Introduction

In 1995, the IUCN (International Union for Conservation of Nature) placed the Pan troglodytes taxon in the red list category of Endangered species. This listing included all four subspecies of chimpanzees: Western (P.t. verus), Central (P.t. troglodytes), Eastern (P. t. schweinfurthii), and Nigeria-Cameroon chimpanzee (P.t. ellioti). Since 1995, populations have continued to decline (Humle et al. 2016a). The current estimate of the total population size of chimpanzees is approximately 200,000 individuals. This estimate indicates a 66% decline over a 30-year span (Kormos et al. 2003). Of the four subspecies, the western chimpanzee is the only subspecies listed as Critically Endangered (Humle et al. 2016a). Since 1990, the population size of western chimpanzees has declined approximately 80% (Kühl et al. 2017). The principal threats to western chimpanzees are habitat loss and/or degradation, hunting, and disease (Humle et al. 2016b).

Chimpanzees in Guinea are the largest remaining population of the western subspecies (Humle et al. 2016a; Kormos et al. 2003). In recognition of the negative effects of habitat destruction and loss of biodiversity, protected areas (PAs) have historically been established in Guinea. There are four PAs in Guinea: Massif du Ziama Strict Nature Reserve, Badiar National Park, Haut Niger National Park, and the Mount Nimba Strict Nature Reserve. An estimated 5–20% of the chimpanzee population in Guinea resides in these areas. The rest lives outside of PAs (Kormos et al. 2003).

Protected areas are impacted by human encroachment and neighboring land-use changes. From 2000 to 2012, the Mount Nimba Strict Nature Reserve (Mt. Nimba SNR hereafter) lost 1.5 km2 of forest within its boundary (approximately 1% of the reserve area) and 21.7 km2 within a 10-km buffer zone around the reserve (Allan et al. 2017). Laurance et al. (2012) found that changes both within and outside PAs influence ecosystem health. For example, changes in the landscape structure of areas surrounding PAs may increase area isolation and edge effects (Laurance et al. 2012). Increasing isolation of chimpanzee communities leads to reductions in gene flow, threatening healthy, viable populations (Morin et al. Morin et al. 1994). Moreover, competition for land and resources leads to increases in human–chimpanzee interaction and conflict (Hockings et al. 2015; McLennan and Hill 2012). For example, in Bossou, Guinea, where there are few forested areas, the chimpanzees rely heavily on cultivars (cassava, papaya, and bananas), terrestrial herbaceous vegetation (Zingeberaceae and Marantaceae families), and oil palm (Elaeis guineensis) during periods of fruit scarcity (Humle 2011). Many of these alternative food sources are in human settlements, so increased reliance accelerates human–chimpanzee conflict (Humle 2011). A decrease in forested areas due to human encroachment will result in an increase in the interaction of humans and chimpanzees.

To effectively protect chimpanzees and their habitats, it is important to understand how chimpanzees respond to their environment, so conservation efforts can focus on areas of highest importance for their long-term survival. Identifying the environmental factors that influence chimpanzee habitat selection is a critical component of developing effective conservation plans (Rushton et al. 2004). Species distribution models (SDMs) (also referred to as habitat suitability models, habitat models, ecological niche models, environmental niche models, etc.) are an informative way to evaluate the importance of environmental variables related to species distribution (Franklin 2009). Species distribution modeling provides a means for mapping chimpanzee habitat. The results of the modeling exercise can be used to guide reserve design, habitat management, and conservation planning. Species distribution models estimate conditions suitable for species survival by examining the relationships between species’ occurrence and associated environmental conditions.

Here, we use direct and indirect evidence of chimpanzee occurrences from fieldwork, medium-resolution remote sensing data, and SDMs to evaluate how the spatial distribution of biophysical variables relates to the distribution of the Seringbara chimpanzee communities in the Mt. Nimba SNR. This modeling effort allows us to test the hypothesis that vegetation is one of the most important factors influencing the occurrence of great apes (Jantz et al. 2016; Junker et al. 2012; Koops 2011; Koops et al. 2012a, b; Torres et al. 2010), specifically within the Seringbara chimpanzee communities. We will also explore the importance of other biophysical variables as they relate to the probability of Seringbara chimpanzee occurrence and compare our modeling results with relevant conservation efforts in the region. To do so, we will quantify and map the spatial distribution of biophysical variables within the study area using remotely sensed images, analyze the importance of each biophysical variable in modeling suitable chimpanzee habitat, and use SDMs to identify areas most suitable for chimpanzees within the Greater Nimba Landscape.

