The authors thank the Strategic Support Programme for Scientific Research in Côte d'Ivoire (PASRES) for funding this study. Also, the authors thank the Centre of Excellence on Climate Change Biodiversity and Sustainable Agriculture (CEA-CCBAD) for offering a supportive work environment and a complementary research grant to the PhD student SIDIBE Ousmane.

Introduction

1Tropical forests cover 6% of the earth’s surface, but contain more than two-thirds of terrestrial plant and animal species, store huge amounts of carbon and provide livelihoods for millions of people (Bradshaw et al., 2009). Fragmentation of these forests has reached an alarming pace in developing tropical regions (Laurance & Laurance, 1999). Habitat fragmentation is defined as a process in which a large area of habitat is transformed into several small patches, smaller areas, isolated from each other by a habitat matrix different from the original (Wilcove et al., 1986). In addition to agricultural expansion, urban sprawl, mining and forestry which are the main causes (Seto et al., 2012), armed conflicts have become factors causing the fragmentation of tropical forests (Gorsevski et al., 2012; Nackoney et al., 2014; Ordway, 2015).

2The consequences of fragmentation of forest landscapes are generally well known. Fragmentation of the landscape increases habitat isolation and reduces forest patches (Bogaert et al., 2004; Barima et al., 2010). This situation not only alters forest dynamics and life cycles (Aguilar et al., 2006), but also the microclimate in forest patches (Laurance et al., 1998). As a result, fragmentation of natural habitats can lead to changes in the configuration, composition and functioning of forest ecosystems (Cabacinha & de Castro, 2009; Barima et al., 2010), and ultimately have several ecological effects on ecosystem (Matsushita et al., 2006). Therefore, it is important for deforestation analysis to include landscape dynamics which provides information on the temporal change in the patch metrics such as size, number, shape, adjacency and the proximity of patches in a landscape. According to Fahrig (2003), a detailed analysis of fragmentation processes requires a combined description of the “habitat loss” and “fragmentation” components in itself. However, these two components, which are generally treated together in deforestation analyses, should be individually described in detail for a better understanding of the fragmentation process (Mazgajski et al., 2010).

3The literature abounds with work on conflict-deforestation relationships (Ordway, 2015). But the question of the influence of conflicts on the degree of forest fragmentation remains little examined, particularly in Africa. Côte d'Ivoire is one of the developing tropical West African countries where the armed conflicts have contributed to the degradation of forest resources (UNEP, 2015). From 2002 to 2011, a political-military crisis caused a degradation of most natural forests, including those of some classified forests and protected areas (Bamba et al., 2018). The Mont Péko national Park (MPNP) is a perfect illustration of this.

4With a view to rehabilitating this protected area, a study on the determination of the impact of the crisis on vegetation dynamics was initiated in order to propose strategies for future action and management. Previous study revealed a sharp decline in forest areas in favor of illegal cocoa crops during and after conflicts (Sidibé et al., 2020). In addition to this pioneering study, this study aims to access the effect of conflicts on forest cover fragmentation in the MPNP. This work is based on the assumption that anthropogenic pressures during the crisis have exacerbated the fragmentation of MPNP forest ecosystems. Specifically, it will analyze the dynamics of landscape structure and then quantify the change in spatial structure resulting from fragmentation using landscape metrics according to the periods before (1988-2002), during (2002-2011) and after (2011-2016) the political-military crisis in Côte d’Ivoire.

Materials and Methods

Study Area

5The Mont Péko national Park (MPNP) is located in the western part of Côte d'Ivoire, more specifically between 6°53' and 7°08' N latitude and 7°11' and 7°21' W longitude. Straddling the departments of Bangolo and Duékoué, the MPNP was located during the crisis period in the non-belligerent “buffer zone” under the control of the UN and French forces (Map 1).

