For each species, we extracted sightings data from the archaeological remains, naturalist’s species lists, catch assessments records, and underwater surveys to reconstruct a time series of sightings (Fig 2). When the specific year was not available, we used the last year of the time period when the occurrence was noted. The equations allow for uncertainty around the prior so the estimate lies in the center of each range where δ = 0.1. For E. coioides, we would have and

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Sightings of historically exploited species reconstructed over time using a suite of disparate data sources.

The colour of the line represents from where the coral reef species data were sourced from: archaeological remains (black), naturalists species lists (purple), catch assessment records (red) and underwater surveys (blue).

https://doi.org/10.1371/journal.pone.0211224.g002

Detectability is for definite sightings (d₁) and was based on the number of definite sightings (S) that occurred during the time that the species was known to be extant (T_N). Definite sightings are fish bones from archaeological records.

In this case of E. coioides, at least one set of fish bones was obtained in 1300 and 1500. For the E. coioides, the detectability for specimens (d₁) is

There is approximately 0.01 probability of obtaining a specimen of E. coioides, if it is extant, since the specimen was only recorded twice in the 200 years it was known to be extant. To account for uncertainty in the estimate, detectability (d₁) lies in the center of the chosen range, that d₁ = [0,0.02].

Detectability for uncertain sightings (d₂) was based on the probability of uncertain sightings being true.

There were three uncertain sightings in 1882, 1996 and 2012. For this step of the analysis we assessed the uncertainty of sighting a species based on the sighting type. To achieve this, we used Birdlife International scoring criteria that have been used on a range of taxa including mammals, fish, and birds [36, 53–55]. The scoring criteria for the various types of uncertainty of each sighting were based on the quality of evidence and circumstance of the record as follows: (1) catch (0.90–0.95 reliability) and (2) underwater surveys and naturalist’s species list (0.80–0.90 reliability).

The 1882 uncertain sighting record is from the naturalist’s species list and has a sighting probability of 0.80–0.90 of being true. We considered the upper and lower limits, S_U and S_L respectively, of the probability of the sighting being true_. Thereby, for every ten sightings of the E. coioides, which was only found in the catch assessment records, eight are considered true. For the E. coioides the detectability for uncertain sightings (d₂) is

The 1996 and 2012 uncertain sighting records are from the catch assessment records and have a sighting probability of 0.90–0.95 of being true. We considered the upper and lower limits, S_U and S_L respectively, of the probability of the sighting being true_. Thereby, for every ten sightings of the E. coioides, which was only found in the catch assessment records, nine are considered true. For the E. coioides the detectability for uncertain sightings (d₂) is

The false detectability for uncertain sightings (f₂) was based on the probability of uncertain sightings being false.

We derived the false detectabilities for the 1882 uncertain sighting record (f₂). Again, we considered the upper and lower limits, S_U and S_L respectively, of the probability of the sighting being true and as before, for the lower limit for every ten sightings, eight are true. In which case, after extinction one would expect the remaining two sightings to be false, that is f₂ = d₂ –d₁. For E. coioides, the false detectability for uncertain sightings (f₂) is

We derived the false detectabilities for the 1996 and 2012 uncertain sighting records (f₂₎. Again, we considered the upper and lower limits, S_U and S_L respectively, of the probability of the sighting being true_. and as before, for the lower limit for every ten sightings, nine are true. In which case, after extinction one would expect the remaining sighting to be false, that is f₂ = d₂ –d₁. For E. coioides, the false detectability for uncertain sightings (f₂) is

We then assessed the likelihood of sighting a species based on the survey effort (s). Survey effort (s) accounted for both survey detectability (d_s1, d_s2), i.e. the probability of a sighting during a survey given that the species is extant, and survey false detectability (f_s1, f_s2), i.e. the probability of a false sighting during a survey given that the species is extinct. The likelihood of sighting a species based on the survey effort was as follows: (1) for catch assessment, a 0.70–0.90 detectability (d_s1, d_s2) if extant and 0–0.70 false detectability if extinct (f_s1, f_s2), (2) for naturalist’s species lists, a 0.40–0.60 detectability if extant (d_s1, d_s2) and 0–0.40 false detectability if extinct (f_s1, f_s2), and (3) for underwater surveys, 0.60–0.80 detectability if extant (d_s1, d_s2) and 0–0.60 false detectability if extinct (f_s1, f_s2). Survey effort values are based on the extent of the species occurring, quality of observers and the frequency, temporal and spatial coverage of the survey (S1 Fig). In this study three of the data sources could be quantified by survey effort.

