Assessing habitat fragmentation of the KwaZulu-Natal Sandstone Sourveld , a threatened ecosystem

Human activities such as urbanisation and agriculture are the predominant drivers behind degradation and habitat loss (Ferrari, Lookingbill & Neel 2007; Hanski 1999; Schooley & Wiens 2003), which in turn threatens biological diversity worldwide (Fahrig 2001; Hanski 1999; Lindenmayer & Hobbs 2008; Schooley & Wiens 2003). Grasslands are amongst the most threatened habitats as a result of total habitat loss and degree of fragmentation (Tarboton 1997). Grasslands provide vital ecosystem services that benefit humans both directly (the provision of traditional medicinal plants and grazing for livestock) (Dzerefos & Witkowski 2001; Egoh et al. 2011, 2016) and indirectly (the sequestration of carbon for climate regulation) (Egoh et al. 2011, 2016). Additionally, South African grasslands host a high diversity of endemic flora and fauna species with 349 endemic Red List taxa of concern (South African National Biodiversity Institute [SANBI] 2015) of which 180 are threatened with extinction (SANBI 2015). South African grasslands have been identified as endangered and in need of urgent conservation (Cilliers, Muller & Drewes 2004). The KwaZulu-Natal Sandstone Sourveld (KZN SS) ecosystem is currently classified as endangered (Mucina & Rutherford 2006:512) and is distributed entirely within the sub-escarpment of KwaZulu-Natal (Figure 1).


Introduction
Human activities such as urbanisation and agriculture are the predominant drivers behind degradation and habitat loss (Ferrari, Lookingbill & Neel 2007;Hanski 1999;Schooley & Wiens 2003), which in turn threatens biological diversity worldwide (Fahrig 2001;Hanski 1999;Lindenmayer & Hobbs 2008;Schooley & Wiens 2003).Grasslands are amongst the most threatened habitats as a result of total habitat loss and degree of fragmentation (Tarboton 1997).Grasslands provide vital ecosystem services that benefit humans both directly (the provision of traditional medicinal plants and grazing for livestock) (Dzerefos & Witkowski 2001;Egoh et al. 2011Egoh et al. , 2016) ) and indirectly (the sequestration of carbon for climate regulation) (Egoh et al. 2011(Egoh et al. , 2016)).Additionally, South African grasslands host a high diversity of endemic flora and fauna species with 349 endemic Red List taxa of concern (South African National Biodiversity Institute [SANBI] 2015) of which 180 are threatened with extinction (SANBI 2015).South African grasslands have been identified as endangered and in need of urgent conservation (Cilliers, Muller & Drewes 2004).The KwaZulu-Natal Sandstone Sourveld (KZN SS) ecosystem is currently classified as endangered (Mucina & Rutherford 2006:512) and is distributed entirely within the sub-escarpment of KwaZulu-Natal (Figure 1).
The connectivity of a landscape is vital for the persistence of species faced with habitat loss and fragmentation (Urban & Keitt 2001).Connectivity is often categorised into two aspects, namely structural connectivity and functional connectivity (Ferrari et al. 2007).Structural connectivity (related to the spatial structure of the landscape) is often easier to measure than functional connectivity (associated with species interactions), as it can be calculated using landscape metrics (Hanski 1999;Minor & Urban 2008;Rayfield, Fortin & Fall 2011;Urban & Keitt 2001).The management of landscape connectivity is a fundamental concern in ecology and biodiversity conservation (Minor & Urban 2008).This has increased the demand for user-driven tools that integrate connectivity within landscape planning (Fahrig 2001;Lindenmayer & Hobbs 2008;Miller Background: The KwaZulu-Natal Sandstone Sourveld (KZN SS) is a grassland ecosystem categorised as endangered by the terms of the National Environmental Management: Biodiversity Act (Act 10 of 2004).Pressure from urbanisation has led to the remaining areas of the KZN SS being physically fragmented, causing low connectivity levels which have diminished the biological persistence of this ecosystem.
2007; Minor & Urban 2008;Visconti & Elkin 2009).The Conefor Sensinode software offers decision-makers a platform in which to quantify the importance of habitat patches for maintaining and improving functional landscape connectivity (Pascual-Hortal & Saura 2006;Pascual-Hortal & Saura 2007).This can assist decision-making in landscape planning and habitat conservation, as the measurement of landscape connectivity requires an approach that emphasises the persistence of both species and ecosystems (Fahrig 2005;Ferrari et al. 2007;Pascual-Hortal & Saura 2006;Rayfield et al. 2011;Visconti & Elkin 2009).
The KZN SS is an interesting and unique vegetation type with a naturally patchy occurrence (Mucina & Rutherford 2006).Presently, the KZN SS is classified as endangered with a small percentage remaining within an urban matrix.Habitat fragmentation has not formally been taken into account for the listing of this threatened ecosystem (Department of Environmental Affairs 2009).This study quantifies the level of habitat fragmentation of the KZN SS utilising measures of structural and functional connectivity, and assesses the influence of fine-scale and broad-scale land cover and vegetation data on habitat fragmentation measurements.

