14 Years of Land Cover Change in Baghdad, Iraq

This was one of four projects I completed in my Digital Image Processing class my senior year of undergrad. I thoroughly enjoyed this project so I just wanted to share it on my blog…

___________________________

Abstract

This study observes the land cover changes in and surrounding Baghdad, Iraq from the year 2000 to 2013. The observed changes are expected to be due to war and conflict. These two years, 2000 and 2013, are before and after the U.S. bombed and invaded Baghdad post 9/11. Changes in urban density and vegetation can be observed.

This study was accomplished by using Landsat 5 TMR L0RP and Landsat 8 OLI/TIRS to acquire remotely sensed images for 2000 and 2013. Landsat 5 is the source for the 2000 images, and Landsat 8 is the source for the 2013 images. The path/row to be analyzed is 139/037. The Landsat scenes are 180km X 180km and 30 meters in resolution.

The objective of this study is to use remote sensing software (ENVI) to reveal these scenery changes in Baghdad by creating thematic maps of the study area. Three images were obtained for each year to create a yearly average for the region. Four thematic maps were created in the conclusion of this study. Two maps for each year, one using a maximum likelihood classification method and one using a minimum distance classification method.

Statistics are discussed and reflected upon.

 

Study Area

The area I have chosen for this study encompasses the city of Baghdad, Iraq. Baghdad is the capital of Iraq and is located on the Tigris River just 40 kilometers to the west of the Euphrates River. Baghdad’s location is depicted in figures 1a and 1b below for spatial context. This is roughly a 40 Kilometer Squared area that is being analyzed, and the time period for this analysis will span 14 years, 2000 through 2013. Baghdad, along with this time frame, was chosen as the study area due to the war and conflict the country experienced during this time period. The first image to be analyzed was captured in 2000. This is three years before the U.S. bombed and invaded Baghdad [Burnham, Gilbert, et al. “Mortality after the 2003 invasion of Iraq: a cross-sectional cluster sample survey.” The Lancet 368.9545 (2006): 1421-1428]. The second image to be analyzed was captured in 2013. This was 2 years after U.S. troops were completely withdrawn from the country. [Lindsay, James M. “George W. Bush, Barack Obama and the future of US global leadership.” International Affairs 87.4 (2011): 765-779]. It may be possible to observe war related damage using these two images. The city of Baghdad is located in a dry Middle Eastern desert environment where the average annual temperature is 22.6 degrees Celsius, and the average annual rainfall for the area is 157 millimeters [http://en.climate-data.org/location/86/]. Baghdad sits 34 meters above sea level and has a flat alluvial plain landscape type [http://www.britannica.com/place/Baghdad]. The estimated population of Baghdad in 1987 was 3,841,268, and this number grew to 6,150,000 by 2011 [http://www.citypopulation.de/Iraq.html]. This is a population increase of roughly 2,308,732. These years were chosen because they were the closest years with available population data for this study.

1a

Figure 1a

1b

Figure 1b

Data

The Landsat scenes acquired for this study came from Landsat 5 Thematic Mapper (TM), and Landsat 8 Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS). These scenes were downloaded from earthexplorer.gov. Three images were chosen from both years, 2000 and 2013. The three images from 2000 were obtained from Landsat 7, and the three images from 2013 were obtained from Landsat 8. The scenes that were selected for this study were chosen for their clarity. It was important for each image to be extremely low in cloud cover, and these were not difficult to locate due to the desert climate of the Middle East. Within each year, three months were chosen. This is one image for each month chosen, and each of these months were from different seasons. This was done to encompass the seasonal changes of the region.

The Landsat scenes that were downloaded for this study are 180km X 180 km and 30 Meters in resolution. The path/row of these Landsat scenes are 139/037, the projection is UTM zone 38 north, and the datum is WGS84. Within these Landsat scenes, three relatively large bodies of water along with their tributaries can be observed. Much of the image consists of pure desert, especially to the west/southwest. The capital city of Baghdad is located on the east/northeastern part of the image, and is observed as a gray color if the image being viewed in true color bands. Something interesting to note at first glance is the significant buffers of vegetation that surround the rivers, and the urban areas that are completely surrounded by vegetation.

Path: 139

Row: 037

Projection: UTM zone 38 north

Datum: WGS84

Resolution: 30 Meters

Day 118 Year 2013 Day 198 Year 2013 Day 358 Year 2013
01 02 03

 

YEAR SENSOR DATE PATH/ROW
2000 Landsat 5 TMR L0RP 2000/04/01 139/037
2000 Landsat 5 TMR L0RP 2000/07/29 139/037
2000 Landsat 5 TMR L0RP 2000/10/01 139/037
2013 Landsat 8 OLI/TIRS 2013/04/28 139/037
2013 Landsat 8 OLI/TIRS 2013/07/17 139/037
2013 Landsat 8 OLI/TIRS 2013/12/24 139/037

 

BD1

Above: NIR Image: Baghdad, Iraq in 2000 containing 8,200,530 pixels.

