Time Series Monitoring of Bush Encroachment by <i>Euclea divinorum</i> in Ol Pejeta Conservancy Laikipia, Kenya

Bush encroachment refers to the invasion of woody species in Savannah ecosystems driven by either anthropogenic and/or natural factors. This study sought to examine land cover changes and topographic features attributable to patterns of encroachment in Ol Pejeta Conservancy (OPC) where, Euclea divinorum unpalatable woody species has colonised former grasslands and other habitats which provide grazing grounds for herbivore wildlife species. Here, we monitored vegetation cover trends in the period 1987-2016 using five vegetation classes on Landsat images acquired during the dry season. Additionally, slope based NDVI maps and digital elevation models were used to identify topographic influences on vegetation change. Results revealed that E. divinorum increased significantly between 1987 and 2016 (Mann Kendall test for trend analysis tau 1, n=6, p< 0.01). On the other hand, Acacia drepanolobium and Acacia xanthophloea decreased from 49.72% and 5.31% in 1987 to 17.00% and 0.29% in 2016 respectively. Further, areas in low elevation were more colonised by E. divinorum. The colonising E. divinorum is unpalatable hence lacks natural predators whilst, A. drepanolobium and A. Xanthophloea which are alternative herbivory species decreased. Understanding dynamics of woody vegetation in savannah is crucial for management of healthy and sustainable ecosystems.


Introduction
Increasingly, savannah ecosystems are gradually changing from open grassland to woody dominated landscapes [1]. This biome shifts consequently result to variations in vegetation composition and structure notably increase in woody species [2]. Research has amassed evidence that savannah ecosystems are changing due to a phenomenon referred to as bush encroachment [3]. Literature has mooted potential drivers of bush encroachment in these ecosystems. As such, over grazing has shown positive correlation with increase in woody species in savannah ecosystems [4]. Other possible causes include precipitation rates particularly, increase in rainfall favouring establishment of woody species [4], fire suppression and favourable edaphic factors [5,3]. This phenomenon varies remarkably making it difficult to revert forested areas to grasslands [6]. Proliferation of woody species in grassland ecosystems threatens the resident biodiversity, biodiversity conservation measures [7] and pastoralism which is major source of livelihood in these ecosystems is also adversely affected [7].
In Ol Pejeta Conservancy (OPC) Laikipia Kenya, increase in unpalatable woody species of Euclea divinorum has become a major concern for conservationists. The encroacher species has been expanding into other habitat types such as Acacia drepanolobium, Acacia xanthophloea, Open grass land among others affecting key resource for mega fauna such as African Elephants (

Study Site Description
The study was undertaken at Ol Pejeta Conservancy (OPC) which covers 90,000 acres (360km 2 ), a classic example of an African savannah. It lies between Mt. Kenya and the Aberdare Mountains (0°7.288'N, 36°42.384'E and 0°8.634'N, 37°0.605'E) (0°1.831'S, 36°46.578'E and 0°5.7025'S 37°2.492'E), at an average altitude of 1810m, mean annual rainfall of 739mm, mean maximum and minimum temperatures of 28°C and 12°C respectively. It is a private owned conservancy chiefly, a sanctuary for black rhinos (Diceros bicornis). However, other wild animals are also conserved in the same property composed of various feeding guilds such as mixed feeders, carnivores, grazers and browsers. It is also a sanctuary for chimpanzee (Pan troglodytes) providing refuge site for seized chimps from black markets. It has one permanent river flowing through the conservancy as well as various man made water holes to supplement water scarcity in dry spells.
Major land cover types include grasslands, Acacia drepanolobium, A. xanthophloea, Euclea divinorum, and mixed bushlands. The conservancy is surrounded by an electric fence with three "corridors" to allow movement of wild animals in and out of OPC (but movement of rhino species is restricted due to the risks involved). The conservancy is surrounded by agropastoral communities and towards the north by other adjoining conservancies. The map of the study site is as in figure 1.

Image Acquisition
Multispectral Landsat images for 1987, 1995, 2000, 2005, 2010 and 2016 captured during dry seasons were acquired from United States Geological Survey (USGS). Auxiliary data were acquired using hand held Global Positioning System (GPS) for ground truthing purposes.

Image Pre-processing and Classification
As light passes through the atmosphere, it interacts with particulate matter such as haze, smoke, water vapour among others hence signal quality can be affected considerably before and after interacting with the target object/surface [8]. As such, in order to get clear Landsat images, top of atmosphere reflectance correction was performed using dark object subtraction algorithm in Quantum Geographic Information Systems (QGIS).

