This study was conducted in Ndoinet Forest (latitude 0˚33' South, longitude 35˚21' East), a sub-block of South West Mau Forest Block in Bomet County (Figure 1). It covers a total of 20,032 ha of land area. The forest and its surrounding areas are characterized by humid and wet climate. The climate is influenced by

prevailing winds and humidity from Indian Ocean and the Congo rain forest through Lake Victoria Basin and the high altitude. The rainfall pattern is bi-modal with long rains experienced between March and July while short rains fall in September to November with January to March being the driest months. The mean annual precipitation is 2000 mm. Ndoinet forest and its environs have moderate temperatures. The mean daily minimum and maximum temperatures are 14˚C and 23˚C respectively [32] .

Ndoinet forest is a catchment area for major rivers including Kipsonoi, Kiptiget, Simbeiwet, Chemosit, Songol, Ndoinet and Chesirere, the main tributaries to Itare River which feeds Sondu Miriu River, important for hydropower production before draining into Lake Victoria. The forest is also the source of river Chepkulo which drains its water into the Mara River which passes through the world-famous Maasai Mara National Reserve. The forest therefore, is a key source of water for Kericho and Bomet counties and other downstream users [32] . The soils in the area are pre-dominantly dark loamy while some parts are characterized by clay and black cotton soils. This is as a result of the variation by the dominant geological formations of deposits and age of parent rock material [32] .

The vegetation type and distribution in the forest is influenced by topography, edaphic characteristics and anthropogenic activities. The forest has four main vegetation types: natural forest, bamboo zone mainly composed of Oldeania alpina (syn. Yushania alpina), grassland and tea zone [32] . The forest has high species diversity with dominant indigenous trees in the forest being Tabernaemontana stapfiana, Neoubotonia macrocalyx, Macaranga kilimandscharica, Podocarpus latifolius and Dombeya torrida. Ndoinet forest is a home to diverse mammals, birds and invertebrates. However, there has been decline in the populations of wildlife due to habitat loss and land use change [33] .

The forest is a habitat to unique animal species such as the Mountain bongo (Tragelaphus eurycerus). The common mammalian species are Colobus monkey (Colobus guereza), Baboons (Papio anubis), Bush pigs (Potamochoerus larvatus), Porcupine (Hystrix cristata), Hyena (Crocuta crocuta) and Honey Badger (Mellivora capensis). Buffaloes (Bubalus bubalis) and Elephants (Loxodonta africana) are seldom sited in deep intact areas. The common bird species are Green Sunbird (Anthreptes rectirostris), Golden-winged sunbird (Nectarina reichenowi), Baglafecht weaver (Ploceus baglafecht), Tullberg’s woodpecker (Campethera tullbergi), Cattle egret (Bubulcus ibis) and Red-chested owlet (Glaucidium tephronotum) [32] [33] . The forest is also a habit for reptiles (lizards and chameleons), amphibians, insects and molluscs.

Ndoinet Forest is gazetted vide the Kenya Gazette Legal Notice No. 44 of 1932 with an aim of forest conservation. The forest is under the jurisdiction of the Kenya Forest Service, the institution mandated to manage all gazetted forests in Kenya. A section of Ndoinet Forest was illegally occupied by humans since 1950s with rapid influx of settlement from 1990s. This human encroachment led to extensive degradation of the forest through deforestation, conversion to crop fields and settlements, and uncontrolled livestock grazing. Consequently, the government carried out evictions between 2005 and 2009. This study was conducted approximately 10 years after cessation of human settlements in the forest. Notably, however, livestock owned by the local communities adjacent to the forest have continued to graze in the forest even after cessation of human settlement.

Selected woody vegetation and grazing-related disturbance attributes were assessed during the rainy season in June 2019 approximately 10 years after the 2009 cessation of human settlements. Vegetation attributes such as species richness, diversity, frequency and similarity, stem density, diameter at breast height (DBH), plant basal area and importance value index were determined. Grazing-related disturbance indicators assessed were presence of livestock, evidence of dung deposits, evidence of plant damage through trampling or browsing, presence of livestock tracks and evidence of soil erosion (e.g. bare ground).