Study site
The KZN SS is an ecosystem currently classified as endangered on a national scale; however, provincially it is considered critically endangered (Mucina & Rutherford 2006).The KZN SS is a short, species-rich grassland with scattered low shrubs and geoxylic suffrutices.The grass and tree proportions vary and depend on many factors, especially fire.Proteaceae trees and shrubs such as Protea, Leucospermum and Faurea are common locally, and can be found on flat, sometimes rolling plateau tops and steep mountains (Mucina & Rutherford 2006).There are 12 endemic taxon species unique to the KZN SS, including Helichrysum woodii, Brachystelma modestum, Cynorkis compacta and Hesperantha gracilis (Mucina & Rutherford 2006).According to Mucina and Rutherford (2006), the altitude of the KZN SS ranges from 500 to 1100 m, but patches are found at lower altitudes (Scott-Shaw & Escott 2011).Although there is some controversy over the classification of this vegetation type, the KZN SS is considered part of the South African grasslands biome (Mucina & Rutherford 2006).The KZN SS experiences mostly summer rainfall, and receives between 700 and 1200 mm of annual rainfall (Mucina & Rutherford 2006).It occurs on the Ordovician Natal Group sandstones, upon which are skeletal, sandy soils that are nutrient-poor (Mucina & Rutherford 2006).
The fauna that inhabit the KZN SS is not very well documented (Sandy Willows-Munro [School of Life Sciences, University of KwaZulu-Natal], pers.comm., 20 October 2014).The KZN SS shares a number of endemic species with the Pondoland-Ugu Sandstone Coastal Sourveld (Mucina & Rutherford 2006).In essence, the KZN SS forms a transitional boundary between the Moist Coast Hinterland Ngongoni veld and the Moist Midlands Mist-belt grassland.
The KZN SS is found both within and outside of the eThekwini Metropolitan (hereafter referred to as EM) area (Figure 1).It originally occupied 179 671 ha (Boon et al. in press;Jewitt 2011), of which 66 611 ha is located within the EM (Boon et al. in press).The KZN SS is naturally fragmented and originally comprised three predominate areas prior to any land transformation.The extant patches of KZN SS cover roughly 2% of the EM, with a large percentage of this ecosystem type converted through agriculture and infrastructure development (Mucina & Rutherford 2006).This is principally because it is a prime agricultural area for timber and sugar cane plantations.These human activities within the EM have led to the destruction and degradation of the KZN SS, resulting in the physical fragmentation of the ecosystem.
Of the 9805 ha of KZN SS that remain within the EM, 993 ha are degraded (Boon et al. in press).Apart from land-use transformation, the extant patches of KZN SS have been degraded through the altering of fire regimes, harvesting of plants for traditional use, and nutrient enrichment from neighbouring land-uses.This has resulted in much of the grassland having been replaced by woody vegetation (Boon et al. in press).

Data and pre-processing
Assessing habitat fragmentation requires up-to-date information on the spatial extent of habitat loss.This is usually derived from land cover maps.The spatial resolution of land cover and vegetation maps influence the quantification of habitat transformation.

