Introduction
The Upper Guinea Forest (UGF) in West Africa is one of the last remnants of tropical rainforest in the region, supporting more than 9000 species of plants and c. 1800 species of vertebrates (Liu et al. Reference Liu, Wimberly and Dwomoh2016). This includes animals listed as Endangered in the International Union for Conservation of Nature (IUCN) Red List of Threatened Species, such as the pygmy hippopotamus (Choeropsis liberiensis; Ransom et al. Reference Ransom, Robinson and Collen2015, Ola & Benjamin Reference Ola and Benjamin2019). Owing to high species endemism and rapid forest loss, the UGF is considered a biodiversity hotspot, but forest loss and degradation have continued (Myers Reference Myers1988, Mittermeier et al. Reference Mittermeier, Gil, Hoffman, Pilgrim, Brooks and Mittermeier2004, Curtis et al. Reference Curtis, Slay, Harris, Tyukavina and Hansen2018).
The UGF is widely fragmented due to human population growth and the expansion of extractive industries, plantations, logging and slash-and-burn agriculture (Hosonuma et al. Reference Hosonuma, Herold, De Sy, De Fries, Brockhaus and Verchot2012, CILSS 2016). The UGF once extended over more than 620 000 km2, from Guinea in the east to Nigeria in the west, but unregulated activities such as logging and shifting agriculture were responsible for c. 90% of the forest loss between 1999 and 2013 (Bakarr et al. Reference Bakarr, Oates, Fahr, Parren, Rödel, Demey, Mittermeier, Gil, Hoffman, Pilgrim, Brooks and Mittermeier2004, CILSS 2016). Shifting agriculture, also known as ‘subsistence agriculture’, is common throughout the UGF region, and it is defined as the cultivation of small- to medium-scale crops maintained by smallholders (Hosonuma et al. Reference Hosonuma, Herold, De Sy, De Fries, Brockhaus and Verchot2012). Logging, agriculture and fire are also drivers of deforestation (Curtis et al. Reference Curtis, Slay, Harris, Tyukavina and Hansen2018). In the UGF region, Ghana lost c. 40% of its forest area between 1990 and 2010 (Hansen et al. Reference Hansen, Potapov, Moore, Hancher, Turubanova and Tyukavina2013); Côte d’Ivoire, Gabon, Liberia and Nigeria have also reported large areas of deforestation since 2005, mostly due to illegal logging (Oduro et al. Reference Oduro, Mohren, Affum-Baffoe and Kyereh2014). Large-scale crops of cocoa and oil palm are well-known drivers of deforestation in the UGF. For example, during 1989–2013, the area of oil palm plantations increased by 63% in Ghana and by 62% in Côte d’Ivoire, and more than 1 000 000 km2 of forests and woodlands in the region are vulnerable to transformation into crops and other forms of development by 2080 (Vijay et al. Reference Vijay, Pimm, Jenkins and Smith2016).
Agriculture is widespread in West Africa and is a major threat to the flora and fauna that inhabit the UGF, including the pygmy hippopotamus. In Côte d’Ivoire, for example, agriculture occupied c. 84% of the usable land until 2016 (Ruf et al. Reference Ruf, Schroth and Doffangui2015, CILSS 2016). Some 60% of the 37 300 km2 of tropical forests that existed in Côte d’Ivoire in 1975 was lost to shifting and industrial agriculture (CILSS 2016). These rates of forest loss have provoked drastic declines in endemic populations of flora and fauna in the region (Grelle Reference Grelle2005, Di Marco et al. Reference Di Marco, Butchart, Visconti, Buchanan, Ficetola and Rondinini2016, Vijay et al. Reference Vijay, Pimm, Jenkins and Smith2016).
