Modeling Change Patterns in Nguru Forest in the Eastern Arc Mountains, Tanzania
This project was initiated as a small grant request from the Global Spatial Data Infrastructure (GSDI) by the University of Kwazulu-Natal and the Wami Ruvu Basin Water Office in Tanzania. The first volunteer conducted the analysis using the TM imagery and now the consortium has acquired new and higher resolution imagery (Rapid Eye) and the second volunteer Linda Delay, a GIS Specialist from New Mexico, conducted similar analysis on the Rapid Eye. The intention of the project was to assess the effects of fragmentation on the landscape using high resolution imagery. The following report was written by Linda DeLay and Mercy Ojoyi. A full copy of the report can be found in here.
By: Linda S. DeLay GISP, GISCorps Volunteer & Mercy Ojoyi, University of Kwazulu-Natal; in affiliation with: ICIPE, Nairobi Kenya & Natural resource Institute, NM, USA
Dynamics in land use and land cover (LULC) is still a serious threat to biodiversity conservation in the tropics. This is the case of Nguru montane forest, rich biodiversity hotspots in Tanzania, in the Eastern Arc Mountains. Change maps of Nguru forest in Morogoro region have been developed for 1984 and 2010 based on LANDSAT and RapidEye satellite data. The spatial fragstats program has been applied in quantification of how the ecosystem faces change at patch and class levels. Change analysis conducted in IDRISI indicates losses in montane forest, closed woodland and grassland. There was a net gain in anthropogenic disturbance (156 km2) leading to habitat loss and net gain in cultivated pastures (457 km2) between 1984 and 2010. Change at class level has been modeled and quantified using the Fragstats spatial statistical program. Change in patch area is seen in montane forest changes from 0.32 in 1984 to 1.61 in 2010. It is regarded as a useful platform for knowledge generation in planning measures for action in areas that remain highly vulnerable and affected in Nguru which is a rich biodiversity hotspot in the Eastern Arc Mountains. This project provides empirical scientific knowledge to authorities managing fragile fragmenting landscapes in sub-Saharan Africa.
Land cover plays an important role in the climate system. It’s key in maintenance of the ecological functioning of the ecosystem and in agricultural production. Land use on the other hand is defined by (Inglis-Smith, 2006) as the manner in which human beings utilize and alter land. Land use change in most ecosystems is caused by human activities (Lambin et al., 2001, Veldkamp and Lambin, 2001). The type of land use systems and practices put in place determines the level of productivity. Modification on land in most cases directly influences human activities such as provision of ecosystem services including food and timber provision, climate regulation, nutrient cycling and cultural identity (Daily, 1997).
LULC is increasingly becoming a global environmental issue of concern (Foody and Cutler, 2003). Many studies have established global consequences associated with land use and land cover change (de Chazal and Rounsevell, 2009, Foley et al., 2005, Haines-Young, 2009, Nagendra et al., 2013, Pereira and Borda-de-Água, 2013, Pérez-Vega et al., 2012, Reidsma et al., 2006, Zebisch et al., 2004). Studies by (Maynard and Royer, 2004, Voldoire, 2006) examined roles of future changes in greenhouse gases under different scenarios of land use and land cover change. They discovered that the relative impact of change in vegetation to greenhouse gas increased from 10% to 30% in the tropics. Change in climate due to change in land cover was significant compared to greenhouse gases at the regional scale where extreme dynamics of cover change were incurred (Meehl et al., 2007, Voldoire and Royer, 2005). They noted that the impact of greenhouse gas absorption was dominated by changes in land cover. Other scientists’ project changes in land use and land cover as major contributors to the expected forces in radiation in future projections (Kettleborough et al., 2007).
Satellite data has been resourceful in the analysis of historical land use and cover change over a series of time periods particularly in areas where changes have been rapid over a shorter time period (Giri et al., 2005). Change detection studies are designed to improve the understanding of regional dynamics, possible drivers of change and planning for further improvements (Boyd and Foody, 2011). The use of spatial tools for change detection over a series of time can create a proper basis in interpreting dynamics and resilience building in sub-Saharan Africa. This is thought to be ideal in effective monitoring, planning and conservation of vulnerable and fragile ecosystems.
