Assessing the Phenological Suitability of Global Landsat Data Sets for Forest Change Analysis

1
The Global Land Cover Facility
Assessing the Phenological Suitability of Global
Landsat Data Sets for Forest Change Analysis
Kuan Song, Chengquan Huang, John R. G. Townshend,
Paul E. Davis, Saurabh Channan, Matthew Smith
Department of Geography Global Land Cover
Facility, University of Maryland
Summary Vegetation phenology is a major
consideration in acquiring satellite images for
forest change analysis, especially in regions
having deciduous forests. When trees have little
or no leaves during the leaf-off season,
spectrally deciduous forests are difficult to
separate from non-forested surfaces, including
disturbed forests. Therefore, use of images
acquired during or near the leaf-off season in
forest change analysis can result in substantial
errors in the derived change products. In this
study, we evaluate the phenological suitability
of two global Landsat data sets centered around
1990 and 2000 for forest change analysis. These
two data sets, together with a third one centered
around 2005, which is being assembled, will be
used to produce an Earth Science Data Record of
global forest cover change.
What does forest phenology look like from
satellite?
How to retrieve the phenology time series?
A series of forest phenology around Washington DC
in year 2000. Mostly this is deciduous forest
1. The phenology time series is shown as the
average NDVI value of the forest canopy in a
calendar year 2. The NDVI value is provided by
the AVHRR GIMMS data set The GIMMS (Global
Inventory Modeling and Mapping Studies) data set
is a normalized difference vegetation index
(NDVI) product available for a 25 year period
spanning from 1981 to 2006. The data set is
derived from imagery obtained from the Advanced
Very High Resolution Radiometer (AVHRR)
instrument onboard the NOAA satellite series 7,
9, 11, 14, 16 and 17. This is an NDVI dataset
that has been corrected for calibration, view
geometry, volcanic aerosols, and other effects
not related to vegetation change. 3. The
forested land is defined as gt30 vegetation
canopy field value in the MODIS MOD44B. Two
forest types are labeled Deciduous and
Evergreen. Shown below are the forest mask used
for the sites of California, Colorado, and DC at
similar latitudes 4. Phenology time series
is calculated preferably on the deciduous forest,
on the dates of GeoCover imagery set. Again for
the sites of California, Colorado, and DC at
similar latitudes
How does forest phenology vary from place to
place?
A series of forest phenology in Northern
California, WRS-2 tile 45/33. Mostly this is
evergreen forest
Bright Green Deciduous Forest Dark Green
Evergreen Forest Siena Non-forest Blue Water
A series of forest phenology in Colorado, WRS-2
tile 34/33. Deciduous Forest phenology shown here
Why do you need forest phenology information, if
you are not studying phenology in particular?
The optical property of forest canopy is such
that dense forests are the dark objects in
remote sensing. The darker the forest canopy, the
easier it is to distinguish forest from other
land cover types. This property is very useful
for automatic image analysis of any land cover
application. For this purpose we generated a
global archive of forest phenology profiles for
each Landsat footprints in the 1990s and 2000s.
How to interpret and use the phenology profile?
A Global Collection of Forest Phenology Profiles
for the Remote Sensing Community
The phenology profiles will be freely
downloadable at www.landcover.org
The growing season and satellite data collection
window
The three sites of Washington DC, Colorado,
and Northern California were deliberately chosen
at the same latitudes. Different growing seasons
show up at very different geographical
conditions. For the evergreen forest of
northern California on the west coast, satellite
data can be collected all year long for land
cover and change mapping. For the deciduous
forest of Washington DC on the east coast, decent
data can be acquired from April to October. And
for the deciduous forest in Colorado, the ideal
observation window is during May to September.
For high resolution satellite imagery,
researchers usually choose the cloud-free dates
for mapping. Less considered is the phenology
situation in the choice of imagery. With our
database, researchers can select leaf-on Landsat
scenes just like they select cloud-free scenes
from USGS. Our study shows that, the GeoCover
global Landsat image set offered by NASA is
excellent in terms of phenology. Most (gt90) of
the imagery were collected in the leaf-on
seasons. This makes the GeoCover imagery set
ideal for land cover mapping and change detection.


www.landcover.org
This project activity is sponsored by NASA
MEaSUREs and LCLUC programs.