Machine-learning fusion of PolSAR and LiDAR data for tropical forest canopy height estimation

M. Pourshamsi (1) , M. Garcia (2) , M. Lavalle (3) , E. Pottier (4) , H. Balzter (1)

1 University of Leicester
2 UAH - Universidad de Alcalá - University of Alcalá
3 JPL - Jet Propulsion Laboratory
4 IETR - Institut d'Électronique et des Technologies du numéRique

This paper investigates the benefits of integrating polarimetric radar variables with LiDAR samples using Support Vector Machine (SVM) to estimate forest canopy height. Multiple polarimetric variables are required as an input to ensure consistent height retrieval performance across a broad range of forest heights. We train the SVM with LiDAR samples and different polarimetric variables based on 5000 samples (less than 1% of the full subset) collected across the images using a stratified random sampling approach. The trained SVM was applied to the rest of the image using the same variables but excluding the LiDAR samples. The estimated height was cross validated versus LiDAR-derived height (RH100) yielding overall good accuracy with r 2 =0.86 and RMSE = 6.8 m. © 2018 IEEE

Mots clés: Forest height; Fusion; LiDAR; Polarimetric synthetic aperture radar (PolSAR); Support Vector Machine (SVM)

Domaines: Sciences de l'ingénieur [physics]

M. Pourshamsi, M. Garcia, M. Lavalle, E. Pottier, H. Balzter. Machine-learning fusion of PolSAR and LiDAR data for tropical forest canopy height estimation. 38th Annual IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2018, Jul 2018, Valencia, Spain. pp.8108-8111, ⟨10.1109/IGARSS.2018.8518030⟩. ⟨hal-02121290⟩