Investigating the effect of different datasets on land cover classification using planetscope satellite data

Land cover classification process determines quantitative decision functions that will provide the determination of different features or objects in the image. In this context, in this master's thesis, the effect of different datasets on land cover classification was investigated in two different study areas (Adana-Karataş and Ağrı-Doğubeyazıt) using PlanetScope (PL) satellite data with 3m resolution. Image processing and visualization processes in this study were performed on the Google Earth Engine (GEE) platform. Controlled classification process was performed using the machine learning algorithm Random Forest (RF). As a result of the classification, user and producer accuracies of land cover classes were calculated. The results obtained from the datasets showed that while the general accuracy (OA) value was measured as 0.76 and the general kappa (OK) value as 0.72 for the Karataş study area in dataset-1, the OA value was measured as 0.84 and the OK value as 0.81, which achieved a high accuracy value in dataset-8. While the OA value was measured as 0.74 and OK value as 0.67 in dataset-1 for Doğubeyazıt area, the OA value was 0.85 and OK value as 0.81 in dataset-9, high accuracy value was achieved. Finally, the importance values of the bands and indices in the datasets were calculated. It is seen that making a choice depending on the satellite image, land cover class type and different datasets decided on in land cover classification can be critical