Land Cover Classification and Feature
Extraction from National Agriculture Imagery
Program (NAIP) Orthoimagery: A Review
Aaron E. Maxwell, Timothy A. Warner, Brian C. Vanderbilt, and Christopher A. Ramezan
Abstract
This review describes the National Agriculture Imagery
Program (
NAIP
), explores strengths and weaknesses of the
data, and summarizes how the data are used in land cover
and feature extraction tasks in order to provide some recom-
mendations for future research and best practices for work-
ing with
NAIP
data.
NAIP
orthoimagery is an often-overlooked
source for remote sensing classification and feature extrac-
tion applications in the contiguous United States (
CONUS
).
NAIP
data are free, or nearly free; are in the public domain;
are available for all of the
CONUS
; comprise a multitemporal
data set that spans more than a decade; and are collected
at a high spatial resolution with generally very low cloud
coverage. However, there are challenges associated with the
use of these data. The low spectral resolution limits spectral
differentiation, while the high spatial resolution results in
very large data sets for study areas even as small as a single
county. Differences in acquisition dates and time of day
result in varying illumination conditions and potentially
even varying phenological state. As a consequence, image
digital number (
DN
) values can vary between adjacent tiles,
and shadow size and direction can be inconsistent between
different tiles and different acquisitions. Therefore, tak-
ing full advantage of this valuable data source requires the
analyst to be cognizant of such concerns and take measures
to deal with such inconsistency and minimize its impact on
the classification results; future research addressing these
concerns would further enhance the value of
NAIP
data.
Introduction
With the growing interest in high spatial resolution remote
sensing, National Agriculture Imagery Program (
NAIP
) ortho-
imagery, which is available across the contiguous United
States (
CONUS
), is a potentially important source of data for
a wide range of land cover mapping applications. Although
there is increasing use of
NAIP
imagery,
NAIP
data do not seem
to be used as widely as one might expect, given that these
data are free, or available at only a nominal cost, and have
high spatial resolution. Additionally, repeat images over more
than a decade are available. On the other hand,
NAIP
data are
acquired from aerial platforms using a variety of sensors, and
thus the data have complexities that are not always encoun-
tered in satellite-acquired imagery. This paper therefore
provides an introduction to
NAIP
data, focusing on the chang-
ing characteristics of the imagery, the applications for which
analysts have found
NAIP
orthophotography useful, and some
of the characteristics that make these data different from other
aerial- and satellite-based imagery datasets.
The selection of remote sensing data is not a trivial task.
Analysts must weigh a wide variety of factors, including data
availability and volume, as well as the necessary spatial,
spectral, radiometric, and temporal resolutions. Additionally,
analysts must consider the needs of the project, the geograph-
ic extent of the study area, the classes to be mapped, and the
required spatial resolution and accuracy of the classification
(Warner
et al
., 2009). A key question centers on data cost, es-
pecially since the price of image data can vary greatly. Much
of the exponential growth in land cover mapping research, at
least as measured by number of publications (Yu
et al
., 2014),
is driven by the increasing availability of large quantities of
low cost or free remotely sensed data, such as Landsat (Wood-
cock
et al
., 2008; Hansen
et al
., 2011). Widely available and
with low or no cost,
NAIP
data could also have a large impact
on high spatial resolution remote sensing, especially if the
challenges in using the data are addressed.
One of the difficulties associated with using
NAIP
data is
that information on data characteristics, and how the data
characteristics have changed over time, can be difficult to
obtain. Unlike other programs, such as Landsat and
MO-
DIS
, which have been extensively documented in a range of
review articles covering both the programs themselves (for
example, Loveland and Dwyer, 2012), and applications that
use data from these sensors (for example, García-Mora
et
al
., 2012), available summary information on
NAIP
data is
more limited in scope, and much of it is in reports that have
not been peer-reviewed. This fact is particularly important,
because in contrast to satellite programs, which nominally
generate consistent data over extended periods of time,
NAIP
data characteristics have evolved over time. Providing a
single, consolidated source for information on
NAIP
is a major
purpose of this paper.
Despite the challenges of using
NAIP
data,
NAIP
data can
be very effective for high spatial resolution mapping. Indeed,
this review was motivated by the authors’ successes in using
NAIP
data for land cover classification. Specifically, we have
had success using these data for mapping land cover associ-
ated with surface coal mining in West Virginia (for example,
Maxwell
et al
., 2014 and Maxwell and Warner, 2015), map-
ping palustrine wetlands (Maxwell
et al
., 2016), and mapping
Aaron E. Maxwell, Timothy A. Warner, and Christopher A.
Ramezan are with West Virginia University, Department
of Geology and Geography, West Virginia University,
Morgantown, WV 26506-6300
.
Brian C. Vanderbilt is with the US Forest Service, Geospatial
Technology and Applications Center, 2222 W 2300 S, Salt
Lake City, UT 84119 and formerly with the Geospatial
Services Branch, USDA Farm Service Agency, Aerial
Photography Field Office.
Photogrammetric Engineering & Remote Sensing
Vol. 83, No. 11, November 2017, pp. 737–747.
0099-1112/17/737–747
© 2017 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.83.10.737
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
November 2017
737
