Geovisualization for Mapping Human Rights Incidents in Northwestern Pakistan

Executive Summary

AAAS worked with Amnesty International USA’s Science for Human Rights program (AI) to create a spatial database of human right incidents occurring as a result of the ongoing conflict in the northwestern Pakistani provinces of the Federally Administered Tribal Areas (FATA) and Northwest Frontier Province (NWFP), seen in Figure One. This spatial database covers the years 2005-2009 at present, although it will continue to be updated in the future, which is explained in latter sections of this report. This report documents the process of creating a spatial database and the creation of an interactive web mapping portal, which has become the centerpiece of Amnesty International’s Eyes on Pakistan website.

Figure One: Study Region

overview

I. Introduction
II. Methods and Technologies
III. Results
IV. Conclusion
V. Further Resources

I. Introduction

In 2008, AAAS and AI began a major project under joint funding from the Oak Foundation to examine the human rights aspects of the conflict occurring in northwestern Pakistan. Specifically, researchers at Amnesty’s International Secretariat were creating a database of human rights incidents from 19 different media sources which were being collected and coded by researchers. The data is organized by province, district, and town, and drawn primarily from international news sources, including the Daily Times and BBC, supplemented by other sources, including the United States Department of State. This database covers January 1, 2005 to December 31, 2009. AI approached AAAS in order to harness the scientific methodologies and research methods of the Association to create a new type of research project, a geographic product for mapping and displaying the information in the database of human rights incidents. The large, comprehensive nature of the database made it inaccessible to the public through AI’s traditional narrative reporting format AAAS saw this as an opportunity to explore geovisualization techniques that better communicate complex human rights violations to the public. The emerging importance of geovisualization is exemplified with projects such as the US Holocaust Memorial Museum’s “Crisis in Darfur” layer on Google Earth, and the Amnesty International USA “Eyes on Darfur” web site, the latter done with significant AAAS satellite imagery input. The spatial perspective brought to the issue through visualization will help make patterns more discernable than would be possible from an examination of lists of incident reports.

II. Methods and Technologies

In a departure from typical satellite imagery analysis, this project has relied on mapping and visualization of data in order to highlight human rights incidents. The database described in the previous section was coded and geocoded using a fuzzy matching search to create spatial databases to use in mapping and data exploration. The Pakistan Fuzzy Matcher determines the locations from media reports within NWFP and FATA. This web interface allowed researchers to input the location mentioned in media reports and to locate its precise coordinates (latitude and longitude), the District/Agency in which it is located, as well as the province. This information is drawn from the publically available US Government National Geospatial Intelligence Agency GEOnet Names Server. After determining the geographic location of the media reports, the database was coded for relevant facts in each report, and contained the following variables:

  • Number of incidents at the village, district/agency, and provincial scales
  • Number of incidents per year
  • Number killed and/or wounded in each incident
  • Agents of the incident: the group or source responsible for the incident in a given observation
  • Attack methods utilized: the tactic or method employed by agents in a given observation

Agent was an important category to include in the database so that incidents could be looked at for trends of particular groups acting in certain locations within the two provinces. Agents were parsed into 14 categories (Table One), ranging from air strikes, to insurgency, to extrajudicial executions.

Table One: Agents

Code Agents
1 Pak military air strike
2 Pak ground military offensive/clash army v Taleban
3 Arrests – Pak army/Security Forces/arrest of militants/ Taleban
4 Collective punishment – Pak political administration/army/Security Forces – (burning of houses, shops, arrests under FCR – of tribes, villagers seen as supporting or harboring Taleban/insurgents)
5 Clash Taleban/militants v lashkars/levies/khassadars/ armed tribesmen (local militia fighters)
6 Inter-armed groups/rival militant groups
7 Insurgency/ attacks on civilians (civilian objects/subjects) – schools, hospitals, shops, markets, mosques, NGOs, “spies”, Gov workers, Tribal elders, Nazims; Taleban unlawful punishment of criminals etc
8 Insurgency – attacks on military (military objects) – police, paramilitary, army, military convoys, Frontier Constabulary, Frontier Corps, NATO supply convoys
9 Inter-tribal/sectarian (Sunni/Shia)
10 Extrajudicial executions (unlawful and targeted killings by the army/insurgents of respectively Taleban, Taleban supporters; or civilians for reprisals)
11 US drone attacks
12 Unknown (when the perpetrator or the target is unclear)
13 Insurgency/destruction of infrastructure/bridges
14 Cross border/NATO-US (mortar shelling, etc from Afghanistan side)

The attack method category (Table Two) was a distillation of the various types of weaponry or attacks used in the incidents reported within the database. Again, it was important to derive spatial trends from the database of media reports, and the ability to see which types of attacks were prevalent in which areas were seen as a crucial set of facts for later data analysis and access by users of the website.

Table Two: Attack Methods

Code Agents
1 Air strike (gunship helicopters, etc)
2 Drones (unmanned aircrafts)
3 Ground base attacks (artillery, mortars, rockets, armed attack, gunfire)
4 Explosives (IED remote control/roadside/car bombs; landmines, hand grenades)
5 Suicide attacks
6 Execution style killings (shooting, beheading, hanging)
7 Kidnapping
8 Arrest
9 Arson/raze (shops/house/schools, etc)
10 Unknown (when method of attack/violence is not mentioned)
11 Other insurgent violent methods (robbery of NGOs, lashing, amputation, occupation of schools)

The final step to creating the spatial database was to create shapefiles of our data. Shapefiles are sets of files which store geographic data (the shape and location of the NWFP, for example) along with the attributes of the geographic data (such as the total number of incidents in NWFP). These files are the basis for the interactive mapping sections for Eyes on Pakistan. The ESRI product ArcServer was used to create the interactive mapping portion of the website. Specifically, the JavaScript API of this software platform was used to create the interactive mapping portal. ArcServer is one of many available options for serving geographic information. Other options considered were GeoserverMapServerSpatialKey, and Google Earth.