Conversions from forest to agricultural land reduce usable habitat, increase fragmentation and limit species movement (Foley et al. Reference Foley, DeFries, Asner, Barford, Bonan and Carpenter2005, Gibson et al. Reference Gibson, Lee, Koh, Brook, Gardner and Barlow2011). For the endemic pygmy hippopotamus, forest loss has contributed to a sustained, region-wide population decline, and further forest loss may drive the species to extinction. Understanding the activities that lead to forest loss in the UGF is therefore critical for the conservation of this and other species. Between 1982 and 1986, there were 19 000 pygmy hippopotami in the region, but by 1997 the wild population consisted of fewer than 5000 individuals (Ransom et al. Reference Ransom, Robinson and Collen2015). Some of this decline was related to hunting for bushmeat and other purposes during times of internal conflict in the region (Conway Reference Conway2013, Conway et al. Reference Conway, Hernandez, Carroll, Green and Larson2015). The pygmy hippopotamus population may now be only 2000–2500 animals (Ransom et al. Reference Ransom, Robinson and Collen2015). Pygmy hippopotami are restricted to primary and secondary forests with high canopy cover, prefer locations close to water bodies and are rarely found in disturbed areas (Eltringham Reference Eltringham1999, Lewison & Oliver Reference Lewison and Oliver2008, Eshuis Reference Eshuis2011); the animals rely on relatively well-preserved forests to survive, and those conditions are found in some protected areas in the UGF.
Even though forest loss in the UGF is probably the major cause of the pygmy hippopotamus population decline, the spatial distribution of the drivers of forest loss in this area remains largely unknown. A driver of forest loss is an activity or process that changes a land with trees to a land without trees (Curtis et al. Reference Curtis, Slay, Harris, Tyukavina and Hansen2018). Understanding such activities is crucial for planning, infrastructure and conservation policies that consider the needs of human communities and endangered species in the UGF. Deforestation is prevalent in the UGF (CILSS 2016); however, spatially explicit information on the proximate drivers of deforestation within the pygmy hippopotamus’s range is still scarce. Remote sensing has enabled the drivers of forest loss to be investigated at global and continental scales. For instance, shifting agriculture was the main driver of forest loss in sub-Saharan Africa between 2001 and 2015 (Curtis et al. Reference Curtis, Slay, Harris, Tyukavina and Hansen2018). However, the spatial resolution of this global analysis, for which the minimum pixel size is 250 m, limits its ability to reveal the proximate drivers of forest loss at finer spatial scales. Our study uses 10–m resolution imagery and proposes a remote sensing classification approach to identify contemporary drivers of tropical forest loss within the range of a threatened species in a region with persistent cloud cover. Assessing the drivers of deforestation is critical for informing conservation efforts for cryptic species such as the pygmy hippopotamus and other large mammals, for which their survival is strictly linked to well-conserved forests (Bowler et al. Reference Bowler, Bjorkman, Dornelas, Myers-Smith, Navarro and Niamir2020). We used Sentinel-2 satellite imagery and the Hansen forest loss dataset (Hansen et al. Reference Hansen, Potapov, Moore, Hancher, Turubanova and Tyukavina2013) to assess proximate deforestation drivers in the pygmy hippopotamus distribution. Specifically, we sought to: (1) identify the main proximate drivers of forest loss in the pygmy hippopotamus’s range; (2) estimate the extent and distribution of forest loss related to each identified driver; and (3) estimate the amount of habitat loss inside protected areas in the UGF region within the pygmy hippopotamus’s distribution area.
Materials and methods
Study area
The known distribution of the pygmy hippopotamus spans 140 221 km2 across Guinea (2335 km2), Sierra Leone (4814 km2), Liberia (53 155 km2) and Côte d’Ivoire (79 917 km2; Fig. 1; IUCN 2015). Within this region there is a mixture of tropical and subtropical moist broadleaf forests, agriculture and human settlements. Rainfall occurs year round (range: 2200–5000 mm per year), with two long rainy seasons separated by two drier periods. Seasonal temperature ranges are 30–33°C during the dry season and 12–21°C during the wet season, with high humidity (≥70%) throughout the year. The natural vegetation is a combination of tropical–subtropical rainforests (closed-canopy forest, epiphytes, lianas and ferns) and tropical–subtropical lowland dry forest and thicket ecosystems (evergreen and drought-deciduous trees with some woody plants; CILSS 2016, Keith et al. Reference Keith, Ferrer-Paris, Nicholson, Bishop, Polidoro and Ramirez-Llodra2022).
