Monitoring extinction risk and threats of the world’s fishes based on the Sampled Red List Index

Global biodiversitytargets require us to identify species at risk of extinction and quantify status and trends of biodiversity. The Red List Index (RLI) tracks trends in the conservation status of entire species groups over time by monitoring changes in categories assigned to species. Here, we calculate this index for the world’s fishes in 2010, using a sampled approach to the RLI based on a randomly selected sample of 1,500 species, and also present RLI splits for freshwater and marine systems separately. We further compare specific traits of a worldwide fish list to our sample to assess its representativeness. Overall, 15.1% of species in the sample were estimated to be threatened with extinction, resulting in a sampled RLI of 0.914 for all species, 0.968 in marine and 0.862 in freshwater ecosystems. Our sample showed fishing as the principal threat for marine species, and pollution by agricultural and forestry effluents for freshwater fishes. The sampled list provides a robust representation for tracking trends in the conservation status of the world’s fishes, including disaggregated sampled indices for marine and freshwater fish. Reassessment and backcasting of this index is urgent to check the achievement of the commitments proposed in global biodiversity targets.


Introduction
In 2020, the UN Decade of Biodiversity came to its culmination, requiring a stocktake of the world's progress towards the Aichi Targets, set by the Convention on Biological Diversity in 2010. Following our failure to reach the previous 2010 target to achieve a signi cant reduction in the rate of biodiversity loss by 2010 1 , many of the Aichi Targets are also not met 2 . This includes Aichi Target 12 which stipulates that "by 2020, the extinction of known threatened species has been prevented and their conservation status, particularly of those most in decline, has been improved and sustained 3 ." Threats continue to increase, resulting in declines in the abundance and distribution of species 4,5 . A recent global assessment of the state of biodiversity by the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES) estimated that up to one million species are at risk of extinction 6 . This estimate relies on inferences from the known extinction risk of species assessed by the International Union for Conservation of Nature (IUCN) Red List of Threatened Species, arguably the most shes, and other freshwater species, on the IUCN Red List. Currently, the IUCN-Toyota Red List Partnership aims to complete the global comprehensive assessment of freshwater shes .
Still, only 59% of sh have been assessed by the IUCN Red List (currently just over 20,000 species), compared to 91% of mammals, 100% of birds, 84% of amphibians and 70% of reptiles 9 . A sampled assessment of 1,500 shes was initially completed in 2010, as part of the development of the sRLI as a global biodiversity indicator 11,56 . Here, we provide the long-overdue presentation and analysis of the results. We present the rst global assessment of extinction risk for a random sample of shes, highlighting major threats impacting shes in marine and freshwater environments, and put the ndings in the context of the conservation status of other non-sh species groups. We investigate the representativeness of the sRLI sample in terms of taxonomy, biological and ecological traits, to ensure that going forward, the sRLI for shes presents a robust indicator which adequately re ects the high trait and ecological diversity of this species group. This global assessment is of particular importance since the sRLI for shes is now due for reassessment to evaluate the trends in the conservation status of the world's shes.

Status of shes
In the 1,500-species sample, 178 species (15.1% of sampled species) are threatened with extinction: 105 are Vulnerable, 40 Endangered and 33 Critically Endangered (Fig. 1a). Another 320 species (21.3%) were assessed as Data De cient (1,180 species were non-Data De cient). Thus, we obtained a lower threat estimate of 11.9%, and an upper threat estimate of 33.3%. Overall, 961 species were assessed as Least Concern, 41 as Near Threatened, and no species were assessed as Extinct or Extinct in the Wild. Just over half of our threatened species (51%) were assessed under restricted geographical range size (criterion B) and 38% under population reduction (criterion A). Only 17.4% of threatened species were classi ed based on criterion D2, while only two and four species were classi ed under criterion D/D1 (very small population size) and criterion C (small, declining populations), respectively (Fig. 1b). Recorded population trends of sampled species were mainly unknown (68%), although 11% showed decreasing populations, 20% were stable, and only three species (0.2%) showed increasing populations.
The most commonly stated threat to shes is exploitation (34.9%), especially for marine shes (44%). However, pollution is the most prominent threat to sh species (49.4% of the 178 threatened species), particularly from agricultural or forestry e uent, domestic and urban wastewater and industrial e fuents, and is particularly prominent for freshwater species (63%) (Fig. 2). Natural system modi cations such as dams and channeling in freshwater ecosystems (9%), invasive species (6.5%), and habitat loss for urban development (4.9%) were also affecting species, especially threatened freshwater species (Fig. 2).
