Habitat loss estimation for assessing terrestrial mammalian species extinction risk: an open data framework

Introduction

Habitat loss and fragmentation significantly impact biodiversity, leading to critical population declines and affecting long-term biodiversity conservation (Fahrig, 2003; de Barros Ferraz et al., 2021; Santos et al., 2019a). Approximately 100 million hectares of rainforest were lost between 1980 and 2000 (IPBES, 2019), and the remaining are strongly affected by anthropogenic activities (Haddad et al., 2015; Leblois, Damette & Wolfersberger, 2017). Decreasing the amount of available habitat may directly affect critical biological processes, such as resource availability (Ryser et al., 2019), dispersal (Cote et al., 2017), pollination (Pavageau et al., 2017) and gene flow (Dixo et al., 2009; Moraes et al., 2018). Additionally, these activities contribute to the increased accessibility of hunters to natural/remote areas, resulting in higher direct exploitation of species (Gallego-Zamorano et al., 2020), and increasing the risk of invasive species introduction. Thus, habitat loss contributes to population reduction (Harfoot et al., 2021; Heinrichs, Bender & Schumaker, 2016; McCormack, Ghani & Ferguson, 2019).

Globally, 28% of all assessed species are threatened with extinction, corresponding to more than 40,000 species, of which 26% correspond to mammalian species (IUCN, 2020). Notably, in addition to being threatened by habitat loss, disturbance, and fragmentation (IPBES, 2019; Bogoni, Peres & Ferraz, 2020; Canale et al., 2012), these species have suffered from anthropogenic impacts and have been victims of several human-wildlife conflicts (Adhikari et al., 2022; Desbiez, Oliveira & Catapani, 2020; Vanak & Gompper, 2010). As mammalian species play a critical role in ecosystem functioning (Jorge et al., 2013; Magioli et al., 2021; Rodrigues et al., 2019), the continuous population reduction of diverse mammalian species over recent decades has directly impacted ecosystem dynamics (IUCN, 2020).

Reducing species extinction risk—particularly for highly threatened taxa—is a global priority and has been featured in various international agreements for biodiversity conservation, such as in the Convention on Biological Diversity Aichi Target 12 (either in its original form: https://www.cbd.int/sp/targets/, as well as in its post-2020 draft framework: https://www.cbd.int/doc/c/abb5/591f/2e46096d3f0330b08ce87a45/wg2020-03-03-en.pdf) and the United Nations Sustainable Development Goal 15 (https://sdgs.un.org/goals). In order to assess the extinction risk for known species, the International Union for the Conservation of Nature (IUCN) conducts global conservation status assessments using a well-established methodology and rigorous theoretical and analytical data (IUCN, 2022). Based on specific parameters and definitions (e.g., population sizes and trends, geographic range and occupancy, population reduction, and generation time), the IUCN’s assessment process has established the globally used extinction risk categories (e.g., Critically Endangered, Least Concern, and Data Deficient), criteria (Criteria A to E) and assessment methodology (IUCN, 2022, 2012).

Although there is a lack of primary biological data for many taxa (IUCN, 2022), the IUCN’s methodology considers a variety of data types from many sources of varying quality. For example, the IUCN’s Criterion A—widely used to assess mammalian species’ risk of extinction—highlights taxa that have experienced intense population reductions either in the recent past or projected for the near future. However, direct observations of population reduction are not available for all taxa. Thus, other data types—such abundance indexes, declines in habitat quality, levels of exploitation, or effects of pathogens—may be used for estimation, inference, or suspicion of population reduction. Habitat loss estimates within a taxon’s range, area of occupancy, or extent of occurrence may be used as an essential tool for extinction risk assessments and as a proxy of population reduction, especially for species that are strictly related to or dependent upon their suitable natural habitat (IUCN, 2022).

