Tree Height Estimation Using Sentinel-1/2 and LiDAR Data: A Case Study in the Dalseong Wetlands, South Korea Baek, G.H..1, Mohamed, S.Y.2, Jang, H.K.3, Choung Y.J. 4, and Jo M.H.5*
1Researcher, Geo C&I Co., Ltd., South Korea
2Researcher, Geo C&I Co., Ltd., South Korea
3Researcher, Geo C&I Co., Ltd., South Korea
4Researcher, Geo C&I Co., Ltd., South Korea
5Director, Geo C&I Co., Ltd., South Korea
Abstract
Accurate estimation of tree height is fundamental to forest resource management, carbon stock assessment, and ecological monitoring. This study presents a remote sensing-based approach to estimate tree height by integrating multi-source satellite data with machine learning, using the Dalseong Wetlands in Daegu, South Korea as a case study area. The region is characterised by heterogeneous forest cover within a protected wetland ecosystem, offering a suitable testbed for evaluating forest height modelling techniques.
We utilised Sentinel-2-derived NDVI, Sentinel-1 SLC backscatter (VV and VH), and a DSM generated from Sentinel-1 InSAR imagery as predictor variables. The target variable, reference tree height, was extracted from a LiDAR-derived normalised digital surface model (nDSM). A Random Forest regression model was developed using both original and derived features, including NDVI-DSM interaction, DSM squared, logarithmic and square root DSM, square root VH, and the VV/VH backscatter ratio.
After removing invalid or noisy pixels and standardising the features, the model was trained on 80% of the data and tested on the remaining 20%. The model achieved a root mean square error (RMSE) of 2.0527 metres and a coefficient of determination (R2) of 0.6161 when compared to LiDAR-based tree heights.
The final tree height predictions were exported as a georeferenced raster map, which may serve as a valuable baseline for long-term monitoring of forest structure, biomass estimation, and habitat mapping in wetland regions. This research demonstrates the effectiveness of combining freely available SAR and optical satellite data with LiDAR and machine learning to estimate forest parameters in ecologically sensitive or data-scarce areas.
Keywords: Dalseong Wetlands, LiDAR, Random forest, Sentinel-1/2, Tree height estimation.