1,427 columns and 1,025 rows

 

BD2

Above: NIR Image: Baghdad, Iraq in 2013 containing 8,200,530 pixels.

1,427 columns and 1,025 rows

 

Initial Observations of NIR images

In observing and comparing these two NIR images, it is easy to notice that the purple shade of the urban area is darker in the 2000 image than it is in the 2013 image. Vegetation is far more colorful in the 2000 image than it is in the 2013 image. I expect this to be due to war and conflict. Rivers did not change color from one year to the other, but the purple shades that directly surround the rivers are significantly lighter in the 2013 scene compared to the 2000 scene. Most greens from the 2000 image are yellow in the 2013 image. This is very noticeable near the airport on the southwest side of the city.

 

Results

Supervised Classification Training Data:

Classes: Urban, Light Urban, Desert, Cropland, Forest, Water, and Clouds (clouds only in 2000)

2000         2013

2000                                                                            2013

 

 

Methods

The two classification methods used to generate these thematic maps were maximum likelihood and minimum distance. Both methods were used for each year resulting in four total maps.

Maximum Likelihood vs Minimum Distance

  • The maximum likelihood method calculates the likelihood or probability for each pixel to belong to a specific class. This classification method assigns each pixel to a class based on highest probability. All pixels will be assigned to a class unless a probability threshold is declared [http://www.exelisvis.com/docs/MaximumLikelihood.html]. In this project no thresholds were specified. All pixels were assigned to a class.
  • The minimum distance method classifies all pixels to the nearest class by calculating the Euclidean distance to the mean vector. Unless a distance threshold is declared, all pixels will be assigned to a class [http://www.exelisvis.com/docs/MinimumDistance.html]. In this project no thresholds were specified. All pixels were assigned to a class.

 

2000-1

 2000-2

2013-1

2013-2

Maximum Likelihood Statistics (2000):

CLASS_NAME         AREA                         PERCENT

Unclassified                0.00                             0.000000

Desert                          1624701600.00           20.071423

Urban                          331227900.00             4.091961

Water                          69236100.00               0.855337

Cropland                     3301883100.00           40.791178

Forest                          119467800.00             1.475895

Light Urban                2112327000.00           26.095505

Clouds                         535757400.00             6.618701

Minimum Distance Statistics (2000):

CLASS_NAME         AREA                         PERCENT

Unclassified                0.00                             0.000000

Desert                          1111798800.00           13.735066

Urban                          1088132400.00           13.442694

Water                          416358000.00             5.143651

Cropland                     1348468200.00           16.658860

Forest                          542764800.00             6.705269

Light Urban                1776638700.00           21.948441

Clouds                         1810440000.00           22.366019

Maximum Likelihood Statistics (2013):

CLASS_NAME         AREA                         PERCENT

Unclassified                0.00                             0.000000

Urban                          382184100.00             5.178312

Light Urban                1584391500.00           21.467332

Deserts                                    1661079600.00           22.506399

Cropland                     3531463200.00           47.848712

Forest                          180773100.00             2.449342

Water                          40585500.00               0.549903

Minimum Distance Statistics (2013):

CLASS_NAME         AREA                         PERCENT

Unclassified                0.00                             0.000000

Urban                          424042200.00             5.745458

Light Urban                1358619300.00           18.408286

Deserts                                    2041870500.00           27.665834

Cropland                     2499678000.00           33.868787

Forest                          575322300.00             7.795191

Water                          480944700.00             6.516445

 

Statistics

No pixels went unclassified for each year. As documented earlier in this report, this is because no probability or distance thresholds were declared.

When comparing the statistics for both methods (maximum likelihood and minimum distance) for the 2000 image, there was not a dramatic difference overall, but a few to note are the differences between cropland and urban. Under the maximum likelihood classification method, cropland covered 40.791178% of the image, and urban covered 4.091961% of the image. This differs greatly from the minimum distance classification method. Under the minimum distance classification method, cropland covered 16.658860% of the image, and urban covered 13.442694% of the image. For cropland, this is a percentage difference of 24.132318. For urban, this is a percentage difference of 9.350733.

