Fig. I. Examples of cloud shadow detection using the CSDI technique. Left panel: HICO image acquired over Virgin Islands on December 20, 2009 (image size: 270 x 400 pixels); (a) true color image, (b) corresponding IV image, and (c) corresponding CSDI image. Right panel: HICO image acquired over Samoa on October 2, 2010 ' (image size: 260 x 260 pixels); (d) true color image, (e) corresponding IV image and (f) corresponding CSDI image. The clouds are shown in white on both CSDI and IV images, the shadows are
Fig. I. Examples of cloud shadow detection using the CSDI technique. Left panel: HICO image acquired over Virgin Islands on December 20, 2009 (image size: 270 x 400 pixels); (a) true color image, (b) corresponding IV image, and (c) corresponding CSDI image. Right panel: HICO image acquired over Samoa on October 2, 2010 ' (image size: 260 x 260 pixels); (d) true color image, (e) corresponding IV image and (f) corresponding CSDI image. The clouds are shown in white on both CSDI and IV images, the shadows are shown in red on the CSDI images and in slightly cooler color on the IV image, and the sunlit regions are shown in blue on the CSDI images and in warmer color on the IV images. The true color, IV and the CSDI images agrees pretty well. The cloud shadows are clearly seen in red structures adjacent to the white clouds on the CSDI images. The shape of the cloud shadow especially for the isolated cloud closely follows that of the cloud as expected.

Background: Spectral information collected by optical satellite sensors can provide important information for various global remote sensing applications. However, clouds cause a serious problem for these sensors, especially over humid tropical regions. Throughout the year about two thirds of the Earth's surface is always covered by the clouds. The problem for the optical sensor is that clouds not only conceal the ground but they also cast shadows and these shadows also occur in the observed images along with the clouds. Unlike airborne imaging where shadows can be minimized by flying at certain times during the day, low Earth orbit satellite-based sensors are limited to acquiring images at fixed times of the day. If the solar elevation is low at the time, then the presence of shadow will be unavoidable. The main problem caused by shadows is either a reduction or total loss of information in an image. Since ocean color algorithms are developed for water pixels illuminated by both direct solar irradiance and sky light, the radiance values in shadow pixels leads to the corruption of biophysical parameters derived at those pixels. Cloud shadow can produce errors of 30-40% in the observed reflectance from the affected pixels over lands [1]. Similar errors can be expected over waters as well, although such studies have never been conducted. Since ocean color products are retrieved based on the assumption that the remote sensing reflectances are accurate, a small inaccuracy in the reflectance can lead to significant errors in the retrieved products. Particularly, since most of the product retrieval algorithms are band ratio algorithms, a small disproportionate alteration in the spectral reflectance amplitude can changes the band ratios considerably hence the retrieved products. However, cloud shadow detection in ocean color scene can be
important and beneficial. For example, the cloud shadowed pixel (pixel illuminated by only skylight photons since direct photons are removed by the cloud) in combination with the neighboring sunlit pixel (pixel illuminated by both direct solar and skylight photons) of similar optical properties can be used to remove atmospheric effects from these scenes. The neighboring sunlit pixel then can be used as known reflectance targets for validation of the sensor calibration and atmospheric corrections. Cloud shadow is important for many other reasons as well. For example, cloud shadow can impact mesoscale atmospheric circulations that lead to major convective storm systems [2]. Furthermore, cloud shadow can also be used to estimate both cloud base and cloud top heights which are still a challenge to estimate reliably from space.

There are numerous algorithms for cloud detection. However, relatively few cloud shadow detection algorithms have appeared in the literature even though accurate detection of cloud shadow is important for many atmospheric and terrestrial applications. Most of the shadow detection techniques described in the literature deals with shadows over land. Few attempts have been made to detect shadows specifically over water, while shadow detection over water is becoming significant as the spatial resolutions of the ocean color sensors are getting finer. This is because the small scale shadow features appear in the acquired images.

The locations of shadows in the image depend on clouds elevation and the incidence angle of the sunlight at that time. The cloud shadow location can be determined by the means of geometrical calculations if the spatial location of cloud, cloud top and bottom heights, and the sun and satellite positions are known. However, geometry based approaches have challenging issues besides requiring too much CPU to run operationally. The main issue for geometry based approach is the estimation of cloud vertical height which is required to determine the relative shadow location. Normally, thermal channels can be used to estimate the cloud-top height. However, it is still a challenge to determine cloud bottom height without cloud profiling measurements. The solar reflective bands cannot provide information about the cloud top height, and the cloud bottom information cannot be reliably estimated from passive
solar-thermal data either. To determine accurate shadow location, both heights are important especially for isolated clouds. In any event, many ocean color sensors such as the Sea-viewing Wide Field-of-view Sensor (SeaWiFS) do not have necessary channels to estimate clouds vertical heights. Therefore, in order to identify the shadow locations from these sensors, we need an algorithm that uses visible channels since these channels are always present on the ocean color sensors.