Multitemporal Landsat images Thematic Mapper, Enhanced
Thematic Mapper+ and Operational Land Imager of 1987,1995,2000,2005,2010 and 2016 were used to study Land cover dynamics. Supervised classification technique was used to classify vegetation cover in to five land cover types namely Euclea divinorum, Acacia drepanolobium, Acacia xanthophloea, Open grassland and Mixed bushlands in the conservancy. This algorithm is preferred over the other algorithms due to its ability to use well developed probability theory. Here, four (4) regions of interest (ROI) per each habitat type identified after different colour composite combination were used to enhance image interpretation. This was made possible due to existence of various habitats types in clumps hence identifiable especially with prior user's knowledge of the spatial extent of cover types. This made it easy to identify and separate species composition prior to actual classification. Assessment of spectral distance (spectral separability) to minimise classification errors was executed using Jefferies-Matusita Distance where if asymptotic distance is 2 the signatures are completely different whilst, if it is 0 signatures are identical [9]. Further, Normalized Difference Vegetation Index (NDVI) display an option in QGIS platform was activated during classification to enhance accuracy of the results. Thirty (30) random points were also generated and loaded in to hand held GPS and points visited thereafter to compare spectral similarity between actual cover and classification output. This further enhanced assessment of species composition on the ground.

Accuracy Assessment
In order to improve on reliability of change detection results, accuracy assessment of the classification output is crucial. Accuracy assessment reveals correctness of the classification results with reality on the ground. Here, 30 random points were generated covering different land cover types and compared with the ground cover to ascertain that there was no landcover classification spectral mix up. Finally, classification reports were generated giving proportions of each land cover type in a given year.

Land Cover Changes with Respect to Euclea divinorum
Classification results revealed that E. divinorum cover increased from 12

Overall Vegetation Cover Changes in OPC
Results from this study revealed that some class cover types either increased or decreased throughout the study period whilst other classes increased or decreased at different times ( figure 4). Particularly, E. divinorum increased in cover throughout the study period whereas A. drepanolobium and

Topographic Features Attributable to Encroachment Patterns by Euclea divinorum
Slope based Normalized Difference Vegetation Index (NDVI) was performed to examine any influence of slope on encroachment patterns. The results showed that areas with high NDVI value (NDVI maps for 1987, 2000, 2005 and 2016) were those along deep channels and valleys, ideally areas in low elevation were covered by dense E. divinorum as in figure 6 a, b, c and d. Here, four Landsat scenes were deemed appropriate for the purposes of exploring encroachment patterns using NDVI ratios and consequently generating NDVI maps. NDVI results revealed that in 1987, areas close to the river creating a thin band covered by evergreen E. divinorum whilst in the year 2000, the species spread outwards notably in the southern and southern eastern parts. In the year 2005, E. divinorum spread further especially in low elevated areas (deep valleys) covering larger part in southern and eastern as well in some parts in the northern region. In the year 2016, E. divinorum spread further towards the northern western parts in the conversancy predominantly in the low elevated areas.
Additionally, topology analyses using DEM and contours revealed the lowest point at about 1762.48 m ASL and highest point at about 1917.64m ASL translating to a range in altitude of 155.16m in the conservancy. Further, overlay of both topographic features, which include contours and elevation (observable features) on vegetation map of 2016 showed that areas below 1800m above sea level (ASL) were covered by E. divinorum as in figure 7 a, b, c and d. However, this encroachment appeared to be expanding towards areas even at higher altitudes than 1800m ASL.

