The Development and Evaluation of a
High-Resolution Above Ground Biomass Product
for the Commonwealth of Puerto Rico (2000)
John S. Iiames, Joseph B. Riegel, Kristin M. Foley, and Ross S. Lunetta
Abstract
This study details the development of a U.S. Commonwealth
of Puerto Rico above-ground forest biomass (
AGB
) product
(baseline 2000) developed by the United States Environmental
Protection Agency (
EPA
) that was compared to another
AGB
product developed by the U.S. Forest Service (
USFS
) for the
same area. The
USEPA
product tended to over-predict in areas
of low biomass and under-predict in high biomass areas when
compared to observed plot data, but compared favorably to
a Forest Inventory Analysis (
FIA
) assessment of structure and
condition of Puerto Rico forests (72.6 Mg/ha versus 80.0 Mg/
ha, respectively).
AGB
estimates were highly correlated with
reference
FIA
biomass for both maps at their native spatial res-
olutions (
USEPA
: r =0.93,
USFS
: r = 0.92).
AGB
mean difference
between both products was 33.5 Mg/ha (
USFS
mean = 106.1
Mg/ha;
USEPA
mean = 72.6 Mg/ha), a difference not out-of-
scope when compared to other biomass comparative studies.
Introduction
The dependence on coastal resources (i.e., fisheries, tourism,
and pharmaceutical products from natural marine resources)
on the island of Puerto Rico (
PR
), has been threatened by mu-
nicipal and agricultural growth (Abuyuan, 1999). This growth
has led to declining quality and availability of drinking water
and increased sediment and nutrient runoff that adversely
affects coastal seagrasses, mangroves, and coral reefs (Warne
et al
., 2005). The United States Environmental Protection
Agency (
USEPA
) and University of South Florida are explor-
ing the relationship between landscape composition and
near-shore turbidity and chlorophyll-a concentrations across
southern
PR
. Both turbidity and chlorophyll-
a
estimates were
developed from the Moderate Resolution Imaging Spectrora-
diometer (
MODIS
) imagery data (Aqua) for 2002 to 2015. The
research approach included both near-term spatially explicit
(watershed-based) analysis for agricultural, urban and rural
watersheds; and long-term temporal correlation analysis for
individual watersheds (2001 to 2015). Landscape composition
correlates included numerous landscape metrics selected to
represent watershed erosion and nutrient loading potential.
One metric, above-ground biomass (
AGB
), allows for the
assessment of the mitigating effect on the reduction of nutri-
ent and sediment loadings during flood pulse disturbances
(Bayley and Guimond, 2008).
AGB
estimates developed under
this research will provide landscape metrics at the watershed,
sub-watershed, and riparian buffer analysis scales. Here we
document the construction of the
USEPA AGB
product (2000),
then compare this product to an
AGB
product created by the
United States Forest Service (
USFS
) for the Commonwealth of
PR
(2000). Both the
USEPA
and
USFS
biomass products incorpo-
rated similar data fusion techniques regressing satellite and/or
satellite-derived (i.e. ‘estimated’) data with
in situ
AGB
esti-
mates. The techniques were dissimilar with respect to the esti-
mation methods implemented and the primary predictor vari-
ables used. The
USFS
method utilized meteorological data and
MODIS
optical data as the primary predictor variables whereas
the
USEPA
method incorporated shuttle radar data. Also, the
spatial scale of the input variables and output predictions (15
versus 250 m) differed between both datasets. The requirement
for a finer resolution
USEPA AGB
product was predicated on the
nutrient flow modeling of sub-watersheds and riparian buffers,
typically narrow in width to exclude the use of the coarse 250
m
AGB
data. The quantitative comparisons described in this
paper only provide descriptors of differences and do not offer
accuracy validation for either of the two products.
Numerous analysis techniques have been used to map
AGB
on regional and global scales (Brown
et al
., 1989; Lefsky
et al
.,
2002 and 2005; Basuki
et al
., 2009). Mapping efforts at these
spatial scales primarily involved correlating remote sensing
data products and ancillary data, with
in situ
above-ground
biomass data (Zolkos
et al
., 2013; Marks
et al
., 2014). Each
data type exhibits strengths and weaknesses in estimating in-
formation about the vertical and horizontal structure existent
within a forested area (Cartus
et al
., 2014). For example, Lef-
sky
et al
. (2002) and Treuhaft
et al
. (2009) mapped
AGB
using
multi-source lidar and radar taking advantage of these sensors
ability to estimate vertical tree canopy structure. Discrepan-
cies in
AGB
mapping can be attributed to a number of factors
including: (a) use of differing allometric equations to acquire
reference
AGB
(Baccini and Asner, 2013); (b) differing scales
and types of geospatial predictor variables (Blackard
et al
.,
2008); and (c) disagreement of predicted land-cover propor-
tions from one technique to the other (Pan
et al
., 2011; Harris
et al
., 2012). Houghton
et al
. (2001) found that different
AGB
estimation techniques can produce
AGB
estimates that differ
by more than two fold within the same region. Understanding
the error components within each process is necessary when
comparing
AGB
products. For example, Asner
et al
. (2011)
noted that
in situ
carbon estimation alone explained 20 to 30
percent of the error in the estimation of
AGB
in Hawaii.
The objectives of this study were to (a) first develop a
USEPA
model utilizing shuttle radar data for predicting a
John S. Iiames, Kristin M. Foley, and Ross S. Lunetta are with
the U.S. Environmental Protection Agency, National Exposure
Research Laboratory, 109 T.W. Alexander Dr., Research
Triangle Park, North Carolina 27711 (
).
Joseph B. Riegel is an Independent Contractor to U.S.
Environmental Protection Agency, National Exposure
Research Laboratory, 109 T.W. Alexander Dr., Research
Triangle Park, North Carolina 27711.
Photogrammetric Engineering & Remote Sensing
Vol. 83, No. 4, April 2017, pp. 293–306.
0099-1112/17/293–306
© 2017 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.83.4.293
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
April 2017
293
