A Remote Sensing Approach to Identifying Coastal Ecosystem Change in the Laurentian Great Lakes Basin from 1985 – 2010.

Abstract:

Over the past several decades, rapid changes in the Laurentian Great Lakes region have impacted ecosystems at the land-water interface. Coastal areas around the Great Lakes range from high density urban areas such as Detroit, Milwaukee, and Buffalo to managed agricultural systems in the south and relatively natural forested landscapes to the north. Increased urbanization, changes in agricultural management, invasion of exotic wetland plants, and lake level volatility have contributed to degraded coastal habitats, loss of property values, and decreased recreational opportunities. A large interdisciplinary effort to integrate linked hydrology and wetland ecosystem models with remote sensing data has sought to quantify historical and likely future impacts of such changes on coastal ecosystem structure. A major piece of this effort is to assess land cover changes, especially in wetlands, using remotely sensed data.

Land cover maps, which include detailed wetland classes, were developed using a combination of multi-spectral and synthetic aperture radar data. These maps leveraged machine learning algorithms and a robust set of field validated wetlands data (>3000 locations) to provide a snapshot of wetland extent in the entire Great Lakes Basin circa 2010. The maps include invasive Phragmites australis and Typha spp., which have spread significantly in previous decades, altering habitat quality and impacting nutrient cycling in the region.

A hybrid change detection approach was utilized to assess changes in the Great Lakes Basin from 1985 – 2010. Categorical information from NOAA’s Coastal Change Analysis Program (C-CAP) land cover product was integrated with radiometric information from the Landsat archive within Google Earth Engine. The hybrid change approach was designed to reduce the amount of detected false change by defining real change as an area where there is both a categorical and spectral change sustained over multiple years. The implementation of this approach resulted in the class-to-class change observed in addition to the year that change was identified, providing a detailed understanding of both where and when ecosystems experienced transition. These products are being used as inputs to process-based models that will project potential future scenarios in the Great Lakes region.