Tropical forests constitute the most important habitats for biodiversity because despite covering less than 7% of the global land surface, they host at least half of all terrestrial species on earth ¹ . However, these habitats also face the greatest threat from human exploitation, destruction or modification with an estimated loss rate of c. 10% every decade especially in areas without formal protection ^{1, 2} . Birds are among the most affected by forest destruction and habitat loss, particularly forest-dependent species ^{3, 4} which may respond to such perturbations in such spatially distinct patterns as to make them suitable for monitoring the quality of the forest habitat and its suitability for other taxa ⁵ .

Sokoke Pipit ⁶ is a forest-floor insectivore of the East African coastal forests of Kenya and Tanzania ^{7– 9} . This globally-endangered species ¹⁰ is generally restricted to the woodland habitat dominated by Brachystegia tree species (Leguminoceae) ^{9, 11} where it feeds on arthropods on the ground or in the understorey ^{9, 12– 14} . It mainly occurs in parts of the woodland with deep floor litter and a fairly closed tree canopy ⁹ .

The species has been more frequently encountered in the coastal forests of Kenya than those of Tanzania, with the most common sites being in the Arabuko-Sokoke forest, hereafter ASF and the Dakatcha Woodland. The main threat to the species is the modification and reduction of its suitable habitat ^{12, 14} , especially the removal of Brachystegia trees, a process to which it is very sensitive ⁹ .

This study aimed to assess the Sokoke Pipit’s response to change and modification of its habitat by comparing its estimated densities across three zones of its Brachystegia woodland stronghold in ASF. Although there have been extensive previous studies on the forest’s biodiversity in general ^{15, 16} ; on other forest-dependent bird species ^{17, 19, 20} and one on forest disturbance ¹¹ no study has been conducted to directly investigate the link between Sokoke Pipit population or distribution and modification of its habitat. Musila et al. examined ¹⁴ the species’ general habitat requirements within Brachystegia woodland, but did not specifically examine its demographic and spatial response to change in structural habitat quality. Our study thus aimed at filling these gaps as well as providing updates on the species’ current population estimate over the past decade since the study by Musila et al. ¹⁴ in 2001. In that earlier study species density was estimated at 2.8 ha ^-1 and 0.7 ha ^-1, in the undisturbed and disturbed areas of the Brachystegia forest respectively, and overall population at 13,000 for the whole forest.

The survey was carried out over 28 days between November 2011 and February 2012 within three blocks in the Brachystegia woodland stratified as follows: the main forest reserve block in the north-east, centered around Narasha (generally regarded as the most highly disturbed area from earlier intensive lumbering which continued until the early 1980s); the southern block of regenerating forest (a reserve regarded as less disturbed) in the Kararacha area; and the smaller strip on the outer north-western part of the forest around the Jilore village, which is considered more disturbed than Kararacha but slightly less so than Narasha ( Figure 1). This classification is based on the methodology of Oyugi et al. ¹¹ . Two 1-km transects were laid randomly in each block. Randomization was achieved by selecting the third track that branched to the left of the main forest track each time ^{31, 32} . When such a track was too short to cover one whole kilometer of forest as was in the case in the Jilore zone, a track was selected to run parallel to the main forest track but maintaining at least 250 m from the main track and the forest edge. In addition, bird surveys were conducted by starting from a different end of the transect each successive time ³¹ . Sampling independence for bird detection was ensured by maintaining at least 1 km from neighboring transects. Bird surveys, vegetation sampling and habitat assessment for tree logging intensity were assessed on separate days.

