GISdevelopment.net ---> AARS ---> ACRS 1990 ---> Water Resource

A cloud detection system using AVHRR HIRS/2 sensor data

Youichi Minowa,Mikio Takagi
Institute of industrial Science University of Tokyo, Japan


Abstract
We propose a cloud detection system using AVHRR (Advanced Very Resolution Radiometer) and HIRS/2 (High Resolution Infrared Radiation Sounder) data from NOAA (National Oceanic and atmosphere administration ) satellites . The purposes of this system is give guarantee the accuracy of the atmospheric correction method for AVHRR data using HIR /2 data there are three stages for cloud detection because two different sensors (AVHRR, HIRS/20 are used in our atmospheric correction method. On HIRS/2 stage gross cloud and channel deferential technique are On the AVHRR stage the gross cloud special coherence multi channel clouds detection technique are used final stage each HIRS/2 pixel uses results of the gross cloud and spatial coherence check of the AVHRR data which is covered with the HIRS/2 pixel to investigation the cloud contamination The relationship between the error of the atmospheric correction and the result of cloud detection in each stage is discussed.

Introduction
Remotely sensed data from meteorological satellite NOAA can provide very useful and important information in meteorology, oceanography and may other scientific fields because of its instantaneous and periodic observation of broad area of the earth. In our laboratory NOAA satellite data are received every day proceed and distributed to many other laboratories and institutes all over Japan The wide and simultaneous observation is one of the most important property of satellite sensor data especially for ocean observation ,it is useful to draw the temperature map. The distribution of the sea surface temperature (SST) observed by the satellite infrared sensors is contaminated by the atmosphere and cloud. In general remotely observed SST are cooler (about 2-6 K) than in situ data the causes are mainly emission absorption and scatter by the atmosphere and cloud. If there are some clouds which is not enough large or thick to cover the satellite sensors field of view the thermal radiance from the surface is partly blocked by them and observed SST's become cool. So in order to determine temperature on good accuracy it is necessary to remove clouds Mc lain et.al (1) discussed about the cloud detection system and its performance for their multi channel correction method.

The NOAA satellites are on board some multi channel sensors (AVHRR, HRIS/2 and other the AVHRR sensor is used for observing the earth surface on high resolution and the HIRS/2 data. But the sounding the atmosphere vertical construction in the same time .So it is expected that the atmosphere effects in the AVHRR data can be compensated by using vertical information from HIRS/2 data. But the AVHRR and HIRS/2 have different resolution and it means that the clouds act different roles for observation accuracy in each field of view of AVHRR and HIRS/2. So it is necessary to remove the clouds on each sensor.

In this paper a cloud detection system for AVHRR and HIRS/2 data is proposed This method is applied to NOAA-10 image of northwest Pacific Ocean for 9 May 1987 (which received in our laboratory) and the results of those experiments show that the almost cloud pixels included in the satellite image can be removed.

The relationship between AVHRR and HIRS/2 sensor coordinate system (2),(3)
The HIRS/2 sensor has nineteen infrared channels and one visible channel. The AVHRR sensor has one visible channel (ch-1, 0.58~0.68µm.) one near infrared channel (ch-2, 0.725~1.1 µm.) and two or three infrared channels (ch3,4 and 5, 3.55~3,93µm) 10.5~11.5µm respectively) to use AVHRR ch-3 we have to compensate the effect of the scattering and sunglint on the sea surface in daytime as the even number NOAA satellites (NOAA-6,8,10) is on board the AVHRR/1 which does not have ch-5 this channel can only be used on the odd number satellites (NOAA-7,9,11). To use the HIRS/2 ch-10~ch-12, these are very important channels because of the sensitivity for water vapor which is an important factor of the absorption of AVHRR thermal infrared channels (ch-4and 5 the weighting function factor are shifted by the difference of the vertical water vapor distribution So it is necessary to estimate the sift and adjust these weighting functions The profile of weighting functions are shown in fig 1.