Diagrammatic map of the study area delimited by the distribution of Choeropsis liberiensis in the Upper Guinea Forest and encompassing parts of Côte d’Ivoire, Liberia, Guinea and Sierra Leone. Red pixels represent areas of 2000–2019 forest (≥50% canopy cover) loss in the study area. Data sources: Hansen et al. (Reference Hansen, Potapov, Moore, Hancher, Turubanova and Tyukavina2013), Ransom et al. (Reference Ransom, Robinson and Collen2015), UNEP-WCMC and IUCN (2021).
Data acquisition and processing
We identified areas of forest loss during 2000–2019 using the Global Forest Change dataset v1.7 (2000–2019; Hansen et al. Reference Hansen, Potapov, Moore, Hancher, Turubanova and Tyukavina2013). We limited our analysis to areas with ≥50% canopy cover in 2000 (89 614 km2) because pygmy hippopotami are mostly found in dense forests and rarely venture into areas with low canopy cover (Eshuis Reference Eshuis2011, Bogui et al. Reference Bogui, Koffi, Koné, Ouattara, Kouakou and Gnagbo2016, Robinson et al. Reference Robinson, Flacke and Hentschel2017). Owing to persistent cloud cover, we opted to use Sentinel-2 due to its higher temporal resolution, which provided more cloud-free pixels and, therefore, a greater number of clear pixels for further analysis (Fig. S1).
We used Google Earth Engine (Gorelick et al. Reference Gorelick, Hancher, Dixon, Ilyushchenko, Thau and Moore2017) to gather all available Sentinel-2 images corrected to surface reflectance (i.e., level 2A, 10–m pixels) acquired for the study area in 2019 (n = 3108; Copernicus Sentinel data, 2019). We followed Braaten (Reference Braaten2021) to remove clouds and cloud shadows, then calculated eight spectral indices from image stacks to discriminate the different drivers of forest loss (Fig. S1 & Table S1). We developed a series of composite metrics to underpin our classification model of proximate drivers of forest loss within the distribution of the pygmy hippopotamus (Table S1); these metrics are frequently used for land-cover mapping to represent pixel-scale spectral dynamics that can be incorporated into statistical pixel classifications to discriminate different land-cover types (DeFries et al. Reference DeFries, Hansen and Townshend1995, Potapov et al. Reference Potapov, Turubanova, Hansen, Adusei, Broich and Altstatt2012). This process yielded a set of 182 covariate layers for use in the classification of the drivers of forest loss (Table S1).
Protected areas inside pygmy hippopotamus distribution
Our analysis focused on ‘declared’ protected areas according to the United Nations Environment Programme – World Conservation Monitoring Centre (UNEP-WCMC & IUCN 2021). There are 87 protected areas covering 30 669 km2 that are partially or totally located inside the pygmy hippopotamus’s range (UNEP-WCMC & IUCN 2021), and pygmy hippopotami are known to inhabit them (Roth et al. Reference Roth, Hoppe-Dominik, Muhlenberg, Steinhauer-Burkart and Fischer2004, Collen et al. Reference Collen, Howard, Konie, Daniel and Rist2011, Garteh Reference Garteh2013, Conway et al. Reference Conway, Hernandez, Carroll, Green and Larson2015, Bogui et al. Reference Bogui, Koffi, Koné, Ouattara, Kouakou and Gnagbo2016, Hillers et al. Reference Hillers, Buchanan, Garteh, Tommy, Fofana and Lindsell2017, Ouattara et al. Reference Ouattara, Bomey, Kone and Paris2018).
