Abstract
Open-source satellite imagery such as Sentinel-2 has been widely proven reliable for various geospatial applications. However, achieving geospatial independence remains crucial for any country to reduce reliance on foreign data sources and strengthen national sovereignty in Earth observation capabilities. In this context, Indonesia initiated a microsatellite development program in 2007, which has now reached its third generation with LAPAN-A3. Despite these efforts, LAPAN-A3 is still considered an experimental satellite, and further evaluation is required before it can be fully adopted for operational applications. This study evaluates the performance of LAPAN-A3 imagery for land cover mapping using machine learning approaches and compares its performance with the well-established global dataset Sentinel-2. Two widely used classifiers, Random Forest (RF) and Support Vector Machine (SVM), were implemented within the Google Earth Engine (GEE) platform and tested using different combinations of spectral features. The results show consistent improvements in classification performance when additional spectral features are incorporated for both LAPAN-A3 and Sentinel-2 datasets. In all feature configurations, RF outperforms SVM, achieving higher Overall Accuracy (OA) and Kappa coefficients. Although Sentinel-2 generally yields slightly better results, LAPAN-A3 demonstrates promising performance despite its experimental nature. These findings highlight the potential of LAPAN-A3 as a national remote sensing asset that can contribute to Indonesia’s long-term goal of achieving geospatial independence and strengthening domestic Earth observation capabilities.