영역 연관규칙을 위한 데이타 탐사 기법

조일래 전남대학교 대학원 1997 국내박사

데이타베이스 활용 분야가 급증하고 업무 의존도가 높아짐에 따라서 데이타베이스에 축적되는 자료의 양이 급속히 늘어나고 있다. 이러한 자료들을 본연의 업무 운영에 적용하는데 그치지 않고, 업무 현장의 특성 분석에 필요한 실질적인 근거로서 활용할 수 있다는 인식이 확산되고 있다. 따라서 대용량의 데이타베이스로부터, 미리 예측할 수 없지만 의사 결정에 유용한 지식을 효율적으로 발견하기 위한 데이타마이닝 연구가 최근 활발히 진행되고 있다. 본 논문에서는 데이타마이닝의 여러 분야 중 특히 사건들의 상호 연관 관계 탐사에 초점을 맞추고자 한다. 사건들의 상호 관련성은 연관규칙(association rules)의 형태로 표현되는데, 연관규칙이란 특정 사건 집합의 발생이 다른 사건의 발생을 암시하는 경향을 표현하는 규칙이다. 기존의 연관규칙은 주로 사건이 발생한 전체 영역에서 성립하는 사건들 간의 연관성만을 고려하고 있다 그러나, 어떤 연관규칙은 비록 전체 영역에 대해서는 신뢰도가 그리 높지 않더라도, 특정 기간 혹은 특정 영역에서 강한 신뢰도로 성립할 수 있고, 그러한 정보를 알 수 있다면 다양한 의사 결정에 매우 유용하리라고 생각한다. 따라서, 본 논문에서는 임의의 부분 영역에서 강한 신뢰도를 갖는 연관성을 영역 연관규칙(ranged association rule)이라 정의하고, 대용량의 데이타베이스로터 영역 연관규칙이 성립하는 부분영역을 탐사하는 효율적인 알고리즘을 제안한다. 먼저, 주어진 이진 연관규칙에 대하여 미리 정의된 고정된 크기가 아닌 임의의 크기이고, 강한 신뢰도를 갖는 부분영역을 탐사하는 방법을 제시한다. 제안된 탐사 기법은 데이타 자체의 분포에 근거하여 가설적인 부분영역을 설정해 가는 데이타 기반(data-driven) 검색 기법을 이용한다. 따라서, 탐사 과정에서 불필요한 부분영역의 검색을 배제할 수 있다. 또한, 중복되는 데이타베이스 스캐닝(scanning)을 줄이기 위해, 주기억장치상에 관리할 수 있는 효과적인 자료구조를 설계한다. 제안된 자료구조는 부분영역의 크기를 확장해 가는 다음 단계의 검색에 필요한 정보를 제시하며, 단 한번의 데이타베이스 스캐닝에 의해 획득된다. 영역 연관규칙의 탐사는 먼저 단일 이진 연관규칙을 대상으로 1차원 사건 발생 영역에 대한 부분 영역의 탐사과정을 제시하고, 복수개의 이진 연관규칙을 수용할 수 있는 탐사 알고리즘으로 확장한다. 알고리즘의 확장 과정에서 연관규칙들이 포함하는 사건 집합에 근거하여 관련된 규칙들을 그룹핑하고, 각 그룹에 대해 단지 하나의 규칙에 대한 탐사만을 수행함으로 알고리즘 수행 성능을 상당히 향상시킨다. 또한, 사건 발생 영역을 다차원으로 확장하여 영역 연관규칙의 적용 범위를 넓힌다. 아울러 실험을 통해, 제안된 탐사 알고리즘에 실제 업무 현장에 적용할 만한 시간 비용으로 수행됨을 보인다. As database systems are widely spread and many business applications are heavily relying on database facilities, the volume of databases are rapidly increasing. It is realized that databases can be used as actual evidence of domain characteristics, rather than only used for their own operational purposes, In this regard, data mining techniques are taking growing attention in many applications, where they discover hidden but potentially useful knowledge for decision making from large databases. Among various data mining areas, this study focuses on the discovery of associations among several events. An association rule expresses the tendency that the occurrence of some events implies the co-occurrence of other events at the same time. Previously announced researches on association rules, mainly deal with associations in the whole domain. Some association rules, however, can have very high confidence in a sub-interval or a subrange of the domain, though not quite high confidence in the whole domain. Such kind of association rules are expected to be very useful in various decision making problems. In this paper, we define a rgnged association rule, an association with hight confidence worthy of special attention in a sub-domain, and further propose an efficient algorithm which finds out ranged association rules. Firstly, we suggest a data mining method that discovers sub-ranges where given binary association rules have high confidence. Note that such subranges are not delimited by predefined boundaries. In addition, the proposed method is data-driven in a sense that hypothetical subranges are built based on data distribution itself. It implies that any unnecessary subranges are not probed in the mining process. To avoid redundant database scanning, we devise an effective in-memory data structure, where essential information for the subsequent mining process is collected through single database scanning. In the mining algorithm of the ranged association rules, we suggest the exploring process of subranges in one dimensional domain for a single binary association rule, and later extend it to accept multiple binary rules. In this phase, we identify several groups of relevant association rules based on their event sets. Since only one association rule per each group is evaluated in the mining process, the performance of the process is significantly improved. The domains of events are extended to multi-dimensional ones, and it enriches the applicability of the algorithm. In addition, our simulation shows that the suggested algorithm has reliable performance at the acceptable time cost in actual application areas.

