Artificial neural networks : applications in morphometric and landscape features analysis

Abstract: In this thesis a semi-automatic method is developed to analyze morphometric features and landscape elements based on Self Organizing Map (SOM) as a unsupervised Artificial Neural Network algorithm. Analysis and parameterization of topography into simple and homogenous land elements (landform) can play an important role as basic information in planning processes and environmental modeling. Landforms and land cover are the main components of landscapes. Landscapes are dynamic systems that involve interrelation between physical characteristics (such as landform, soil) and anthropogenic processes (such as land use). In morphometry (as general term of geomorphometry) - the qualitative and quantitative measurement of topography - morphometric parameters are calculated such as profile curvature and longitudinal curvature. They are then used in morphometric analysis to identify morphometric features like plane, channel, ridge, peak or pit. In February 2000 the Shuttle Radar Topography Mission (SRTM), collected data over 80% of the Earth's land surface, to derive a consistent digital elevation model (DEM) for allland areas between 60 degrees N and 56 degrees S latitude. This DEM with about 90 m grid spacing was used to generate morphometric parameters of first order (slope) and second order (minimum curvature, maximum curvatures and cross-sectional curvature) by fitting a bivariate quadratic surface. These surface curvatures are strongly related to landform features and geomorphological processes. The thesis starts with an overall introduction and literature review. Then two methods for morphometric analysis are compared: morphometric parameterization and feature extraction proposed by Wood (1996a), calculated with Geographic Information Systems (GIS) software and our method implemented with Self Organizing Map (SOM) as an nsupervised artificial neural networks paradigm. Finally in our method for landscape element analysis morphometric parameters and remotely sensed spectral data are combined. The emphasis is on morphologically homogeneous landscape elements characterized by similar slope and curvature conditions. SOM is used to reduce large multidimensional data sets to one output layer consisting of 20 map units. These map units are interpreted in terms of morphometric features, slope and land cover to identify and characterize landscape elements or geoecological units Both studies have demonstrated valuable methods for extraction of land information that can be used in geomorphologic applications and geoecosystem modeling. These methods allow important savings in field work and can be used as alternative to labor intensive manual methods. But results may depend on scale and quality of the DEM and the topographic situation; caution should be used in interpretation. Evaluation of these methods in other areas with different morphometric conditions and with multi-scale DEM remains to be done.