The study area, Afi-Mbe-Okwangwo forest landscape, is composed of four protected areas namely Afi Mountain Wildlife Sanctuary (AMWS), Afi River Forest Reserve (ARFR), Mbe Mountains and Cross River National Park (CRNP) (Okwangwo division) all situated in the northern part of Cross River State (Figure 1). The entire landscape covers an approximate area of 176,835 hectares and is geographically located between latitudes 6˚04'19" to 6˚28'35" and longitudes 8˚49'55" to 9˚21'47".

The AMWS is the westernmost protected area that serves as home to estimated 25 - 30 Cross River Gorilla sightings. The sanctuary created in 2000 was established to protect wildlife in approximately 100 km2 of lowland and hill forests in the north-western corner of the 380 km² Afi River Forest Reserve (ARFR) [3] . The protected area is currently managed by the Cross River State Forestry Commission with assistance from the environmental NGO Wildlife Conservation Society (WCS). Other NGOs that have been involved region include Fauna and Flora International, Nigerian Conservation Foundation and Pandrillus. The AMWS is dominated by rocky hills with sparse vegetation and large grasslands making it a suitable roost for migratory European barn swallows (Hirundo rustica). [4] estimates 20 million winter birds, particularly swallows, roost in the suitable grasslands of the AMWS each year. The sanctuary is surrounded by sixteen villages and farms which has led to cases of encroachment beyond its boundaries, scattered logging in the lowland areas, occurrence of forest fires during farm clearing by farmers in farming seasons, poaching of endangered wildlife (such as the Cross River Gorilla, chimpanzee etc.) and over exploitation of non-timber forest products such as herbs for commercial and household use [1] [3] [4] [29] .

To the east of the Afi Mountains and adjourning the ARFR is the Mbe Mountains protected area, a forest dominated hilly landscape with sightings of gorillas since 1983 was established a wildlife sanctuary in 2000 [30] . It covers an estimate area of 100 km² and has hills rising to elevation of 900 metres. To the northern, western and southern parts of Mbe Mountains are nine villages (namely, Kanyang 1, Kanyang 2, Bamba, Abo Mkpang, Abo Obisu, Abo Ogbagante, Bokalom, Wula 1 and Wula 2) which have led to problems of encroachment by surrounding communities. The Mbe Mountains is an important corridor between AMWS and east of the CRNP (Okwangwo Division). The Mbe Mountains has been managed as a community wildlife sanctuary by the Conservation Association of the Mbe Mountains (CAMM) established by the earlier mentioned nine communities. The community wildlife sanctuary is managed by CAMM and supported by the CRS Forestry Commission and Wildlife Conservation Society (WCS). As is the case with AMWS, similar challenges from agricultural active-

ties, poaching, bush burning and fragmentation of existing forest landscapes are key issues of concern in the protected area. [30] confirmed the severity of threats from hunting, farm encroachment, fire and logging as threats to the ecological integrity of the landscape which left unchecked could affect the continued existence of the Cross River Gorilla and other primates present.

Further to the east of Mbe Mountains, divided by the Okon River is CRNP Okwangwo Division. The CRNP located in south eastern Nigeria consists of two divisions, the Oban division to the south of Cross River state and the Okwangwo division to the north as shown in Figure 1. The Okwangwo division of the CRNP, created in 1991, covers an approximate area of 4000 km² and comprises of the former Boshi, Okwangwo, Boshi Extension Forest reserves and some inhabited enclaves (Okwangwo and Okwe 1) [2] (Figure 1). The CRNP Okwangwo division landscape, dominated by mixture of primary and secondary high forests and savanna woodland formations, has a high biodiversity particularly with respect to primates like the Cross River Gorilla [31] . Although the wildlife in the park are in theory under protected status, hunting in the park is widespread and cases of fire in the montane forests on the edge of the Obudu Plateau a reoccurring event [3] . The gorilla habitats in the CRNP (Okwangwo division) are under immense threat from farming encroachment as the population within both enclaves has greatly increased over the years.