6The MPNP is part of the Guinean Forests of West Africa Hotspot, one of 36 tropical zones of global importance for their biodiversity (Goné Bi et al., 2013). It is a natural habitat for several animal and plant species, many of which are endemic and/or rare and threatened with extinction (Goné Bi et al., 2013; Lauginie, 2007). Examples include two large mammals at risk: Pan troglodytes, the Common Chimpanzee and Loxodonta cyclotis, the forest elephant (Lauginie, 2007). However, the political-military crisis triggered in 2002 in Côte d'Ivoire led to the collapse of the monitoring system for this Hotspot of biodiversity. This situation led to the occupation of the park by an armed band that encouraged the massive infiltration of the peasant population (Sidibe et al., 2018). In 2014, at the request of the Government of Côte d'Ivoire, the Ministry of Solidarity, Family, Woman and Child carried out a census of persons illegally infiltrated in the MPNP. This census supported by the Ivorian Office of Parks and Reserves (OIPR) counted about 20 622 people in 6 397 households (MINSFFE, 2014). This census also showed that 99% of the infiltrated population in the MPNP was of foreign origin, 96% of whom were from Burkina Faso and 3% from Guinea, Mali and Ghana. These people infiltrated in the MPNP created camps, large cocoa plantations, and engaged in poaching and logging among others (Sidibe et al., 2018). Despite the end of the conflicts in 2011, the park remained illegally occupied and exploited by the population until 2016.

7Creation Decree No. 68-79 of 9 February 1968 gives the MPNP an area of 34,000 ha. The topography of the park is mainly characterized by barely hilly plateaus of 300 to 500 meters of altitude and three well individualized summits in the North part (Avenard, 1971). It is the Mont Péko (1,004 m) to which the park owes its name, the Mont Gueï (996 m) and the Mont Kahoué (1,125 m; (Lauginie, 2007)). The climate of the region is subequatorial humid with two seasons, with the rainy from March to October and the dry from November to February (Eldin, 1971). The vegetation consists of semi-deciduous wet forest (Photo 1), degraded forest and a mosaic of crops and fallow land resulting from agricultural pressures (Sidibé et al., 2020).

Map Data Collection

8This study required the use of four Landsat satellite images of 30 m spatial resolution, covering the periods before (1988-2002) during (2002-2011) and after (2011-2016) the crisis. These images were downloaded from the United States Geological Survey (USGS) website https://earthexplorer.usgs.gov. These are the images Landsat Thematic Mapper (TM) of 14/12/1988, Enhance Thematic Mapper (ETM) of 13/12/2002 and 19/12/2011 and Operational Land Imager (OLI) of 27/12/2016. All scenes were acquired during the month of December, with 5 to 14 years in between acquisition dates. This month corresponds to the dry season in the area, settling from November to February with permanent sunshine and some isolated rainfall. The preprocessing of images and the development of land cover maps were carried out by Sidibé et al. (2020). A total of three land cover classes was taken into account: forests (primary or secondary), non-forests (crops/fallow) and bare soils and rock outcrops (Map 2). The maps derived from the supervised classification technique with maximum likelihood algorithm, through visual interpretation, field survey and Google Earth archive images of the area. The confusion matrix of classified Landsat images gives values between 89.38% - 97.60% and 0.82 - 0.91, respectively, for overall accuracy and Kappa coefficient (Sidibé et al., 2020). These values indicate better discrimination between different land cover classes (Landis & Koch, 1977).

Identification of spatial processes in landscape transformation (SPLT)

9For each land cover class, the spatial process in landscape transformation (SPLT) was determined before, during and after the crisis, based on the number of patches (NP), the area (A) and cumulative perimeter (P) of patches using a decision tree (Figure 1) proposed by Bogaert et al. (2004). According to this decision tree, there are ten possible processes of spatial transformations in the landscape. These are aggregation (merging of patches), attrition (disappearance of patches), creation (formation of new patches), deformation (change in the shape of patches), enlargement (increase in the size of patches), perforation (formation of holes in patches), shift (translocation of patches), shrinkage (reduction in the size of patches), fragmentation (break in the continuity of patches resulting in several loose patches) and dissection (subdivision of patches by small lines).

10NP_x, A_x, P_x and NP_y, A_y, P_y are respectively the number, area and cumulative perimeter of land cover class patches in year x and y where y > x. In order to distinguish between fragmentation and dissection, is t_Obs = A_y / A_x calculated and compared to a predefined threshold value t_Obs = 0.5; Barima, 2009). The dominant process is fragmentation if t_obs < t and dissection if not.

Forest fragmentation analysis

11There are many metrics in landscape ecology used to explain land cover change impact on landscape fragmentation (McGarigal & Marks, 1995). Metrics selected in this study were: number of patches (NP), mean patch size (MPS), edge density (ED) and landscape shape index (LSI). These metrics are commonly used to measure fragmentation (Venturelli & Galli, 2006; Lambert, 2010). NP and MPS are measures of fragmentation at landscape composition level, while ED and LSI are measures of fragmentation associated with landscape configuration (McGarigal et al., 2002). Table I provides a brief description of these metrics adopted to characterize the structural dynamics of landscape in MPNP throughout the periods before, during and after the crisis.