Many natural habitats in East Africa continue facing increased anthropogenic disturbances altering the complexity of intact habitat structures and their functioning (Burgess et al., 2007, Shirima et al., 2011, Swetnam et al., 2011). Land use change continues to be a major threat in conservation of biodiversity hotspots. Nguru forest ecosystem, one of the key blocks rich in plant and animal species: both threatened and vulnerable globally face threats from anthropogenic activities such as farming, settlements, and urban sprawl. This most probably is projected to have effects on future conservation and monitoring plans. One key aspect to effective conservation requires knowledge of threats facing the ecosystem. It is envisaged that results emerging from this project will be significant towards establishment of an important basis relevant in conservation of the Nguru Montane ecosystem that remains fragile following increasing synergistic effects.
1.1 STUDY AREA
The Eastern Arc Mountains has been globally ranked among the top biodiversity ‘rich hotspots’ in the world (Hall, 2009, Menegon et al., 2008). Over the past years, previous researchers carried out documented biodiversity inventories of threatened, extinct and vulnerable species (Burgess et al., 2007), with an alarming increase in the rate of endemism and extinction based on the IUCN red list criteria (Hall et al., 2009). It supports important ecosystem functions and contributes to the world’s carbon storage (Swetnam et al., 2011). The block hosts the world’s endemic plant and animal species (Brooks et al., 2006, Burgess et al., 2007, Myers et al., 2000, Shirima et al., 2011, Swetnam et al., 2011). The region around Nguru Mountain range in Tanzania covers a total area of 31, 409 hectares. The Nguru South mountain range consist of four forest reserves namely Mkindu (Mikindu), Kanga, Magotwe (Mtibwa) and Nguru South (figure 1). It is dominated by lowland sub-montane and montane forest, woodland, agricultural farmlands, sugar plantations and human settlements and supports a population of 61,250 (Menegon et al., 2008). Major conservation threats include: subsistence agriculture, pit sawing, charcoal production and frequent fires and evapotranspiration problems affecting riparian sources (Doggart and Loserian, 2007). Recent research recorded potentially threatened species by IUCN in Nguru mountain ranges. Tree and shrub species such as Mussaenda microdonta, Memecylon cogniauxii, Syzygium micklethwaitii, Coffea mongensis Lasianthus pedunculatus, Zenkerella capparidacea, Polyscias stuhlmannii were identified as threatened and the Allanblackia ulugurensis as vulnerable in both mountain ranges (Platts et al., 2010). Besides, changing climatic patterns have led to changes in the habitat including biomass losses, displacement of people, lower river flows, and biodiversity losses. Further human-induced processes in Morogoro rural district have led to drying of major rivers, opening up intact forests, rapid species losses and conversion of communal land into farmlands.
Figure 1 Location of Nguru Forests
1.2.1 DATA PROCESSING
Landsat TM satellite images from 1984 and 2010 were selected to assess changes in land use and land cover across the years. The approximated area of interest was 10,500 square kilometers in areas surrounding Nguru Forest Reserves in the Eastern Arc Mountains of Tanzania. Landsat images on the USGS Glovis website (http://glovis.usgs.gov/) for the fewest cloud cover in the growing season of May and June. ERDAS IMAGINE 2011 was used to minimize haze in the images. Clouds and their shadows were masked out of the analysis by assigning no data values (zero) to these classes after using a supervised classification to aid in the delineation process. A cloud-effects mask which was merged from both images was applied to the change detection analysis. Examples of land cover categories were selected to train the supervised classification to identify spectral signatures and associate them with a specific user-defined category. Topographic maps for 1970s were also used in identification of land cover such as crops or plantations, grasslands, and forests. Elevation from digital elevation models (DEM 30m) aided decisions regarding montane forest locations. We equally used Google Earth software and the compiled satellite imagery to select land cover types. Within ERDAS IMAGINE, band combination 5 (short-wave infrared), 4 (near infrared), 1 (blue) were used in the R, G, B color guns to help distinguish between vegetation categories.
The land use / cover classification was derived from the USGS classification. The classification involved 8 categories: anthropogenic disturbance, cultivated pasture/ crops, open woodlands, closed woodlands, montane forest, grassland/shrub, wetlands/riparian and water/wet soil. Anthropogenic disturbances are a group of identified human activities such as settlements, subsistence farms, urban areas and agricultural fields. The samples included a mosaic of bare soil, structures, trees, rectangular crops, and roads. Cultivated pasture / crops is also a type of anthropogenic disturbance but was treated as separate land use. This category did not show the same pattern as the rectangular agricultural fields combined as anthropogenic disturbance. Perhaps these sites were a succession of tree crops. Wetlands/riparian sites were associated with some waterways identified in the image. This category was excluded in a subsequent supervised classification due to uncertainty of training samples. This classification was mapped and used in the IDRISI change detection analysis. The original classification with the riparian category was used in the fragstat analysis. Open woodlands were composed of mixed soil and grassland spectral signatures. The large plantations found were delineated from the Landsat images and exported as vector files in order to calculate change in number and area between years. The ERDAS IMAGINE raster function eliminate was used to decrease the amount of noise in the resulting classification by eliminating very small areas of a defined number of pixels by filling in with the nearest neighboring land use/cover class. We chose to eliminate and merge areas less than 10 pixels (300 sq. meters).