Purposive stratified sampling technique was employed [34] [35] . Prior to sampling, an exploratory survey was conducted to stratify the forest based on the degree of human-induced disturbance, majorly cultivation. Three distinct strata (light, moderate and heavy disturbance) were identified based on historical reconstruction, through participatory mapping, focusing on intensity and duration of disturbance and canopy cover [35] [36] . The light disturbance stratum comprised of areas with minimal short lived disturbance and served as baseline for the study and exhibited relatively closed canopy cover (>50%). The moderate disturbance stratum comprised sites which were deforested and cultivated for about 5 - 10 years before cessation of encroachment and occurred as semi-open forest sites with moderate (~50%) canopy cover. The heavy disturbance stratum was exposed to intensive and continuous disturbance of more than 10 years and were still largely grassland with scattered residual trees with none or minimal (<10%) tree canopy cover. The disturbance levels declined with increasing distance from the human settlement.

Vegetation and grazing disturbance attributes were assessed along line transects located in three of the forest’s administrative units (also known as “beats”), namely, Kapkembu, Kipkoris and Chematich. For each administrative unit, sampling was performed along a transect that extended from the edge of the forest bordering the farmlands towards the interior forest core. Each transect traversed the three delineated human-induced disturbance strata, with the outer (forest edge), middle and inner (forest core) segments of the transect coinciding with the heavy, moderate and light disturbance strata, respectively. Nested quadrats were employed, with different-sized plots being used to sample different woody plant age classes; big trees (DBH > 10 cm), saplings (DBH > 2 but <10 cm) and seedlings (DBH < 2 cm). Specifically, 10 m by 10 m plots were used to sample the big trees and grazing disturbance attributes, whereas 5 m by 5 m and 1m by 1m sub-plots were used to sample saplings and seedlings, respectively. The different-sized plots were nested at each location such that they all shared a common corner. Plots were located systematically along each transect at an interval of about 100 m. In situations where the sample plot fell on the glades, the plot was not sampled; rather, a subsequent plot was sampled 100 m from the edge of the glade. Overall, for each plot size, a total of 28 plots were sampled, with 10, 9 and 9 of these plots being located on heavy, moderate and light disturbance strata, respectively. The overall layout was a split-plot design, with administrative unit, disturbance stratum and age class representing the blocking, whole-plot and split-plot factors, respectively.

Woody plant species were identified by their botanical and local names with the assistance of an experienced botanist. We recorded presence/absence of different tree species in sample plots. Plant DBH of all live trees within the plot was measured at a height of 1.3 m above the ground using a diameter tape. Tree and saplings which had two stems were measured as separate individuals; for trees and saplings with multiple stem (3 or more), the numbers of stems were counted and mean DBH recorded. At each sampling location, all individual plants belonging to the different age classes within their respective plot size categories were counted. Presence of grazing-related disturbance attributes was recorded based on visual assessment.

The total species richness of the forest was computed as the total number of species recorded in the entire sampled area. The diversity was computed using Shannon Weiner diversity index [37] , Equation (1).

H ′ = − ∑ i = 1 s ( p i ln p i ) (1)

where H' is Shannon Weiner diversity index; P_i is the proportion of the i^th species within the sample (p_i = n_i/N, n_i is the number of individuals of the i^th species sampled and N is the total number of all the tree species sampled.

Species similarity was computed using Jaccard index of similarity [38] as described by Equation (2).

S j = n c n a + n b + n c (2)

where S_j is the similarity index, n_c is the number of shared species between the two sites and n_a and n_b are the number of species unique to each site.

The frequency, stem density (stems ha⁻¹) and basal area (m²·ha⁻¹) were computed using the formulas as described in Equations (3), (4) and (5) respectively [39] .

F = N p T p (3)

where F is Frequency; N_p is number of plots in which a particular species occur and T_p is total number of plots sampled.

D = N O A S (4)

where D is the stem density (stems ha⁻¹), N_o is number of individuals, while, A_S is the total area sampled (Ha).

BA = π ∗ D 2 40000 (5)

where BA is the basal area (m²·ha⁻¹), π is pie and D is the diameter at breast height (cm).

Importance value index as indicator of species dominance in each disturbance level was computed as summation of relative density, relative frequency and relative basal area.

The above attributes were calculated separately for each of the three age classes.