Normalised species richness in the sample is shown in Fig. 3. Highest species richness and threatened species richness for marine shes in our sample are located in Southeast Asia, and secondarily around tropical islands in the Caribbean ocean. Data De cient species richness of our sample was highest in Southeast Asian waters and along the north and western Australian coast. Freshwater species richness in our sample was highest in Southeast Asia and the Amazon Basin, whereas threatened species are concentrated mainly in Southeast Asia. Data De cient freshwater species richness in our sample was highest in parts of Southeast and Eastern Asia.

Representativness of the Sampled List
In both the Global and the Sampled List of shes, the largest order is the Perciformes, which includes both freshwater and marine species. The next largest orders are represented primarily or exclusively by freshwater species, such as the Cypriniformes, Siluriformes and Characiformes (Fig. 4). In terms of spatial distribution, both lists show that species richness is highest in the Western Central Paci c and Northwest Paci c. In inland waters, tropical areas of Asia and South America are the most species-rich areas. In terms of habitat preference, most sh species are in pelagic and benthopelagic habitats, followed by reefassociated ecosystems, in both the global and sample lists (Fig. 4). Chi-square tests showed no signi cant differences in the proportion of species among orders, FAO areas and preferred habitats between the Global and the Sampled List (Table 1). Similarly, there are no signi cant differences in estimated biological traits (life span, generation time, trophic level and vulnerability index) comparing Global and Sampled Lists (Table 2). Table 1 Values of Chi-square distribution (χ 2 ), degrees of freedom (df) and probability values (P) comparing Global (33,112 spp.) and Sampled List (1,500 spp.) by orders, FAO areas and habitats, and considering total, and marine and freshwater species separately. Data was obtained from Fishbase (

Discussion
Our study provides a baseline of the exticntion risk of the world's sh, against which to track trends in the conservation status of sh in future. Overall, we show that 15% of sh species in our sample are estimated to be threatened with extinction and that threat is higher in freshwaters compared to marine systems. Our results con rm previous ndings on the truly alarming conservation status of freshwater shes 57,58 ; in the same way that other freshwater biodiversity is highly threatened with extinction 13,20,48 .
The overall sRLI for shes is similar to the Red List Index for birds 12 , and dragon ies and damsel ies 18 , and higher than for the remaining evaluated taxonomic groups (Fig. 1). Overall, the sRLI for marine shes is the highest of the Red List indices calculated so far (bar the historical index estimated for reef-building corals pre-1998 24 ). Our estimate of 4.7% of marine sh threatened with extinction is lower than threat levels found in other studies on extinction risk of marine shes: regional shore shes (5-9.4% 41,46 ), sharks and rays (17.4% 34 ), hag shes (12% 30 ), groupers (12% 33 ), tarpons, lady shes and bone shes (12.5% 35 ), porgies (8.6% 37 ), and puffer shes (7.9% 38 ). But we are not surprised as threat tends to be greatest in shallower waters and a random sample of marine shes will have disproportionately greater deepwater species than the shallowwater taxa that have dominated early assessment priorities. The sRLI calculated for freshwater shes is much lower and similar to the RLI for mammals and plants 19 . It is slightly higher than RLIs for other freshwater groups, with a lower estimated threat level (cray shes 48 , freshwater crabs 47 and shrimps 59 ). However, note that the sRLI protocol was not developed to accurately estimate threat levels in a species group, but to accurately detect extinction risk trends in a species group over time 11 . Thus, any threat estimates from our sample should be treated with caution and may only be broadly indicative of overall levels of threat within shes. However, work is ongoing to test the accuracy of threat estimates from sRLI samples.
The level of data de ciency in our Sampled List of shes is comparable to that found in other species groups such as cray sh 48 and reptiles 17 , and lower than that observed in freshwater crabs 47 and freshwater molluscs 20 . While Data De cient species should be considered as potentially threatened until their status can be assessed 8 , they cannot contribute to the Red List Index 10 unless their status can be predicted using trait-based methods. Reducing data de ciency is thus important to produce more robust knowledge on extinction risk patterns and RLI values in future 60 . With data de ciency in our sample highest in parts of Southeast Asia, this region would make a logical place to target to reduce DD, speci cally for marine sh for which DD currently produces wide margins of uncertainty around estimated threat levels.