Methods

Study site

The Mt. Nimba SNR is a UNESCO World Heritage Site in Danger (World Heritage Committee 2017). The Mt. Nimba SNR encompasses most of the Nimba Mountain range in Guinea and parts of Côte d’Ivoire on the southeastern side of the mountain range. Covering approximately 175 km2, the reserve is dominated by wet, evergreen forests with diverse topographical features including rocky peaks, rough cliffs, bare granite, steep river valleys, high-altitude savannahs, and rounded hilltops (Guillaumet and Adjanohoun 1971; Koops 2011; Kormos et al. 2003). The reserve is home to a variety of flora and fauna, including the critically endangered endemic Mt. Nimba viviparous toad (Nimbaphrynoides occidentalis) and the Critically Endangered western chimpanzee (P.t. verus) (World Heritage Committee 2017).

The study site (N07.634°, W08.425°), spanning 30 km2, is located on the Guinean side of the Nimba Mountains within the Mt. Nimba SNR (Fig. 1). The site is largely composed of primary tropical forests, but as the terrain becomes steeper, it transitions to a mosaic of terrestrial herbaceous vegetation, montane forest, and high-altitude grasslands (Koops 2011). The elevation ranges from 595 to 1511 m. The climate is characterized by a rainy season from February to October and a dry season lasting from November to February (Koops et al. 2012a, b, Koops et al. 2013). The site is adjacent to the small village of Seringbara, located about 6 km from Bossou at the foot of the Nimba Mountains (Koops 2011). Bossou is home to a community of chimpanzees (currently seven individuals) that have been the focus of research for over 30 years by the Kyoto University Primate Research Institute (KUPRI) (Matsuzawa and Humle 2011). The Mt. Nimba SNR and Bossou are separated by savannah that few chimpanzees traverse (Matsuzawa et al. 2011b). This study focuses on at least two communities of chimpanzees within the Mt. Nimba SNR, known as the Seringbara communities (Koops unpublished data). The Seringbara communities have been the focus of habituation efforts since 2003 (Koops 2011) and intermittent ecological studies and surveys since 1996 (Matsuzawa and Yamakoshi 1996; Humle and Matsuzawa 2001, 2004). The chimpanzees remain mostly unhabituated to humans (Koops 2011; Koops and Matsuzawa 2006; Matsuzawa et al. 2011a).

Fig. 1
figure 1

Location of the Seringbara study area on the Guinean side of the Mt. Nimba SNR in West Africa

Full size image

Occurrence data

Between January 2012 and April 2014, a team of research assistants and local field assistants collected data on chimpanzee behavior at the Seringbara study site on the Guinean side of the Mt. Nimba SNR. Research teams maintained a nearly constant presence in the forest during this period, only missing data collection for 1–2 days a month. Field days focused on tracking and directly observing chimpanzees to obtain data on ranging, grouping, diet, nest building, and tool use. Direct observations of wild chimpanzees can be difficult, especially when communities are not fully habituated, such as the Seringbara communities. For this reason, nests, fecal samples, ant dipping sites, and feeding traces (i.e., wadges) were considered indirect indicators of chimpanzee presence and included as occurrence points along with direct chimpanzee sightings.

Direct and indirect evidence of chimpanzee presences were recorded using handheld global positioning system (GPS) devices during daily tracking of the chimpanzees. Sampling effort within the study area was comprehensive, as we covered the whole study area when searching for chimpanzees by splitting into teams and exhaustively surveying the study area using opportunistic sampling. In total, 1385 occurrence points were recorded. Occurrence points were not evenly distributed throughout the study area due to the behavior of the chimpanzees and perhaps also due to sampling bias. In a study comparing the different methods commonly used to correct for sampling bias, Fourcade et al. (2014) found that systematic spatial filtering consistently outperformed most other methods regardless of the species or type of bias. Systematic spatial filtering uses a grid of a user-defined cell size and randomly keeps one occurrence point per cell. We used R 3.3.2 (Supplementary Appendix A) to place a grid (30-m resolution) over the study area and randomly select one occurrence point from each grid cell. The total occurrence points were filtered and reduced to N = 947 for use in the final model (Fig. 2). Filtering to include only one occurrence point per cell did not influence our results because this study does not address the frequency nor magnitude of use by chimpanzees. Absence data were not available for this study. In addition, we chose to combine direct and various indirect types of evidence of occurrence for modeling, because (1) we wanted a robust sample size and classifying occurrences into behavior categories would drastically decrease sample size for the model and (2) it is unclear the behavior category for which we would attribute the feces occurrences, given that they can be found at feeding locations, along movement routes, at resting spots, as well as other locations of use (Supplementary Appendix A Table 1).