Bush Encroachment by Euclea divinorum
Through the study period E. divinorum has increased whilst other habitat types have reduced in cover. The patterns exhibited by spread of E. divinorum are consistent with diffusion model [10] for invasive species whereby at the start there is low recruitment rates but over time the recruitment rate increases consequently and cover increases exponentially (encroachment from infested zones to transition zones and finally establishment in the un-infested zone).
Whilst there has been increase in cover by E. divinorum, on the other hand A. xanthophloea and A. drepanolobium have also reduced in coverage notably. As these dynamics in cover changes take place, other habitats have increased or decreased over the entire study period as for the case of mixed bushland majorly composed of other woody species such as Scutia myrtina, Rhamnus staddo, Euclea divinorum, Acacia drepanolobium, Rhus natalensis and Carissa edulis with no relative abundance. Remarkably, E. divinorum has been spreading in to vacant niches in habitat with low densities or areas devoid of the encroaching species [11]. The finding of this study is in tandem with other research done in OPC where they reported increase in spatial coverage by E. divinorum [11] though they reported that there were no significant changes. Given that the woody species under consideration is unpalatable to both wild and domestics animals, there is a potential of affecting their resource base indirectly by augmenting loss of resources through replacement.
Disturbances in savannah ecosystems have been mooted as possible driver for changes in savannah landscapes [12]. These disturbances can range from human induced land cover changes such as prescribed burning, climatic induced factors such as droughts and rainfall to herbivory and pastoralism [13,12]. As such in OPC, where there is an increasing population of elephants, giraffes and black rhinos (personal communication, EMU) their herbivory (with preference towards A. drepanolobium) impact is giving E. divinorum an advantage over A. drepanolobium with regards to their reestablishment.
Earlier research work conducted in the conservancy reported that there was high levels of damage (herbivory) to the A. drepanolobium whose net reduction can potentially explain the encroachment by E. divinorum [14]. Further, illgrazing management systems/regimes such as over stocking can lead to degradation of the ecosystems functions thus altering grass-woody interactions. To this end, such land management systems can potentially suppress grass biomass consequently facilitate encroachment by woody species in arid and savannah ecosystems.
However, herbivory effect especially by mixed feeders has potential to determine heterogeneity of savannah fauna composition. This holds true if the woody species in question are palatable. According to past research by Wahungu et al [11] mega herbivores such as elephants have the ability to open up bushy habitats and regulate woody species density thus minimize net effect of encroachment. In OPC there is low levels E. divinorum damage owing to the fact that the species is unpalatable [15].
As such, elephant's herbivory net effect on this plant species is insignificant hence may not contribute significantly to opening up or suppressing of habitats under E. divinorum. This observation is consistent with other work done in Seregeti that elephants had no significant effect on E. divinorum [16]. In summary, elephants can suppress and/or open up closed woody vegetation especially if composed of palatable species whilst on the other hand pure grazers such as cattle can suppress grass hence alter its competitiveness with other plant species. Such interactions coupled with rainfall and other disturbances have potential to augment encroachment by woody species as well as determine dominant plant cover [17].
Increasingly, changes in structure of savannah and semiarid ecosystems from grass to increased bushy or woody species has remained a subject of debate. As such, possible theories and supposition have been postulated to elucidate this phenomenon as observed over time in savannah ecosystems. Climate change, high levels of herbivory, changing fire regimes (fire severity, duration and frequency), changes in competitiveness of grass, seed dispersal by animals and combination of all these factors have been suggested as responsible for encroachment [18][19][20]. Pejeta Conservancy Laikipia, Kenya Chiefly, introduction of cattle (pure grazers especially in large numbers) in grassland ecosystems has been cited as the major driver of encroachment [18]. However, relatively low herbivory pressure can be tolerated by plants without conspicuous changes in plant productivity, biomass reproduction but higher pressure can affect these factors. In OPC, where there are mixed grazers, browsers and pure grazers, there is a potential that their herbivory effect has benefited spread of E. divinorum over time. However, the role of cattle in their indirect facilitation of encroachment in OPC has not been investigated and up to date remains unknown.

Topographic Features Attributable to Encroachment Patterns by Euclea divinorum
The study revealed that areas in low elevation had higher vegetation coverage as opposed to those in high areas. Micro topographic feature influences vegetation cover, distribution and even species present [21]. Relief and topographic variables such as slope, aspect and elevation can exert site specific microclimates hence affect landcover in some area [22]. This is in consistence with observed E. divinorum encroachment patterns and cover. This changes in elevation and slope indirectly affect net effect of solar radiation as such influencing soil temperature, near surface soil temperature and soil moisture [23]. Overall such differences on earth surface net effect is manifested in form on vegetation structure, distribution and growth [23,24,22]. These differences in topographic feature also influences water infiltration, run off, erosion, seed migration and other debris [25]. Elevation at regional scales constrains vegetation distribution [26] which is true for the case on bush encroachment by E. divinorum in OPC which is mainly in low elevations. However, it is important recognising there are other factors contributing significantly towards these vegetation distribution types such as soil factors though not examined in this study.

Conclusions
Based on the study findings we conclude that E. divinorum has increased in cover significantly whilst reducing coverage of A. drepanolobium and A. xanthophloea substantially. Additionally, elevation is attributable to encroachment patterns as revealed by the digital elevation models and NDVI results. As such, there is need to actively manage the encroaching woody species which comprises nearly half of the land cover in the conservancy. Further, some aspects have been identified hence recommend additional investigation on role of chemical and physical properties of soil in determining vegetation cover which will help mapping areas that can be potentially under threat of encroachment, as well as impact of bush encroachment on grass biomass and diversity.