Social surveys were conducted to assess the extent to which local communities grazed their livestock in the forest, and local community perceptions on palatability of different tree species to livestock (particularly cattle and sheep) at seedling stage. This is because livestock have insignificant effect on trees with a height of about 2 m and above. To assess the extent of community livestock grazing in the forest, household survey was conducted comprising a sample of 381 households drawn from a total of 5235 households located within 5 km of the forest boundary. A two-stage random sampling procedure was used to select the sample. First, from each of the four sub-locations bordering the forest (i.e., Kapkembu, Kapno, Chematich and Kipkoris), nineteen villages were randomly selected since the population was homogeneous. Secondly, a given number of households from each village were randomly selected based on the proportional representation of that village in the total household population. Random selection of households within each village was performed using a detailed household list provided by the village elders in consultation with the local administration and community leaders. A semi-structured questionnaire was administered face-to-face to the household head or, in his/her absence, the spouse or the eldest adult child was interviewed. The information collected from the households was on livestock ownership and types, level of dependence and seasonal utilization of the forest for grazing.

To assess local perceptions on the palatability of different tree species to livestock; three Focus Group Discussions (FGDs) were held with the chairpersons of forest grazing user group, Kenya Forest Service officers, local administration officers and knowledgeable people with long-term grazing history in the forest. The FGD discussants were selected solely based on their knowledge on forest grazing. Three palatability categoriess, namely, highly palatable, palatable and unpalatable were applied in the assessments. Highly palatable plants comprised plants that were deemed to be mostly preferred and frequently browsed by cattle and sheep at seedling stage. Palatable plants included those that were considered to be less preferred and are occasionally browsed by livestock when fodder resources are limited. Unpalatable plants were those that were reported to be totally avoided (i.e., not browsed) by livestock.

For all vegetation response variables other than frequency, species similarity and importance value, data was analyzed using linear mixed models (LMMs). In these analyses, disturbance stratum, plant age class, and their interaction were specified as the fixed factors, whereas transects (administrative units) and plots nested within transects were specified as the random factors. Before analysis, the data sets were tested for normality using both Kolmogorov-Smirnov and Shapiro-Wilk test to ensure that the model conformed to the required assumptions for analysis. Pairwise comparisons were done across the disturbance regimes and means separated with Bonferroni post hoc test. Statistical significance was accepted at P < 0.05. All statistical analyses were carried out using SPSS version 21 software. Data on species similarity, importance value and frequencies of different plant species and grazing disturbance indicators were analysed descriptively. Likewise, data from household surveys were analysed descriptively. Additional information from FGD was used to triangulate household findings.

Majority of the households (89%) adjoining Ndoinet forest kept livestock as a source of livelihood. The main types of livestock kept were cattle (55%), sheep (37%), goats (6%) and donkeys (2%). A household kept on average about 16 livestock composed of 8 ± 1 cattle (4 dairy, 2 beef and 3 dual purpose cattle), 6 ± 1 sheep, a goat and a donkey. The owners grazed their livestock both in the forest (88%) and at home. Grazing in the forest occurred throughout the year (96%), but it was proportionately higher (99%) in January which coincided with the peak dry period. About 73% of the households that owned livestock grazed exclusively in the forest, 12% grazed exclusively at home, while 15% grazed both in the forest and at home. According to the FGDs, the forest fodder resources mostly utilized were the grasses, bamboo, shrubs, herbs and tree regenerates. Bamboo shoots and grass formed the main feed resources during the rainy season as they flourish and occur in abundance while the herbaceous and woody vegetation were browsed on mostly during dry seasons.

The combination of all livestock induced disturbance indicators were notably high in heavily and moderately disturbed forest areas with relative disturbance of 50% and 30% respectively compared to lightly disturbed sites (20%). About 97% of the sampled plots showed evidences of vegetation damage through livestock trampling, browsing and grazing. Apart from cattle which were spotted in both heavy and moderate disturbance regimes, the rest of the livestock types (sheep, goats, donkeys) were not recorded in the light and moderate forest disturbances, but all livestock types were present in the heavy disturbed areas. Dung as an indicator of livestock presence was 79% and 46% higher in heavily disturbed forest sites than in lightly and moderately disturbed levels respectively ( Table 1). The presence of soil erosion was not recorded in the sampled areas, but was only observable in the main livestock tracks leading into the forest and watering points. According to the knowledgeable forest grazers and validation through literature review, 55% of the sampled woody plants were classified as unpalatable, 36% as palatable and only 9% perceived to be very palatable ( Table A1).

