Uncertainty of Forested Wetland Maps
Derived from Aerial Photography
Stephen P. Prisley, Jeffery A. Turner, Mark J. Brown, Erik Schilling, and Samuel G. Lambert
Abstract
Forested wetlands (
FWs
) are economically and environmen-
tally important, so monitoring of change is done using remote
sensing by several U.S. federal programs. To better under-
stand classification and delineation uncertainties in
FW
maps,
we assessed agreement between National Wetlands Inventory
maps based on aerial photography and field determinations
at over 16 000 Forest Inventory and Analysis plots. Analy-
ses included evaluation of temporal differences and spatial
uncertainty in plot locations and wetland boundaries. User’s
accuracy for the wetlands map was 90% for
FW
and 68%
for nonforested wetlands. High levels of false negatives were
observed, with less than 40% of field-identified wetland plots
mapped as such. Epsilon band analysis indicated that if
delineation of
FW
boundaries in the southeastern U.S. met the
data quality standards (5 meters), then the area within uncer-
tainty bounds accounts for 15% to 30% of estimated
FW
area.
Introduction
Forested wetlands (
FWs
) are ecologically and economically
important. They support exceptional biodiversity, flood stor-
age, high water quality, and can produce valued products in
a sustainable manner (Loehle
et al.
2009; Richardson 1994;
Walbridge 1993). Therefore, understanding changes in the
area, distribution, and functional status of
FWs
is crucial.
Federal agencies in the U.S. monitor land cover change (in-
cluding
FWs
) with different remote sensing systems and areas
of emphasis. Three such programs are the Multi-Resolution
Land Cover Characteristics Consortium (
MRLC
), the National
Resources Inventory (
NRI
), and the National Wetlands Invento-
ry (
NWI
). The
MRLC
produces periodic lan
Land Cover Data base [
NLCD
]) for the con
30 m resolution satellite imagery and inc
category of “woody wetland”. The
NRI
program of the Natural
Resources Conservation Service publishes periodic reports
on land-based natural resources on nonfederal lands using a
sample of aerial photo plots. The
NWI
program of the U.S. Fish
and Wildlife Service (
USFWS
) produces both a complete inven-
tory (polygon map) of wetlands in the conterminous U.S. and
periodic Status and Trends (
S&T
) reports documenting changes
observed on photointerpreted sample blocks. The wetlands
map produced by
NWI
is often considered the standard source
of information about wetlands extent, due to the extensive cov-
erage in the United States and the program focus on wetlands.
Mapping
FWs
with remote sensing poses significant challeng-
es. The National Research Council (
NRC
) noted that mapping of
FWs
is difficult due to foliage obscuring the ground and, because
for most of the year, the water table is below the ground surface
(
NRC
1995). This statement is echoed by Tiner (1997), who
reported that
FWs
are “conservatively mapped” due to these dif-
ficulties, and that temporarily flooded or seasonally saturated
FWs
may not appear on
NWI
maps especially along the Coastal
Plain, where many of the
FWs
in the southern U.S. are located.
Given the difficulties in mapping
FWs
, it is important to
assess and report accuracy and reliability of change detection.
Accuracy assessment, reported quantitatively in standard
formats, is needed to provide evidence that reported differ-
ences are beyond those that might be expected from sampling
error, photointerpretation (classification) error, or boundary
(delineation) uncertainty.
Despite the importance of wetlands information, there are
few current, large-area, quantitative assessments of wetlands
map accuracy. In many of the published studies, comparisons
are made between wetlands delineated according to different
definitions (e.g., Stolt and Baker 1995; Morrissey and Sweeney
2006; Gage, Cooper, and Lichvar 2020). Furthermore, many stud-
ies are focused on a narrow geographic scope (Stolt and Baker
1995; Kudray and Gale 2000; Morrissey and Sweeney 2006).
Nichols (1994) examined
NWI
accuracy in Maine using
point intercept transects involving a total of 1800 sample
points, reporting an overall accuracy of 95%. Nearly all of
the wetlands identified in the field that were omitted in the
NWI
map were
FWs
. Kudray and Gale (2000), in a study in a
national forest in Michigan involving 148 field plots, reported
all nonforested wetlands (
NFWs
) were identified correctly, as
were slightly over 90% of
FWs
. In a 20 000-ha area in Vermont,
Morrissey and Sweeney (2006) found that
NWI
underestimated
FW
area by 64%, and overall wetland area by 39%.
ponents crucial to understanding the
n map product such as
NWI
: (1) the ac-
ation of individual polygons and (2) the
uncertainty of areas derived from polygon delineation. We
sought to address these components through a comprehensive
analysis of uncertainty in
FW
maps using an extensive valida-
tion data set. Specifically, our objectives were to: (1a) charac-
terize the level of classification agreement between an aerial
photo-based map of
FWs
and a field sample; (1b) assess the im-
pact of spatial uncertainties and temporal mismatch on clas-
sification agreement; (1c) examine patterns of disagreement
across forest conditions such as forest type, stand age, and
physiographic class; and (2) develop estimates of delineation
uncertainty for photo-delineated wetland polygons (Figure 1).
We addressed classification agreement using data from
NWI
as the remote sensing source, and data from the U.S. Forest
Service Forest Inventory and Analysis (
FIA
) program as a field
validation data set. To assess delineation uncertainty, we con-
ducted an epsilon band analysis using the
NWI
polygons.
Stephen P. Prisley is with the National Council for Air and
Stream Improvement, 541 Washington Ave. SW, Roanoke, VA
24016
.
Jeffery A. Turner, Mark J. Brown, and Samuel G. Lambert are
with the U.S. Forest Service Southern Research Station, 4700
Old Kingston Pike, Knoxville, TN 37919.
Erik Schilling is with NCASI, 104 East Bruce St., Aubrey, TX
76227.
Photogrammetric Engineering & Remote Sensing
Vol. 86, No. 10, October 2020, pp. 609–617.
0099-1112/20/609–617
© 2020 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.86.10.609
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
October 2020
609
