Part 1 of a 2 part analysis by Scott A. Drzyzga and Daniel McCormack


The National Land Cover Database (NLCD) has been a reliable source of consistent and spatially-explicit land cover information for more than a decade. These data are derived from LANDSAT satellite images and provide GIS professionals with a seamless classification scheme with 20 classes at 30-meter resolution (Homer et al., 2014, USGS, 2014a). Other data layers associated with NLCD program include Tree Canopy (USGS, 2014b) and Developed Imperviousness (USGS, 2014c).

Figure 1 shows the portion of the 2011-vintage NLCD that covers Dover Township and Borough, in York County, Pennsylvania. We chose this area because one of our undergraduate student analysts, Daniel McCormack (who just graduated in December), grew up in Dover and wanted to study this landscape.

New High Resolution Data

The University of Vermont Spatial Analysis Laboratory (2016) released a new High-Resolution Land Cover (HRLC) dataset for the portions of Pennsylvania that overlap the Chesapeake Bay and Delaware River Watersheds. The primary sources used to derive these new data were 2006-2008 leaf-off LiDAR data, 2005-2008 leaf-off orthoimagery, and 2013 leaf-on orthoimagery. Ancillary data for roads and hydrology were used to augment the mapping process. These data provide GIS professionals with a seamless classification scheme with 12 classes at an incredible 1-meter resolution.

We began comparing the 2013 HRLC against the 2011 NLCD. Table 1 highlights both classification schemes. The NLCD uses a hierarchical two-digit classification scheme, which can be collapsed to the first digit, and an 8-bit pixel depth to accommodate the range of digital numbers. The HRLC uses a simpler coded value scheme and a 4-bit pixel depth, which mitigates file size.

Figure 1: 2011-vintage National Land Cover Data for Dover Township & Borough. The 30-m pixel footprints are apparent in the zoomed area.

Figure 2: 2013-vintage National Land Cover Data for Dover Township & Borough. The 1-m pixel size is not apparent in the zoomed area.

The NLCD scheme includes multiple forest and wetland categories as well as some information about developed and agricultural land uses. It does not contain any explicit information about roads, buildings, or other impervious surfaces, which often co-occur within the footprint of a 30-meter pixel and, so, are represented together as developed land.

The HRLC is a land cover dataset and not a land use dataset. Accordingly, the HRLC scheme does not include classes indicating industrial or agricultural land uses. For example, notice how all the areas mapped as ‘Cultivated crop’ (brown) or ‘Pasture /hay’ (yellow) in the NLCD (Figure 1) have been mapped as ‘Low vegetation” (lime green) in the HRLC (Figure 2). Instead, the HRLC dataset contains valuable information about built structures, roads, and other impervious surfaces. Also indicated are structures, roads, or other impervious surfaces that are covered by tree canopy. These dual codes {10,11,12} provide the GIS analyst with the ability to re-classify and use these pixels to support different kinds of analyses, which might require an on-the-ground land cover perspective or a higher bird’s-eye perspective.

Table 1: The NLCD and HRLC classification schemes.

NLCD 2011* HRLC 2013
Code Class Code Class
11 Open water 1 Water
21 Developed, open space 2 Wetlands (emergent)
22 Developed, low intensity 3 Tree Canopy
23 Developed, med intensity 4 Scrub-Shrub
24 Developed, high intensity 5 Low Vegetation
31 Barren 6 Barren
41 Forest, deciduous 7 Structures
42 Forest, evergreen 8 Other Impervious Surfaces
43 Forest, mixed 9 Roads
52 Shrub / scrub 10 Tree Canopy Over Structures
71 Grassland / herbaceous 11 Tree Canopy Over Other Impervious Surfaces
81 Pasture / hay 12 Tree Canopy Over Roads
82 Cultivated crops
90 Wetlands, woody
95 Wetlands, emergent herbaceous

* Five codes/classes that are limited to Alaska are not shown here.

