Land Cover and Wetlands of the Gulf of Maine Watershed
go to: USFWS Gulf of Maine Watershed Habitat Analysis
Arnold Banner, August 2002

Land cover from five interpretations of Landsat data, and wetland cover from photo-interpretations were combined to yield a 31-class raster (grid) digital coverage, "GOMLC7", for the Gulf of Maine watershed. This is projected as UTM zone 19, NAD83, dimensions in meters, and has a 30m cell size. The following narrative explains how the coverage was developed. Coverage attributes and their designations are shown in Table 5.

Upland Cover

SOURCES (In order of relative suitability of the classes, and/or apparent accuracy):

1. NOAA CCAP scenes; NH/ME was based on 1993 imagery, Penobscot Bay on 1995 imagery, Massachusetts on 1997 imagery, and Cobscook Bay on 1992 imagery.

2. University of Maine GAPS landcover; this used imagery obtained between 1990 and 1993; covers all of Maine.

3. EPA-MRLC; this covers all of New England

4. New Hampshire GAPS landcover; this was based on scenes from 1992 through 1995, covering all of New Hampshire and Vermont.

5. Southern New England (SNE) GAPS landcover (sneveg4n; files dated 1/2/99).


Initially I projected all of these to UTM zone 19, NAD27 to match our other data sets. Additionally, I made spatial corrections by shifting source 2 by 40 m and source 3 by 50 m to the south, and "warped" source 5 to better fit 1:24000 USGS road network coverages.  The final grids were 'shifted' to a NAD83 projection.


Photo-interpretation has been shown to be relatively more accurate for wetland delineation than image processing. Therefore, information about the wetland cover classes was derived primarily from National Wetlands Inventory and other photo-interpreted digital data (see description below). The 5 landcover sources were used primarily for delineating upland types. Because the CCAP classification was relatively simple yet proved to be appropriate for habitat modeling in our southern Maine analysis, I reclassified or translated the upland classes (landcover types) of sources 2, 3, 4, and 5 to match those of CCAP. This typically involved aggregation of classes, which also tends to improve the classification accuracy. I combined the low and high density development classes of CCAP into a "developed" class, since the non-developed portion of a 30 m square pixel would not offer significantly more habitat value than an entirely developed pixel.

Translation of the University of Maine GAPS landcover to CCAP is given in Table 1; the MRLC translation is shown in Table 2; the New Hampshire GAPS to CCAP translation is shown in Table 3, and the SNE GAPS shown in Table 4.

Combination of Upland Data Sources:

The upland classes were selected from the above sources as follows:

CCAP data was used as the primary source. Where CCAP wasn't available secondary sources were developed. For Maine the following sequence was used; shrub/scrub was identified from Maine GAPS, then development from MRLC, then agriculture from GAPS, and the remaining classes were derived from MRLC.

The secondary source for New Hampshire and Massachusetts was MRLC, but NH GAPS and SNE GAPS were used to adjust MRLC wetland interpretations (see below).

Table 1. Relationships used for translating Maine GAPS upland cover types into CCAP classes


(High or
Low Density)



sparse resid.

dense resid.



crops/ground grassland

blueberry field

abandoned field

deciduous forest


coniferous forest


heavy partial

light partial cut

late regen.

early regen

alpine tundra



rock shore

Table 2. Relationships used for translating MRLC upland cover types into CCAP classes


(High or Low Density)


low intensity residential

high intensity residential

commercial/ transportation


small grains

bare soil



row crops

other grasses

deciduous forest evergreen forest mixed forest deciduous shrubland

evergreen shrubland

mixed shrubland


bare rock/sand


Table 3. Relationships used for translating NH GAPS upland cover types into CCAP classes

(High or
Low Density)


barren/urban nonforest (agriculture/oldfield/lawn) deciduous dominant coniferous dominant mixed forest

Table 4. Relationships used for translating SNE GAPS upland cover types into CCAP classes


(High or
Low Density)



agriculture grassland red maple dom.

birch dom.

oak dom.

oak/maple/birch codominant.n. hardwoods

conifer oak/maple/conifer codom. coastal scrub nonforest;

bare rock/sand

beach, mud flat

urban open land

Wetland Cover

Maine. Full NWI digital coverage was available and was the sole source.

