Leveraging GEDI and multi-temporal Sentinel data integration for detecting rubber tree plantations in Northeast Thailand

Despite the economic and environmental significance of rubber plantations in Northeast Thailand, accurate and continuous monitoring remains challenging due to the strong dependence on optical remote sensing, which is highly affected by cloud cover and seasonal variability. Existing land-use datasets are often static and fail to capture temporal dynamics and stress-induced canopy changes, frequently leading to misclassification with other perennial crops and forested areas. This limitation highlights the need for a robust, high-temporal-resolution monitoring approach capable of reliably mapping rubber plantations under different agroclimatic conditions. This study aimed to map the spatial distribution and age structure of rubber in Northeast of Thailand through the integration of satellite-based earth observation and the Random Forest (RF) algorithm supported by a large reference dataset. The RF classification model was developed using multi-temporal Sentinel-1 and Sentinel-2 imagery combined with canopy height derived from the Global Ecosystem Dynamics Investigation (GEDI) mission, using a total of 30,000 sample points. The final classification achieved an overall accuracy of 82.63%, and an F1 score of 0.84, although performance declined for classes characterized by high spectral variability, such as other crops and non-crops. The results revealed an uneven spatial distribution, with the highest concentrations located in Bueng Kan, Loei, and Nong Khai. Age-structure analysis indicated that plantations older than seven years dominate the landscape, whereas the extent of young rubber plantations (<5 years) remains comparatively limited. Significant challenges were observed in the classification of young plantations, mainly due to soil background effects, spectral heterogeneity, and mixed-pixel influences associated with spatial resolution. Overall, this study demonstrates the potential of integrating GEDI-derived canopy height data with multi-temporal Sentinel imagery and an optimized RF classification framework for rubber plantation mapping. At the same time, the findings highlight the methodological challenges that remain in accurately classifying younger plantation age groups.