We applied this method to obtain the probability that species evaluated under our framework as threatened with local extinction were, in fact, still extant in 2015. We also applied this method to those species we considered extant but too rare to obtain a probability that they were extant today. The year 2020 was used as the year of analysis, as the model failed in cases where we sought the probability that a species was extant in 2015 when the sighting input was 2015. We entered our range of estimates into Lee’s [36] numerical tool. We showed the implementation of the E. coioides example in a screenshot of the Excel spreadsheet (S2 Fig). The example shows the inputs and outputs of the model and how the p-value was determined. The result is a p-value, a statistically assigned probability that a species is still extant, given the observed data. The lower the p-value, the less likely a species is still extant.

Comparing extinction risk methods

We compared the results of our framework to identify fish species threatened with local extinction with those of three other established extinction risk methods. The extinction risk methods were functional vulnerability [56], vulnerability to fishing [23], and the International Union for the Conservation of Nature (IUCN) Red List status.

Functional vulnerability is the decrease in functional diversity following the local extinction of several species from the same functional group [57]. To assess functional vulnerability we classified species into functional groups defined in Wilson et al. [56], using existing literature and Fish Base [49]. Functional groups directly differentiate among groups of species based on trophic guilds, dietary specialization and life history characteristics. We used the number of species threatened with local extinction per group as a simple surrogate for functional vulnerability.

Vulnerability to fishing is the assessment of the intrinsic extinction vulnerabilities of a species to fisheries exploitation. To assess vulnerability to fishing, we classified species into vulnerable groups as defined by Cheung [23] using the existing literature, Fish Base [49] and expert knowledge. Vulnerability to fishing uses a fuzzy logic expert system that takes into account eight life-history and ecological characteristics (body size, longevity, age at maturity, von Bertalanffy growth parameter, natural mortality, fecundity, spatial behavior, geographic range) that makes species vulnerable to fisheries exploitation. Species were placed into categorical groups based on level of vulnerability to fishing from the estimate produced from the fuzzy logic system [23].

The IUCN red list assessment is a global assessment of a species risk of extinction and is widely accepted as the definitive index. The IUCN approach is to assess a species’ risk to extinction using five quantitative criteria to establish the Red List of Threatened Species [58]. These criteria are used to classify species into different levels of extinction risk such as endangered, vulnerable or least concern [58]. Species are assigned a status or category of extinction risk based is based on a combination of extinction theory, threatening processes and species differences [59]. All Red List status data are from the IUCN Red list website.

We classified the species our analysis found as threatened with local extinction into their respective functional group, vulnerability to fishing groups and IUCN Red List status using data available in FishBase [49] and IUCN Red list. Each respective extinction risk method was used to determine a level of extinction risk for each species. We compared our prediction for the species identified as potentially locally extinct from the Bayesian extinction analysis with the extinction risk assigned under the three other extinction risk methods.

Vulnerable species

At-risk species often elude detection because of the lack of species monitoring programs [59]. Surprisingly, our study identified a large number of well-known and conspicuous, rather than rare or inconspicuous, species as at-risk. Many of these species were identified as at risk as they were not sighted recently in fish transects and fish landings (Tables 1 and 2). A number of species identified by our method as threatened were commercially important carnivores. Additionally, piscivores and macro-invertivores are among the most vulnerable functional groups because they tend to be large bodied and exhibit long generation times [19, 24]. Finally, many carnivores are attracted to fishing bait, which can make even non-target or by-catch carnivores vulnerable to overexploitation [13]. The loss of carnivores found here reflects declines reported in other coral reefs and ecosystems [30, 33, 60, 61].

