TRANSCRIPT

Michelle Pasco: As populations expand and stressors on the environment increase, once pristine landscapes are transformed. These changes are very apparent in the widespread degradation or repurposing of wetlands seen around the country. 

Michelle Pasco: While advancements in technology have driven some of these negative changes, they have also enabled us to create tools for increasing our knowledge of the environment and how our actions affect it.

Michelle Pasco: Phase II of the North Carolina Ecological Forecasting project harnessed one of these technologies, utilizing remote sensing in order to explore and classify the land use types of the Albemarle - Pamlico Sound Estuary System.

Benjamin Charlem: Located in Southeastern Virginia and Northeastern North Carolina, the Albemarle - Pamlico Sounds Watershed is one of the largest estuary systems in the United States and   demonstrates some of the negative effects an increasing population can have on an estuarine environment.  Many of its 31,000 square miles have been transformed from forest and wetland to cropland and urban areas as human population has continued to increase -  36% in just the years between 1990 and 2010.  Today, the dominant land uses are 40% forest, 25% cropland, and 15% wetland.

Benjamin Charlem: Of particular concern are the many wetlands that filter water, stabilize shorelines, and provide protective habitats.  Phase I of the North Carolina Ecological Forecasting project demonstrated that wetland health is deteriorating and suggested that wetlands are becoming less effective at performing these ecosystem services.  With rising sea levels and increasing nutrient runoff from agriculture, it's more important than ever to revitalize our wetland regions. 

Ben Roberts-Pierel: To best protect these wetlands, we must first understand their role in the surrounding environment. 

Ben Roberts-Pierel: We collected data publically provided by the federal government. 

Ben Roberts-Pierel: NOAA's C-CAP produces land classifications of coastal regions at 30 meter resolution every 5 years through an unsupervised classification methodology. 

Ben Roberts-Pierel: Cropland data is created annually by the USDA using remote sensing combined with extensive ground-truthing.

Ben Roberts-Pierel: SSURGO, also provided by the USDA, is assembled from field sampling. 

Ben Roberts-Pierel: Finally, we assembled two Landsat 8 scenes from June 23rd and July 25th, which served as the basis for our classification. 

Michelle Pasco: The bands on the Landsat scenes were composited in several ways to best visualize separate types of landcover.

Michelle Pasco: Using the different data sources, we outlined areas on the Landsat scenes in ArcGIS that clearly represented a specific landcover classification. 

Michelle Pasco: These areas were used as training site inputs to run a random forest supervised classification tool in Google Earth Engine. 

Michelle Pasco: We then checked our results by conducting an accuracy assessment within Erdas Imagine.  The accuracy assessment tool generated hundreds of stratified random points, which we attributed by hand using our GIS data layers and Google earth. Next, these referenced locations were input back into ERDAS Imagine, which compared the class values we assigned to those it had produced in the original classification. The assessment outputs a series of statistics including percent accuracy and a kappa coefficient. 

Ben Roberts-Pierel: With the resulting classification we are able to make comparisons with the original C-CAP dataset and highlight areas where our output differed from C-CAP.  

Ben Roberts-Pierel: As our primary focus was on wetlands, we were able to show areas where wetland type or extent was mis-qualified in the C-CAP dataset. 

Ben Roberts-Pierel: The hope is that these results can be used going forward to better focus protection and restoration efforts on wetland types most threatened by stressors found throughout this basin.