Because of the relatively small number of replicates in the study (two transect runs for birds and one set of habitat variable samples) preliminary data exploration showed departure from normal distribution. As such, all count data such as for live stems and cut tree stumps were transformed by logarithm and ratio or scale data such as by arc-sine before analyses proceeded ^{31, 35} . Sokoke Pipit densities were determined per hectare using DISTANCE v 6 software ³³ while their encounter rates were also calculated from the relationship R _E = n/L _t where R _E = encounter rate; n = mean abundance of Sokoke Pipits along the transect; and L _t = total length of transect in kilometers. Due to high variance in detection of Sokoke Pipit in the Jilore block compared to Kararacha and Narasha ( Table 1), which was likely a result of differences in understorey structural characteristics, Multiple Covariate Distance Sampling (MCDS) was preferable to Conventional Distance Sampling in estimating Sokoke Pipit density even with the relatively small sample size as the mean cluster sizes were quite constant at two individuals per sighting ^{36, 37} . We selected the cosine adjusted half-normal detection functional model with the lowest value based on Akaike Information Criterion in the density estimations ³⁷ . Species richness for all birds was evaluated as the total cumulative number of different species recorded in each transect during all the bird sampling sessions. Bird diversity was worked out using the reciprocal of Simpson’s index of the form: 1/S = 1/[(Σn(n-1)/N(N-1)] where S = Simpson’s Index, n = the total number of organisms of a particular species and N = the total number of organisms of all species. Simpson’s index of diversity was chosen as it is suitably robust for non-numerous replicate sampling such as was the case in the study ^{32, 35} .

The mean number of live stems and tree stumps/cut stems were derived from all stems counted in the three size classes in all quadrats in transects and transformed into densities per hectare. Percent canopy cover scores were coded such that open canopy, moderately open canopy and closed canopy scored 1, 2 and 3, respectively. These were then transformed to ratios scaled with ‘3’ as the maximum. Canopy height, floor litter cover and litter depth measurements were averaged from all quadrats in all transects.

Due to high preliminary-test covariance amongst the various size classes of live tree stems and tree stump counts, the size classes were pooled together into ‘total live stems’ and ‘total stems cut’ for subsequent analyses. For habitat variables that showed particularly strong correlations to bird variables, simple linear regression was performed to test the actual correlations and relative strengths of predictability. Differences of means of the key habitat (independent) variables were compared on the spatial scale by one-way Analysis of Variance (ANOVA) using the forest blocks as the categorical treatment effects on the bird (response) variables. The relationships between the independent and response variables (ANOVA and regressions) were analyzed in SPSS version 18.

In all surveys, a total of 308 birds were encountered, distributed across 55 species belonging to 25 families. This included three of the globally-endangered species: Sokoke Pipit, Clarke’s Weaver and Amani Sunbird; two globally near-threatened species: East Coast Akalat and Plain-backed Sunbird; and one regionally-vulnerable species: Little Yellow Flycatcher Erythrocercus holochlorus ( Supplementary Table 1). There were 17 encounters of Sokoke Pipit with an overall abundance of 30 individuals. The Sokoke Pipit occurred at a mean overall density of 0.72 birds/ha across the blocks surveyed, with a projected overall population estimated at 5,544 individuals ( Table 1). The density was higher in the moderate to high disturbance Brachystegia forest zone represented by Jilore and Kararacha blocks (0.89 birds ha ^-1) compared to the more disturbed Narasha block (0.71 birds ha ^-1). Nevertheless, there was no significant evidence of clumped distribution of the species across the blocks (χ ² _{(2, 0.05)} = 2.061).

Similarly, the species had the highest encounter rate in Jilore (4 km ^-1) while Kararacha had 2.0 birds km ^-1 and Narasha 1.5 birds km ^-1. The mean cluster size observed for Sokoke Pipit was 2 birds. For all birds, Kararacha had the highest species diversity (1/ S = 0.69) followed by Narasha (1/ S = 0.721) then the Jilore area (1/ S = 0.724), S being the reciprocal of Simpson’s diversity index. Jilore was the most bird species-rich (38 species) followed by Narasha (35 species) and then Kararacha (34 species).

