Thematic Accuracy of Agriculture in
Land Cover Layers of Select Virginia Counties
Ioannis Kokkinidis, Steven C. Hodges, and Randolph H. Wynne
Abstract
NLCD
and
NASS
CDL
are freely available moderate resolution
land cover datasets, but their accuracies vary widely and
are untested for agricultural land in Virginia. We performed
validation through aerial photointerpretation of agricul-
ture at the field level, using cadastral parcels as proxies
for fields, over Albemarle, Charles City, Chesterfield, and
Henrico Counties for
NLCD
1992, 2001, 2006 and
CDL
2002,
2008, 2009, 2010, and 2011. The extent of agricultural land
remained generally stable over 19 years but is generally
overestimated by the datasets in all four counties, rang-
ing from 70.06 percent to 697.48 percent of validation layer
extent. Extent of agricultural land in our validation layer
also differed from the Census of Agriculture, likely due to
differing definition. Comparison of layer pairs on extent
of agricultural land mostly reveals classification artifacts
rather than change. The limited extent of agriculture and
mixed land cover characteristics of the region suggest the
use of multitemporal data to extract agricultural land cover.
Introduction
In this study we investigate accuracies of major and often used
land cover products applied to describe the locations and distri-
butions of agricultural land in four counties in eastern Virginia.
The distribution and extent of the different land covers are
important for the emerging science of ecosystem service assess-
ment as land cover has been used as an input for climate model-
ing (Wilson and Henderson-Sellers, 1985), natural resource
inventory (Anderson
et al
., 1976), spatial modeling (Pontius and
Schneider, 2001), land use/land cover change detection (Lam-
bin, 1997) and multiple other purposes. Furthermore, quantita-
tive data on the extent and location of agricultural land has been
used as an input in agricultural flow models to quantify food
and fiber provision (Santelmann, 2004), primary production
(Prince
et al
., 2001), carbon fixation, nutrient cycling (Boody
et
al
., 2005), soil erosion and deposition (Wei
et al
., 2008), water
quality (Johnson, 2012) and to show how change affects the
ability of agroecosystems to provide its various benefits (Boody
et al
., 2005). Since the outputs of models are dependent on their
inputs, it is very important to understand the uncertainty asso-
ciated with land cover datasets, especially in areas with diverse
and heterogeneous covers and landscapes.
Published studies dealing specifically with the quality of
agricultural land cover layers in this region are few and tend to
focus on large polygons such as states (Johnson, 2013), entire
counties (Goslee, 2011; Maxwell
et al
., 2008) or at best arbitrary
circular plots (Hollister
et al
., 2004) as units of analysis. These
are not spatial scales relevant to assessing the flow of land use
related ecosystem services at the field scale; hence a validation
study at the field scale is needed. We used several assessment
methods to assess the accuracy of the agricultural layer of
several of the highest resolution classifications available over
parts of the relatively humid and spatially heterogeneous Com-
monwealth of Virginia with the eventual aim of using them to
track changes over time in the extent of agriculture. Our assess-
ment is based on various validation methods: traditional point
to point validation, point to polygon and polygon to polygon
using the cadastral parcels as the polygon unit of analysis. We
explore the use for our unit of analysis of the relatively stable
polygon of cadastral parcels, proxies for agricultural fields,
while still allowing visual inspection of each polygon. Cadas-
tral parcels represent consistent boundaries across time, and
we hypothesize the degree of land use change will remain,
much like state and county boundaries, socio-politically stable.
Their resolution is finer than polygons used in previous stud-
ies. To the best of our knowledge, the use of cadastral parcels as
a unit to analyze land cover over time and assess its accuracy
has not been previously reported. For this reason we also eval-
uate the suitability of cadastral parcels in our study area for use
as aggregation units, attempting to understand the variability of
land use inherent within these units and if their use improves
the ability to study land use change or simply introduces more
complexity and errors into the validation process. Finally we
use our findings to evaluate the evolution of agricultural land
cover over time to track changes in its extent.
Study Area
Virginia, part of the Mid-Atlantic region, lies socially at the
intersection of the urbanized Northeast and the more rural
South and climatically in the transition zone between thermic
and mesic climate regions. It thus supports complex cov-
ers including evergreen, deciduous, annual and perennial
plants of both climatic zones, as well as complex mixtures
of natural, managed, and urban land uses (VA DCR, 2016).
Its population has been increasing (US Census Bureau 2014)
resulting in conversion of different land covers into urban
use. Furthermore, being in the humid East Coast of the United
States, it has been difficult to spectrally differentiate agri-
culture, which after all is a socially rather than spectrally
defined land cover, from other vegetated classes, as opposed
to irrigated crops in the desert (El-Magd and Tanton, 2003).
Ioannis Kokkinidis is with the City of Fresno, Information
Services Department, Geographic Information Systems
Division, 2600 Fresno Street, Room 1059, Fresno, CA, and
formerly with the Virginia Polytechnic Institute and State
University, Geospatial and Environmental Analysis Program
(
).
Steven C. Hodges is with the Virginia Polytechnic Institute
and State University, Crop and Soil Environmental Sciences
Department, 339 Smyth Hall, Blacksburg VA 24061.
Randolph H. Wynne is with the Virginia Polytechnic Institute
and State University, Forest Resources and Environmental
Conser-vation Department, 319 Cheatham Hall, Blacksburg,
VA 24061.
Photogrammetric Engineering & Remote Sensing
Vol. 83, No. 10, October 2017, pp. 679–692.
0099-1112/17/679–692
© 2017 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.83.10.679
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
October 2017
679
