Mapping and tree species diversity of the forest savanna mosaic in the Ashanti region, Ghana

The objectives of the study are; (1) To stratify the proposed plantation site into defined land cover types using satellite imagery; (2) To assess the level of canopy cover and (3) Ecological stratification of forest and savanna according to vegetation cover based on a terrestrially assessed parameter canopy cover was conducted as a supervised classification of satellite image. Two vegetation strata were identified and classified as forest and savanna. The forest vegetation featured an open canopy structure with an estimated average canopy cover of 21.6 ± 3.2%. The canopy cover ranges from a minimum of 10% to a maximum of 51.8%. A tree species assessment recorded a total of 65 tree species belonging to 49 genera and 23 families. A comparison of the tree composition of the forest and savanna vegetation strata revealed no significant differences. A literature review of the inventoried tree species revealed that, the majority of tree species are characteristic forest species (47.7%), while typical savanna woodland species and species found in both vegetation types recording 36.9 and 15.4% respectively.


INTRODUCTION
Globally the tropical forests are estimated to host more than four-fifth of the world's biodiversity (FAO, 2001) but despite the high populations and rapid deforestation in recent years (FAO, 2005), the tropical forest still retains about half of the world's species, even though they occupy only seven percent of the land area (Wilson, 1992). African forest resources suffer from continuous degradation caused by overutilization (e.g., for fuel wood collection, hunting, logging) and the regular occurrence of fires. The rate of disappearance of tropical forests is alarming with West Africa being the hardest hit by this trend (Schroeder et al., 2010). In a typical year, Africa accounts for more than half of the global forest area damaged by wildfires (FAO, 2007). Fire therefore plays an important role in shaping the vegetation. In recent decades, West Africa has developed wood-based industries (that is, sawmilling and panel-products discouragi manufacture) stimulated by policies and legislation discouraging the export of unprocessed logs. West Africa has been a major source of tropical hardwood however; the sub region's ability to be a major producer of tropical hardwood is on the decline. From being exporters of tropical logs in the 1980s and 1990s, many countries are finding that their forest resources have depleted to such an extent that they are unable to even meet the domestic demand for wood and wood products (FOSA, 2003). The conversion of degraded tropical forests into forest tree plantations is widely practised throughout the tropics and is taking place in Ghana on moderate extent. The estimated area of forest plantations, a subset of planted forests defined as those consisting primarily of introduced species is about 52 million hectares worldwide and accounts for app. 18 million hectares in the tropics (FAO, 2010). The purpose of tree plantations are generally grouped into economic, *Corresponding author. E-mail: tom_dery@yahoo.co.uk. Tel: +233203669027. social and environmental benefits (Buck, 1995;Ruitenbeek and Cartier, 1998) but most often the economic benefits supersede the rest. It is expected that tree plantations will play an increased role as a sustainable, energy efficient, environmentally and socially friendly source of world roundwood, fibre, fuelwood, nonwood forest products and other social and environmental values (Gelsenan, 2003).
Tree species composition studies are critical in the context of understanding the extent of plant diversity in various ecosystems (WCMC, 1992;Addo-Fordjour et al., 2009). They are important for the assessment of sustainability of ecotypes since various ecotypes play a vital role in the conservation of plant species, and ecosystem management (Tilman, 1988;Ssegawa and Nkuutu, 2006). The general objective of the study is to evaluate the various land cover types and tree species diversity in a heavily degraded tropical semi-dry savanna forest which is planned for conversion into a forest plantation. The specific objectives of the study are; (1) To stratify the remaining natural forest into defined land cover types using satellite imagery, (2) To assess the level of canopy cover and (3) To assess tree species richness of the degraded forest.

