Presenter: Sophia Shuler – Geography, Spatial Data Science and Technology
Faculty Mentor(s): Lucas Silva
Session: (In-Person) Poster Presentation
Satellite based remote sensing is one of the most accessible methods for implementing large-scale terrestrial change detection. However, cloud cover contamination of images is a frequent barrier to the use of change detection algorithms, particularly in places where cloud cover is frequent, such as in tropical mountains. In this project, I offer a method for cloud detection that can improve the quality of satellite image time series in tropical regions. Using both a cloud mask and a cloud index, I detected clouds in a set of Landsat-5 TM and Landsat-7 ETM+ time series from a tropical montane forest in Oaxaca, Mexico to a higher degree of accuracy than would be achieved by using the cloud mask alone. This method was used in sequence with the Breaks For Additive Season and Trend (BFAST) method in order to detect forest disturbances. After using a cloud index threshold of 2.8, the percentage of clouds detected increased from 91.8% to 94.4%. Additionally, this method yielded a 161% increase in the number of forest disturbances detected by BFAST. These results are applicable to change detection projects in regions with frequent cloud cover, where accuracy is limited by the climate conditions.