A number of the threatened species belong to the carnivorous family of Epinepheline groupers and Haemulidae sweetlips. Epinepheline groupers represent a valuable fishery resource that are inherently vulnerable to collapse due to fishing pressure [62,63]. One of their attributes that predispose them to intense fishing and population collapse is the formation of predictable spawning aggregations that fishers frequently learn to exploit. Sweetlips are less studied but are considered as vulnerable because both families share similar biological attributes such as large body size, extended longevity, late sexual maturation, and aggregating behavior, which correlate positively with extinction vulnerability [23, 26, 64]. A number of the other identified species are more difficult to classify in terms of the specific traits that make them vulnerable at the family level and may require more investigation to better understand the specific causes. Sweetlips and groupers are more easily observed and counted than some identified species and their higher abundance in fisheries closures indicates their vulnerability to even moderate levels of fishing [65, 66].

Limitations of the method and data

Our study highlights some of the limitations in our methods and data for detecting the possible local extinction of species. The disparate data collection methods have different levels of reliability, sampling effort, and inherent biases [67]. Temporal resolution, for example, varied greatly among our data sources. Despite covering a 1,250-year time period, there were time and data gaps of 280 year (1500–1780) and 90 year (1900–1990). Spatial resolution varied among data sources. The archaeological records were confined to two sites and only represented captured species (Fig 1). In contrast to the archaeological records, the Kenyan contemporary catch assessments and underwater surveys were intensively and extensively sampled, but only for two decades.

Archaeological data were useful for exposing a number of threatened species, but cannot fully evaluate the extinction status of species arising during the contemporary period of intense fishing and climate change. The archaeological data had fewer species than the contemporary sources and there is now a new set of species that cannot be evaluated with historical data. For example, we did not find evidence that vulnerable functional groups such as bioeroding parrotfish (Scarine in the Labridae) were threatened. While some may argue that they have the ability to resist local extinction due to short generation times [66], we believe that many well-known species were not exploited during the historical period (750-1500CE) and cannot be evaluated using archeological data. Effective capture of parrotfish was made easier by the onset of modern fishing methods, including the use of scuba, snorkeling, and spear guns. If so, failure to detect vulnerable parrotfishes exposes a weakness of relying on archaeological data. In these cases, recent catch assessment records are needed to evaluate the status of parrotfish and similar species. Consequently, species subjected to recent exploitation will need their vulnerability evaluated by other methods, including life history considerations [23, 66], but also the need for continued monitoring of these species.

Factors affecting extinction risk, such as fishing intensity and climate are among influential factors not accounted for in this method. Past environments may have differed from current conditions and therefore influenced species composition. The archaeological data, for example, included periods during the Medieval Warm period when temperatures were increasing [31]. The modern assemblage was also likely to be influenced by warm and intensified climate, which alters species composition and vulnerabilities to extinction [5, 19,22,68,69]. Graham et al. [22], for example, noted that species differed in their vulnerabilities to climate disturbances and fishing intensity; small-bodied coral-dependent species were sensitive to climate disturbances while large-bodied and carnivorous were sensitive to fishing. Clearly, there is a need to further investigate the role that these changes and biases played in evaluating species vulnerabilities to extinction.

We are taking a precautionary approach in designating the 15 Kenyan reef species as threatened with local extinction rather than locally extinct. Incorrectly classifying a species as locally extinct can occur due to taxonomic uncertainty, observation error and process error [70]. Misclassification of extinction risk could lead to a false alarm, incorrectly classifying a species as being a conservation concern. Absence of a Kenyan species from the contemporary data sources does not necessarily reflect absence of a species from the region. Absence may represent a change in the behaviours of the fish or fishers’ records. Fishing effort, gear technology, efficiency, targeting, management, and market behaviours have changed and may have influenced presence/absence records [6,40,43,71–73]. However, the increasing fishing pressure and broad exploitation of reef fishes, irrespective of species and size, in modern Kenya makes these possibilities less likely [42].

Further evidence is required before classifying the Kenyan reef species as locally extinct. Further sources of uncertainty, natural variability and observation error, need to be accounted for. While contemporary catch assessment and underwater survey data were plentiful, observation error remains an uncertainty. Species absence needs to be verified by unexplored sources. For example, non-traditional data sources such as expedition reports, cookbook recipes and fisher knowledge have revealed the local and regional extinction of species [74,75]. Another source of uncertainty that needs to be accounted for is natural variability over time and space. Reconstructing the direction and rate of change in abundance provide a quantitative estimate of the decline and further evidence of local extinction [76–78].

