Human Settlements, version 3.0
Presence of human settlements, defined as aggregations of man-made built-up protruding from the ground. Version 3.0 updates the previously delivered version.
What is the “Human Settlements” layer?
The Human Settlements layer reports the presence of human settlements across the world in systematic and globally consistent manner. Human settlements are aggregations of man-made built-up protruding off the ground. The generalization scale controlling the extent of settlements is customizable and settlements can be defined at the scale of interest and depending upon the application. The Human Settlement includes three layers:
(1) Built-up Index (BuI): a gray scale layer with intensity extremes of 0.0 and 1.0 (floating point values). The intensity of each pixel / unit area specifies the percentage of the unit area covered by built-up. The latter is defined as all man-made structures protruding from the ground.
(2) Built-up Surface (BuS): a binary layer reporting the presence of built-up in the ground. BuS is assigned a non-zero value for any unit area confirmed to contain built-up with confidence of 15% or higher.
(3) Built Surface Aggregation (Built-Surface): a vector layer aggregated at 50m with area thresholding.
The Human Settlement layer is available for the five focus countries in East Africa for the Rockefeller Foundation's Interconnected Pathways to Development (RF IPD) project.
What is the spatial and temporal resolution of the Human Settlement layer?
| Temporal Extent | 2016 - 2021 |
| Temporal Resolution | Annual |
| Spatial Coverage | RF IPD Focus Countries (Rwanda, Uganda, Nigeria, Ethiopia and Kenya) |
| Spatial Resolution | 10m x 10m, and aggregated into vector format |
We produce the Human Settlements layer for every year from 2016 to 2021 inclusive. Datasets are available in raster format (.tif) at a maximum spatial resolution of 10 m x 10 m in a systematic grid and in vector (.geojson) format.
What are the units of measurement for Human Settlements?
Binary layer showing the precense of human settlements or not
How do we estimate the Human Settlements Layer?
The AHS layers are produced by a deep learning workflow that makes use of a popular encoder-decoder architecture referred to as U-Net. The design of the AHS model is a customization of the U-Net producing two outputs; one via regression (BuI) and one via segmentation (BuS). The latter uses a combination of the Dice and Focal Loss functions to account in part for class imbalance present in the training data, if building types and sizes are to be imagined as class labels.
AHS version 3.0 differs substantially from AHS version 2.0 that was previously delivered. It is a product of high accuracy based purely on ML practices and with no dependencies to annually issued external layers. That is by contrast to the AHS vs 2.0 that was based on corner detections, aggregated to an application suitable scale and then masked by WorldPop, a global population estimation layer produced annually (temporarily frozen since 2020).
AHS is delivered at the native resolution of the input and can be aggregated in real time to support application-suitable definitions of human settlements.
What are the main data sources we use?
We use the following publicly available data sources for model inputs, ground truthing, and model calibration.
| data source | how we use |
|---|---|
| Sentinel-2 Annual Composite | |
| SENTINEL-2 is a European wide-swath, high-resolution, multi-spectral imaging mission. The full mission specification of the twin satellites flying in the same orbit but phased at 180°, is designed to give a high revisit frequency of 5 days at the Equator. | |
| Microsoft Building Footprints | |
| Google Research developed this large-scale open dataset contains the outlines of buildings derived from high-resolution satellite imagery in order to support these types of uses. |
If you have any questions, please reach out to RF IPD Project Manager, Emily Logan