Geomatics Indaba Proceedings 2015 – Stream 2 Geomatics in conservation: Mapping woody cover in Botsalano Game Reserve, North West Province by Chris Munyati and Mogomotsi Moeng, North-West University Abstract Woody vegetation is a vital component of savannah ecology, serving as a food source for browsers. It exists in inter-specific competition with the herbaceous component of the savannah vegetation. The herbaceous component in turn supports a diversity of grazers. Mapping the distribution of the woody vegetation in conserved savannah areas is potentially of use to the management authorities in assessing carrying capacities for both grazer and browser species, and in assessing the long-term carrying capacity dynamics. Updated maps of the distribution of woody vegetation are often unavailable to the conservation authorities, resulting in reliance on outdated maps instead. In this study the distribution of woody vegetation in the 5700 ha Botsalano game Reserve, North West Province, was assessed using the spatial technologies of remote sensing and GIS. A 10 m-resolution SPOT 5 HRG multispectral image (K/J 128/401) acquired in April 2012 (late rain season) was used. During field work, woody cover at purposefully sampled 30 m x 30 m plots in Botsalano Game Reserve was quantified using a canopy cover quantification field technique. Dominant woody species, topographic position, and GPS location were recorded at the sampling plots. Field-identified woody cover end member locations served as seed data for sub-pixel mapping of woody cover on the multispectral image. Using thematic layers of geology, soil, and vegetation type the distribution of the woody vegetation was assessed using GIS overlay analysis in order to determine the factors influencing it. In addition, a dBASE object oriented database of woody species and their location context was developed. An illustrative SQL query answering a hypothetical conservation need was constructed and implemented on the database in ArcMap 10.3. Based on illustrative hypothetical conservation quests, the remote sensing and GIS derived results were found to be potentially useful in expediting some of the North West Parks and Tourism authorities’ conservation planning activities. Keywords image classification, GIS databases, SQL, GIS analysis, mapping Introduction Woody vegetation is important to the ecology of savannas [1]. This is especially so in game reserves in the savannah areas of the semiarid North West Province of South Africa where tree cover is scanty. In general, savannah woody vegetation is of importance since it provides food to browsers, and fulfils a number of functions to avian fauna. It coexists in inter-specific competition with the grass component [1, 2] on which grazers depend. Quantification and mapping of woody vegetation in the game reserves is, therefore, potentially beneficial in support of the nature conservation effort through the provision of habitat status data upon which carrying capacity assessments can be based. Updated maps of the distribution of the woody vegetation are lacking and, often, conservation management decisions rely on old maps as a consequence. The spatial technologies of remote sensing and GIS can help in the mapping and quantification of the woody cover, which can result in frequently updated maps. This paper aims at demonstrating the feasibility of woody cover mapping, using the woody vegetation in Botsalano Game Reserve as a case study. The ecological utility of the woody cover mapping is then examined using hypothetical nature conservation tasks. The distribution of woody vegetation on African savannas is influenced by a number of factors, primarily fire and browser (e.g. elephant) herbivory [3] but also soil (geology-related) factors such as nutrients and soil moisture, as well as rainfall and inter-specific competition [1, 4]. Depending on the location context these various factors can interact to result in varying patterns of woody vegetation [5, 6]. In the Kruger National Park, South Africa, Scholtz et al. [6] evaluated a number of environmental variables (rainfall, temperature, aspect, slope, geology, fire frequency and elephant density) to establish how they affect woody species distribution in relation to structural height classes. The