H12+V10+-+Western+Brazil

Changes in Western Brazil Page by: Amanda Hoffman

Western Brazil Landsat Data

The area I have chosen to study is MODIS tile H12 V10. The overlay of this tile can be seen the GoogleEarth image below. The area includes cities such as Rondonia and Mato Grosso in Western Brazil (I most likely won't look too much at the Bolivia area of my tile). As you can see in the Classification map below from 2000 (made using Land Cover Type 1 MODIS Classification) Rondonia is in an area of mostly forest, specifically, the Amazon Rainforest. Mato Grosso in 2000 was in more of a crop land/agricultural area. As shown in the GoogleEarth image though, Rondonia no longer looks like a rainforest, it looks more urban. I expect that when I create a classification map for 2008 the area of Rondonia will be classified as more urban than in this 2000 map. (I would have done this today but a blizzard hit so I couldn't make it to school! Boo!) From the general map of change on the main page of this wiki site I know that change is occurring in this area. My hypotheses as to why can be found below the two images.



Before the 1970's all of Rondonia was covered completely by the Amazon Rainforest. Then intensive deforestation and settlement occurred. From 1978 to 1988, 15,000 sq km per year of land was deforested (source: http://earthshots.usgs.gov/Rondonia/Rondonia). (NOTE: This conflicts with my 2000 classification map, which shows no urban in the area where Rondonia should be. When I am able to get back into the lab I am going to try different classification types.) This area has been studied extensively, but it seems as though all research occurred before 2000. However it is evident from the global change map that change is still occurring. I believe that this is due to the same things that caused deforestation previously in this area. Growing populations have been forcing people to move to more open spaces. In search of more land on which to grow food, people have been cutting down the rainforest to clear pasture and cropland. I think that the intensity of this operation has either slowed, which has caused less research interest, or that the same is happening now and people just stopped caring about it since it wasn't stopping.
 * Hypotheses**

__**February 8th, 2011 Assignment**__ For my three maps I chose to use EVI. I chose this vegetation index because NDVI too easily saturates. Since I am working in the Amazon Rainforest, where there are very slight difference in vegetation, NDVI would not have been a good choice. Especially because everything is generally green. I chose the the 105th (S. Hemisphere Fall), 201st (S. Hemisphere Winter) and 281st (S. Hemisphere Spring) days of 2010. I did this to show seasonal variations in vegetation in my area, and also these days had the best data available. Many of the fall dates were missing a lot of data due to atmospheric interference. I made three maps, 1 with ArcMap and 3 with ENVI. I chose not to put the ArcMap map here (or make all three with ArcMap) because I could not figure out how to get the same scales for comparison (and since I knew how to do it in ENVI I didn't bother).

As seen in the first two maps, my tile is much more vegetated in the fall time than in the winter. The largest change occurs in the southeast, which is to be expected because the temperature gradient rises from South to North. However, the changes are not huge because my tile is part of a rainforest ecosystem, which is typically warm all year. The missing data areas are also in the same place, which is interesting, I'm not sure why this is. The brown spot in the middle is also the same. I hope to investigate this spot more in depth later.

In the third map (Spring) it is much greener in the north, but not as green in the south. This makes sense seasonally, because the growing season in Spring in the south will start later than in the north. I also think the reason it is greener in Fall in the south rather than in Spring is due to the start of the growing season along with that in fall, it hasn't gotten cold enough to kill off a lot of vegetation yet. This is also the map with the largest amount of missing data, which is likely due to this being the rainy season, and therefore more cloudy than in the fall and winter. Despite the missing data the main point can still be seen.

In all three maps a large white (sparesly vegetated) area can be seen in the middle. This area is associated with shrubland and crop lands. It is obvious in winter than less crops are planted as this area grows in winter, and shrinks in spring and fall.

1. MODIS Time Series – 263 Columns, 202 Lines, 184 Bands Anthromes – 159 Columns, 121 Lines, 1 Band
 * __Lab 4__**

2. The number of columns and lines for the Anthrome image are less than the number for the MODIS time series. They are different because they are at different resolutions. They show the same area of land, one is just more “zoomed in” than the other, which makes it appear larger. My time series has 23 observations per year (though some are missing quite a bit of data, in some cases all of it) and they are 16 day composites. __K comment: The pixel size is different.__

3. If you were to shrink the Anthrome pixels it would be worse than enlarging the time series pixels because since the time series is floating data, it won’t be really distorted by blowing it up. The Anthrome data is classification data so if you tried to shrink it, it would get very messy because it would go from (for example) a pixel with class four and class five to a pixel with half class four/half class five and it would just be very confusing and weird.


