Leveraging Time-Series Imageries and Open Source Tools for Enhanced Land Cover Classification

Abstract

The urgency of regularly updated land cover information is crucial for monitoring the dynamics of regional landscapes. Current land cover classification techniques are typically based on snapshots taken on specific dates; therefore, potential variations over time are ignored. This limitation is exacerbated in regions that require land cover classification professionals. To address these challenges, we propose the use of existing innovative and open-source tools. This study presents a comparative analysis of two methods: the first uses machine learning classification using Google Earth Engine and the second uses time-series imagery using R’s sits package. Both methods were tested to classify land cover in Sragen Regency, Central Java, Indonesia, and our results demonstrated the superiority of R’s sits package in terms of accuracy and reproducibility. This facilitated the acquisition of a more precise and comprehensive land cover classification and demonstrated a robust application. The sits package also benefits from being open -source, encouraging wider accessibility and collaboration in the scientific community. The implications of these results underscore the potential of leveraging time-series imagery and open-source tools such as the sits package in R for more accurate and reproducible land cover classification. This approach can significantly improve our ability to monitor and understand the dynamics of regional landscapes, particularly in areas with limited skilled human resources. Future research should refine these techniques and examine their applicability to different geographic contexts.