results suggested that the patterns and processes driving woody composition and structure were largely decoupled and that the distribution of different structural classes of a particular species may not be driven by the same environmental variables [6]. For habitat assessments in support of conservation remote sensing is particularly useful because of repetitive coverage of sensors aboard space orbiting platforms [7], which makes it possible to update habitat data. GIS on the other hand is useful because of the ability to store many spatial layers pertaining to habitat factors, update 84 Geomatics Indaba Proceedings 2015 – Stream 2 and manipulate the data layers and then jointly use them to answer specific conservation needs [e.g. 8]. An illustration of these capabilities of GIS is an analysis that quantified suitable habitat for an endangered species whose primary habitat requirement in Langtang National Park (Nepal) was forest cover as influenced by relief [8]. For the analysis by Yonzon et al. [8] spatial layers of forest cover, relief height, slope and aspect derived from a Triangular Irregular Network (TIN) surface were employed in GIS overlay analysis that also used a buffer analysis-derived layer of distance from human settlements. Forest or woody cover, as a spatial layer in such analyses, can readily be generated on remotely sensed imagery using appropriate mapping algorithms. The common approach to the mapping of the woody cover is to employ a high-accuracy image classification algorithm, for example pixel-based classifiers [e.g. 9, 10]. The disadvantage of pixel-based classifiers in mapping of savannah woody cover is that they allocate whole pixels to the woody cover class when in reality only parts of the pixel are occupied by woody individuals. The result would be an overestimation of the woody cover, particularly if low spatial resolution (i.e. large pixel size) imagery is employed. Sub-pixel classification can avoid this overestimation [11]. Sub-pixel classifiers calculate the fraction of cover (as a decimal) by land cover classes of interest per pixel [12]. As requirement locations that represent pure coverage by the cover classes of interest (i.e. “end members”) are identified and specified during the generation of the spectral signature for the sub-pixel classification [12]. Data and methods Study area Botsalano Game Reserve lies just north of Mafikeng (Mahikeng) in the North West Province of South Africa (Fig. 1a), about 18 km north east of the Ramatlabama border post. The closest human settlements to the game reserve are the villages Madutle and Khunoswane (Fig. 1a). The game reserve was established in 1984 as an amalgamation of four farms. The reserve covers some 5700 ha and is managed by the North West Parks and Tourism Board (NWPTB). The large browser and grazer herbivores that are conserved at Botsalano are buffalo, blesbok, duiker, eland, gemsbok, giraffe, impala, kudu, reedbuck, hartebeest, springbok, steenbok, waterbuck, rhino, wildebeest, and zebra. Fig. 1: Location and physical features of Botsalano Game Reserve The land at Botsalano Game Reserve slopes gently in a southwestly direction (Fig. 1b). The geology consists mainly of limestone, with only a small western segment consisting of quartz porphyry lithology (Fig. 1b). The soils resulting from this predominant limestone lithology are Haplic Lixisols (Fig. 1c) on which is vegetation 85 Geomatics Indaba Proceedings 2015 – Stream 2 that is classified by Mucina and Rutherford [13] as Klerksdorp Thornveld [Fig. 1d]. Haplic Lixisols are strongly weathered soils with low levels of plant nutrients. Klerksdorp Thornveld vegetation is characterised by open to dense short Acacia karroo bush clumps in dry grassland; other dominant trees including Acacia caffra, Celtis Africana, Rhus lancea, Ziziphus mucronata, and tall shrubs including Acacia hebeclada, Diospyros lycioides, Grewia flava, Gymnosporia buxifolia, Rhus pyroides, Asparagus laricinus, Felicia muricata, Anthospermum hispidulum, etc., [13]. Portions of the small western quartz porphyry segment have Ferallic Arenosols as the soil type [Fig. 1c], and are covered by Mucina and Rutherford’s [13] Mafikeng bushveld vegetation type [Fig. 1d]. Ferallic Arenosols are highly weathered and iron-rich sandy textured