중요도를 고려한 연관규칙 탐사 알고리즘에 관한 연구

황병웅 濟州大學校 經營大學院 2003 국내석사

Recently, a s the large-scale database building has been generalized, the Data Mining area, which analyze stored data and find helpful knowledge of existing but unexposed in database, has been spotlighted a s a new strategic technology for companies' marketing, electronic con-lmerce and etc. Especially, it is helpful for management decision making by performing the role of reassuring the existing experiential knowledge retained by companies in corporate management and at the same time by offering new information and knowledge unrecognized till now. The investigation technology for association rule in the area of Data Mining has been studied most actively, and applied to marketing, business management, and decision making of companies. This research suggested the investigation algorithm for association rule that makes possible faster detection for more helpful information by considering the relative frequency and importance of database item in the 56 mass storage database at the same time. The algorithm suggested in this research found to be better than existing one a s result of simulation test.

유전자 알고리즘을 이용한 원격탐사 자료의 감독분류 기법 연구

유성곤 전북대학교 대학원 1998 국내석사

The classification of remotely sensed imagery is important in GIS analysis. The most widely used classification methods are statistical methods such as maximum likelihood classification. In recent years the neural network is applied to classify remotely sensed imagery. The neural network is typically trained by the backpropagation algorithm. Genetic algorithm is one of optimization methods based on the concepts of genetics and natural selection. In this paper, a neural network approach to the supervised classification of remotely sensed data was presented. The training algorithm for neural network is a genetic algorithm instead of the traditional gradient descent-based backpropagation algorithm. Genetic operators are tested for implementing a genetic algorithm on the aforementioned study. The results of the error convergence and the classification accuracy using genetic algorithm are compared with the backpropagation algorithm. And the training results of these algorithms are examined varying the size of training data. The results of the study on the implement of genetic algorithm are as follows. 1. The weights in the neural network were encoded as a real numbers. The tournament selection method was better than roulette wheel selection method and uniform arithmetical crossover method was better than one point or two point arithmetical crossover methods. 2. The number of hidden-layer node in the network and parameters for genetic algorithm must be chosen through the heuristic method. The results of the comparative study on the supervised classification of remotely sensed data using the genetic algorithm and the backpropagation algorithm are as follows. 1. Overall training time of the genetic algorithm was longer than the backpropagation algorithm. Errors of these algorithms were decreased well in the end of the training. The classification result of the genetic algorithm was more accurate than that of the backpropagation algorithm. It shows that it is possible to use the genetic algorithm for the supervised classification of remotely sensed data. 2. The training using these algorithms was carried out according to the variation of training data size. In the case of the backpropagation algorithm, errors were not decreased well according to reducing the data size. But the genetic algorithm was relatively superior to the backpropagation algorithm. On the occasion of the large data size, the classification results of both algorithms were more correct than those of the small data size. Especially when the data size is 20, the classification result of the genetic algorithm was better than that of the backpropagation algorithm. So using the genetic algorithm is recommended for the supervised classification of remotely sensed data if it is not possible to get sufficient training data.