The sampled forest area constituted a species richness of 33 belonging to 22 families ( Table A1). Fewer tree species shared the same plant families. The plant families which had multiple tree species were Euphorbiaceae (12.1%), Fabaceae (12.1%), Rutaceae (9.1%), Araliaceae (6.1%), Oleaceae (6.1%) and Salicaceae (6.1%). The rest of the plant families were represented by a single species. Generally, the species richness was 83% and 84% lower in heavy disturbance (3) than in moderate (18) and light disturbance (19) respectively. The mature trees richness sampled in the forest were 20, while the seedlings and saplings had species richness of 16 and 22 respectively. Markedly, species richness varied across the age class in each disturbance regime ( Table 2). The tree species richness was 75% and 25% lower in heavy disturbance levels than in light and moderate disturbance regimes. Seedlings and saplings were not recorded in the heavily disturbed forest sites. The sapling richness was 50% higher in moderate disturbance than light disturbance level, while seedling richness was 42% lower at moderate levels of disturbance compared to light disturbance ( Table 2).

The most common tree species in the forest were Neoubotonia macrocalyx, Podocarpus latifolius, Syzygium cordatum and Tabernaemontana stapfiana each with a relative frequency of 10.87%. The key species of high economic value

Values denoted by the same superscript letter indicates no significant variation while those that have different superscript letters denotes significant mean difference as per Bonferroni post hoc test at 0.05 significance level.

which were least frequent in the forest were Hagenia abyssinica, Olea capensis and Zanthoxylum gilletii each with relative frequencies of 2.17%. In heavily disturbed areas, common species were Acacia melanoxylon, N. macrocalyx and Olea africana each with relative frequency of 33%. The species that dominated the moderately disturbed areas were N. macrocalyx and S. cordatum each relative frequency of 78%, while T. stapfiana, M. kilimandsharica, P. latifolius and S. cordatum were common in lightly disturbed areas each with relative frequency of 67%.

The Jaccard similarity index showed that species similarity was low among the disturbance regimes; heavy versus moderate and light disturbance intensities showed a similarity index of 0.05 and 0.12 respectively, while the moderate versus light disturbance exhibited similarity index of 0.09. The overall species diversity of the forest was H' = 2.72 and declined significantly with increasing disturbance levels. The diversity was 59% and 61% lower in the heavily disturbed sites than in the lightly and moderately disturbed sites. Pairwise comparisons of means showed that, the species diversity was only significantly lower for heavily disturbed forest sites vis. avis moderate and light disturbance intensities, while variation was not detected between moderate and light disturbance levels (Table 2). The species diversity did not vary across the age classes (trees H' = 2.34, saplings, H' = 2.55, seedlings, H' = 2.34, F_2,92 = 2.273, p = 0.1.09) but variation existed in species diversity of age classes across the levels of forest disturbances, with the means being significantly lower for heavy disturbance.

Ndoinet forest generally depicted a low stocked forest with a mean stem density of trees being 439 ± 72 stems ha⁻¹. The tree species that existed in high densities in the forest were the early and mid-successional species such as N. macrocalyx (129 ± 37 stems ha⁻¹), T. stapfiana (82 ± 37 stems ha⁻¹) and M. kilimandsharica (39 ± 25 stems ha⁻¹). Podocarpus latifolius and Nuxia congesta recorded the least stems density of about 36 ± 18 and 4 ± 4 stems ha⁻¹ respectively. Pairwise comparisons of means revealed significant variation in stem density of trees across the disturbance regimes, with heavily disturbed sites having significantly low stem density. Stem density was 80% - 84% lower in the heavily disturbed sites than in the lightly and moderately disturbed sites ( Table 3). Neoubotonia macrocalyx had the highest density in heavily and moderately grazed areas with 60 and 267 stems ha⁻¹ respectively, while T. stapfiana had the leading density in lightly disturbed areas with 256 stems ha⁻¹ ( Table A2).