Fig. 2
figure 2

Location of chimpanzee occurrence points used in the model (N = 947)

Full size image

Predictor variables

Raster layers of predictor variables (Table 1) dealing with landscape structure and land cover, herein referred to as biophysical variables, were prepared at a 30-m spatial resolution. An initial set of 17 biophysical variables (Table 1) was assessed, as detailed below, before being narrowed down to 12 variables in the final model. Minimum distance, supervised classification of a Landsat 8 image, obtained during the study period (December 26, 2013), was developed in ENVI 5.0.2 to delineate five land-cover types: dense forest, mixed forest, bare ground, village, and savannah (Supplementary Appendix B). These five classes were chosen based on expert knowledge of the region after analyzing the spectral groupings of the supervised classification. Dense forests consist of mostly primary, undisturbed forest comprised of tree species such as Parinari excelsa, Parkia bicolor, Antiaris africana, and Aningeria altissima. Mixed forests are mostly secondary, disturbed forests with less dense vegetation and less canopy cover. Tree species common in mixed forests include Musanga cecropioides, Elaeis guineensis, and Uapaca sp. Bare ground includes cleared areas, sparsely vegetated grasslands, and bare rock. Savannah consists of very dense, tall grass areas lacking trees. The village class includes buildings, huts, and other anthropogenic structures interspersed with bare ground. The minimum distance land-cover classification procedure performed well (overall accuracy of 90.78%) in distinguishing between macrohabitats, such as savannah and forest, but was not able to distinguish microhabitats, such as vegetation types, at the spatial resolution of the image (30 m) (Supplementary Appendix B). Because chimpanzees have sophisticated mental mapping capabilities (Ban et al. 2014; Boesch and Boesch 1984; Normand et al. 2009; Normand and Boesch 2009) and are able to perceive their surroundings at the level of individual trees and forest patches, vegetation indices were calculated to capture differences at microscales (Pintea et al. 2003; Torres et al. 2010). Landsat 8 imagery from six different dates within the data collection period was used to derive an average normalized difference vegetation index (NDVI) raster. NDVI is an indication of relative biomass (i.e., healthy, photosynthetically active vegetation) within each raster cell and can range from − 1 (water or bare ground) to 1 (healthy, dense vegetation). It is calculated from the near-infrared and red bands of a satellite image ((NIR – R)/(NIR + R)) (Campbell and Wynne 2011). In addition, we captured microhabitat characteristics within the study area using a tasseled cap transformation of the original Landsat 8 image. This process transforms the original spectral data into a new coordinate system with four orthogonal axes (Campbell and Wynne 2011). Each of these axes carries specific information that can be interpreted as (1) soil and surface brightness (brightness), (2) photosynthetically active vegetation (greenness), (3) soil moisture (wetness), and (4) atmospheric noise (Crist and Cicone 1984).

Table 1 Biophysical predictor variables evaluated for use in modeling habitat suitability for the Seringbara communities
Full size table

Studies of the Seringbara chimpanzees (Koops 2011; Koops et al. 2007, 2012a, b, 2013, 2015), as well as other non-human primates (Clee et al. 2015; Gregory et al. 2014; Hickey et al. 2013; Plumptre et al. 2010; Serckx et al. 2016; Torres et al. 2010; Wich et al. 2012), indicate that climate, vegetation, and anthropogenic factors may play a significant role in identifying suitable habitat. In particular, the dietary preferences of Seringbara chimpanzees indicate that the availability of fruit affects their ranging patterns (Koops 2011; Koops et al. 2013). Many of the tree species producing fruit utilized by the chimpanzees occur in primary forests and at elevations higher than 800 m (e.g., Parinari excelsa) (Koops 2011). Moreover, the Seringbara chimpanzees prefer to nest at locations with lower humidity (Koops et al. 2012a, b). For example, they tend to nest at higher altitudes (above 1000 m) where relative humidity is low and avoided nesting in areas of high humidity (below 800 m) (Koops et al. 2012a, b). Therefore, the other biophysical variables included in the initial model were chosen for their ability to serve as proxies for these (i.e., climate, vegetation, and anthropogenic) factors (Franklin 2009).

