Prediction of LAI in Scots Pine Forests of Türkiye Using UAV and Sentinel 2 Images

Abstract

Monitoring the structural characteristics of vegetation cover is critical for understanding ecosystem functioning and sustainable forest management. Effective Leaf area index (LAIe) is directly related to photosynthetic capacity, carbon cycle, and water balance of ecosystems. In this study, LAIe prediction was performed using random forest algorithms with five different datasets obtained from unmanned aerial vehicle (UAV) images and Sentinel 2 (S2) satellite images. The study aimed to (1) determine which UAV features perform best in the prediction of LAIe, and (2) to compare prediction accuracies obtained while using UAV and S2 features. To evaluate the contribution of different data sources, five datasets were considered, namely UAV_RGB (DS1), UAV_3D (DS2), UAV_ALL (DS3), S2 (DS4), and ALL (DS5), with corresponding RMSE% values of approximately 27.3, 29.6, 24.4, 32.1, and 24.8, respectively. The results showed that the DS3 dataset provided the highest accuracy and the most balanced error distribution. The DS5 resulted in more stable prediction performance compared to individual datasets, enabling the model to demonstrate similar accuracy levels for both low and high LAIe values. We conclude that low-cost UAV-RGB data, utilizing both 3D and 2D features, is an effective alternative to high-cost remote sensing techniques for LAIe prediction.