The forest generally demonstrated a reverse exponential curve with high number of regenerates and few mature trees. However, regeneration was not occurring at all in heavily disturbed areas as shown by complete absence of seedlings and saplings (Table 3, Figure 2). The overall seedling and sapling density in the forest was about 43,571 ± 14,191 and 1129 ± 246 stems ha⁻¹ respectively. This represents about 3% of the seedling being recruited to sapling stage successfully. Syzygium cordatum seedlings showed highest density in the forest with 22,143 ± 12,508 stems ha⁻¹ while palatable Indigofera sp had the least density of 357 ± 357 stems ha⁻¹. Syzygium cordatum also led in seedling density in lightly and moderately disturbed forest sites (58,889 ± 37,098 and 10,000 ± 3333 stems ha⁻¹ respectively). The seedling density varied significantly across the disturbance classes, with seedling density being 66% lower in moderately disturbed sites than lightly disturbed regime. Pairwise comparisons of means showed that variation did not exist in seedling density between moderate and light disturbance but was significantly lower for the heavily disturbed sites (Table 3).

Neoubotonia macrocalyx and S. cordatum were the most common sapling species in the forest with densities of 342 ± 114 and 129 ± 77 stems ha⁻¹ respectively.

Values that share a superscript letter indicate no significant variation while those with different superscript letter denotes significant mean difference as per Bonferroni post hoc test at 0.05 significance level.

Podocarpus latifolious and D. torrida had tailing sapling density of 14 ± 14 and 29 ± 29 stems ha⁻¹ respectively. In lightly disturbed forest, S. cordatum had the highest density of 356 ± 226 stems ha⁻¹ while, N. macrocalyx occurred in high densities in moderately disturbed forest with 756 ± 278 stems ha⁻¹. The sapling stems densities varied across the disturbance regimes, with significant variation detected between heavy vs. moderate and light, while difference were not detected between moderate and light disturbance levels. Sapling density was 18% higher in moderately disturbed sites than in light disturbed sites. Generally, about 1% and 6% of the seedlings in light and moderately disturbed forest were being recruited to saplings stage respectively (Table 3, Figure 2).

The DBH class distribution of trees varied across the forest disturbance categories. The heavily disturbed areas had trees with DBH range of 10 and 19.9 cm while lacking trees with DBH above 20 cm. The moderately disturbed areas had large numbers of individual trees within the DBH size classes of 10 - 39.9 cm. Trees with diameter size classes greater than 60 cm were only represented in lightly disturbed areas which were missing in both heavily and moderately disturbed areas (Figure 3).

The mean DBH of trees in Ndoinet forest was 16.48 ± 2.56 cm. The mean DBH declined significantly with increase in disturbance gradient (Table 3). The mean DBH was 86% and 74% lower in heavy disturbance level than was in light and moderate disturbance respectively. The mean sapling DBH was 3.04 ± 0.48 cm. Although the saplings mean DBH declined with increase in disturbance, variation was not detectable between light and moderate levels of disturbance (Table 3), as the DBH was only 7% lower in moderate than in light disturbance level.

The mean basal area of trees in the forest was 27.73 ± 7.13 m²·ha⁻¹. The dominance was majorly contributed by T. stapfiana (7.23 ± 4.02 m²·ha⁻¹), P. latifolius (7.13 ± 3.59 m²·ha⁻¹) and S. cordatum (4.14 ± 2.34 m²·ha⁻¹). Across the forest disturbance categories, basal areas were significantly different. The difference existed between heavy versus moderate and light disturbances, however differences did not exist between moderate and light disturbance levels. Plant basal area was 98% and 90% lower in the heavily disturbed sites than in the lightly and moderately disturbed areas, respectively, and 80% lower in the moderate disturbance than was in light disturbance regime. The tree species that contributed to high dominance in heavily, moderately and lightly disturbed areas were O. africana (0.62 ± 0.62 m²·ha⁻¹), N. macrocalyx (4.59 ± 1.35 m²·ha⁻¹) and T. stapfiana (22.49 ± 4.02 m²·ha⁻¹) respectively. The overall sapling basal area was found to be 2.25 + 0.53 m²·ha⁻¹ with significant variation across the disturbance regimes (Table 3). The sapling basal area in moderately disturbed sites was 28% higher than in lightly disturbed sites. Tabernaemontana stapfiana and N. macrocalyx were the most ecologically important tree species in the forest with important value index of 72.09% and 66.61% respectively. The high importance value index of T. stapfiana was highly contributed by basal areas, whereas that of N. macrocalyx was contributed by frequency. Neoubotonia macrocalyx had the highest IVI value in the heavily disturbed forest (129.33%). In light and moderately disturbed areas, the tree species that were most important were T. stapfiana and N. macrocalyx with IVI of 83.52% and 103.35% respectively. Across the forest disturbance intensities, the high IVI of the listed species were majorly contributed by high densities and basal areas.