The field data on endangered primates (gorillas and chimpanzees) presence across the study area was collected using cyber-tracking technology employed by patrol officers of WCS and CRSFC [32] . The signs of gorilla and chimpanzee presence were based on vocalisation; presence of nests, food remains, dung and direct encounters. The use of cyber tracking based systems was introduced in the Mbe Mountains in 2009 and has since allowed for ease in tracking gorillas across the protected area since [32] . Other sources of endangered primate sightings were obtained from the WCS annual reports available on the official website (http://www.wcsnigeria.org/). The gorilla and chimpanzee presence data from 2011 to 2014 was used for this study. Past studies on gorilla distribution in Cross River observed that combining earlier surveys did not alter the current distribution of the primates, as recent surveys showed that all earlier known areas were still occupied [4] .

The change in the forest landscape across the study area was performed using satellite remote sensing data for two epochs (2000 and 2014). For this study Landsat 7 Enhanced Thematic Mapper (ETM+) with World Reference System path 187, row 056 (10 December 2010) and UK-DMC satellite data (7 January 2014) were used. The Landsat data was downloaded from the United States Geological Survey GLOVIS web-page (http://glovis.usgs.gov/) while the UK-DMC satellite data was supplied by Nigeria Space Agency, NASRA (National Space Research and Development Agency), based on sharing agreement with UK partners (Disaster Monitoring Constellation Company, DMCii). The Landsat image was processed to Level 1 Terrain-corrected, implying it was already ortho-rectified making it a suitable product for direct image-to-image comparison in change detection analysis. Similarly the UK-DMC satellite image was radiometrically and geometrically corrected. Since the images were already radiometrically and geometrically corrected on delivery, the Landsat and UK-DMC images were atmospherically corrected [33] and the digital numbers converted to ToA (i.e. Top of Atmosphere Reflectance) [34] [35] . This conversion process was done to remove all forms of noise present in the satellite images (such as instrument error, errors from view changes during data acquisition and effects of illumination geometry) [34] [36] [37] . A similar approach was implemented on the UK-DMC image used in the study. Before image classification, Landsat bands with same spectral ranges as the UK-DMC image were selected for input during image classification (Figure 2).

For this study, the supervised maximum likelihood classifier (MLC) was applied to each satellite image. A total of six broad classes were used in the study, based on the Intergovernmental Panel on Climate Change land use classification scheme [38] . The six broad classes comprised of forestland, cropland, grassland, wetlands, settlements and other land classes. For purpose of this study theses classes were further re-categorized into two distinct classes namely Forest and Non-forest. The process of image classification and accuracy assessment was performed using independent training and testing data. The training and ground truth data were obtained using a variety of sources namely Google Earth, GPS (global positioning system) data over the study area, historic/recent aerial photographs and visual interpretation of the satellite images.

In this study, the IDRISI Selva Land Cover Modeler was used to analyses forest cover change over the study area. Using results of the forest cover maps for 2000 and 2014, forest cover change maps across the protected areas were determined for 2000 to 2014. The forest cover change maps for both time intervals showed deforested, unchanged forest and afforested landscapes classes. In order to calculate, the annual deforestation and forest changes for the two time periods were used.

The open source software, FRASTATS (version 4.3.1.603), [39] was used to calculate the landscape and class metrics needed to better understand the spatial structure of forest cover classifications. The class metrics used in this study included the number of patches (NP), mean patch area (MPA) and patch density (PD) (which is the number of patches of a patch type divided by the total landscape area).

An important component of this study, with particular emphasis on the management of protected areas and eventual benefits to endangered primate habitats, was predicting the future status forest landscapes across the study area using current trends of forest cover transition. To this end, the Markov Chain Analysis and Cellular Automata Analysis (CA_MARKOV) module in IDRISI Selva was used to model forest cover scenario over a 14-year future projection (i.e. from 2014 to 2028). The Markovian process models the future state of any chosen landscape using the immediate preceding state as the basis for forward prediction [22] . For this study, the forest cover maps for 2000 and 2014 were used to model future predictions of forest cover over a 14-year period. Using the forest cover maps for 2000 and 2014 as inputs, the MARKOV module was used to produce the transition probability matrices and conditional probability images. To implement the CA_MARKOV module, the transition area text file, transition suitability image and 2014 forest map were used to generate a 14-year forward simulation of forest cover. The Cellular Automata iteration was specified as 14, allowing the model to be based on 1-year increments. The default cellular Automata filter of 5 × 5 was used for the study.

