Battle of Marawi: Using Object Detection to Observe Trends Before, During, and After Urban Combat

Battle of Marawi: Using object detection to observe trends before, during, and after urban combat

Understanding what is happening to a region experiencing conflict is time-consuming and the scant data available is often latent or does not always provide a complete picture.

This causes delays to humanitarian aid. Proxies such as car counts and their spatio-temporal patterns can be indicative of what is actually happening on the ground in regards to civilian and military movements. Using our Orbital Insight GO Platform, we automatically counted approximately 35,000 cars in available imagery over an 8 year time period. GO counted cars before, during, and after the Battle of Marawi. The results show a change in car counts over time as well as differences in the city’s spatio-temporal distribution of car clustering and dispersion. This GO project started creating data within hours of project creation.

Detections of cars (blue dots) produced from Orbital Insight’s Car Detector Algorithm. (Imagery source: Digital Globe)

Background on The Battle of Marawi

The Battle of Marawi was a five-month long conflict that started 23 May 2017 and is considered the longest urban battle in modern Philippine history (Singh, 2018). The main belligerents involved included the Armed Forces of the Philippines (AFP) and militants affiliated with The Islamic State (ISIS), including the Maute and Abu Sayyaf Salafi jihadist groups (Franco, 2017). These terrorist organizations took over part of the city spurring the Republic of the Philippines to declare martial law. The AFP’s mission was to then clear and retake the city. For this use case, the GO Platform’s car detection algorithm was used to automatically detect cars from DigitalGlobe imagery. The area examined is the southeastern part of Marawi, which is separated from the “safe zone” to its northwest by the Agus River and bordered to its south/southwest by Lake Lanao. This area was most affected by the fighting (Gunaratna, 2017).

Time Series Data

Based on the time series of this area, raw car counts immediately decline once the battle starts on 23 May 2017. These trends are consistent with the expectation that commercial and passenger traffic through Marawi are severely disrupted due to the battle. By setting user-defined thresholds, GO subscribers can be alerted of when there is a relevant change in objects, such as a 15% decrease in cars. Historical counts are used to create a baseline and observations deviating outside the user-defined thresholds triggers an alert. This automation allows a user to spend more time conducting a deeper analysis rather than spending time counting cars.

GO’s objectivity, transparency and speed of data revealed two major insights:

  • Notable drops in raw and rolling mean values during the initial start of the battle in late May 2017.
  • As of July 2018, car counts have not completely recovered to their pre-battle numbers, possibly indicating that the city is still rebuilding, lacking infrastructure, and not suitable for repopulation.

Car counts within the region can be monitored and used as a metric to evaluate re-population patterns by humanitarian or government organizations.

Spatio-temporal Data

In addition to the overall impact the battle has on traffic within Marawi, spatial patterns are analyzed to derive additional conclusions or open new lines of analytic questioning. GO allows for the retrieval of detected object locations, which can be visualized in a geographical information system (GIS) like ArcGIS or QGIS.

Prior to the start of the battle, the most densely populated car counts were found in the west/southwest part of the conflict area AOI. Once the battle began, civilians were displaced, causing a change in the civilian pattern of life (PoL). Car counts decreased overall within the kinetic area, as did car dispersion. By understanding the dispersion of cars, an analyst can better identify civilian and/or military patterns. GO makes it possible to quickly acquire this data and incorporate it into any pre-existing intelligence models.

Potential users of the data include humanitarian organizations doing post-battle reconstruction analysis, or military commanders using car locations as a proxy for civilian locations. These personas could use this data to plan for aiding displaced persons or to help minimize collateral damage during missions, respectively.

A) 14 MAR 2017: Before battle; a normal pattern of life with car counts clustered in the CBD. B) 29 MAY 2017: One week into battle; severe decline in car counts overall, still some clustering in CBD, a potential exodus of civilians. C) 18 SEP 2017: One month before the end of the battle; continued decline in car counts, less clustering, potential removal of most civilian vehicles. D) 07 JUN 2018: Seven months after the battle; car counts beginning to increase, no CBD clustering, potential reconstruction efforts but no return of civilian POL.

References

Franco, J. (2017). The Maute Group: New Vanguard of IS in Southeast Asia? Retrieved from https://dr.ntu.edu.sg/handle/10220/42544

Gunaratna, R. (2017). The Siege of Marawi: A Game Changer in Terrorism in Asia. Counter Terrorist Trends and Analyses, 9(7). Retrieved from http://www.jstor.org/stable/26351533

Singh, J. (2018). One Year After Marawi: Has The Threat Gone? Retrieved from https://dr.ntu.edu.sg/handle/10220/44949


Orbital Insight’s mission is to help the world understand what is happening on and to the Earth. Through multiple sources of geospatial and proprietary AI, we deliver a 360 degree view into the physical world at the speed of change in an approachable and versatile platform designed to answer your specific questions.

To ask your first question with Orbital Insight GO, email sales@orbitalinsight.com.

Back to Blog