Towards Digital Earth 1
— Proceedings of the International Symposium on Digital Earth
Science Press ,1999
Development of the Cloud Detection and Rehabilitation Operational System
Applied to NOAA Satellite Image
Jianguo Tan1 Hongmei Zhou1 Xian Lu1 Chongjun Yang2
Shanghai Meteorological Institute, Shanghai, 200030
Institute of Remote Sensing Application, CAS, Beijing, 100101
ABSTRACT In this paper, the techniques and models of cloud detection and cloud rehabilitation applied to
AVHRR image of NOAA meteorological satellite and the basic functions of the operation system on base of
Visual Basic language has been introduced. By experimentation and comparing with the fact it have
acquired the high operation efficiency and precision of the cloud detection and cloud rehabilitation. Thereby,
the cloud detection and cloud rehabilitation operation system can provide the guarantee of credible NOAA/
AVHRR images in the fields of remote sensing application.
KEY WORDS NOAA satellite, cloud detection, cloud rehabilitation, operation system
1. Introduction than in short infrared band, etc.
The data of NOAA satellite have been paid more In addition, the differences of reflectivity
attention and applied to the many fields such as between low cloud or thin cloud and underlying
meteorology, hydrology, geology, agriculture and mediums.
environment for its characteristics of short period, All these are advantage in detecting the cloud.
broad coverage. For reason of weather there is So based on the difference between cloud and no
seldom cloudless sky, it is difficult to get a no cloud cloud area in characteristic in spectrum and the
satellite image covering whole interesting area. threshold value are used to distinguish the cloud
Through the methods of cloud detection and area and no cloud area.
rehabilitation, we can detect cloud, eliminate cloud
disturbance and enhance the availability of NOAA 2.1.1. Cloud Detection through Reflectivity in Visible
satellite. It can provide data resource assurance to Band or Near-Infrared Band
NOAA satellite remote sensing dynamic watching. The discriminant is as follow:
Based on many experimental investigation, a cloud Ai A1i (1)
detection and rehabilitation optional system has Where AI is the reflectivity in visible band(i＝1）or
been developed and had a good effect in near-infrared band；A1I is the threshold value of
enhancement of data utilization of AVHRR. reflectivity.
In order to get good effect and automation of
2. Theory and Methods cloud detection, we analyze the reflectivity in visible
band, which is sensitive to cloud and find that the
2.1 Methods and Models of Cloud Detection reflectivity intensity of cloud affects the reflectivity of
In fact, cloud detection is target recognition and underlying mediums that not cover with cloud. The
classification process, which can be used statistical thicker of the cloud, its reflectivity is higher, while
method and threshold value. In this system the reflectivity of underlying medium is higher. The
threshold value has been used in cloud detection enhanced reflectivity intensity relates to the cloud
based on experimental analysis of spectrum reflectivity. In addition, climate also affects the
characteristic. reflectivity. In the clear days or the heat days, cloud
The basic theory of threshold value is that the and underlying mediums have higher reflectivity
cloud has bigger reflectivity in visible band and near because of the intense radiation intensity. Through
–infrared band and lower brightness temperature in repeated experiments and analyses based on
thermal infrared band than that of other underlying above characteristics, we obtained the comparative
mediums such as vegetation, soil, and water etc. In relationship between minimum reflectivity of
generally cloud area has several characteristic as underlying mediums and maximum reflectivity of
follow: cloud in visible band and achieved the automatic
Bad horizontal homogenization in the top of cloud detection in visible band. The relationship
cloud; between reflectivity and threshold in visible band
Good relationship between different infrared shows in Table 1.
bands in clear day destroyed by cloud;
Absorption of vapor in long infrared band more
2 Jianguo Tan et al./Development of the Cloud Detection and Rehabilitation Operational System
data have little change between same period, near
Table1 threshold value in cloud detection through time image. Therefore, considering objectivity of
reflectivity of visible band( NOAA satellite) data and continuity of image, it is substituted the
cloud data by the cloudless data of same satellite
Reflectivity orbital, near-time AVHRR image after being
Minimum Maximum Threshold value reversibly calculated according to variation between
65 15 two image. The method is as follow:
<3 65<max90 16 Given the original cloud image is X ， and
>90 18 substituting image is Y
3min<6 45<max90 16
x max x min m
6 70 23
y m a y min
then, the pixel value of no cloud area is ：
2.1.2. Cloud Detection through Brightness
Temperature in Infrared Band xi fi * ( x max x min) x
The discriminant is as follow: the pixel values of cloud area is ：
Ai A2i (2) xi Fi * ( x max x min) x
Where AI is the brightness temperature in different where xi,yi,xmax,ymax,xmin,ymin are maximum and
infrared bands (i＝3，4，5） 2I is the brightness minimum value in X and Y respectively.