The Maximum Entropy Distribution (Maxent) [26] [27] was used to model the presence of endangered gorilla and chimpanzee primates across the study area. Maxent is a machine learning programme that utilises maximum entropy to estimate species distribution probabilities’ using the presence only data and environmental variables (categorical or continuous) [26] . For this study endangered primates’ presence data across the study area and environmental variable (continuous and categorical) layers were used as inputs into the modeling process. The open source software Maxent version 3.3.3k (http://www.cs.princeton.edu/~schapire/maxent/) was used to generate estimates of probabilities for primates presence varying from 0 to 1 (where 0 is the lowest and 1 the highest probability respectively). The presence data, Global Positioning System (GPS) positions of gorilla and chimpanzee nests and sightings for 2009 and 2013 were sample layer inputs used in Maxent to model potential distribution of environmental suitability for the endangered primates. Topographical data used in modeling was taken from the Shuttle Radar Topography Mission Digital Elevation Model (SRTM DEM). The SRTM DEM was downloaded from the Global Land Cover Facility (GLCF) website (http://glcfapp.glcf.umd.edu:8080/esdi/).

The modeling of environmental suitability for endangered primates using the presence data of gorilla and chimpanzee sightings and nests and three key environmental variables namely, DEM (for terrain information), Slope map (derived from the DEM) and forest cover map of the study area. The recommended default values for maximum iteration numbers (50), convergence threshold (100,000) and regularization values (10,000) were used. The random test percentage of value used was 30 (meaning that 30% of the presence data was used for validation while 70% for training of the model). For analysis a threshold was specified to help perform binary predictions of habitat suitability conditions with suitable habitats predicted above such threshold and unsuitable indicated as below. In this study a convergence threshold of 0.00001 was used with maximum iterations of 500. The ASCII raster outputs of predicted primate habitats were further analyzed using ESRI ArcGIS software version 10.2 [40] .

The overall land cover classification accuracy results were approximately 89 percent (2000) and 91 percent (2014) respectively. Table 1 presents the classification accuracy reports for the classifications performed. The classification accuracy results were within VCS REDD Methodologies specification [41] .

The classification result shows that total forest cover for the entire study area was 152,509 hectares in 2000 and 129,339 hectares in 2014 (Table 2). The total area of forest cover across the project area landscape declined from 86% in 2000 to 73% in 2014. Overall, the four protected areas experienced a general decline in the spatial

cover of forest landscape for the 14-year period investigated (Table 2).

The forest cover transition results indicated that overall deforestation for the study area from 2000 to 2014 was 34,620 hectares (Table 3). The total forest landscape that remained unchanged and afforested for the same period were 117,730 hectares and 11,574 hectares respectively. In terms of percentage landmass, 13.3% of the total forest landscape was deforested over the entire study area. Amongst the four protected areas, the most affected by deforestation were AMWS and Mbe Mountains with annual deforestation rates of 2.6% and 2.2%, followed by CRNP (1.1%) and ARFR (1.5%). The AMWS and Mbe Mountains, both known to have gorilla and chimpanzee sightings, are under immense pressure from illegal logging, encroachment farming and bush burning. The recent WCS annual report (2014) on AMWS and Mbe Mountains identified rise in the number of hunting incidents in the area, a major threat to the existence of endangered primate population. Furthermore, the close proximity of surrounding communities to the primate habitats has further heightened the pressures of farming encroachment and hunting activities, (particularly in forms of wire snares across the sanctuary). These factors have greatly contributed to the rise in deforestation and consequent disturbance of the gorilla and chimpanzee habitats across the protected area.

Using spatial metrics measures (NP, PD and MPA) over time, the changes in spatial structure over the 14-year period was investigated for the study area The NPs for the entire study area increased from 1255 in 2000 to 2843 in 2014. Similarly, the PD of the entire study area increased from 0.48 in 2000 to 1.12 in 2014 while MPA decreased from 122 hectares (in 2000) to 46 hectares (in 2014) (Table 2). The implication of these results was that the forest landscape across the study area experienced high levels of forest fragmentation over the 14-year period. Results of spatial structure dynamics for all four protected areas indicated that there were significant rises in NP and PD and a general decline in MPA (particularly for Mbe Mountain which experienced a sharp MPA decline from 4472 hectares in 2000 to 48 hectares in 2014). The effects of large decline in forest cover and increased levels of forest fragmentation over the 14-year period are indications of habitat loss, decreased connectivity of remaining vegetation (which prevents migration and movement of primates) and reduction in forest patch sizes (unsuitable for primate habitation). Other key deforestation drivers that have ravaged the Afi-Mbe- Okwangwo landscape include over-exploitation of non-timber forest products and occurrence of forest fires (a common practice in seasonal farming activities in the lowland forests) [1] [3] [4] . The combined effects of the afore-mentioned drivers of deforestation are serious threats to the conservation of the forest ecosystem, upon which existing primates depend for survival. Hence, the major challenge of the Cross River State Government and Nigeria at large is the establishment of suitable measures to combat or possibly mitigate these effects of de-

forestation across existing protected areas.

