I. Introduction
In response to a request by Amnesty International (AI-USA), the Geospatial Technologies and Human Rights Project of the American Association for the Advancement of Science (AAAS) undertook an analysis of satellite imagery to examine incidents of reported violence in Mogadishu, Somalia, between February 2007 and January 2010 (Figure One). For this investigation, three areas of interest were identified: the KM4 junction, the presidential palace (Villa Somalia), and the Bakara market (Figure Two). Each of these locations was used as a central point for areas of analysis, each defined by a circle one kilometer in radius from that central point. Although the region surrounding the KM4 junction was discrete, there was a slight degree of overlap between the two other areas.
Figure One: Overview of Somalia
Figure Two: Overview of Mogadishu
The city of Mogadishu, Somalia, including the study area. Image © 2010 DigitalGlobe.
II. Methods and Technologies
Dense urban areas such as Mogadishu present special problems for damage assessment. In particular, the visual complexity of the terrain makes small-scale damage, such as would result from mortar fire, extremely difficult to detect. Except in ideal cases, this type of destruction must therefore be evaluated through its secondary effects, such as the removal or replacement of roofs and small structures. These observations can be problematic, however, as more mundane causes such as home renovation or nomadism can also produce visually similar results. As such, any identification based on these secondary effects is termed “potential damage.”
Within the regions of interest, the following feature classes were mapped:
- potential damage or destruction
- increases and decreases in the number of temporary structures (possibly shelters for Internally Displaced Persons)
- new construction or rehabilitation of existing damage
Table One: Imagery Information
Image Date | Sensor | Resolution | Image Bands |
02/05/2007 | QuickBird | 0.6 | 4-band multispectral |
12/16/2007 | QuickBird | 0.6 | 4-band multispectral |
01/21/2008 | QuickBird | 0.6 | 4-band multispectral |
06/24/2009 | GeoEye-1 | 0.5 | 4-band multispectral |
07/09/2009 | QuickBird | 0.6 | 4-band multispectral |
09/26/2009 | WorldView-1 | 0.5 | 1-band multispectral |
11/17/2009 | QuickBird | 0.6 | 4-band multispectral |
01/28/2010 | WorldView-2 | 0.5 | 4-band multispectral |
In total, eight images were utilized, and the parameters for each are listed in Table 1 below. For each time interval, the extent of the destruction was quantified, and then normalized for the differing durations between images. The resulting damage rate has units of incidents per day, and is expressed as a percentage for clarity (Table Two).
Table Two: Image Analysis Results
Start Date | End Date | Duration (days) | Damage Rate (incidents * day ^-1)*100 | Notes |
02/05/2007 | 12/16/2007 | 314 | 1.9 | 2.9 including ambiguous |
12/16/2007 | 01/21/2008 | 36 | 16.7 | 25 including ambiguous |
01/21/2008 | 07/09/2009 | 535 | 4.2 | |
07/09/2009 | 01/28/2010 | 203 | 6.9 |
III. Results
Little potential damage is visible within the area of interest between February 5 and December 16, 2007 (Figure Two), and no clear trends are discernible in the geographic distribution of the damage. Most potential damage appears as structures with roofs removed, such as could occur following structural compromise due to shelling. However, the lack of damage documented could be due to the small number of analysis sites. The estimated damage rate during this 314-day period is calculated at 1.9 percent, rising to 2.9 percent if ambiguous detections are included (Table Two).
Six instances of potential damage were counted between December 16, 2007 and January 21, 2008, excluding ambiguous detections. Five of these six occurred within one kilometer of the western edge of the Bakara market, though little clustering was observed. As before, most damage is apparent through damage to or removal of roofs, with only one incident where an entire building has disappeared. Estimated damage flux during this 36 day period is 16.7 percent, rising to 25 percent if ambiguous detections are included. This relatively high damage rate is consistent with Amnesty’s reporting of “heavy shelling” in the Bakara market area from January 17-18, 2008.
Between January 21, 2008 and July 9, 2009, twenty-three instances of potential damage were documented. In the Bakara Market subregion, seven damaged buildings were counted, all of which are located to the north and/or west of the market. In the area surrounding the Villa Somalia, eight damaged structures were counted, primarily in the north and west of the subregion. The final grouping of eight damaged buildings occurs in the area northeast of the KM4 Junction, and is the most localized, with all instances of damage occurring within 700 meters of one another. Over the entire study area, the damage rate during this period is calculated at 4.2 percent. As before, most destruction took the form of damaged or destroyed roofs.
Figure Three: Damage observed in study area, 2007-2010
In the top image, brown points indicate damage present by December 16, 2007, orange indicates damage by January 21, 2008, and red indicates damage by July 9, 2009. In the bottom image, yellow points indicate damage that occurred between July 9, 2009 and January 28, 2010. Image © 2010 DigitalGlobe.
