Measuring urban tree canopy: NLCD maps, photo-interpreted points and i-Tree Canopy
Tree cover in a city comes from a satellite map, a random-point count on aerial photos, or both. The Forest Service found the 30 m national map ran 9.7 percent low on tree cover.
Checked against the sources at the bottom of this page on October 15, 2026. Rules, fees and pay change: the source has the last word.
Urban tree canopy is reported as the percent of an area covered by trees. The USGS National Land Cover Database (NLCD) maps it from satellite imagery at 30 m. The USDA Forest Service counts it by sampling random points on aerial photos. i-Tree Canopy is the tool that does that counting on a screen.
The NLCD tree canopy layer
The USGS record describes the 2001 tree canopy layer as a 30-meter data set for the conterminous United States, split into four tiles for GIS use. It came from the Multi-Resolution Land Characteristics (MRLC) Consortium. The members are federal agencies: USGS, NOAA, EPA, USDA, the Forest Service, the National Park Service, the Fish and Wildlife Service, the Bureau of Land Management and the Natural Resources Conservation Service.
The record says the country was cut into 68 mapping zones. Nowak and Greenfield, writing in Environmental Management in 2010, count 65 zones and describe the method: twelve mapping teams processed circa 2001 Landsat 7 imagery. Maps built from 1-m digital orthoimagery were used to develop a separate algorithm for each zone's canopy percentage.
Each layer ships with error estimates from cross-validation. In the preliminary figures Nowak and Greenfield quote, average error on tree canopy ranges from 6 to 17 percent, depending on the zone.
Checking the map against photo-interpreted points
In the same 2010 paper the Forest Service compared the map with photo-interpretation across all zones. Each zone got 1,000 random points, viewed in Google Earth and classified by trained interpreters. Four interpreters worked the zones, and another interpreter re-did 100 points per zone as a check. Of all points, 99.6 percent fell on imagery that could be read.
| Measure | Tree canopy | Impervious cover |
|---|---|---|
| Zones where NLCD ran low | 64 of 65 | 44 |
| National average underestimate | 9.7 percent (SE 1.0 percent) | 1.4 percent (SE 0.4 percent) |
| Largest underestimate | 28.4 percent, mapping zone 3 | 5.7 percent, mapping zone 56 |
An earlier comparison with sampled Census places and counties, cited in the same paper, put the tree canopy shortfall at 9.7 percent as well. Its impervious shortfall was 5.7 percent in places and 1.3 percent in counties.
Point sampling and standard error
The Forest Service method is simple. Lay random points on the city, classify each one, and divide the points that hit trees by the points that could be read. The standard error is the square root of p times (1 minus p) divided by n, where p is the cover share and n the number of points.
Nowak and Greenfield's 2012 paper on 20 U.S. cities used 1,000 points in 18 of them. That gives a maximum standard error of 1.6 percent when every point is classified. Baltimore got 2,500 points and Spokane 2,000. Every city had more than 97.2 percent of its points read, and a second interpreter re-checked a five percent sample, agreeing on 97 percent.
i-Tree Canopy
The i-Tree Canopy tool page describes random sample points over Google Maps aerial photography in a project area. The user picks a cover type for each point. The tool suggests 500 to 1,000 survey points. Its report gives percent cover and standard error for each class. The report credits the concept to David Nowak, Jeffery Walton and Eric Greenfield of the Forest Service, and the current version to The Davey Tree Expert Company.
A sample report from version 6.1, run on June 3, 2015, shows the output. Of 399 points, 20 were Tree, which is 5.01 percent plus or minus 1.09. The other 379 were Not a Tree, 95.0 percent plus or minus 1.09.
The report states its own limits. Accuracy depends on the user classifying each point correctly. More points shrink the standard error. If too few points are classified, the standard error is too high to give any real certainty.
Cover change between two dates
For change, the Forest Service pairs two aerial photos and interprets the same points on both. The 2012 paper reports that urban tree cover in the conterminous United States is declining by about 7,900 ha a year, or 4.0 million trees a year. Tree cover declines were statistically significant in 17 of the 20 cities. Only Syracuse, New York, had a statistically significant gain. The average city lost about 0.27 percent of its area in tree cover a year, while impervious cover grew about 0.31 percent a year.
The paper lists limits too. Image dates and resolution vary, though most city images were 1 m or finer. A point can seem to change class when a tall tree leans one way in one photo and another way in the next. Pairing the points lets the interpreter correct those false changes. The paper also warns that comparing two cover maps can invent change through misclassification. Its example is Seattle, where maps from 0.6 m imagery put tree cover at 22.5 percent in 2002 and 22.9 percent in 2007, with map accuracy unknown.
City-level cover also feeds tree inventories and the i-Tree suite.
Sources
- USGS, National Land Cover Database 2001 (NLCD01) tree canopy layer, Tile 2, Northeast United States, publication record dated January 1, 2010, read October 15, 2026
- USDA Forest Service, Nowak and Greenfield, Evaluating the national land cover database tree canopy and impervious cover estimates across the conterminous United States, Environmental Management 46(3), 2010, read October 15, 2026
- USDA Forest Service, Nowak and Greenfield, Tree and impervious cover change in U.S. cities, Urban Forestry and Urban Greening 11(1), 2012, read October 15, 2026
- i-Tree, Nowak and Greenfield 2012 paper as hosted by i-Tree, read October 15, 2026
- i-Tree, i-Tree Canopy tool page, read October 15, 2026
- i-Tree, i-Tree Canopy v6.1 cover assessment and tree benefits report, June 3, 2015, read October 15, 2026