Capability of GLAS/ICESat data to estimate forest canopy height and volume in mountainous forests of Iran

[Departement_IRSTEA]Territoires [TR1_IRSTEA]SYNERGIE [Axe_IRSTEA]TETIS-ATTOS

International audience

Inglés. The importance of measuring biophysical properties of forest for ecosystem health monitoring and forest management encourages researchers to find precise, yet low cost methods especially in mountainous and large area. In the present study Geoscience Laser Altimeter System (GLAS) on board ICESat was used to estimate three biophysical characteristics of forests located in north of Iran: 1) maximum canopy height (Hmax), 2) Lorey's height (HLorey), and 3) Forest volume (V). A large number of Multiple Linear Regressions (MLR) and also Random Forest (RF) regressions were developed using different set of variables: waveform metrics, Principal Components (PCs) produced from Principal Component Analysis (PCA) and Wavelet Coefficients (WCs) generated from wavelet transformation. To validate and compare different models, statistical criteria were calculated based on a five-fold cross validation. The best model concerning the maximum canopy height was an MLR with an RMSE of 5.0 m which combined two metrics extracted from waveforms (waveform extent "Wext" and height at 50% of waveform energy "H50"), and one from the Digital Elevation Model (Terrain Index: TI). The mean absolute error (MAPE) of maximum canopy height estimates is about 16.4%. For Lorey's height, a simple MLR model including two metrics (Wext and TI) represents the highest performance (RMSE=5.1 m, MAPE=24.0%). Totally, MLR models showed better performance rather than RF models, and accuracy of height estimations using waveform metrics was greater than those based on PCs or WCs. Concerning forest volume, employing regression models to estimate volume directly from GLAS data led to a better result (RMSE=128.8 m3/ha) rather than volume-HLorey relationship (RMSE=167.8 m3/ha).