Evaluating MODIS Land Products at a Tropical Moist Forest Site

Evaluating MODIS Land Products at a Tropical Moist Forest Site
David P. Turner1, Warren B. Cohen2, Stith T. Gower3, Marcos H. Costa4, William D. Ritts1, Thomas K. Maiersperger1, Maosheng Zhao5
2USDA
PARTICIPANTS
Principal Investigators
Other Participants
Warren B. Cohen
Marcos Heil Costa
Stith Tom Gower
David P. Turner
Steve W. Running
Mônica Carneiro Alves Xavier
Ricardo Guimarães Andrade
Thomas K. Maiersperger
Robert E. Kennedy
Alan Kirschbaum
Dulce Castleton
Maosheng Zhao
David Ritts
CONTEXT FOR
LBA-ECO CD-15
The overall goal of the BigFoot project is to validate a
variety of MODIS science products at nine sites covering
a range of biomes. The BigFoot campaign at the Tapajós
primary forest tower site (TAPA) is manifested in LBAECO as project CD-15.
1Department of Forest Science, Oregon State University, Corvallis OR 97331, [email protected]
PNW Research Station, Corvallis OR 97331, 3Department of Forest Ecology and Management, University of Wisconsin, Madison WI 53706,
4Department of Agricultural Engineering, Federal University of Viçosa, Av. P. H. Rolfs, s/n, Viçosa MG 36570-000, Brazil
5School of Forestry, University of Montana, Missoula MT 59812
APPROACH & OBJECTIVES
Each study site contains an eddy flux tower;
BigFoot provides spatial context for the flux data
by characterizing a greater footprint area (7x7 km).
• Make multi-year in situ measurements of NPP, LAI, FPAR
and land cover.
Our sampling scheme consists of
multiple transects. They provide for a
field-based ecological characterization
of heterogeneous areas within the site.
Multi-year measurements ensure that
the temporal validity of MODIS
products can be accurately assessed.
• Develop multi-year land cover, LAI, and FPAR surfaces by
linking in situ measurements to Landsat ETM+ data
Surfaces are generated using a combination of empirical methodologies
appropriate to each site. Errors in these surfaces are quantified and the
surfaces summarized to provide a map-based characterization of the site.
EARLY RESULTS
Land Cover / LAI / FPAR
EARLY RESULTS
GPP / NPP
Land Cover
Gross Primary Production
The MODIS land cover classification for
all 25 1-km2 cells in the study area is
“Evergreen Broadleaf”. The BigFoot
Landsat-based classification is the same.
Higher resolution analysis reveals roads
and a significant proportion of the study
area in shadows associated with
emergent trees.
The temporal trajectory of the MODIS GPP product
generally follows that of the MODIS FPAR product.
The absolute magnitude of the MODIS-based 8-day
mean GPP reaches 10 g C m-2 d-1. The GPP estimated
at the eddy covariance tower shows a more seasonal
temporal pattern but maximum values are of about the
same magnitude. Output of the Biome-BGC model (to
be used for scaling GPP in the BigFoot products) when
driven by meteorological data from the flux tower is
similar to the flux tower observations although without
as strong of a dry season reduction.
IKONOS 1 meter PanSharpened Color-IR Image
Landsat ETM+ 30 meter
Color-IR Composite
Estimates of maximum daytime light use efficiency for
gross primary production from the flux tower
measurements suggest a value (1.5 gC MJ-1) similar to
that used in the MODIS GPP algorithm (1.3 gC MJ-1).
Flux Tower (km67)
Comparisons of shortwave solar radiation (SW) and vapor pressure deficit (VPD) from the Data
Assimilation Office (DAO) climatology (used in the MODIS GPP algorithm) with observations
at the eddy covariance flux tower show similar trends for SW but less day-to-day variation in the
DAO estimates than the tower measurements for VPD.
Transects
BigFoot 30m Land Cover
www.fsl.orst.edu/larse/bigfoot
MODIS 1km Land Cover
ABSTRACT
The satellite-borne Moderate Resolution Imaging
Radiometer (MODIS) is operationally producing
estimates of land cover, leaf area index (LAI), 8-day gross
primary production (GPP), and annual net primary
production (NPP) for each 1 km2 of the Earth’s
terrestrial surface. Validation of these products requires a
combination of ground measurements, modeling, and
fine resolution remote sensing. The BigFoot Project is
carrying out validation studies using a consistent scaling
approach at 9 sites representing different biomes. The
tropical moist forest site is at Km 67 in the Tapajós
National Forest in Brazil, and ground measurements were
begun in June of 2003. Fifty plots were established over
a 5 km x 5 km area that included an eddy covariance flux
tower. Measurements of forest cover, LAI, litterfall, and
bolewood production are on-going. Flux tower
measurements of net ecosystem exchange are being used
to estimate daily GPP. The MODIS land cover
classification for the 25 km2 study area is Evergreen
Broadleaf Forest, which is in agreement with ground
observations. Mean dry season LAI from MODIS is
about 6.0. For 2002, the MODIS GPP in the area was
2,664 gC m-2 yr-1 which compares with 3,164 from the
flux tower. MODIS products for 2003 will be compared
to a comprehensive set of BigFoot products. The
BigFoot scaling approach permits analysis of the multiple
components of the MODIS GPP/NPP algorithm and
will be used in efforts to refine its parameterization.
• Estimate NPP/GPP in multiple years using land cover and
LAI surfaces, climate data, and ecosystem process models
Net Primary Production
The process models are calibrated at
each site and performance is assessed
by comparisons of modeled NPP
against in situ measurements of NPP.
Modeled GPP and water vapor are
compared to tower-based calculations
of these variables.
• Validate MODIS land cover, LAI, FPAR, and NPP/GPP
products
Validation involves direct spatial comparisons of MODIS surfaces to
BigFoot field data and derived surfaces over time.
Evergreen Broadleaf Forest
Road
The mean MODIS NPP in the study area is 580 gC m-2 y-1 and the ratio of NPP to GPP was
0.22. The NPP and NPP/GPP ratio appear to be low relative to other estimates for intact
forests in the region. The BigFoot NPP measurements are in progress.
5 kilometers
Leaf Area Index / FPAR
The spatial heterogeneity of the MODIS LAI and FPAR products is low
during the dry season but high during the wet season. Maximum LAI values
are on the order of 6 and maximum FPAR is close to one. The BigFoot LAI
and FPAR products are in preparation.
CONCLUSIONS
•
•
•
•
MODIS land cover classification is in agreement with BigFoot observations.
MODIS LAI/FPAR products are of approximately the right magnitude in the dry season but are
anomalously low and show artificial spatial heterogeneity in the wet season.
MODIS GPP estimates are low relative to flux tower observations in the wet season but similar in
the dry season.
MODIS NPP estimates appear to be low, related in part to a relatively low ratio of NPP to GPP.
ACKNOWLEDGEMENTS
The BigFoot Project is funded by the NASA Terrestrial Ecology Program. Special Thanks to Steve
Wofsy (Harvard University), Scott Saleska (Harvard University), and Scott Miller (UC, Irvine) for the
meteorological and GEE data from the Santarem flux towers.