To use HIRS/2 data related with AVHRR data it is necessary to decide the pixel correspondence fig.2 shown the relationship between AVHRR and HIRS/2 sensor coordinate system(3) but qA is very small so it can be omitted from consideration so the relationship can be expressed as


Where (XA,YA) and (XH YH) are pixel and line number on each sensor image DX and DY are the resolution between two sensors.




Cloud detection System.
Some algorithms for AVHRR to detect the cloud are investigated by Saunders(5) We use the three algorithms (gross cloud, special coherence channel differential) investigated in his paper in theAVHRRclouddetection stage The HIRS/2 has three window channels in infrared region (ch-8-18-and 19) and they can be use to detect the cloudy pixels (6) and we refer (6) making theHIRS/2 stage The fastest and simplest technique is the gross cloud technique it used the empirical range of sea surface temperature and the brightness temperature threshold for the thermal infrared sensor image is defined by the range ,The special coherence technique uses the standard deviations (SD) for the small area (we use 3x3 pixels here) because the cloud top temperature often vary over small space scales whereas the sea surface temperature is uniform over large area The SD of the cloud free area is smaller than the of the cloudy area. Using empirical threshold for the SD the cloudy area removed in this technique The brightness temperature of low clous or fog at 3.7 µm spectrum region is less than at 11µm and this difference is used to detect the low cloud or fog at night ( channel differential technique).

Our system has three stages because of using two different sensors, processing respectively and finally to remove the cloud contaminated pixels of the low resolution sensor (i.e HIRS/2) using the result of the high resolution sensor (i.e AVHRR) data which is included in the low resolution sensors field of view The black diagram of the system is shown in fig 3 of course the visible channel can't be used in night time as the thermal radiance and scattered sunshine are both observed by infrared 3.7µm hand such as AVHRR channel -3 so it is not used in day time in order to ignore the sunlight on the first stage HIRS/2 sensor data are processed using channel 8,9,10 this stage simple and first to remove cloud contaminated HIRS/2 pixels. On second stage AVHRR sensor data are processed using channel 3,4,5 in night time and channel 1,2,4,5 in daytime This stage removes cloud contaminated VHRR pixels using gross cloud spacial coherence and channel differential technique finally HIRS/2 data are checked using AVHRR data which is included in the HIRS/2 pixel and checked in previous stage if the clear AVHRR pixel ratio in the HIRS/2 pixel is cover a certain threshold RC the HIRS/2 pixel is regarded as cloudy.

Atmospheric correction method
In a cloud free nonscrattering atmosphere the remotely sensed brightness temperatures are effected by the radiation and absorption of the atmosphere under local thermodynamic equilibrium the remotely sensed radiance Rn at the top of the atmosphere is the sum of the radiance from the earth surface and the radiance of the atmosphere so it can be written as


Where Bn(T) is the Planck function at frequency n for emitted radiance of blackbody at temperature T, tv(q,p) is the atmospheric transmittance from pressure p to the top of the atmosphere at frequency p and zenith angle q subscript S means the value at the earth surface and Wn(p) is the weighting function which can be written as


en is the emittance of the earth surface at frequency n the property of en changes greatly in the different spectrum regions in microwave region it strongly depends on the surface conditions I infrared region it can be regarded as a constant and value is about 0.95 in the 11µm window region and 0.98 in the 3.7µm window region on the sea surface .

In this method the atmosphere is simplified by assuming that the optical properties of the atmosphere vary only method vertical direction that is the distribution of atmosphere constituents in the horizontal direction assumed to be homogenous such an atmosphere is known as a plane parallel atmosphere on the assumption eq (4-1) can be written as


Solving eq.(4-4), Bnr(Ti) can be decided, so the temperature of each layer Ti is known by the solution of th inverse-Plank function.

