Satellite Remote Sensing for Stand Age Estimation in Tree Plantations: A Case Study Using Landsat Images and Machine Learning Algorithms

Stand age information is essential for the efficient monitoring and sustainable management of plantation forests. Satellite images from remote sensing, with their free and data availability, provide spatial and temporal information over large extents of an observed area. Its applications include land use/landcover mapping, resource monitoring, change detection carbon and biomass assessment, leaf area index (LAI) prediction and disease detection, and estimating forest-related attributes like stand age. In this work, the utilization of satellite remote images and machine learning algorithms was explored for detecting and estimating the stand ages of Falcata plantations. Specifically, Maximum Likelihood Classifier (ML), Support Vector Machine (SVM), and Random Forest (RF) algorithms were trained and applied for Falcata plantation stand age estimation, using the Landsat 8 OLI image bands, Normalized Difference Vegetation Index (NDVI), as well digital elevation model (DEM) and slope data layers. Multitemporal layer stacking of Landsat 8 OLI spectral bands, NDVI, along with the additional data layers - DEM and Slope produced the best classification results of 93.67% using Random Forest Classifier. RF and SVM provided better performance than ML when multiple data layers were used in classifying Falcata plantation stand ages and Falcata from non-Falcata. The ML performed best when applied to single-year data with NDVI. It may imply that using more variables or data layers is not effective in improving the classification accuracy of ML. Although ML failed to differentiate more Falcata stand ages, it could still differentiate Falcata from non-Falcata. Among the classifiers used, RF achieved the best classification performance in two classifications-Falcata Plantations stand ages and Non-Falcata, and Falcata and Non-Falcata Plantations. Improvement in the classification accuracy may be achieved if these parameters were optimized.