Study area
The study was carried out in a privately owned degraded forest, foreseen for conversion into a forest plantation. 100 km north of Kumasi, in the Ejura -Sekyedumase district of the Ashanti region of Ghana. The site is located in the forest-savanna transition zone which is characterized by the co-existence of the two distinct ecosystems forest and savanna. The ecosystems occur in an intertwined structure referred to as the "forest-savanna mosaic" (Hopkins, 1963). The district is marked by two rainfall patterns; the bi-modal pattern in the south and the uni-modal in the north. The main rainy season is between April and November. The annual rainfall varies between 1, 200 and 1, 500 mm. The proposed plantation site is approximately 17 km from the Ejura Township and is bordered to the north by the Nkrama and Bisiu, Boami is located to the south, while to the west it is bordered by the Awura forest reserve and to the east by the Afram river (Knoell, 2004). The soil types were classified according to WRB/USDA Soil Taxonomy as Aibic Plinthosols/Typic Plinthustalf. The soil-texture comprises predominantly of sandy loams. The soils are relatively shallow, encountering a hard ironstone pan at the petroferric contact at an average depth of 42 cm (Werner, 2008).

Pre-processing and processing of satellite images on the proposed plantation
A pan-sharpened IKONOS-2 satellite image clipped to the area of the proposed plantation site (recorded on October 10 th , 2000) was purchased from European Space Imaging in 2006 ( Figure 1). The IKONOS-2 satellite is a commercial satellite which is able to collect panchromatic images of 1 m resolution and multi spectral color images of 4 m resolution. Images from both s ens ors w er e m erg ed by t he proc ess of p an-sharp en ing t o creat e c olor im ag ery wit h 1 m res olution. Pan-sharpened satellite images are commonly used for presentation purposes, due to the high resolution, which in case of Ikonos corresponds to 1 m 2 per pixel. The satellite image was geo-referenced with Ground Control Points (GCPs) which were taken with Global Positioning System (GPS) in 2006 -and the satellite image was mapped with existing Geographic Information System (GIS) data, that is, boundaries of the proposed plantation site, already planted area, infrastructure, etc. (Figure 1).
The pan-sharpened IKONOS-2 satellite image was stratifiedaccording to the parameter vegetation cover -with a supervised classification procedure, which belongs to the spectrally oriented classification procedures for land cover mapping (Lillesand et al., 2004). In this case the three bands for red, green and blue light (RGBbands) were used for the pixel-based classification process. The assessed canopy cover of the sample plots was used as input to compile a numerical "interpretation key" for the description of spectral attributes. The stratification classes used were "forest" and "savanna ", assuming a normal distribution. The "Minimum-Distance-to-Means" algorithm was used for the comparison of the spectral information between known and unknown pixels (Wulder and Franklin, 2003;Lillesand et al., 2004). Due to the high resolution (1 m 2 per pixel) and the high cloud cover of the satellite image, image preprocessing procedures had to be carried out prior to the supervised classification. The low pass filter was used to modify the satellite image prior to the supervised image classification. Low pass filter belong to the category of spatial filters which emphasize low spatial frequency features and deemphasize high spatial frequency components of an image (Lillesand et al., 2004). In this case, a 25 × 25 low pass filter (based on the "moving window" algorithm) was used to modify the satellite image.
The error matrix in accordance to Lillesand et al., (2004) revealed two classification errors; 1) exclusion of forest vegetation and 2) other inclusion of forest vegetation. The overall accuracy of the supervised classification was 89.5%. Forest vegetation is classified in relation to FAO (2005) as land with trees higher than 5 meters and a canopy cover of more than 10%, or trees able to reach these thresholds in situ (it does not include land that is predominantly under agricultural land use).

Sampling design
Long term research is intended for this proposed plantation site which resulted in the use of reconnaissance forest inventory with the general objectives of indexing the location and extent of forested areas, as well as registering accessibility, species composition, tree dimensions, the distribution of various forest types, and initial data on timber quality (Kohl et al., 2006). A systematic sampling methodwas used because the systematic distribution of sample plots is more efficient for a given number of sample plots and detecting phenomena such as variations in species composition and stand structure (Synnott, 1979). A total of 55 circular sample plots ( Figure 2) were laid on the intersections of the 600 m UTM-grid. Each plot of 15 m radius (706.5 m 2 ) was set up on the field with the use of the ultra sonic device Vertex III. Smaller circular sub-sample plots of 5 m radius (78.5 m 2 ) were set up using the same centre point for the assessment of regenerating trees with a height < 1.30 m (seedlings and small saplings).