Brazil is the fifth largest country in the world, with the greatest global biodiversity and is responsible for roughly 14% of the world’s biota, even though many species are yet to be discovered and cataloged (Costa et al., 2005). There are around 770 mammalian species (Abreu et al., 2021) in Brazil, accounting for the second-greatest mammalian diversity in the world and the largest for some mammalian orders, such as Primates (Mittermeier et al., 2013) and Xenarthra (Santos et al., 2019b). Despite housing this great level of mammalian biodiversity, exceptional levels of deforestation affect mammal species in all Brazilian phytogeographic domains (Bogoni, Peres & Ferraz, 2020; Magioli et al., 2021), including the Amazon, which presents the greatest species richness; the Atlantic Forest, which presents increased rates of endemism (Mittermeier et al., 2011); and the Pantanal, which has lost approximately 17 million vertebrates over the past 2 years due to wildfires (Tomas et al., 2021). An area of approximately 82 Mha of natural habitats was lost in Brazil between 1985 and 2020, including a notable increase in deforestation during recent years in almost all phytogeographic domains, directly affecting multiple taxa (Souza et al., 2020; MapBiomas Project, 2021). The future trend is uncertain, with or without the compliance of Brazilian public policies such as the Brazilian Forest Code (Brasil, 2012). In a modeling study, Soterroni et al. (2018) presented a spatially explicit assessment of Brazil’s 2012 Forest Code through the year 2050 with a detailed representation of the agricultural sector and spatial land-use change. Their results showed that, if rigorously enforced, the Forest Code could prevent a net loss of 53.4 Mha of forest and native vegetation by 2050, 43.1 Mha (81%) of which are in the Amazon alone.

Throughout the years, the high number of threatened species and unassessed taxa indicate the necessity for more efficient extinction risk assessments, including the production of new data and metanalyses to support IUCN categories and criteria application. Since land-use changes very fast globally, automatic methods with satellite images make the assessment of habitat loss more dynamic and efficient. Moreover, these methods may be a promising tool for supporting extinction risk assessments and in guiding decision-making (de Barros Ferraz et al., 2021). Furthermore, performing metanalyses with open data would ultimately contribute to greater reproducibility, replicability, and data cycling—the basis of the open science philosophy (Gallagher et al., 2020)—within the assessment process. South America currently depends on the MapBiomas Project, a collaborative network formed by NGOs, universities and technology startups (MapBiomas Project, 2021). The MapBiomas project has produced land use and land cover maps (LULC) since 1985—in addition to other products such as fire and deforestation reports. Within this project, LULC maps are updated yearly and include all Brazilian and many South American phytogeographic domains (Amazon, Chaco, Pampa and Atlantic Forest), in addition to recently mapping Indonesia. Along with the Google Earth Engine technology, a cloud-based platform with a vast satellite image collection catalog (Gorelick et al., 2017), the MapBiomas LULC maps may be used to produce land cover change estimates at a faster and more efficient rate, without requiring local supercomputers. Therefore, we used a series of land-cover maps from the MapBiomas Project to generate a cloud-based, open data framework—using the Google Earth Engine platform (GEE)—that allows habitat loss estimation across different time scales.

This study established a systematic approach to estimating habitat loss for terrestrial species using a time series of land use classification maps to generate necessary, reliable, and prompt information in order to subsidize the application of the IUCN categories and criteria for extinction risk assessments. The novelty here is that we created a method to obtain an accessible and standardized analysis of habitat loss to assess species’ risk of extinction, directly applicable to South American and Indonesian species but which may also be adapted for other tagged landcover Datasets on Earth Engine. The data generated is of great value to the process of qualifying species under the IUCN criterion A, which is based on population reduction estimates for the past, present, and/or projected for the future, considering the sub-criterion c (habitat loss), thereby strengthening the results, as well as informing conservation decision-making.

Materials and Methods

Habitat loss estimate

We used each taxon’s geographical distribution, generated within the Brazilian species extinction risk assessment, to estimate habitat loss. This area consists of a polygon delimited by occurrence records (obtained through literature or personal communication with the assessors), and adjusted with the known biogeographical limits for the distribution of each taxon (e.g., rivers or relief), according to the available literature or expert knowledge, when possible (ICMBio, 2022).

We used the MapBiomas Project—Collection 6 (2021), which has produced LULC data from 1985 to 2020, to estimate habitat loss. We first identified the LULC classes that constitute each species’ suitable habitat; we then remapped the pixels within the distribution ranges to either 0 or 1 values, representing the non-habitat (0) or habitat (1) classes. Finally, after this pixel reclassification, we estimated habitat loss using the following equation:

$\mathrm{H}\mathrm{L}=[1-(\mathrm{H}\mathrm{L}\mathrm{Y}/\mathrm{H}\mathrm{F}\mathrm{Y})]\ast 100$where HL = percentage of habitat loss, HFY = total habitat area of the first year of the time window (which is the sum of all pixels equal to 1 in the first year’s map), HLY = total habitat area of the last year of the time window (sum of all pixels equal to 1 in the last years’ map). We established 1985 as the first year of the time frame for the taxa with 3GL greater than 35 years as this is the oldest spatial data provided by the MapBiomas Project (2021). Therefore, habitat loss might be underestimated for these taxa.