1.2.2 DATA ANALYSIS
IDRISI Selva remote sensing software (Clark Labs, 2012) and Land Change Modeler was used to quantify change in land cover/use between 1984 and 2010. Each supervised classification was used as inputs into IDRISI Land Change Modeler. Gains, losses and persistence of classes between years were based on cross-tab comparisons of each category area.
Fragstats metrics with a strong reference to fragmentation were extracted from a series of LANDSAT data sets. Four dominant classes classified were used in the analysis. They included less dense (LDF), dense forest (DF), woodland (WD) and grassland (GR). The majority filter rule was applied in ArcMap 10.2 in preparation of the classified image scene for processing in Fragstats. Fragstats is a spatial statistics program useful in computing metrics at patch, class and landscape levels (McGarigal & Marks, 1995). It is an effective spatial analyst program in forest fragmentation studies (Vogelmann, 1995). All classified images were converted into ASCII format in ArcGIS 10.2 then modeled in Fragstats. We used the raster version of the C program in Fragstats, which has the ability to extract nearest neighbor metrics. Fragstats diversity of metrics is efficient in explaining fragmentation with diverse applications (McGarigal, 2006).
It is evident from results that the habitat is undergoing loss which is a result of human activities (156.47km2) between 1984 and 2010 (anthropogenic disturbance, see figures 2-4). The greatest density of anthropogenic disturbance lies in the northwest corner of the scene near North Nguru, in the mid-elevation expanse between Mkuli and South Nguru, and in the southeast corner wrapped around South Nguru and Mikindo (figure 2). Density of disturbance moves toward the base of the forest reserves and the mid-section expands to fill in area between the reserves during 1984 to 2010. A road network circles the forest reserves. A closer examination of the images, reveal an expansion of village and urban development along these major roads.
Land use/cover which has undergone losses include: montane forest, closed woodland and grassland/shrub (figure 4). The area between North Nguru and Kilindi and south to Rudewa Forest Reserves transitioned primarily from grassland/shrub and closed woodlands into cultivated pasture and open woodlands (figure 2). There is association between anthropogenic expansion in the landscape and greater expansion of open woodland based on their distribution in the landscape between the two years of imagery. The area of open woodland may be the result of the opening up of closed woodland by anthropogenic disturbances such as village, farm and pasture development. As populations became denser and urban areas grew to fill in lowlands, perhaps native grasslands were converted to agriculture and pasture.
Closed woodlands were replaced by or exchanged with open woodland (254 Km2) in 2010, figure 3. The pattern is scattered along the west side of the Closed woodlands were replaced by or exchanged with open woodland (254 km2) in 2010, figure 3. The pattern is scattered along the west side of the study area and concentrated south of Mkuli. An exchange between open and closed woodland (135 km2) occurred east of North Nguru and Kilindi (figure 3). Perhaps, over this 26 year time period, the woodlands had grown denser due to regrowth. More likely, with increased settlement, land use had changed resulting to an increase in cultivation. Higher resolution imagery or field validation would assist in the interpretation.
Figure 2 LULC in Nguru Montane Ecosystem
Figure 3 Patterns of gains, losses and persistence
Figure 4 Net change between 1984 and 2010
Table 1 Patterns in Land Cover
Montane forest, woodland and grassland cover types have suffered losses. This could, in part, be a result of forest clearing in Nguru montane ecosystem in support of farming activities such as cocoa, maize, banana farms. There was a gain in cultivated pasture / crops by 457.33km2 between 1984 and 2010 (table 1). Expansion of built-up areas, urban settlements at Turiani and Mtibwa sugar plantation is another reason that could be a possible driving force to loss of these ecosystems. The total area coverage for plantations in the year 1984 was 50.53 km2 which grew to 118.96 km2 in 2010. That was a net gain of 68 km2 in plantation size.
Fragmentation metrics illustrate a landscape that is undergoing fragmentation. Change in patch area, edge density and PARA is seen in montane forest from 0.32 in 1984 to 1.61 in 2010.
