Tree Species cClassification by Multiseasonal High Resolution Satellite Data
Elatawneh, Alata; Rappl, Adelheid; Schneider, Thomas; Knoke, Thomas
Technische Universität München, GERMANY

Accurate forest mapping is a fundamental issue for sustainable forest management and planning. Especially, due to the ongoing conditions of the climate change and their risk on the forest tree species survival. Accurate mapping of forest tree species with the means of remote sensing data is still a topic to be investigated. At the local level, alternatives methods to the terrestrial inventory for the managed forests by the remote sensing means can be used for mapping the forest cover annually. Nowadays, in Bavaria, the State Office for Survey and Geoinformation (LVG) acquires triennially very high resolution digital aerial images of the whole state. Such data are used intensively in the forester community for forest management and planning, however, using aerial images is still cost- and time-consuming. The new space-born sensor generations, with a very high temporal resolution, are promising for the forester community to improve their forest mapping. Multiseasonal data available by the currently operating sensors of the RapidEye system and the upcoming sensors of Sentinel-2 have the potential to improve the forest mapping, especially the tree species mapping.

In this study, we investigated the potential of multiseasonal RapidEye data for mapping tree species in a Mid European forest in Southern Germany, north of the Alps. The study area of "Heiligengeist" district of the Traunsteiner Stadtwald, a mixed, highly structured community forest and such a good representative of the "forest of tomorrow" as propagated from the Bavarian forest administration. For remote sensing such structure rich mixed forests are a real challenge.

The RapidEye data of level A3 were collected on ten different dates in the years 2009, 2010 and 2011. For data analysis, a model was developed, which combines the Spectral Angle Mapper technique with a 10-fold-cross-validation. The analysis succeeded to differentiate four tree species; Norway spruce (Picea abies L.), Silver Fir (Abies alba Mill.), European beech (Fagus sylvatica) and Maple (Acer pseudoplatanus). The model success was evaluated using digital aerial images acquired in the year 2012 by the LVG and inventory point records from 2008/09 inventory. Model results of the multiseasonal RapidEye data analysis achieved an overall accuracy of about 80%. However, the success of the model was evaluated only for all the identified species and not for the individual. This study shows the ability of the upcoming sensors (e.g. Sentinel) with high temporal capability in enhancing forest tree species mapping.