Methods
To survey the target area we took Landsat images found on GloVis from different years in order to provide a proper before and after of Virunga National Park and the surrounding area. When selecting years that would be most important to focus on we planned around the years of the First Congo War (October 24, 1996- May 16,1997) and the Second Congo War (August 2, 1998- July 18, 2003). The methodology behind this decision was that the span of 1996-2003 was the period in which Congo experienced the most human disturbance since it was in the middle of a war period. Additionally, there were a number of volcanic eruptions from Mount Nyiragongo and Mount Nyamuragira which are both found in the southern portion of Virunga National Forest. It’s also during this time span the Soco International PLC began taking serious interest in oil exploration around the region. Given all this, we chose Landsat images before and after to monitor the ongoing changes resulting from these disturbances in order to show the impact on Africa’s oldest national park.
The first step we took in this process was selecting images from the same standardized set in order to create the best controlled experiment. The reason we selected GloVis and Landsat images was due to the fact that they were periodically taken with the same camera, as opposed to ASTER images which need to be requested- thus adding more variability to the data. After that there was some difficulty in selecting images without cloud cover, however we were luckily to find images of the Northern, Central and Southern portions of Virunga National Park that had minimal or no cloud cover to corrupt the data. Landsat 7 (ETM+) was used to collect images of 2005 and 2014, and Landsat 5 (TM) was used to acquired the images of 1987 and 1995.
Once the images were downloaded, we began our analysis using the ENVI Classic software. We first loaded bands 1 through 5 and band 7 for each image. We then proceeded to make a layer stacking for all the images. Due to a failure in the ETM+ sensor, the images from Landsat 7 displayed black stripes of missing data along the edges of the image. We corrected the gaps by going to the Topographic function in the menu bar and selecting Replace All Bad Values. We set the bad value as zero and the minimum and maximum values were also set to zero.
We then performed a Normalized Difference Vegetation Index (NDVI) analysis on every image to determine the vegetation cover of the Virunga National Park. We loaded the layer stacked files in ENVI Classic, and then selected the Transform function in order to create the vegetation indices. We made a spatial subset of our study area using the boundary vector obtained from the Protected Planet website. We identified the type of the input file as TM (Thematic Map), and changed the band values to Red: 4 and NIR: 3 due to the order of the bands. For every NDVI image, we created a density slice so we could distinguish the health of the vegetation. In the Display menu bar, we selected Overlay and then Density Slice. We set the range minimum value to 0.000001 and the maximum value to 1.000000. We chose six ranges to get a more detailed image. The following table contains the values of each range and its respective color:
The first step we took in this process was selecting images from the same standardized set in order to create the best controlled experiment. The reason we selected GloVis and Landsat images was due to the fact that they were periodically taken with the same camera, as opposed to ASTER images which need to be requested- thus adding more variability to the data. After that there was some difficulty in selecting images without cloud cover, however we were luckily to find images of the Northern, Central and Southern portions of Virunga National Park that had minimal or no cloud cover to corrupt the data. Landsat 7 (ETM+) was used to collect images of 2005 and 2014, and Landsat 5 (TM) was used to acquired the images of 1987 and 1995.
Once the images were downloaded, we began our analysis using the ENVI Classic software. We first loaded bands 1 through 5 and band 7 for each image. We then proceeded to make a layer stacking for all the images. Due to a failure in the ETM+ sensor, the images from Landsat 7 displayed black stripes of missing data along the edges of the image. We corrected the gaps by going to the Topographic function in the menu bar and selecting Replace All Bad Values. We set the bad value as zero and the minimum and maximum values were also set to zero.
We then performed a Normalized Difference Vegetation Index (NDVI) analysis on every image to determine the vegetation cover of the Virunga National Park. We loaded the layer stacked files in ENVI Classic, and then selected the Transform function in order to create the vegetation indices. We made a spatial subset of our study area using the boundary vector obtained from the Protected Planet website. We identified the type of the input file as TM (Thematic Map), and changed the band values to Red: 4 and NIR: 3 due to the order of the bands. For every NDVI image, we created a density slice so we could distinguish the health of the vegetation. In the Display menu bar, we selected Overlay and then Density Slice. We set the range minimum value to 0.000001 and the maximum value to 1.000000. We chose six ranges to get a more detailed image. The following table contains the values of each range and its respective color:
Range | Color |
---|---|
0.000001 to 0.166668 | Coral |
0.166669 to 0.333334 | Yellow |
0.333335 to 0.500000 | Green1 |
0.500001 to 0.666667 | Green2 |
0.666668 to 0.833334 | {0, 122, 139} |
0.833335 to 1.000000 | Purple3 |
A change detection analysis was also performed to determine the changes in the vegetation over the years. We used the same NDVI images that we previously created. Using the ENVI software, we went to Basic Tools, then Change Detection and next to Compute Difference Map. We left the default settings except for the Change Type: we selected the percent difference option since it gives more accurate data. We did a change detection map for the following time periods: 1987 to 1995, 1995 to 2005, 2005 to 2014, and 1987 to 2014.
The final step we took was performing an unsupervised classification of Virunga National Park and the surrounding area. Utilizing the Iterative Self-Organizing Data Analysis Technique (ISODATA) provided from the images we were able to create a visualization of the land cover for the target region. The Thematic Mapper(TM) and Enhanced Thematic Mapper Plus (ETM+) systems aboard the Landsat satellites were able to capture an array of spectral signatures for the region from the photos we collected. After that we settled on 11 different classifications to properly account for each feasible result that could occur in the region, while still remaining as detailed as possible. After obtaining and identifying the spectral signatures received we then obtained post classification Change Detection Statistics to provide a numerical example of the change that occurred in the Northern, Central and Southern portion of Virunga. The best results differed in each region due to the cloud cover that was represented in the target area. However, maintaining that we were looking to get a before and after around the 1996-2003 range, we were able to obtain exceptional results from February 1995 and January 2014 for the Northern region; February 1995 and July 2005 for the Central region; and August 1987 and February 2005 for the Southern region. The results of this research dictated an overall shift in the landscape from areas with lush vegetation to that consisting mainly of low grassland. The result of which is better represented in the ISODATA section.