Remotely sensed satellite imagery contains an abundance of spectral data that can be used to monitor terrestrial landscape change at continental to global scales. To detect landscape change from spectral data a large quantity of imagery is needed to create image composites free of spectral anomalies such as clouds and cloud shadows. Image compositing is a process that reduces a stack of imagery to a single image that is filled with median or medoid values on a pixel-wise basis, effectively reducing image size and export time. Our research objective was to automate a workflow that efficiently collects large quantities of Landsat 8 and Sentinel-2 imagery, masks spectral anomalies from collected imagery, and generates annual-image composites at a spatial extent that optimizes image size and export time. To test our workflow, Google Earth Engine was leveraged to collect 2018-2019 imagery for date range June 1st - August 31st from the continental United States.
Results/Conclusions:
Our automated workflow efficiently collected and processed enough Landsat 8 and Sentinel-2 imagery to create annual-image composites free of spectral anomalies. The majority of medoid pixels used to create composites were filled with 2019 imagery (i.e. > 99.0%), with some composites filled entirely with 2019 imagery. Maximizing the number of pixels filled with imagery from one year opposed to several years of imagery can augment accurate landscape change detection at annual-to monthly time steps. We determined the composite size that optimized export time was 23,040,000 ha, about the area of Minnesota. From start to finish our workflow exported composites in about hour. During the hour composite-medoid values were calculated on a pixel-wise basis from around 200 Landsat 8 and Sentinel-2 images, resulting in a composite size of about 2.4 GB. The workflow presented here significantly reduces the time needed to collect and process large quantities of remotely sensed imagery. We encourage those interested in using remotely sensed spectral data to detect terrestrial landscape change for large spatial areas at annual-to monthly time steps consider our big data solution.