The New England chapter of Urban and Regional Information Systems Association (NEURISA) in conjunction with…
Assessing Land Use/Land Cover Change in the Eastern Arc Mountains in Morogoro Region
Project coordinated by Mercy Ojoyi, Environmentalist, University of Kwazulu-Natal and Wami Ruvu Basin Office, Tanzania; Techical support by GISCorps Volunteer, Imroz Raihan
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 application requested a Remote Sensing Specialist who works for the United Nations to provide technical assistance in assessing land use change. As a partner of GSDI in its small grant program, GISCorps recruited and deployed Mr. Imroz Raihan, a Remote Sensing Specialist, to conduct image processing and analysis. Mr. Raihan contributed a very specific part of the project which includes vegetation change analysis. In line with his contribution it is expected that there will be further exploration using Remote Sensing techniques to assess effects of land use and land cover change using LANDSAT and then apply high resolution images to assess fragmentation effects on the ecosystem.
Vegetation dynamics are considered to be of primary importance for assessing ecosystem fragmentation. This research tries to employ multispectral remote sensing data to measure vegetation change. Tri-decadal Landsat images were processed to extract vegetation coverage at 3 different time periods. Extracted vegetation coverage for the years 1991, 2000, and 2011 were overlaid to delineate the process of deforestation in the selected study area. To improve the extraction of vegetation coverage, raw Landsat images were calibrated and atmospheric corrections were applied using the radiative transfer model ‘MODTRAN4’. Normalized Difference Vegetation Index (NDVI) values were calculated from the processed images. By selecting different threshold levels from NDVI values, four different land cover classes were developed. These included bare soil, grassland, shrubs, and dense forest. In parallel to this, unsupervised classifications were applied and compared with the NDVI analysis. Finally, year-wise land cover class statistics were calculated and compared in order to present the vegetation dynamics within the time frame.
The Morogoro region is located in the Eastern Arc Mountains of Tanzania, which is regarded as one of Africa’s renowned hotspots with endemic animal and plant species. It is dominated by lowland submontane and montane forest, woodland, agricultural farmlands, sugar plantations, and human settlements. It supports a population of 61,250. Major conservation threats include: subsistence agriculture, pit sawing, charcoal production, frequent fires, and evapotranspiration problems affecting riparian sources. Charcoal production and agricultural farming have been major contributors to forest degradation with an impact on the region’s average precipitation (IUCN, 2010). Considering the ecological importance first the administrative boundaries of Morogoro Urban, Morogoro Rural, Mvomero, and Kilosa region were selected for the study area (see Figure 1).
Figure 1: Extent of the study area. Yellow line depicted on Landsat image represents the extent. Locality names following first and second level administrative boundaries are also plotted
Data and Data Sources
Tri-decadal Landsat 30 meter resolution images were used for this project. Selected scenes were captured on the years 1991, 2000, and 2011. To eliminate seasonal variations, selected Landsat scenes were captured for the same month of the year; and to make better representation of vegetation dynamics, selected scenes were captured for the vegetation growing months. Gap-filled 2011 Landsat imagery products were used to compensate for scan line error.
Free Landsat images were downloaded and processed to drive vegetation dynamics. Data processing involved a series of structured steps. All three Landsat scenes were calibrated as per the guidelines provided by the Landsat Science Team. Calibration parameters and procedures were also varied for three different scenes since they were captured using different Landsat Sensors (Landsat ETM, TM).
Digital number (DN) values of the downloaded scenes were converted to radiance as prescribed in the Landsat hand book. The Radiance to Reflectance conversion yields at satellite reflectance which is still affected by the atmospheric effect. To calculate ground level reflectance, the state-of-the-art atmospheric band model radiation transport model, MODTRAN4, was applied for the test site images. More details about MODTRAN4 are available at the following link http://modtran5.com/. Land cover samples were taken from calibrated and atmospherically corrected images and then plotted against their wavelength. Six spectral bands were plotted – 1,2,3,4,5 and 7. The wavelength of spectral bands ranged from 0.485 to 2.22 micrometers.
Figure 2: Spectral profile of four different land cover samples. Wavelength of spectral bands ranged from: 0.485 to 2.22 micrometers plotted on X axis and raw DN value of six spectral bands (1,2,3,4,5 and 7) plotted on Y axis.
