Cloud Masking Techniques for AVHRR Data for Oceanographic Applications in Malaysia
Data
AVHRR scene acquired between March 1996 to March 1997 were used in this study. Sub scenes consisting of 850 pixels and 800 lines covering the sea water areas around Malaysia ( the South China sea and the Straits if Malacca ) were extracted for detailed analyses. These data sets were captured by NOAA-12 and NOAA-13 satellites series.
Cloud Detection Technique
Different clouds have different optical and thermal properties. Each cloud filtering technique has to unique capability. No single technique can eliminate all cloud types form the satellite data. A combination of these techniques is essential for very high certainty in cloud masking. Day time and night time cloud detection schemes are briefly described below.
-
Infrared Threshold Test or Gross Cloud Check
This method uses 12 mm ( channel 5) brightness temperatuerr4. channel 5 is used because clouds have greater optical thickness at this band . pixel will be flagged as cloudy if the calculated temperature is less than a pre-defined threshold value. The thresholds have to be determined form the histogram of the brightness temperature generated image. this test is applicable for both day and night scenes. This method is not capable of detecting a warm low cloud above the sea.
- Spatial Coherence Technique
This method was developed by Coakley and Bretherton ( 1982) which uses the spatial structure of the infrared radiance field for obtaining cloud cover and determining clear sky radiances. It relies on the fact that cloud top radiances often vary over small spatial scales whereas the radiance form the sea surface is uniform over large areas. If cloud-free pixels do to exhibit uniformity, such as in coastal areas where there are usually large variations in surface temperatures, this test will have difficulty to identify them. A modified version of this technique was introduced by Thiermann and Ruprecth ( 1992).
- Dynamic visible and Near Infrared Threshold Technique
The reflectance of most cloud types is higher than that of the sea at visible and near-infrared wavelengths away form areas of secular reflection . radiation of shorter wavelengths ( channel 1) is more affected than radiation at longer wavelengths by aerosols and Rayleigh scattering. Land and cloud have higher albedeo at longer wavelengths. Therefore channel 2 radiances are preferred for cloud detection over the sea because the higher radiances reflected form the land can also be masked out. The land surface has much lower reflectance than clouds in channel 1, thus giving high contrast between land and cloud at this wavelength. This favous the use of channel 1 for cloud detection over land.
The threshold for cloud is a function of solar zenith angle, satellite zenith angle and azimuthal angle of the view point. It can be determined dynamically from the reflectivity histogram. The visible histogram can two cloud-free peaks over coastal regions ( sunders 1986) to which the dynamic threshold cannot be applied successfully. To overcome this problem a constant threshold can be used.
- Ratio of Near -Infrared Reflectance to visible Reflectance
This test is only applicable during the day. The bi-directional reflectance is defined as
Rn= (GnCn+Yn) /cos (1)
Where Gn is the gain, Yn is the intercept, Cn is the raw count for channel n and is the solar zenith angle ( Saunders and Kriebel 1988). The ratio used in the test is defined as
Q=R2/R1 (2)
Over clouds the value of Q is close to unity. Enhanced molecular and aerosol backscattering at shorter wavelength over cloud -free water causes the visible reflectance to be often twice that in the near-infrared. The value of Q is then about 0.5. however over land the reflectance at the near-infrared wavelength is always higher than the visible reflectance and thus Q is always greater than unity. Cloud-free pixels can be determined by using pre-defined or dynamically defined thresholds. The test cannot be applied on sunglint areas because the value of Q can approach unity and this will falsely flag clear pixels as cloudy.
- Channel Difference
The emissivity difference of low cloud and fog at 3.7 mm and 11 mm provides an effective test at night for the detection. The emissivity of stratus or fog is greater at 11mm than at 3.7mm. this technique is use full since uniform low cloud at night is the most difficult to detect with either the infrared or the spatial coherence method.
The test using T3.7-T12 is useful for detecting pixels partially filled with cloud, semitransparent cloud an most medium-level and high-level clouds . thin cirrus cloud is easily detected by this method because this cloud provides higher differences in transmissivity an single scattering albedeo between 3.7mm and 12mm.
The temperature differences between T11 and T12 can be applied to both day time and night-time. This test is effective in detecting thin cirrus cloud and the edges of thicker cloud. This technique can be used to detect most types of clouds except for uniform low cloud. For clear sky radiances, the difference are less ( often <1 degK) but will vary with total column water amount and satellite zenith angle ( saunders and Kriebel 1988). Over cloud the brightness temperature differences are higher than over the clear sea water. All pixels with T11-T12>Tdiff ( threshold for temperature difference ) are then identified as cloud contaminated.
