2. REMOTE SENSING
• The diagram shows the general process of gathering information by optical sensors.
• Light from source A falls on the landmass C which is reflected to the sensor D.
• The signals are then sent to the receiving station where the images are processed
referring to F which is then passed as data to the human interpreter G.
• B and E refer to the Electromagnetic waves.
3. Role of Clouds in Remote
Sensing
•Clouds cause a serious problem for
the sensors. They not only conceal the
ground but also cast shadows.
•These shadows also occur in the
observed images along with the
clouds.
•The main problem caused by shadows
is either a reduction or total loss of
information in an image.
•But shadows can also be used to
estimate both cloud base and cloud-
top height which are still a challenge
from space.
•It can also impact mesoscale
atmospheric circulations that lead to
major convective storm systems.
Clouds and their shadows over land.
Clouds and their shadows over water.
4. SENSORS
HYPERION
• A hyper spectral sensor on board NASA’s
EO-1 satellite.
• Spatial resolution about 30 meters.
• Spectral configuration of 430nm-2400nm.
• Designed for land operations.
• Useful for coastal areas and Navy
operations.
• Not suitable for water studies.
MODIS
• Moderate Resolution Imaging
Spectroradiometer(MODIS) is currently
aboard the Terra and Aqua spacecraft.
• Two spatial resolution bands of 250
meters each.
• Wide spectral range: 0.41-14.24µm.
• Used for global monitoring of terrestrial
ecosystems, fires, ocean biological
properties and sea surface temperature.
5. SHADOW DETECTION OVER LAND
• Location of shadows depends on:
– Cloud elevation.
– Incidence angle of the sunlight at that time.
• For geometrical calculations ,we need:
– Cloud-top and cloud-bottom heights.
– Sun and satellite positions.
• Main issue is determination of cloud vertical height.
• Thermal channels can be used to estimate cloud-top height but cloud-bottom
height estimation is still a challenge.
• Another method is to use brightness thresholds of clouds but it is a difficult
process as the brightness values can be very close to those of their neighbors.
6. The image on the left depicts a sun-cloud geometry for shadow detection.
The image on the right tells the projections of clouds at discrete heights from
sea level.
7. SHADOW DETECTION OVER WATER
• The brightness of shadows over water pixels varies with atmospheric conditions. Therefore
the brightness values from shadow and close-by sunlit regions over water can provide
information if a small portion of the image is examined at a time.
• It is because water-leaving radiance over sunlit pixels results from both direct and diffuse
solar irradiance while the water-leaving radiance over shadowed pixels results from only
diffuse solar irradiance.
Sunlit pixel
having direct and
diffuse radiance
Shadow pixel having only
diffuse solar radiance
8. •The diffuse part of the incident radiation is radiation from the sky, exclusive of the
sun. It comes from clouds or from the blue sky, i.e. from many directions
simultaneously.
•The direct part comes directly from the direction of the sun and can, therefore, cast
shadows.
•Water-leaving radiance is the light emitted from the water pixels.
•This paper does not use any angular information or cloud height estimation. It is
a cloud shadow detection technique acquired over water by satellite/airborne
sensors.
It is derived for optical imageries entirely based on measurements in the optical
channels.
9. DATA OF THE RESEARCH
• Hyperspectral Imager for the Coastal Ocean(HICO)
– The HICO has been operating aboard the ISS since 24 September, 2009.
– Provides hyperspectral images at 100 meters resolution optimized for the coastal ocean.
– Collects radiance at 128 contiguous spectral channels from 350 to 1070nm range.
– Each HICO scene is 50km in width by 200km in length.
– HICO data flow from the ISS provides 15 scenes per day and managed by U.S. Naval Research Lab.
– Has high spectral resolution, thus contrast between shadowed and adjacent sunlit regions would be
higher after integrating the spectra which is advantageous for shadow detection.
10. Lsdw w (λ)
From shadowed region
Lsny w(λ)
From sunlit region
•Assuming sensor is at nadir, i.e. directly below and a thick cloud over water preventing direct
solar photons on the sea surface and generating a shadow region.
•The total radiance measured by the sensor from the sunlit area is:
Lsny t (λ) = La (λ) + t(λ) Lsny w (λ)---------------------->(1)
La from the atmosphere
11. where t(λ) represents the diffuse transmittance of the atmosphere for the water-leaving radiance.
The total radiance over the shadowed region is :
Lsdw t (λ) = La (λ) + {t(λ) +Δt(λ)} Lsdw w (λ)------------>(2)
where Δ represents the perturbations due to the differences in illuminations between the sunlit and
shadowed regions.
