환경 빅데이터 분석 및 서비스 개발 Ⅴ = Big Data Analysis: Application to Environmental Research and Service Ⅴ

다국어 초록 (Multilingual Abstract)

Ⅰ. Background and Aims of Research
□ Continue to build up ‘Environmental Policy monitoring System’
○ Periodic repetition of environmental policy need identification, timeliness assessment and effectiveness assessment
○ composed of ‘Deep Learning Based Pollution Prediction algorithm’, ‘Real Time Environmental Text Analysis algorithm’ and ‘Issue Based Database’
□ 2021 Research Goal: Strengthening the components of the environmental policy monitoring system
○ Deep Learning Based Pollution Prediction algorithm
- Enhancement of uncertainty prediction and causal analysis
- Development of an algorithms with better long-term predictive performance in predicting particular matter
○ Issue Based Database
- Composition of issue bulletin board for carbon neutrality issues
○ Real Time Environmental Text Analysis algorithm
- Analysis of Similarities and Differences in Carbon Neutral Strategies by Country
Ⅱ. Deep Learning-based Particular Matter Concentration and Uncertainty Prediction
□ Research purpose
○ Prediction of particular matter concentration and uncertainty simultaneously based on deep-learning
- Improving the reliability of PM prediction
□ Research Method
○ Collect data from Air Korea and Korea Meteorological Administration
○ Missing value preprocessing with spatial interpolation
○ Development of a deep learning algorithm capable of uncertainty prediction
- Interpolated Convolutional Neural Network using Monte - Carlo Dropout(ICNN-MCDO)
□ Development of algorithms that can estimate uncertainty as well as high prediction performance
○ Applying Monte Carlo Dropout (MCDO) to Interpolated Convolutional Neural Network (ICNN)
○ High prediction performance reflecting spatio temporal characteristics through CNN using multidimensional arrays
- High performance in classifying high-concentration of particular matter
○ Estimate predicted values and uncertainties simultaneously even at non-observation points by expressing the entire Republic of Korea as a grid at regular intervals
○ Confirm that there is a positive correlation between the deviation of the actual value and the predict value and the estimated uncertainty
○ Providing a threshold while evaluating uncertainty in prediction of particular matter
Ⅲ. Mid- to Long-term Prediction of Ultra-fine Particles Using Graph- GRU: Focusing on South Korea
□ Research purpose
○ Considering the characteristics of ultra-fine particles and the seasonal
and geographical characteristics of South Korea
○ Development of mid- to long-term (72 hours later) ultra-fine particles prediction model based on Graph Neural Network
□ Construction of space-time 3D dataset
○ Input data: Meteorological, air pollution, and Chinese ultrafine dust concentration data
- Weather and air pollution data: Satellite-based reanalysis data provided by the European Center for Medium-Range Weather Forecasts
- Ultra-fine particles data from China is based on measurement stations
□ Establishment of ultra-fine particles prediction model
○ Build with Graph-based model
○ It consists of a stage of combining surrounding information, a graph attention stage, and a GRU stage for time series learning
□ Prediction results and application
○ Improved prediction performance compared to other deep learning prediction models
○ The prediction model developed in this study can be used to predict other air pollution and can be used as base data when establishing a preemptive air pollution response plan
Ⅳ. LEDS Document Analysis Using Text Analysis
□ Research purpose
○ Analysis of similarities and differences in carbon-neutral strategies by country through LEDS document analysis of major countries including South Korea
※ LEDS (Long-term low greenhouse gas Emission Development Strateges
○ Investing the application range and usability of text analysis results
- LEDS Document characteristics: Long document length and multiple topics in one document
- Prior research for establishing policy document DB and text analysis automation procedure
□ Process
○ LEDS document collection → Keyword frequency analysis and related keyword review → Cross-country similarity network and keyword network analysis bt country → result analysis and implication
□ Results
○ Differences in national strategies in the LEDS document
- South Korea: Emphasis on energy and target alternative resources by sector with a focus on energy management and demand
- Japan: Energy efficiency and transportation targets to be reduced. Emphasis on decarbonization keywords
- Can be applied as policy analysis data
○ Establishment of case of application of text mining for global policy issue documents
- Lay the foundation for digital transformation with examples of quantifying and using atypical data such as text data
Ⅴ. Carbon-neutral Issue Based Database
□ Built a datamap on Carbon-neutral issue
○ Development of an evaluation model for monitoring the flow and relationship of variables according to the policy input, implementation, and production process
- Building a datamap under the theme of ‘carbon neutrality’ following ‘particular matter’ in 2019 and ‘climate change’ in 2020
○ Close linkage of questions, methodologies and data for policy evaluation and analysis
□ The difference between carbon neutrality and greenhouse gas reduction
○ No significant difference in terms of concepts and policy measures
○ Carbon neutrality is focused on goal-oriented discussion while greenhouse gas reduction policy is focused on specific indicators
- Carbon neutrality emphasizes strengthening the institutional foundation, changing the reduction target to an absolute value method, realizing regional carbon neutrality and seeking ways to raise public awareness.
□ Implication
○ Despite the situation in which an active low-carbon strategy is required, the proposed policy measures are not significantly different from the existing greenhouse gas reduction policies
○ Differences in regional industrial structure, final energy consumption and policy effects
- It is necessary to establish a policy that takes into account differences in regional industrial structure and carbon emission levels
Ⅵ. The Impact of Air Pollution Long-term Exposure on the Mortality of COPD Patients
□ Research purpose
○ Advancement of research on the impact of air pollution long-term exposure on the mortality of COPD patients, conducted from 2019
○ Refinement of Interpolation Modeling of Air Pollution
- Advancement of kriging-based air pollution estimation into a machine learning-based hybrid model
□ Data and methodology
○ Interpolation variable: PM₁₀, PM_2.5, O₃, NO₂, SO₂
- Target period: 2009-2018 yr.
- 1km × 1km points for the entire area of South Korea
○ Input variables: Air pollution data from Air Korea and Seoul Information Communication Plaza, satellite-based air data, meteorological data
- Missing values of satellite data are processed through convolutional neural network analysis
□ Results
○ Interpolation modeling performance of pollutant sources (PM₁₀, PM_2.5, O₃, NO₂) excluding SO₂ is good with R² value of 0.8 or higher
- The PM_2.5 interpolation model in Seoul has an R² value of 0.92, which is superior to previous studies of 0.75 to 0.90
○ The interpolation results of this study will be used for subsequent studies of health impact of air pollution
Ⅶ. Conclusions and Academic Achievements
□ Expanding and deepening the components of the environmental policy monitoring system
○ Reinforcement of environmental policy monitoring system components by adding two particular matter prediction studies of ‘Deep Learning Based Pollution Prediction algorithm’, one study of ‘Real Time Environmental Text Analysis algorithm’ and one study of ‘Issue Based Database’
□ Academic Achievements
○ Develop an algorithm that can simultaneously predict uncertainty while maintaining high predictive performance in particular matter prediction, and develop an algorithm with better mid- to long-term prediction efficiency of 3 days or longer than the existing neural network model by using a graph model
○ Through text mining analysis of Long-term low greenhouse gas Emission Development Strateges documents, we identify low-carbon strategies for each country including South Korea and suggest the need to establish DB for policy documents
○ Promote close linkage of issues, methodologies, and data for carbonneutral policy evaluation and analysis. Development of time series analysis automation program module and analysis of policy effect
○ Lay the foundation for more improved estimation of individual air pollution exposure through the development of improved air pollution interpolation models