12First, metric value changes at the forest class level were calculated on the total park area scale to characterize the process of forest fragmentation during the different study periods. Secondly, in order to better study the park’s territory, the grid approach was used (Saura et al., 2008). Thus, the study area was plotted in 291 grids. These grids consist of 1 km² cells, except those located along the park boundary in areas slightly greater than or less than 1 km² (Map 3). Mapping of the metrics at the level of the «forest» class was carried out in the software ArcGis 10. This mapping was based on the results of a “moving window analysis” in the 1 km vicinity performed in the FRAGSTAT software (McGarigal & Marks, 1995). The visualization of metrics gives an overview of the structural dynamics of forest cover. Third, metric changes at the forest class level were calculated on the 1 km² MPNP grid over the study periods to assess the impact of the crisis on the observed forest fragmentation. The evolution of the selected metrics was evaluated according to the different study periods. Given the non-normality of the data, the Kruskal & Wallis (1952) test was used to test whether there is no statistically significant difference between the median likelihood in metric changes during the periods before, during and after the crisis. Pairwise comparisons using the Wilcoxon rank sum test of medians with Bonferroni adjustment to α = 0.05 were carried out to determine whether the trends in the fragmentation rate through metrics are identical for the different periods studied. In addition, the effect sizes were estimated using the “A₁₂” measure of the stochastic superiority of Vargha & Delaney (2000) for analyzing significance and magnitude of the result obtained. The Vargha-Delaney’s A₁₂ expresses the effect size on the basis of the probability that a random observation of one group scores higher than a random observation of the other group, that is:

where # is the count function Y₁ and Y₂ are vectors of scores for the two groups, and n₁ and n₂ are group sizes. Its measure is a value between 0 and 1: A₁₂ measure of 0.5 means that the groups are stochastically equal; A₁₂ measure less than 0.5 means that an observation of the first group is smaller than an observation of the second group; A₁₂ measure more than 0.5 means that an observation of the second group is larger than an observation of the first group. The interpretation of the A₁₂ statistic according to Vargha & Delaney (2000) is as follows: 0 – 0.29 => large effect; > 0.29 – 0.44 => medium effect; > 0.44 – < 0.56 => small effect; 0.56 – < 0.71 => medium effect; 0.71 – < 1 => large effect.

13The landscape metrics were calculated from the FRAGSTAT software and the statistical analyses were performed from the R software.

Results

Spatial processes in landscape transformation (SPLT)

14In MPNP, considerable changes were found in the landscape at class level between 1988 and 2016 (Table II). Different fluctuation of the number of patches (NP), the area (A) and cumulative perimeter (P) of patches in the landscape at the different dates was noted. Overall, forests have experienced a decrease in area with an increase in the number of patches and cumulative perimeters. This phenomenon is very marked between 2002-2011 and 2011-2016 covering the crisis period and the post-crisis period respectively. The non-forest (Crops/Fallow land) show an increase in area from 1988 to 2016 and in the number and cumulative perimeters of patches from 1988 to 2011. Both numbers and cumulative perimeters of patches decreased from 2011 to 2016. Bare soil/rocky outcrops with marginal proportions in the landscape experience a decline in area, patches and perimeters between 1988-2002 and 2011-20216 with large increases between 2002-2011. Comparative analysis of these fluctuations in these metrics made it possible to determine the spatial processes in landscape transformation (SPLT) carried out in the different land cover classes, before (1988-2002), during (2002-2011) and after (2011-2016) the crisis (Table III). For the forest class, the recorded SPLT is dissection (subdivision of patches by small lines accompanied by a decrease in total area and an increase in cumulative perimeters) before and during the crisis, then fragmentation (break in continuity in several disjointed patches accompanied by a decrease of total area and cumulative perimeters) after the crisis. The class of non-forest has had the same SPLT before and during the crisis,creation (formation of new patches accompanied by an increase of total area and cumulative perimeters) which then turned into an aggregation (melting of patches accompanied by an increase in total area and a decrease in cumulative perimeters) after the crisis. The SPLT of bare soils and rocky outcrops was first attrition (disappearance of one or more patches accompanied by a decrease in total area and cumulative perimeters) before the crisis, then creation during the crisis and finally attrition again, after the crisis. Overall, the SPLT observed in the MPNP show a phenomenon of forest fragmentation in favor of non-forests, particularly crops and fallow land (Photo 2).