Floor litter was deepest in the Kararacha block (2.52±0.83 cm) followed by the Jilore block (2.21±0.73 cm) and Narasha (1.75±0.58), F = 6.839, p = 0.002 (see Figure 2). Mean litter cover was generally within the middle category (33–66%) in the Kararacha and Jilore blocks and below the lower category (0–33%) in the Narasha block (F = 9.937, p = < 0.001).

Other significant spatial variations in means of habitat variables were observed in overall tree removal (total cut stems), removal of small poles (small-sized trees), and density of live mid-sized trees and removal of mid-sized trees ( Table 2). Thus overall tree removal rate was highest in the Kararacha block and lowest in Narasha both for small poles and large mature trees. The same pattern was observed for the density of mid-sized live woody vegetation.

Overall, the Brachystegia habitat was dominated by small-sized trees of 30 cm diameter at breast height (dbh) or less especially in the Jilore area ( Table 3). These were also the most intensely logged tree sizes with most of them cut in the Kararacha block ( Table 2).

Sokoke Pipit abundance was strongly correlated to forest floor litter depth (R ² = 0.719, β = 0.848, p = 0.033) and floor litter cover (R ² = 0.769, β = 0.877, p = 0.021) although litter depth was the better predictor of the species’ abundance ( Figure 3), with a predictive equation:

     Sokoke Pipit abundance = 0.727 + 0.485 * Mean litter depth

Furthermore, litter depth was positively correlated to logging intensity of small trees (R = 0.787, p = 0.063) suggesting that pruning of small trees in the forest by tree poachers might be a significant source of forest floor litter. Sokoke Pipit density appeared adversely affected by overall logging intensity (R ² = 0.663, β = -0.814, p = 0.048) see Figure 4. However, there was no significant effect of percent canopy cover (R = 0.5798, p = 0.228) or canopy height (R = 0.174, p = 0.742) on Sokoke Pipit density.

The density of the Sokoke Pipit from this study are lower than the values from studies in the same habitat about a decade ago in which the undisturbed Brachystegia forest had 2.8 birds ha ^-1 and disturbed zones had 0.9 birds ha ^{-1 14} . The same applies for the previous estimated total population of 13,000 birds. This is because of the continued degradation and modification of the species’ habitat in the Brachystegia spiciformis zone through disturbance, especially in the form of tree cover loss, which has continued over the past decade as observed by many investigators ^{9, 11, 15, 17– 20, 38} . Human activity and related encroachment effects are strongly presumed by all these investigators as the sole and direct source of the disturbance. The results of the present study confirm this but suggest that other complementary causes could be responsible for this habitat degradation processes.

To put this into perspective, it is important to understand the current general categorization of two main regions within the study area (the woodland dominated by Brachystegia spiciformis) as used by ecologists and forest managers. The first region is the middle section known as Narasha, which is generally characterized as “disturbed” while the second region is Kararacha to the south east, which is characterized as “undisturbed” ^{11, 18– 20} . The Jilore region, which is located to the far northwesterly end of the ASF across the larger Cyanometra forest stand, is rarely studied and is generally uncharacterized using such criteria even though it harbours Brachystegia tree stands. These regions are labelled solely on the basis of the comparative intensities of decades of intensive selective logging, which was spurred by high demand for valuable timber species, leading to loss of much of the primary indigenous stands of Brachystegia trees during the forest’s pre-protection era ^{11, 18} . The effect of this logging impacted the Brachystegia forest so heavily that it is yet to recover its original near-pristine forest status. It is this readily apparent scar of differential or selective “disturbance” between these regions that still guides most scientific habitat stratification in comparative study design. But this characterization is driven by such visible vegetation structural evidence as comparative densities, extent of spatial cover or openness of the understorey ^{11, 17, 20, 26} rather than the more functional and ecologically consequential processes and dimensions such as species or community dynamics, inter-trophic interactions, effect of anthropogenic encroachment, forest management systems or even climate change.

