2.1 Study Area

There are three conservation areas in the forest agroecological zone of Ghana, that is, the Ankasa, Bia and Kakum conservation areas. This study was conducted in the ACA and BCA (Figure 1). The ACA and BCA were chosen because of the availability of cloud-free Landsat dataset for the epochs under investigation. Bailey et al. (2016) emphasised that conservation areas should not be analysed in isolation, as they are influenced by the surrounding landscape's land use, species and ecological processes. Therefore, a buffer of 5 km was created around the conservation area's boundary to define the study area. The reason for choosing this distance is that the Ghana Wildlife Division considers communities and related activities within a 5 km radius of conservation areas for management purposes. Additionally, the 5 km buffer zone around conservation areas is adequate for assessing the direct impact of anthropogenic drivers on conservation areas. The study area was divided into two regions: the regions inside the conservation area's boundaries, hereafter referred to as ‘on-reserve’ areas and delineated by the boundaries of the conservation areas. The ‘off-reserve’ represents the areas outside the delineated boundaries of the conservation areas but within the 5 km buffer.

2.2 Satellite Data Acquisition, Processing and Classification

The methodological flowchart for data analysis is illustrated in Figure 2.

A survey was conducted using a Global Positioning System (GPS) receiver to collect coordinates for the selected land use and land cover classes. Sampling points for the classes were extracted from high-resolution Google Earth images to complement the field sampling efforts. At least 50 training points were collected for each class and distributed evenly across the study areas. Seventy percentage of the coordinate points collected were used as training data, with the remaining 30% used for validation. The pixel-based non-parametric machine-learning Random Forest classification algorithm was used to classify the images into the land cover categories using the ‘random forest’ package in R software. The random forest classifier was chosen because it has been shown to yield a higher classification algorithm in the land cover mapping of Ghana's forest landscape compared to other classifiers (Ashiagbor et al. 2023). The error matrix was then calculated by overlapping the validation points on the resulting classified image. The error matrix was then used to compute the final LULC maps' overall accuracy, producer accuracy, user accuracy and Kappa coefficient.

2.3 Characterisation of Landscape Structure Using Metrics

Six complementary landscape metrics were used to characterise the structure (i.e., composition and configuration) of each landscape. The six metrics include patch density (PD), edge density (ED), landscape shape index (LSI), aggregation index (AI), largest patch index (LPI) and mean proximity index (MPI). PD is the number of patches per 100 ha (1 km²) and ranges from 0 (Forman 1995; Turner and Gardner 2015; Mathur and Mathur 2010; Asabere et al. 2020). A higher PD indicates fragmentation, or the division of a class into multiple patches. The ED, measured in metres per hectare with a minimum value of 0, which rises with increasing fragmentation and shape complexity, provides valuable information about the composition and configuration of a landscape (Asabere et al. 2020). The LSI compares the complexity of spatial object shapes to standard geometric objects such as a square or circles. It increases as the shape becomes more complex than a standard geometric unit. Increased LSI indicates increased fragmentation (Asabere et al. 2020; Forman 1995; Turner and Gardner 2015). The AI, which ranges from 0 to 100, measures how much the landscape is aggregated. AI is zero when the patches are wholly separated (no similar adjacent patches) and increases as the landscape becomes more grouped, reaching 100 when it becomes a single patch (Meng, Cai et al. 2023). The LPI represents the proportion of the entire landscape the largest patch occupies. The MPI measures isolation and fragmentation. This value increases as the area within the specified search radius becomes more occupied by patches of the same type, as those patches become closer and more contiguous and as their distribution becomes less fragmented. A low MPI indicates isolation, whereas a high MPI indicates proximity (Forman 1995; Turner and Gardner 2015). The landscape metrics were calculated using the spatial statistical tool FRAGSTATS (version 3.3). The categorical LULC raster for 1980, 2000 and 2020 of the BCA and ACA were used as input data sources for the landscape structure analysis.