Water-leaving radiance can be expressed as two parts:
•Part caused by the backscattering of the diffuse skylight
•Part caused by the backscattering of the direct solar beam
For sunlit and shadowed regions ,the water-leaving radiance can be expressed as:
Lsny w (λ) = Lsny wsky (λ) + Lsny wdir (λ) and
Lsdw w (λ) = Lsdw wsky (λ)
Where Lsny wsky (λ) / Lsdw wsky (λ) and Lsny wdir (λ) / Lsdw wdir (λ) represent the water-
leaving radiances caused by diffuse skylight and direct solar beam / shadowed region
in the sunlit region.
12. According to P.Reinersman , K.L. Carder and F.R. Chen, “Satellite-sensor calibration
verification with the cloud shadow method” vol. 37, no. 24, pp.5541-5549, Aug. 1998., Lsny
wsky (λ) can be expressed as
Lsny wsky (λ) = Lsdw wsky (λ) + ΔLsdw wsky (λ)
•From the analysis , it can be expected that the water-leaving radiance from the shadowed
pixel reaching the satellite sensor is lower than the water-leaving radiance from the
neighboring sunlit pixels.
•Thus, the total radiance measured over the shadowed pixel is lower than that measured
over the neighboring sunlit pixel.
•An example of HICO image taken over Guam
island in 2009. the adjacent sunlit pixel has
higher digital counts(green line) over
shadowed pixels(red line).
13. DEVELOPMENT OF CSDI
• The above proposed technique indicates that the spectral shape or amplitude alone is not
adequate to separate the two regions for an entire image. It can be only done if a small
portion is examined.
• So , we introduce a cloud shadow detection technique called the CSDI as
CSDI=IV c / <IV ASB >---------------->(3)
– Where IVc represents the IV index of the pixel (the central pixel of ASB which is the
small portion ) that needs to be classified as a shadowed or sunlit pixel.
– The <IV ASB > represents the spatial mean of the IV indices within the selected ASB of
this pixel.
– The IV index is defined as
• IV=∫ Lt(λ) dλ from 400nm to 600 nm---------------------(4)
• CSDI is mainly used for deep waters. Before applying CSDI , cloud needs to be removed
properly. The ASB needs to be selected carefully so that it only contains shadowed and sunlit
pixels or only sunlit pixels because to make the denominator of CSDI larger than the
numerator for shadowed and vice versa for sunlit pixels.
14. •It is important to select the ASB in such a way that it is bigger than the shadowed region.
•This can be achieved by using the cloud size information since cloud is generally bigger than
the shadow and relatively easy to detect.
•If the selected ASB contains only sunlit pixels and the pixel under examination is also a
sunlit pixel, the CSDI value for this pixel would be around one since the mean of the ASB
[denominator of (3)] and the IV index [numerator of (3)] would be about the same.
•If the ASB contains both shadowed and sunlit pixels and the pixel under examination is a
sunlit pixel, the CSDI value will be greater than one since the mean of the ASB will be slightly
lower than the IV index of the pixel under examination.
•On the other hand, if the pixel under examination happens to be a shadowed pixel, the
CSDI value would be less than one since the IV index of this shadowed pixel would be
smaller than the mean of the ASB. Now, if the ASB contains only shadowed pixels, it can be
problematic since the CSDI value will be around one, like the case of only sunlit pixels. They
will be classified as sunlit pixels if the CSDI threshold is put less than one.
17. BENEFITS AND DRAWBACKS OF CSDI
•Benefits
•Relatively easy to use and faster than geometry based approach.
•Does not require thermal channels which are not always present on ocean sensors.
•Based on top-of-the-atmosphere readings or airborne sensors.
•Drawbacks
•May give spurious results in non-homogenous or shallow waters.
•Cannot detect shadows in the edge pixels of satellite images. But an IV image can be
used to visually identify the shadows in those pixels since shadows appear dimmer in
an IV image.
18. CONCLUSION
•A cloud shadow-detection technique (CSDI) has been developed and applied to HICO data
collected from various locations to isolate shadowed pixels. The shapes of the clouds and
cloud shadows observed in the CSDI images closely resemble those of clouds and cloud
shadows in the corresponding true color and IV images. The agreement between the true
color, IV, and CSDI images is very reasonable over open ocean.
• This suggests the potential of the cloud shadow detection using the proposed technique
which only uses the top-of-the atmosphere optical readings of the space borne or airborne
imagers. Although the proposed CSDI threshold works reasonably well on the selected HICO
images, further studies are necessary to fine tune the threshold and the selection of optimal
ASB size based on image scene content for automated processing.