Driver classification
We developed a classification scheme of potential drivers of forest loss across the region (Fig. 2). Initial testing of the classification model revealed that the low number of cloud-free images over the study area, together with similarities in the spectral signatures of several drivers of forest loss, often resulted in driver misclassification. Therefore, we combined several driver classes into three driver categories that are informative of the drivers of forest loss in the UGF. The three categories are defined as ‘Urban expansion’, ‘Shifting agriculture’ and ‘Intensive agriculture’ (Fig. 2). While cattle grazing was not widespread in the UGF due to a range of diseases (National Research Council 1993), over 20 million people in West Africa now depend on livestock production and trade (Thébaud et al. Reference Thébaud, Corniaux, François and Powell2018, Valerio et al. Reference Valerio, Walther, Eilittä, Cissé, Muneepeerakul and Kiker2020). Cattle, buffalo, donkey, goat and sheep are common livestock in pastoral, transhumance and ranching practices in Côte d’Ivoire and Guinea, which combine livestock grazing with a range of crops (Inter-réseaux Developpment rural 2017, Valerio et al. Reference Valerio, Walther, Eilittä, Cissé, Muneepeerakul and Kiker2020). Livestock grazing areas were merged into the ‘Shifting agriculture’ class because their spectral signatures and image interpretation cues were similar from satellite imagery. We used the full 2019 Sentinel-2 image archive supplemented by high-resolution imagery available in QGIS to develop training data for each driver class. In total, we acquired 1150 training points for all classes (Fig. 2) and 450 points for independent accuracy assessment (Figs 2, S1 & S2; Foody Reference Foody2009). The training and validating points were collected by visually interpreting high-resolution imagery from Google Earth, Bing Maps and ESRI available in QGIS. The georeferenced training points correspond to the driver classification scheme of forest loss in our study area. We did not collect training data where high-resolution imagery was unavailable or in places where clouds were prevalent (e.g., southern Liberia and western Côte d’Ivoire).
Definition of the drivers of forest loss and training classes alongside two examples of each driver from high-resolution imagery. High-resolution imagery and other interpretation cues were used to identify drivers of forest loss within the pygmy hippopotamus distribution. Pixels from Sentinel-2 images of these areas were selected for training and validating the classification model. Imagery sources: Esri, DigitalGlobe, GeoEye, CNES/Airbus, Maxar Technologies, USGS, Google Earth.
Classification model
We used the Hansen et al. (Reference Hansen, Potapov, Moore, Hancher, Turubanova and Tyukavina2013) forest loss dataset to develop a mask of areas that had experienced forest loss in the region between 2000 and 2019. We used this mask to mask our covariate layers such that the model was deployed only in areas where recent deforestation has occurred (Table S1). A random forest classification model with 300 trees was used to classify these areas into one of the three drivers of interest. After applying the driver classification model, we post-processed the classified map with a majority (i.e., mode) filter to remove single, isolated pixels. Then, we computed the total area of each driver by country, both inside and outside protected areas. Accuracy was assessed using a confusion matrix; overall, pixels in the classified image were correctly assigned to drivers 87% of the time (overall accuracy). Class accuracies were high, with ‘Urban expansion’ achieving 96% accuracy, followed by ‘Intensive agriculture’ (86%) and ‘Shifting agriculture’ (81%; Fig. S2 & Table S2).
Results
In the year 2000 there were c. 89 600 km2 of forest with 50% canopy cover in the study area. Côte d’Ivoire and Liberia accounted for 94% (14 948 km2) of the net area of forest lost, c. 15 times greater than the forest loss areas of Sierra Leone and Guinea combined (953 km2; Table 1). Côte d’Ivoire accounts for 57% of the pygmy hippopotamus distribution, out of which 9825 km2 (61.8%) have been lost. Here, forest loss was concentrated in the north and south within the study area, as well as near the Liberian border, and was sparse in the central region (Fig. 3). Similar trends were detected in Liberia, where intense forest loss was evidenced in the southern and western regions of the country. Only 3.4% (4814 km2) of the pygmy hippopotamus distribution falls within Sierra Leone’s territory, of which 765 km2 (4.8%) of forest were lost between 2000 and 2019. Out of the four patches that pygmy hippopotami are expected to inhabit, the most degraded lies on the border with Liberia (Figs 1 & 3 & Table 1). When normalized by the area within the pygmy hippopotamus distribution, Sierra Leone has lost more forest than Côte d’Ivoire or Liberia despite an order of magnitude difference in net loss, followed by Côte d’Ivoire, Liberia and Guinea (Table 1 & Appendix S1).
Net and normalized areas of forest loss per country (km2) between 2000 and 2019. Net area of forest loss refers to the area of forest with >50% canopy cover. Normalized area of forest loss is defined as the ratio between the net area of forest loss (km2) and the area within the pygmy hippopotamus distribution (km2) × 100. Total area of forest loss attributed to each driver (per country; km2) is shown.