Result of experiments
for the experiments to estimate this cloud detection system the NOAA 10 satellite image of northwest Pacific Ocean for the image received at 9 May 1987 (JST) and HIRS/2-1,2,….7 channels and AVHRR-4 channel are used for the atmospheric correction The sensor coordinate parameters DI, DJ and qAH for NOAA-10 satellite are used as -0.5, 2.5 and 0.00 respectively (see eq-2-1)

The in situ SSTs in the weather reports of northwest Pacific Ocean by the Meteorological Agency of Japan which contains data observed by may kinds of ships and buoys are used for estimation Thus the in situ data observed between 6 hours before and after the satellite observation The relationship between the corrected remotely sensed temperature and the in situ SSTsare shown in fig 4 The Statistical parameters are shown in table 1 The standard distribution depends on the detection method of cloud free area because the cloud existence is one of the most effective error of SST measurement The weather reports have the cloud observations (cloud amount cover type etc.) The observations at 9 May 1987 6.00 (JST) are used to estimate the cloud detection feature the relation between the correction error cloud cover type cloud amount are shown fi 5 and features of each observation point are shown in table 2 the symbol (o) is the detected by the AVHRR cloud detection stage the symbol (D) is the point detected by the HIRS/2 cloud detection stage and the symbol (X) is the point not detected by this cloud detection system .

The images of cloud detection at (a) May 1987 13:50 (JST) and (b) 9 May 1987 18:20 (JST) are shown in fig 6 HIR/2 stage detect the central part of cloud part and AVHRR stage detect the remaining cloud area . IT means that the HIRS/2 stage canuse as a preprocessor to save processing time. Inthis paper the in situ SSTs are bulk temperature but the satellite sensors only observe the skin surface temperature it is necessary to estimate the relationship between bulk and skin temperature as discussed in ref 7 and 8 by these papers the difference is 0.1 ~0.5K is usual.

References
  1. Mc claim , et al : " Comparative Performance of AVHRR based multi channel sea surface temperature" J. of Geophysical Research Vol.90.No.C6 pp.11587 -11601, 1985.

  2. L. Lauritson ,et. Al: "Data Extraction and Calibration of TIROS /n/NOAA radiometers" NOAA Technical memorandum NESS 107 1979.

  3. T.Aoki: "Procedure and Results of AVHRR -HIRS Picture Matching " proc of the 1st Australian AVHRR Conference pp 280-285- 1986

  4. A Werbowetzki :" Atmospheric Sounding Users Guide "NOAA Technical Report NESS 83 1981.

  5. R.W Saunders ; An Automated Scheme for the Removal of Cloud Contamination from AVHRR Radiances over Western Europe " INT J Remote Sensing Vol.7.No.7,pp.867-886,1986.

  6. C.Wachiche et. Al.: "Cloud Detection and cloud parameters Retrieval from the Satellites of the TIROS -N series" Annales Geophusicae 4,B,2,pp.207-222 1986

  7. Y.Takayama et.al.: "Measurements of the water surface Temperature at the 11µm Window region " , Papers in Meteorology and Geophysics Vol.-33, No.2 pp.79-83, 1982.

  8. I.S Robinson et. Al "The Sea Surface Thermal Boundary Layer and its Relevance to the Measurement of Sea Surface temperature by Airborne and Space borne Radiometers" Int J.Remote sensing Vol.5 No1 pp 19-45 1984.

  9. L.Lavani "AVHRR/HIRS/2 Coupling " Fifth International TOCS Study Conference 247-259-1990.

Table. 1 Relation between the satellite temperature error and sea-truth
Detection Step Amount of points Standard Deviation Correlation
Not Detected 55 4.023 0.801
HIRs/2 only 32 3.072 0.888
AVHRR only 18 4.338 0.854
All Process 10 1.199 0.971


Table.2 feature of the cloudy observation point
Observation Detected Stage Class of cloud Cloud Amount
1
2
3
4
5
6
7
8
9		 (not detected)
AVHRR
(not detected)
HIRS/2
HIRS/2
HIRS/2
AVHRR
AVHRR
AVHRR		 Low
High
Low
Low,middle,high
Middle
Low,high
High
Low
Low		 2
1
9
5
7
2
2
2
1









Fig4 relation between the satellite observation error and sea-truth





Fig.6 cloud detection image