Continuous Fields Tree Cover Project
Entry ID:
GLCF_TREE_COVER_AVHRR
|
Summary
From the Documents-Dataset Descriptions located at the following URL:
http://glcf.umiacs.umd.edu/data/treecover/description.shtml
Origin.
An important requirement for global models of the earth system is
reliable, geographically-referenced, consistent data on global
... vegetative cover. The only truly synoptic view of the earth is
provided by satellites and may improve the quality, consistency and
reproducibility of global land cover information. To this end this
project aims at providing an alternative to traditional land cover
classification by representing vegetative cover as a gradient over the
landscape using satellite data. Data acquired in 1992-93 from NOAA's
AVHRR at a 1km spatial resolution and processed under the guidance of
the International Geosphere Biosphere Programme (IGBP) (Eidenshink and
Faudeen 1994) were used to derive the tree cover, leaf type and leaf
longevity maps.
Description.
Characterization of terrestrial vegetation from the Advanced Very High
Resolution Radiometer (AVHRR) on the global to regional scale has
traditionally been accomplished using classification schemes with
discrete numbers of vegetation classes. Representation of vegetation
into a limited number of homogeneous classes does not account for the
variability within land cover, nor does the portrayal recognize
transition zones between adjacent cover types. An alternative paradigm
to describing land cover as discrete classes is to represent land
cover as continuous fields of vegetation characteristics using a
linear mixture model approach. This prototype data set contains 1km
cells estimating:
1. Percent tree cover;
2. Percentage cover for two layers representing leaf longevity
(evergreen and deciduous); and
3. Percentage cover for two layers estimating leaf type (broadleaf and
needleleaf).
Each pixel in the layers has a value between 10 and 80 percent. These
layers can be directly used as parameters in models or aggregated into
more conventional land cover maps. For the latter, the product offers
the flexibility to derive land cover maps based on the user's
requirements for a particular application. The product is intended for
use in terrestrial carbon cycle models, in conjunction with other
spatial data sets such as climate and soil type, to obtain more
consistent and reliable estimates of carbon stocks.
The aim of this research is to 1) develop methodologies for global
representation of vegetation characteristics and 2) produce continuous
fields of vegetation characteristics at 1km which are accessible to
the global change research community.
Legend and Values.
For:
treecover
---------------------------------------------------------------
10 - 80 percent tree cover
254 non-vegetated
255 tree cover less than 10%
For:
evergreen; deciduous; broadleaf; and needleleaf
---------------------------------------------------------------
10 - 80 percent cover for indicated leaf longevity and type
(% evergreen + % deciduous = % tree cover; and
% broadleaf + % needleleaf = % tree cover)
-Band interleaving: Band sequential
-Mask: Sea mask applied=255
Quantization:
8-bit unsigned integer
Downloadable file formats:
UNIX compressed (".gz" for UNIX) also works with winzip for Windows.
Geographic Coverage
Spatial coordinates
| N: 90.0 |
|
S: -90.0 |
|
E: 180.0 |
|
W: -180.0 |
Data Set Citation
Dataset Creator:
DeFries, R., Hansen, M., Townshend, J.R.G. and Sholberg, R.
Dataset Title:
Global land cover classifications at 8km spatial resolution: The use of training data derived from Landsat imagery in decision tree classifiers
Dataset Series Name:
International Journal of Remote Sensing
Dataset Release Date:
1998
Issue Identification:
Volume 19
Online Resource:
http://glcf.umiacs.umd.edu/data/treecover/index.shtml
Dataset Creator:
DeFries, R.S., Hansen, M., Townshend, J.R.G., Janetos, A.C. and Loveland, T.R.
Dataset Title:
A New Global 1km Data Set of Percent Tree Cover Derived form Remote Sensing
Dataset Series Name:
Global Change Biology
Issue Identification:
In Press
Online Resource:
http://glcf.umiacs.umd.edu/data/treecover/index.shtml
Dataset Creator:
Hansen, M., DeFries, R., Townshend, J.R.G. and Sholberg, R.