Kohonen의 SOM 알고리즘을 이용한 원격탐사 자료의 영상 분류

박헌종 경상대학교 2002 국내석사

Recently, remote sensing technology is being made rapid progress. In special, the range of using the remote sensing by satellite is gradually being expended because of its merits. and after private division are allowed to achieve the remotely sensed data, many researchers have been continue their study of processing and analysis of it form that time onward. Image classification is one of the important analysis methods in the field of remote sensing, our country, which is at a geographical disadvantage and a meteorological disadvantage, needs to research about this. Some new methods have been applied to improve classification accuracy and pass the limit that the established statistical classification techniques have had, the classification method using neural networks is exemplary. The artificial neural networks is a copy of human brain's neural activity. It is used in the classification and prediction of output variables, and has a good result in expressing relation between input and output variables with complicated data. The SOM neural networks algorithm, which can learn by using its self-organizing feature was invented by Teuvo Kohonen. As the advantage of faster learning, this model is very useful. The use of Kohonen's Self-Organizing Map algorithm helps a analyst of classification and reduce the time in satellite image processing. For the reasons mentioned above, in this study, we experiment with classification of multi spectral band data by implemented SOM method, which was invented by Teuvo Kohonen, and analyze the results in more ways than one. Finally, such a study of this will be helpful to create data for the land-use classification of remotely sensed data and to establish the foundation of GIS database which is on the rise recently.

뉴로-퍼지 알고리즘을 이용한 원격탐사 화상의 지표면 패턴 분류 시스템 구현

한종규 韓南大學校 1999 국내석사

Improvement of aerosol optical properties retrieved from visible and near-infrared measurements of MODIS and MISR

이재화 연세대학교 대학원 2011 국내박사

New, improved aerosol retrieval algorithms using visible and near-infrared bands observed from passive satellite sensors including Moderate Resolution Imaging Spectroradiometer (MODIS) and Multi-Angle Imaging Spectroradiometer (MISR) are developed in this study. To retrieve aerosol optical properties (AOP) over ocean and land, measurements of MODIS and MISR are used, respectively. These algorithms basically retrieve aerosol optical depth (AOD), fine-mode fraction (FMF), and single scattering albedo (SSA) of atmospheric aerosols, and then derive aerosol type information from the retrieved FMF and SSA. In these algorithms, a unique technique is used to create aerosol models that is, the inversion products of aerosol robotic network (AERONET) sunphotometer observations are used exclusively to extract AOP of typical aerosol types classified by FMF and SSA for each AOD bin to resolve hygroscopic growth of aerosols in addition to their size and radiation absorptivity. Using the new aerosol models, AOP are retrieved from MODIS data over the global ocean for 7 years from 2003 to 2009, which are compared with AERONET observations to evaluate the accuracy of AOD retrievals. The percentage of data within 8% error range decrease slightly from 62% to 57%, compared with the operational algorithm. However, the slope of regression line between the two variables increases significantly from 0.86 to 1.00. Moreover, the validation result is improved significantly at Anmyon station at the west coast of Korea with the Pearson coefficient from 0.71 to 0.94 and the regression slope from 0.69 to 1.09 by adopting turbid water correction. For the MISR land algorithm, AOD is retrieved in high spatial-resolution by constructing three different databases for surface reflectance. In addition to the high resolution capability, the accuracy of the algorithm shows significant improvement compared with the current operational algorithm at four different locations. The slopes of regression equations between AERONET observations and MISR retrievals are improved significantly regardless of location and selected surface reflectance database. The Pearson coefficients show comparable or better performances with the operational algorithm for dark dense vegetation (DDV) areas. The use of MISR surface BRF results in the best performance among the three different surface reflectance databases, followed by wavelength-adjusted MODIS BRDF. Moreover, the inter-comparisons of aerosol types from the MODIS ocean algorithm are performed to evaluate accuracy of aerosol type information in global coverage over ocean. Comparisons of aerosol types among the derived aerosol type, AERONET, Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP), and MODIS-OMI algorithm (MOA) show that the new algorithm detects 78% of dust aerosols observed from AERONET sunphotometer, 66% to 92% from CALIOP for AOD greater than 0.3 with different height constraints, and 81% from MOA. In case of anthropogenic aerosols, the overall agreement shows 84% between aerosol types classified from MODIS and AERONET. However, the MODIS algorithm shows limitation to detect absorptivity accurately. Taking the advantage of aerosol type retrieval from MODIS data, the developed algorithm can be used to evaluate radiative forcing of different aerosol types which has usually been performed by chemical transport model (CTM). In addition, the MISR algorithm is expected to contribute to air quality monitoring over land with its high resolution capability. Further validation of the retrieved products is required to analyze error sources and to improve the algorithm.

