New and improved C-CAP land cover


The National Land Cover Database (NLCD), produced by the United States Geological Survey (USGS), and NOAA’s very own Coastal Change Analysis Program (C-CAP) land cover data have had a long-standing and cooperative relationship.  Starting with our 2001 land cover, through the 2010 update, we have provided our coastal data to the USGS for incorporation into the NLCD.  For the production of the 2016 land cover, things changed.  The USGS undertook a major renovation in their processing approach, resulting in the production of a nationwide 2016 land cover prior to C-CAP completing the coastal portions.  Additionally, the USGS re-mapped their 2001, 2006, and 2010 data sets to incorporate improvements to the whole time series.

With a 2016 NLCD land cover completed for the coastal areas, and new versions of all existing NLCD dates, we were faced with a few challenges to bring the C-CAP data in-line with the new NLCD.

  • Keep the best improvements from the new NCLD mapping process
  • Ignore any errors from the new NLCD mapping process
  • Keep the additional wetland categories specific to C-CAP
  • Update the 2016 NLCD to create a 2016 C-CAP update
  • Incorporate all these changes and re-release all C-CAP data holdings

Our first step was to compare all the 2010 land cover products (NLCD original, NLCD new, C-CAP).  The NLCD and C-CAP products were visually, and geographically, compared against the base Landsat imagery.  The purpose of this step was to identify changes made to NLCD 2010 and compare these products to the existing 2010 C-CAP.    For example, in South Carolina, the original NLCD and C-CAP 2010 products were ~85% similar.  The USGS remapping resulted in over 25% of the area in South Carolina changing. We needed to examine these differences and determine what to keep, change, or ignore.

Improvements in the new NLCD

The new NLCD has a number of improvements, including:

  • Captured water (rivers, little ponds) more consistently
  • Better, more complete, Developed classes
  • Modeled forest transitions well (grass to shrub to forest)
  • Captured forest burns and insect damage well
  • Cleaned-up the mapping of Cultivated and Pasture
  • Removed speckled land cover
Image showing California forest burn area
NLCD refinements did a good job at capturing the loss of forest from fires.  These images show current 2010 C-CAP land cover, representative Landsat TM imagery, and new 2010 NLCD.  C-CAP had much of the area captured as Evergreen Forest (dark green), while NLCD has Grassland (beige).
NLCD_cultivated_cleanup
Changes to existing NLCD resulted in more complete capture of cultivated (brown) and pasture (tan) fields. Revised NLCD (left) compared to Original NLCD (right).

Issues in the new NLCD

Although the new NLCD has a number of advantages, there are several issues too.

  • Occasional seam lines and blunders in land cover
  • Some incorrect flipping of forest type (Evergreen to Deciduous)
  • Topographic shading causing some false water in areas
  • Small/rural roads often captured as Open Space Developed
  • Incorporate additional C-CAP wetland classes
  • Missing wetland loss in areas
LA_marsh_loss3
Areas of salt marsh loss were not adequately mapped in the new NLCD map.  The image on the left shows the C-CAP data with more water (dark blue) and less marsh (purple) compared to the NLCD on the right.  The middle image is the corresponding 2010 Landsat TM image.
NH_water_in_shadows
2010 Landsat TM image on the left for an area in New Hampshire and the corresponding new NLCD map with Water shown as red. Much of the mapped water should actually be Evergreen forest.  The steep terrain, combined with the dark spectral signature of Evergreen, resulted in false Water classification.

The changes we saw in the new NLCD products, both good and bad, can be tied to the greater automation in their mapping process.  While this automation introduced some errors, it also has some significant advantages, including more consistency, lower cost, and faster production times.  Overall, we feel the trade-offs resulted in better starting NLCD products, which we could then tweak for C-CAP.

A multistep modeling process was developed to try and keep the best changes from the NLCD product, improve the above identified issues, and tweak other minor issues with the existing C-CAP product.  A combination of logic models, object-based image analysis, spectral analysis, and finally hand-edits were performed to create a new 2010 C-CAP product.

Updating the older, creating the newer

With the new 2010 C-CAP completed, we turned our focus to our older dates of land cover and the update to 2016.  Going back in time, we were able to use existing land cover change masks from both the NLCD and C-CAP product lines.  We also had corresponding land cover calls from both of these data sets.  A series of logic models were created to assign final land cover classes, very similar to the approach used to create the 2010 product.

The process to create the 2016 C-CAP product was slightly different in that there was only one date of land cover and land cover change to rely upon.  Most of the steps were quite similar: isolated 2011-2016 NLCD land cover change and use logic, modeling, and editing to incorporate into a 2016 C-CAP product. We incorporated additional wetland categories through the use of ancillary data and existing C-CAP calls.

We have reviewed the results of our effort qualitatively through the comparison of before and after maps, comparing against source imagery, and are in the process of performing an accuracy assessment.  Overall, we feel that the C-CAP product line has improved overall, as can be seen in examples below, and initial statistical checks have shown some improvements in accuracy.

pre_post_CCAP_comparison
Three land cover maps, showing the new NLCD (left), original C-CAP (middle), and new C-CAP (right).  This area highlights many of the changes between Cultivated and Pasture, speckle removal, and highlights the differences in Open Space Developed (gold).
pre_post_CCAP_comparison2

Data Access

Now that you know about the changes and improvements we have made to ALL our C-CAP data holdings, you need to know where to get it!  We have two primary means to access the data, as nationwide files, or seamless custom areas.

The goal of C-CAP has been to produce consistent and accurate land cover data products for the coastal portion of the US on a 5-year cycle.  The current update resulted in not only a 2016 land cover product, but improvements to the full C-CAP data holdings as well.

One comment

Leave a Reply. Comments are moderated.

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.