Instructor-derived Example
Contact: David Meek
(dmeek@uga.edu)
Department of Anthropology
University of Georgia
Part 1. Longitudinal Landscape Change Analysis
In this step we conducted a land cover classification of 5 Landsat images. Below is a map of land cover change, and then follows an interpretive analysis of the mapping process.
 |
| Fig. 1 Supervised Classification of Land Cover Change |
Narrative reflections on Land Cover Change Trajectory:
The 2009 image shows what appears to be
a very heterogenous landscape. Large areas seem to be barren of
vegetative cover. Smaller areas do have forest cover, but they are
the minority. There is a small urban area in the center of the image,
and small roads criss-cross the landscape. When additional imagery is
added (1986, 1997, 2000, 2005, 2009), a narrative of landscape change
begins to emerge. We notice that as of 1986 there was significant
forest cover in this settlement, but at the same time a large amount
of forest cover had already been cleared. This trajectory continued,
and in subsequent images the ration of cleared land to forest cover
continued to increase until present. Two points of note are that the
area that would become the settlement in 1997 was already
significantly deforested by the time that the MST settlers arrived in
1997. This observation challenges narratives of sole deforestation on
the part of the Landless peasants. This preliminary finding
highlights the importance of temporal scale: for instance, looking
just at the 2009 image, one might think that the landscape was singly
deforested by the MST, expanding the scale to consider imagery prior
to their occupation, however, one gains a more nuanced understanding
of the effect of temporal scale on analysis of remotely sensed
imagery.
Narrative Reflection on Creating Training Sample:
Creating training samples was an
interesting, and somewhat problematic process. For the 1989 image it
was fairly obvious what area was forest cover, and what area was
non-forest, and what area was generally non-forest.
 | |||
| Fig. 2 Area in 1986 image |
 | ||
| Fig. 3 Non-forested Area in 1986 image |
However, as shown in Fig. 3, and more so in Fig 4. that binary division became somewhat complicated as the land cover was not completely cleared or completely forested as homogenous areas (based upon pixel color, and grain size).
 | ||
| Fig. 4 Questionable Area in 1986 image |
For other images that fairly simple (although still problematic)
differentiation became more complicated. For example, in the 2000
image there are a variety of areas that seemed to look like forest
cover, however had different coloring, and different texture.
 | ||
| Fig 5. Two polygons drawn around forest cover of likely different classes in 2000 image |
My
interpretation is that this might be secondary growth resulting from
burning, or it could be agroforestry. In general the spatial scale of the Landsat imagery made it
somewhat difficult to differentiate anything other than broad feature
classes. For instance, I was fairly comfortable classifying forest
cover, but it was unclear whether some of that forest cover was
anthropogenic, i.e. orchards. In general, I found that the Landsat
imagery made it difficult to observe the human presence in this
landscape. For example, in Fig. 6 it is unclear what form of anthropogenic development this is, is it a school, a house, a ranch, a plant nursery? The scale, and lack of ethnographic data, makes it impossible to determine.
Part 2:
Complicating
a Simple Picture: Adding Very-High Resolution Imagery and
Ethnographic Data
First, we overlayed a shapefile of GPS points gathered during ground truthing. We explored the land cover classes in the 2009 image (when the GPS points were taken) in the Landsat imagery. Below are screenshots of each GPS point alongside the interpretation based on the Landsat imagery.
 |
| GPS Point 1 Landsat Image |
 | |||||||||||||||
| GPS Point 2 Landsat Image |
The dark color of this land cover area led me to believe that is was also an agroforestry area. Given the difference in shading and texture, it might be a different crop from the previous point.
 | ||
| GPS Point 3 Landsat Image |
The coloring of this area suggests that it is deforested, however, not much else is definitively clear, i.e. whether or not there is habitation, or it was recently burned.
 |
| GPS Point 4 Landsat Image |
 |
| GPS Point 5 Landsat Image |
The coloring and proximity to other deforested areas suggest that this area has been cleared and is under some form of cultural management. Whether that is a subsistence crop or another form of land use is not clear.
 |
| GPS Point 6 Landsat Image |
Critical Reflections Following Addition of GPS Points:
The process of adding the GPS points somewhat complicated my original interpretation. They didn't have any observations attached to the attribute table, however and so made me wonder what was observed at these points. For example, the land cover surround point 1 and point 2 were fairly similar visually, what was the reason for collecting two points? The points suggest that there is detail which is not accurately visually interpretable through the Landsat imagery.
Addition of Very-High Resolution GeoEye Imagery:
At this step we added very-high resolution GeoEye imagery, and then compared that imagery with the Landsat imagery on each GPS point.
