Assessing the global impact of targeted conservation actions on species abundance

In recent years, vertebrate population abundance has declined at unprecedented rates (WWF 2020). In response, targeted conservation measures – such as breeding programs or species-specific habitat management – have been applied to halt population declines, aid population recovery, and reduce and reverse the loss of biodiversity (Salafsky et al. 2008; Bolam et al. 2020). Until now, assessments of conservation actions have focused on the extent to which they reduce extinction risk, impact populations within protected areas, or increase the global area of land under protection (Hoffmann et al. 2010, 2015; Barnes et al. 2016; Maxwell et al. 2020; Bolam et al. 2020; Grace et al. 2021a). Here, we record and analyze conservation actions for 26,904 vertebrate populations from 4,629 species, to measure the impact of targeted conservation on vertebrate abundance. Using a counterfactual approach to represent population trends in the absence of conservation, we demonstrate that targeted actions have delivered substantial positive effects on the abundance of recipient vertebrate populations worldwide. We show that, in the absence of conservation, a global indicator of vertebrate abundance would have declined even more. Positive population trends were associated with vertebrate populations subject to species or habitat management. We demonstrate that targeted conservation actions can help to reverse global biodiversity loss and show the value of counterfactual analysis for impact evaluation – an important step towards reversing biodiversity declines.


Introduction 35
Alterations to global ecosystems have caused widespread declines in biodiversity worldwide (Díaz et  Ideally, the impacts of conservation interventions would be assessed using experimental designs 47 Table 1 Types of population abundance data that meets the data standards for tracking trends in the 116 abundance of species populations (Accepted) and that does not meet the data standards (Not 117 accepted) 118 119 Accepted population abundance data Not accepted population abundance data Full population counts Occupancy (unless it is used specifically as a proxy for abundance) Estimates (e.g. population size estimated from measured parameters) Data from experimental observations Densities (including converted camera trap data) Survival rates Indices Recruitment data e.g. number of eggs or young Proxies (e.g. breeding pairs, nests, tracks) Catch or hunting data with no measure of effort Measures per unit effort (e.g. fish caught per net per hour) Data where method has changed (unless corrected for) Biomass (e.g. spawning stock biomass) Opportunistic sighting data Samples (e.g. where a proportion of the population is regularly monitored In the LPD, each population is stored with a unique ID and contains annual population data alongside 121 additional information that covers eight broad categories relating to the species or the population. 122 The first category is 'Base information', which contains information about the source, the year that 123 the source was published or accessed, the reason for data collection, if the data overlaps with other 124 populations, and the reference for the data. The second category is 'Taxonomy'. Here, taxonomic 125 information is stored such as the common and Latin species name, Class, Order, Family and Genus. 126 The taxonomic authorities used are: 127 The sixth category contains 'Protected Area' information that describes if the population is inside of 150 a protected area, and the type of protected area if so. The seventh category covers 'Management' 151 aspects, which indicates if the population is managed, the type of management (see Table 1  trained personnel record the 'Management', 'Threat' and 'Reasons for increase' information from 158 the original sources using a set of guidelines. This aims to reduce potential bias, but there is likely 159 still to be some individual interpretation of the information in the data source. 160

