Türkiye can expand solar by 120 GW through rooftops
In the analysis, the publicly available Microsoft Building Footprints database, which contains more than 1 billion roof coordinates stored as polygons, was used to identify roofs in Türkiye. For this study, the May 2022 update, stored as a “Feature Collection” in the Google Earth Engine, was used. The dataset contains 18,058,257 polygons within the borders of Türkiye, including all kinds of structures with roofs.
More than one source was examined regarding the surface area a 1 kW solar power plant installed on the roof will occupy. According to a UK-based company that lists the most suitable solar panel options for homeowners and enables them to get price quotes, the required roof area is calculated as 6.4 m 2 per kW, assuming panels with less than 20% efficiency and with a capacity of 260 watts. According to a US-based website established to assist consumers in the solar energy sector, and an Australian service provider that lists solar panel suppliers to facilitate obtaining price quotes, the roof area required for 1 kW with panels of 330–400 watts and an efficiency of at least 20%, is from 4.1 to 5.6 m2. When ten random rooftop SPP projects completed in Türkiye in 2021-2022 were examined using satellite images, it was observed that the average area required for 1 kW of rooftop SPP capacity was 6.3 m2. Therefore, taking a conservative approach for the calculations, it was assumed that a 1 kW panel would cover an area of 6.4 m2.
The process of classifying roofs into three separate categories started with a training set including all three types identified in a satellite image containing only roofs for a selected province. In creating the training set, a sufficient number of randomly selected roof images were manually labeled according to the three roof types (flat empty/pitched empty/full). The decision to create a sufficient number of training sets was made using a validation set created completely independently of the training set. The validation set, selected from eight different regions of Türkiye, required the manual labelling of thousands of points in each region according to whether they were flat/pitched/full. Then, a visual classification algorithm designed on Google Earth Engine (GEE) was run to calculate accuracy rates in the validation set, and the training set was expanded and improved to maximize accuracy. The training set was created in Ankara, and the provinces covered by the validation set included Istanbul, Ankara, Izmir, Antalya, Konya, Erzurum, Trabzon and Şırnak. During validation, the final model achieved accuracy scores of 97% for empty pitched roofs, 83% for empty flat roofs, and 89% for full roofs.
Some corrections were applied to the roof areas following classification into three separate categories. The first correction was to reclassify areas with insufficient space for a panel as unsuitable roofs. Another adjustment was made for regions such as Antalya and Mersin, where greenhouse cultivation is common. For these regions, the coordinates of greenhouse roofs were manually identified and likewise classified as unsuitable.
