QUANTIFYING UNCERTAINTY: CALCULATING INTERVAL ESTIMATES USING QUALITY CONTROL RESULTS

摘要

EPA's Great Lakes National Program Office (GLNPO) is leading one of the most extensive studies of a lake ecosystem ever undertaken. The Lake Michigan Mass Balance Study (LMMB Study) is a coordinated effort among state, federal, and academic scientists to monitor tributary and atmospheric pollutant loads, develop source inventories of toxic substances, and evaluate the fate and effects of these pollutants in Lake Michigan. A key objective of the LMMB Study is to construct a mass balance model for several important contaminants in the environment: PCBs, atrazine, mercury, and trans-nonachlor. The mathematical mass balance models will provide a state-of-the-art tool for evaluating management scenarios and options for control of toxics in Lake Michigan. At the outset of the LMMB Study, managers recognized that the data gathered and the model developed from the study would be used extensively by data users responsible for making environmental, economic, and policy decisions. Environmental measurements are never true values and always contain some level of uncertainty. Decision makers, therefore, must recognize and be sufficiently comfortable with the uncertainty associated with data on which their decisions are based. The quality of data gathered in the LMMB was defined, controlled, and assessed through a variety of quality assurance (QA) activities, including QA program planning, development of QA project plans, implementation of a QA workgroup, training, data verification, and implementation of a standardized data reporting format. As part of this QA program, GLNPO has been developing quantitative assessments that define data quality at the data set level. GLNPO also is developing approaches to derive estimated concentration ranges (interval estimates) for specific field sample results (single study results) based on uncertainty. The interval estimates must be used with consideration to their derivation and the types of variability that are and are not included in the interval.