Evidence from this study indicated that about 89% of the households adjoining the forest kept livestock mainly for economic and socio-cultural purposes and as safety nets. The livestock ownership were comparable with the figures (90%) reported by [40] in the same study area. The mean number of livestock owned by a household was slightly higher compared to findings by [41] within the study site; who reported a mean of 12 livestock (6 cattle, 5 sheep, 2 goats). Further, majority of the households (88%) grazed their livestock in the forest during the year, which was within the range (85%) found by [41] . The high dependence on the forest for grazing could be attributed to abundance of forage and lower demand for pasture management compared to high cost of production of on-farm fodder; and the declining farm sizes due to the growing population hence less land is allocated for fodder as described during FGD. Similarly, several authors have also reported high dependence of forest adjacent communities on natural ecosystems for grazing in most African countries, for example moist montane forests of Tanzania [42] [43] and Kenya [44] [45] [46] [47] .

Furthermore, it was recorded that the local communities grazed their livestock in the forest throughout the year, with influx of livestock grazing in the forest during dry season. The increase in livestock numbers during drought is also reported by [48] [49] . This increase could be due to the shortage of forage on-farm during this period which triggers the switching of grazing into the protected forest. However, during the dry season, woody plants species (especially the regenerates) are key forage resource, thus vulnerable to damage and loss and may ultimately halt natural recovery and resilience of the forest in the long term. The persistent damage and loss of vegetation through trampling, defoliation and browsing may have significant effect on vegetation attributes.

Although the Kenya’s Forest Conservation and Management Act 2016, only allows grazing of cattle and sheep in the forest, it was observed in this study that a number of goats (mainly browsers) and donkeys (non-selective grazers as well as browsers) were grazed in the forest. The feeding habits of goats and donkeys are known to cause detrimental effect on plant composition and structure in the long term [50] which pose a great challenge to the forest integrity.

Analysis of the livestock related forest disturbance indicators showed that heavy disturbed sites showed high relative disturbance (50%), followed by moderate (30%) and least in light (20%) disturbed forest. This indicates the high dependence of livestock in heavily disturbed sites, which may commensurate to high grazing intensity levels, because the open canopies encouraged proliferation of pasture while the closing canopies in less and moderately disturbed forests discouraged growth of forage due to shading effect and thus moderate and light grazing intensities.

Findings from this study indicated that unpalatable species were gaining dominance over the palatable species. For instance, over fifty percent (50%) of the sampled tree species were unpalatable to livestock. Grazing being the leading disturbance in the South West Mau forest presently as opined by [33] [41] , may have caused selective feeding of the palatable species leading to their elimination at early stages of development and leaving the unpalatable species to dominate. Accordingly, the high dependence of livestock in the heavy and moderate disturbed sites may explain the absence of very palatable species at young age such as Dombeya torrida, Indigofera sp and Podocarpus latifolious due to persistent selective grazing which may have hampered their regeneration and recruitment, but were abundant in light disturbance areas which are exposed to minimal grazing. Moreover, the unpalatable species had the leading IVI. This confirms the findings by [51] that species with high IVI are mostly unpalatable. The dominance of unpalatable species suggests that these species will continue to gain dominance while the palatable will continue to be suppressed and ultimately lead to altered species composition and existence of homogeneous stands in future if the condition persists. This is in agreement with the finding by [42] [52] , that persistent and heavy grazing contribute to the disappearance of palatable species and the subsequent dominance by other species mostly, less palatable plants-that thrive under extremely modified habitat conditions. The low abundance of the palatable species can therefore be largely attributed to the browsing effect by livestock and are at danger of exclusion from the ecosystem as a result of uncontrolled livestock grazing in the forest.