Canopy covers
The crown position of all trees ≥ 1.3 m was assessed according to Dawkins and Field (1977). The crown diameter was measured by projecting the exterior shape of the crown to the ground and measuring the diameter with the Vertex III from north to south and from east to west. The mean diameter was used for calculating the crown projection area assuming a circular crown shape. To assess the crown projection area of trees with a DBH < 5 cm, a mean value was calculated. A total number of 170 trees with a DBH < 5 cm were measured resulting in an average value of 1.05 ± 0.08 m 2 . The measured trees were distributed over 20 sample plots which were randomly chosen using the random number generator of Microsoft Excel 2007.
The canopy cover (in %) was calculated by dividing the accumulated crown projection area per sample plot with the area per sample plot. The crown projection area of trees of the crown positions 1 and 2 -that is trees which do not receive direct overhead light were excluded from the calculation of the canopy cover, since they were fully shaded by trees of the higher crown positions 3 to 5. The Kolmogorov-Smirnov test was used to test for differences in canopy cover of the vegetation types, forest and savanna.

Tree species inventory
Tree species were identified in a vegetative state, with the aid of a forest botanist from the Forestry Commission (Kumasi) following the identification key by Irvine (1961), Hawthorne (1990) and Hawthorne and Jongkind (2006). In cases of uncertainty specimen were taken to the herbarium of the Forestry Commission Department in Kumasi. The tree species were given vernacular names up to the point of identification by the herbarium. All trees with a height < 1.30 m -which occurred in the inner sub sample plot were identified and counted. Each specimen was assigned to one of the 3 heightclasses; seedlings < 30 cm, small saplings 30 to 50 cm and big saplings 50 to 130 cm.

Vegetation mapping
The mapping of the foreseen plantation site according to vegetation cover, conducted as a supervised classification resulted in a ratio of land covered by forest vegetation to

Canopy cover
The forest vegetation has an open canopy structure with an estimated average canopy cover (cc = 0.05) of 21.6 ± 3.2%. The canopy cover ranges from a minimum of 10% (sample plot 36) to a maximum of 51.8% (sample plot 17). Only the sample plots 2 and 17 feature a canopy cover > 40%, while thirty sample plots (93.8% show a cc > 10%) feature a canopy cover between 10 and 40%. An area of 629 ha have (32.9% of the planned plantation area) could not be classified due to the high amount of areas covered and shaded by clouds ( Figure  2). Fifty five sample plots were systematically distributed over the proposed plantation site, but due to the aforementioned, only 35 sample plots could be utilized for the mapping of the proposed plantation site. 25 sample plots contained forest vegetation (cc > 10%) and were used for the supervised classification of the forest vegetation stratum. 10 sample plots contained savanna vegetation (cc < 10%) and were used for the supervised classification of the savanna vegetation stratum. The Box-Whisker plot (Figure 4) for canopy cover of the vegetation strata illustrates the distribution as being univariate with stratum 1 resembling the forest vegetation and stratum 2 being savanna. The one sample Kolmogorov-Smirnov the test (Table 1) illustrates test distribution as normal and no significant change in canopy cover of the vegetation types, forest and savanna.

