Journal of Global Research in Computer Sciences
MenuISSN: 2229-371X
ISSN: 2229-371X
| T. Gomasathit Department of Physics, Faulty of Science, Mahasarakham University, Khamriang, Kantarawichai, Mahasarakham, 44150, Thailand |
| Corresponding Author: T. Gomasathit, E-mail: tosak.g@msu.ac.th, msuscienceteam@hotmail.com |
| Related article at Pubmed, Scholar Google |
Visit for more related articles at Journal of Global Research in Computer Sciences
The primary objective of this study is to identifycloud coverage over the Thailand region by using satellite data and the K-mean clustering algorithm. The successful development software tools for K-mean unsupervised classification were based software ERDAS Imagine usingthe ERDAS Macro Language and Spatial Modelling Language. The clear rate for the Thailand region was identification from cloud coverage analysis of 8,760 digital data items from meteorological satellite data on October 2010 to September 2011. The analysis of the mean rate and the standard deviation of the clear rate data for each month were related to weather conditions in Thailand.
Keywords |
| Cloud coverage, identification, satellite data |
INTRODUCTION |
| Clouds are one of the most important forces of Earth’s heat balance and hydrological cycle, and at the same time one of the least understood. It is well known that low clouds produce negative feedback while high, thin clouds produce positive feedback on the amount of radiation. The net effect of clouds, however, is still unknown, which causes large uncertainties in climate models and climate predictions [1], [2]. Cloud detection from satellite imagery is important for remote sensing retrieval algorithms. Cloud free pixels must be identified before the retrieval of the atmospheric and surface variables (e.g. land surface temperature and vegetation index). [3], [4]. Satellite Remote Sensing data, including high spatial coverage and high temporal sampling, plays an important role in monitoring precipitation in flood-prone areas and other severe weather events [5]. Many researchers have developed numerous methods of cloud detection which have been successfully used to detect clouds for satellite data. These methods include International Satellite Cloud Climatology Project [6], Automated cloud screening of AVHRR[7], Clouds from AVHRR (CLAVR) [8], the AVHRR Processing method above the Clouds, Land and Ocean (APOLLO)[9], MSG/SEVIRI cloud mask [10], Cloud Detection with MODIS and improvements in the MODIS Cloud Mask for Collection [11], Estimation of instantaneous net surface long wave radiation from MODIS cloud-free data [12], synergistic use of POLDER and MODIS for multilayered cloud identification [13], and Generation of high-resolution East Antarctic land fast sea-ice maps from cloud-free MODIS satellite composite imagery [14]. The detection of cloud coverage using a full scene of MTSAT-2 data is computationally expensive because of the significant volume of data to be handled for each scene multiplied by the need throughout the whole year. |
THE STUDY AREA |
| A. The study area |
| Thailand’s boundaries are located from latitudes 05° 37’ to 20° 27’N and longitudes 97° 22’E to 105° 37’E, with an area of 513,115.029 km2, in the warm and tropical climate region (see in Figure.1). The average annual rainfalls range from 1000 to 1500 mm for the north eastern and central parts. But on the eastern tip and the southern peninsula, rainfall is higher, averaging from 2000 to 3000 mm [15]. |
 |
| B. Data usage |
| The Multi-functional Transport Satellite-2 (MTSAT-2) is a Japanese geostationary satellite with an operation period from 2010 to 2015 [16]. It has five spectral channels, which are in the following wavelength ranges: visible channel VIS (0.55-0.90 μm), Thermal-Infrared channel IR1 (10.3-11.3 μm), IR2 (11.5-12.5 μm), Water Vapor channel IR3 (6.5-7.0 μm), and Shortwave Infrared channel IR4 (3.5-4.0 μm). The spatial resolution of the visible image is about 1x1 km2 (nadir) and the TIR image is about 4x4 km2. An example of the satellite’s infrared image above Thailand is shown in Figure 2. The MTSAT-2 has provided information to most countries in the Asia-Pacific region including |
| (1) Data for monitoring the distribution and motion of clouds |
| (2) Sea surface temperatures, and |
| (3) Distribution of water vapor. |
| Its data are currently available on the internet in several places, such as JMA’s MTSAT data website (http://dmss. tksc.jaxa.jp/sentinel/ contents/ MTSAT.html). |
 |
METHODOLOGY |
| A. Cloud covers identification method |
| To determine cloud coverage, the pixels in the image were classified into various classes according to their digital numbers. Unsupervised classification using the K-means algorithm was used to distinguish between cloud and non-cloud pixels. The K-means algorithm is a non-hierarchical clustering unsupervised classification method [17], [18], [19], [20]. An arbitrary number of clusters should be temporally selected. Using the “least distance method” number belonging to each cluster, the author assigns the cluster to the group that corresponded to the cluster mean nearest to it. At the end of the first iteration, the mean for each group is calculated and the process is repeated again. The non-hierarchical clustering method has been used for cloud, land and water area detection because it is advantageous for a huge volume of data over large areas, as it is a computer-generated automated technique. |