We developed the script for the habitat loss calculation using Google Earth Engine (GEE), and the code is available at: https://code.earthengine.google.com/?accept_repo=users/maributti/HabitatMammalsBR and https://github.com/MariellaButti/GEE-HabitatLossMapbiomas. This script allows the user to access the MapBiomas Project database online. Then, in order to calculate habitat loss, the user may select both the initial (ano_ini) and final (ano_fin) years—corresponding to the time window (3GL) for the target species, the scale (minimum = 30 m, which is the default resolution of MapBiomas Project), and the corresponding habitat(s) which is suitable for the target taxon (according to the MapBiomas Project Classes ID code). Additionally, it is possible to select upper and lower elevation limits to apply a mask to the habitat layer based on this range. Therefore, all habitat classes outside the selected range are reclassified as non-habitat (0). We used the NASA-SRTM (Shuttle Radar Topographic Mission) dataset (Jarvis et al., 2008) to implement this mask layer. In this study, we set up the limits out with Brazil’s altitudinal range (−1,000, 4,000) as no species have elevation boundaries (Fig. 1).

We uploaded the taxon range file in vector format to perform the analyses on the GEE platform. The vector file must contain: .shp, .shx, .dbf and .prj extensions. Our habitat loss calculation script also makes it possible to draw a random polygon directly on to the map displayed on GEE in order to calculate the habitat loss in a specific area. Each taxon was analyzed separately, and outputs can be found in File S1. After running the script model, three data files containing the results are generated and are available for download: a CSV file, with all the requested information, and two raster files in .tiff, with the spatialized habitat of both the first and last years, corresponding to the taxon time window. It is also possible to save the files on Google Drive, in the cloud storage, or in the GEE Asset, as such, there is no need to save the files on the user’s local computer.

Results

We successfully developed an open script that estimates the habitat loss for each species of interest using a series of LULC maps from the MapBiomas Project through the GEE platform. We tested the application of the habitat loss calculation script on 190 mammalian taxa, covering different orders, habitat types, and conservation statuses. The mean habitat loss was 6.46% (±8.23%), but with a high data variation, where the maximum and minimum values varied from 34% to −19.92% (Table 1). When observing the three target classes (Conservation Status, Conservation Category, and Order), the mean habitat loss followed the general trend of a high degree of variation, in terms of habitat loss (Table 1).

When comparing Conservation Status, the three categories presented a similar habitat loss (Fig. 2), where species with the status “None” suffered the greatest loss (Table 1). The NT category presented the highest habitat loss for the Conservation Category, followed by the “NA/NE” (Fig. 2; Table 1). Finally, habitat loss was similar for all five analyzed groups, with the order Carnivore presenting the highest values, followed by the Cingulata Order (Fig. 2; Table 1).

Most taxa included in this study (71%) suffered between 0–10% of habitat loss within their distribution, followed by 14% of the taxa that suffered between 10–20% of habitat loss. Only four taxa (2% of all included in this study) experienced habitat loss greater than 30%—three of which are already considered threatened with extinction and one which was previously categorized as Near Threatened (File S2). About 6.8% of the analyzed taxa’s ranges (13 species) suffered between 20% and 30% habitat loss (Fig. 3), including five species (38.46%) that are already threatened with extinction (File S2). Less frequently, some taxa (6%–11 species) exhibited an increase in their habitat cover over 3GL (represented as negative values) (Fig.3; File S1).

Discussion

Our results revealed that many mammalian taxa suffered habitat losses to differing degrees and intensities over the past three generations, while only a few species experienced increases in habitat cover, highlighting the pervasive effects of habitat loss on species distribution and the potential effects on taxa persistence which have been previously presented in many studies (Crooks et al., 2017; Heinrichs, Bender & Schumaker, 2016). Overall, habitat loss occurred regardless of Conservation Category, suggesting that some of the current Non-threatened or Non-evaluated species might experience an uplisting (change to a higher threatened category) in the subsequent extinction risk assessment, solely based on these habitat loss analyses by inferring a consequent proportional population reduction. For example, Tolypeutes matacus, the southern three-banded armadillo, currently classified as NT, may suffer an uplisting due to the high habitat loss estimated for its range. This species only inhabits the Pantanal region, which has experienced severe wildfires over the past 2 years (2019–2020), compromising a significant number of refugees for this species (Silva et al., 2020).