temperature threshold. In addition we find the difference of variation
Cloud has sensitive response in thermal infrared reflectivity between ocean and land in order to
band, we chose images takes in different seasons enhance the precision of cloud rehabilitation, we
to analyze, and found that cloud has same pattern replace the cloud area in ocean and land
in CH4 and CH5. For southern China, the respectively and get better effect than before.
threshold of thermal infrared temperature is higher
in summer and it is about 10℃, below which is 3. Main Function of the System
cloud, while in spring and autumn, the threshold is The main function is as follow:
lower and it is about 5 ℃, while in winter, the Image Show: single band image or three-band
threshold is lowest and it is below 0℃. In different composition image can be show in this system, also
region and different season there are different its gray value can be gotten.
threshold. For example, northern China has many Calculation of Reflectivity： solar angle correction
lower temperature thresholds than that in southern of gray value to visible or near-infrared band and its
China. Therefore, In order to detect accurately thin conversion to reflectivity.
cloud and low cloud, based on the detecting results Calculation of Brightness Temperature:
of many data different season. We generally correction of boundary-dark to middle-infrared or
employed the threshold of temperature which is thermal infrared band ， calculation of brightness
lower 1℃ to 2℃ than minimum temperature of temperature using Planck formula.
ground surface or sea surface of research region in Cloud Detection:
same season, which obtained good results. 1) Single channel threshold detection: an image of
any channels is chosen, and a gray value threshold
2.2. Cloud Edge Model for cloud pixel is determined, the pixel whose value
In order to eliminate the effect on the edge of cloud is higher than the threshold id identified as cloud
area, w extend the cloud area, for a cloud area, we pixel.
give an coefficient of expansion K and get an 2) Threshold detection through reflectivity of visible
Boolean modular which 0 express cloud area and 1 channel and near-infrared channel: calculate pixels’
express the no cloud area. While original image reflectivity in visible channel and near-infrared
multiply this modular we can get a no cloud edge channel, the pixel whose value is higher than a
affection image. given threshold of reflection is identified as cloud
2.3. Cloud Rehabilitation Model 3) Automatic cloud detection through visible
In order to use AVHRR data more effectively, apart channel: according to the temporal and spatial
from detecting cloud image, the pixels of cloud distribution characteristics of cloud, analyze pixels’
should be rehabilitated. We find that the relative reflectivity in visible channel using method of
Jianguo Tan et al./Development of the Cloud Detection and Rehabilitation Operational System 3
automatic computer identification, detect cloud with rehabilitation.
different thresholds. Moulding Board Smoothness ： convolution
4) Threshold through brightness temperature operation and bilinear interpolation.
detection of infrared channel: calculate pixels’
brightness temperature and real temperatures using 4. Application Case
Planck formula, the pixel whose temperature is In order to test the function of this system, we
lowed than a given threshold of temperature is chose the NOAA satellite images in September 7
identified as cloud pixel. and September 8,1998, the result of cloud detection
5) Automatic cloud detection of thermal infrared and rehabilitation is show in figure 1. It shows that
channel: owing to the temperature of cloud in through the method of cloud detection and
different region as well as different season, identify rehabilitation in this system, we have get the
cloud with computer automatically. continuous, objective and reliable image so it also
6) Automatic cloud detection with combination of enhance the serviceability of NOAA /AVHRR data
visible and thermal infrared channels: detection of greatly.
thermal infrared is mainly sensitive to temperature,
and not sensitive to low cloud and thin cloud, for 5. Conclusion
their temperature are almost equal with that of The development of the operational system in cloud
ground objects, but they can be clearly seen in detection and rehabilitation applied to NOAA
visible channel. Therefore, considering respective satellite image provides an objective and advanced
advantage of visible and thermal infrared detectors, science method to eliminate the Cloud disturbance
we combined the reflection of visible channel with and enhance the utilization of NOAA satellite data.
temperature infrared channel to identify The model of cloud detection and rehabilitation is
automatically cloud. feasible either in theory or in practice. But in fact we
Image Matching: three-band image composition, find that the reflectivity of seawater in along the
image enhancement, image stretch, GCP fetch, coast which abounded with sand is bigger or equal
re-sampling etc. than that of thin cloud, so we will focus on research
Image Substitution: including cloud area extension, in distinguishing the sand area and thin cloud area.
ocean and land distinguishment, cloud
4 Jianguo Tan et al./Development of the Cloud Detection and Rehabilitation Operational System