When comparing the statistics for both methods (maximum likelihood and minimum distance) for the 2013 image, there was little difference. The differences here are even smaller than they were for the 2000 image. The largest difference for the 2013 methods were within the cropland class. The maximum likelihood cropland class covered 47.848712% of the image, and the minimum distance cropland class covered 33.868787% of the image. This is a percentage difference of 13.979925.

 

 

Maximum Likelihood: Year 2000

Overall Accuracy = (169347/174481) 97.0576%

Kappa Coefficient = 0.9619

 

Ground Truth (Pixels)

Class            Desert         Urban         Water         Cropland      Forest 

Desert        52967            0            0            0            0

Urban            0        50756            0            0            0

Water            0           18         4693            0            0

Cropland            0           38            0        24135            0

Forest            0          301            0            0         4917

Light Urban           19         1019            3          169           86

Clouds            4            0            0            0            0

Total        52990        52132         4696        24304         5003

 

 

Ground Truth (Pixels)

Class         Light Urban     Clouds      Total 

Desert            0            0        52967

Urban         2984            0        53740

Water            0            0         4711

Cropland          427            0        24600

Forest           58            0         5276

Light Urban         9036            8        10340

Clouds            0        22843        22847

Total        12505        22851       174481

 

 

Ground Truth (Percent)

Class          Desert       Urban         Water         Cropland      Forest

Desert        99.96         0.00         0.00         0.00         0.00

Urban         0.00        97.36         0.00         0.00         0.00

Water         0.00         0.03        99.94         0.00         0.00

Cropland         0.00         0.07         0.00        99.30         0.00

Forest         0.00         0.58         0.00         0.00        98.28

Light Urban         0.04         1.95         0.06         0.70         1.72

Clouds         0.01         0.00         0.00         0.00         0.00

Total       100.00       100.00       100.00       100.00       100.00

 

 

Ground Truth (Percent)

Class         Light Urban   Clouds        Total 

Desert         0.00         0.00        30.36

Urban        23.86         0.00        30.80

Water         0.00         0.00         2.70

Cropland         3.41         0.00        14.10

Forest         0.46         0.00         3.02

Light Urban        72.26         0.04         5.93

Clouds         0.00        99.96        13.09

Total       100.00       100.00       100.00

 

 

 

Class   Commission     Omission          Commission            Omission 

(Percent)    (Percent)            (Pixels)            (Pixels)

Desert         0.00         0.04             0/52967            23/52990

Urban         5.55         2.64          2984/53740          1376/52132

Water         0.38         0.06             18/4711              3/4696

Cropland         1.89         0.70           465/24600           169/24304

Forest         6.80         1.72            359/5276             86/5003

Light Urban        12.61        27.74          1304/10340          3469/12505

Clouds         0.02         0.04             4/22847             8/22851

 

 

Class   Prod. Acc.    User Acc.          Prod. Acc.           User Acc. 

(Percent)    (Percent)            (Pixels)            (Pixels)

Desert        99.96       100.00         52967/52990         52967/52967

Urban        97.36        94.45         50756/52132         50756/53740

Water        99.94        99.62           4693/4696           4693/4711

Cropland        99.30        98.11         24135/24304         24135/24600

Forest        98.28        93.20           4917/5003           4917/5276

Light Urban        72.26        87.39          9036/12505          9036/10340

Clouds        99.96        99.98         22843/22851         22843/22847

————————————————————————————–

Minimum Distance: Year 2000

Overall Accuracy = (145464/174481) 83.3695%

Kappa Coefficient = 0.7894

 

Ground Truth (Pixels)

    Class            Desert        Urban         Water         Cropland      Forest 

Desert        48173            0            0            0            0

Urban            0        40418            0         1856          122

Water            0           21         4659            0            0

Cropland            0            0            0        17461            0

Forest            0          522           37         1221         4881

Light Urban            0        11097            0         3532            0

Clouds         4817           74            0          234            0

Total        52990        52132         4696        24304         5003

 

 

Ground Truth (Pixels)

Class            Light Urban   Clouds        Total 

Desert             0         1233        49406

Urban         3686            0        46082

Water            0            0         4680

Cropland          107            0        17568

Forest          369            0         7030

Light Urban         8254            0        22883

Clouds           89        21618        26832

Total        12505        22851       174481

 

 

Ground Truth (Percent)

Class            Desert        Urban         Water         Cropland      Forest

Desert         90.91         0.00         0.00         0.00         0.00

Urban         0.00        77.53         0.00         7.64         2.44

Water         0.00         0.04        99.21         0.00         0.00

Cropland         0.00         0.00         0.00        71.84         0.00

Forest         0.00         1.00         0.79         5.02        97.56

Light Urban         0.00        21.29         0.00        14.53         0.00

Clouds         9.09         0.14         0.00         0.96         0.00

Total       100.00       100.00       100.00       100.00       100.00

 