The following variables were generated using ArcMap 10.2.2 (ESRI 2011) and R 3.3.2 (R Core Team 2005) and derived from a digital elevation model (DEM) from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) Global Digital Elevation Model Version 2 (NASA 2009): elevation, slope, aspect, topographic position index, roughness, integrated moisture index, heat load index, landform curvature, compound topographic index surface relief, and hierarchical slope position (Table 1). The R script for calculating hierarchical slope position is found in Supplementary Appendix C. The distance to rivers variable was generated in ArcMap 10.2.2 using a shapefile of rivers within the Greater Nimba Landscape and calculating the Euclidean distance of each 30-m2 cell from the nearest permanent river or stream.

We examined correlation between variables to reduce the effect of collinearity on interpreting Maxent results (Dormann et al. 2013; Kumar et al. 2014; Rödder et al. 2013). Correlation was calculated using Pearson’s product moment correlation (r). For a set of highly correlated variables (|r| > 0.7), the variable with the highest predictive power (training gain), in the preliminary model using all 17 biophysical variables, was retained (Dormann et al. 2013; Estes et al. 2010; Hickey et al. 2013).

Modeling technique

To map suitable chimpanzee habitat and analyze biophysical variables contributing to suitability, we used Maxent 3.3.3 software based on the maximum entropy framework (Phillips et al. 2004). Maxent estimates relative probability of species presence given data on occurrence and user-selected predictor variables (Franklin 2009; Phillips et al. 2006). Maxent performs well with presence-only data and frequently outperforms other SDM methods (Elith et al. 2006, 2011; Phillips and Dudík 2008; Wilson et al. 2013). The result is a best-fit model classifying locations in the study area according to probability of presence (0–1, with 1 indicating highest probability of presence). The model’s predictive performance is evaluated using the area under the receiver operating characteristic curve (AUC). AUC was chosen over other evaluation measures because it does not require an arbitrary selection of a threshold (Phillips et al. 2006). For presence-only data, AUC describes the probability that the model scores a presence site higher than a background site (Phillips et al. 2009). An AUC of 1 indicates perfect predictive power and an AUC of 0.5 indicates random prediction. A model with a high AUC, such as 0.70, indicates that there is a greater-than-random chance that a randomly selected presence site will be given a higher value than a randomly selected background site (Elith et al. 2006). Thus, a model with a high AUC has more discriminative power. A k-fold cross-validation procedure was replicated ten times to obtain a mean AUC value for the final model (Dormann et al. 2013; Kumar et al. 2014; Wich et al. 2012). Additionally, Maxent was used to generate response curves showing the relationship between each predictor variable and predicted probability of chimpanzee presence (i.e., predicted habitat suitability). Percent contribution and permutation importance were reported for each variable. Percent contribution is a measure of the amount of explained variance each variable contributes to the model. Permutation importance is a measure of how AUC changes when a variable is removed from the model and it is not sensitive to the order variables are put into the model (Songer et al. 2012; Wilson et al. 2013).

The model was projected beyond the study area to better assess habitat suitability for the Seringbara chimpanzees within the larger landscape. This geographically projected model is hereafter referred to as the final model. The extent, referred to as the Greater Nimba Landscape for this study, includes the majority of the Nimba Mountain range in Guinea, Liberia and Côte d’Ivoire, as well as the regions surrounding a few of the closest villages to the study site and an iron ore mining site (Fig. 3). The total area of the Greater Nimba Landscape is 992 km2. By including these villages, namely Bossou, Seringbara, Nyon, and Zuoguepo, and their near surroundings, the model is better able to capture the landscape heterogeneity of the region and its influence on habitat suitability beyond the protected area. It is important to note a few limitations of projecting, or transferring, a model into a geographic region where data were not collected (Warren and Seifert 2011). One issue in model transferability is the difference in predictor variable ranges between the sampled area and the area into which the model is projected (i.e., Greater Nimba Landscape). If the ranges in the sampled region are narrower, it can cause the response curves to be truncated (Randin et al. 2006). In addition, transferring a model can reduce the model’s predictive ability in the new region (Eger et al. 2016). For this reason, the results from the study should be interpreted carefully while keeping these limitations in mind.

