Support vector machines regression for estimation of forest parameters from airborne laser scanning data

Jean-Matthieu Monnet (1) , Frédéric Berger (1) , Jocelyn Chanussot (2)

1 UR EMGR - Ecosystèmes montagnards
2 GIPSA-SIGMAPHY - GIPSA - Signal Images Physique

Estimation of forest stand parameters from airborne laser scanning data relies on the selection of laser metrics sets and numerous field plots for model calibration. In mountainous areas, forest is highly heterogeneous and field data collection labour-intensive hence the need for robust prediction methods. The aim of this paper is to compare stand parameters prediction accuracies of support vector machines regression and multiple regression models. Sensitivity of these techniques to the number and type of laser metrics, and use of dimension reduction techniques such as principal component and independent component analyses are also tested. Results show that support vector regression was less accurate but more stable than multiple regression for the prediction of forest parameters.

Mots clés: Support vector regression; airborne laser scanning; forest parameters estimation

Domaines: Ingénierie de l'environnement; Traitement du signal et de l'image [eess.SP]; Traitement du signal et de l'image [eess.SP]

Jean-Matthieu Monnet, Frédéric Berger, Jocelyn Chanussot. Support vector machines regression for estimation of forest parameters from airborne laser scanning data. IGARSS 2010 - IEEE International Geoscience and Remote Sensing Symposium, Jul 2010, Honolulu, Hawaii, United States. pp.2711 - 2714, ⟨10.1109/IGARSS.2010.5651702⟩. ⟨hal-00521400⟩