2.4 Analysis of Changes in Landscape Structure

The post-classification change matrix was utilised to examine human-induced changes in forest cover. The non-parametric Wilcoxon signed-rank test was used to analyse statistically significant differences (p < 0.05) in landscape structure. The analysis was conducted pairwise by comparing changes from 1980 to 2000 and 2000 to 2020. The analysis was conducted at the landscape level. This was done by randomly subsetting 30 1 km × 1 km paired sample grids from the 1980, 2000 and 2020 landcover maps. The six landscape metrics, PD, ED, LSI, AI, LPI and MPI, were calculated for each grid, and the changes in metrics were compared statistically.

3.1 Accuracy of the Land Use and Land Cover Maps

The 1980 land cover map of the ACA achieved an overall accuracy of 78.7% and a kappa value of 0.73, indicating good agreement, as shown in Table 2. Similarly, the 2000 and 2020 LULC maps had overall accuracies of 82.2% and 90.3%, respectively, and kappa values of 0.77 and 0.88, respectively. On the other hand, the 1980 LULC map of the BCA had an overall accuracy of 77.5% and a kappa value of 0.71, also indicated in Table 1. The 2000 and 2020 LULC maps of the BCA achieved overall accuracies of 81.4% and 90.7%, respectively. These accuracies were deemed sufficient for the study.

3.2 Forest Cover Changes in the Ankasa and Bia Conservation Area

3.2.1 Ankasa Conservation Area

In 1980, closed-canopy forests covered 98,057.4 ha of the ACA. Of this, 49,718.1 ha (50.7%) were within the conservation area boundaries, with the remaining 49.3% in the off-reserve areas (Figures 3 and 4). Between 1980 and 2000, closed-canopy forests decreased to 72,040.2 ha, marking a 26.6% loss. Subsequently, closed-canopy forests further declined to 64,019.2 ha. In contrast, open-canopy areas expanded from 2884.2 ha to 22,820 ha between 1980 and 2000, but then decreased to 20,391.4 ha by 2020, with 1351.2 ha within the conservation area and 19,040.2 ha in off-reserve areas.

During the first change period, the increase in open-canopy forests in the off-reserve regions is linked to human activities such as logging, which has transformed closed-canopy forests into open-canopy ones. Figure 4a illustrates that the loss of closed-canopy forests mainly occurred in the off-reserve regions. However, during the second change period, open-canopy forests decreased due to agricultural conversions in the off-reserve regions. The study area saw an increase in agricultural conversions to food crops and tree crops, particularly in the off-reserve regions. No food crops and tree crops were identified within the boundaries of the ACA (on-reserve area). Tree crops, particularly cocoa and rubber, expanded from 1017.6 to 5095.6 ha between 1980 and 2000. The area under tree crop plantation increased from 5095.6 to 14,936.7 ha between 2000 and 2020. Food crops increased from 474.9 to 2849.8 ha over the 40-year study period (Figures 3 and 4). The rise in agricultural activities in the off-reserve landscape led to a decline in closed-canopy and open-canopy forests in those areas.

3.3 Changes in Landscape Patterns in the ACA and BCA and Their Catchment

In the following sections, the results of the landscape structural analysis are presented at two levels: (1) the landscape level and (2) the class levels. The class level changes show the dynamics of anthropogenic conversions (agricultural and logging activities) on forest spatial structure of forest cover in the study landscape. The metric-based landscape characterisation analyses are presented for each study area.

3.3.1 Ankasa Conservation Area

Figure 5 shows the results of changes in landscape level in the ACA from 1980 to 2000. In the ACA study area, the PD of focal classes increased from 28.02 patches/100 ha to 30.4 patches/100 ha between 1980 and 2000 and increased again from 39.58 patches/100 ha by 2020, which indicates a division of the classes into multiple patches. AI decreased from 84.5% in 1980 to 72.76% in 2020 and LPI also reduced from 58.6 to 32.45 between 2000 and 2020. This shows that focal and dominant classes in the ACA have been subdivided into smaller and fragmented patches with increasing levels of fragmentation across the landscape. The shapes of dominant classes in the ACA have become more irregular and complex, and this has been characterised by the continual increase in ED and LSI. ED increased significantly from 103.36 to 181.62, and LSI increased from 74.55 to 130 (Figure 5). The MPI, however, increased from 3698.47 to 7058.16, indicating that the dominant class in the landscape remains more contiguous and less fragmented. In summary, the results indicate a landscape experiencing fragmentation, with the focal patches becoming isolated, complex and irregular.

