The climate change is a popular issue. People understand the concept of climate change. For instance, the polar bear on the melting sea ice is well-known image of the climate change. The climate has really be changed and it causes disasters without fo...
The climate change is a popular issue. People understand the concept of climate change. For instance, the polar bear on the melting sea ice is well-known image of the climate change. The climate has really be changed and it causes disasters without forecasting. Make it worse, the scale and frequency of disasters has been increased. To predict and prepare future climate, many researches have investigated the climate system. The computer simulation is a useful tool for this research. The goal of modeling is to guess more reliable future climate trend from many climate parameters. In this manner, recent climate models are contain properties of the sea ice. Because the sea ice is closely connected with the global energy budget of the earth and a circulation of the ocean and the atmospheric.
In fact, the sea ice an essential factor in the polar climate investigation. First of all, the sea ice and snow on surface has a high albedo. Most solar radiation energy are reflected on sea ice surface. Otherwise, the open water region has low albedo and absorbs more solar energy. In this manner, the albedo of sea ice controls the energy budget of polar region. Secondly, the sea ice acts as an insulator between the ocean and the atmosphere. In this case, the thicker sea ice is, the better insulator. Thick sea ice effectively prohibits a heat exchange between the ocean and the atmosphere. Thus the circulation of the ocean and the atmospheric are affected. Third, the sea ice controls the salinity of the ocean. Sea ice contains a salinity as a small droplet called the brine cell. It is slightly going down by its high salinity when time goes by. This falling process ends up when the salinity enters to the ocean. On the other hand, a fresh water is added into the ocean when the melting process of sea ice is occurred without brine cells. By these means, the salinity of the ocean is affected by the sea ice. Consequently, the sea ice executes important roles for the polar climate and global climate system. In these manner, measuring properties of the sea ice is important for reasonable climate system simulation. This modeling process is closely connected with the climate change prediction.
For decades, many satellite missions has been operated for monitoring the sea ice region. The Synthetic Aperture Radar (SAR) system has been included among these observations. The SAR has a good characteristics on the polar region observation. The SAR system is rarely affected by the atmospheric or weather condition. This feature can be archived by using the microwave frequency. The microwave band is located in the atmospheric window. It does not absorbed by particles in atmosphere likewise water vapor or ozone. Another advantage of the SAR mission is a wide swath. SAR system could monitor the wide area in one observation process. The other characteristics of SAR system is an active sensor. The active means that it can radiate and receive the signal. Thus SAR system can be operated in night time without sun light. In addition, it can make clear image by controlling a strength of the signal. With these advantages, the SAR system is suitable for monitoring the polar region.
Due to these observations, a lot of sea ice data has been accumulated in present. The data should be operated for extracting the parameters what we need. In this study, I want to make parameters by segmentation technique. The segmentation of sea ice has still been remained a challengeable task. The difficulty of sea ice segmentation is connected to the scattering mechanism of the sea ice in electromagnetic radiation. In the view of the microwave sensor, the sea ice surface is rough, so sea ice is represented as a heterogeneous bulk region in SAR images. In addition, the speckle noise in SAR images makes it worse the sea ice segmentation. I focused on this characteristics and tried to parameterize this feature.
The Fractal Geometry was introduced by Mandelbrot. It can measure the randomness in the nature or self-similarity. The Local Fractal Dimension (LFD) can determine a heterogeneity of the sea ice region by fractal geometry. For this reason, the LFD was adopted in the segmentation process in this study. After the LFD calculation, the sea ice pixels have high LFD value, otherwise the open water pixels have relatively low LFD value. With LFD calculation, pixels were separated by sea ice and open water. In this manner, the LFD is shown a useful tool for segmentation purpose.
In addition, the Markov Random Field (MRF) segmentation method was also utilized in this study. The MRF segmentation method is based on a probability model and uses statistical parameters of input image. MRF segmentation method uses two aspect of image. One is feature energy and another is label energy. Conventionally, the feature energy has been measured by the Gaussian maximum likelihood function. It assume that the pixels in image are followed the Gaussian distribution. By this means, the feature energy determines a similarity of initial classes and interested pixel. MRF segmentation also uses the clique potential to establish the label energy. It describes a relationship between adjacent pixels in local region. These two aspects of the image establish the cost function of the MRF segmentation. With cost function, the segmentation task is same as an optimization process. To resolve this, the MRF segmentation method has used the Expectation-Maximization (EM) algorithm in operation.
The proposed segmentation method is composed following two phases. First, the LFD of input SAR image was derived. In second phase, the LFD image was inserted to the MRF segmentation method. For comparing reason, a hybrid image was made by original SAR image and calculated LFD image. The segmentation results were resembled in both LFD input and hybrid input cases. Though, segmentation results from hybrid images were shown more similar to human interpretation. These segmentation results were evaluated by an unsupervised evaluation method for segmentation. This method uses inter-class disparity and intra-class disparity for making criterion values.
In this study, the TerraSAR-X images are used for testing the proposed segmentation method. The SAR images are taken in the Arctic sea in summer season. In the summer season, sea ice suffers melting process. So the shape of sea ice is irregular and chaotic. Even though, the proposed method could produce successful segmentation results from these SAR images. And successful segmentation results are could be converted to the ice concentration rate in that region. This calculation also performed in present study. For guarantee the result of ice concentration, ice concentration data from the Advanced Microwave Scanning Radiometer - Earth Observing System (AMSR-E) and aerial photographs were used in this study.
The result of this study shows that the proposed segmentation method could provide advanced segmentation results. The LFD could capture the characteristics of sea ice and separate sea ice and open water pixels in SAR images. Especially, the segmentation results from hybrid images were similar to human interpretations. From segmentation results, it is possible to establish the ice concentration. Ice concentration from segmentation results were more accurate than AMSR-E which used a passive sensor. The reason is originated from a difference in resolution. High resolution SAR images allows more detailed estimation of the ice concentration. The proposed method is anticipated for being utilized in other sea ice researches. in addition, an estimation method for ice concentration is going to be used for researching the accurate surface albedo. Therefore I think it is possible that the future climate change research will uses more accurate sea ice distribution and other properties with the proposed method.