Table II. Value of the number of patches (NP), area (A) and perimeter (P) of the different land cover classes of the Mont Péko national Park in 1988, 2002, 2011 and 2016

Metric	Types of land cover	Year
		1988	2002	2011	2016
NP	Forest	96	250	602	1139
	Non-Forest (Crops/Fallow land)	794	903	1413	359
	Bare soil/Rocky outcrops	151	144	1004	185
P (km)	Forest	624.06	873.96	908.72	1 571.88
	Non-Forest (Crops/Fallow land)	482.76	785.76	2 292.84	1 785.6
	Bare soil/Rocky outcrops	99.54	96.6	619.62	171
A (km2)	Forest	286.23	270.05	189.77	92.32
	Non-Forest (Crops/Fallow land)	12.80	29.13	94.71	201.78
	Bare soil/Rocky outcrops	3.45	3.30	18.00	8.39

Table III. Spatial processes in landscape transformation (SPLT) of the Mont Péko national Park before (1988–2002), during (2002–2011) and after (2011–2016) the political-military crisis in Côte d’Ivoire

Types of land cover	Metric	Period
		1988–2002	2002–2011	2011–2016
Forest	Number of patches (NP)	N2002 > N1988	N2011 > N2002	N2016 > N2011
	Area (A)	A2002 < A1988	A2011 < A2002	A2016 < A2011
	Perimeter (P)	P2002 > P1988	P2011 > P2002	P2016 < P2011
	SPLT	Dissection (tObs = 0,94)	Dissection (tObs = 0,70)	Fragmentation (tObs = 0,47)
Crops/Falow land	Number of patches (NP)	N2002 > N1988	N2011 > N2002	N2016 < N2011
	Area (A)	A2002 > A1988	A2011 > A2002	A2016 > A2011
	Perimeter (P)	P2002 > P1988	P2011 > P2002	P2016 < P2011
	SPLT	Creation	Creation	Agregation
Bare soil/Rocky outcrops	Number of patches (NP)	N2002 < N1988	N2011 > N2002	N2016 < N2011
	Area (A)	A2002 < A1988	A2011 > A2002	A2016 < A2011
	Perimeter (P)	P2002 < P1988	P2011 > P2002	P2016 < P2011
	SPLT	Attrition	Creation	Attrition

Spatial patterns of forest cover fragmentation

15In order to describe spatial patterns of landscape change, NP, MPS, ED and LSI metrics are computed at forest class level in MPNP. The evolution of metrics is presented in figure 2. The NP increased by 127, 303 and 550 respectively before (1988-2002), during (2002-2011) and after (2011-2016) the conflicts. This fragmentation of forest patches is accompanied by a reduction in their mean patch sizes (MPS) of 319 ha, 104 ha and 30 ha respectively before and after conflicts. The contour density (ED) of patches has increased of 9 m/h and 35 m/ha respectively before and during conflicts, followed by a decrease of 10 m/ha after conflicts. The landscape shape complexity (LSI) from class also has increased by 4, 25 and 6 from the front of the post-conflict. Overall, the evolution of the metrics throughout the periods studied indicates a trend of fragmentation of the forest cover which has intensified during and after the conflicts.

16The overall change (GC) is the difference in the values of a metric between two dates.

Effect of the crisis on forest cover fragmentation

17Map 4 shows the mapping of the NP, MPS, ED, and LSI forest class metrics obtained from the moving window analyses carried out in the FRAGSTAT software during the periods before, during and after the crisis. Visual analysis of the maps reveals a clear modification of the metrics in a 1 km neighborhood during and after a crisis.

18The distribution of NP, MPS, ED and LSI change scores by grids of 1 km² shows differences between the periods before, during and after the conflicts (Figure 3). With high standard deviations, these distributions exhibit high dispersion around the averages. As a result, A Kruskal-Wallis test was conducted to evaluate these differences among the periods on median change in metric variation scores by 1 km² grids. The test showed that there was a statistically significant difference in metric score variation by grids between the periods (Table IV). A post-hoc test using Wilcoxon rank sum tests with Bonferroni adjustment indicated significant differences between the periods studied, except for the pair periods “during-after” and “before-after” the crisis for the NP and LSI metric changes respectively (Table IV).