To this end, the Nigerian Government is actively involved in the United Nations REDD+ (Reducing Emissions from Deforestation and Forest Degradation) programme aimed at preserving existing forest landscape across the country, particularly forest dependent communities [42] . As part of the REDD+ programme in Nigeria, portions of Afi-Mbe-Okwangwo protected areas have been assigned as one of the three main REDD pilot sites for Cross River State. The implementation of the REDD+ programme across the state would facilitate sustainable management of forest landscape and its resources whist promoting wildlife conservation. Other approaches include creating public awareness on the need to refrain from killing endangered primates and avoidance of human activities such as agricultural encroachment, forest fires, illegal logging and over-exploitation of non-timber forest products. Figure 3 shows the forest cover (2000 and 2014) and forest transition (2000-2014) maps of the entire study area.

Figure 4(a) shows the spatial extent of forest cover across the entire study area, including the four protected areas from 2000 to 2028. The results show that for the entire study area, forest landscape experienced a gradual decline over the 28 year period. The spatial extent of forest decline from 2000 to 2014 and 2014 to 2028 were 23,170 hectares and 41,338 hectares respectively. The results of forest landscape decline were representative of approximately 15% and 27% of the earlier forest cover in 2000. Figure 4(c) and Figure 4(d) show the 2014 forest cover map and 2028 forward simulated forest cover map of the study area. At a glance there’s an obvious decline in the spatial coverage of forest cover in 2028.

The results of forest cover change analysis for the entire study area based on the 2014 and 2028 predicted forest maps showed that 28,121 hectares of forest cover would be deforested. The spatial extent of unchanged forest and afforested landscape between 2014 and 2028 would be 101,183 and 9988 hectares respectively. These

predicted results are based on the current trend of forest cover change driven by current forest management practices and pressures from human activities (such as farming encroachment, bush burning and illegal logging). Figure 4(c) shows forest landscape transition highlighting deforested, unchanged and afforested landscapes (for 2000 to 2014 and 2014 to 2028) across the protected areas. Overall, the results show that forest cover transition across the study area is prone to high levels of deforestation particularly outside and within the protected areas (Figure 4(d)). Given the current trend of forest transition, measures should be put in place to mitigate associated threats to endangered primates and human population.

The omission and predicted area plots of the endangered primates (Figure 5) show how testing and training omissions and predicted areas vary with the choice of cumulative threshold. Figure 5(a) and Figure 5(b) showed that the omission on test samples (for both primates) was a very good match to the predicted omission rate. Figure 5(c) and Figure 5(d) showed the receiver operating curve (ROC) for both training and test data used in the modeling process. The red (training) line shows the “fit” of the model to the training data while the blue (testing) indicated the fit of the model to the testing data (note, this is the real test of the models predictive power). Given that the blue line in graph 5(c) and graph 5(d) were towards the top left of the graph, it was an indication that the model was better at predicting the presence contained in the test sample of the data [43] . The accuracy of the predictive modeling for both primates was determined by the test area under the ROC (AUC).

The AUC and standard deviation for the modeled prediction of gorillas and chimpanzees habitats were 0.921 (standard deviation 0.006) and 0.921 (standard deviation 0.007) respectively. The gorilla suitability habitat modeling was performed using 359 presence records for training and 153 for testing. For the chimpanzee habitat suitability modeling 217 presence records was used for training and 93 for testing. The environmental variables used in the model included the study area DEM, classified forest map and slope map datasets.