Species that have been recently described but are still not evaluated, may also enter the list as threatened when first assessed. One example is the Cyclopes rufus, first described only 5 years ago (Miranda et al., 2017), which may be classified as Vulnerable since habitat loss within its range was estimated to be greater than 30% (File S2). This species lives in one of Brazil’s most heavily deforested areas—the Arc of Deforestation in the Amazon, which is responsible for about 75% of the total habitat loss in the Amazonian region (Fearnside, 2005; Oviedo, Lima & Augusto, 2019). Similarly, considering the estimated habitat loss calculated in this study, the recently described Leopardus munoai and Plecturocebus grovesi may be classified as NT or threatened in the next extinction risk assessment (File S2).

Our results demonstrated that habitat loss also affects taxa that are already categorized as threatened. Nevertheless, despite being categorized as threatened, the vast majority of taxa only lost a moderate amount of their habitats. This may reflect taxa that experienced habitat loss too many years ago (i.e., before the 3GL time window needed for the application of Criterion A) but have suffered from the secondary effects of habitat loss and may have been categorized as threatened under another criterion. Additionally, some taxa may inhabit the Atlantic Forest, which has already lost more than 70% of its forest coverage (Rezende et al., 2018; Souza et al., 2020), despite being protected by specific law regulations, which can make further deforestation difficult (Atlantic Forest Law, n° 11.428, December 22, 2006). Furthermore, some taxa may be categorized as threatened as a result of other factors, such as population reduction due to pathogens, competitors, human conflicts, pollutants, or parasites, or even due to small geographic ranges or population sizes, along with other threats like population decline. In such cases, habitat loss may only be one of the threats leading to threatened categorizations. Indeed, for many taxa, habitat loss estimation must be accompanied by other types of data, such as population size reductions that are observed, estimated, or inferred by other sources—such as roadkill and poaching estimates, potential levels of exploitation, number of locations, among others (IUCN, 2022). Furthermore, most of these threatened taxa are included in National Action Plans, which may mitigate direct threats to the taxa.

The order Carnivora experienced the greatest habitat loss (Fig. 3C). This group is composed predominantly of predator species with low population densities and high spatial requirements. Therefore, this order is among the most threatened in Brazilian territory (Mittermeier et al., 2013). Following Carnivora, the orders Cetartiodicatyla and Perissodactyla, which compose the ungulates group, suffered the highest habitat loss intensity (Fig. 3C). These taxa include susceptible species with low reproduction, extended parental care, and high requirements for natural areas (Beca et al., 2017; Jorge et al., 2021; Keuroghlian, Eaton & Longland, 2004).

Contrarily to the Atlantic Forest biome, which has historically suffered extreme habitat loss, the Amazon and Cerrado biomes have suffered greater levels of natural habitat loss in the more recent past. The Amazon still concentrates a high proportion of natural remnants; however, it has lost more than 74 billion hectares of its forest habitat since 1985 (MapBiomas Project, 2021). The Cerrado, a Brazilian hotspot, has already lost 50% of its natural areas (WWF- World Wide Fund For Nature, 2019). The lack of specific legislation to protect this biome, which already exists for the Atlantic Forest, exacerbates the degradation of Amazon and Cerrado biomes, which is projected to continue in the future (Soterroni et al., 2019; Strassburg et al., 2017; Velazco et al., 2019). Conversely, all taxa that have experienced habitat expansion in their range inhabit the Atlantic Forest (File S2; Fig. S1). Over the past few years, this phytogeographic domain has experienced an increase in its natural areas (Rezende et al., 2015; Júnior, Santos & Eugenio, 2012), due, in part, to specific environmental laws, social-economic shifts, and conservation pacts (Baptista & Rudel, 2006; Fundação SOS Mata Atlântica, 2021).

Conclusions

In this study, we successfully used the accessible dataset of the MapBiomas project through the Google Earth Engine platform to generate a user-friendly, open-source, and replicable script to calculate habitat loss for terrestrial mammalian taxa, generating accurate and reliable data to support extinction risk assessments. The script’s usefulness was demonstrated by estimating habitat loss for 190 mammalian species in Brazil. Improving habitat loss analyses is a crucial step for improving the conservation status assessment for Brazilian mammals and other terrestrial taxa in the future. Additionally, our results provide new information which can guide future research and conservation efforts. Finally, the script and protocol we present here can be directly applied or adapted for use in other similar assessments while remaining open to continuous development.

Supplemental Information