Fig. 3
figure 3

Landsat 8 satellite image of the Greater Nimba Landscape

Full size image

Since absence data were not available, a maximum of 10,000 background points were randomly generated to represent the availability and range of environmental conditions within the study area (Wilson et al. 2013). A minimum-convex polygon around the occurrence points was created to restrict background point generation to only the area covered while collecting data in the field. This procedure ensures that sampling of background points is restricted to the same region from which occurrence points were collected and helps account for sampling bias (Phillips et al. 2009).

The final Maxent output is a gradient model classifying each pixel according to probability of presence or habitat suitability. In many cases, SDMs are converted to binary models, delineating suitable versus unsuitable habitat, which are used by conservationists and land managers (Fourcade et al. 2014; Escalante et al. 2013). Reclassification to create a binary model requires the identification of a threshold, above which a location is considered suitable for a species (Liu et al. 2005). There is not a single method for threshold selection that is better than all others regardless of the species or study objective (Liu et al. 2005). For this study, we reclassified the final model output to create binary maps of habitat suitability for the Seringbara chimpanzees based on three commonly used threshold selection approaches: minimum training presence, 10 percentile training presence, and equal training sensitivity and specificity (Escalante et al. 2013; Fourcade et al. 2014; Norris 2014; Pearson et al. 2007). The purpose of these binary maps was to visually and quantitatively assess the amount of suitable and unsuitable habitat for the Seringbara chimpanzees in the Greater Nimba Landscape, while also emphasizing the importance of carefully choosing a threshold approach.

Results

Correlation analysis

The following variables were highly correlated (|r| > 0.7): TPI and curvature (r = 1), slope and roughness (r = 0.85), NDVI and greenness (r = 0.91), NDVI and wetness (r = 0.74), wetness and greenness (r = 0.74), and LCC and wetness (r = − 0.72) (Supplementary Appendix D). For each highly correlated pair, the variable retained in the test models was chosen because it had the higher permutation importance when an initial model was run using all variables. Thus, the final model was created using only 12 of the original 17 biophysical variables: NDVI, elevation, HSP, brightness, DTR, aspect, HLI, CTI, IMI, roughness, curvature, and relief (Table 1).

Gradient Habitat Suitability Model

The fit of the final chimpanzee habitat suitability model for the Greater Nimba Landscape was 0.721 with a standard deviation of 0.023. Models with AUC values greater than 0.70 are considered to have fair discriminative abilities and are ecologically useful (Araujo et al. 2005; Pearce and Ferrier 2000; Swets 1988). The resulting map from the final model (Fig. 4) highlights areas of highest predicted suitability for chimpanzee habitat. The biophysical variables contributing most to the model, as measured by permutation importance, were NDVI (37.8%), elevation (27.3%), HSP (11.5%), brightness (6.6%), and DTR (5.4%) (Table 2).

Fig. 4
figure 4

Chimpanzee habitat suitability model showing the geographic distribution of suitable chimpanzee habitat throughout the Greater Nimba Landscape. This is a gradient model displaying habitat suitability on a scale from 0 (low suitability) to 1 (high suitability). This figure illustrates the importance of the Mt. Nimba SNR in providing habitat for chimpanzees within the Greater Nimba Landscape

Full size image
Table 2 Permutation importance and percent contribution of each biophysical predictor variable used in creating the final habitat suitability model
Full size table

Variable response curves

The spatial distributions for the biophysical variables of highest importance were mapped and displayed above the corresponding response curves (Fig. 5). The response curve for NDVI shows a positive relationship between probability of presence and NDVI, as healthy, photosynthetically active vegetation increases, so does the probability of chimpanzee presence (Fig. 5a). The response curve for elevation shows that probability of presence is highest between 800 and 1200 m (Fig. 5b). The response curve for hierarchical slope position indicates that probability of presence fluctuates in mildly exposed areas (HSP values between 0.3 and 0.65), whereas probability of presence is relatively low in valley bottoms and toe slopes (low HSP values) and is lowest in topographically exposed areas, such as cliff faces and ridges (high HSP values) (Fig. 5c). For brightness, probability of chimpanzee presence peaks at an index value of 0.35 before declining sharply at higher brightness values (Fig. 5d). There is a negative relationship between DTR and probability of presence, with a sharp decline in probability of presence for areas farther than 500 m from a river (Fig. 5e). Response curves and maps for all other biophysical variables used in the final model can be found in Supplementary Appendix E.