Figure 5 shows changes in structure in the closed-canopy and open-canopy forest classes as an indicator of the impact of agricultural expansion on the structure of forest cover in the ACA. Closed-canopy forests experienced a significant increase in PD from 0.25 patches/100 ha in 1980 to 5.9 patches/100 ha. This shows that the closed-canopy forest class in the ACA has been subdivided into smaller and fragmented patches. The LPI for the closed canopy forest class was reduced from 91.1 to 53.02 in the 40-year study period. ED, LSI and AI increased significantly from 1980 to 2020 (Figure 6). This shows that the closed-forest class in the ACA has become fragmented, and the patches have become irregular and complex in shape as a result of agricultural expansion. A visual inspection of the land cover classes shows how human agricultural activities have fragmented the forested landscape, mainly cocoa, rubber and food crops (Figure 7). The forest class has significantly reduced, and the patches have become isolated and disjointed.

3.3.2 Bia Conservation Area

Likewise, the BCA also experience fragmentation, with the focal patches becoming isolated, complex and irregular in shape (Figure 8: Figure 10). The PD, ED, LSI and MPI increased significantly between 1980 and 2020. The increase in PD from 28.02 patches/100 ha to 39.59 patches/100 ha shows that the landscape is subdivided into smaller fragments, increasing the number of patches per unit area. This is corroborated by the decrease in AI from 84.9% to 72.76%, implying increasing fragmentation. At the class level, the forest classes, that is, the closed-canopy and open-canopy forest classes, also experience some levels of fragmentation. The closed-forest patch continued to fragment into isolated patches, as indicated by the increased PD and declined AI (Figures 9 and 10). The decline in the LPI from 58.6 to 32.45 indicates that the largest closed-forest patch is getting smaller, indicating a loss of contiguous closed-forest habitat (Figure 8). The significant drop in the MPI for the closed-forest patch means that, on average, the close-canopy forest classes are further apart, indicating increased patch isolation. A visual inspection of the changes (Figure 10) in the landscape shows the fragmentation and disruption of forest habitat, loss of contiguous habitats and reduced patch size. The results (Figures 9 and 10) also point to a complex and irregular landscape and reduced connectivity amongst forest habitat classes.

4.1 Changes in Total Forest Area and Landscape Fragmentation in the ACA and BCA

The research revealed a significant loss in forest habitat to agricultural expansion in the off-reserve regions of the study area within the 40-year study period. This finding is confirmed in a related study by Abu et al. (2021), who observed that the concentration of cocoa farms is higher and more concentrated within the 5 km buffer of the protected areas in Ghana. The observed conversion of forests to cocoa and other crops in the off-reserve areas could be attributed to the significance of ‘forest rent’ whereby farmers utilise forest soils as a source of nutrients to enhance agricultural productivity (Ashiagbor et al. 2022). Also, as human populations grow, human activities continue to increase in the landscapes, increasing LULC conversions (Belete, Maryo, and Teka 2023; Rawat et al. 2021). Gockowski and Sonwa (2008) and Ajagun et al. (2021) observed that there are no longer large tracks of uncultivated land readily available for development by cocoa farmers and other tree crops, except for forest lands. Consequently, the expansion of these activities is bound to be at the expense of Ghana's tropical forest (Ajagun et al. 2021; Gockowski and Sonwa 2008).

