Browsing by Subject "Random forest"

Land use and land cover maps are vital sources of information for many uses. Recently, the use of high resolution and open access satellite images are being preferred for mapping large areas. Sentinel satellites exhibit such valuable traits. This study was designed to analyze the potential of Sentinel-1A SAR images for land use mapping in Pakistan. Machine learning methods were employed for image analysis. Random forest classifier algorithm performed significantly better than others in the training step. Thus, we took the model for tuning parameters. After several image processing steps, we classified the final image to 23 classes and achieved 42 % of an overall accuracy. The present study showed the potential advantages of using Sentinel-1 images in land use mapping besides highlighting some characteristics of Sentinel-1A images. This study also compares the results with an earlier study using Landsat-8 optical multispectral images over the same area. Similar to the prior study, overestimation in dominant classes and underestimation in rare classes were observed. The method and findings of this study could be beneficial for future studies in the use of Sentinel-1A images for land use/cover mapping over large areas.