Using Sentinel-2 Imagery and Machine Learning Algorithms to Estimate Deforestation (Case Study: a district in Malang Regency)

Abstract Forests are an inseparable element of human life. As time passes, the population of humans and other living beings will continue to increase which may induce the conversion of forest land will be widespread. This land conversion activity causes deforestation. In 2019, protected forest land in Sumbermanjing Wetan District, Malang Regency was reduced by approximately 200 Ha. To achieve Sustainable Development Goal (SDG) number 15 on sustainable forest management, a deforestation analysis is conducted to assess the rate of forest land reduction and predict forest land reduction from 2024 to 2028. This research classifies the rest land using Sentinel-2 satellite imagery from 2019 to 2023. Based on this classification, this research estimates the rate of reduction and prediction of forest area using the polynomial regression during 2024 – 2028. This research applies machine learning algorithms; random forest, SVM, and CNN as the classification method. The dataset is split into training and testing data with a ratio of 70:30. The results witness that the best algorithm for classifying forest areas is a random forest with an average accuracy of 0,957. The forest area based on the classification results using random forest from 2019 to 2023 is consecutively 14342,98 Ha, 12243,44 Ha; 13033,36 Ha; 14677,75 Ha; and 15347,88 Ha. This research concludes that the forest in Sumbermanjing Wetan District has no deforestation, with an average rate of forest addition of 438.084 Ha/year and the prediction of forest area in the period of 2024 – 2028 is consecutively 15228,73 Ha; 15666,814 Ha; 16104,898 Ha; 16542,981 Ha; and 16981,065 Ha.