 

Ground Truth (Percent)

Class            Light Urban   Clouds        Total 

Desert         0.00         5.40        28.32

Urban        29.48         0.00        26.41

Water         0.00         0.00         2.68

Cropland         0.86         0.00        10.07

Forest         2.95         0.00         4.03

Light Urban        66.01         0.00        13.11

Clouds         0.71        94.60        15.38

Total       100.00       100.00       100.00

 

 

 

Class   Commission     Omission          Commission            Omission 

(Percent)    (Percent)            (Pixels)            (Pixels)

Desert          2.50         9.09          1233/49406          4817/52990

Urban        12.29        22.47          5664/46082         11714/52132

Water         0.45         0.79             21/4680             37/4696

Cropland         0.61        28.16           107/17568          6843/24304

Forest        30.57         2.44           2149/7030            122/5003

Light Urban        63.93        33.99         14629/22883          4251/12505

Clouds        19.43         5.40          5214/26832          1233/22851

 

 

Class   Prod. Acc.    User Acc.          Prod. Acc.           User Acc. 

(Percent)    (Percent)            (Pixels)            (Pixels)

Desert         90.91        97.50         48173/52990         48173/49406

Urban        77.53        87.71         40418/52132         40418/46082

Water        99.21        99.55           4659/4696           4659/4680

Cropland        71.84        99.39         17461/24304         17461/17568

Forest        97.56        69.43           4881/5003           4881/7030

Light Urban        66.01        36.07          8254/12505          8254/22883

Clouds        94.60        80.57         21618/22851         21618/26832

————————————————————————————–

Maximum Likelihood: Year 2013

Overall Accuracy = (316842/328090) 96.5717%

Kappa Coefficient = 0.9426

 

Ground Truth (Pixels)

Class            Urban         Light Urban   Deserts       Cropland      Forest

Urban        42604         3624            0          120          133

Light Urban         1309         6113         1188         3664          125

Deserts            0            0       189242            0            0

Cropland           24          244          609        68114           16

Forest            0            4            0          184         8081

Water            0            0            0            0            0

Total        43937         9985       191039        72082         8355

 

 

Ground Truth (Pixels)

Class               Water      Total 

Urban            0        46481

Light Urban            0        12399

Deserts            0       189242

Cropland            3        69010

Forest            1         8270

Water         2688         2688

Total         2692       328090

 

 

Ground Truth (Percent)

Class            Urban Light   Urban         Deserts       Cropland      Forest

Urban        96.97        36.29         0.00         0.17         1.59

Light Urban         2.98        61.22         0.62         5.08         1.50

Deserts         0.00         0.00        99.06         0.00         0.00

Cropland         0.05         2.44         0.32        94.50         0.19

Forest         0.00         0.04         0.00         0.26        96.72

Water         0.00         0.00         0.00         0.00         0.00

Total       100.00       100.00       100.00       100.00       100.00

 

 

 

Ground Truth (Percent)

Class         Water       Total 

Urban         0.00        14.17

Light Urban         0.00         3.78

Deserts         0.00        57.68

Cropland         0.11        21.03

Forest         0.04         2.52

Water        99.85         0.82

Total       100.00       100.00

 

 

 

Class   Commission     Omission          Commission            Omission 

(Percent)    (Percent)            (Pixels)            (Pixels)

Urban         8.34         3.03          3877/46481          1333/43937

Light Urban        50.70        38.78          6286/12399           3872/9985

Deserts         0.00         0.94            0/189242         1797/191039

Cropland         1.30         5.50           896/69010          3968/72082

Forest         2.29         3.28            189/8270            274/8355

Water         0.00         0.15              0/2688              4/2692

 

 

Class   Prod. Acc.    User Acc.          Prod. Acc.           User Acc. 

(Percent)    (Percent)            (Pixels)            (Pixels)

Urban        96.97        91.66         42604/43937         42604/46481

Light Urban        61.22        49.30           6113/9985          6113/12399

Deserts        99.06       100.00       189242/191039       189242/189242

Cropland        94.50        98.70         68114/72082         68114/69010

Forest        96.72        97.71           8081/8355           8081/8270

Water        99.85       100.00           2688/2692           2688/2688

————————————————————————————–

Minimum Distance: Year 2013

Overall Accuracy = (317523/328090) 96.7792%

Kappa Coefficient = 0.9460

 

Ground Truth (Pixels)