Population trends were lacking for many marine shes beyond coastal areas, as most of our knowledge on the marine realm comes from coastal, intertidal or neritic habitats: for example, 73% of marine sh species assessed on the IUCN Red List occur in these habitats 9 . This is especially problematic since marine shes were predominantly assessed as threatened under criterion A (Fig. 1B), i.e. because of a population reduction over ten years or three generations. Results offered by the Living Planet Index, a measure of the trends of global biodiversity based on population trends of vertebrate species from around the world 61 , showed an average decline of around 52% for monitored marine vertebrate populations since 1970 62 , compared to 84% for freshwater vertebrate populations 5 . This suggests that the risk of population declines for those species with unknown populations trends in our sample should not be underestimated, and that we need to push efforts towards better monitoring and estimating populations.
Fishes are among the most diverse classes of vertebrates with signi cant differences between marine and freshwater environmental realms. Despite differences between realms, our results consistently show exploitation and pollution are the main threats to both marine and freshwater shes (Fig. 2). In the marine realm, overexploitation is overwhelmingly prominent in assessments of nearshore and epipelagic shes 30,34,38 . Despite low overall threat levels of marine shes in our study, in 2015 only 7% of globally assessed stocks were under shed according to the FAO 63 , and increases in exploitation pressure in future may lead to further declines in species. Safeguarding marine sh diversity needs the urgent engagement of different stakeholders to ensure the sustainability of this resource while also addressing the United Nations Sustainable Development Goals, e.g. such as SDG2 on combatting hunger and malnutrition.
The impact of human settlements and cities around aquatic ecosystems and increasing water demand have led to the degradation of freshwater biodiversity 57,58 , especially through water pollution, dams and water exploitation, river fragmentation, habitat loss, and establishment of non-native species 58 , all threats which were prominently recorded in the sRLI assessments. Rivers are highly connected linear structures 64 : they are collectors of terrestrial impacts of the landscapes they drain, conducting them downstream. Management plans therefore need to consider the unique characteristics of freshwater systems and their high connectivity 64,65 .
Our study provides the rst in-depth test of representativeness of the sRLI -including the separate disaggregated indices obtained for marine and freshwater shes-in terms of geographic, ecological and trait diversity. This is particularly important since the sRLI method at present randomly draws species from the species list; strati cation of the sample was originally considered, but was rejected as a workable strategy due to the general lack of knowledge on any of these factors prior to the assessment process 11 .
Thus far, tests have only been carried out to show that the recommended sRLI sample sizes are large enough to accurately re ect species group attributes regarding biogeographic realm, ecosystem types and taxonomy 11,16 . Representativeness is important since, for example, marine shes that are restricted to the continental shelf, and especially those that occupy shallow habitats of less than 50 m depth, have a signi cantly higher proportion of threatened species compared to marine shes that occur in waters deeper than 300 m 45 , while on the other hand, deep sea shes are often assumed to be Least Concern because of a lack of intense shing pressure on these shes, although low growth rates, late maturity, low fecundity and long lifespans of many deep sea shes make them particularly vulnerable to any level of exploitation 66 . Here, we again showed that there were no signi cant differences in the proportions among taxonomic groups, geographic regions and habitat types between the Sampled and Global List of shes, while also showing representativeness of other biological traits (life span, generation time, trophic level and vulnerability index). The Sampled List seems to not only be su ciently large to accurately detect trend direction in the extinction risk of the world's shes 11,16 , but also to be representative of the world's sh taxonomic, trait and ecological diversity.
Recent work has shown that where assessments occur every ten years, samples of 400 non-DD species may be su cient to accurately show direction of RLI trend of a group 16 . According to this, our study con rms the suitability and representativeness of calculated sRLIs for shes, including the separate disaggregated indices obtained for marine (598 non-DD species) and freshwater shes (610 non-DD species). Once completed, the data generated by the global freshwater assessment (carried out through the IUCN-Toyota Red List Partnership) can be uses to re-evaluate the representativeness of our sRLI sample of shes, especially to see whether the spatial representativeness of our freshwater sample is broadly representative of overall freshwater sh species richness patterns (Fig. 3).
In this study, we calculated the baseline sRLI for 2010, the year in which the assessments of the selected 1,500 species was concluded. The index results published here provide the baseline towards monitoring global extinction risk in this highly species-rich group, allowing us to track future changes and trends in the conservation status of the world's shes. Speci cally, with the addition of subsequent assessments this index lets us track improvements or deteriorations in the status of the world's shes, considering separately freshwater and marine species. A current reassessment would allow us to check how shes fared against Aichi Target 12, and provide a starting point for better conservation action and management for these vital aquatic resources. Many of the original assessments have already undergone (20,878 species) or are in the process of reassessment of their IUCN Red List status. As such, a rst step for reassessment of the sRLI is to collate recent assessments and update the status of those species which have undergone non-genuine changes in their assessment status in recent years (i.e. changes because of improved data rather than actual improvements or deteriorations in extinction risk status). Secondly, we need to prioritise reassessments of those species which were in threatened or Near Threatened categories in 2010, to allow in depth reassessment. As in other assessment processes, Least Concern species may be fast-tracked more rapidly through the assessment process 67 . Thirdly, in the absence of a reassessment for 2020, application of retrospective assessments to assess past extinction risk status from a present perspective [68][69][70] should be considered to derive long-term trends in extinction risk over time.
