SEMI-AUTOMATIC CRESTLINE MAPPING OF SMALL BEDFORMS ON HIGH RESOLUTION SATELLITE IMAGES

Features with small aerial extent on satellite images are hard to map for detailed analysis. This is more significant for bedforms and features that occur in large numbers and huge fields. In addition, increased application of high resolution satellite or Unmanned Aerial Vehicle (UAV) images from Earth, as well as High Resolution Imaging Science Experiment (HiRISE) images from Mars, highlights the demand for developing automatic image-processing techniques in order to facilitate and speed up image analysis, along with the improvement of the accuracy of the results. This study introduces an applicable and simple framework based on an unsupervised Artificial Neural Network (ANN) algorithm called Self-Organizing Maps (SOM) for mapping features with small footprints.

SOM translates information relationships of high dimensional input data to a two dimensional output grid in what is called the map. In this study, high resolution satellite images have been used as the primary input data and other layers were extracted from these images depending on the properties of their bands and the characteristics of the features throughout the study area. The study area for this research is a unique mega-ripple field in Iran which hosts millions of mega-ripples with various spatial patterns and crest morphologies. Mega-ripples in this area are of great interest to planetary scientists because of their similar horizontal length scale to mysterious aeolian bedforms on Mars known as Transverse Aeolian Ridges (TARs).

Millions of the mega-ripple features were automatically mapped by this technique and the resulting crestline map was highly precise based on an accuracy assessment of the results. The introduced methodology with its associated high degree of accuracy can save a lot of time and should aid quantitative studies in diverse Earth and planetary science projects. This technique could benefit different environmental management issues such as model validation and change detection studies based on multi-temporal images for evaluating climate change effects, aeolian studies and many other purposes.