미리 보기를 이용한 연관 규칙 탐사 알고리즘

최영란 성신여자대학교 2000 국내석사

Retrieval of aerosol effective height using O4 band from OMI with its type detection from MODIS over East Asia

박상서 Graduate School, Yonsei University 2013 국내박사

Aerosol is one of the important elements in understanding the radiative forcing for the climate studies and monitoring air quality issues. For this reason, aerosol amount and its optical properties have been observed and analyzed extensively by using ground-based and space-borne measurements. These measurements have provided aerosol optical depth (AOD), single scattering albedo (SSA), and size parameters such as fine mode fraction (FMF), and ?ngstr?m exponent (AE) over long time period. Although limited in spatial coverage, active measurements such as Light Detection and Ranging (LIDAR), has provided observation on the vertical distribution of aerosol, which depends on transport path or aerosol types. However passive space-borne measurements still have not provided vertical height information of aerosol, quantitatively, due to the limitation of observation techniques. Recently, hyperspectral instruments onboard satellite have been launched to measure trace gas concentrations by using Differential Optical Absorption Spectrometer (DOAS) technique. In this study, radiance spectrum in the UV and visible range was analyzed to investigate its sensitivities to the changes of aerosol optical properties and its altitude by using a radiative transfer model. Furthermore, the slant column density (SCD) of oxygen-dimer (O4) was estimated by using the O4 band located at 340, 360, 380, and 477 nm from the DOAS technique through simulated spectra. Reasonable correlation was found between the difference in O4 SCD and the aerosol effective height at the 477 nm. From the error analysis, the retrieved aerosol effective height is largely affected by its vertical distribution and SSA of aerosol, and also influenced by the AOD and surface albedo. On the other hand, the errors from the amount of atmospheric gases, instrument condition, and the variation of the cross section database for O4 are found to be negligible. Overall, the total error budget for the retrieval algorithm of aerosol effective height is estimated to be about 25% for absorbing aerosol and 55% for scattering aerosol, except the error due to the change of vertical distribution of aerosol. The effective height of aerosol is retrieved in several cases over East Asia by using radiance spectrum from OMI. To determine the aerosol properties, aerosol type and AOD information are obtained from the aerosol type classification algorithm using visible and IR channels of MODIS. Compared with the LIDAR observation, the retrieved height tends to overestimate about 30%, which can be attributed to difference in the definition between aerosol effective height and that from LIDAR, the cloud contamination, spatial inhomogeneous of AOD, and uncertainty in aerosol optical properties.

Improvement of aerosol optical properties retrieval over asia using visible and UV measurements from geostationary satellites

김미진 Graduate School, Yonsei University 2015 국내박사

Two algorithms to retrieve aerosol optical properties (AOPs) over Asia were developed in this study. Continuous monitoring of aerosol amount and transport over the region plays significant role in improving the prediction of air quality change as well as in understanding radiative effect of aerosol. From the reason, a single channel algorithm using visible measurement of a Meteorological Imager (MI) onboard a geostationary orbit satellite, Communication, Ocean, and Meteorological Satellite (COMS), was developed to retrieve aerosol optical depth (AOD) over Asia. The single channel algorithm is based on widely used look-up table (LUT) approach for inversion, and clear sky composite method for surface reflectance estimation. Since the algorithm has limitation in detecting aerosol optical type, the seasonally dependent aerosol model was adopted to calculate the LUTs. The aerosol models consists of refractive indices and volume size distribution with integrated from long-term measurements of Aerosol Robotic Network (AERONET) sun-photometer. For quantitative validation of the algorithm, the AOD from the single channel algorithm was compared with the result of the AERONET direct measurement. The comparison showed good agreement especially in spring. However, accuracy of the algorithm has fundamental limitation in the aerosol type selection as mentioned. Thus, the uncertainty induced by aerosol model assumption was analyzed, and improved by modifying inversion dataset. Furthermore, an effect of existence of background aerosol was corrected to improve the accuracy of surface reflectance estimation. Similarly as other retrieval uncertainties, inversion error in the LUT approach and instrument calibration error was analyzed in quantitative manner. Besides, critical reflectance method was introduced to overcome the limitation in aerosol type selection. In spite of the advantage of the continuous monitoring of AOD change, lack of information about scattering property results in increaseing uncertainty in understanding radiative effects of aerosol over Asia. From the reason, a multi-channel algorithm utilizing near ultra-violet (UV) measurements was developed to retrieve AOD and single scattering albedo (SSA). The algorithm used the optimal estimation (OE) method to reduce the retrieval uncertainty induced by assumption of aerosol loading height. The algorithm retrieved a priori states of retrieval values by using two channel LUT approach, and the a priori states were applied to find optimized solution to minimize difference between measured and simulated spectrum. The algorithm was developed for the planned Geostationary Environment Monitoring Spectrometer (GEMS) measurement to be launched in 2018, but was tested by using the measurement of Ozone Monitoring Instrument (OMI). An advantage of the UV-Vis algorithm lies in the self-sufficiency in a priori state. However, the large sensitivity to the a priori value in the retrieval of aerosol loading height is one of limitation of the algorithm. This study focused on the improvement of the satellite measurement of aerosol information over Asia, and presents the two aerosol retrieval algorithms developed by using UV-Vis and visible measurement. After the launch of GEMS, continuous monitoring of AOPs from multi-geostationary measurements is expected to be realized. The results can contribute to estimate its effects on regional climate change, and also can be applied to improve the air quality forecast via data assimilation system.

위성원격탐사를 통한 오염기체와 에어러솔의 지역적 상관성 분석

목정빈 연세대학교 대학원 2008 국내석사

에어러솔은 그 화학적 유형별로 태양광선을 흡수, 산란 시키는 작용이 다르기 때문에, 유형에 따라 대기환경 및 기후에 미치는 영향도 다르다(IPCC, 2007). 이 연구에서는 에어러솔 유형구분 알고리즘(Kim et al., 2007)을 사용하여 에어러솔을 구분하고, 위성에서 관측한 대류권의 화학기체와 유형 구분된 에어러솔 간의 계절별, 지역적 상관성을 밝히고자 한다. 또한 이산화황(SO2), 이산화질소(NO2), 일산화탄소(CO)를 포함하는 특정 오염기체의 변화경향(trend)과 오염기체의 분포가 지역별로 어떠한 차이를 갖고 있는지 보이고자 한다.연구에서 사용한 자료는 주로 위성원격탐사 자료로, Terra 위성의 MOPITT (Measurement of Pollution in the Troposphere)에서 측정되는 CO, 같은 위성에 탑재된 MODIS(Moderate Resolution Imaging Spectroradiometer)에서 측정되는 AOD(Aerosol Optical Depth) 자료 및 산불계수 분석자료(fire counts), Aqua에 탑재된 MODIS의 AOD 자료, SCIAMACHY(SCanning Imaging Absorption spectroMeter for Atmospheric CHartographY)와 OMI(Ozone Monitoring Instrument)에서 측정되는 NO2와 SO2 농도 값들이다. 위성자료를 활용한 MODIS-OMI 에어러솔 유형구분 알고리즘(Kim et al., 2007; Lee et al., 2007)을 통해 에어러솔을 흑탄소 입자(black carbon; BC), 광물성 먼지(mineral dust), 황산염(sulfate), 해염(seasalt)으로 나눌 수 있으며, 이 연구에서는 유형구분 알고리즘을 통하여 구해진 BC와 황산염 자료를 사용하였다.1996년 3월부터 2007년 11월까지의 NO2 대류권 수밀도의 장기경향을 일차 회귀식을 이용하여 전구를 7개의 지역(아시아, 오스트레일리아, 유럽, 북아프리카, 남아프리카, 북아메리카, 남아메리카)에 대해 나타내면, 아시아 지역에서 NO2 대류권 수밀도는 0.022×1015 molec/cm2/year로 증가하며, 특히 Beijing과 Shanghai 같은 대도시의 경우 계절 효과를 제거했을 경우, 1×1015 molec/cm2/year 이상 증가하고 있는 경향을 보였다. 그러나 아시아를 제외한 다른 모든 지역에서는 감소경향을 보이며 유럽, 북아메리카 등은 세계의 주요 공업지역 및 대도시가 분포하고 있는 지역임에도 각각 -0.068×1015 molec/cm2/year, -0.061×1015 molec/cm2/year로 감소하는 경향을 보였다. 2000년 3월부터 2007년 10월까지의 CO 전체기둥 수밀도의 장기경향을 일차 회귀식으로 계산한 결과, 7개의 모든 지역에서 증가 경향을 보였다. 2004년 9월부터 2007년 11월까지의 행성경계층 내의 SO2 수밀도를 동아시아지역에 대해 일차 회귀식으로 나타내면, 중국의 공업지역을 중심으로 평균 약 0.13 DU/year 증가하는 경향을 보였다. 이와 같이, 아시아 지역은 중국의 빠른 공업화로 인해 오염기체인 NO2, CO, SO2 모두 계속 증가할 것으로 예상되며 앞으로 계속적인 감시가 필요 하겠다.오염기체의 경우와 마찬가지로 전구를 7개의 구역으로 나누었을 때, CO와 BC AOD의 상관관계를 보면, 생체질량의 연소시기인 6~8월(JJA), 9~11월(SON)에 CO와 미세모드 AOD의 상관계수가 0.5 이상이었으며, 특히 남아메리카의 JJA 시기에는 0.7로 높은 상관관계를 보였다. CO와 미세모드 AOD의 상관계수와 CO와 BC의 상관계수를 비교한 결과, 거의 모든 지역에서 CO와 미세모드 AOD의 상관계수보다 CO와 BC의 상관계수가 높았고, 특히 남아메리카의 12~2월(DJF)에는 상관계수가 2배 이상(r=0.75) 높은 값을 보였다. 동유럽에서 2006년 5월 1일부터 8일까지의 산림화재로 인해 발생한 CO와 미세모드 AOD의 상관계수는 0.58, CO와 BC의 상관계수는 0.69, CO와 산불이 발생한 지점에서의 BC의 상관계수는 0.86으로 상관계수는 에어러솔 유형구분을 한 후, 그리고 화재지역으로 국한한 후 계속 좋아졌으며, 일차 회귀식의 기울기도 0.33, 0.52. 0.94로 그 기울기도 단계적으로 점차 증가하였다.중국의 공업지역을 중심으로 SO2와 황산염의 상관관계를 SO2와 미세모드 AOD와 비교하였는데, 모두 0.5 이하의 약한 상관성을 보였지만 SO2와 황산염의 상관계수와 기울기가 SO2와 미세모드 AOD의 그것보다 DJF 시기를 제외하고는 모든 기간에서 상관성이 같거나 좋아졌다. 또한 SO2와 CO의 상관계수를 비교한 결과, JJA 시기를 제외하고는 SO2와 CO의 상관계수가 SO2와 미세모드 AOD의 상관계수 보다 높았다. 종합하여 보면, JJA 시기에는 OH기 수밀도의 증가로 SO2에서 황산염으로의 화학과정이 효율적으로 발생하여 SO2와 황산염의 상관성이 다른 계절에 비해 높으며, 겨울철에는 OH기 수밀도의 감소로 인해 SO2가 황산염이 되는 과정보다는, 난방을 위한 화석연료 사용의 증가로 SO2의 발생량이 증가하는데, 이때 BC의 발생량 또한 증가하여, SO2와 BC의 상관성이 다른 계절에 비해 높았다. Recent development in satellite remote sensing, with its global coverage now enables us to investigate correlation between aerosol and pollutant gases. Increase of pollutant gases in tropospheric atmosphere modifies chemical, physical, and climatological properties. Aerosol has different types and each type has different