Forest Ecology and Management 260 (2010) 1843–1852 Contents lists available at ScienceDirect Forest Ecology and Management journal homepage: www.elsevier.com/locate/foreco Tree species classification from fused active hyperspectral reflectance and LIDAR measurements Eetu Puttonen a,∗ , Juha Suomalainen a , Teemu Hakala a , Esa Räikkönen a,b , Harri Kaartinen a , Sanna Kaasalainen a , Paula Litkey a a Department of Photogrammetry and Remote Sensing, Geodeetinrinne 2, 02341, Kirkkonummi, PO Box 15, Finland b Klastech GmbH, Konrad-Adenauer-Allee 11, 44263, Dortmund, Germany article info Article history: Received 4 June 2010 Received in revised form 13 August 2010 Accepted 18 August 2010 Keywords: Classification Data fusion Forestry Hyperspectrum LIDAR abstract A new terrestrial laser system was tested for tree species classification. A dataset consisting of shape parameters of three boreal tree species was collected with Light Detection and Ranging (LIDAR) and integrated with an actively measured reflectance hyperspectra. Tree species were classified using param- eters derived from reflectance spectra and point cloud shape distribution. Classification performance was tested with individual, paired, and mixed combinations of both reflectance and shape parameters. The best classification results were obtained with combined datasets consisting of two reflectance and two shape parameters. Of all tested classification parameter combinations, 67.5% were able to classify all trees with over 90% accuracy. The best reflectance spectrum bands for the examined species were located around 550 and 700 nm. The best shape parameters described the upper midsection or the tops of the trees. This study was a successful step in developing classification algorithms for integrated LIDAR and hyperspectral data. © 2010 Elsevier B.V. All rights reserved. 1. Introduction Accurate tree species information is needed in several fields in forest management. Environmental planning, resource man- agement and forest industry all benefit from these types of data. Traditionally, species information has been collected by field measurements that are very accurate, but expensive and time- consuming. Laser scanning studies, both airborne (ALS) and terrestrial (TLS), have experienced a rapid development during the last decade. Hyyppä et al. (2009) have made an extensive review of several ALS applications in forestry. Terrestrial laser scanners are another valu- able tool for forest attribute measurements on single tree and plot levels. Henning and Radtke (2008) give an extensive overview of the different methods used in TLS research. Laser scanning mea- surements provide detailed spatial data over large areas with high speed. Several parameters can be measured accurately from TLS-data, for example, tree location and diameter at various heights. Shad- owing can cause problems when only one scan is considered, but this can be often solved by taking several scans. Typically one full ∗ Corresponding author. Tel.: +358 9 29555239. E-mail address: eetu.puttonen@fgi.fi (E. Puttonen). URL: http://www.fgi.fi/envilaser/hyper.html (E. Puttonen). scan of the environment takes a couple of minutes, and from these data tree trunks can be extracted to several tens of meters with centimetre accuracy. Using TLS, stem diameter can be measured as accurately as it is measured in traditional field measurements with steel calipers (Vastaranta et al., 2008). Single-wavelength laser scanners are built to provide accurate spatial data, which allows object classification based on their shape properties. However, a single wavelength and the monitoring of the intensity changes of the laser beam are not optimal for spec- tral target characterization which is another approach to object classification. Therefore, measurement systems that can operate on more than one wavelength are of great interest. Morsdorf et al. (2009) have studied the concept of multiwavelength, full- waveform canopy LIDAR instrument and its uses in biochemical canopy parameter extraction through simulations. Both the instru- mentation and the trees were simulated in the study. The obtained results showed that canopy, understory, and soil physiology were separable with multispectral data, and that the seasonal variation of the chlorophyll was also traceable in contrast to using a single- wavelength LIDAR alone. One way to collect and measure several laser wavelengths simultaneously is by using frequency multiples of the used laser beam. For example Tan and Narayanan (2004) have developed the Multiwavelength Airborne Polarimetric LIDAR (MAPL) that oper- ates on two wavelengths and is designed for vegetational remote sensing. There is also a new type of laser systems that can pro- 0378-1127/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.foreco.2010.08.031