Maasai Skywatch is a proposed online tool that uses publicly available satellite imagery and sophisticated object-detecting algorithms to monitor, detect, and analyze land-related injustices against the Maasai people.

The Maasai are a pastoral people who have occupied the regions of Southern Kenya and Northern Tanzania for generations.

In recent years, precarious land-rights laws have made it possible for governments and private individuals to easily and illegally acquire traditionally owned and occupied Maasai land.

As a result, unannounced village burns, forced evictions and illegal land grabs have plagued the Maasai community for decades.

By leveraging the latest A.I. technology and satellite imagery, Maasai Skywatch provides an online tool to quickly identify, locate and monitor many of these illegal land-based crimes.

Whether they are unannounced village burns, forced evictions, or sudden changes in land-use, Maasai Skywatch can provide the robust, unbiased, and hard evidence needed to fight many of these social injustices and protect Maasai livelihood and land for many more generations to come.

1. Autonomous Monitoring

• Village locations, counts & densities

• Estimated population counts

• Changes over time

  • sudden movements of people

  • sudden drops or increases in village densities

  • changes in average house counts per village

  • changes in historically or currently disputed regions

  • land use changes

• Monitor over various scales (village > region > country)

2. Rapid Detection

• Illegal village burns

• Forced evictions

• Potential land grabs

• Land use changes

• Quick notification of possible injustices

3. Instant Analysis & Forecasting

• Past and current population movement patterns

• Historic and current trends (eviction, migration, change in village sizes, density and house count)

• Future predictions based on past and current patterns and trends

Tight Security & Reliability

• Provides autonomous, reliable and unbiased data based on publicly available satellite imagery.

• Does not rely on witnesses who may fear reprimand, punishment, or harm.

• Completely independent of government and potentially biased “official” reporting outlets.

• Provides transparent high-level data and insight that would have otherwise been unavailable

Step 1: Collect satellite imagery

By using publicly available satellite imagery as our input data source, we are able to monitor, detect, and analyze Maasai villages and associated land-based injustice from above. As both historic and current satellite imagery can be collected, we can detect both past and present injustices committed against the Maasai.

The following satellite image sources were used.

• CNES / Airbus, Maxar Technologies (super high-res.)

• Sentinel-2, MSI (high-res.)

• Landsat-98 (medium-res.)

Step 2: Gather training data

In order to autonomously detect Maasai villages, home burnings, and other injustices from above, we used a sophisticated neural network algorithm called Yolo-V3. This particular algorithm has the innate ability to learn what specific objects look like and then accurately detect them at an extremely rapid pace and over large swaths of land. Such a rapid detection system is capable of locating, detecting and reporting land-based data at a rate far beyond human capacity. These kinds of algorithms are called Object Detection Models and are commonly used for real-time tasks such as detecting vehicles for driverless cars or faces for phone cameras.

In order to teach our Object Detection Model to detect villages, we trained it on a custom dataset of 600 satellite images of Maasai villages compiled and augmented within Roboflow, a dataset builder application. Our final training dataset contained images of villages in various states (burned, not burned, etc.) , landscapes (desert, grassland, etc.) and contexts (dense, not dense, etc.), allowing our model to detect villages in varying environments and locate instances of potential injustice.

Step 3: Train object detection model

By using FastAi, we trained a Yolo-V3 neural network algorithm to detect Maasai villages and various other aspects such as burn evidence, number of houses, and house materials.

Yolo-V3 was chosen as it can quickly and accurately detect multiple objects in real-time scenarios. As we are scanning vast tracts of land for multiple object types, Yolo-V3 was the ideal algorithm to use.

After training, our model was able to rapidly and accurately locate and detect villages, determine which ones have been burned, provide a house count per village, and detect whether homes were made out of traditional cow dung or metal.

Step 5: Apply model to real world scenarios

After training and testing our YoloV3 object detection model, it was time to apply Maasai Skywatch it to a real-life scenario.

Maasai Skywatch’s first task was to detect and provide burned village data for a disputed Maasai region in Tanzania called the Loliondo region.

in 2014, Tanzanian officials announced plans to illegally sell 1,500 square kilometers of Maasai land to a private luxury hunting and safari company based in the United Arab Emirates. After the deal went through in 2017, over 5,800 Maasai homes were illegally burned or damaged and 20,000 Maasai were forcefully displaced and left homeless.

By feeding historic 2017 satellite imagery of the Loliondo region into our trained Yolo-V3 model, it successfully found and detected both burned and intact Maasai villages at an extremely rapid rate. It also determined the number of homes in each individual village, which if changing over time, may indicate rapid drops in population and potential forced evictions.

Village detection result sample

"the app is symbolic as it is not just about politics. It's about protecting our land, and that is important for all Maasai. I know they will accept that".

- Member of Maasai diaspora community