2. MATHOD
Clouds are generally characterized by higher reflectance and
lower temperature than the underlying earth surface. So visible
and infrared threshold approaches can be used for detection
cloud cover. However, there is some surface when this
characterization of clouds is inappropriate, most notably over
snow and ice.
Clouds have three radiative properties in the visible and infrared
spectral range:
1) Clouds are white in the visible and near infrared;
2) Clouds are bright in the visible and near infrared;
3) Clouds are cold in the thermal infrared.
Water and ice clouds show some distinct extinction features at
1.6um and 3yum.
But some surface like snow, ice has spectral properties that are
similar to the clouds properties. Fig.l shows the repectivity of
snow and clouds.
Snow on the ground reflects the sun strongly at one of the
MODIS wavelength bands (0.6jm) but very little at another
band (1.6um). Clouds reflect the sun well at both wavelengths.
1.0
SONVAST 1438
8
i
A
2.5 3.0
1.5 2.0
WAVELENGTH rpm)
Figure 1. Spectral properties of snow and cloud
But clouds seldom show all the properties at the same time.
Thin clouds show a portion of the underlying surface spectral
properties and low clouds are warmer than the background.
Additionally some surface types, like snow, ice and deserts have
spectral properties that are similar to clouds. Therefore simple
threshold algorithms often don’t work. Cloud detection
algorithms use many different cascaded thresholds. (Saunders,
1988; King, 1992; Ackermann, 1998)
2.1 Clouds detection
Spectral test for cloud detection are based on the fact that
clouds are highly reflective in the visible, near, and mid-IR
bands and are cold in the thermal bands (Hulley, 2008). These
characteristics are used to build thresholds to detect most types
of clouds. The Landsat-7 Cloud cover algorithm is based on
Lansat-4 and Landsat-5 and MODIS cloud mask. This
algorithm uses eight different filters in four bands to distinguish
clouds and eliminate problematic land surfaces such as snow
and other highly reflective surface.
The MODIS cloud mask is a science data product that will be
produced regularly as a Earth Observing System(EOS) standard
product. Its main purpose is to identify scenes where land,
ocean and atmosphere products should be retrieved based on
the amount of obstruction of the surface due to clouds and thick
aerosol. (Kathleen, 2005)
MODIS cloud mask algorithm uses a series of sequential tests
on the passive reflected solar and infrared observations.
Band Spectral
range(um)
Application field
Band1 0.620-0.670 Land, cloud, Aerosol boundary
Band6 1.628-1.652 Land, cloud, aerosol
Band8 0.405-0.420 Ocean color, phytoplankton
Band26 | 1.360-1.390 Cirrus, vapor
Band29 | 8.400-8.700 Cloud characteristics, temperature
Band31 | 10.780-11.280 | Land cover, Cloud top
temperature
Tablel.the spectral range and application field of the MODIS
bands for the cloud detection
Cloud detection employs the normalized difference cloud index
(NDCI) defined as the difference of reflectances observed in
two bands divided by the sum of the two reflectances.
Here have two type of NDCI. One is defined as the difference of
reflectances observed in a visible band (0.66jm) and a near
infrared band (0.936um). The other is defined as the difference
between 0.66pm and 1.64um.
Cloud is high repectivity in visible band (0.66um), and it is very
appropriate for the discriminate the edge of land and cloud at
this band. For the near infrared band (0.936um) the Spectral
characteristic of cloud has relations with the vapors from the
atmosphere. It is the vapors absorption valley. So the NDCI is
defined as below.
_ Band0. 66 — band0. 936
NDCI =
Band0. 66 + band0. 936 (1)
When the NDCI is positive the surface is cloud; when the NDCI
is near to zero the surface is soil; when the NDCI is negative the
surface may be vegetation.
For the 1.64um band, the repectivity of snow is less than the
cloud.
According to the experience a pixel is mapped as bare soil when
NDCI<0; A pixel is mapped as water when NDCI is between 0
and 0.1; NDCI of cloud pixel is from 0.1 to 0.5; NDCI value of
snow is great than 0.5.So it is possible to set a suitable
threshold to detect cloud and other land cover.
2.2 Cloud removal methodology
Cloud removal is different from cloud detection. The results
from cloud detection are cloud mask. But cloud removals are
further more. The cloud mask pixel will be replace by its 'real
pixel', such as soil, vegetation, water, snow or other land cover
types with a great deal repeating scenes coming from the same
area.
General
and onc
compos
compos
removal
interval
weather
the time
clear pi:
equals ‘
31 Si
Our stu
The ge
80?0'4:
3274.5
Tibet i:
Qingha
on the
on the
borders
It has a
average
-Mt. Q
In its s
of the
meter |
Figure
map. T
data cc
The ra
N40. "
MODI
500m :