The study found that despite increased forest conversion to agriculture within the 5 km buffer zone of the conservation areas, the ACA and BCA have benefitted from strong protection measures, as no agricultural encroachments were observed within their boundaries. This suggests that the management strategies of these protected areas have been effective (Bailey et al. 2016). This finding corroborates the study by Naughton-Treves, Holland, and Brandon (2005), confirming that parks are generally effective at curbing deforestation within their boundaries. Furthermore, Abu et al. (2021) investigated encroachment in conservation areas in Ghana and found no evidence of cocoa encroachment in the ACA and BCA. This may be attributed to the strict protective measures the Ghana Forestry Commission implements, which prevent encroachment.

However, the absence of agricultural encroachment within the boundaries of these conservation areas does not necessarily mean alleviating biodiversity threats or guaranteeing resident fauna (Bailey et al. 2016; Redford 1992). It has been observed that trees often remain in forests where the fauna has been depleted by poaching and hunting, as highlighted by Redford (1992). Holbech et al. (2018) reported that between 1990 and 2009, there was a decline in hornbill abundance in Ankasa by 39% and in Bia by 72%. They specifically reported that the IUCN Vulnerable Upper Guinea endemic Yellow-casqued Hornbill (Ceratogymna elata) was only detected in Ankasa, even though this species was once common in Bia and the Western Region, especially in Ankasa in the early 1990s. Similarly, the abundance of the Black-casqued Hornbill (Ceratogymna atrata) and the Western Piping Hornbill (Bycanistes fistulator) also declined by 61% and 85%, respectively, by 2009 compared to 1995. In addition, Oates et al. (2000) reported the extinction of Miss Waldron's red colobus monkey (Procolobus badius waldroni). This primate species was historically relatively abundant in the Bia and Ankasa landscapes.

Generally, the results of the fragmentation analysis point out that the forest patches outside the boundaries of BCA and ACA have become fragmented, disjointed and significantly reduced in size. This observed change was driven primarily by agricultural expansion, mainly cocoa, in both study areas. Agricultural commodities such as cocoa, oil palm and rubber are the most prevalent drivers of landscape transformation in Ghana and Cote d'Ivoire (Afriyie et al. 2021). Agriculture has been observed and discussed in the literature as the primary driver of landscape changes in Ghana (Kleemann et al. 2017; Manzoor et al. 2022). In both Ankasa and Bia, cocoa dominated and drove the transformation in the structure of the landscape in the once forest-dominated landscape.

4.2 Implications of Forest Cover Loss and Fragmentation for Conservation of the Bia and Ankasa

These forest-agricultural conversions in the buffer of the conservation areas can reduce the effective size of the conservation areas, reducing their ability to safeguard biodiversity and ecosystem functions (Bailey et al. 2016). Available scientific literature has related anthropogenic land use conversions around protected and conservation areas to human–wildlife conflict, bushmeat hunting, poaching and species extinction (Bailey et al. 2016; Meng et al. 2023). A total of 49 elephant crop damage occurrences involving 44 farms owned by 36 farmers were documented in a research by Sam et al. (2005) in the BCA. Various studies, including those by Sam et al. (2005) and Harich et al. (2013), have documented multiple conflicts between humans and wildlife in the landscape, and according to Sam et al. (2005), the frequency of raids increases as proximity of the farm to the park boundary reduces. According to Bempah, Dakwa, and Monney (2019), habitat loss through forest conversions to cocoa and other agricultural land uses has reduced the average species encounter rate from 10.2/km in 2008 to 3.7/km in 2016 in the ACA.

As indicated by the results, the conversion of open forests to cocoa and rubber could significantly impact the landscape's biodiversity. Cocoa and rubber, as monoculture crops, are reported in the literature to have low levels of biodiversity and significantly contribute to the loss of many species (Mang and Brodie 2015; Phommexay et al. 2011). The conversion of forest ecosystems to monoculture cocoa and rubber results in the destruction of diverse habitats and the displacement of various plant and animal species (Donald 2016). For example, the BCA has experienced significant biodiversity loss due to the expansion of cocoa. Daniel et al. (2018) found that converting forested areas into cocoa farms led to a decline in the abundance, richness and diversity of plant species within the BCA. Similarly, a report by the IUCN titled Protecting Ghana's Natural Heritage: A Biodiversity Assessment of Cocoa Farms within the Bia Conservation Area identified cocoa farming as a significant driver of biodiversity loss in the Bia region. This report highlighted the loss of critical forest habitats, the decline in plant and animal species and the disruption of ecological processes caused by cocoa expansion. These studies provide evidence that the expansion of cocoa, rubber and other monoculture plantations drives biodiversity loss and the overall degradation of ecosystems.