III. Results

The final data set utilized by the Eyes on Pakistan mapping portal contains a total of 2,512 incidents which are mapped at the district/agency and provincial levels. Of the incidents, only 660 were not coded at the village/city level- just over 26% total. This is an acceptable level of results for international geocoding endeavors, given the remoteness of the areas in question, in addition to the fact that some media incident reports only contained district/agency level information, not village level. The results of geocoding to the district/agency level were extremely high- 99.5%, as only 14 incident reports did not provide information at this geographic scale. Finally, only one incident was not attributable to the provincial level.

For the Eyes on Pakistan website, all data were aggregated to the district/agency and the provincial levels to include the largest set of data possible (99.5% mentioned above). This aggregation was also important in examining trends across larger geographic areas to look for differences in data fields including attack types, agents, and number of dead or wounded. Table Three shows the range of total incidents by district/agency, which provides an understanding of the main areas of activity. Particularly of interest are the seven districts/agencies which had over 100 incidents in the 2005-2009 time frame.

Table Three: Incidents by District or Agency

No Incidents <20 Incidents 20-100 Incidents >100 Incidents
Tribal Area Adjacent Peshawar Haripur Mardan Orakzai (102)
Tribal Area Adjacent Dera Ismail Khan Swabi Tribal Area Adjacent Kohat Mohmand (127)
Tribal Area Adjacent Bannu Batagram Upper Dir Khyber (175)
Tribal Area Adjacent Lakki Marwat Chitral Bannu North Waziristan (181)
Tribal Area Adjacent Tank Tank Dera Ismail Khan South Waziristan (287)
Karak Manshera Malakand Bajaur (292)
Kohistan Nowshera Kohat Swat (545)
Abbottabad Buner
Lakki Marwat Lower Dir
Shangla Kurrum
Charsadda Hangu

Another important trend for analysis was the increase in incidents over the course of the study period. Figure Two tracks the incidents recorded by year and by province. While initial incidents were constrained to FATA, by 2007, they began to spread to NWFP and to also increase in absolute numbers through the end of the study period.

Figure Two: Incidents per Year

1

While the database was based upon media reporting sources and therefore unlikely to encompass the entire scope of the killings and wounded people in the region, these numbers were collected to gain an understanding of the general scale. As can be seen in Figure Three below, the numbers of killed and wounded for FATA exceed those of NWFP in absolute terms, with approximately 9,000 killings counted in FATA and 5,800 in NWFP.

Figure Three: Total Killed or Wounded by Province

2

Figures Four and Five show further detailed statistics from the database of incidents, highlighting the 14 Agent types and 11 Attack method types, from Tables One and Two, respectively.

Figure Four: Agent Types by Province*

3
*Refer to Table One for Agent Codes

Figure Five: Attack Methods by Province**

4
**Refer to Table Two for Attack Method Codes

AAAS has conducted further analysis of data for the region to identify trends. A major data trend is the density of the total number of incidents in the database (Figure Six). Seven Districts/Agencies have experienced the highest level of incidents, and all except for one (Swat) are in the FATA region, which borders Afghanistan. Swat has nearly double the number of incidents of the next highest ranking district/agency, Bajaur. Nearly half of the incidents in Swat have been ground-based attacks. Ground- based attacks are the major incident type in all seven of these areas. North and South Waziristan, however, exhibit air strikes and drone attacks as prevalent methods in addition to ground-based attacks. Analysis has also shown a major increase in the number of incidents from year to year. The year-to-year increase from 2008 to 2009 shows an incident number increase of 330% (540 to 1787 incidents) and a 600% increase from 2007-2008 (90 to 540 incidents).

Figure Six: Density Map of All Incidents Recorded Between 2005-2009

blog_incident_graphic

There were 93 drone attacks collected from media sources for the entire study period incorporated into the spatial database for Eyes on Pakistan (Figure Seven). In order to verify that the database captured the scope of drone incidents, AAAS compared drone attack data to that published in the Washington Post in early 2010. As can be seen in Table Four, the information collected by Eyes on Pakistan is very close to that reported in the Post.

Figure Seven: Recorded Drone Attacks

drone_attacks_crop

Table Four: Comparison of Eyes on Pakistan Recorded Drone Attacks to Washington Post Findings

Year Eyes on Pakistan Washington Post
2009 52 55
2008 35 Mid-30s

IV. Conclusion
AAAS utilized a variety of geographic techniques in the creation of the Eyes on Pakistan spatial database and interactive web mapping portal. These techniques included the creation of a fuzzy matching system for geocoding international place-name information, compilation of over 2,500 media reports and the associated map files, utilization of the ArcServer development platform, and spatial-statistical analysis. The result of these efforts is the Eyes on Pakistan interactive map portal for use by Amnesty International, USA and a variety of geospatial outputs, particularly detailed maps of the study area, density analysis to determine hot spots of different types of activities occurring in the region, and the detection of major trend in the data that provide insight into the situation in northwest Pakistan.

V. Further Resources