Land Use Analysis Using Machine-Learning Based on Cloud Computing Platform

Land use analysis can provide a foundation for successful and efficient regional planning and environmental monitoring. The application of machine-learning on a cloud computing platform (Google Earth Engine, GEE) in land use analysis enables efficient and rapid processing of spatial data on a wide scale. It overcomes the constraints inherent in conventional approaches. The purpose of this study was to identify land use and estimate its level of accuracy using GEE and a Random Forest machine-learning method. The data utilized were the administrative boundaries of Bangkalan Regency (1:25,000) and Landsat 8 SR L2 C2 T1 satellite images from 2022. Satellite image analysis using the Random Forest algorithm on the GEE platform with the JavaScript API, including masking, cloud masking, class and sampling, training, and testing sample data. Land use study using the Random Forest algorithm yielded the following results in order of area: vegetation 65,040.39 ha (49.98%), agricultural land 31,817.16 ha (24.45%), settlements 20,578.05 ha (15.81%), open land 6,683.94 ha (5.14%), and water bodies 6,021.09 ha (4.63%). The accuracy test in GEE revealed an overall accuracy (OA) of 91.39% and a kappa score of 88.39%, or 0.88. At the same time, validation in the field gave an OA of 88.68% and a Kappa of 85.53%. The findings of this study can be applied to land use evaluation and fundamental decision-making. Keywords: land use, random forest, geographic information system, remote sensing