For instance, not only is earlier intensive deforestation still discernible in the structure of much of the forest, but in addition human populations around the forest have grown steadily and rapidly over the years ^{9, 18} . The increasing dependence of these adjacent communities on forest products have resulted in significant negative ecological impacts on the forest’s biodiversity ^{9, 14, 17, 20} . In addition, forest management intervention methods by the Kenya Forest Service (KFS) and Kenya Wildlife Service (KWS), have been in place for more than three decades since the era of official intensive logging began and this has further influenced overall ecological processes in the forest including the increasing population of elephants ^{18, 25} . Partly due to this forest surveillance, much of the illegal logging in the ASF is now predominantly focused on smaller trees or poles that are easier to cut and remove from the forest. Thus, effective characterization of the forest into spatial zones should therefore account for both structural and functional elements in the ASF.

In this study, the main driver of Sokoke Pipit habitat degradation was tree removal that results in opening up the understorey, a process that may expose individuals to the risk of predation through increased edge ^{39, 40} , reduction in patch substrate ^{14, 26} or change in micro-climate ³⁹ . One of the main reasons for lower logging rates in Narasha is that it is closest to the KWS and KFS stations and thus enjoys higher levels of surveillance against tree poaching compared to Kararacha and Jilore where logging rates were higher. These patterns conform to patterns observed by Ngala and Jackson from surveys carried out in 2009 and 2010 ^{25, 41} . Secondly, it is the region with the highest elephant activity ³⁸ which is a further deterrent to illegal loggers.

However, the effect of tree removal on Sokoke Pipit abundance was offset by the positive influence of forest floor litter cover and depth. Floor litter harbors much of the arthropod and other invertebrate biomass on which many insectivorous birds such as Sokoke Pipit depend ^{15, 42} . Secondly, the process of removing small trees appeared to be a significant additional source of floor litter, due to cumulative layers of discarded leaves and twigs left behind by tree poachers during pole harvesting, in addition to the slow rate of decomposition of organic matter typical of many forests along the eastern coast of Africa ^{43, 44} .

The Jilore block’s predominance in Sokoke Pipit encounter rates, in spite of its proximity to human settlements and farmland, may in part be due to its comparatively small size and low canopy with an understorey dominated by small regenerating trees ( Table 3). Harvesting of small trees for poles, which was highest in this block might also contribute to litter density that is conducive for harboring the Sokoke Pipit’s arthropod prey. On the other hand, the Kararacha block’s high Sokoke Pipit abundance ( Table 1) in spite of high logging rates indicates that human-driven selective tree removal is not the sole determinant of Sokoke Pipit population abundance or distribution across the Brachystegia habitat. For Narasha block, low litter depth coupled with lower percent canopy cover and low overall tree density due to poor regeneration all contributed to relative non-suitability for the Sokoke Pipit, despite apparent intensive surveillance against tree logging due to its proximity to the KWS and KFS stations. This is in contrast to Kararacha block’s comparative habitat suitability with respect to these variables is evidenced not only by higher a Sokoke Pipit density but also greater overall birds species richness despite a comparatively higher degree of logging pressure targeting small poles.

Evidence of the role of elephants in modifying the forest habitat is borne by our numerous direct chance observations across the study area, particularly in the Narasha block during which we made frequent sightings of trees felled or broken and the ground dug up by elephants. Analyses of data (see Data File) from these incidental observations was not attempted since counts were made only for the Narasha and Kararacha blocks. However, the spatial distribution of elephant damage noted here is consistent with similar earlier studies and observations conducted by ASFMT ¹⁸ , Ngala ⁴¹ and Banks et al. ³⁸ , all of which recorded the highest elephant activity intensity in Narasha. Such intensive activity results in more open canopy, exposed understorey and an increased area of edge habitat that may limit the dispersal capability of species that avoid crossing gaps, or may increase rates of general or nest predation ^{45, 46} .