GPS Point 1: GeoEye image (left) and Landsat image (right). With the addition of the GeoEye imagery, it became clear that Point 1 was forest cover. Additionally, the very-high spatial resolution of this image, and the shape of the trees which are now visible, suggests that these are agroforestry plantings.
| |||||||
GPS Point 2: GeoEye image (left) and Landsat image (right). With the GeoEye image it appears that this is a stream surrounded by forest. The dark color in the Landsat image appears to have been misinterpreted.
| |||||||||||||||
| ||
GPS Point 4: GeoEye image (left) and Landsat image (right): The addition of the GeoEye image really changed my interpretation of this area. Whereas I interpreted this as a forested area in the Landsat image, in the GeoEye image it appears that it is not only deforested, but contains the burned remnants of large trees. Notice in the image below the fan shaped shade. Given that there is forest cover in other areas, indicating that this is not an artifact of seasonality, one might surmise that ta standing burned tree trunk, and is a visual artifact of deforestation. Again this detail is completely obscured in the Landsat image.
|
GPS point 5: GeoEye image (left) and Landsat image (right): The GeoEye image suggests that this is an agricultural field of some sort. There appears to be several standing live trees in the vicinity, and a variety of trail criss-crossing the landscape. None of this information is viewable in the Landsat image.
|
GPS Point 6. GeoEye image (top) and Landsat image (bottom): The GeoEye imagery enables the interpretation that this is a point in a cleared area of some sort. There does not to be any built structures in the area, but there are several trails. To the side of one of the trails appears to be primary forest, and to the other side, where the clearing is, there appears to be secondary growth of various ages.
|
| GPS Point 6 Landsat Image |
__________________________________________
GPS Point 1: GeoEye image (left) and Landsat image (right).
Ethnographic data:
At this step we added ethnographic data, and critically assessed its impact on our analyses.
GPS Point 1: GeoEye image (left) and Landsat image (right).
Ethnographic data:
| Land Cover Observation | Additional ethnographic data | ||
| Agroforestry (mango, cupu, inga) | M. has been living here for 15 years and has several hundred fruit trees in this area. Approximately 40 species of fruit trees are planted. In addition to agroforestry, which is mainly for subsistence, M. derives income from raising a small number of cattle. |
| |||||||
After considering these ethnographic data, it makes sense that this is an agroforestry location, and that it is in close proximity to the cattle raising. What is interesting is that there seems no way to get any sort of accurate estimate of species diversity, even from the high resolution GeoEye image.
GPS Point 2: GeoEye image (left) and Landsat image (right).
| Land Cover Observation | Additional ethnographic data |
| Acai agroforestry plantings along river margins | B. has been planting acai saplings along the river bank for the last several years. They are still small, approximately knee high. |
|
The ethnographic data in this context is interesting as there is no way, at least at this stage, that even with the high-resolution imagery that one would be able to identify B's actions of planting acai. Perhaps in several more years such an analysis would be possible
GPS Point 3: GeoEye image (left) and Landsat image (right).
| Land Cover Observation | Additional ethnographic data |
| Path between areas of primary forest cover | This is on a hunting trail within an area of primary forest cover on B's land. |
These observations support my analysis of the GeoEye image, but also point to the complicating process of attributing usage, i.e. I was able to identify this was a trail, but not that it was for hunting.
GPS Point 4: GeoEye image (left) and Landsat image (right):
| Land Cover Observation | Additional ethnographic data |
| Standing burned out castanha (Brazil nut) tree | This hulking relic seems like out of a Dr. Seuss book. Towering burned out Brazil nut trees like this dot the landscape, and are an artifact of the deforestation. B. indicates that his land was largely burned accidentally by a neighbor who had set a fire to clear a section of forest, and that fire got out of control on his land. This is extremely common in this area. |
|
The ethnographic data support my interpretations of this image.
GPS point 5: GeoEye image (left) and Landsat image (right):
| Land Cover Observation | Additional ethnographic data |
| Agricultural field (corn/beans/squash/okra) | This is an agricultural field containing various crops planted together. |
|
The ethnographic data support my observation; however, it is noteworthy that it would likely be impossible to determine intercropping at any spatial scale with the imagery.
GPS Point 6. GeoEye image (top) and Landsat image (bottom):
| Land Cover Observation | Additional ethnographic data |
| Empty reservoir (dirt pit) | This is an interesting spot: B. tells me that he had been part of a credit program to build watering holes for cattle. However, the extension agent didn't properly assess the soil quality, and the water ran out. Now all that is left is the empty resevoir, and B's debt to the bank. |
|
The ethnographic data is very interesting in this case as it shows there is no way to accurately determine the history of certain features on the ground. In this 2-dimensional image, one can not tell this is a sunken dirt pit, much less than it ever had water, or that was created as part of a misconducted credit program.
Conclusions: Through this lab we have explored the impact of spatial and temporal scale on land use and landscape change studies. In terms of temporal scale, it was clear that expanding the chronosequence provided important insights into the trajectory of landscape change in the study region. In addition, comparing Landsat imagery (with a pixel resolution of 25 meters) with GeoEye imagery (.5 meter resolution) demonstrated the value of spatial scale in interpreting the cultural landscape. However, adding ethnographic data also problematized the value of scale. For example, irregardless of how high resolution imagery gets, there is no way to obtain the scale of nuanced information that one gets through ethnographic data.