It can be easy to identify the cloud regions simply by using brightness thresholds, but it is difficult to identify the shadow regions this way because their brightness values can be very close to those of their neighbors or some other regions. Distinguishing shadows over water bodies based on spectral reflectance shape and amplitude information is also very difficult or possibly even impossible. Shadows over water pixels do not have any specific spectral features while the brightness varies with atmospheric conditions and imaging geometry. Therefore, the brightness or the spectral shape alone may not be appropriate for shadow detection. However, brightness values from shadow and close by sunlit regions over water can provide a great deal of information if a small portion of the image (where optical properties of water and atmosphere is uniform) is examined at a time. This is because the 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. The path radiance from the shadowed pixel to the sensor is also slightly lower depending on how much of the atmospheric path radiance is shadowed. Therefore, the total radiance at the top of the atmosphere measured over the shadowed pixels is slightly lower compare to the adjacent sunlit pixels. Assuming the optical properties of water and atmosphere is homogeneous around shadowed and adjacent sunlit regions, examining the radiance difference amongst these small uniform regions together enable us to separate the shadowed regions.

Invention: A cloud shadow detection technique called the Cloud Shadow Detection Index (CSDI) for optical imageries acquired over water by satellite/airborne sensors. This technique does not require any angular information (viewing or solar), or any estimation of clouds vertical heights. It is entirely based on measurements in the optical channels. This approach is for homogeneous water bodies such as deep waters where shadow detection is very challenging due to the relatively small differences in the brightness values of the shadows and neighboring sunlit or some other regions.

The CSDI technique is developed based on the small differences between the total radiance reaching the sensor from the shadowed and neighboring sunlit regions of similar optical properties by amplifying the differences through integrating the spectra of the two regions. The Integrated Value (IV) is then normalized by the mean of the IVs within a spatial Adaptive Sliding Box (ASB) where atmospheric and marine optical properties are assumed homogeneous. Using a predefined threshold shadow pixels are separated from the sunlit pixels.

The CSDI technique will not only improve ocean color products retrieval such as chlorophyll, it can also enable us to do tasks such as automating sensor's calibration and atmospheric conection using the cloud-shadow approach, improve local weather forecasting, and estimate cloud base and cloud top heights.

Advantages: The CSDI method has several advantages over geometry based or reflectance threshold based techniques. It does not require any thermal or short wave infrared channels which are not always present on ocean color sensors. The CSDI method is entirely based on visible channels, the most important spectral region for ocean color studies, which always exist on ocean color sensors. Although angular (solar and viewing) information can be acquired from the satellite sensors, it is still a challenge to estimate clouds vertical height accurately from satellite sensors. Some sensors such as Moderate Resolution Imaging Spectroradiometer (MODIS), can estimate cloud vertical height however they are not very reliable. Nonetheless, most ocean color sensors do not have capability to estimate clouds vertical heights. So, geometry based approach is not appropriate for these sensors. On the other hand, the CSDI method does not require any estimation of clouds vertical heights or even any angular information. It is based on the top of the atmosphere readings of the space borne or airborne sensors. Since it is based on the true measurement without any estimation, the detected shadow locations are more precise. Furthermore, the CSDI is relatively easy to use and should be faster than the geometry based approach since it requires less computation. Moreover, since the CSDI method uses the top-of-atmosphere radiance measured in raw digital counts, the method will still work well even with the sensors with high radiometric calibration errors unlike the reflectance threshold based methods.

Applications: The CSDI detected shadows can enable us to estimate both cloud base and cloud top heights which are still a challenge to estimate reliably from space. This will make cloud- shadow data usable even from ocean color sensors which can be important and beneficial. For example, cloud shadow can impact mesoscale atmospheric circulations that lead to major convective storm systems. This has potential to improve local weather forecasting. This technology can facilitate satellite sensors calibration (in a very cost-effective way) without any expensive airborne and ground based measurements. The technique will also facilitate atmospheric correction using cloud-shadow method for ocean color sensors. The technique can be slightly modified to detect shadows over land and also shadows on digital images from typical cameras. This technique will also improve ocean color products retrieval such chlorophyll by eliminating shadow pixels.

Market: Multispectral Imaging is used to detect the phenomenon of clouds. The patented technique is not implemented in a product so the market is large for the potential applications of the technology. It is difficult to identify a specific market and its value until the technology is implemented with an existing system. The forecast for the global market of space-based platforms is expected to rise to $4.5 billion in 2019. In 2013 is estimated to be a $3 billion global market.

Patents: US Patent No. 8,509,476 entitled "Automated system and method for optical cloud shadow detection over water." Issued August 13, 2013 to Amin; Ruhul, Gould; Richard W., Hou; Weilin, Arnone; Robert, Lee; Zhongpong

References:

  • l. Ruhul Amin, Richard W. Gould, Weilin Hou, Robert A. Arnone, and ZhongPing Lee "Optical Clowl Shadow Detection Algorithm over Water," IEEE Trans. Geosci. Remote Sens. Vol. 51, No.2, (2013)
  • J. J. Simpson and J. R. Stitt, "A procedure for the detection and removal of cloud shadow from AVHRR data over land," IEEE Trans. Geosci. Remote Sens. 36, 880-897, 1998.
  • R. T. McNider, J. A. Song, and S. Q. Kidder, "Assimilation of GOES-derived solar insolation into a mesoscale model for studies of cloud shading effects," Int. J. Remote Sensing, 16, 2207-2231, 1995.

Click here for a more detailed PDF version of this technology.

Technology Status: Patent has issued.
Licensing Status: Licensing available to companies with commercial interest.
Lead Inventor: Ruhul Amin


Back to Oceanography
Technology Tags: