Informal Settlements in Turks and Caicos

Over many years, several virtually self-contained communities have been established outside of the national development efforts of the national authority of Turks and Caicos. Informal settlements such as these usually have the following issues as stated by the UN.

Informal settlements are defined by the United Nations Human Settlements Programme (UN-Habitat) 2020, as: 1. inhabitants have no security of tenure regarding the land or dwellings they inhabit, 2. neighbourhoods usually lack, or are cut off from, basic services and city infrastructure, 3. housing may not comply with current planning and building regulations, and 4. housing is often situated in geographically and environmentally hazardous areas.

Methods are needed to identify the locations of theses settlements, quantify the number of structures present, and determine whether or not they reside on private or Crown Land

The images below speak to the scope and severity of the issue. In just 5 years a new settlement has been established on Crown land.

5cm Drone Imagery provided courtesy of PLACE

Parcel boundaries provided courtesy of the Survey and Mapping Divisions, Turks and Caicos

The Imagery provided by Place was extremely high resolution. We decided to test the tools on just one image chunk. That image chunk was roughly 50GB and had 5cm resolution.

Esri ran deep learning models trained on Africa, Australia and USA imagery to extract the building footprints from the drone imagery. After several rounds of testing, the model trained over the US result in more accurate extraction of building footprints. The model was then run on the entire image. In 2.5 hours, the model tagged nearly 12,000 buildings. The building footprints were then regularized to give them sharp corners and define their boundaries. The model did well detecting buildings in both urban and suburban settings. It did detect buildings where none existed and failed to detect others. Because of this we determined that the model needs to be fine tuned for Turks and Caicos for better results. Overall, the results were encouraging, especially compared to the effort and time required for human imagery extraction.

• Fine tuning the model
• QC and Masking
• Land Classification
• Statistical Analysis
• Surface modeling
• 3D building creation