characteristics such as absorbing and scattering light. This study investigates correlation between aerosol and pollutant gases which are regarded as a precursor of aerosol, and their dependence on season, and region. In addition, the trend and distribution of pollutant gases are also investigated.The data which are used in this study are mostly satellite remote sensing MOPITT (Measurement of Pollution in the Troposphere) onboard Terra satellite launched in December of 1999 has observed carbon monoxide density, and MODIS (Moderate Resolution Imaging Spectroradiometer) has observed AOD (Aerosol Optical Depth). In addition, SCIAMACHY (SCanning Imaging Absorption spectroMeter for Atmospheric CHartographY) and OMI (Ozone Monitoring Instrument) have observed sulfur dioxide and nitrogen dioxide density. Black carbon and sulfate aerosols are classified through MODIS-OMI aerosol classification algorithm (Kim et al., 2007; Lee et al., 2007).To investigate the trend of pollutant gases and aerosol, globe is divided into 7 areas – North America, South Africa, Europe, North Africa, South Africa, Asia, and Australia. In all regions except Asia, NO2 tropospheric column density decreases from March 1996 to November 2007. NO2 tropospheric density in Asia increases by 0.022×1015 molec/cm2/year and by over 1×1015 molec/cm2/year in mega cities such as Beijing and Shanghai in particular. CO total column density through MOPITT from March 2000 to October 2007 increases in all areas. SO2 PBL column density also increases mainly in Chinese industrial regions and large cities by average of 0.13 DU/year.Correlation between black carbon AOD retrieved by MODIS-OMI aerosol classification algorithm and CO becomes better than that between fine mode AOD and CO in most regions. By comparing these results with MODIS fire counts, enhanced correlations are found for CO with black carbon aerosol in the region of biomass burning and wild fires. Good correlation for such case suggests possibilities that CO can be used as surrogates of BC and validate the classified black carbon aerosol from remote sensing. However the correlation between black carbon AOD and CO in South Africa becomes worse than that between fine mode AOD and CO. It is one of reasons that black carbon and CO are transported well by the trade wind and it is far from source regions. Local meteorological condition and efficiency of burning could be the other reasons. In addition, black carbon could be mixed with other particles (e.g., sulfate) through coagulation, condensation of secondary aerosol compounds and cloud processing (Mikhailov et al., 2006) and it would affect the result of MODIS-OMI aerosol classification algorithm. The correlation between SO2 and sulfate AOD in China is not good because of longer time in converting SO2 into sulfate. Comparing correlation coefficients according to season, the correlation between SO2 and sulfate in JJA was the highest and the correlation between SO2 and black carbon in DJF was the highest.