Master's Theses

Date of Award


Document Type



Earth and Atmospheric Sciences

First Advisor

Kyle C. McDonald

Second Advisor

Reiner Zimmermann

Third Advisor

Benjamin Black


White Sand Vegetation, SAR, classification


White sand vegetation communities are wide spread across South America; found in Peru, Venezuela, Brazilian Amazon and Guyana. They are distributed in patches ranging from <1 km2 to greater than tens of square kilometers and their origins and locations are still not well understood. The communities are related to a variety of factors (soil type, flooding, nutrient content and fire); hence a precise definition for the ecosystem is still not fully defined. Nevertheless, the result of these variations creates a unique environment for endemic plant and animal species to thrive. Furthermore, analysis of these areas has been very scattered and identification of local white sand areas (<1 km2) have not been accomplished. In addition, identification of these locations has currently only used optical satellite imagery (Landsat, MODIS). Hence, in this project, we have attempted to use synthetic aperture radar to create a classification system to locate the white sand vegetation systems. The goal is to be able to apply this method to identify white sand vegetation distribution across South America. The region of focus for this thesis has been in Aracá, a large white sand area located in Brazil in the State of Amazonas. Due to the lack of ground reference data, a classified map by Capurucho et al. (2013), generated using Landsat data, was used as a comparison and reference. JAXA’s ALOS-1 PALSAR (L-band), ESA’s Sentinel-1A (C-band) and NASA’s SRTM sensors were used for land classification. As microwave signals penetrate clouds and haze, the advantage of using sensors with this wavelength allows for an unobstructed coverage of the landscape all year round. Different combinations of polarizations and wavelengths were used during the analysis to try and separate the white sand vegetation from water and terra firme forest. The resulting classification images showed a 30% agreement with the classification map by Capurucho et al. It is important to note, that this number is in fact an agreement percentage as the map used was a classification image and coarse in resolution (due to the lack of reference data). Therefore, this value does not imply a bad classification. Future work will include time-series data, precise ground reference points and data from other sensors such as ALOS-2 PALSAR, to improve the classification accuracy.

Available for download on Friday, August 25, 2017