soils that lack any significant soil profile development. The Mafikeng bushveld is characterised by well-developed tree and shrub layers, dense stands of Terminalia sericea, Acacia luederitzii, and Acacia erioloba in certain areas, with shrubs that include Acacia karroo, Acacia mellifera, Acacia hebeclada, Dichrostachyas cinerea, Grewia flava, Grewia retinervis, Rhus tenuinervis, Ziziphus mucronata, etc., [13]. Spatial data The following spatial data layers were used in the analysis: A SPOT 5 (Systéme Pour l’Observation de la Terre 5) high resolution geometric (HRG) multispectral image (K/J 128/401) of 1 April 2012 with a spatial resolution of 10 m, obtained from the South African National Space Agency (SANSA). A geology layer, extracted from the Council for Geoscience geology map of South Africa (Fig. 1b). A contours layer at 20 m vertical interval (Fig. 1b), extracted from a 2006 National Geo-spatial Information (NGI) layer. A triangular irregular network (TIN) layer was then created from the contours layer using the 3D Analyst extension’s tools in ArcMap 10.3, so that slope data could be generated. For the surface feature type in the TIN layer a hard-line designation was specified, in order for the TIN edges to represent distinct breaks in slope based on the contours. A soils layer, extracted from the Agricultural Research Council (ARC) of South Africa soils map of South Africa (Fig. 1c). A vegetation types layer, extracted from Mucina and Rutherford’s [13] vegetation map of South Africa. Field data Field work was conducted in Botsalano Game Reserve for the purpose of deriving woody cover image classification training data, for use in mapping the woody cover on the SPOT image. The woody cover was quantified at purposefully sampled 30 m x 30 m (900 m2) plots, using a detailed canopy cover quantification field technique [11]. The sampling sites were selected on the basis of homogeneity of woody cover. A measuring tape was used to measure the canopy width (diameter, d) of woody individuals (i.e. shrubs, trees), and the woody canopy area computed using Eqn. 1: Canopy area r 2 (1) Where r = canopy radius (= d/2) The total area of all woody individuals in the 900 m2 plot gave an indicative percentage of canopy cover value. Thus, the field quantification procedures estimated canopy cover as opposed to woody biomass. The size of the field data plots was set to 30 m x 30 m in order to accommodate GPS location error, as well as the root mean square (RMS) error that arises during image geometric correction and can possibly result in shifts in pixel location. At each plot the centre coordinates were recorded using a Garmin eTrex 30 hand-held GPS with location accuracy of 3 m. The common woody species and topographic position at the sampling plots were also recorded. A total of 13 sampling plots were used. Image processing Using ground control points (GCP’s) the SPOT 5 HRG multispectral image was geometrically corrected to the UTM projection (zone 35S, WGS84 datum). The RMS error was less than 10 m (i.e. less than one pixel), which meant that the pixel locations were not spatially offset by a distance the equivalent of a pixel or more. On the UTM projected image the field sites at which woody cover was quantified were located. The field sites with over 90% canopy cover were specified as woody cover end members for use in sub-pixel classification of 86 Geomatics Indaba Proceedings 2015 – Stream 2 the woody cover, using ERDAS Imagine 2015 software. Manual generation of signatures was employed, by drawing polygons around woody cover end member pixels. The spectral signature of woody vegetation can be confused with that of herbaceous vegetation, but this error was minimised by using an April (late rain season) image when the herbaceous layer was entering senescence. With the woody cover spectral signature generated the actual woody cover (material of interest) classification was then performed, with a specification of 8 fractions of cover (0,20-0,29; 0,30-0,39; 0,40-0,49; …. 