Two main reasons support the contribution of elephants to habitat modification in the ASF’s Brachystegia woodland. First, the forest is estimated to hold between 126 and 184 individuals, giving a density of 0.44 animals km ^{-1 29, 30} . Not only does this make the ASF the 7 ^th highest elephant density site of all 30 elephant habitats across Kenya ²⁹ but is also fast approaching the 0.5 km ^-1 recommended maximum carrying capacity, to ensure stability and sustainability of the vegetation in the habitat ⁴⁷ . This density is a conservative estimate as it represents a projection for the whole forest; considering that the elephants seem to favour the Brachystegia forest zone ³⁸ , the carrying capacity will likely be exceeded much sooner than for the ASF overall, with negative consequences for the Sokoke pipit for which this is a critical habitat.

Secondly, the forest management has began erecting an electric fence, in 2006, which already covers a substantial portion of the forest boundary for the purpose of keeping the animals within the forest to reduce conflicts with forest-adjacent farmers who have previously incurred heavy crop losses to the elephants. This physical barrier to dispersal has had the effect of nearly doubling elephant density in the forest, further stretching the carrying capacity and worsening the habitat degradation process ⁴⁷ . The pressure is particularly high in the ASF due to its small size in comparison to other elephant sites in Kenya ³⁰ and given the peri-urban nature of the forest with its surrounding agricultural land and human settlements ¹⁸ .

In addition, the Brachystegia vegetation zone of the ASF has the lowest vegetation regeneration rates along the entire eastern coast of Africa due to soil with a functionally poor structure ²⁴ , low nutrient content, low moisture level and limited micro-organism activity that is necessary for nutrient cycling ^{11, 44} .

Thus, in terms of forest habitat health, in addition to human-driven tree removal the high elephant density and the restrictive nature of the electric fence are compounded by the slow forest regeneration rate, which may aggravate the role of elephants as drivers and accelerators of overall habitat change in the Brachystegia woodland. In many sections along the transects in the Narasha area, the frequency of elephant-felled trees outnumbered those cut down by humans. Without implementing measures to regulate the elephant population and movement in Brachystegia woodland concurrently with a halt in illegal logging in the forest, the rate of tree removal is likely to increase in the medium term with serious ramifications for the Sokoke Pipit and other forest-dependent species.

The Sokoke Pipit’s favoured habitat is an open understory with deep litter cover, often but not always with dense vegetation. Its density and estimated population in Brachystegia woodland is lower than it was a little more than a decade ago, suggesting increased pressure on the species’ habitat through increased loss or continued modification. The main cause of this habitat degradation is still illegal logging. Habitat degradation may be further accelerated by elephant–mediated habitat damage through tree felling, as evidenced by several incidental observations during the study, and consistent with findings of other recent studies though more in-depth studies on this are needed to underpin its scale and impact patterns on Sokoke pipit and forest specialist birds. Tree poachers mainly target small trees/poles which are taken mainly from parts of the forest farthest away from patrol bases and with minimal elephant numbers, from where they are easily carried out of the forest. A sound long-term conservation strategy would involve significantly reversing tree logging trends through increased surveillance; effectively managing local elephant populations and their movement; and stepping up habitat restoration through reforestation of heavily damaged Sokoke Pipit sites.

The deterring effect of proximity of Narasha to the KWS and KFS stations indicates the potential benefits of increased patrol and surveillance as an immediate check on habitat destruction by humans not only in the Brachystegia woodland zone but also throughout the forest.

The data referenced by this article are under copyright with the following copyright statement: Copyright: � 2014 Otieno NE et al.

Data associated with the article are available under the terms of the Creative Commons Zero "No rights reserved" data waiver (CC0 1.0 Public domain dedication).

figshare: Arabuko-Sokoke forest ecological data: Sokoke Pipit abundance, vegetation survey results, floor litter measures and elephant damage in three forest blocks, http://dx.doi.org/10.6084/m9.figshare.924690 ⁴⁸ .