Forest disturbance is vital in influencing plant diversity, forest structure and regeneration [53] [54] . The effect of disturbance on stand structure and composition is highly influenced by disturbance intensity [55] . Our study found that the species richness declined with the increase in disturbance levels. Several authors have reported few species counts with increase in disturbance gradient [30] [53] [55] [56] [57] [58] . The decrease in species richness in this study at high levels of disturbance could be attributed to intensive anthropogenic disturbances such as over-exploitation and grazing which may have affected regeneration and recruitment as also suggested by [55] [59] . The heavily disturbed areas in this study experienced high intensities and prolonged disturbance before cessation of human settlement and are presently exposed to intensive grazing as it exists as grassland. Although several factors such as deteriorated site quality conditions and seed dispersal and colonization limitation may affect post disturbance natural recovery, persistent grazing and trampling by livestock is acknowledged as the critical factors that delays or halts post disturbance forest recovery after cessation of major disturbance agents or events [60] which is in agreement with results from this study. This results are also conforms to findings by [61] [62] [63] and [64] who found that high grazing pressure often leads to the disappearance of some species, especially those which are sensitive to grazing at the recruitment stage. The low species richness recorded in this study may therefore be linked to recruitment limitation due to prolonged past human induced disturbances and persistence of grazing following cessation of human settlement. The high intensity of disturbance may have led to loss of seedbank coupled with the lack of peripheral vegetation to initiate recovery. Further, the site being heavy utilized by livestock may have led to soil compaction coupled with persistent browsing which hinders natural regeneration and recruitment. Contrary to other studies, [65] found low species richness on undisturbed forests which suggests that human induced disturbance may favor colonization and establishment of new species.

The ecological stability of a community is dependent on the species diversity. Higher diversity values indicate a more stable and resilient community [64] . According to [66] , the Shannon-Weiner diversity index normally varies between 1.5 and 3.5 and rarely exceeds 4.5. The overall diversity values (H' = 2.7) recorded in this study indicate a moderate species diversity [67] , however, diversity was very low [67] in heavy disturbance. The results are in agreement with the findings by other authors [34] [53] [55] [58] [68] , who also reported decline in species diversity with increase in disturbance intensity. The extremely low species diversity in heavily disturbed sites may have been contributed by high and repeated disturbance via grazing coupled with intensive deforestation hence arrested succession. Correspondingly to this study, other studies have also reported decreased plant diversity with increased grazing intensities [42] [69] [70] and highest plant diversity at transitional levels of grazing [71] [72] . The relatively high plant species diversity in moderately and lightly grazed areas may be due to moderate effect of grazing disturbance which created small canopy gaps that encouraged the growth and establishment of gap opportunistic plants as reported by [73] , and which further supports the concept of intermediate disturbance hypotheses as espoused by [74] whereby, in intensively disturbed areas, the remnant species tend to re-colonize the site after similar disturbance resulting in prolonged low diversity. Further, the selective browsing of palatable species may have reduced their ability to achieve seed production as found in other studies [75] hence low species diversity and ecosystem stability. Our findings imply that the intensive and persistent disturbance has the potential to create homogeneous vegetation in heavy disturbed forests.

The Jaccard similarity index is used to express and compare the ecological similarity between two sites under different management regimes based on species lists [76] [77] . A species similarity value of 1 indicates completely similar species and 0 if they do not share a common species. Our study recorded very low similarity in species across the disturbance regimes. Furthermore, fewer tree species shared same plant families an indication of distant relation between species in the forest. This dissimilarity and distant relationship may be due to different reorganization/successional trajectory depending on the initial level and duration of disturbance, proximity to remnant forest for increased seed rain and tree recruitment, presence of seed banks and the recurring disturbance as the forest recovers naturally.

Frequency is used as an indicator of heterogeneity and species distribution in a stand. Frequency showed that most species in the heavily and moderately disturbed forest were early and mid-successional species, suggesting secondary succession. The lightly disturbed areas were dominated by mid and late succession species. High disturbances may have caused loss of late successional species and abundant establishment of pioneer species in the abandoned cultivated fields, while little disturbance led to the exclusion of early successional species that colonize sites immediately after disturbance as noted by [78] . However, the repeated disturbance in heavy disturbed sites had led to exclusion of colonizing species at regeneration stages. The presence of late succession species in light disturbed site may be an indication of full recovery of those sites as it was exposed to short-lived and less intense disturbance.

The high degree of disturbance was found to not only affecting species diversity, richness and frequency but also it promoted colonization and growth of woody invasive species (Acacia melanoxylon). The dominance of this species in the highly disturbed site and absence in other disturbance levels support the findings by [79] who reported that intensive disturbance favours the plant invasion because it provides a pulse of resources and pave the way for their seedling establishment and growth. In addition, most woody invasive species are shade intolerant and rapidly colonizes open forest gaps following disturbance. Their unpalatability to the livestock may encourage their establishment and recruitment. According to [80] , when grazing grounds are intensively grazed as is the case in the heavily disturbed forest, it may lead to invaders or undesirable plants to increase. The creation of more gaps through past disturbance, coupled with the recurrent overgrazing and deposition of dung (nutrients) may have provided suitable conditions for colonization of the invasive species. The abundance of this species in the forest is an important bio-indicator for intensive disturbance and native species succession limitation.