Data selection, management data and coding 161
We extracted data for every population in the LPD (LPD 2020) -26,904 populations representing 162 4,629 species from 11 taxonomic classes. For each of these populations we used the additional data 163 stored in the LPD indicating whether a population was managed, utilised, located inside a protected 164 area, or likely benefitting from conservation action (Table 1) Scenario two assumed that, in the absence of conservation, a conservation targeted population 212 would show similar trends in abundance to any other population from the same species and country. 213 Relative to the first scenario, exact matching on country and species reduces the sample size but 214 imposes stricter conditions in terms of similarity between the treatment and control populations. 215 Scenario three assumed that, given no conservation, a targeted population would have developed 216 similarly to a non-targeted population from the same taxonomic class and political region. In 217 addition, we matched each population on the populations' year of first record and time series length, so that each was matched to a single non-targeted population from the same class and 219 region, and with similar time-series characteristics. We did this using a combination of exact and 220 propensity score matching. Propensity score matching uses logistic regression to predict a 221 probability of receiving treatment, which in this case is targeted conservation, given a set of 222 observed predictors ( The 95% confidence intervals were generated with 10,000 bootstrap replicates across species-level 245 annual growth rates (Collen et al. 2009). 246 Our approach is similar to that used to create the LPI, except that no taxonomic or other weighting 247 was applied, so that each species was weighted equally. The LPI aims to characterize global trends in 248 vertebrate populations in a balanced fashion, and therefore applies weighting to account for 249 taxonomic and geographical inequalities in the sampled data (McRae et al. 2017). However, we set 250 out to estimate the impact of conservation actions on target populations using a matching approach 251 which similarly corrects for bias. Furthermore, we created the four counterfactuals from matched 252 subsamples of the LPD. The weighted approach is not suitable because these samples are much 253 smaller than the actual LPD, and not randomly selected. Therefore, we did not apply the LPI 254 weighting, as this could potentially exacerbate the effect of any selection biases in ways that would 255 be difficult to interpret.

Country
The country (or countries) that the population occurs in from the list. Marine data are allocated a country if it is within its EEZ, or as International Waters. Multiple countries are selected in order of proportion of the population it represents starting with the greatest.

First year of observation
The first recorded year with an abundance estimate for a given population Time series length The number of years from first to last population abundance estimate Managed A population that receives targeted management (some of which involves sustainable use). This is usually to promote recovery in a population or can incentivise it's use for conservation.   Table 2. For this 283 subsample, we replaced the observed abundance estimates with a constant, thus assuming that 284 trends for these populations would have at least remained stable without conservation. The index 285 was then calculated from all of the available LPD population data, but with constant annual 286 abundance estimates for the selected subset of populations. 287 Finally, we calculated a population index where the impacts of targeted and collateral conservation 288 were excluded (Hoffmann et al. 2015). This population index, excluding the impact of both targeted 289 and collateral conservation, was calculated similar to the second index but in addition excluding the 290 impact of collateral conservation. By collateral, we mean that a population could have benefitted 291 from conservation without being specifically targeted, which we defined as having increasing 292 population trends within a protected area, while not being specifically chosen for any of the targeted 293 actions. This index was therefore calculated using the full LPD data but assuming stable trends for 294 the same populations as in the second index and additionally populations without targeted 295 conservation, but which were inside a protected area and had a reason for population increase 296 recorded. Population indices were calculated using the rlpi package ) without 297 applying taxonomic or geographical weighting. To visualize the impact of conservation, we plotted 298 the difference between the reference population index, calculated using all of the population trend 299 data in the LPD, and the three potential scenarios that represent the reference index without the 300 likely impact of conservation.

Mixed model 302
We compared the effects of the seven primary targeted conservation actions on abundance trends 303 (the log of the summed rate of population change) using a mixed model framework (McRae et al.  304 2020). The rlpi package generates a matrix of annual rates of population change for each population. 305 We summed these rates into a logged value of total change for each population. To test the effects 306 of conservation actions in general, and of the different types of interventions, two separate models 307 were specified. 308 In the first, conservation actions were aggregated into a single fixed effect binary variable (1 = 309 targeted or 0 = not targeted by conservation) to estimate the overall effect of actions regardless of 310 action type. 311 In the second, we specified a binary variable for each of the seven main conservation actions. This Conservation + Class + (1|Family/Binomial) + (1|Location); Model two: lambda_sum ~ 0 + Utilised + 319 ts_length + land_water_protection + land_water_management + species_management + 320 education_awareness + law_policy + incentives + external_capacity + research + Class + 321 (1|Family/Binomial) + (1|Location)). 322 collateral conservation remained stable (Fig 4, Fig S7). for species were highly associated with population increases, suggesting a particularly strong effect 386 of actions within these three conservation categories (Fig. 5). Research actions were negatively 387 associated with population trends. Longer population time series were more likely to have 388 increased, while utilized populations did not display a clear trend. 389  Table S1 for 392 parameter estimates, standard errors and t values.