During the training phase of species prediction modeling Maxent software keeps track of environmental variables that contribute to fitting the model. Consistent with ecological knowledge the model predicted that Cross River Gorilla and chimpanzees are found mostly in forested areas (particularly tropical high and montane forests) in mountainous locations across the site. Figure 6(a) and Figure 6(b) show the predicted habitat suitability map outputs of endangered gorilla and chimpanzee primates generated using presence data and key environmental variables in Maxent modeling software. The most important variables in order of importance for both primates as indicated by the Maxent model were DEM, Slope derived map and Forest map of study area.

The raster datasets showing predicted primate habitats generated using Maxent model were imported into ArcGIS software for further analysis and visualization. The raster outputs were reclassified to binary images showing only predicted primate habitats with possible probability of occurrence (i.e. 0.4 - 1.0) (Figure 6(c)). The binary images were digitized and compared with current and historic primate presence data. The results indicated that the total coverage of predicted suitable habitat for endangered primates over the entire study area was 30,940 hectares (54.4% situated in CRNP―Okwangwo division, 29.4% in AMWS, 14.3% in Mbe Mountains and 1.9% in ARFR). The spatial coverage of modeled primate habitats for both gorillas and chimpanzees for

AMWS, ARFR, Mbe Mountains and CRNP (Okwangwo division) were approximately 9094 hectares, 573 hectares, 4428 hectares and 16,846 hectares respectively. These results would prove vital in the effective management and development of conservation plans for endangered primate habitats across the study area required by wildlife conservation organizations and the State government.

Following the modeling of suitable gorilla and chimpanzee habitats, the authors investigated the effects of deforestation trends (2000-2014) on the spatial coverage of predicted primate habitats across the study area. Using spatial analysis techniques in ArcGIS, the forest transition map (Figure 3(c)) and predicted primate habitat map (Figure 6(c)) were reclassified to binary images, each having two classes (deforestation map with 1 = deforested landscape & 0 = others landscape and primate habitat map with 1 = primate habitats & 0 = others habitats). Using the afore-mentioned binary raster, a map showing deforestation hotspots across the predicted primate habitat were generated (Figure 6(c)). The results indicated that approximately 6468 hectares of predicted primate habitats were affected by deforestation (21% of the predicted forested primate habitat) in 2014. The effects of deforestation and forest fragmentation on the predicted primate habitats reflects the poor management of forest landscape in the protected areas across the study area, justifying the urgent need for existing management practices across the protected areas to be reviewed

The high levels of forest fragmentation across the four protected areas investigated in this study are largely dependent on the current prevailing agricultural practices across the region. The presence of intensified pressure from agricultural encroachment by surrounding villages and communities are key catalysts to the disturbance of existing gorilla and chimpanzee habitats. Over the years the combined effects of forest fragmentation and deforestation has resulted in extensive restriction of primates to isolated and highly fragmented forest patches. This pattern of forest transition combined with uncontrolled hunting in protected areas such as the CRNP (Okwangwo division) has resulted in rapid loss of endangered primates (particularly gorillas and chimpanzees) and in drastic cases extinction of some wildlife species such as the grey-cheeked mangabey (Lophocebus albigena) and crowned guenon (Cercopithecus pogonias) [2] [44] . The problems of hunting and agricultural encroachment are further heightened by the presence of large enclaves in CRNP (Okwangwo division). Similar occurrences are observed in AWFS, ARFR and Mbe Mountains.

In the Mbe Mountains, a key catalyst asides from the above mentioned factors, is the presence of the Ikom―Obudu highway which traverses across Ago Ogbagante, Kayang-1, Kayang-2, Wula-1 and Wula-2 communities. Results of past studies [2] [4] and current interviews have noted a distinct pattern in forest clearance along the Ikom―Obudu highway for agricultural practice by surrounding communities.

The devastating impact of deforestation and other human related activities on remaining forest landscape and wildlife (in particular endangered primates) justifies the need for Cross River State (CRS) Government (in collaboration with the National Government) to put in place mitigative measures and policies/laws that would protect all wildlife and forests. In this regard, the CRS government on February 2014 established a mobile court under the CRSFC charged with the responsibility of prosecuting any individual found guilty of killing or hunting any animal (or endangered species) as listed in the CRS Forestry Commission Law No. 3 of 2010 [45] . Another avenue explored by CRS is the implementation of the UN-REDD+ programme. In 2010 CRS Government demonstrated its commitment to the REDD+ programme by enacting a deforestation moratorium, which is still in force till date, banning all forms of logging in the state [46] [47] .