Fig. 5
figure 5figure 5figure 5

Plots of the response curves showing the dependence of probability of presence on a given biophysical variable. Each plot represents a Maxent model using only the corresponding variable. The plots are given for the five biophysical variables with highest permutation importance (percent shown on plot). The plots show the average response (red line) and the standard deviation (blue interval around the average). X-axes show the units of the corresponding variable. Y-axes indicate the logistic output. The maps above each response curve illustrate the spatial distribution of the biophysical variable in the Greater Nimba Landscape

Full size image

Binary habitat suitability models

The final model was reclassified to create three binary models based on different threshold levels: minimum training presence (0.08), 10 percentile training presence (0.33), and equal training sensitivity and specificity (0.46) (Fig. 6a–c, respectively). Using a threshold allowed the amount of suitable versus unsuitable habitat to be delineated and quantified within the Greater Nimba Landscape (992 km2) (Table 3). For the minimum training presence threshold (0.08), 42% of the landscape was classified as suitable and 58% was classified as unsuitable for the Seringbara chimpanzees. The equal training sensitivity and specificity threshold (0.46) lends itself to a different interpretation of the Greater Nimba Landscape, as only 3% was classified as suitable habitat and 97% was unsuitable. Similarly, the 10% training presence threshold (0.33) delineated 7% of the Greater Nimba Landscape as suitable and 93% as unsuitable.

Fig. 6
figure 6

Final model output showing the distribution of suitable chimpanzee habitat throughout the Greater Nimba Landscape as a series of binary models of three different threshold values: a minimum training presence, b 10% training presence, and c equal training sensitivity and specificity

Full size image
Table 3 The amount of area (km2) within the Greater Nimba Landscape that was delineated as not suitable and suitable based on the assigned threshold value
Full size table

Discussion

Data on habitat requirements of chimpanzees are needed for effective management and conservation. Constant advancements in technologies, such as remote sensing and GIS, combined with modeling techniques, such as Maxent, allow researchers to assess the influences on habitat suitability for many different species. In this study, we modeled the habitat suitability for the Seringbara chimpanzees in the Greater Nimba Landscape and identified the most important biophysical variables contributing to habitat suitability. The results indicate that NDVI, elevation, hierarchical slope position, brightness, and distance to rivers contributed most to predicted habitat suitability (Table 1).

The most important variable in predicting chimpanzee habitat suitability was NDVI. This index indicates the presence of photosynthetically active vegetation (Campbell and Wynne 2011). The positive relationship between NDVI and probability of occurrence suggests that chimpanzees prefer forested areas with dense, healthy vegetation (Fig. 5a). A study by Koops et al. (2012a, b) showed that the Seringbara chimpanzees prefer larger trees with dense leaf cover in primary forests to build nests. In addition, many of the tree species, utilized by the chimpanzees for feeding, are found predominantly in primary forests (e.g., Parinari excelsa, Parkia bicolor, Antiaris africana, and Aningeria altissima) (Koops 2011). This relation indicates that the habitat suitability model presented here is capturing important biological signals from the Seringbara chimpanzees’ use of the landscape. Similar studies at other locations have also shown that vegetation influenced chimpanzee behavior (Jantz et al. 2016; Torres et al. 2010), as well as great ape behavior in general (Junker et al. 2012).

