A Land Cover Classification Using a Random Forest Model

Authors- Davaasuren Bayarmagnai, Professor Bayanjargal Darkhijav, Professor Tsolmon Renchin

Abstract-Earth’s surface forms the outermost layer of our planet, and it changes due to both natural processes and human activities. Therefore, the classification of land surface features is essential for many environmental applications and serves as the basis for survey studies. Various methods are used for land cover classification using satellite data, such as random forest classification and the decision tree method of machine learning. In this study, we used the Random forest(RF) classification, which shows robustness and provides high accuracy compared to other image classification methods in remote sensing. The study area is in Khangal soum of Bulgan province of Mongolia, a forest-steppe zone with mountains and hills. Land cover types of the study area include bare land, forest, and grass. Spectral bands of Blue, Green, Red, and Near Infrared(NIR) of Landsat 8 data and ground observation data were used in the research. A confusion matrix was obtained by comparing the results obtained by the random forest method with the ground observation values, and the result was 86.4 percent. Using the results, we applied Random forest results to create a land cover map for 2017-2021. However, larch forests are estimated to be the most significant percentage in the study area. RF can be applied to different forest classifications in any forested region of Mongolia to save time and money. Ontslogiig bichih. Tsoon erdemten GEE.

DOI: /10.61463/ijset.vol.12.issue3.189