Page 138 - Çevre Şehir ve İklim Dergisi İngilizce - Özel Sayı

National Scale Land Cover Classification and Monitoring System

Land cover classes that are not automatically classified through satellite
images or exhibit low classification accuracy are directly classified through the
use of auxiliary data of the stakeholder institutions. To ensure the dynamism
of the system, the data will be continuously obtained from stakeholder
institutions for the relevant map production years.
With the system created within this framework, 32 of the 79 UASIS land
cover classes shown in Table 3, indicated in italics, through the use of the
aforementioned artificial intelligence-based approaches. Other classes are
directly added to the map through the data prodived by the institutions
stated in Table 4.
Although it is anticipated that some classes that are not automatically
identified can be identified through remote recognition methods, there is a
possibility of confusion with different classes. Examples include the sub-class,
airports, which is in the main class, artifical surfaces.An airport consists of an
administrative building resembling any other building, a runway that resembles
a road, and an open area surrounding these structures, usually covered with
grass. These parts are classified separately as settlements, roads, and grassland
vegetation in algorithms that produce data in a 0.25-hectare mapping unit. On
the other hand, an airport is a land cover unit made up of these three coming
together. For this reason, some sub-classes like airports and harbords are
provided not through classification algorithms but through auxiliary data.

2.4. Forest Density and Settlement Areas Impermeability,
CORINE Transformation Map, Change Monitoring

The UASIS project aims for the creation of land cover data as well as
forest density maps, settlement areas impermeability maps, and CORINE
transformation maps. Moreover, a land cover change map is also one of the
target outcomes.
Copernicus High Resolution Layer (HRL) maps, Tree Cover Density (TCD),
and Copernicus High Resolution Imperviousness Density (HRL IMD) maps
are being usedto train the algorithms developed for the production of forest
density and settlement areas impermeability maps, For each point, if there
is nodata in the HRL TCD data, new labels are created and forest density
calculations are performed by assigning an average density of 5 for areas
labeled as forest soil (OT) in the Forest Stand Data, for labels with a low
density of 1-10 in the HRL TCD data. HRL-TCD data are saved for other points.
An algorithm based on raster data has been developed for the transformation
between UASIS and CORINE land cover classes in the production of the
CORINE transformation map. The smallest unit to be mapped as CORINE

125
Special Issue / 2024