Overview of MEPs big data applications Presentation Outline

Overview of MEP’s big data applications

Presentation Outline I. Background II. Plan and Progress III. Key Programs/Projects IV. Questions 2 2

Background • The Chinese Government attach great importance to the role of big data in promoting the development of ecological civilization. • The State Council has released the Outline of Action Plan for Promoting the Development of Big Data, which makes arrangement for the activities on big data. • In the chapter on “Information Disclosure and Public Participation”, the new Environmental Protection Law requires promoting the application of information technology in environmental protection. • MEP Minister Chen Jining hopes applying big data to enhance synergy of relevant work and promote transformation of environmental management. 3 3

Environmental Information Center (EIC) of MEP • In charge of the collection, storage, maintenance, analysis, publication and management of national environmental data and information. • Having more than 50 staff in 6 divisions, including Technology, Website, Network, Application Development, Information Security and General Affair. • Establishing and maintaining the National Environmental Information System (NEIS) for Digital Environmental Protection 4 4

Presentation Outline I. Background II. Plan and Progress III. Key Programs/Projects IV. Questions 5 5

Three objectives Systematic and comprehensive Scientific decision making Open & innovative Accurate & reliable Accurate supervision Synergy Rational & effective Big Data Public service People oriented 6 6

Six Tasks 1. Promote integration and sharing of data resources 2. Strengthen scientific decision making in eco-environment 3. Make innovation in the mode of supervision on eco environment 4. Improve public services for eco environment 5. Develop a big data platform 7 7

Task 1: Promote integration and sharing of data resources Improve capacity in obtaining data resources 01 02 Construct infrastructure for big data Strengthen integration of data resources 03 04 Promote sharing of data resources Promote public access to eco environment data 05 8 8

Task 2：Strengthen scientific decision making Driven by Way of decision making experience Comprehensively judge environment situation Analyze environmental load and carrying capacity Driven by data quantification Predict and give early warning of environment risks Assess the performances of key work 9 9

Task 3：Make innovation in the mode of supervision on eco environment Early warning E Q Environment Eco environment Video monitoring Satellite remote sensing Energy EIA review New pollution sources of approved construction projects release pollutants to the environment and affect the environment Mag of sources Supervision Pollutants On-line monitoring Pollution source Accurate supervision Scientific monitoring & analysis Three dimension supervision APP Environmental complaints by letters & visits 10 10

Task 4: Improve public service for eco environment Improve on-line services • • • Improve service quality for disclosing information • Enhance government comprehensive service capacity • • • Construct the e-administration hall and develop the mode combining both on-line and off-line services. Develop the “one stop” platform for on-line review and approval Strengthen inter-department sharing of administrative review and approval data and support combined review and approval. Disclose more environment information and improve the quality and timelessness of disclosed information. Expand the interactive channel between government and the public, create convenience for access to environment information and improve public participation. Employ industry and social data to develop environmental information service products. Develop convenient applications including APP, enrich service contents and expand service scope. 11 11

Task 5：Develop a big data platform Big data management platform Big data application platform Collection, transmission, management, analysis, sharing and access of data resources, which provide unified data support to the application of big data. Innovative applications such as comprehensive decision making for eco environment, environmental supervision and public services. 12 12

Presentation Outline I. Background II. Plan and Progress III. Key Programs/Projects IV. Questions 13 13

Analysis and prediction of particle size of urban air pollutants based on big data (I) Weather Traffic Historical air quality data Human activities Real-time air quality monitoring report POIs Road network Air quality of local urban areas 14 14

Analysis and prediction of particle size of urban air pollutants based on big data (II) • Partition a city into disjoint grids Extract features for each grid from its affecting region – – – • Meteorological features Traffic features Human mobility features POI features Road network features Co-training-based semi-supervised learning model for each pollutant – Predict the AQI labels – Data sparsity – Two classifiers 15 15

Analysis and prediction of particle size of urban air pollutants based on big data (III) 16 16

Demonstration • http: //urbanair. msra. cn Urban Air Windows Phone App • MEP Information Center big data model on temporal and spatial distribution of local key urban air pollutants in Beijing-Tianjin. Hebei region http: //kqzl. chinaeic. net: 8080/u. Air/p/air-state 17 17