Elevation was the second most important biophysical variable in predicting habitat suitability for the Seringbara chimpanzees. The relationship between elevation and the probability of Seringbara chimpanzee occurrence is bell shaped (Fig. 5b). Increasing elevation up to 900 m is associated with increasing probability of occurrence. Above 900 m, increasing elevation is associated with decreasing probability of occurrence. Within the Greater Nimba Landscape, elevation serves as a good proxy for climate and vegetation, as well as anthropogenic disturbance. Unfortunately, there is not sufficient data on anthropogenic disturbance for the whole reserve, but based on personal observations, we noticed that many of the villages and cultivated fields surrounding the study site are all located below 700 m. Thus, as elevation increases, so does the distance from anthropogenic disturbance. In addition, the protected status of the Nimba Mountains increases this effect because the mountains are within high-elevation areas. Although protected status does not directly indicate a lack of anthropogenic disturbance, the Mt. Nimba Strict Nature Reserve is remote, hunting pressures tend to decrease with distance from villages, and illegal hunting is targeted at animals other than the Seringbara chimpanzees (pers. obs., Koops and Fitzgerald). Moreover, as elevation increases above 1200 m, the landscape is dominated by high-altitude grasslands (Lamotte 1998), which may not provide ample resources for chimpanzees (Koops 2011). Thus, resulting in the bell-shaped curve of the relationship between elevation and probability of Seringbara chimpanzee occurrence.

The HSP (a measure of topographic exposure) was the next most important variable in predicting chimpanzee presence. Topographic exposure is the degree to which a location is surrounded by high relief terrain. A high HSP value indicates that a location is not surrounded by areas of higher relief (i.e., exposed), such as a cliff face or ridge top. A low value indicates that the landscape is surrounded by high relief terrain (i.e., not exposed), such as valley bottoms and toe slopes. The relationship between topographic exposure and the probability of Seringbara chimpanzee occurrence is generally negative, where an increase in topographic exposure is associated with a decrease in the probability of occurrence (Fig. 5c). Thus, Seringbara chimpanzees are more likely to occur in less-exposed areas, such as mild slopes, not surrounded by high relief terrain. Exposure can serve as a proxy for temperature and vegetation similar to the other important biophysical features, but it might also relate to the ease of movement through an area. Non-human primates have been found to distinguish between topographic features when traveling. For example, Gregory et al. (2014) found that bearded saki monkeys use ridge tops and slopes near ridges, because it may reduce the energetic cost of travel and/or serve a function in route-based mental mapping. This behavior is yet to be explored for chimpanzees in the Greater Nimba Landscape. Future studies examining the role of topography in chimpanzee movement would contribute greatly to our understanding of their perception and utilization of the landscape.

Another important variable in predicting chimpanzee habitat suitability was the tasseled cap brightness index. As brightness values increase, it indicates an increase in open canopy and an increase in bare ground (Campbell and Wynne 2011; Cohen and Goward 2004; Cohen et al. 1995). Cohen et al. (1995) showed that closed forest stands tend to have moderate brightness values. Previous studies from other chimpanzee research sites indicate that mature, closed forests are preferred by chimpanzees (Torres et al. 2010). Thus, the results from this study, showing highest probability of presence at moderate brightness values support previous findings. Nevertheless, caution must be taken when interpreting brightness values, because this index is responsive to topographic variation in addition to forest condition (Cohen and Goward 2004). For example, in our study site, some of the high savannah areas have very low brightness values despite very minimal canopy cover (Fig. 5d). Other very similarly vegetated savannah regions have much higher brightness values. Thus, the low brightness value in some high savannah areas might be explained by the steepness of the terrain and the incidence angle of the radar from the satellite collecting the image (Cohen et al. 1995).

Habitat suitability is also affected by the proximity of an area to the nearest river. As distance increases, the probability of chimpanzee presence decreases. This biophysical variable may serve as a proxy for vegetation (Hickey et al. 2013; Koops 2011). In evaluating the distribution of the variable distance to river throughout the Greater Nimba Landscape (Fig. 5d), many of the places that are more than 500 m from rivers are in the high savannah areas of the Nimba Mountains or in areas outside of the Mt. Nimba SNR, where the terrain is slightly flatter and rivers are more dispersed. Riverine areas may also provide food resources not available elsewhere in the landscape (Koops pers. comm.).