Distribution of drivers of forest loss in the study area during 2000–2019. The bar chart shows the area (km2) of forest loss by driver per country (for display purposes smaller patches of the pygmy hippopotamus distribution located in Sierra Leone are omitted).
Our driver classification model indicated that almost all forest loss within the pygmy hippopotamus distribution is attributable to ‘Shifting agriculture’ (93.4%), with minor contributions from ‘Urban expansion’ (5.6%) and ‘Intensive agriculture’ activities (1%; Table 1). ‘Shifting agriculture’ activities account for 13 960 km2 in Côte d’Ivoire and Liberia, whereas Sierra Leone and Guinea have lost fewer than 1000 km2. ‘Shifting agriculture’ in Guinea has accounted for 170 km2 out of 188 km2 of forest loss, making it disproportionally large when compared to the other drivers assessed here (Table 1). In contrast, Côte d’Ivoire has lost almost two orders of magnitude more area to ‘Shifting agriculture’ activities (9019 km2; Appendix S1).
Forest loss due to ‘Urban expansion’ was concentrated along the border, in western Côte d’Ivoire and north-eastern Liberia, with a small extent in Sierra Leone and Guinea (Fig. 3 & Table 1). Most of the ‘Urban expansion’ areas were associated with existing settlements, infrastructure and industries (e.g., mining) and were close to roads. We also found areas classified as ‘Urban expansion’ close to ‘Shifting agriculture’ and ‘Intensive agriculture’ patches, primarily in the form of land clearing for crops and infrastructure in oil palm plantations (Fig. 3 & Table 1). Within the pygmy hippopotamus distribution area, Côte d’Ivoire and Guinea were more impacted by ‘Urban expansion’ (0.88% and 0.73%, respectively) than Liberia (0.27%) or Sierra Leone (0.07%).
‘Intensive agriculture’ accounted for c. 1% (144 km2) of the total area of forest loss, with most expansion occurring in Côte d’Ivoire and Liberia (Fig. 3 & Table 1). As the main crop in the ‘Intensive agriculture’ category, oil palm was easily identifiable from high-resolution satellite images and is thus considered to be one of the main industrial crops responsible for forest loss. This crop is primarily found in southern regions of Liberia and Côte d’Ivoire and extends from east to west in the pygmy hippopotamus distribution (Fig. 3).
Protected areas within the pygmy hippopotamus’s range accounted for 30 669 km2 (22%) of the study area. During 2000–2019, 4835 km2 (15.8%) of forest within protected areas were classified as forest loss by the Hansen et al. (Reference Hansen, Potapov, Moore, Hancher, Turubanova and Tyukavina2013) dataset (Table 2). Côte d’Ivoire had the largest protected area network within the pygmy hippopotamus distribution (24 636 km2) and accounted for the largest area of forest loss within protected areas (3809 km2; Table 2). When the normalized area was considered, Liberia had lost the most forest (33.7%), followed by Côte d’Ivoire (15.5%), whereas Guinea and Sierra Leone had lost the least both in net surface area and in proportion of their protected areas (Fig. 4 & Table 2). Forest loss inside protected areas mainly occurred in northern, southern and western Côte d’Ivoire, as opposed to eastern and central protected areas within the study area. In Liberia, western protected areas were the most affected (Fig. 4).
Net and normalized areas of forest loss inside protected areas per country (km2) between 2000 and 2019. Net area of forest loss refers to the area of forest with >50% canopy cover. Normalized area of forest loss is defined as the ratio between the area of forest loss within protected areas (km2) and total protected area within pygmy hippopotamus distribution (km2) × 100. Areas of drivers of forest loss inside protected areas partially or totally within the pygmy hippopotamus distribution (per country; km2) are shown.
Distribution of drivers of forest loss within protected areas in the study area during 2000–2019. The bar chart shows the area (km2) of forest loss by driver inside the protected areas of each country (smaller patches of the pygmy hippopotamus distribution located in Sierra Leone are omitted).
‘Shifting agriculture’ was the main driver of forest loss inside protected areas across all countries, but especially in Côte d’Ivoire and Liberia, which accounted for 94% (4568 km2) of total forest lost. ‘Urban expansion’ was the second largest driver of forest loss and represented c. 5% (254 km2) of all forest lost inside protected areas. ‘Intensive agriculture’ represented <1% (13 km2) of deforestation caused inside protected areas (Fig. 4 & Table 2).
