hand, increased by more than 60,000 ha yr
−1
, while mixed for-
est cover remained relatively constant (gaining roughly 15,000
ha yr
−1
). In agreement with previous literature (Drummond
and Loveland, 2010; Napton
et al
., 2010), this forest cover
loss did not come as a result of expanded cultivated cropland,
which also decreased at an average rate of about −130,000 ha
yr
−1
from 1986-2006 (Figure 5d). The loss of forested land and
cultivated cropland were offset by large, monotonic increases
in Developed land (+118,000 ha yr
−1
), Shrub/Scrub (+120,000
ha yr
−1
), and Grassland/Herbaceous (+49,000 ha yr
−1
) between
1986 and 2006. However, since we were only able to estimate
accuracy for aggregated classes and were not able to estimate
accuracy of the 1986 or 1991 classifications, the areas re-
ported above are based solely on the land cover areas mapped
by
AASG
(Figure 5), which may be biased due to error in the
classifications (Olofsson
et al
., 2014).
Conclusions
Here, we provide a new, long-term land cover dataset in the
dynamic and rapidly-changing southeastern United States,
which is quite accurate and broadly consistent with the
NLCD
but over a much longer time period. Our new land cover
product was produced using a novel algorithm for temporal
extension of existing land cover classifications (Automatic
Adaptive Signature Generalization) that is robust to atmo-
spheric and phenological differences among images from
different dates and is relatively simple to automate over large
areas. Additionally, the
AASG
method maintains semantic con-
sistency in class definitions, which makes it particularly well-
suited to studies of long-term land cover and land use change.
In this work, we show that the
AASG
algorithm is well suited
for automated, large-scale application and, when paired with
a robust classifier, produces highly accurate land cover maps,
especially when considering its relative simplicity.
In future work, we intend to address several weaknesses
in our Southeast land cover product. First, while we demon-
strated both the consistency of our new land cover product
with the
NLCD
and the relatively high accuracy of a simplified
classification scheme for the 1996-2006 classifications, we
were not able to estimate accuracy for 1986 or 1991 classifica-
tions or for the more detailed Anderson Level 2 classification
scheme. While the overall accuracies and producer’s/user’s
accuracies of most classes remained relatively consistent from
year-to-year, the limitations of the current assessment have
important implications for the interpretation of the classified
maps, and a more thorough accuracy assessment therefore re-
mains a priority for future work. Second, we employed a very
simple spatial filtering technique that eliminates contiguous
clusters of pixels of a given class that form an area smaller
than a class-specific minimum mapping unit threshold.
While this filtering approach is consistent with those used
in previous large-scale land cover classifications (e.g., the
NLCD
), it may be possible to further improve the accuracy and
consistency of the dataset by using an advanced set of spatial
and temporal filters based on the posterior distribution of the
ensemble-based random forest classifier (e.g., Pouliot et
al
.
,
2014; Waldner et al
.
, 2017) or on logical rules that eliminate
unlikely land cover transitions (e.g., Franklin et al., 2015).
The work presented here will provide an unprecedented
moderate-resolution perspective on the biophysical, biogeo-
chemical, and ecological consequences of land cover change
in the Southeast. To date, land cover information over so
large an area has only been available at moderate resolution
over short time periods (e.g., the
NLCD
) or at coarse resolu-
tions over longer time periods (e.g., from
AVHRR
). Our initial
analyses of land cover change over the 1986-2011 period sug-
gest several promising areas for future research that could be
facilitated with our new land cover product. How has a shift
from deciduous hardwood to evergreen forest affected carbon
uptake in the Southeast? What are the consequences of rapid
urbanization on hydrological processes (e.g., storm flow) and
ecological processes (e.g., phenology and landscape connec-
tivity)? What social, political and demographic processes are
driving the relatively rapid decline in cultivated cropland in
the Southeast, and what are the consequences for food supply
chains, farm labor, and the regional energy balance? Given the
high likelihood that these historical trends will continue in
the near future, it is more important than ever to assess their
causes and consequences.
Acknowledgments
This study was partly supported by
NASA
grant NNX-
17AE69G. The authors thank Tim Warner (West Virginia Uni-
versity) and three anonymous reviewers for helpful comments
that improved the manuscript.
References
Beven, K.J., and M.J. Kirkby, 1979. A physically based, variable
contributing area model of basin hydrology,
Hydrological
Sciences Journal
, 24:43–69.
Birdsey, R., K. Pregitzer, and A. Lucier, 2006. Forest carbon
management in the United States: 1600-2100,
Journal of
Environmental Quality
, 35(4):1461–69.
Bonan, G.B., 2008. Forests and climate change: forcings, feedbacks,
and the climate benefits of forests,
Science
, 320:1444–49.
Bounoua, L., R. Defries, G.J. Collatz, P. Sellers, and H. Khan, 2002.
Effects of land cover conversion on surface climate,
Climatic
Change
, 52:29–64.
Breiman, L., 2001. Random forests,
Machine Learning
, 45:5–32.
Bronstert, A., D. Niehoff, and G. Bürger, 2002. Effects of climate and
land-use change on storm runoff generation: present knowledge
and modelling capabilities,
Hydrological Processes
, 16:509–29.
Canadell, J.G., C. Le Quéré, M.R. Raupach, C.B. Field, E.T. Buitenhuis,
P. Ciais, T.J. Conway, N.P. Gillett, R.A. Houghton, and G.
Marland, 2007. Contributions to accelerating atmospheric CO2
growth from economic activity, carbon intensity, and efficiency
of natural sinks,
Proceedings of the National Academy of
Sciences of the United States of America
, 104(47):18866–70.
Crist, E.P., and R.C. Cicone, 1984. A physically-based transformation of
Thematic Mapper data - The TM Tasseled Cap,
IEEE Transactions
on Geoscience and Remote Sensing
, GE- 22(3):256–63.
Dannenberg, M.P., C.R. Hakkenberg, and C. Song, 2016. Consistent
classification of Landsat time series with an improved automatic
adaptive signature generalization algorithm,
Remote Sensing
,
8(8):691.
Diffenbaugh, N.S., 2009. Influence of modern land cover on the
climate of the United States,
Climate Dynamics
, 33(7–8):945–58.
Drummond, M.A., and T.R. Loveland, 2010. Land-use pressure and
a transition to forest-cover loss in the eastern United States,
BioScience
, 60(4):286–98.
Ellis, E.C., K.K. Goldewijk, S. Siebert, D. Lightman, and N.
Ramankutty, 2010. Anthropogenic transformation of the biomes,
1700 to 2000,
Global Ecology and Biogeography
, 19(5):589–606.
Fischer, J., and D.B. Lindenmayer, 2007. Landscape modification
and habitat fragmentation: A synthesis,
Global Ecology and
Biogeography
, 16:265–80.
Foley, J.A., R. Defries, G.P. Asner, C. Barford, G. Bonan, S.R.
Carpenter, F.S. Chapin, M.T. Coe, G.C. Daily, H.K. Gibbs,
J.H. Helkowski, T. Holloway, E.A Howard, C.J. Kucharik, C.
Monfreda, J.A Patz, I.C. Prentice, N. Ramankutty, and P.K.
Snyder, 2005. Global consequences of land use,
Science
,
309:570–74.
566
September 2018
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
