Klasifikasi Bencana Kekeringan di Jawa Timur Menggunakan Citra Satelit Landsat 8 dengan Metode XGBoost

Abstract

Drought disasters cause major impacts such as reduced clean water, damage to food crops and the destruction of environmental ecosystems that occur in very large areas for a very long time. In monitoring droughts today, it takes quite a long time. Usually the time used in monitoring is carried out within a month. From the existing problems, the need for drought information to the community and authorities is needed to reduce losses by anticipating drought and mitigating the social, economic and environmental impacts caused by drought disasters. So a system is needed that can classify drought disasters quickly and accurately. This study uses the XGBoost algorithm, which has the ability to handle spectral variability in satellite imagery data. The data used includes Landsat 8 satellite imagery with Band 4, Band 5, and Band 10 for calculating the vegetation index, as well as SPI and DEM data. The dataset consists of 168 raster images from 2014 to 2022, which are then divided into 80% for training data and 20% for testing data. The XGBoost model was trained and evaluated using the data to classify images into five categories of drought severity: extreme drought, severe drought, moderate drought, mild drought, and no drought. The results showed that the XGBoost model achieved an overall accuracy of 89.40%. The high precision, recall, and F1 score values for each class indicate the model's ability to identify various levels of drought severity well.