Abstract:
Soil is an important non-renewable source. Its protection and allocation is critical to
sustainable development goals. Urban development presents an important drive of soil
loss due to sealing over by buildings, pavements and transport infrastructure.
Monitoring sealed soil surfaces in urban environments is gaining increasing interest
not only for scientific research studies but also for local planning and national
authorities.
The aim of this research was to investigate the extent to which automated classification
methods can detect soil sealing in UK urban environments, by remote sensing. The
objectives include development of object-based classification methods, using two
types of earth observation data, and evaluation by comparison with manual aerial
photo interpretation techniques.
Four sample areas within the city of Cambridge were used for the development of an
object-based classification model. The acquired data was a true-colour aerial
photography (0.125 m resolution) and a QuickBird satellite imagery (2.8 multi-spectral
resolution). The classification scheme included the following land cover classes: sealed
surfaces, vegetated surfaces, trees, bare soil and rail tracks. Shadowed areas were also
identified as an initial class and attempts were made to reclassify them into the actual
land cover type. The accuracy of the thematic maps was determined by comparison
with polygons derived from manual air-photo interpretation; the average overall
accuracy was 84%. The creation of simple binary maps of sealed vs. vegetated surfaces
resulted in a statistically significant accuracy increase to 92%. The integration of
ancillary data (OS MasterMap) into the object-based model did not improve the
performance of the model (overall accuracy of 91%). The use of satellite data in the
object-based model gave an overall accuracy of 80%, a 7% decrease compared to the
aerial photography.
Future investigation will explore whether the integration of elevation data will aid to
discriminate features such as trees from other vegetation types. The use of colour
infrared aerial photography should also be tested. Finally, the application of the object-
based classification model into a different study area would test its transferability.
