Mallon Technology



Artificial Surface Detection using Sentinel Satellite Imagery

Using our expertise in Earth Observation, we are exploring how readily available satellite data can assist in detecting, tracking and monitoring environmental issues.

Our latest blog examines how Earth Observation can be used to detect artificial surfaces with satellite imagery.  Our colleague Rafal Marciniak created the following case study using data captured from ESA’s constellation of Sentinel satellites.

Artificial Surface Detection using Sentinel Satellite Imagery in Northern Ireland

With the improved accessibility of satellite imagery and data, mapping and tracking urbanised areas has been a key focus for many fields.

Satellite imagery has been used to help identify areas of urban sprawl and environmental degradation by monitoring changes in land cover, vegetation, and soil erosion.  The captured data can then be applied to update GIS maps, with further analysis possible to measure urbanisation.  By utilising publicly available satellite data, it is possible to map and measure urbanisation across time and space, which can inform critical planning and decision–making.

Artificial Surfaces Northern Ireland

To show how Satellite imagery can be used to detect artificial surfaces, we have created the map below of Northern Ireland, with imagery captured in 2022.  This map was generated using three polarisation modes from Sentinel 1 and 10 bands from Sentinel 2, and 6 indexes based on these bands.  Any clouds and their shadows were removed using a cloud probability mask, near–infrared band, and solar azimuth angle.  Time–series data and a Random Forest machine–learning model were used for classification.

Map of Artificial Surfaces Northern Ireland 2022
Map of Artificial Surfaces Northern Ireland 2022

The map differs from those available as open source because it shows the real probability of artificial surface detection and is characterised as high accuracy.  A darker area indicates a high likelihood of that area being an artificial surface, whereas a light or white–coloured area indicates a low probability.  In tests for another region, a probability score above 76% corresponded to an accuracy of 98% (while maintaining a high recall and precision).

The map was generated using Google Earth Engine Environmental and Google Colab.  Using the Google engine allowed us to quickly generate a map covering large areas in a 10–meter resolution.

Below, we have created a small interactive sample map from the data above focused on the Belfast region.  This has been done by importing the data into Azimap, our web–based GIS.  This clearly illustrates how satellite data can be used within a GIS.  With the satellite data imported into a GIS, it is possible for further analysis to be undertaken with additional data.  The resulting visualisation can help to inform key policy decision–making.  

Further Information

Please get in touch with us below for further information on the methods used in this case study or to enquire about our Earth Observation services or Azimap web GIS.