0,90-1,00). These fractions of cover translate to 20 to 29%, 30 to 39%, 40 to 49%, etc. Pixels that were not classified into any of these fractions belonged to the class < 20% woody cover. Woody species database A tabular (object oriented) database of the woody species data that were derived from the field sampling plots was created using Microsoft Access. The table fields were: woody species, UTM easting, UTM northing, topographic position, slope (°), aspect (°), and woody cover (%). Slope and aspect values were obtained from the TIN layer that was generated using the contours layer (Fig. 1b). The Microsoft Access database was then exported to dBASE format so that it could be queried in ArcMap 10.3 using the software’s Structured Query Language (SQL), to answer hypothetical conservation management questions. Although Microsoft Access can answer such queries using its SQL prompts, analysis using ArcMap was necessary because the software is widely used in industry. GIS analysis and database querying Overlay analysis The woody cover map which generated from the SPOT HRG multispectral image was then converted from raster to vector format to facilitate GIS overlay analysis. The overlay analysis sought to establish whether high woody cover was associated with certain geology, soils, or vegetation type. Therefore, the geology, soils and vegetation type spatial layers (Fig. 1) were used in conjunction with the woody cover layer during the overlay analysis. Hypothetical carrying capacity analysis Although there are protagonists and antagonists with regard to vegetation-based ecological carrying capacity on African savannah rangelands, the carrying capacity is closely dependent on the vegetation as food source [14]. This ecological carrying capacity can be calculated, for example based on the large animal unit (LAU) concept; and 12 LAU per hectare (ha) is an indicative carrying capacity value on savannah rangelands [15]. Open grassland without tree cover provides key habitat for grazers. Therefore, the area of Botsalano Game Reserve in the <20% woody cover category was computed and assessed for herbivore carrying capacity based on the LAU criterion. Hypothetical habitat search for a faunal species based on woody cover criteria Part of the conservation effort in savannah game reserves involves rehabilitating fauna and releasing them into the wild. A number of conservation organisations, both state entities like the NWPTB and nongovernmental organisations (NGOs), do undertake such endeavours in conservation. A hypothetical scenario is a species that requires high woody cover (e.g. at least 90%) and Acacia woody species. An answer would be required as to whether that species can be release into Botsalano Game Reserve and, if so, where? The appropriate SQL for the species’ requirements as criteria is as in Eqn. 2, and it was run on the dBASE database table of the woody cover using ArcMap 10.3. "WOODY_COVER" >= 90.0 AND "WOODY_SPECIES" LIKE 'Acacia%' (2) Results Woody cover mapping The woody cover map of Botsalano Game Reserve that resulted from sub-pixel classification of the SPOT image (Fig. 2a) is as in Fig. 2b. Table 1 summarises the area sizes per woody cover category for the woody cover in Fig. 2b. Most of the game reserve had low woody cover of less than 20% on the date of the image. The sub-pixel classification generally mapped the woody cover with high classification accuracy, with all the field 87 Geomatics Indaba Proceedings 2015 – Stream 2 sampling sites being correctly assigned to the woody canopy cover classes that were measured in the field. High woody cover was generally in valley locations (Fig. 2). From the GIS overlay analysis, high or low woody cover was not associated with particular geology, soil or vegetation type. For example, pockets of high woody cover (> 70% canopy cover) existed both in the eastern predominantly limestone (Klerksdorp Thornveld vegetation) sector as well as in the narrow quartz porphyry (Mafikeng Bushveld vegetation) sector. Proximity to the human settlements Madutle and Khunotswane to the north and east (Fig. 1a), respectively, did not seem to influence the woody cover patterns. Fig. 2: The SPOT 5 HRG multispectral image (RGB: 321) of Botsalano Game Reserve (a) that was used in mapping the woody cover. The resulting woody cover map is depicted in (b). Woody cover < 20% 20 – 29% 30 – 39% 40 – 49% 50 – 59% 60 – 69% 70 – 79% 80 – 89% 90 – 100% Area (ha) 5314,1 130,3 111,0 71,5 38,9 16,8 7,3 3,4 6,7 Table 1: Woody cover area estimates for the image classification in