Between July 9, 2009 and January 28, 2010, most damage is clustered in the area northeast of the KM4 junction, with only occasional incidents detected elsewhere. In total, 14 incidents were detected within the study area, indicative of a damage rate of 6.9 percent. In general, the destruction followed the same pattern seen before of damaged or absent roofs.
Figure Four: Overview of damage outside of study area, 2007-2010
The red circles indicate the destruction of housing in northern Mogadishu. Image © 2010 DigitalGlobe.
Additional incidents also occurred near this time which, while beyond the defined study area, were visible in the imagery provided to AAAS SHR. These included destruction of numerous homes in the areas in northern Mogadishu between January 21, 2008 and June 24, 2009 (Figure Five), and the shelling of the Martini Hospital, as was reported by Amnesty on September 11, 2009, and confirmed by AAAS SHR imagery, which show fresh patterns in the nearby terrain consistent with bursting artillery shells (Figure Five).
Figure Five: Damage observed in study area, 2007-2010
On January 21, 2008 (top), the neighborhood is intact. By June 24, 2009 (bottom), large numbers of dwellings have been destroyed. These image dates encompass the period of time between the withdrawal of Ethiopian troops and the beginning of insurgent activities in the area. Images © 2010 DigitalGlobe, GeoEye.
During the study period, many sites where temporary structures were present experienced a decrease in number, and many cases were observed with all structures removed (Figure Six). Several of these sites were located in relatively close proximity to clusters of damage to permanent structures, which may indicate that they share a common cause, such as shelling. Despite the instability that continues to plague Mogadishu, this analysis also indicates that new construction, as well as the rehabilitation of damage that predates the study period, is a major phenomenon within the city. Particularly prevalent in the area surrounding the presidential palace, many formerly empty ruins have now been covered with new roofs, at a rate that appears to substantially outpace that of any potential damage (Figure Six), which corresponds to the documentation of Ethiopian troop withdrawal ending in January 2009 by AI and major news sources. Additionally, areas of new construction and renovation are generally distinct from the damage clusters that indicate shelling, which could indicate wariness on the part of residents to return and rebuild in areas with a higher risk of further destruction.
Figure Six: Evidence of shelling near Martini Hospital, Mogadishu
July 9, 2009, no damage is visible. By September 26, however, numerous clustered marks with rayed striae, suggestive of mortar craters, are visible. Images © 2010 DigitalGlobe.
AAAS also found evidence of population movements in Mogadishu (Figure Seven). In early 2007 (images at left), numerous temporary structures can be seen crowding the interior of a walled compound, and the empty shells of buildings are present near the presidential palace. By 2010 (images at right) all temporary structures have disappeared from the compound, and many gutted buildings show signs of rehabilitation. Whether these phenomena are related remains unclear, but they underscore the complexities of population movements in and around Mogadishu.
Figure Seven: Population Movements in and around Mogadishu
IV. Conclusion
The results of this analysis indicate that sites of potential damage exist in all three regions of interest. In many cases, these sites are clustered in close proximity to one-another, a pattern that would not be expected in the case of individual building renovations, suggestive that shelling may indeed be a more likely explanation. When observed over the whole study interval, three major clusters occur northeast of the KM4 junction, and another is present in a nearby area that is southwest of the Bakara market. Other areas of potential destruction are scattered throughout the city; which may be false positives. While source reports of “heavy shelling” were sometimes associated with a significant increase in the overall damage rate visible in the imagery, at other times no clear change was apparent. Part of the explanation for this may lie in the ambiguity associated with the terminology used. Whether “heavy” shelling refers to the intensity of bombardment, the caliber of the weapons, or both, remains unclear. Knowledge of such details could have a substantial impact on AAAS’s ability to effectively verify on-the-ground reporting through high-resolution satellite imagery analysis. At present, while the observed increase in damage rate can be used to corroborate source reporting, the absence of such damage increase is merely ambiguous given the previously discussed problems of damage assessment in highly fluctuating urban areas. For more imagery of damaged observed in this assessment, refer to Figures A, B, and C in the Appendix.
V. Appendix: Imagery of Additional Damage
A. Damage to a Structure south of KM4 Junction
In early 2007 (top), the roof of this large structure south of KM4 junction appears damaged, but largely intact. By June of 2009, however, large portions have been completely destroyed. Images © 2010 DigitalGlobe, GeoEye.
B. Damage to Structures near the Presidential Palace
In 2008, all houses in this neighborhood north of the presidential palace appear intact. By June 2009, however, two of them (indicated by arrows) appear to have lost their roofs. Images © 2010 GeoEye.
C. Damage to Structures north of KM4 Junction
In 2008, this neighborhood north of the KM4 junction appears undamaged. By June of 2009, multiple structures appear to have sustained considerable damage (indicated by arrows), including one house where two massive holes have appeared in the roof, and another with no roof at all. Images © 2010 GeoEye.