Dataset Title:
Global cover classifications at 1km resolution using a decision tree classifier
Dataset Series Name:
International Journal of Remote Sensing
Issue Identification:
In Press
Online Resource:
http://glcf.umiacs.umd.edu/data/treecover/index.shtml
|
Temporal Coverage
Start Date:
1992-01-01
Stop Date:
1993-12-31
|
Location Keywords
Data Resolution
Latitude Resolution:
1 km
Longitude Resolution:
1 km
Horizontal Resolution Range:
1 km - < 10 km or approximately .01 degree - < .09 degree
|
Science Keywords
ISO Topic Category
Platform
Instrument
Project
Quality
Combining a land cover map based on International Geosphere Biosphere
Programme IGBP land cover definitions (Hansen et al. in press) with
percentage cover estimations of trees, leaf longevity and leaf type
(DeFries et al. in press) produced the layers provided here. The land
cover map was joined with the ... mixture model results to bind the
minimum and maximum value defined for each cover type to the IGBP
definitions of that cover type. For example, if all pixels classified
as woodland according to Hansen et al. (in press) in a particular
continent ranged in percent woody cover from 30 to 70 percent
according to DeFries et al. (in press), the values would be scaled to
within the 40 to 60 percent range as defined as woodland. Values
obtained for leaf type and leaf longevity in DeFries et al. (in press)
were adjusted so that the total percentage summed to the adjusted
percent woody value. Furthermore pixels classified as agriculture in
the tropics were adjusted to range from 10 to 25 percent tree cover
because of the known! ! presence of mixed farming with an
overstory. Both approaches use the same IGBP processed data set from
the AVHRR in 1992-1993. The merging of the data sets was performed to:
1. Overcome the difficulty in distinguishing percentage tree cover,
leaf longevity and leaf type at extreme high and low percent
cover. The difficulty in determining high and low values is likely due
to cloud contamination in humid forests and saturation of the spectral
signature at high percent cover and the overwhelming influence of soil
and understory background on the spectral signature at low percentage
cover. Therefore the layers of tree cover, leaf type and leaf
longevity have values ranging from 10 to 80 percent, with a value of
80 percent cover being equal or greater than 80 percent and a value of
10 being equal or less than 10 percent cover.
2. Constrain the observed overestimation of tree cover in locations
known to have intermediate values of percent canopy coverage
characteristics of woodland (defined by the IGBP as 40 to 60 percent
canopy cover). Proportions of trees, leaf type and leaf longevity were
(Defries et al. in press) adjusted so that the percent woody coverage
in each pixel is within the range of percent canopy coverage defined
by the classification result (Hansen et al. in press).
3. Adjust the range of agricultural pixels form 10 to 25 percent cover
because of known mixed farming with an overstory.
The two data sets merged in this study were developed using different
methods to describe vegetation. A brief explanation of each of the
methods used to derive the data sets is given below:
1. Global maps of proportional cover for three vegetation
characteristics, leaf form (percent woody vegetation, percent
herbaceous vegetation and percent bare ground cover), leaf type
(percent needleleaf and percent broadleaf) and leaf longevity (percent
deciduous and percent evergreen). The procedure for deriving the
continuous fields of vegetation characteristics is fully explained in
DeFries et al. (in press) and utilizes a linear mixture model approach
applied to 1km AVHRR data. A set of 156 Landsat Multispectral Scanner
data were used to train the linear models for vegetation
characteristics permitting estimation of endmember values (DeFries et
al. 1998). The spectral response of the AVHRR data is then unmixed
using the endmembers and estimates of leaf longevity (percent
evergreen and percent deciduous), leaf type (percent broadleaf and
percent needleleaf) and percent tree cover are identified. A separate
model was developed for each continent to determine the mixtur! ! es
of broadleaf evergreen, broadleaf deciduous, needleleaf evergreen, and
needleleaf deciduous woody vegetation depending on which forest types
are present in each continent. The approach is based on the annual
phenological cycle of vegetation derived from 30 metrics acquired from
the AVHRR. These metrics are the annual maximum, minimum, mean and
amplitude for the annual time series of the Normalized Difference
Vegetation Index (NDVI), and channels 1 through 5 of the AVHRR. The 24
metrics were calculated from April to April (1992-93) to account for
the full growing season in both hemispheres, but only the eight months
with the highest NDVI are used to describe green vegetation. Six
metrics based on surface temperature were also derived from channel 4
of the AVHRR to account for snow cover at higher latitudes. Linear
discriminates or linear combinations of the weighted metrics were then
made to reduce the statistical complexity and error associated with
using 30 metrics in the linear model. The resulting data set thus
represents a percentage map where each cell is composed of between 10%
and 80% of the respective vegetation characteristic.
