Data Management Plan (Deprecated)

The NOAA Coastal Change Analysis Program (C-CAP) produces national standardized land cover and change products for the coastal regions of the U.S. C-CAP products inventory coastal intertidal areas, wetlands, and adjacent uplands with the goal of monitoring changes in these habitats. The timeframe for this metadata is summer 2016. These maps are developed utilizing high resolution National Agriculture Imagery Program (NAIP) imagery, and can be used to track changes in the landscape through time. The C-CAP program will use addtional image sources such as leaf off or tide controlled low tide imagery if it is available. This trend information gives important feedback to managers on the success or failure of management policies and programs and aid in developing a scientific understanding of the Earth system and its response to natural and human-induced changes. This understanding allows for the prediction of impacts due to these changes and the assessment of their cumulative effects, helping coastal resource managers make more informed regional decisions. NOAA C-CAP is a contributing member to the Multi-Resolution Land Characteristics consortium and C-CAP products are included as the coastal expression of land cover within the National Land Cover Database.

Process Steps:

• 2020-08-01 00:00:00 - This dataset was created by NOAA's Ocean Service, Office for Coastal Management (OCM). Random Forest Classification: The initial 1m spatial resolution 6 class high resolution land cover product was developed using a Geographic Object-Based Image Analysis (GEOBIA) processing framework. This involves taking each image to be classified and grouping the pixels based on spectral and spatial properties into regions of homogeneity called objects. The resulting objects are the primary units for analysis. Additionally, these objects introduce additional spectral, shape, textural and contextual information into the mapping process and are utilized as independent variables in a supervised classification. Each object is labeled using a Random Forest Classifier which is ensemble version of a Decision Tree. Training data for the initial 6 classes (Herbaceous, Bare, Impervious, Water, Forest and Shrub) were generated through photo interpretation. The resulting Random Forest model was applied to the input data sets to create the initial automated map. Impervious Surface refinement: To create an impervious surface class with more spatial detail, 2012 impervious surface data (http://cteco.uconn.edu/projects/ms4/impervious2012.htm) from the UConn Center for Land Use Education and Research (CLEAR) was incorporated into the land cover map. Updates from the random forest classification were isolated and retained through a series of spectral and contextual threshold models. Forest refinement: Forest features from the initial classification were refined using 2016 lidar data (http://cteco.uconn.edu/data/flight2016/index.htm) collected through The Capitol Region Council of Governments. Point clouds were processed into normalized digital surface models to clean up the boundaries between forest, grass and shrub features. Water Refinement: Commission and omission errors associated with the Water class were addressed through manual interpretation and clean up in ERDAS Imagine. Once the review and edits were complete, the refined Water class was incorporated back into the land cover data set.
• 2020-08-01 00:00:00 - Water Refinement: Commission and omission errors associated with the Water class were addressed through manual interpretation and clean up in ERDAS Imagine. Once the review and edits were complete, the refined Water class was incorporated back into the land cover data set. Unconsolidated Shore: Unconsolidated substrate features were extracted primarily through unsupervised classification and manual editing in ERDAS Imagine. This process relied on tide contolled imagery collected by National Geodetic Survey (NGS). Once these features were inserted into the land cover additional object based algorithms were applied to clean up the results. Agriculture: Cultivated land and Pasture/Hay features were mapped from the grassland and bare categories using a Convolutional Neural Network (CNN). The CNN was training using existing high resolution C-CAP data and NAIP imagery available in the region. Predictions made by the CNN were post-processed and inserted into the land cover which was finalized through internal QA and manual editing. Wetlands: Wetlands were derived through a modeling process which used ancillary data such as Soils (SSURGO), the National Wetlands Inventory (NWI) and topographic derivatives. Forest, shrub and grassland objects within the initial land cover that exhibited hydric characteristics based on the input ancillary layers were designated to their appropriate wetland category. The process relied mainly on the NWI to determine palustrine and estuarine distinctions. Open Space Developed: Managed grasses and other low lying vegetation associated with development were derived from the grassland category of the draft 2016 land cover through intersections with ancillary vector-based land use data as well as analysis of impervious surface coverage within those land use polygons. Herbaceous land cover areas that intersected select features from the OpenStreetMap landuse and points of interest layers were converted to Open Space Developed. Additionally, parcel features from CoreLogic parcel data were used along with relative area and adjacency thresholds of the impervious surface class to further map the Open Space Developed class. Manual edits were performed to fine tune the classifications.