Massachusetts. NWI digital coverage was available for 34 quads out of 64 full or partial quads needed. The missing quads were filled in with surrogate wetland information derived from aerial photo-interpreted digital data and re-interpreted into NWI classes. In order of relevancy and quality these were:

1) Massachusetts orthophoto wetlands (from MAGIS); attributes translated into NWI types.

2) USGS 1:24000 hydrology polygon features; coded from the respective features on NWI paper maps.

3) Town landuse information (from MAGIS); non-forested wetlands translated into NWI "PEM".

The sequence of development of the wetland coverage was: start with the wetlands from town landuse; update (overwrite) this with USGS polygon features interpreted as wetland vegetation; update this with orthophoto wetlands; then update with USGS water features. This tended to leave just features in which I had the highest level of confidence. This final coverage was merged with the NWI wetland coverages.

New Hampshire. Full NWI digital coverage became available during this study. and was the sole source for wetland mapping.

Combination of Upland and Wetland Data Sources

Because both Maine GAPS and MRLC utilized some NWI data in their landcovers, I reduced the extents of all those wetland types by expanding the adjacent uplands into them by one cell prior to overlaying my selection of NWI types. This improved the coincidence of upland and NWI boundaries and eliminated single pixels identified as wetlands by image processing, which did not correspond with the NWI interpretation.

After I overlaid the wetland data from NWI and other sources (see development of wetland data, above), I found that there were image-interpreted wetlands beyond those photo-identified in NWI, USGS hydrology and Massachusetts landcover. Where I had comprehensive wetlands cover from other sources I regarded these as having a high likelihood of being in error, and replaced them with upland classes interpreted from the alternate landcover sources.

In Maine, where I had complete NWI coverage, I substituted MRLC-interpreted uplands for CCAP wetlands extending beyond the NWI polygons; if MRLC also called these areas wetlands, I alternatively used the GAPS upland interpretation; if that also "confirmed" an area as being wetland, I retained the original CCAP designation.

For those areas of New Hampshire where I had digital NWI, I followed the above sequence, ending with NH GAPS upland and forest covers. Where I initially had only a less complete alternate or "surrogate" wetland cover, I retained the CCAP designation unless MRLC classified the area as upland.

For those areas of Massachusetts where I had digital NWI, I followed the above sequence, ending with SNEV upland and forest covers. Where had to use a less complete alternate or "surrogate" wetland cover, I retained the CCAP designation unless MRLC classified the area as upland.

Intertidal and Subtidal Wetlands

The inter- and sub- tidal designations of coastal (estuarine and marine) wetlands were updated using bathymetric information developed from a variety of sources (see gombathydoc.htm). The mean low water line based on bathymetry was used to "split" areas NWI had designated as intertidal and subtidal aquatic bed, and unconsolidated shore or bottom. For example, areas of NWI-mapped "E1UB" (estuarine subtidal unconsolidated bottom) found to be above mean low water were renamed "E2US" (estuarine intertidal unconsolidated shore). The following types were so adjusted:


All of the E2EM and rocky shore/bottom areas were found to be intertidal.

Estuarine and Freshwater Wetlands

The palustrine/riverine vs. estuarine wetland designations of near-coastal wetlands were updated using salinity information developed from a variety of sources (see gomsalinity83.htm).  Adjustments typically involved wetlands and head of tide regions along coastal rivers.