19The effect sizes in metric by the periods studied oscillated between small and large effects (Figure 4). Between 1988 and 2011, the value increases of metrics per grid of 1 km² were higher during the crisis period, compared to the post-crisis period in 78%, 76% and 70% of the time, respectively for the LSI, ED and NP metrics and lower at 80% of the time for the MPS metric. Also, the value increases of the metrics during the crisis period, compared to the post-crisis period were higher at 84% and 75% of the time, respectively for the ED and LSI metrics and lower at 49% and 36% of the time respectively for the NP and MPS metrics.

Table IV. Kruskal-Wallis test and Post-hoc using the Wilcoxon test for metric changes comparison at the level of forest class in Mont Péko national Park between 1988 and 2016

Metric	Summary statistics		Period		df	chi- square	p-value
		Before crisis (1988-2002)	During crisis (2002-2011)	After Crisis (2011-2016)
ΔNP	Median	0a	1b	2b	2	76.18	***
	95% CI	[0, 0]	[1, 2]	[1, 3]
	Mean rank	344	503	492
ΔMPS	Median	-0.63a	-34.07b	-11.20c	2	211.09 .	***
	95% CI	[-1.42, -0.36]	[-54.2, -24.1]	[-15.3, -8.64]
	Mean rank	615	315	409
ΔED	Median	2.67a	38.92b	-9.97c	2	264.08	***
	95% CI	[1.84, 3.73]	[32.1, 45]	[-14.2, -5.51]
	Mean rank	429	626	284
ΔLSI	Median	0.06a	1.17b	0.23a	2	168	***
	95% CI	[0.04, 0.1]	[1.05, 1.28]	[0.09, 0.33]
	Mean rank	372	363	604

Note. * = p < .05, *** = p < .001., CI = Confidence interval, Medians with differing subscripts within rows are significantly different at the p < .05 based on the Wilcoxon rank sum test. NP: Number of patches, MPS: Mean patch size, ED: Edge density, LSI: Landscape shape index. N = 291 for all analyses. The metrics were calculated from the study area grid using a 1 km x 1 km cell.

Discussion

Structural dynamics of the Mont Péko national Park between 1988 and 2016

20The analysis of landscape dynamics in MPNP between 1988 and 2016 revealed significant fragmentation of forest cover. The fragmentation rate has been greater during the post-crisis period of 2011 to 2016 than the crisis period of 2002 to 2011. Overall, the evolution of NP, MPS, ED, and LSI metrics throughout the periods studied confirmed that trend. This increase in forest fragmentation is largely related to the expansion crops and fallow land in the MPNP.

Forest cover dynamics

21The forest SPLT was recorded as dissection before and during the crisis, then fragmentation after the crisis. In reality, both phases are not strictly separated from each other. The dissection processes observed are intermediate phases of forest cover fragmentation in MPNP among others such as perforation, shrinkage and attrition (Forman, 1995; Jaeger, 2000). Fragmentation is materialized by a bigger gap between patches hence a lower area of patches compared to dissection (Barima et al., 2016). With t_Obs = 0.94 and 0.70 between 1988-2002 and 2002-2011 respectively, it could be concluded that dissection was less pronounced before the crisis than during the crisis. The dissection seems a key process of landscape “anthropisation” in MPNP since it directly increases the accessibility of the landscape through the creation of road networks. According to August et al. (2002), accessibility constitutes the trigger or a priori condition of landscape dynamics; without accessibility, the probability of anthropogenic effects on pattern is expected to decrease sharply. In MPNP, the fragmentation of forest during the post-crisis period is the phase which has aggravated the initial changes initiated by dissection before and during the crisis. If nothing is done to stop that trend, it will inexorably lead to the process of attrition of forest patches in the landscape.