    Class            Urban Light   Urban         Deserts       Cropland      Forest

Urban        42835         3691            0          113          132

Light Urban         1091         6101         1117         3355          104

Deserts            0            0       189337            0            0

Cropland           11          190          585        68454           11

Forest            0            3            0          160         8108

Water            0            0            0            0            0

Total        43937         9985       191039        72082         8355

 

 

Ground Truth (Pixels)

Class              Water        Total 

Urban            0        46771

Light Urban            0        11768

Deserts            0       189337

Cropland            3        69254

Forest            1         8272

Water         2688         2688

Total         2692       328090

 

 

 

 

Ground Truth (Percent)

Class            Urban         Light Urban   Deserts       Cropland      Forest

Urban        97.49        36.97         0.00         0.16         1.58

Light Urban         2.48        61.10         0.58         4.65         1.24

Deserts         0.00         0.00        99.11         0.00         0.00

Cropland         0.03         1.90         0.31        94.97         0.13

Forest         0.00         0.03         0.00         0.22        97.04

Water         0.00         0.00         0.00         0.00         0.00

Total       100.00       100.00       100.00       100.00       100.00

 

 

Ground Truth (Percent)

Class              Water       Total 

Urban         0.00        14.26

Light Urban         0.00         3.59

Deserts         0.00        57.71

Cropland         0.11        21.11

Forest         0.04         2.52

Water        99.85         0.82

Total       100.00       100.00

 

 

 

 Class   Commission     Omission          Commission            Omission 

(Percent)    (Percent)            (Pixels)            (Pixels)

Urban         8.42         2.51          3936/46771          1102/43937

Light Urban        48.16        38.90          5667/11768           3884/9985

Deserts         0.00         0.89            0/189337         1702/191039

Cropland         1.16         5.03           800/69254          3628/72082

Forest         1.98         2.96            164/8272            247/8355

Water         0.00         0.15              0/2688              4/2692

 

 

Class   Prod. Acc.    User Acc.          Prod. Acc.           User Acc. 

(Percent)    (Percent)            (Pixels)            (Pixels)

Urban        97.49        91.58         42835/43937         42835/46771

Light Urban        61.10        51.84           6101/9985          6101/11768

Deserts        99.11       100.00       189337/191039       189337/189337

Cropland        94.97        98.84         68454/72082         68454/69254

Forest        97.04        98.02           8108/8355           8108/8272

Water        99.85       100.00           2688/2692           2688/2688

———————————————————————-

Histogram

Percent Urban: Year 2000 Percent Urban: Year 2013
30.187466% 26.645644%

 

Total Urban Area: Year 2000 Total Urban Area: Year 2013
10,571,970 2,609,733

 

According to the processes and methods applied in this study, Baghdad, Iraq has decreased in overall urban area within the 14 years being analyzed. Urban area was broken up into two different classes in this study due to the dramatic difference in NIR pixel color between denser urban areas and less dense urban areas.  As a result of this multi class urban area, the two classes were simply added together when analyzing overall urban statistics. The percentage of overall urban area decreased 3.541822% within the 14 years in Baghdad. This is a gradual decrease that may be cause by war, but this is not a fact. It may also be due to the historic nature of how Middle Eastern cities change. The Arab world has historically urbanized or deurbanized at slow paces, especially during the 19th and early 20th centuries [Ibrahim, Saad EM. “Over-urbanization and under-urbanism: The case of the Arab world.” International Journal of Middle East Studies 6.01 (1975): 29-45].

To gain a visual of the urban area lost, the two images below can be compared.

z1.jpg                               z2.jpg

Thematic map: Baghdad, Iraq 2000                         Thematic Map: Baghdad, Iraq 2013

 

There is a significant amount of urban area (gray colors) that is missing in the upper right portion of the 2013 image as compared to the 2000 image. Making this comparison supplies a visual to the urban area loss, but this also supplies a visual to a vegetation increase in 2013 (green colors).

 

 

Works Cited

Burnham, Gilbert, et al. “Mortality after the 2003 invasion of Iraq: a cross-sectional cluster sample survey.” The Lancet 368.9545 (2006): 1421-1428

Lindsay, James M. “George W. Bush, Barack Obama and the future of US global leadership.” International Affairs 87.4 (2011): 765-779

http://en.climate-data.org/location/86/

http://www.britannica.com/place/Baghdad

http://www.citypopulation.de/Iraq.html

Harris Geospatial Solutions: http://www.exelisvis.com/docs/MaximumLikelihood.html

Harris Geospatial Solutions: http://www.exelisvis.com/docs/MinimumDistance.html

Ibrahim, Saad EM. “Over-urbanization and under-urbanism: The case of the Arab world.” International Journal of Middle East Studies 6.01 (1975): 29-45.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s