The forest structure also varied across the disturbance regimes. The tree stem density declined with increasing disturbance gradient as also found by [9] [58] . The density was significantly lower for the heavily disturbed sites while variation was not detected between light and moderately disturbed sites. The stem density of heavily disturbed forest areas (110 stems ha⁻¹) was below the average range (435 - 934 stems ha⁻¹) for indigenous tropical forests [9] [81] [82] . The low density may due to prolonged disturbance intensity and persistent disturbance after cessation of human settlement which hindered germination, recruitment and growth.

Although the forest generally demonstrated a reverse exponential curve with high density of regenerates and few mature trees depicting a healthy forest undergoing regeneration, heavily disturbed forest showed even aged j-shaped curve, with high density of mature trees than the regenerates, hence a poor population structure, while seedling and sapling regeneration and recruitment was maintained in the moderate and light disturbance gradients thus the uneven aged J curve. A study by [53] found inverse relationship between disturbance and regeneration for both seedling and sapling. Tree regeneration is determined by available resources (species specific environmental attributes) and species regeneration attributes (pioneer or shade tolerant) that influences seed production and survival of trees. The survival of regenerates is largely influenced by the presence of suitable microsite conditions [83] [84] . The existence of regeneration in moderately and lightly disturbed areas indicates presence of suitable site conditions that promoted natural regeneration and recruitment. The inhibited natural regeneration and recruitment in heavily disturbed sites areas could be attributed to a number of factors: effects of livestock grazing via persistent trampling, soil compaction which inhibits seed germination and browsing on recruits as it halts survival, growth, recruitment and density; lost seed banks due to repeated disturbance which results in disrupted succession patterns and processes. In another study [30] also related the low regeneration in heavily disturbed forest to lack of regeneration from soil seedbank or re-sprouts from stumps, ongoing grazing pressure and distance from the seed source. Accordingly, as perceived that the main limiting factor is intensive and persistent grazing [85] [86] [87] also found significant decline in regenerates’ density with increasing grazing intensity. The inhibited seedling establishment and subsequent growth into the saplings may encourage even aged stands and hinder the formation of canopy in future.

A similar trend as tree stem density was observed for the basal area with declining tree basal area with the increasing level of disturbances as also reported by [55] [58] . The basal area of trees at heavy (1.43 m²·ha⁻¹) and moderate (14 m²·ha⁻¹) disturbance level were lower than expected range (26 - 74 m²·ha⁻¹) for tropical natural forest ( [81] [82] , however, the differences were not significant between moderate and light disturbance regimes. The declining basal areas with the increase in disturbance are largely due to the low tree density and dominance by small diameter trees at moderate and heavy disturbance as indicated by the variation on the mean DBH values and DBH class distribution across the forest disturbance categories. The DBH diameter distribution for heavy, moderate and light disturbance ranged from 10 and 19.9 cm, 10 - 39.9 cm and 10 - over 60 cm respectively. This suggests that heavy anthropogenic disturbances can alter long term diameter structure of forest stands. Interestingly, the sapling basal area was significantly higher in moderate than the light disturbed forest; this may have been contributed by the high sapling density at moderate levels of disturbance rather than high mean DBH of saplings at moderate levels of disturbance. The shading effect from the closing canopies in the light disturbance may have hindered the recruitment of most seedling to saplings, while the canopy gaps in the moderate disturbance encouraged the recruitment of seedlings to saplings, while the heavy disturbance halted seed germination, survival and growth hence the lack of seedlings and sapling on such sites.

Corroborated evidence from this study suggests that the intensity and duration of disturbance, lack of seed sources/dispersal and persistent post disturbance perturbations can strongly modify forest recovery trajectories, including species composition and structure. Those unfavorable conditions, can push the system towards an alternate steady state where active interventions are required to restore a forest ecosystem. The continuity of heavy disturbance trend will likely hamper the development, integrity and sustainability of the forest and may ultimately lead to irreversible forest condition.