| Due the huge number of images to be processed, it would be very demanding on time and effort to perform the above-discussed processes on each and every image manually. Consequently, batch processing was considered as a better option to accomplish the various tasks. SML and EML, as a kind of macro language supported by ERDAS, are used to defining the client interface structure and content. It can provide some editing capabilities of the basic process [21], [22], [23], [24]. In this work, the author successfully developed a novel GUI by using SML and EML scripting language, for the K-mean unsupervised classification. In Figure 3 shows an example of the novel GUI. |
 |
| The MTSAT-2 browse image was formed using three bands (visible, mid-infrared and thermal infrared) to classify the image data. The false color composite images were generated by assigning the red, green and blue colors to three channels of a MTSAT-2 image: visible is red, mid-infrared (or water vapor channel) is green, and thermal infrared is blue. The classified image was randomly interpreted by RGB color image (false color composite). Initially, the number of classes was increased to eight to study the effect of the number of classification classes on giving more accurate results. Finally eight classes were divided into two groups, namely cloud and non-cloud, using the cluster nearest mean value technique. The percentage of the cloud pixels over the total pixels present in the image was noted. The identification of the study area over Thailand is shown in Figure 4. It is clear that there are distinct differences in cloud coverage percentage along Thailand’s boundary. |
 |
| B. Clear rate equation |
| In this study the author uses the clear rate technique (see in equation 1) as follows: |
| CL=10-(C/10) (1) |
| Where CL is the clear rate and C is cloud coverage (expresses as a percentage). |
| When C = 0%, then CL = 10, and when C = 100%, then CL = 0. A clear rate of 10 is considered as clear weather, and conversely, a clear rate of 0 is considered as bad weather (overcast). Clear rate was estimated from the non-cloudiness of the data for each pixel of each image. Clear rate was used instead of a cloud percentage for the simplicity analysis of weather conditions. The estimated clear rates for the three regions are shown in Figure 5. |
 |
| In Figure 6. Illustrates the relationship between the standard deviation and the mean value of the clear rate for each month of the year. The relative frequency of occurrence of clear rate for Thailand regions was estimated and different probability distribution function were tried for a best-fit curve, the results of which are shown in Figure 7. |
 |
 |
RESULTE |
| A. Clear rate analysis |
| Thailand experiences the effects of monsoons from the South China Sea. It experiences northeast monsoons from November to March with a large amount of rainfall due to water vapor and clouds brought by the northeast wind from the South China Sea. It experiences southwest monsoons from April to September. These monsoons are relatively drier as most of the air comes from Sumatra instead of from a large sea body. It experiences inter-monsoons of moderate rainfall and light winds during the months of March and October. |
| B. The clear rate of mean value and standard deviation |
| Figure 6 shows the relationship between the mean value and standard deviation of clear rate for Thailand regions. A high mean value for the clear rate indicates good weather and a low mean value indicates cloudy weather. A high standard deviation indicates frequently changing weather and a low standard deviation indicates more stable weather conditions. Figure 6 also shows that the Thailand region experiences a more consistent weather pattern, with a narrow dispersion range of mean clear rate values. |
| C. Probability Model |
| Different probability distribution functions (normal, lognormal, extreme, and value) were tested to find the best fitted function of the clear-rate data for each region. The interval ranges used for clear rates and relative frequency of occurrence are shown in Figure. 7. The Figure 7 illustrates only the considered probability distribution function for the Southeast Asian region. The RMS error was used to decide the best-fitted curve of probability distribution function (see in Table 1). The RMS error rates obtained from the different probability distribution functions that were tested for each of the Thailand region. One can estimate the probability of occurrence of good or bad weather at one position by using the probability distribution function. |
 |
CONCLUSION |
| The aim of this study is to identifycloud coverage over the Thailand region by using satellite data and the K-mean clustering algorithm. The successful development software tools for K-mean unsupervised classification were based on the commercial software ERDAS Imagine usingthe ERDAS Macro Language and Spatial Modelling Language. The clear rate for the Thailand region was identificatied from cloud coverage analysis of 8,760 digital data items from October 2010 to September 2011. The results would be more conclusive if more data covering a longer span of between 3 to 5 years were available. However, the result described in this study provides some preliminary information about the possibility of weather analysis from cloud cover estimates using MTSAT-2 data, and would be useful for further studies when more data will be available. |
ACKNOWLEDGMENT |
| This research was financially supported by budget fiscal year 2012 of the Faculty of Science, Mahasarakham University |
References |
|