Fig. 2b. 88 Geomatics Indaba Proceedings 2015 – Stream 2 (a) (b) Fig. 3: ArcMap implementation of the SQL query in Eqn. 2 for the habitat requirements of a hypothetical species (a), and (b) part of the woody cover dBASE table with the result of the SQL implementation and showing suitable locations for the species. Hypothetical carrying capacity analysis The < 20% woody cover class in Fig. 2b encompasses open grassland that provides key habitat for grazers. The area of cover by this woody cover class was 5314,1 ha (Table 1). Using the guideline of 12 LAU/ha as carrying capacity, the 5314,1 ha low tree cover rangeland in Botsalano Game Reserve had a carrying capacity of 63 769 large animal units. This does not translate into 63 769 grazers only, because each herbivore has an LAU value. For example, one buffalo constitutes 1 LAU, whereas 13 duikers equate 1 LAU [15]. Hypothetical habitat search for a faunal species based on woody cover criteria The SQL query in Eqn. 2 was implemented in ArcMap as in Fig. 3a. Running the query resulted in an indication of locations that would be suitable for the species that requires high woody cover (≥ 90%) and Acacia woody species (Fig. 3b). The result of the query includes the UTM coordinates of the potential sites. The coordinates can then be input to a GPS and navigated to. This process appears to be a rapid way of yielding the desired result in the search for a suitable habitat for the hypothetical species in question. Discussion and conclusion The lack of association between woody cover and the geology, soil or vegetation type is likely to be due to the farm land use history of Botsalano Game Reserve. However, the abundance of woody vegetation in valley locations fits the pattern of savannah woody vegetation observed in other savannah locations in South Africa [10, 11]. The analysis in this paper underscores the importance of conservation authorities embracing the spatial technologies of remote sensing and GIS. Many conservation agencies have GIS units, including the North West Parks and Tourism Board. Embracing the capabilities of the spatial technologies in these agencies should involve more than the mere drawing and labelling of maps using vector datasets. Remotely sensed raster datasets like satellite imagery would enhance their conservation operations, particularly problem solving analyses like the ones illustrated in this paper. The major pre-requisite for accurate answers to result from such analyses is that the datasets used are accurate and up-to-date [16]. With up-to-date and accurate data, spatial technologies do yield rapid answers to conservation targeted queries that are designed and executed correctly, as illustrated by studies even in the developing world context. In Madagascar GIS was used in a rapid fauna survey and habitat modelling procedure that was designed to generate information for reserve selection and design, with lemur conservation as the focus [17]. The spatial data in the GIS framework included lemur habitat variables and environmental and land use 89 Geomatics Indaba Proceedings 2015 – Stream 2 data derived from maps and satellite images. For the Sango Bay area, Uganda, field surveys of plants and animals were combined with satellite remote sensing of broad vegetation types to map biodiversity and thereby help plan conservation, resulting in a biodiversity map that has been used to aid conservation planning [18]. There are technical constraints associated with the adoption of spatial technologies in conservation, including those pertaining to manpower, software and hardware costs particularly in the developing world. Once these are overcome, conservation then benefits from the use of spatial technologies. It can, therefore, be concluded that embracing spatial technologies by conservation authorities will enhance the effectiveness of the conservation effort. Acknowledgements Mogomotsi Moeng received funding from the National Research Foundation (NRF) of South Africa, towards postgraduate studies in the Department of Geography and Environmental Science at the Mafikeng Campus of NWU, which made this work possible. References [1] AJ Belsky: "Influences of trees on savannah productivity: Tests of shade, nutrients, and tree-grass competition", Ecology, Vol. 75 No. 4, pp. 