Land cover change in Badung Regency 2016-2020 : An approach using machine learning method: Random Forest Extreme Gradient Boost (XGB)

ABSTRACT

Badung Regency is one area that mostly suffered from Covid-19 pandemic. Their gross regional domestic product has decreased 21.5% from 2019 to 2020 because of sluggishness of the tourism sector. It also affects the physical development of Badung Regency as a fast-changing area. To map the change of its land cover, satellite imagery-based classification was conducted. Both optical and radar imagery has its own deficiencies due to cloud cover in optical imagery and difficulties in interpretation in radar imagery. Therefore, combining optical and radar imagery and classifying the land cover through machine learning (ML) algorithm is necessary. In this study, we compare two methods of ML which are Random Forest and Extreme Gradient Boost. Sentinel 1 and 2 imageries utilized as the input to derive land cover change from 2016 to 2020. The data is classified into five classes dense vegetation, sparse vegetation, bare land, water body, and urban, using supervised classification. As for training and validation, the field survey data was conducted. With similar input and set of training data, Extreme Gradient Boost (XGB) methods yield higher average accuracy than Random Forest (RF). The XGB has around 93% of accuracy, while RF has around 76% accuracy. From the result of land cover change using XGB method, bare land and water bodies are decreasing 22.9% and 4.1% consecutively. While urban areas and sparse vegetation, slightly develop around 5.6% and 1.26%. Dense vegetation has almost not changed with increasing 0.34% of its area. © 2021 IEEE.