Discussion
‘Shifting agriculture’ was the dominant driver of deforestation within the pygmy hippopotamus range across all countries and inside protected areas. Shifting agriculture has been identified as the major driver of deforestation in tropical African forests; however, recent estimates by country are limited (DeFries et al. Reference DeFries, Rudel, Uriarte and Hansen2010, Fisher Reference Fisher2010, Hosonuma et al. Reference Hosonuma, Herold, De Sy, De Fries, Brockhaus and Verchot2012, Bodart et al. Reference Bodart, Brink, Donnay, Lupi, Mayaux and Achard2013, Giam Reference Giam2017, Curtis et al. Reference Curtis, Slay, Harris, Tyukavina and Hansen2018, Simmons et al. Reference Simmons, Law, Marcos-Martinez, Bryan, McAlpine and Wilson2018). While Hosonuma et al. (Reference Hosonuma, Herold, De Sy, De Fries, Brockhaus and Verchot2012) state that deforestation in Africa is driven by subsistence agriculture (c. 40%) and industrial agriculture (35%), our study found that the class ‘Urban expansion’ plays a more significant role than ‘Industrial agriculture’. In West African countries, where forest cover is high and economic development is low, employment, food and energy demands are met by clearing forest for different activities (Leblois et al. Reference Leblois, Damette and Wolfersberger2017). Human population density and growth are thus ultimate drivers of deforestation (DeFries et al. Reference DeFries, Rudel, Uriarte and Hansen2010).
The human population in the four countries has increased 307% over the past 70 years, which translates into more demand for land to supply food and natural resources (United Nations 2019, World Bank 2019). These patterns of deforestation caused by subsistence and shifting agriculture are more evident in developing than developed countries (McDowell et al. Reference McDowell, Allen, Anderson-Teixeira, Aukema, Bond-Lamberty and Chini2020), and, once forests are cleared for shifting agriculture activities, less work is required to further transform these areas into industrial agriculture or urban centres (Bren d’Amour et al. Reference Bren d’Amour, Reitsma, Baiocchi, Barthel, Güneralp and Erb2017). More research into how drivers of forest loss have changed over the past decades, informed by the historical empirical record from satellite imagery, is needed to investigate how best to manage human impacts on the remaining forests in West Africa and the species that inhabit them.
West Africa has been forecast as a major centre of rapid human population growth (Seto et al. Reference Seto, Güneralp and Hutyra2012). Although ‘Urban expansion’ was not the most important driver of forest loss in our study area, rapid population growth will probably involve further deforestation, even in protected areas (Fig. 4). According to Bogui et al. (Reference Bogui, Koffi, Koné, Ouattara, Kouakou and Gnagbo2016), increasing human presence close to remaining forests is likely to reduce pygmy hippopotamus populations, with human–wildlife conflict expected to increase as some hippopotami venture into farmland and plantations to find food (Robinson Reference Robinson1970, Bülow Reference Bülow1987, Hentschel Reference Hentschel1990, Conway Reference Conway2013). Further investigation is needed to better understand human–wildlife interactions (Conway et al. Reference Conway, Hernandez, Carroll, Green and Larson2015, Robinson et al. Reference Robinson, Flacke and Hentschel2017) and the proximate drivers of pygmy hippopotamus habitat quality across the UGF region.
Our results are consistent with Curtis et al. (Reference Curtis, Slay, Harris, Tyukavina and Hansen2018), who reported 13% forest loss in Africa, and Global Forest Watch (Hansen et al. Reference Hansen, Potapov, Moore, Hancher, Turubanova and Tyukavina2013), where Côte d’Ivoire, Liberia, Sierra Leone and Guinea have lost 26%, 5.9%, 11% and 6.6% of humid primary forest since 2002, respectively, meaning that the UGF has been heavily exploited in the past two decades (Table S3). Our results indicate that industrial agriculture in the study area corresponds mainly to oil palm, tea and cocoa; however, cocoa plantations are common across West Africa, with Côte d’Ivoire producing 40–50% of the world’s cocoa (Smith Dumont et al. Reference Smith Dumont, Gnahoua, Ohouo, Sinclair and Vaast2014, Wessel & Quist-Wessel Reference Wessel and Quist-Wessel2015).