Tree species diversity
A total number of 65 tree species (Table 2) were encountered belonging to 49 genera and 23 families. Leguminosae is the biggest family represented by 63% followed by Moraceace and Rubiaceae both with 25%. The remaining 21 families are represented with an abundance of less than 5 tree species. Two important trends were identified during the assessment of the species composition. 32 tree species (48.5%) show an average abundance of only 1 individual per hectare or less, out of which 28 tree species (42.4%) occurred in only 2 sample plots (= 4.3% of sample plots) or less. This indicates that the majority of the remnant tree vegetation comprises of a relatively homogeneous species composition. The species composition may vary on a small scale, since almost half of the encountered tree species recorded a low abundance in combination with a limited range of distribution. About 30.3% of encountered species (20 tree species) recorded an abundance of at least 5 individuals per hectare. Six tree species (9.1%) recorded an abundance of 20 individuals per hectare and more. A total of 47.7% of all tree species (Table 2) recorded in the study belong to the biome "forest" while 36.9% are savanna species. However 15.4% of encountered species exist in both forest and savanna (Table 2). Bridelia ferruginea, Vitellaria paradoxa, Terminalia laxiflora. The most abundant tree species were 30 individuals per hectare and more. Pterocarpus erinaceus, Daniellia oliveri, Ficus sur recorded abundance of at least 20 individuals per hectare. The relative abundance of these six tree species accounted for 45.8% of the total number of surveyed trees.

DISCUSSION
The current extent of vegetation distribution is only inferred because the classification was carried out with a precision of 89.5% and a wide time lapse of more than ten years between taking the satellite images and the implementation of the forest reconnaissance inventory. In this time lapse, the rate of deforestation could not be estimated and it can be assumed that the forest area has rather decreases than increased. Therefore, it is presumed that the ratio of forest to savanna has shifted towards savanna due to ongoing transformation of vegetation to different anthropogenic activities, mainly wild fires, swidden agriculture and charcoal production. 32.8% of the satellite images (629 ha of the planned plantation site) were not classified due to cloud cover and areas shaded by clouds. Field observations revealed that the large coherent area of savanna vegetation extends almost to the southeastern boundary of the proposed plantation site, but the actual extent of the vegetation stratum cannot be presented. It can be concluded from the forest reconnaissance inventory,that the extent and distribution of the savanna vegetation stratum is underestimated, but by how far cannot be estimated at this point. A GPSassessment of the major forest savanna boundaries was not carried out whichcould have provided relevant information on the current extent and distribution of the vegetation strata. According to FAO (2006), forest vegetation with a canopy  A further stratification of the forest vegetation ( Figure 5) into open and closed forest was not attempted, due to the limited number of sample plots. Forest canopy structure is a key determinant of forest ecosystem processes, and its measurement is essential for ecosystem monitoring, modeling, and management (Sexton et al., 2009). The species recorded in this study are similar to those recorded recently by other studies in semi-deciduous forest of Ghana (Vordzogbe et al., 2005;Anning et al., 2008;Addo-Fordjour et al., 2009;Tom-Dery and Schroeder, 2011). Vordzogbe et al. (2005) which recorded a high number of 172 plants belonging to 67 families however this include herbs and grasses that were excluded from the current study. Addo-Fordjour et al. (2009) likewise recorded a higher number of 107 species belonging to 37 families which also included lianas, grasses and herbs. However, Tom-Dery and Schroeder (2011) recorded a lesser number of 46 tree species belonging to 14 families in two small semideciduous forest fragments in the same region.
The dominance of the families Leguminosae, Moraceae and Meliaceae in some semi-deciduous forests has been reported (Vordzogbe et al., 2005;Anning et al., 2008;Addo-Fordjour et al., 2009). FAO (2008) reviewed D. oliveri, V. paradoxa and Terminalia spp. as characteristic tree species of natural and derived Guinea woodland savannas. Interestingly the dominant family recorded in recent studies of the savanna was also Leguminosae (Asase and Oteng-Yeboah, 2007;Asase et al., 2009;Tom-Dery et al., 2012). This can be taken as further evidence that the planned planation site is located in the transition zone since its tree composition contains species of savanna and forest origins.
In view of the open nature of the classified forest canopy and the comparison of the tree species composition of the forest and savanna vegetation strata revealing no obvious (significant) differences, a literature search of inventoried plant species was carried out. This revealed a majority of the tree species being forest but a substantial number being either savanna species or species that occur in both forest and savanna. It is therefore concluded that the biomes planned plantation site can be classified as a transition zone. It is recommended that an inventory design of soil-vegetation catena be implemented, to facilitate the identification of indicator species that will enable the categorization of site requirements of indigenous timber species for sustainable tree growth.