 * # || Class || # pixels || % pixels ||
 * 11 || Urban || 7 || 0.037 ||
 * 12 || Dense settlement || 9 || 0.047 ||
 * 21 || Rice villages || 0 || 0 ||
 * 22 || Irrigated villages || 0 || 0 ||
 * 23 || Cropped pastoral villages || 0 || 0 ||
 * 24 || Pastoral villages || 29 || 0.152 ||
 * 25 || Rainfed villages || 2 || 0.010 ||
 * 26 || Rainfed mosaic villages || 21 || 0.110 ||
 * 31 || Residential irrigated cropland || 6 || 0.031 ||
 * 32 || Residential rainfed mosaic || 475 || 2.485 ||
 * 33 || Populated rainfed cropland || 40 || 0.209 ||
 * 34 || Populated rainfed cropland || 1096 || 5.733 ||
 * 35 || Remote croplands || 127 || 0.664 ||
 * 41 || Residential rangelands || 184 || 0.962 ||
 * 42 || Populated rangelands || 2590 || 13.547 ||
 * 43 || Remote rangelands || 1408 || 7.365 ||
 * 51 || Populated forests || 3258 || 17.042 ||
 * 52 || Remote forests || 8574 || 44.848 ||
 * 61 || Wild forests || 1283 || 6.711 ||
 * 62 || Spare trees || 6 || 0.031 ||
 * 63 || Barren || 0 || 0 ||

Most Dominant Classes 1. Remote Forests 2. Populated Forests 3. Populated Rangelands 4. Remote Rangelands 5. Wild Forests

4. The area covered by the 5 most dominant classes is 89.513%.



5. The NDVI graphs are basically the oposites of the Variation graphs, they look very similar but just upside down. However the peaks and drops seems to be much more prenounced in the variation graphs. The variation graphs for anthromes like Wild Forsests, stay pretty low, until the end where it gets as high as 0.55, indicating high variability. However the variation graph does get higher (more varied) for wild forests year after year. The same is not as true for the other anthromes, however every single anthrome had a huge drop in NDVI and was very unstable in 2007.

6. August/September seems to be the start of the growing season for every anthrome, which makes sense because that is the end of winter and begining of spring time in this area. March is the usual end of the season, which also makes sense since it is Fall there at that time. The start of growing season seems to be the same throughout the years but the end of the growing season starts at the same time each year, but de-greens more steeply/quickly in the later years.

7. I think that the first few years in the Wild Forests graph give a good indictation of how this landscape would look without humans in it. Very stable, without very many peaks and valleys. Since this area isn't really very seasonal you would expect to see not a huge die off of vegetation each year like we are seeing in the later years.

8. I think the Wild Forests graph is the most shocking, you can really see the changes. It is obvious that even in the hieght of the growing seasons NDVI has decreased from 0.85ish to 0.80ish. Also, the valleys in the data are huge compared to the first 3 years. Doesn't seem like the "wild" forests are so wild anymore. 2007 has the largest dip in NDVI for all of the anthromes. Hopefully I can find some data to follow up on that and see if it's the start of a trend or an anomaly. It certainly looks though that the valleys have been getting lower and lower every year. The NDVI values in the valleys are not numbers you would expect to see in a rainforest ecosystem.

9. I don't really have much else to say other than point out again that there should not be this much variablity in a rainforest ecosystem. And the fact that these anthromes cover 85% of a 10 degree x 10 degree swath of land means that this change isn't just a small centralized problem, perhaps around Rondonia, which is what I previously thought. It is instead affecting the entire area!


 * __Lab 5 - Due March 9, 2011__**

For this lab I decided to look at the EVI values of my tile in the winter time. From Lab 4 I noticed that there was the most variability in the winter NDVI values, compared to the summer ones. EVI is better for areas of high vegetation because it saturates less, which is why I chose to use it, since I am working in a rainforest ecosystem. I chose three 8-day BRDF reflectance composite MODIS images per year (2000 through 2010), one in May, June and July. I then found the EVI values for these images, and then stacked each three images per year together. I then found an average EVI for each winter season and graphed it in excel. I also created mean images for 2000 and 2010. The graph and two images can be seen below.

As you can see, the images really do not look that different. However the graph clearly shows that EVI values have been dropping each year. The values don't change a huge amount, but for a typically stable rainforest ecosystem, this is quite significant change. I thought this change may be due to deforestation for agriculture or urbanization so I download MODIS classification data to investigate further. I downloaded one image from 2001 and one from 2009 (2010 was not available at the time) and color coded them using the Land Cover Type 2 Classification from the University of Maryland. The images can be seen below.



After I classified the images I ran class statistics on each and found that Evergreen Broadleaf Forest pixels had decreases from 25% to 23%. Open shrubland had decreased from 16% to 13%. Savannas (both woody and non-woody) had increases from 36% to 43%. Croplands had increased slightly from 4.7% to 5.8%. Urban had remained the same with the exact same number of pixels for the different years. These findings I think show that urban and cropland expansion has not taken place, which is what I previously thought was the main reason for the drop in EVI. It appears however, that with the forest decreasing and the woody savannah increasing so dramatically, that deforestation is occuring, not because they are clearing land for development, but for timber. Deforestation here has been occuring for years but I have not been able to find why it is still occuring in the 2000s. All the studies I found focus on the 1970s through 1990s. Hopefully through the use of landsat imagery I will be able to see more clearly what is occuring so that I can research why a little more.