Quantifying habitat fragmentation
Many of the metrics developed for determining connectivity only address structural connectivity or the physical connections amongst patches of a particular habitat type (Ferrari et al. 2007;Lindenmayer & Fischer 2008;Miller 2007;Minor & Urban 2008).The use of graph theory provides a more flexible set of metrics (Ferrari et al. 2007;Lindenmayer & Fischer 2008;Miller 2007;Minor & Urban 2008).The flexibility of graph theory stems from the fact that edge lengths can be defined in any way, not just by Euclidian distance between patches (Lindenmayer & Fischer 2008;Pascual-Hortal & Saura 2006;Rayfield et al. 2011;Visconti & Elkin 2009).Graph-based metrics can therefore measure functional connectivity, which accounts for species-specific habitat preferences and movement behaviours.Graph theory and spatially explicit metapopulation models that combine the physical attributes of the landscape with limited species information can be used to provide a measure of potential connectivity (Ferrari et al. 2007;Lindenmayer & Fischer 2008;Miller 2007;Minor & Urban 2008).
Conefor Sensinode has been suggested to have the greatest benefit for conservation problems regarding landscape connectivity as it includes new connectivity metrics based on the habitat availability concept (Pascual-Hortal & Saura 2006, 2007).The habitat availability concept enables a patch to be considered as a space where connectivity occurs.This enables habitat patch area and the connections between different patches to be integrated

Analysis
Landscape connectivity is species-specific and different species have diverse levels of dispersal and movement.Dispersal data on key KZN SS species were not available so a range of different dispersal distances (50, 100, 250, 500, and 1000 m) were taken into consideration for the study.This enabled the representation of a wide range of dispersal distances for grassland flora and fauna.The 50 m dispersal distance represents the dispersal range of smaller mammals such as grassland mice (Kollmann & Schill 1996) and the autochorous dispersal of seeds in plants (Higgins, Nathan & Cain 2003), whilst the 1000 m distance is more representative of medium-sized mammals such as duiker (Lawes, Mealin & Piper 2000).The range of dispersal distances used can be applied to many species, but the results produced may not necessarily reflect the movement patterns of all the organisms in the landscape.The study primarily focused on the 250 m dispersal distance to indicate the impact of habitat fragmentation at an intermediate dispersal range, as preliminary analysis using different distances exhibited minimal changes from one distance to the next.The attributes of the four data sets used in the analysis -namely broadscale KZN SS (full extent); broad-scale KZN SS (EM only); fine-scale KZN SS in good condition (EM only); and finescale KZN SS in any condition (EM only) -are summarised in Table 1.The study did not consider species-specific data as our knowledge of the KZN SS is limited and fauna that inhabit the KZN SS are not well documented.

Determining the connectivity of different fragments
Connectivity can be quantified as either binary or probabilistic.Binary considers each habitat patch as either connected or unconnected and does not factor in the strength of the connection between patches.Probabilistic, however, considers the movement of fauna as a means to estimate the strength of the connection between habitat patches (Pascual-Hortal & Saura 2007).There are nine different types of binary indices available for use through the Conefor software.The integral index of connectivity (IIC) was chosen as the best binary index for calculating landscape connectivity measurements (Pascual-Hortal & Saura 2007).The advantage of the IIC lies in the formula, which allows for patch quality and connectivity to be integrated, allowing each habitat patch to be considered as an area where connectivity takes place (Pascual-Hortal & Saura 2007).The IIC is calculated by the formula indicated below, with the output values ranging from 0 to 1, with 1 highlighting optimal connectivity.
Where: n is the total number of nodes (the habitat area considered for the connectivity analysis) (Pascual-Hortal & Saura 2006), a i and a j are the attributes (e.g.habitat area) of nodes i and j, nl ij is the number of links (the connection between habitat nodes in the landscape) in the shortest path between patches i and j, and A L is the maximum landscape attribute (the total landscape area, consisting of both habitat and non-habitat areas) (Pascual-Hortal & Saura 2006, 2007).
The individual patches were arranged and prioritised according to their overall contribution to the connectivity of the landscape.Larger and less fragmented patches were viewed as better connected, with fewer components (a set of nodes which contain pathways between every pair of nodes) representing a more connected landscape (Pascual-Hortal & Saura 2007).

Quantifying overall landscape connectivity
In order to determine the extent of the overall landscape connectivity of the KZN SS, two connectivity indices were used, namely the number of components and the IIC.These specific connectivity indices were chosen as they do not demonstrate the same problems associated with many other connectivity indices.For instance, where there is an increase in connectivity with increased fragmentation, or no connectivity predicted for a landscape occupied by one large habitat patch, or a lack of response of the index to the loss of a large isolated patch (Lindenmayer & Fischer 2008;Pascual-Hortal & Saura 2007;Visconti & Elkin 2009).As connectivity across the landscape increases, the number of components will decrease.More connected landscapes will tend to consist of one large component in which all the habitat patches are connected.As the landscape becomes more connected, the percentage of the available habitat area that is in the largest component will also increase (Fahrig 2001;Lindenmayer & Hobbs 2008;Miller 2007;Visconti & Elkin 2009).