The immediate loss of forest cover in the off-reserve landscape and the significant levels of forest fragmentation, resulting in the loss of forest connectivity, have significant implications for wildlife habitat loss and overall degradation (Laurance, Vasconcelos, and Lovejoy 2000; Cagnolo, Cabido, and Valladares 2006; Cote et al. 2017). A plethora of studies have demonstrated that the conversion of both closed and open canopy forests to other land uses has profound effects on crucial ecological processes such as changes in species composition, distribution, abundance, primary production and biotic interactions (Ebagnerin and François 2015; Muhammed and Elias 2021; USEPA 1999).

The patches of open-canopy forest in the landscape are essential features, primarily ones found around conservation areas, as they serve as a habitat for some species, predominantly arboreal species in the area (Faria et al. 2006; Smith-Ramirez et al. 2010). These patches connect isolated areas of closed-canopy forests, enhancing the habitat quality (USDA 2004). They also serve as corridors, linking conservation areas to forest reserves and facilitating the movement of animals across the landscape, and also giving them access to a variety of habitat resources, which are often distributed across the landscape, enabling them to search for food and other essential resources (Collinge 1996; USDA 2004; McIntyre and Wiens 1999). However, the mobility of species is restricted by increased fragmentation in open-canopy forests (Yates and Muzika 2006). As it becomes difficult for individuals to travel across habitat patches, there may be an increase in inbreeding and a loss of genetic diversity (Laurance, Vasconcelos, and Lovejoy 2000). As a result, the population becomes less healthy over time, which increases its susceptibility to disease and the likelihood of extinction (Benedick et al. 2006). For example, the loss of forest connectivity, mainly driven by cocoa expansion and logging, has been attributed to the population decline of the Roloway Guenon (Cercopithecus diana roloway) and the white-naped mangabey (Cercocebus atys lunulatus), both of which are considered to be primarily restricted to the BCA (Asare et al. 2014; Oates, Struhsaker, and Whitesides 1997; Oates et al. 2000).

In 2000, the Wildlife Division of Ghana introduced the Community Resource Management Areas (CREMAs) programme, a Collaborative Community-Based Wildlife Management policy, to address the many challenges associated with wildlife conservation and habitat conversions around the protected areas in Ghana. One of the goals of the programme was the community's willingness to reserve remnant forests outside protected areas and to regulate further agricultural expansion (Osei-Owusu and Frimpong 2019). Despite this, these CREMAs have been characterised by a lack of participation and have yet to meet the implementation goals (Bempah, Dakwa, and Monney 2019).

Replacing monoculture cocoa with cocoa agroforestry in off-reserve areas can enhance the connection between PAs and forest reserves, leading to more available habitats for wildlife (Asare et al. 2014). Critchley et al. (2022) pointed out that increasing tree coverage in cocoa-growing regions in Ghana can help address fragmentation issues and improve forest connectivity. Cocoa agroforestry systems can potentially contribute to producing various ecosystem services compared to monoculture systems. Critchley et al. (2022) recommended prioritising the promotion of agroforestry systems that are more heavily shaded in regions near protected areas and conserved forests to facilitate the connection of habitats and the provision of ecosystem services.

The study did not consider the impacts of climate change, seasonal fluctuations, or other environmental factors that drive changes in the landscape. It also did not analyse the complex issues related to human–wildlife conflict, which are linked to human agricultural activities near conservation areas. Additionally, the study did not assess the effectiveness of conservation efforts or consider the perspectives and knowledge of local communities and conservationists. By recognising these limitations, managers of the ACA and BCA, as well as conservationists, can gain a better understanding of the complexities and challenges associated with conservation efforts in these areas.