Other than the classification scheme, the biggest difference between these two datasets is the spatial resolution. The NLCD is composed of 30-meter pixels and the HRLC is composed of 1-meter pixels, which means the HRLC has a data density that is 900 times greater than the NLCD. Again, the HRLC uses a 4-bit pixel depth, which mitigates the files sizes required to store so many pixels. The finer pixel size allows land cover boundaries to be mapped with greater precision. The finer pixel size also allows narrow roads or small patches of forest, water, or wetlands to be mapped explicitly, whereas they may be combined or confused within the much larger NLCD pixel.

Landscape patches

A ‘patch’ is a homogeneous area of land that differs from its surroundings. In a land cover dataset, a patch is represented by a pixel or group of adjacent pixels that share the same code/class and are surrounded by pixels with different codes. In this exploratory analysis, we observed the patch size distributions in Dover and calculated summary statistics for each dataset (Table 2). As expected, the smallest patch size in the NLCD is 900 sq.m (0.22 acres) and the smallest patch size in the HRLC is 1 sq.m. The largest patch size observed in both datasets is approximately 7.1 (1760 acres) large. The median patch size in the NLCD (1800 sq.m, or 2 pixels) is 300 times larger than the median patch size in the HRLC (6 sq.m, or 6 pixels, which reflects the different resolutions. Figure 3 illustrates both frequency distributions. In the NLCD, nearly 85% of all patches in Dover were between 900 sq.m (0.22 acres) and 10,000 sq.m (2.5 acres) large. In contrast, 82% of all patches in the HRLC layer were 100 sq.m (0.02 acres) or smaller. Many of the tiny patches are associated with the dual cover classes (i.e., tree canopy over something); they can be dissolved into adjacent patches after the analyst decides how to re-classify and use them.

Table 2: Summary patch size statistics

NLCD 2011 HRLC 2013
Statistic sq.meters acres sq.meters acres
Largest 7,103,884 1,755.42 7,153,198 1,767.61
Median 1,800 0.44 6 > 0.00
Smallest 900 0.22 1 > 0.00


Figure 3. Patch size frequency distributions. The dashed grey line represents 1 acre.


In this brief comparison, we learned that the University of Vermont Spatial Analysis Laboratory’s new High-Resolution Land Cover (HRLC) dataset can provide GIS professionals and others with land cover information that is much more spatially-precise than the NLCD dataset can provide. The abilities to interpret small patches of forest, individual buildings/structures, narrow roads, and other impervious surfaces are major improvements over NLCD. The trade-offs include larger file sizes and fewer categories. The HRLC data can, however, add value to the typical parcel or zoning layer, which are attributed with land use codes. Our next comparative analysis, part 2, will take a look at how the HRLC land cover classes compare with the NLCD land use/land cover classes. Do they agree?

References cited

Homer, C.G., Dewitz, J.A., Yang, L., Jin, S., Danielson, P., Xian, G., Coulston, J., Herold, N.D., Wickham, J.D., and Megown, K. 2015. Completion of the 2011 National Land Cover Database for the conterminous United States-Representing a decade of land cover change information. Photogrammetric Engineering and Remote Sensing 81(5): p345-354.

US Geological Survey. 2014a. National Land Cover Database 2011. Available via the Multi-Resolution Land Characteristics Consortium. Last accessed on January 10, 2017 at

US Geological Survey. 2014b. USFS Tree Canopy. Available via the Multi-Resolution Land Characteristics Consortium. Last accessed on January 10, 2017 at

US Geological Survey. 2014c. Percent Developed Imperviousness. Available via the Multi-Resolution Land Characteristics Consortium. Last accessed on January 10, 2017 at

University of Vermont Spatial Analysis Laboratory. 2016. High-Resolution Land Cover, Commonwealth of Pennsylvania, Chesapeake Bay Watershed and Delaware River Basin, 2013. Available via Pennsylvania Spatial Data Access. Last accessed on January 10, 2017 at