Improving Land Cover Classification using Texture Patterns Derived from Micro-Scale Digital Elevation Models




Klier, John D.

Journal Title

Journal ISSN

Volume Title



<p>Historically the determination of land cover types has relied upon techniques that analyze reflected spectral energy in the visible or near infrared wavelengths. Often, these wavelength bands are used in combination to produce vegetation indices like the Normalized Difference Vegetation Index (NDVI).</p> <p>Data derived from sources such as Light Detection and Ranging (LiDAR) or Structure from Motion (SfM) have been primarily utilized for construction of elevation models and contour maps. The resolution of this data generally allows for detailed reconstruction of the subject area terrain where one can see objects such as rivers, streams, buildings, stands of trees, etc.</p> <p>In the past 10 years drone technology has become available to the general public and with it a researcher is able to gather high resolution data in a fixed flight path where the camera orientation for each photo in relation to the ground is known. This detail subsequently has allowed for the creation of elevation models that show detail in individual tree crowns. This dissertation uses drone technology to examine the role of tree canopy texture, derived from a hyperspatial digital elevation model, in identifying individual tree species. The research questions addressed in this dissertation include: (1) Are texture patterns derived from hyperspatial digital elevation models (DEM) of the tree canopy indicators of individual tree species? (2) What is the role of texture in determining species-level assemblages and/or individual tree entities? (3) Can texture alone match reflectance-based land-use/land-cover (LULC) detection methods in accuracy of classification? To answer these questions three classification techniques are compared for a mixed canopy environment in the Texas, USA Hill Country: (1) object based image analysis of drone-based canopy texture, (2) maximum likelihood classification of multispectral drone imagery, and (3) object-based image analysis of NDVI derived from National Agricultural Imagery Program aerial photography. Findings from this comparison suggest that an analysis of texture alone can match the results of multispectral image classification techniques. Findings lead to the conclusions that canopy texture is a key indicator of individual tree species and that hyperspatial DEMs adequately capture unique differences in tree species.</p>



Object-based analysis, Remote sensing


Klier, J. (2022). <i>Improving land cover classification using texture patterns derived from micro-scale digital elevation models</i> (Unpublished dissertation). Texas State University, San Marcos, Texas.


Rights Holder

Rights License

Rights URI