Aichi Target 12 for biodiversity has not been met 2 . Considering the existing priorities and limited conservation resources to establish an e cient reassessment of larger samples, the selected subset of species can inform current and future policy targets about trends on sh species conservation and help to allocate efforts and resources. Given that to date, shes have been largely neglected in large-scale conservation analyses, likely due to an apathetic public perception of these animals 71 where W is the category weight (category weights increase from 0 for Least Concern in equal steps to 5 for Extinct and Extinct in the Wild) for S species at time t; and N is the total number of assessed species, excluding those considered DD. Thus, RLI values can vary from 0 (all species are Extinct) to 1 (all species are Least Concern). In this way, we produced an sRLI for all sh. A recent re-visit of the sRLI sample size, analysing data for a broader set of species than in the original sRLI paper by Baillie et al. 11 , suggested that Page 13/19 200 to 400 non-DD species are su cient to accurately detect trend in RLI 16 . Thus, we also produced sRLI values for freshwater and marine sh separately.
The threats impacting each species were recorded during the Red List assessments, following the IUCN's uni ed threats classi cation scheme 77 . We summarised the frequency of threats for threatened (VU, EN, CR) and non-threatened species (LC and NT). We also analysed species population trends, which are recorded as unknown, decreasing, stable and increasing populations on the IUCN Red List 8 .
Species distribution was mapped -where possible -for all assessed species for which the distribution could be mapped (n = 1,484). For some species, speci cally DD species, distribution data was too uncertain to allow mapping. To visualise the distribution pattern of our Sampled List, we selected only those parts of a species' distribution map where the species was considered extant or probably extant, resident, and native or reintroduced 67 , resulting in 1,473 species remaining. We mapped species richness, threatened species richness and Data De cient species richness of our sample by overlaying a grid with 1° grid cells onto the respective aggregated species' distribution and summing the number of species occurring in each grid cell. We normalised species richness relative to the richest cell to derive a synthetic pattern of species richness ranging from zero (no species present) to one (highest species richness), as described in Collen et al. 13 . We created richness maps for freshwater (n = 715) and marine species (n = 799), separately. All maps were created in R Studio v. 1.2.1335 and R Studio v. 3.6.0 78 .

Taxonomic, ecological and biological trait data
To obtain a full picture of trait and ecology of the world's shes, we extracted information on taxonomy, distribution, preferred habitat and biological traits for 33,112 sh species (hereafter termed Global List) from the FishBase online database 79 . We determined the number of species in each order according to the FishBase taxonomy, and obtained the number of species for marine and inland waters per FAO Major Fishing Areas (http://www.fao.org/ shery/area/search/). We extracted the following habitat information (particular habitat preferred by each species, adapted from Holthus and Maragos, 1995): pelagic, benthopelagic, demersal, reef-associated, bathypelagic and bathydemersal, according to the glossary of FishBase 79 . We summarised the number of species in each habitat type.
We collected the following biological traits from Fishbase 79 : life span, generation time, trophic level and vulnerability index. Life span is the approximate maximum age that sh of a given species are estimated to reach, and generation time is the average age of parents within the cohort. Trophic level is the position of sh in the food chain, determined by the number of energy-transfer steps to that level 79 . Trophic levels reported in FishBase are derived from Ecopath 80 . The index of intrinsic vulnerability to sheries presented in FishBase is calculated via an expert system developed for shes that integrates life history and ecological characteristics 81 .
Subsequently, we compared the taxonomic, geographic, ecological and biological representativeness of this Global List against traits of our Sampled List, to assess whether the randomly Sampled List adequately represents taxonomic, spatial, and biological trait diversity of global shes. Global and Sampled lists of sh species were tested for differences in the number of species among taxonomic orders, among FAO areas and among habitat types, using chi-square tests 78 . To assess the representativeness of biological traits, we used non-parametric analyses because the normal distribution assumption was not met in these data sets, even after data transformation. First, two-tailed (Wilcoxon) Mann-Whitney U test was used to examine whether the medians of the two samples were different. Second, Kolmogorov-Smirnov tests were used to assess whether the distributions were equal, independently of differences in other descriptive parameters as mean or variance 78 .