bmng.pdf

(4609 KB) Pobierz
The Blue Marble Next Generation - A true color earth
dataset including seasonal dynamics from MODIS
Reto St¨ckli
∗
Eric Vermote
†
Nazmi Saleous
‡
o ,
,
,
Robert Simmon
§
and David Herring
June 28, 2005
Contents
1 Introduction
2 Methods
2.1 Data . . . . . . . . . . . . . . . .
2.2 Temporal snow-free correction . .
2.3 Temporal correction of snow . . .
2.4 Spatial gap filling . . . . . . . . .
2.5 Scaling to RGB and NDVI space
3 Results
4 Discussion
5 Dataset User’s manual
5.1 Map projection . . . .
5.2 Data format . . . . . .
5.3 Relief shading . . . . .
5.4 Caveats . . . . . . . .
5.5 Citation of the BMNG
6 Acknowledgements
3
4
4
5
6
6
6
6
8
8
8
8
9
10
11
11
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Corresponding author: Reto St¨ckli, IACETH, Winterthurerstrasse 190, 8057 Z¨rich,
o
u
Switzerland. e-mail: rstockli@climate.gsfc.nasa.gov, phone: +41 44 635 5209. Affiliations: NASA
Earth Observatory and ETH Institute for Atmospheric and Climate Science
†
Department of Geography, University of Maryland, College Park, MD 20742, USA
‡
Department of Geography, University of Maryland, College Park, MD 20742, USA
§
NASA Earth Observatory, Mail Code 613.2, Goddard Space Flight Center, Greenbelt, MD
20771, USA
NASA Earth Observatory, Mail Code 613.2, Goddard Space Flight Center, Greenbelt, MD
20771, USA
∗
1
Abstract
Satellite remote sensors provide the science community with high quality
datasets to better understand and monitor Earth’s environment and climate
system. When used in education, these datasets allow visual exploration of
the planet at moderately high spatial and temporal resolutions. However, na-
tive data file formats and file sizes are not well suited for public distribution.
To help circumvent these issues, the Blue Marble Next Generation (BMNG)
is a new series of 12 monthly cloud-free, global-scale images. We created the
BMNG using NASA Terra MODerate resolution Imaging Spectroradiometer
(MODIS) science data collected in 2004. We temporally adjusted the data
using a discrete Fourier technique. This correction method removed cloud dis-
turbances, but snow reflectance remains a significant challenge. The BMNG
visualizes seasonal changes of the land surface (spring greening, snowmelt,
drought, etc.) as true-color images and Normalized Difference Vegetation In-
dex (NDVI) maps in monthly steps at 500-m spatial resolution. The images
are available at no cost from NASA.
1
Introduction
Space exploration changed our visual perception of planet Earth. In 1950s, satel-
lites revolutionized weather forecasting when they began beaming home television
images of cloud patterns. Astronaut photography in the early 1970s showed us
the whole Earth“disk” for the first time in true-color – the so-called “Blue Mar-
ble” (Figure 1 left). Since 1972, satellite sensors have been acquiring atmosphere,
land, ice and ocean data with increasing spectral, spatial and temporal resolution.
Satellite remote sensing systems like NASAs Earth Observing System (EOS) help
us to understand and monitor Earth’s physical, chemical, and biological processes
[Wielicki et al., 1995, Kaufman et al., 1998, Running et al., 1999].
Figure 1: Earth views: Apollo 17 astronaut photograph (left, 1972), Blue Marble
false color composite (center, 2000), MODIS Blue Marble true color composite
(right, 2002)
The false-color Earth image [St¨ckli et al., 2000] shown in Figure 1 (center)
o
was created from Advanced Very High Resolution Radiometer (AVHRR), Geo-
stationary Operational Environmental Satellite (GOES-8), and Sea-viewing Wide
Field-of-view Sensor (SeaWiFS) data. Named the “Blue Marble”, that image was
created for public outreach and it demonstrated the high value of such visual-
izations. New sensors like MODIS, aboard NASAs Terra and Aqua satellites,
observe and measure a wide range of geophysical parameters. In 2002 we cre-
ated the true-color Earth image in Figure 1 (right) using MODIS land, ocean,
ice and atmosphere science products [St¨ckli et al., 2002]. This image has been
o
widely used in museums, print media, TV documentaries, in movies, by mapping
agencies, and in NASA’s public communications about its missions and research
initiatives.
The success of the Blue Marble imagery motivated us to continue the project.
The BMNG is a cloud-free true-color dataset at 500-m spatial resolution and
monthly temporal resolution. Whereas cloud-free Earth imagery is commercially
available (e.g., The Living Earth Inc., WorldSat International Inc.), the BMNG
aims at providing freely available data from the Earth surface in true color, derived
from scientific data as a value-added product. Although the spatial resolution of
the BMNG is comparable to other datasets, seasonal variations with monthly
time-steps have not been shown before in seamless true-color composites. A visu-
alization of seasonal variations (snowfall, droughts, wet seasons, spring greening,
etc.) has good potential to enhance education. Furthermore, the BMNG can help
to increase public understanding (and therefore acceptance) of satellite missions
and awareness of causes and effects of changes in Earth’s climate system.
The creation of seamless cloud-free spatial and temporal composites of Earth’s
surface is, however, not a trivial task. It is dependent on sophisticated atmospheric
corrections [Vermote et al., 1997] and cloud screening. Even then, cloudy pixels,
missing time periods, and processing artifacts [Los et al., 2000], can disturb satel-
lite derived data. Temporal compositing based on empirical filters (Maximum
value composite, Holben [1986]) can be used to remove such irregularities. An
ecosystem-dependent spatio-temporal interpolation has been successfully applied