Côte d’Ivoire and Ghana have the largest areas of oil palm plantations in Africa (200 000 and 100 000 ha planted, respectively) and are the largest exporters of palm oil in the UGF region (Descals et al. Reference Descals, Wich, Meijaard, Gaveau, Peedell and Szantoi2021, FAO 2021). Plantations of oil palm in Côte d’Ivoire and Ghana increased by c. 60% between 1989 and 2013 (Vijay et al. Reference Vijay, Pimm, Jenkins and Smith2016), and the world’s demand for palm oil is expected to grow threefold by 2050 from 2000 levels (The Economist 2014). West African countries, including Liberia and Côte d’Ivoire, are expected to increase their production to meet this demand. Despite this, Vijay et al. (Reference Vijay, Pimm, Jenkins and Smith2016) suggest that the increased area of these plantations may not be linked to deforestation but to conversion of previously deforested areas. This claim, however, cannot be verified without official figures of deforestation and oil palm plantations or via independently conducted studies of deforestation and land-change dynamics across the region. Our results suggest that although oil palm contributes to considerable forest loss in the UGF, it has been primarily concentrated in southern Côte d’Ivoire. If Vijay et al. (Reference Vijay, Pimm, Jenkins and Smith2016) are correct, only negligible new forest loss should occur due to industrial agriculture expansion in the coming years, potentially reducing the pressure on the remaining forest. Despite this, it is crucial to further understand how the pygmy hippopotamus and other threatened species use different ecosystems in the region to improve understanding of the impacts of various forest-loss drivers on threatened species.
A classification model that includes more drivers of forest loss (i.e., individual crops, fire) could improve conservation efforts (e.g., targeted policies and resources) by allowing decision-makers to understand not only the location of forest loss but also the frequency and intensity of the activities that drive deforestation, as well as to effectively monitor and evaluate the impacts of their policies, interventions and programmes. The high prevalence of clouds in the region and the spectral similarity of forests and plantations often lead to driver misclassification (e.g., see Curtis et al. Reference Curtis, Slay, Harris, Tyukavina and Hansen2018). These two limitations could be addressed with Earth observation data obtained from a variety of recently launched sensors that have improved temporal and spatial resolution. However, acquiring these data can be costly, and their use is likely to involve a number of trade-offs, such as reduced spectral resolution compared to Landsat and Sentinel-2 sensors, which may limit their ability to generate the diverse covariates necessary for a pixel classification approach such as that developed in our study. Our study therefore represents a first step to understanding how land cover is changing in the UGF and will help inform conservation efforts in it.
Forest loss inside protected areas
Almost 16% of the forest with ≥50% canopy cover inside protected areas was lost between 2000 and 2019, with ‘Shifting agriculture’ being the main driver of deforestation. The magnitude is this driver is 10-fold greater than the second most significant driver of deforestation, ‘Urban expansion’, inside the protected areas of all countries analysed here. Despite some methodological differences, our results align with those of Spracklen et al. (Reference Spracklen, Kalamandeen, Galbraith, Gloor and Spracklen2015), who found that Côte d’Ivoire had some of the highest rates of deforestation inside and within 10 km of protected areas (0.5% year–1) in West Africa between 2000 and 2012.
Our findings suggest that forest loss inside Taï (Côte d’Ivoire), Gola Rainforest (western Liberia) and Sapo (eastern Liberia) national parks is negligible and unrelated to the identified drivers assessed, which is not the case for other protected areas in the region. Taï, Gola Rainforest and Sapo national parks host several pygmy hippopotami populations (Bogui et al. Reference Bogui, Koffi, Koné, Ouattara, Kouakou and Gnagbo2016); therefore, these results are encouraging. There is, however, large-scale fragmentation alongside the western border of Taï National Park related to timber extraction and the establishment of new crops (Chatelain et al. Reference Chatelain, Bakayoko, Martin and Gautier2010). According to Chatelain et al. (Reference Chatelain, Bakayoko, Martin and Gautier2010), the remaining patches of forest to the west of Taï were expected to disappear, driven by land clearing for the expansion of cropland and existing settlements, and our results suggest that ‘Shifting agriculture’ activities are increasing in the area, but they are more prominent to the north-west and south-west of the National Park (Fig. 4).