Despite limitations, we demonstrated that combining data sources, using a two-step framework involving Bayesian extinction analysis, improved the detection of possibly locally extinct species. The archaeological data used greatly expanded the temporal depth and uncovered a number of threatened species not previously identified.

Limitations of existing extinction risk assessments

Our study showed that limitations are evident in existing extinction risk methods. The IUCN Red List is the definitive index of extinction risk and has assessed a 7563 marine fishes [70]. However, the IUCN Red List assessments examine extinction risk at a global scale but apply the extinction risk independent of local factors. For example, three Kenyan species, which we identified as threatened with local extinction, were listed by the IUCN as “Least Concern”. The justification given was that despite local declines in certain areas of their range these species are common and abundant in other parts of their range [79]. This approach overlooks the possibility of local extinction.

Another limitation of the IUCN Red List assessment is the level of quantitative data required to carry out the assessment. This requirement is not always achievable, particularly for non-charismatic species or species with low commercial value. For these species records can be scarce. Our study identified three species as threatened with local extinction that the IUCN Red List categorised as “Data Deficient”. These species could go extinct before sufficient data is available for the IUCN Red List assessment.

Furthermore, the IUCN use a category “Not assessed” for species that haven’t been assessed for extinction risk. Six of the species we identified as threatened with local extinction are included in this IUCN category. There is a considerable risk that some of these six species will disappear before the IUCN can carry out an extinction risk assessment. Limitations to IUCN’s resources are understandable given the scale of their task; however relying solely on the IUCN Red List could result in these species going extinct without formal protection or recognition of their threatened status.

Similar to the IUCN Red List limitation concerning global level assessments, the vulnerability to fishing assessment method fails to reflect local differences in a species extinction risk, such as vulnerability to climate change [22]. Furthermore, recent work has shown that spatial variation in fishing needs to be taken into account when using life history traits [80]. These limitations of the vulnerability to fishing assessment method do not account for differential responses in species extinction risk at a local scale [81–83].

The other method, functional vulnerability, simplifies the species into groupings. While this is a useful indicator at an ecosystem level, it fails to produce a threatened list at the species scale [22]. This presents a challenge where incorporating functional vulnerability findings into conservation measures as these measures are typically set at a species level.

Value and application of the framework

Our approach has advantages over existing extinction risk assessments as it can be used for data poor species and estimates probability of extinction at a species level. Our Kenyan case study highlights the importance of exploring and incorporating historical and disparate data sources. First the archaeological data produced a list of unaccounted for species, species that are no longer present in contemporary records. Secondly, extracting heterogeneous occurrence data from disparate sources enabled a reconstruction of species absence/presence over time overcoming the IUCN’s issue regarding minimum quantitative data requirements. The naturalists’ species list verified the occurrence of historically exploited species by taxonomic specialists between 1712 and 1940. Thirdly, the reconstruction of species occurrence showed the decline and possibly the local extinction of historically exploited species from Kenya’s reefs today.

The Bayesian extinction analysis proved a suitable approach for data poor species, i.e. when certain sightings are rare. Our framework succeeded in comparing heterogeneous data from disparate sources and accounted for uncertainty, in a demonstrable way. The Bayesian extinction analysis proved an effective model by explicitly accounting for several sources of uncertainty. Specifically the model accounted for the varied observation error among disparate data sources. The observation errors were as follows: true and false sightings and survey effort. The model accounted for the variability among types of sightings and survey effort. Another source of uncertainty that was accounted for was uncertainty around the prior, the prior belief that the species still occurs.

The framework used should be of value to conservation management efforts. The assessment of local extinction was carried out at a species level at a local scale which will allow easier application into species management policies and regulations. Furthermore, the framework proved an inexpensive and effective for identifying threatened species when historical data, particularly archaeological records, were available. Such assessments are especially critical for data- poor areas where inexpensive management advice is needed the most. Combining these data and applying Bayesian extinction risk methods can strengthen other species assessment efforts.

Local extinction rates are increasing among marine species [1,2] and here we have added fifteen new species potentially disappearing from Kenya’s reefs. Some of these species are likely to play important roles in ecological processes and services that may erode as they disappear. The first step towards avoiding local extinction is providing status information at a scale where decision-makers and managers can intervene. This is frequently at the local to national fisheries levels where policies can introduce size limits, protection of breeding seasons and locations, and outright bans to protect species most at risk.