22The changes in the spatial structure of the forest class quantifying in NP, MPS, ED and LSI metrics are coherent with the different SPLT identified before, during and after the crisis. During the periods studied between 1988 and 2016, the NP, ED and LSI metrics increased in value, except between 2011 and 2016 when the ED index decreased in value. At the same time, the MPS metric decreased in value over the entire study period. These trends point to fragmentation (McGarigal et al., 2002). The difference between the dissection processes recorded before and during the crisis through these metrics could be explained by their rate of evolution. These rates are of the order of 2.38, 3.89 and 5.26 times higher, respectively for the NP, ED and LSI metrics and 3.07 times lower for the MPS index during the conflicts relative to the developments recorded before the conflicts. As for the distinction between the dissection processes and that of the fragmentation recorded after the crisis, the decline in the value of the ED index between 2011 and 2016 would be the discriminating factor. The ED index reflects the complexity of the forms present in the landscape: the higher the density of the contours, the more complex the landscape and the units are intertwined. Thus, the decline in the value of the ED index after the crisis would suggest a decrease in the aggregation of patches at the level of the forest class. This would be linked to the significant loss of forest area recorded at that time (Sidibé et al., 2020).

23The landscape metrics provide information on the average trends of fragmentation. But they do not give spatially explicit information. To this end, the moving window analysis on NP, MPS, ED and LSI metrics are carried out in this study. The maps of the metrics at the level of forest class in 1988, 2002, 2011 and 2016 provide visual information about forest cover structural dynamics in MPNP. There has been an overall modification in metrics, but this does not occur uniformly across the landscape and in the time. The land cover dynamics maps of the MPNP (Sidibé et al., 2020), considered together with the results of the moving window analyses, show that the metrics have been more impacted during and after the crisis than before the crisis, in terms of landscape fragmentation. The modification in metrics is most detectable where there has been identified the loss and the smaller of forest patches in the based land cover mapping of Sidibé et al. (2020) between 1988 to 2016. This would suggest that there has been a considerable change in forest cover structure during and after the crisis. This is supported by the increasing in NP, ED, LSI range value and decreasing in MPS range value in the moving window analyses maps of 2002, 2011 and 2016.

Non-forest (crops and fallow) dynamics

24Contrariety to forest class, the non-forest has recorded creation before and during the crisis, then aggregation after the crisis. With the process of aggregation after the crisis, the mosaic of crops and fallow land becomes wider and better connected. Currently, a patchy landscape, dominated by a man-made matrix, is now observed in MPNP. Given that bare soils and rocky outcrops represent a marginal class in the landscape, it can be suggested that agricultural expansion is the predominant factor in forest fragmentation in the MPNP. The results of Sidibé et al. (2020) study confirmed this assertion. Several authors (Geist, Lambin, 2002; Lepers et al., 2005) identify agricultural land expansion as one of the main immediate causes of forest fragmentation, specifically in developing tropical countries (Geoghegan et al., 2001; Ochoa-Gaona, 2001). Generally, in Côte d'Ivoire, particularly in the MPNP, illegal cultivation of cocoa is recognized as the main factor in the use of forest land (Sidibé et al., 2018; Sidibé et al., 2020).

Bare soils and rocky outcrops dynamics

25The SPLT of bare soils and rocky outcrops was first attrition before the crisis, then creation during the crisis and finally attrition, after the crisis. The processes of attrition of the bare soil class and rocky outcrops observed before and after the crisis, could be explained by the conversion of certain areas of this class to cultivation and fallow. Indeed, the transition matrix of the land use dynamics of the MPNP during the different periods studied indicates a conversion of the patches of this class into cultivation and fallow (Sidibé et al., 2020). Therefore, it should be noted that the process of creating soils and rocky outcrops recorded during the crisis (2002-2011), would be due to the establishment of camps inside the MPNP (Sidibé et al., 2018) and forest clearing or weeding of existing parcels.

Political-military crisis as a factor for forest cover fragmentation

26This study shows that the crisis period and the post-crisis period influenced forest fragmentation in the MPNP more than the period before the crisis. This result confirms that of Barima et al. (2016) which shows that it was during the crisis period and the post-crisis period that the fragmentation of the classified forest of Haut-Sassandra was more intense. Robert-Charmeteau (2015) made the same observation in his study of the landscape dynamics of A Lưới forests after the Vietnam War. In Côte d’Ivoire, several studies on factors influencing forest fragmentation have also identified conflicts as a driving force (Sangne et al., 2015; Barima et al., 2016; Bamba et al., 2018).

27The intensification of fragmentation in the MPNP during these periods would be due to the occupation and accelerated exploitation of forest resources by an armed band that previously controlled the park and the infiltrated peasant populations. In fact, between 2011 and 2013, the MPNP was still entirely controlled by the armed band, which, moreover, undertook the various forms of exploitation in order to obtain maximum profits. In 2013, despite the exfiltration of this strip, the park remained occupied by the peasant populations who maintained the existing plantations and were engaged in new clearing.