The final habitat suitability model illustrates the isolation of high suitability areas within the Greater Nimba Landscape. The areas of highest predicted habitat suitability for the Seringbara chimpanzees are located almost entirely within the Nimba mountain range. This is highlighted in the binary classification of the habitat suitability map into areas of suitable and not suitable habitat based on various threshold values (Fig. 6). A comparison of the three binary models also highlights the importance of carefully selecting a threshold value. In this study, the amount of suitable habitat within the landscape ranged for 3–42% (Table 3). This variation in amount of suitable habitat based on threshold values may result in very different conservation strategies and threshold selection should be carefully considered based on local knowledge, research, and specific conservation goals. Although binary models can be arbitrary and over simplify the landscape for behaviorally flexible and dynamic species that may not perceive the landscape in binary terms, the ability to identify suitable versus unsuitable habitat is useful for conservation practitioners (Escalante et al. 2013; Ferrer-Sánchez and Rodríguez-Estrella 2016; Liu et al. 2005). For example, Torres et al. (2010) delineated suitable from unsuitable chimpanzee habitat to assess changes in habitat suitability over time as well as temporal changes in the most important ecogeographical factors influencing chimpanzee habitat in Guinea-Bissau. Their results provide a basis for practitioners to adapt their strategies based on past changes as well as forecasted changes to chimpanzee habitat suitability.

Additionally, within the Nimba mountain range, high suitability areas are fragmented by terrain features such as high ridgelines and anthropogenic disturbances, such as the iron-ore mining concession in the NE region of the Nimba mountain range (Fig. 4). Thus, not only are the Seringbara chimpanzees isolated from other chimpanzee communities outside the Mt. Nimba SNR, they are at risk of becoming isolated from other communities within the Mt. Nimba SNR. Isolation and fragmentation of suitable habitat hampers gene flow between groups and can lead to further decline in chimpanzee populations in the region.

Maintaining viable, healthy chimpanzee populations requires movement between communities, thus the creation of corridors is one solution to restoring connectivity. One of the current efforts in the Greater Nimba Landscape is the Green Corridor Project. This project was established in 1997 with the aim of connecting chimpanzee populations in Bossou with those in the Nimba Mountains by planting trees species utilized by chimpanzees in the savannah between the sites (Matsuzawa et al. 2011b). Despite difficulties with fires, the Green Corridor Project has made and continues to make progress. One sign of this progress was the video recording of two male chimpanzees traveling into the corridor and the use of the corridor by monkeys (“The Green Corridor Project” 2017). The project is ongoing and technologies such as remote sensing (e.g., use of unmanned aerial vehicles and satellite imagery) and modeling may prove useful for monitoring and expanding the corridor. In addition, as the vegetation in the corridor matures, its NDVI value will increase. NDVI was the most important biophysical variable in our model and increasing NDVI was related to increasing probability of occurrence. Our modeling effort supports the hypothesis that the corridor will increasingly provide more suitable habitat for chimpanzees as the vegetation within the corridor matures. Future plans for new corridors might additionally consider locations with low topographic exposure that are near rivers.

While conservation efforts can use the methods and results from this study and expert knowledge of the region to more effectively and efficiently promote the long-term viability of chimpanzees in the region, these efforts should also recognize the limitations of this study. Since the model was projected into a novel geographic area where data on chimpanzee occurrences were not collected, the response curves may not encompass the full range of variables. In other words, interpretation of how the probability of chimpanzee presence will respond to a predictor variable beyond the range of the collected data is unknown. This is a limitation for many predictive SDM studies, yet there are few generally applicable solutions (Eger et al. 2016; Elith et al. 2010; Peterson et al. 2007; Zurrell et al. 2012). Future research might be able to mitigate this by surveying more areas within the greater landscape so the sampling effort is more representative of the range in predictor variables. Moreover, given that vegetation and proxies for vegetation greatly influence chimpanzee habitat suitability, this model might be improved with data that are better able to capture vegetation characteristics at a higher spatial resolution. Likewise, the model results could be improved by (1) additional surveys in the Greater Nimba Landscape beyond the study area used to create the model, (2) systematic data pertaining to anthropogenic disturbance, and (3) ground truthing of the variables used and results.

Conclusions

In conclusion, this study demonstrates that species distribution modeling is a useful tool for identifying suitable chimpanzee habitat within montane rainforests. More specifically, the results indicate that (1) biophysical variables quantifying the landscape structure within the Greater Nimba Landscape were useful predictors of chimpanzee presence, (2) NDVI, elevation, hierarchical slope position, brightness, and distance to rivers had the greatest influence on habitat suitability for the Seringbara chimpanzees, (3) suitable chimpanzee habitat within the Greater Nimba Landscape is fairly isolated and does not make up a large portion of the landscape, and (4) enforcing the protection of the Mt. Nimba SNR and adjacent areas is vital to supporting chimpanzee populations.