GISdevelopment.net ---> AARS ---> ACRS 1990 ---> Oceanography

Extraction of the sea - ice information from NOAA Satellite Imagery

Huang Runheng, Wang Qiang, Jin Zhengang
National Research center For Marine Environment Forecasts
Dahuisi 8 Beijing 100081, PRCC


Abstract
This paper introduces a method to derive the ses-ice information and some remote sensing products from NOAA operational meteorological imagery According to the difference in albedo the criterion to identify sea ice and water was deduced .The sea ice was classified on the basis of the relationship between the thickness of sea-ice and its albedo The concept of distinguishing interval was introduced to overcome the difficulty in concentration calculation of sea-ice and water mixed pixels which results from low resolution of NOAA satellite imagery And the concentration and thickness distributions averaged in the grid was derived .In addition , the apparent displacement of sea-ice was obtained by using of the maximum cross correlation method these quantitative data are useful to monitoring and analysis of sea ice and the initial data of numerical prediction model of sea-ice.

Introduction
The Bohai Sea of China is located between 370-41O N icing takes place region three months every year The sea ice makes a great impact to marine transportations and engineering facilities Mean while the sea ice has an influence on the global energy budget of atmospheric system. Therefore it is necessary that we know the situation of sea ice in the Bohai sea for both economic developing and study of east Asia weather.

The satellite technique is a new useful means for large scale monitoring of sea-ice The study have shown that the Synthetic Aperture Radar is the most ideal instrument for monitoring sea-ice [1] But it is difficult to use it operationally because of the complex technique and large capacity information Taking the advantage of continuous observations, VIS and IR scanning radiometers mounted on the operational meteorological satellites can be used to detect earth surface characteristics though they have low resolution and are influenced from this kind of satellite data. Based on our work of monitoring the sea ice this paper introduces the method to derive sea ice quantitative information from NOAA satellite data Some results have been given.

Orginal satellite data.
Before application of NOAA satellite data the following steps must to be done real time ingesting of AVHRR data pre-processing (it consists of data separations radiance calibration geographical location and solar zenith angle calculation ) Mercator projection transformation and land mark rectification . Then we can get a digital remote sensing image of bohai sea north Huanghai Sea.

There arefive channels on satellite AVHRR (advanced very high resolution radiometer) of which channel 1 and 2 are VIS and near IR channel respectively . They are used to detect solar reflective characteristics of underlying surface. By using these two channel data after calibration and solar angle correction we can get the albedo is possible to distinguish between sea ice and water and to classify the sea ice according to the magnitude of its albedo.

The spectral characteristics of sea ice and the criterion of identification and classification
We selected a series of ice images respecting different phase of sea ice Then we sampled a certain number of pixels of sea ice and water randomly and made a statistical analysis on the reflective characteristics of ice and water.

Table 1 shows the albedo of channel 1 (A1) and the ratio of channel 2 ( R ) because of reasonableness of assumption of normal distribution on most remote sensing analysis [2] according to the data on table -1 we get the probability curves of albedo of seaice and water under the assumption of gauss distribution fig 1 shows that sea and water differ greatly in spectral characteristics we can distinguish between them by using data of channel 1 and 2 According to the Bayes classification regulation[2] we used the point of intersection of probability as the criterion That is R =2.25 with 10.1% error identification probability .

Table 1. Mean and root of mean square deviation of albedo of sea-ice and water.
  sea-ice(%) Water (%)
A1 1.7 ± 4.8 8.2 ±1.4
R 1.7 ± 0.26 3.0 ± 0.5





On the basis of identification of sea-ice and water we classified the sea ice according to the difference inits albedo both the theoretical model of optical property of sea ice [3] and experimental observation [4] have shown that the albedo of sea ice increases as its thickness increases SHIROZWA hasobserved the thin and oneyear formed sea ice in Hokkaido of Japan [4] According to his observation results we deduced the curveof the thickness of ice versus its albedo shown in fig 3 and fig 4 is an example of sea ice satellite remote sensing image its result of classification in the liaodong Bay on12of February 1990 respectively The sea ice were classified into three types the thickness less than 10 cm 10-20 cm and more.






Than 20 cm those correspond `approximately rind ice grey ice respectively cording to the sea ice observation regulations of China These quantitative data are valuable for monitoring and analysis of sea ice.