Table 2 Fragstats metrics of eight land use/cover categories
Dynamics in patch metrics are good indicators of a landscape undergoing fragmentation (Fahrig, 2003, Jha et al., 2005, McGarigal, 2006, Vogelmann, 1995). Mean patch area for instance for montane forest changes from 0.32 in 1984 to 1.61 in 2010. This is an indication of a landscape that is undergoing negative transformation. In ecology, change in patch area has been identified with a fragmenting landscape (McGarigal & Marks, 1995). The patch number, patch density and mean patch area characterize the extent of fragmentation of natural vegetation (Jorge and Garcia, 1997). With a fragmenting landscape, a decrease in the habitat and patch size and an increase in patch number is dominant (Fahrig, 2001, Fahrig, 2003, McGarigal, 2006, Wiens, 1995).
Supervised Classification based on RAPIDEYE
A RapidEye satellite image was acquired for the area around Nguru North Forest Reserve in order to better determine land use/cover and placement of anthropogenic disturbance within the reserve (figure 1). The RapidEye scene was acquired in December 19, 2010 with a pixel resolution of 5 m. The image is composed of 5 bands that include a red edge and near-infrared. The red edge section of the spectrum (690-730 nm) is reported to increase discrimination between vegetation cover classes very useful in assessing effects of disturbances.
ERDAS IMAGINE 2011 was used in a supervised classification of the satellite imagery. Clouds and their shadows were masked out of the analysis by assigning no data values (zero) to them after an initial classification to aid in identification. Examples of land use/cover categories were selected to train the supervised classification to identify spectral signatures and associate them with a specific user-defined category. We had no data from ground surveys to use in classification of the Nguru area but used information and land cover maps from literature to determine predicted locations of cover types (Kindt, et. al, Vol. 3 and 4; United Republic of Tanzania Ministry of Natural Resources and Tourism). Maps from literature were georeferenced within ESRI ArcMap 10.1 and overlaid onto the satellite image to better select training samples. The 5 meter pixel resolution of the Rapid Eye imagery and 1-5 meter resolution of Google Maps imagery helped identify features. The ERDAS IMAGINE raster function eliminate was used to decrease the amount of noise in the resulting classification by eliminating very small areas of a defined number of pixels by filling in with the nearest neighboring land use/cover class. We chose to eliminate and merge areas less than 20 pixels (100 sq. meters).
Figure 5 RapidEye Analysis in Nguru North
The classification included 13 land use/cover categories: bush land, cultivated lands (anthropogenic disturbance), escarpment, grassland/meadow, montane forests, woodlands, open woodlands, riverine woods, urban development (anthropogenic disturbance), sandbars / bare soil, open water, floodplain, and deciduous woods (figure 2). Originally, cultivated lands were split into dark and light classes based on the apparent predominates of plant or soil cover. Montane forest was also originally split into montane forest 1 and 2 categories based on the distinctive spectral signatures and height differences (table 1). The classes were generalized and merged before fragmentation analysis for better interpretation. Compared to the Landsat classification, we were able to make finer distinctions among land use/cover types with RapidEye’s finer image resolution. We confined the supervised classification to 206,585 hectares of land use / cover surrounding all of the forests reserves that fell within the satellite scene. Of that area of interest, 125,088 hectares of that analysis fell within the Nguru North mountain block bounds (figure 6).
Figure 6 LULC based on supervised classification
Moderate-dense woodlands made up 50.9% of cover whereas the sparser open woodlands made up 7.8% of the total land cover. The second most abundant category was anthropogenic disturbance (urban development and cultivation) which comprised 24.3% of total cover. Floodplain (soil and sparse shrubs/grasses), escarpment (rocky/sparsely vegetated areas), and grassland/meadow cover made up, respectively 6.9%, 3.2% and 1.6% of the total.
It is evident from study findings that Nguru montane ecosystem is undergoing severe fragmentation. There is an urgent need to incorporate appropriate measures which will minimize further habitat destruction and loss. The use of remote sensing has been useful in generation of change maps for 1984 and 2010. Possible driving forces have been documented. Fragstats program has been used in quantification of class and patch area dynamics. It is envisaged that knowledge generated from this project will be useful in planning measures for action in fragile ecosystems in Nguru Forest.
We thank the GSDI co-ordination team for their great support in successful implementation of the project. RapidEye Imagery was provided by DLR Germany. Fragstats Program, ArcMap 10.2, IDRISI Selva and ERDAS IMAGINE 2011 were applied in the analysis. Field support was provided by Sokoine University of Agriculture, forestry department, Tanzania Government Forestry departments in Morogoro region and the Wami/Ruvu Basin Office in Tanzania.
References are included in the paper.