28Overall, the effect of the crisis on the fragmentation of MPNP forests was greater during the crisis than before the crisis and after the crisis than during the period of the crisis. This dominance is clear between the period before and during the crisis. Indeed, the A₁₂ measures for this pair of periods accounted for metric changes is generally large in favor for the period of crisis. However, the MPS A₁₂ value measured is particularly low. Since the analyses of the effects covered the differences in the value of the metrics for each period, for the MPS metric which naturally experiences a drop in value when the fragmentation rate increases, the effect sizes (A₁₂ statistic) is decreasing with fragmentation. Thus, lower values of A₁₂ statistic at the MPS level indicate a sharp reduction in the mean patch size in the landscape. For “during-after” crisis pair period, the metric changes are different with “large” effect for the ED metric (A₁₂ = 0.84) and LSI metric (A₁₂ = 0.75), “moderate” effect for the MPS metric (A₁₂ = 0.36) and “small” effect for the NP metric (A12 = 0.75). The A₁₂ value of 0.49 recorded for NP metric during the crisis, indicates that the probabilities of increases in the number of patches per km² are substantially equivalent to that of the post-crisis (A₁₂ = 0.51). However, as the number of patches increased by 303 during the crisis and by 550 after the crisis period, it can be deduced that the fragmentation process was more intense per km² after the crisis than during the crisis.

29The inferiority of ED an LSI metrics per grid of 1 km² during the post-crisis period in 84% and 75% of the time would be due to the disappearance of several forest patches at this time. The A₁₂ value recorded during the crisis for the MPS value is lower than the value recorded after the crisis. This shows that the reduction in the mean patch size per grid of 1 km² is greater during the crisis than after the crisis. This situation would be explained by the fact that the forest patches during the post-crisis period were smaller in size compared to the crisis period.

30Each of the metrics used confirmed the effect of conflicts on forest fragmentation. This finding supports the strong correlation between landscape metrics and their redundancy found by some authors such as McGarigal & Marks (1995) and Bogaert et al. (2002). In addition, the results of this study confirm the sensitivity of the landscape metrics to changing levels of fragmentation and their ecological relevance, as highlighted by the results of the work of Bogaert (2003). The use of these metrics is therefore appropriate for the assessment and monitoring of forest fragmentation. The choice of grid system adopted in this study is not new. According to Bamba (2010), grids allow the area to be subdivided into several samples in order to generate a proximity database that best relates information about the study area. In this grid approach, the size of grids varies among authors (Bamba, 2010; Sader et al., 1994). Compared to this study, some authors have also used grids with 1 km x 1 km cells in forest fragmentation studies (Barima, 2009; Lambert, 2010). But this method is limited by the high proportion of spatial autocorrelation that could exist between neighboring grids (Bamba, 2010). However, since the Kruskal-Walis test was based on a comparison of median averages, this spatial autocorrelation effect between grids was mitigated.

Conclusion

31This study provides a good understanding of the forest fragmentation in Mont Péko national Park between 1988 and 2016 for judging the impacts of human pressures associated with the political-military crisis in Côte d'Ivoire from 2002 to 2011. It used landscape metrics to analyze the effect of the periods before (1988-2002), during (2002-2011) and after (2011-2016) the crisis on the structural dynamics of the forest cover. The aim was to test the hypothesis that human pressures during the crisis accentuated the fragmentation of the forest cover in the Mont Péko national Park. Analysis of the periods before and after the crisis shows that armed conflicts have, to a large extent caused the fragmentation of the forest cover in the Mont Péko national Park. The impact was very strong after the conflicts only during the conflicts. This increase in forest fragmentation is largely related to the expansion of illegal cocoa crops in the park. Currently, the protected area’s forest ecosystems are presented as islets in a landscape matrix dominated by a mosaic of crops and fallow land. The impairment of the integrity of the park’s forest ecosystems has affected both its composition and configuration. As a result, studies on the functioning of these disturbed ecosystems should be conducted as a complement to this study to assess their dynamics, microclimate and life cycles. In addition, more comprehensive and continuous study on habitat fragmentation and its harmful effects may provide necessary information to identify potential areas for systematic conservation planning. Besides, monitoring of the Mont Péko national Park deserves to be strengthened in order to compensate for the various degradation phenomena of its forest ecosystems.