The concentration of sea-ice
Sea-ice concentration is defined as the fraction of sea ice in the field of view its equivalence used in the numerical predication model is the percentage in the grid area . In order to provide the initial field for numerical prediction model of sea ice the unit area is taken to be 6' X 6' there are 63 pixels in this grid every pixel is identified by using the criterion described in the last section the number of sea ice pixel divided by 63 is the sea ice concentration in this grid.

However the image must taken in to account when the sea ice concentration is calculated the resolution of sea ice remote sensing imagery is 1.5 X 1.5 square kilometer in the Bohai Mercator projection image. It is possible that there are both sea ice and water in this is large area the albedo of mixing pixel is a mixing contribution of sea ice and water. This is one of error sources in sea ice identification in order to partially overcome the effect of mixed pixel defined twice mean square deviation of probability distribution as distinguish interval (Ri, Rw) it corresponds that the error identification probability of sea ice and water are 2.2 % respectively and the total error identification probability is 4.4 % Referring to the equation for the calculation of cloud coverage of radiometer observation we can calculate the percentage of sea ice in a mixed pixel (p);

R = p Ri + (1 - p) Rw

Where R is the measured ratio of albedo of channel 1 to albedo of channel 2 similarly the albedo of mixed pixel (A) is given by.

A = p Ai + (l - p) Aw

Substituting the measured albedo of mixed pixel (A) and the lower limit of the albedo of water (aw) we can obtain the albedo of sea ice in the mixed pixel (Ai) According to the relationship between the thickness of sea ice its albedo the thickness of sea ice in this pixel was derived We averaged the concentration and the thickness of sea ice of the all pixels the gird average concentration and average thickness of sea ice of all pixels in very useful numerical predication model.

The apparent displacement of sea ice.
We have the template matching method to track the motion of sea ice [5] Two adjacent sea ice remote sensing images (1 and 2 ) were used we selected asub image A in image 1 and we called it "template" in image 2 we can selected a sub image B which has same size and position with sub image A- The similarity between sub image A and B can be measured by their correlation coefficients Inside argion ( i.e the searching region dotted line rectangle in fig 5 ) in image 2 we move the template step by step and get a correlation coefficient matrix if the feature of sub image A has been changed obviously when the template move to some position the correlation coefficient will appear to be a maximum the displacement of template a is simple the difference of position between sub image B and C this placement vector is the apparent displacement of sea ice in template A during dt.

We have used two tracking methods manual and automatic shows the automatic tracking result for febuary 5-6 1989 the advantage of this method is that we can obtain the displacement field distribution if sea ice inn grid point . These data can be used to examine the displacement result of sea ice numerical forecast model and from those we can know the general situation of sea ice placement in whole region is the manual tracking result for February 9-10 1989 Although the displacement field distribution is not. Well distributed like the results ofautomatic method they can show the sea ice displacement more objectively . By using method we can keep away the cloud manually to derive seaice motion in clear region.

Summary
On the basis of operational monitoring of sea ice using NOAA satellite data we have studied the methods for deriving information the products include the classification the concentration distribution the thickness distribution and the apparent displacement to suit for different user.

It should be noted that there is a difficulty in comparison of these quantitative remote sensing products with ordinary sea ice observation data firstly almost all the data ofcoastal observation are obtained by observer's eyes and limited within the limits of distance off seashore secondly ice breaker only takes the limited wioth the limits of distance off now we are collecting the aerial remote data measurement of typical sea ice sampling examination and correction according to the forecast application of these two. Consistent with experience and analysis of forecasters.

Reference
  1. F. Carsey 1989 : Review and Status Sensing of sea ice IEEE journal of oceanic Eng 14,127 138.

  2. P.H Swain & S.M Davis remote Sensing the quantitative Approach McGRAW -Hill Inc 1978.

  3. T.C Grenfell 1983 A Theoretical Model of the optical properties of sea ice in the visible and near infrared J.Geophys Res 88 c14, 9723 -9735.

  4. MASUZAWAJIOTARO physical oceanography v.4.

  5. R.M Nininis, 1986 Authomatated Extraction of Pack Ice Motion from AVHRR imagery geophys Res ,., 91 c9 10725 -10734