The degree of habitat fragmentation
The KZN SS is highly fragmented at a small dispersal distance and as the dispersal distance increases from 500 m onwards, there is a much lower level of fragmentation.The majority of the KZN SS is located outside the EM.This explains the high number of components and greater level of fragmentation seen with the 'broad-scale KZN SS (full extent)' data set (Figure 2), which has an excess of 1200 components, in comparison to the other data sets (Figures 2 and 3), which have less than 200 components.The number of components is considerably less when fine-scale data are used.Furthermore, the number of components can be observed to have changed substantially when small dispersal distances are used for fine-scale data; however, with the use of large distances it can be noted that fine-scale data do not show such a major difference.The slight variation between 'finescale KZN SS in good condition (EM only)' and 'fine-scale KZN SS in any condition (EM only)' in Figure 3 implies that the inclusion of patches regardless of their ecological condition does not have a substantial impact upon the level of fragmentation experienced.

Connectivity levels
The overall landscape connectivity of the KZN SS for each of the different data sets is depicted in Figure 4.The general trend displayed for each data set is an increase in connectivity as the dispersal distance increases.The inclusion or exclusion of degraded patches had an important impact upon the overall landscape connectivity.The 'fine-scale KZN SS in any condition (EM only)' data set (Figure 4), which included degraded patches, showed a considerably higher level of connectivity.This outcome was expected and it indicates that it is possible for degraded patches to be stepping-stones between more desirable habitats.In addition, degraded patches are particularly important for species with a dispersal distance in excess of 500 m, evident by the higher IIC.There is an important difference between the uses of either broad-scale or fine-scale data evident in Figure 4.At 1000 m, both broadscale and fine-scale data that dealt with patches within the EM have a substantial difference of 85% in their connectivity levels.The broad-scale data set only captured large patches and did not capture many of the smaller patches.The smaller patches (included in the fine-scale data) increased the level of connectivity.Broad-scale data tend to underestimate levels of connectivity.The patches located within the EM show a higher level of connectivity, possibly owing to the smaller distance covered within the municipal area as compared to the distance covered by the entire extent of the KZN SS.The individual importance of each patch concerning the level of connectivity of the landscape is depicted in Figure 5, showing the significance of each patch and therefore the loss of connectivity should this patch be lost.The regions highlighted in red are considered the most connected and hence are essential in ensuring the connectivity of the region.Most of the patches essential to connectivity are located outside the EM, with connectivity levels slightly increasing with dispersal distance.There are several patches located outside the EM that are crucial to connectivity, of which the patch located to the south-east of KoShange is the most crucial in the landscape (Figure 5).Patches located near Assagay and Crest Holme are crucial for connectivity within the EM (Figure 6A).However, when placed in the broader context of the landscape, these patches hold only a moderate level of importance for connectivity (Figure 5).
The majority of vital patches are located away from the N3, as roads have the ability to act either as conduits or as a barrier, and in so doing, inhibiting movement of mammals and reptiles.The inclusion of degraded patches (fine-scale KZN SS in any condition (EM only)) has shown to increase the connectivity levels of a number of different patches within the landscape, which is apparent upon examination of the patches located at Sterkspruit, Vutha, Olwambeni and Crest Holme (Figure 6B and 6C).These patches increase in connectivity with the inclusion of degraded patches regardless of dispersal distance.There are noticeable differences between the uses of broad-scale and fine-scale data, which is discernible between Figure 6A (broad-scale KZN SS (EM only)) and Figure 6B (fine-scale KZN SS in good condition (EM only)).One can see that the patches of KZN SS located at Assagay and Crest Holme have been broken down into smaller fragments not holding as much importance as they did in Figure 6A (broad-scale KZN SS (EM only)).In addition, the group of fragments located near Sterkspruit (fine-scale KZN SS in good condition (EM only)), which held a relatively high importance, is absent (Figure 6A and 6B).