• Ground surface PM 2. 5 concentration with high temporal and spatial accuracy is estimated with satellite remote sensing monitoring data on the thickness and scope of haze combined with monitoring data of ground monitoring sites • At present, ground air quality monitoring sites in China are established based on administrative region with very limited amount of monitoring sites. They cannot reflect spatial distribution of pollutants and are easy to be affected by the environment surrounding monitoring sites. • Satellite remote sensing technology could offset the regional limit of ground monitoring and make real-time monitoring of each inch of land. In addition, it could help real-time calculation and display PM 2. 5 concentration across the whole country. 18 18

http: //54. 191. 242. 127: 8200/ 19 19

Weather and pollution analytics aims to investigate the correlations between API (Air Pollution Index) and the atmosphere states of cities in China. Both the air quality and the weather data are measured by monitoring sties distributing over the whole country. We gathered the big data, and then employed various machine learning methods to analyze them in distributed computing platforms, such as Spark. After the exploration of big data, we can use the knowledge to forecast the air quality and further provide decision support to our government. Raw Data ETL The Correlation Features between Air Quality and Atmosphere States of Cities Filter by Fields Significance Test Hypothesis Testing of Data Distribution Correlation Analysis 20 20

Weather and pollution analytics This chart shows the correlations between PM 2. 5 and wind speed at ground in 370 cities of China. The warm color and cool color indicate positive and negative correlation respectively. The correlations of cities have an obviously regional distribution. 21 21 21

Backward Trajectory Analytics Regional pollutant sourcing and pathway analysis (national wide and city based), especially for high pollution events Typical big data analytics use case for AQM on pollutant sourcing between major cities, especially for AQM guidance on finding out major pollution pathway and sources for high pollution events. E. g. , Haikou (tourist city, no local pollution sources) high pollution events during 2014 New Year holidays, the local gov. desire to find out the pollution sources. Pollution injection pathway cluster for Haikou city during 2014 new year holidays Guangdong 75% Zhejiang 12. 5 % Hunan 12. 5% 22 Haikou Based on our sourcing results, two major pollution pathways found for Haikou city (northeast straight flow and west golf circulation) West pacific subtropical high 22 22

Backward Trajectory Analytics Beijing high pollution events during 2015 National Day: the pollution cause can be analyzed based on backward trajectory (with passedcity’s PM 2. 5 concentration) at beginning time of event Based on our results, major pollution transport pathway found to cause the high pollution event for Beijing city at beginning Shandong -> Hebei -> southeast of Beijing 10 -04 20: 00 PM 2. 5 Pollution transport pathway at beginning time of event Beijing: PM 2. 5 207 (10 -04 20: 00) Heng. Shui: PM 2. 5 189. 5 (10 -04 05: 00) 23 23

Case Study on the control of water pollution at river basin level --- TMDL Model (I) Definition: • Total Maximum Daily Load refers to the daily maximum load of certain kind of pollutant that water body can accept to meet water quality standard Application scope: • It is applicable to contaminated water body. With meeting water quality standard as the aim, it includes point sources and nonpoint sources and is a key link of the framework for water quality management 24 24

Case Study on the control of water pollution at river basin level --- TMDL Model (II) Big data model is employed in prediction of surface water concentration at river basin level, TMDL allocation and so on. In many circumstances, it has better simulation accuracy and applicability compared with traditional hydrology and water quality models and is applied in the management of water quality of several river basins of the United States. 25 25

Case Study on the control of water pollution at river basin level --- TMDL Model (III) v Big data model developed by USGS could simulate BOD and NH 3 discharge of point sources and DO discharge of non-point sources. v It provides support to the granting of pollution discharge license of the region. 26 26

Background: The condition of pollutant sensors used by factories is hard to examine and maintain. The reported sensor readings often consist of a significant amount of erroneous data which degrades its overall quality. Careless usage of such sensor readings may mislead the conclusions and misguide the decisions made for environment protection. Solution: Aims to develop a ubiquitous framework for detecting, filtering, and pattern mining the anomalies in the time series data captured with pollutant sensors. Challenges: • Lack of precise ground truth labels on the anomaly data • Capturing and distinguishing among various types of anomalies including that caused by device failures, forged readings, and abnormal manufacturing activities, etc. 27 27

Case study: Industrial pollutant emission automatic monitoring data Time Series Analysis Pollutant Sensor Normal Data Anomaly Detection Time Series Data • • • Forged readings Device failures Extreme values Mean shifts Pattern changes … Trend analysis Anomaly Data Filter Abnormal Data Pattern mining 28 28

Thanks a lot!