Based on the data in Figure 2, an upward trend of the blue band (0.485 micrometers) is considered the most prominent atmospheric affect. This is likely caused by atmospheric aerosol scattering, also known as ‘skylight’. An accurate atmospheric correction should be used to compensate for the skylight in order to produce spectra that more truly depict surface reflectance. After applying MODTRAN4 on calibrated images, the vegetation reflectance curve (as shown Figure 3) displays a more characteristic shape, with a peak in the green, a chlorophyll absorption in the red, and a sharp red edge leading to higher near infrared reflectance.
Figure 3: Spectral profile of four different land cover samples. Wavelength of spectral bands ranged from: 0.485 to 2.22 micrometers plotted on X axis. Calibrated and atmospherically corrected reflectance of four land cover sample class plotted on Y axis.
Extraction of Vegetation Cover
The Normalized Difference Vegetation Index (NDVI) is a simple graphical indicator which was used to analyze remote sensing measurements and to assess whether the target being observed contains live green vegetation. To extract vegetation coverage NDVI values were calculated on three different scenes. Negative values of NDVI (values approaching -1) correspond to water. Values close to zero generally correspond to barren areas of rock, sand, or snow. Lastly, low positive values represent shrub and grassland while high values indicate temperate and tropical rainforests (values approaching 1).
Other classification techniques were also applied. Unsupervised classifications were also separately applied inside the extent of the study area on the three images. Four primary clusters were found. Initial investigation and local knowledge were used to define the four clusters – bare soil, grassland, vegetation, and dense forest cover. NDVI threshold values were determined with the help of the land cover classes found from the unsupervised classification. The NDVI algorithm subtracts the red reflectance values from the near-infrared and divides it by the sum of near-infrared and red bands*. The following formula was used to calculate NDVI:
NDVI = (NIR-RED) / (NIR+RED
Figure 4: True color composition of tri decadal Landsat image shown on the first row. Calculated NDVI image shown on the second row. Images on the middle captured on 2001, left 1991 and on right 2011.
Although the land cover classification made from NDVI images is significantly affected by a number of variables, the captured image quality is the only determinant of the quality of the classification result. Everything drawn here is from the three images found in the Landsat archives.
For the 2000 image, vegetation coverage was surprisingly low and many places were converted in bare soil when compared to 1991 image, especially in the middle and south western parts of the image. Development activities create many clear-cut pockets inside dense forest in 2011. In many cases clear-cut pockets inside dense forest are comparatively small considering the resolution of Landsat images, which ultimately failed to trace deforestation at the micro level by 2011. Outside the Morogoro urban area the natural afforestation process over red colored dry-land is much higher, especially on the southwestern part of the 2011 image. The thickness of a linear pattern of vegetation was quite high in 2011 compared to 2000.
Figure 5: Extracted land cover class derived from NDVI images.
Land cover classification statistics were developed by counting the number of pixels falling inside each category of land cover. The total number of pixels varied from year to year since the three images were captured using different Landsat sensors with slight variation of pixel size. The results might change if we concentrate only on developed areas. A similar analysis would result in a different change pattern if focused only on the eastern or western side of the test site.
An overlay analysis was performed by overlaying two different years of vegetation coverage. The 2011 vegetation cover was overlaid on the 1991 vegetation cover. Thus the deforested vegetation will appear in a different color underneath the 2011 image. Similarly, by overlaying the 1991 vegetation on 2011 vegetation delineated afforested areas in between those two years will be evident. Identification of afforestation and deforestation were based on comparisons among all three study years – 1991 to 2001, 2001 to 2011, and 1991 to 2011. See Figure 6.
Lastly exported PDF map elements were kept in vector format; therefore the host agency can print the layouts to any suitable paper size without compromising quality. Extracted land cover classification data and raw data were delivered to the host agency.
This study provides a general picture of change in the area. Clearly, human induced processes are contributing to the decrease in areas formerly covered under forests. More analysis is needed. There are many variables affecting measurable changes from one Landsat image to the next, including sun illumination, aerosol content, precipitation, and vegetation phenology at the particular time the image was captured. Accounting for these factors will require more time and staffing than was available for this project. However, the results of this initial study should inform the next stage in the work of the Wami Ruvu Basin Water Office.
Figure 6: Vegetation Change Analysis Map.
Partner Agency’s Future Plans
These results will be used for planning, monitoring and conservation of vulnerable and affected habitats. It is envisioned that the results will be useful to stakeholders of the project who are directly involved in conservation of water resources. Stakeholders include the Wami Ruvu Basin Water Office in Morogoro region, Uluguru Conservation and Monitoring Office-Eastern Arc Mountain Office in Tanzania, and Kitulangalo Forest Office in the Morogoro region. In the next steps, another volunteer will be recruited to assist with remote sensing tasks requested.