Discussion
The results indicate that, compared to its naturally occurring state, the KwaZulu-Natal Sandstone Sourveld is heavily fragmented.This could have severe implications for ecosystem functioning, especially for species with shorter dispersal distances (Cushman 2006;Whitcomb et al. 1981).These species may not be able to move freely if patches of KZN SS are not well connected, leading to local species extinction within isolated patches (e.g.Trenham 2001).Taxa which have a small dispersal radius -e.g.arachnids (Bonte et al. 2003), juvenile amphibians (Trenham 2001), invertebrates (Kotze & Samways 1999) and rodents (Kollmann & Schill 1996) -are more likely to be vulnerable to this high level of fragmentation.Species with larger dispersal distances (usually predators) are less likely to be affected by fragmentation (LaRue & Nielsen 2008;Wells et al. 2008).Future studies on habitat fragmentation and connectivity would benefit from species-specific data to incorporate movement patterns of species.
Well-connected fragments can form a corridor to facilitate the movement of species between the different habitat fragments (Bennett, Henein & Merriam 1994;Doko et al. 2011;LaRue & Nielsen 2008;Urban & Keitt 2001) and enable species persistence.However, the study ascertained that the KZN SS is not well connected -the highest IIC value amongst the data sets was recorded at 0.0063 (fine-scale KZN SS in any condition (EM only)) (Figure 4).This IIC value is considerably low in comparison to data obtained by Fourie, Rouget and Lotter (2015) for the Mpumalanga grasslands, where the lowest IIC value recorded was 0.049 for a 1000 m dispersal distance.The low connectivity level for the KZN SS is directly linked with the high degree of habitat loss and fragmentation currently being experienced.A similar study by Garcia-Feced, Saura and Elena-Rossello (2011) yielded minimum IIC values of 0.45 for forests in the Mediterranean region.
Spatial resolution of the land cover and, to some extent, habitat condition influenced the quantification of habitat fragmentation and connectivity.A comparison between the broad-scale data for both in and outside the EM showed that the habitat patches outside the EM had a higher level of connectivity.The proposed approach can be used to identify key patches (such as the patch located at Hoqweni (Figure 5)) to maintain habitat connectivity in the landscape.Degraded patches improved connectivity levels, providing stepping stones for species with smaller dispersal distances to gain access to pristine patches.Fourie et al. (2015) concluded that degraded patches are important for the connectivity of grasslands in Mpumalanga.This study also suggests that key degraded patches, if rehabilitated, would improve the connectivity of the landscape.
Broad-scale data over-estimated the level of habitat fragmentation, which is counter-intuitive to other studies (e.g.Payet et al. 2013).Thompson et al. (2009) showed that the use of broad-scale data could underestimate the quantification of degraded habitat.Map misclassification, which can result in errors with landscape pattern indices (such as mean patch size, patch shape and clumpiness), could have influenced the level of habitat fragmentation experienced with broad-scale data (Langford et al. 2006).The use of fine-scale data showed a higher IIC, possibly as a result of more stepping stone patches, and displayed a higher level of connectivity at smaller distances.The finescale data better depicted the current state of the KZN SS.High spatial resolution data are vital to adequately monitor and delineate important fragments (Boyle et al. 2014).
The proposed approach enabled the identification of essential patches to maintain habitat connectivity of this threatened ecosystem.

FIGURE 1 :
FIGURE 1: Original extent of the KwaZulu-Natal Sandstone Sourveld in and outside the eThekwini Metropolitan, based on the 2006 SANBI vegetation map.
into a single measure(Pascual- Hortal & Saura 2006).The Conefor Sensinode software has several connectivity indices available for use, of which the integral index of connectivity (IIC)(Pascual-Hortal & Saura 2006) and the probability of connectivity (PC) are the most highly recommended indices(Pascual-Hortal & Saura 2007).The software considers connectivity from a functional point of view.Hence, not only is the spatial configuration of the landscape (structural connectivity) taken into account, but also it encompasses the dispersal distances of species and their interaction with the physical structure of the landscape (functional connectivity) in the analysis(Pascual-Hortal & Saura 2007;Theobald 2006;Tischendorf & Fahrig 2000).The required inputs for the software are, therefore, both the spatial structure of the landscape and the dispersal distance of the study species.