to MODIS albedo data by Moody et al. [2005]. For the BMNG dataset, we used a
temporal adjustment based on discrete Fourier series. This method is well suited
to creating consistent NDVI time-series images from AVHRR data [Los et al.,
2000, St¨ckli and Vidale, 2004], but had not been applied to true-color imagery
o
until now.
The next sections describe the science behind the BMNG imagery and provide
examples, showing the performance of the method we used.
2
Methods
We used a temporal correction based on discrete Fourier series to create monthly
land surface reflectance composites from MOD09A1 (TERRA satellite) data. This
method assumes that the snow-free land surface changes on a seasonal timescale
with a yearly periodicity [Moulin et al., 1997]. Second order Fourier series allow
to model this temporal pattern, which was successfully demonstrated by Los et al.
[2000] and St¨ckli and Vidale [2004]. Fourier series can be fitted to a disturbed
o
or incomplete measurement time series and still represent the seasonal changes
very well, while ignoring short-term disturbances. Some artificial landscapes, like
cropfields, can have more than one growing cycle during a year. We used third
order Fourier series to account for this pattern when we had enough data to fit
such a curve. Some areas do not show substantial seasonal change (such as lakes,
ocean, and permanently snow-covered areas) nor do they have sufficient temporal
coverage (tropical rainforests) due to persistent cloud cover or aerosols. For those
areas we used weighted averaging to temporally composite their reflectance values.
First, we quality checked individual MODIS bands, followed by a snow-free Fourier
adjustment. As a last step, we added the reflectance values for snow-covered areas
to the snow-free dataset and then transformed the data into true-color imagery
and NDVI maps.
2.1
Data
The BMNG is based on MODIS land surface reflectance data (MOD09, Justice
et al. [2002]) – the reflectance that would be measured coming off the land surface
if there were no atmosphere. This measure is calculated from MODIS top-of-
atmosphere reflectances by applying corrections to remove the effect of gaseous
absorption, molecules and aerosol scattering [Vermote et al., 1997]. In compar-
ison to earlier satellite-derived surface reflectance products (e.g. from AVHRR;
[James and Kalluri, 1994, Csiszar and Gutman, 1999]), the MODIS land surface
reflectance is corrected for aerosols. This novel correction is important for the
monthly BMNG dataset, since aerosol effects can remain in uncorrected datasets
even after monthly temporal compositing periods [Vermote et al., 2002].
All MOD09 8-day composites (MOD09A1, Collection 4) from 2004 were used.
Bands 1 (red), 2 (near-IR), 3 (blue) and 4 (green), the internal snow cover flag,
and the internal cloud mask at 500-m spatial resolution were quality screened
through the use of the provided quality flags:
1. bands 1,2,3 and 4 extracted if quality assurance (QA) bits are ideal quality
(sur refl qc 500m bits 0 and 1) and none of the bands have fill values
2. snow masked if the internal snow mask flagged (sur refl state 500m bit 12)
3. cloud masked if the internal cloud mask or the cloud shadow mask flagged
(sur refl state 500m bits 2 and 10)
4. steps 1-3 only performed if the aerosol quality is high (sur refl state 500m
bits 6 and 7)
We used a time-invariant land cover map where the surface reflectance product
provided insufficient temporal information – specifically, the MODIS landcover
dataset (MOD12Q1, year 2001, Collection 4, IGBP classification) derived through
the use of a supervised classification methodology [Friedl et al., 2002].
2.2
Temporal snow-free correction
We applied a weighting scheme to all valid snow-free and cloud-free 8-day MOD09A1
reflectance values. This method eliminates MODIS cloud and snow mask failures.
Contamination of snow and cloud pixels is problematic for fitting discrete Fourier
series. The weights
W
i
are based on the standard deviation of reflectance values
X
i
b
between bands
b
= 1
. . . nb,
where
nb
= 4 is the number of bands. Large
weights are given to pixels showing a high standard deviation (relative to their
magnitude: e.g. dense green vegetation) and small weights are given to pixels,
where all color values are in the similar range (e.g. clouds, snow);
1
W
i
= ¯
X
i
nb
¯
(X
i
b
−
X
i
)
2
b=1
(1)
where
i
= 1
. . . n
and
n
= 46 is the number of 8-day reflectance composites per
year. Using this scheme, non-vegetated areas such as deserts, salt lakes, and
permanent snow receive a constant low weighting during the course of a year.
Weights
W
i
are therefore normalized by their yearly mean value. The Fourier
matrix
(F
T
F
)c =
F
T
W X
(2)
is solved for the Fourier coefficients
c
j
b
by use of singular value decomposition,
where
F
i
=
W
i
1 cos
φ
i
sin
φ
i
· · ·
cos(m
−
1)φ
i
sin(m
−
1)φ
i
(3)
is a matrix
n
x (2m
−
1),
φ
i
= (2πi)/n, and
j
= 1
. . .
2m
−
1.
m
is the order of
the resulting discrete Fourier series
Y
i
b
=
c
1
b
+
c
2
b
cos
φ
i
+
c
3
b
sin
φ
i
+
. . .
+
c
2m
b
cos(m
−
1)φ
i
+
c
2m+1
b
sin(m
−
1)φ
i
(4)
and
m
= 2 when the longest gap in
X
is between 1-3 months and
m
= 3 when the
longest gap is shorter than one month. Time-series with longer gaps (but with at
least 3 valid reflectances) are linearly interpolated, only using values with
W
i
>
0.5. Monthly snow-free surface reflectances are created from
Y
i
b
by averaging.
Water pixels (Discriminated by the MOD12Q1 land cover dataset) are averaged
over the entire year, discarding values where
W
i
<
1.0.

Plik z chomika:

Inne pliki z tego folderu:

Inne foldery tego chomika:

Zgłoś jeśli naruszono regulamin