922-932, June 1994. [2] RJ Scholes and SR Archer: "Tree-grass interactions in savannas", Annual Review of Ecology and Systematics, Vol. 28 No. 1, pp. 517-544, 1997. [3] M Sankaran, NP Hanan, RJ Scholes, J Ratnam, DJ Augustine, BS Cade, J Gignoux, SI Higgins, X Le Roux, F Ludwig, J Ardo, F Banyikwa, A Bronn, G Bucini, KK Caylor, MB Coughenour, A Diouf, W Ekaya, CJ Feral, EC February, PGH Frost, P Hiernaux, H Hrabar, KL Metzger, HHT Prins, S Ringrose, W Sea, J Tews, J Worden and N Zambatis: "Determinants of woody cover in African savannas", Nature, Vol. 438, pp. 846-849, December 2005. [4] M Sankaran, J Ratnam and N Hanan: "Woody cover in African savannas: the role of resources, fire and herbivory", Global Ecology and Biogeography, Vol. 17 No. 2, pp. 236-245, March 2008. [5] P Couteron and K. Kokou: "Woody vegetation spatial patterns in a semi-arid savannah of Burkina Faso, West Africa", Plant Ecology, Vol. 132 No. 2, pp. 211-227, June 1997. [6] R Scholtz, G.A. Kiker, IPJ Smit and FJ Venter: "Identifying drivers that influence the spatial distribution of woody vegetation in Kruger National Park, South Africa", Ecosphere, Vol. 5 No. 6, Article Number 71, June 2014. [7] H Nagendra, R Lucas, PJ Honrado, RHG. Jongman, C Tarantino, M Adamo and P Mairota: "Remote sensing for conservation monitoring: Assessing protected areas, habitat extent, habitat condition, species diversity, and threats", Ecological Indicators, Vol. 33, pp. 45-59, October 2013. [8] P Yonzon, R Jones and J Fox: "Geographic information systems for assessing habitat and estimating population of red pandas in Langtang National Park, Nepal", Ambio, Vol. 20 No. 7, pp. 285-288, November 1991. [9] AE Gaughan, RM Holdo and TM Anderson: "Using short-term MODIS time-series to quantify tree cover in a highly heterogeneous African savannah", International Journal of Remote Sensing, Vol. 34 No. 19, pp. 6865-6882, October 2013. [10] AT Hudak and CA Wessman: "Textural analysis of high resolution imagery to quantify bush encroachment in Madikwe Game Reserve, South Africa, 1995-1996", International Journal of Remote Sensing, Vol. 22 No. 14, pp. 2731-2740, November 2001. [11] NI Sinthumule and C Munyati: "Quantifying savanna woody cover in the field and on historical imagery: A methodological analysis", South African Journal of Geomatics, Vol. 3 No. 2, pp. 113-127, August 2014. [12] W Liu and EY Wu: "Comparison of non-linear mixture models: sub-pixel classification", Remote Sensing of Environment, Vol. 94 No. 2, pp. 145-154, January 2005. [13] L Mucina and MC Rutherford (eds.): Vegetation map of South Africa, Lesotho and Swaziland: An illustrated guide, strelitzia 19, South African National Biodiversity Institute (SANBI), Pretoria, South Africa, 2006. 90 Geomatics Indaba Proceedings 2015 – Stream 2 [14] M Smet and D Ward: "A comparison of the effects of different rangeland management systems on plant species composition, diversity and vegetation structure in a semi-arid savanna", African Journal of Range & Forage Science, Vol.22 No.1, pp. 59-71, March 2005. [15] C Munyati, P Shaker and MG Phasha: "Using remotely sensed imagery to monitor savannah rangeland deterioration through woody plant proliferation: a case study from communal and biodiversity conservation rangeland sites in Mokopane, South Africa", Environmental Monitoring and Assessment, Vol. 176 No. 1-4, pp. 293-311, May 2011. [16] M Brambilla, F Casale, V Bergero, GM Crovetto, R Falco, I Negri, P Siccardi and G Bogliani: "GIS-models work well, but are not enough: Habitat preferences of Lanius collurio at multiple levels and conservation implications", Biological Conservation, Vol. 142 No.10, pp. 2033-2042, October 2009. [17] AP Smith, N Horning and D Moore: "Regional biodiversity planning and lemur conservation with GIS in western Madagascar", Conservation Biology, Vol. 11 No. 2, pp. 498-512, April 1997. [18] RM Fuller, GB Groom, S Mugisha, P Ipulet, D Pomeroy, A Katende, R Bailey and R Ogutu-Ohwayo: "The integration of field survey and remote sensing for biodiversity assessment: a case study in the tropical forests and wetlands of Sango Bay, Uganda", Biological Conservation, Vol. 86 No. 3, pp. 379391, December 1998. Contact Chris Munyati, North-West University, Tel 018 389-2325, [email protected] 91
© Copyright 2026 Paperzz