Our results provide regional support for agricultural expansion and ‘Urban expansion’ being unambiguously linked to forest loss inside protected areas globally (Heino et al. Reference Heino, Kummu, Makkonen, Mulligan, Verburg, Jalava and Räsänen2015). Protected areas in West African countries are generally less effective at preventing forest loss than in other regions, with deforestation often occurring both outside and inside protected areas (Leberger et al Reference Leberger, Rosa, Guerra, Wolf and Pereira2020). We have demonstrated that protected areas within the pygmy hippopotamus’s range are affected by forest loss, and this may lead to human–wildlife conflicts.
Implications for conservation of the pygmy hippopotamus
We provide evidence that pygmy hippopotamus populations are and will continue be threatened by the expansion of shifting agriculture, grazing and human settlements. By identifying the proximate drivers of forest loss, we understand better the risk of extinction to the pygmy hippopotamus and other species in the UGF (Chatelain et al. Reference Chatelain, Bakayoko, Martin and Gautier2010). Our results indicate that Taï, Gola Rainforest and Sapo national parks account for 25% of the protected habitat in the pygmy hippopotamus distribution but only for 2% of the deforestation inside protected areas. In other words, these three protected areas provide almost all of the intact habitat for pygmy hippopotami and therefore should be given high priority for conservation. Importantly, these protected areas are spatially distributed across the region, hence ensuring the conservation of populations with different genetic pools. These protected areas present excellent sites to assess the demography, viability and extinction risk of pygmy hippopotamus populations as well as the effects of changing climate and forest loss on this species. The spatially explicit distribution of deforestation drivers presented here can inform conservation actions aimed at reducing forest loss while maintaining the remaining forests in the region. Critically, the reduction of agriculture and grazing activities inside and close to protected areas must be prioritized in Côte d’Ivoire and Liberia.
Conclusion
Our spatially explicit classification of deforestation drivers in the pygmy hippopotamus distribution area using the Sentinel-2 image archive and a random forest model shows that habitat loss primarily by ‘Shifting agriculture’ but also by ‘Urban expansion’ is pervasive. The protected areas in the UGF may not be fulfilling their role in providing pygmy hippopotami with protection from forest loss; therefore, it is imperative to prioritize conservation efforts in areas where deforestation is low (e.g., Taï, Sapo and Gola Rainforest national parks) and to assess the viability of pygmy hippopotamus populations in those areas.
Supplementary materials
For supplementary material accompanying this paper visit https://doi.org/10.1017/S0376892923000310.
Acknowledgements
We thank Dr Stephanie Duce, Centre Suisse de Recherches Scientifiques en Côte d’Ivoire, Google Earth Engine and NIESGI Cia. Ltda. We acknowledge the European Space Agency (Pesaresi et al. Reference Pesaresi, Ehrlich, Florczyk, Freire, Julea and Kemper2015) for making all Sentinel-2 images freely available. We also thank the anonymous reviewers who made useful comments to improve this manuscript.
Author contributions
EE-M, DP and NM conceived the study. EE-M, NY and NM developed the methods. EE-M and NY analysed the data. EE-M and NY led the writing of the manuscript. All authors critically reviewed and edited the manuscript and gave final approval for publication.
Financial support
EE-M was supported by a James Cook University Postgraduate Research Scholarship and HDRES Grant. NM was supported by an Australian Research Council Discovery Early Career Research Award (DE190100101).
Competing interests
The authors declare that they have no known competing financial, professional, contractual or personal interests or personal relationships that could have appeared to influence the work reported in this paper.
Ethical standards
None.
Data sharing
Google Earth Engine code for the analyses are accessible through Zenodo: https://doi.org/10.5281/zenodo.7091804 (Erazo Mera, Younes Cárdenas, Murray, 2022). All Sentinel-2 imagery is publicly available via Sentinel Hub: https://sentinel.esa.int/web/sentinel/sentinel-data-access or through Google Earth Engine.
Open access
