Neural networks, multitemporal landsat thematic mapper data and topographic data to classify forest damages in the Czech republic

Summary, in English

This study uses multitemporal Landsat Thematic Mapper data and topographic data for the purpose of classifying coniferous forest damage in the Czech Republic using an artificial neural network. Comparing the neural network-based classification with earlier studies and a multinominal logistic regression using identical training and test data indicates that the back propagation algorithm is comparable, but not superior, to conventional methods. The dependence on the randomly set input weights and the more time-consuming back propagation training make neural network less useful for classification of forest damages than conventional classification algorithms. However, the ability to integrate and extract information from multisource data with different or unknown distributions are advantages of neural networks.

• Author: J. Ardö; P. Pilesjö; A. Skidmore
• Department/s: Dept of Physical Geography and Ecosystem Science
• Publishing year: 1997
• Language: English
• Pages: 217-229
• Publication/Series: Canadian Journal of Remote Sensing
• Volume: 23
• Issue: 3
• Links: http://dx.doi.org/10.1080/07038992.1997.10855204
• Document type: Journal article
• Publisher: Taylor & Francis
• Topic: Environmental Sciences
• Status: Published
• ISBN/ISSN/Other: ISSN: 0703-8992