運用機器學習法特徵化潛在污染場址之受體暴露地圖

Using Machine Learning to Map and Characterize Exposure Zones for Potentially Contaminated Sites

10.6499/JSGR.2017.0401.004

陳怡君(I-Chun Chen)；賈皓鈞(Hao-Chun Chia)；馬鴻文(Hwong-Wen Ma)

機器學習 ； 暴露行為 ； 社會經濟 ； 土地利用 ； 土壤地下水 ； machine learning ； exposure behavior ； socio-economics ； exposure maps ； land use ； soil and groundwater

土壤及地下水污染整治

4卷1期（2017 / 01 / 01）

53 - 71

繁體中文

基於風險預防概念，土壤地下水風險評估應找尋可特徵化暴露與污染間關連性方法，台灣污染場址暴露調查少有長期與大量行為統計，區域或場址性的暴露調查無法有效洞察全台受體暴露脆弱特徵地區。機器學習能建立良好近似系統（approximation）找出特徵（Characteristic）、規則（Patterns）或關聯性（Relationship）。本研究運用機器學習探討受體暴露特徵，以大數據概念特徵化台灣居民生活型態。結合民眾生活習慣問卷、社會經濟公開資料（如自來水普及率、人口密度、教育程度、年齡、收入等）、及土地利用圖資等資料，將複雜、困難、異質性資料進行分析，找尋特徵化受體暴露潛勢區域。利用決策樹與地理圖資顯示受體暴露潛勢地圖，結果顯示土壤與地下水接觸特徵不太相同，但兩者與土地利用特徵有強度關連，建議健康風險評估暴露參數應強化與土地利用連結。地下水接觸潛勢判定結果除土地利用外，教育程度高低與自來水接管率亦具有顯著性判定依據，建議可由本研究篩選結果（教育程度比例小於30%與自來水普及率小於44%區域），優先進行地下水暴露型態調查，以降低調查上時間與經費的花費。此外暴露問卷設計應新增家戶年收入、教育程度、土地利用、自來水接管率等資料勾稽，以作為未來暴露潛勢區域劃定的參考依據。

1. Introduction Health risk assessments play an important role in dealing with soil and groundwater contamination events. However, such assessments often take a lot of time and money, especially in soil and groundwater cases, for which it is more difficult to conduct pollution survey and remediation work. In addition, variability is an inherent characteristic of a population; people vary substantially in their exposures and the induced susceptibility to harmful effects. Addressing this variability is critical for health risk assessments, and thus it should be better characterized. 2. Material and Methods A decision tree, one of the most popular methods in machine learning, is used to extract the main characteristics that affect exposure behavior. In order to define the relationship between socioeconomic information and exposure characteristics, the main database chosen includes 4,188 historical questionnaires (age, gender, weight, occupation and acceptor type) and socio-economic factors (income, tap water connection ratio, education, population density, elderly) and land-use. This study produces exposure maps including soil and groundwater contacts and six exposure types. 3. Results and Discussion The results show that the exposure characteristics of soil and groundwater are not the same, and thus need to be discussed separately. Land-use is an important characteristic of soil and groundwater exposure, especially in agricultural land and for those employed as farmers. Additionally, the tap water connection ratio is highly related to exposure types, such as inhalation and dermal exposure when taking a shower or bath. Moreover, the rate of education and income of households may help predict the exposure zone with regard to contacting groundwater. 4. Conclusion This study proposes a solution for conducting exposure assessments, and can be used as a guidance for working with potential contaminated sites. This machine learning based method can be done quickly and screen the relationship between socio-economic open data and exposure behaviors. Finally, if the exposure zones are determined to be more vulnerable, relevant policies should put in place and more resources assigned top further investigations to protect the population and achieve environmental justice.

1. Adger, W.N.(2006).Vulnerability.Global Environmental Change,16,268-281.
2. Hossain, M.M.,Piantanakulchai, M.(2013).Groundwater arsenic contamination risk prediction using GIS and classification tree method.Engineering Geology,156,37-45.
3. National Research Council(2009).Science and Decisions: Advancing Risk Assessment.Washington, DC:The National Academies Press.
4. National Research Council(2014).Review of EPA’s Integrated Risk Information System (IRIS) Process.Washington, DC.:The National Academies Press.
5. Neshat, A.,Pradhan, B.(2015).Risk assessment of ground water pollution with a new methodological framework: Application of Dempster – Shafer theory and GIS.Natural Hazards,78,1-21.
6. Pouzaud, F.,Ibbou, A.,Blanchemanche, S.,Grandjean, P.,Krempf, M.,Philippe, H.J.,Verger, P.(2010).Use of advanced cluster analysis to characterize fish consumption patterns and methylmercury dietary exposures from fish and other sea foods among pregnant women.Journal of Exposure Science & Environmental Epidemiology,20,54-68.
7. Son, J.Y.,Lee, J.T.,Kim, H.,Yi, O.,Bell, M.L.(2012).Susceptibility to air pollution effects on mortality in Seoul, Korea: A case-crossover analysis of individual-level effect modifiers.Journal of Exposure Science & Environmental Epidemiology,22,227-234.
8. Tian, N.,Xue, J.,Barzyk, T.M.(2013).Evaluating socioeconomic and racial differences in traffic-related metrics in the United States using a GIS approach.Journal of Exposure Science & Environmental Epidemiology,23,215-222.
9. Wilson, W.E.,Mar, T.F.,Koenig, J.Q.(2007).Influence of exposure error and effect modification by socioeconomic status on the association of acute cardiovascular mortality with particulate matter in Phoenix.Journal of Exposure Science & Environmental Epidemiology,17,11-19.
10. 行政院環保署(2012)。,未出版
11. 行政院環保署 (2014)，土壤及地下水污染場址健康風險評估方法。http://intra.ctsp.gov.tw/EBulletin/InternetUpload/26078_47143.pdf
12. 林軒田 (2014)，機器學習基石。http://www.ntumooc.org/single-post/2014/07/22/%E6%A9%9F%E5%99%A8%E5%AD%B8%E7%BF%92%E5%9F%BA%E7%9F%B3
13. 黃文,王正林(2014).利用 R 語言打通大數據的經脈.臺北:佳魁資訊.