FIGURE 3 :
FIGURE 3:The change in number of components experienced in relation to dispersal distance (the greater the number of components, the higher the level of fragmentation) for broad-scale KZN SS (EM only), fine-scale KZN SS in a good condition (EM only), and fine-scale KZN SS in any condition (EM only).

FIGURE 2 :
FIGURE 2:The change in number of components in relation to dispersal distance (the greater the number of components, the higher the level of fragmentation) for broad-scale KZN SS (full extent) and broad-scale KZN SS (EM only).

FIGURE 4 :
FIGURE 4: Overall landscape connectivity levels of the KZN SS (expressed as the change in the integral index of connectivity [IIC] with dispersal distance) using both broad-scale and fine-scale data.

FIGURE 5 :
FIGURE 5:The importance that each patch of KZN SS in a natural condition contributes towards the connectivity (expressed as the delta integral index of connectivity (dIIC), the output values range from 0 to 1, with 1 highlighting optimal connectivity) for the entire KZN SS within the 2006 SANBI vegetation map (broad-scale KZN SS (full extent)), based on 250 m dispersal distance.

FIGURE 6 :
FIGURE 6: The importance that each patch of KZN SS contributes towards the connectivity (expressed as the delta integral index of connectivity (dIIC), the output values range from 0 to 1, with 1 highlighting optimal connectivity) of KZN SS patches that fall within the EM at a dispersal distance of 250 m, based on: (a), broad-scale KZN SS (EM only); (b) fine-scale KZN SS in good condition (EM only), and (b) fine-scale KZN SS in any condition (EM only).

FIGURE 6 (
FIGURE 6 (Continues...): The importance that each patch of KZN SS contributes towards the connectivity (expressed as the delta integral index of connectivity (dIIC), the output values range from 0 to 1, with 1 highlighting optimal connectivity) of KZN SS patches that fall within the EM at a dispersal distance of 250 m, based on: (a), broadscale KZN SS (EM only); (b) fine-scale KZN SS in good condition (EM only), and (b) fine-scale KZN SS in any condition (EM only).
Jewitt et al. 2015)African National Biodiversity  Institute 2006).The land cover and vegetation maps were combined to ascertain the level of habitat loss.The broadscale land cover data had been classified into either 'natural' (predominantly made of indigenous plant species in varying degrees of ecological states as a result of soil erosion, invasive plant species and overgrazing) or 'transformed' (conversion of natural habitats to cultivated or urban areas where indigenous plant cover is no longer dominant).This did not differentiate habitat degradation and offered very little information pertaining to the ecological condition of the KZN SS.The fine-scale data used in the study comprised a 2011 land cover map (overall accuracy of 83.5%) (Ezemvelo KZN Wildlife, GeoTerraImage 2011;Jewitt et al. 2015)and a 2011 vegetation map(Scott-Shaw & Escott 2011).The finescale data used in the study offered more information regarding the ecological state on the KZN SS.The fine-scale data categorised the KZN SS into either 'pristine', The broad-scale data used in the analysis consisted of a 2008 land cover map (overall accuracy of 79%) (Ezemvelo KZN Wildlife, GeoTerraImage 2008) and a 2006 vegetation map (South African National Biodiversity Institute 2006).The 2008 land cover map was developed using SPOT 5 imagery (Ezemvelo KZN Wildlife, GeoTerraImage 2008; Jewitt et al. 2015).The 2006 vegetation map was developed using multiple sources such as ancillary vegetation maps, imagery (SPOT 5 2006 and 2006 aerial imagery), topographical maps and expert knowledge (Scott-Shaw & 'intermediate' or 'degraded' based on expert knowledge and site visit (Scott-Shaw & Escott 2011).

TABLE 1 :
Summary of data used for analysis, with variation of resolution, geographical extent and ecological condition.

KZN Sandstone Sourveld dllC
(Roberts et al. 2012)ce System (DMOSS) framework albeit based on a different approach(Boon et al. in press).The DMOSS, incorporated in eThekwini's town planning schemes, is a critical instrument to ensure that the development planning and assessment process adequately consider biodiversity concerns(Roberts et al. 2012).Moreover, http://www.abcjournal.orgOpen Access Source: Ezemvelo KZN Wildlife, GeoTerraImage 2008; South African National Biodiversity Institute 2006 KZN SS, KwaZulu-Natal Sandstone Sourveld.