Identification of Barchan Dune Locations from Satellite Imagery Using Deep Learning Methods in Kuwait

Keywords: Kuwait, Deep Learning, Barchan Dune, Hazard
Abstract Type: Virtual Poster Abstract

Abstract

Deep Learning methods are now available that allow the identification of features within Satellite Imagery based on well-defined sets of training chips or samples. With the advent of Cloud storage of satellite data with pixel sizes of 10 meters of less, we now have multi-year and, in some cases, decadal time series of Satellite imagery available for analysis. In the recent past (< 10 years ago), it would take months to gather and prepare imagery for use before we could begin the analysis process for change detection. Deep Learning tools and modern image analysis methods are now available and integrated into Commercial software, making it available for use by the non-technical community. In this analysis ArcGIS Pro's Image Extension and Deep Learning module are used to identify barchan dune locations within the country of Kuwait and in the future will be used to measure their migration rate and direction of movement. Initial assessments based on data from 2019-2022 indicated the dunes in northcentral Kuwait are migrating southeast (155 degrees) at the rate of approximately 14 meters per year. In this study the area of interest was a 320 Km2 portion of the Al’Huwaimliyah dune field. When analyzed with the model developed here, >95% of the barchan dunes used as training samples were successfully identified.