2. A global land cover classification (Hansen et al. in press)
containing 12 cover types based on requirements identified by the
IGBP. The classification methodology used to make this map is
described in Hansen et al. (in press) and DeFries et
al. (1998). Briefly approximately 150 Landsat scenes were interpreted
with ancillary data and consultation with experts on land cover to
obtain training data for the classification method. A decision tree
algorithm was then employed using 41 metrics derived from the annual
temporal profile of the NDVI and the five individual bands of the
AVHRR to obtain a classification of cover types.
Access Constraints
None
Use Constraints
Please use the following references to cite the data:
DeFries, R.S., Hansen, M., Townshend, J.R.G., Janetos, A.C. and
Loveland, T.R. (in press) A New Global 1km Data Set of Percent Tree
Cover Derived form Remote Sensing, Global Change Biology,
Hansen, M., DeFries, R., Townshend, J.R.G. and ... Sholberg, R. (in press)
Global cover classifications at 1km resolution using a decision tree
classifier, International Journal of Remote Sensing,
DeFries, R., Hansen, M., Townshend, J.R.G. and Sholberg, R. (1998)
Global land cover classifications at 8km spatial resolution: The use
of training data derived from Landsat imagery in decision tree
classifiers, International Journal of Remote Sensing, 19, 3141-3168.
Ancillary Keywords
Data Set Progress
Originating Center
Data Center
Distribution
Distribution Media:
Online (HTTP)
Fees:
None
|
Personnel
RUTH
S.
DEFRIES
Role:
INVESTIGATOR
Phone:
301-405-4884
Fax:
301-314-9299
Email:
rd63 at umail.umd.edu
Contact Address:
Laboratory for Global Remote Sensing Studies
Department of Geography
2181 LeFrak Hall
University of Maryland at College Park
City:
College Park
Province or State:
MD
Postal Code:
20742-8225
Country:
USA
GENE
R.
MAJOR
Role:
DIF AUTHOR
Email:
gsfc-gcmduso at mail.nasa.gov
JOHN
R. G.
TOWNSHEND
Role:
INVESTIGATOR
Phone:
301-405-4050
Fax:
301-314-9299
Email:
jrtownsh at geog.umd.edu
Contact Address:
Laboratory for Global Remote Sensing Studies
Department of Geography
2181 LeFrak Hall
University of Maryland at College Park
City:
College Park
Province or State:
MD
Postal Code:
20742-8225
Country:
USA
|
Related URL
Link:
GET DATA
Description:
This is the location from which the Goodes Projection Tree Cover
Data as BSQ Binary Files may be downloaded.
|
Publications/References
DeFries, R.S., Hansen, M., Townshend, J.R.G., Janetos, A.C. and
Loveland, T.R. (in press) A New Global 1km Data Set of Percent Tree
Cover Derived form Remote Sensing, Global Change Biology,
Hansen, M., DeFries, R., Townshend, J.R.G. and Sholberg, R. (in press)
Global cover classifications ... at 1km resolution using a decision tree
classifier, International Journal of Remote Sensing,
DeFries, R., Hansen, M., Townshend, J.R.G. and Sholberg, R. (1998)
Global land cover classifications at 8km spatial resolution: The use
of training data derived from Landsat imagery in decision tree
classifiers, International Journal of Remote Sensing, 19, 3141-3168.
Creation and Review Dates
DIF Creation Date:
2000-05-25
Last DIF Revision Date:
2008-08-14
|
|
Links