Table 5. Class Names and Designations of GOMLC7 Cover Types

bare ground open land, quarries, beach
cultivated active agriculture other than hay
developed residential, industrial, commercial, roads
grassland pasture, hay, lawns, old field
upl conif forest predominantly coniferous upland forest
upl decid forest predominantly deciduous upland forest
upl mixed forest mixed upland forest
upl scrub/shrub upland shrub, regenerating forest
E1AB estuarine subtidal aquatic vegetation
E1UB estuarine subtidal unconsolidated bottom
E2AB estuarine intertidal aquatic vegetation
E2EM estuarine intertidal emergent vegetation
E2RS_R1RS estuarine, riverine intertidal rocky shore
E2SS estuarine intertidal scrub/shrub
E2US_R1US estuarine, riverine intertidal unconsolidated shore
L1UB_PUB lake or pond
L2RS lake, rocky shore
L2US lake, unconsolidated shore
M1AB marine subtidal aquatic vegetation
M1UB marine subtidal unconsolidated bottom
M2AB marine intertidal aquatic vegetation
M2RS marine intertidal rocky shore
M2US marine intertidal unconsolidated shore
PAB_L2AB lake, fresh marsh, or pond, aquatic vegetation
PEM_L2EM lake, fresh marsh, or pond, emergent vegetation
PFOcon coniferous forested wetland
PFOdec deciduous forested wetland
PSScon coniferous shrub wetland
PSSdec deciduous shrub wetland
R1UB riverine tidal unconsolidated bottom
Rper riverine perennial


Accuracy of the upland cover types was tested by comparing the structure and vegetation type depicted in the image-interpreted coverage with actual conditions onsite. The examination was limited to southern Maine, with sites ranging up to 53 miles from Falmouth. Two field data sets were used: one set of 107 point descriptions and GPS coordinates had been developed for testing land cover data of the Casco Bay watershed, in 1994; another set of 117 were developed during the Fall of 2000 for this study. Data from the 1994 survey were compared to the relevant classes of the current land cover; inappropriate types or sites at the borders of different cover types were ignored. The recent site surveys including acquisition of one to 5 digital images of the site, and extensive field notes. Because the number of sites inspected for each of the 8 cover types was small, we combined the observations of the two data sets. This gave a sampling of 11 to 38 sites for each of the image cover types. The results are shown in the following table.


Cover of interpreted image

site conditions

grass bare developed agriculture decid. forest conif. forest mixed forest upland shrub proportion correct overall accuracy
grass 34 1 0 15 0 0 0 4 0.630 0.751
bare 0 12 0 0 0 0 0 1 0.923 0.837
developed 0 1 12 0 0 0 0 0 0.923 0.923
agriculture 3 0 0 2 0 0 0 0 0.400 0.259
decid. forest 0 0 1 0 18 0 0 6 0.720 0.810
conif. forest 0 0 0 0 0 14 0 3 0.824 0.912
mixed forest 0 1 0 0 1 0 11 1 0.786 0.851
upland shrub 2 1 0 0 1 0 1 23 0.821 0.713
proportion correct 0.872 0.750 0.923 0.118 0.900 1.000 0.917 0.605

Overall, the level of accuracy was regarded as satisfactory. Three of the 8 classes were under 80% accuracy: agriculture, shrub, and grassland. The agriculture class was confused only with grassland: about half of the agricultural sites (row crops) were interpreted as grasslands, and 88% of the sites interpreted as agriculture proved to be grasslands. There were two reasons for such errors. First, row crop stubble may have a "signature" similar to that of grasses. Second, crops are rotated and the vegetation at the time of the inspection may be different from that current when the Landsat image was collected. Also, hayfields often are manured, and this may make them appear to be a more intensively managed crop. Since agricultural fields were regarded as of minimal habitat value for most of our species, although they actually become or have the structure of grasslands after harvest, the consequences of misidentification of agricultural fields were minimal. On the other hand, about one third of the real grassland sites were called agriculture. Again, many of these are intensively managed at some periods, and are cut several times during the season, so assigning a reduced habitat value is not entirely erroneous. Moreover, we supplemented the image interpretation with of an independent "high value grassland" coverage to ensure that the preferred grassland habitats were not omitted.

Areas that actually were upland shrub were identified quite accurately, but the upland shrub class of the image included about one third forested types. That is, forested areas were frequently identified as shrub. This is not surprising since most of the upland shrub cover in this area is regenerating forest that has been cut over or thinned. Since the point at which regrowth should be reclassified from shrub to one of the forest types is not distinct, this level of confusion may need to be accepted. In our habitat models we usually treated shrub as 'open' forest, so this error has been accommodated to a degree.