Quantifying Indicator Uncertainty Jay Messer U.S. EPA - National Center for Environmental Assessment November 23, 2018
Importance of indicator uncertainty Uncertainty in indicators can lead to mistaken ideas about current environmental status Uncertainty in indicators can lead to mistaken ideas about environmental trends Uncertainty in performance measures can lead to mistaken ideas about program performance (e.g., were performance goals met or not?)
Importance of indicator uncertainty Reviews of EPA’s Report on the Environment 2008 - Quantitative treatment of uncertainty is essential! EPA’s Science Advisory Board Public comment (Federal Register) Interagency government review
Importance of indicator uncertainty
Prevalence of uncertainty estimates in environmental indicator reports Quantitative treatment of indicator uncertainty remains the exception, rather than the rule: Very few US environmental indicator reports include uncertainty estimates. Where they do, the information is seldom used in an interpretive manner.
EPA’s Report on the Environment-2008 Has only qualitative information on uncertainty (“limitations”) Of the 85 ROE indicators, only 15% include uncertainty estimates in their original sources.
EPA’s Strategic Plan & Performance Reports No targets or performance measures include uncertainty estimates
ROE Indicator Pilot Study Objectives Better characterize and quantify uncertainty in current state, change, and trend estimates in ROE indicators Develop the most appropriate ways to scale ROE indicators so that they are relevant and comparable at regional or sub-regional scales Determine how sensitive ROE indicators are to environmental management in the face of natural and other phenomena
Major sources of uncertainty Measurement error Modeling error Non-comparable measurements Treatment of missing data Lack of representative sampling “Elephants” (very large entities) Sampling error and statistical models
Let’s take a look at some examples
Importance of sampling error Individual sites - Brasstown Creek, NC Stream restoration
Populations of sites: Chesapeake Bay
Populations of sites – Gulf of Mexico v1 v2 v3 WQI 1 3 SQI CH BI 2 FTCI 5 All 1.8 2.4 2.2
EPA’s Report on the Environment-2008 Among the indicators in ROE08 31 are based on “inventories” 28 are based on probability samples 26 are based on other designs
Indicators used in Uncertainty Pilot Inventory High priority cleanup sites where contamination is not continuing to spread above levels of concern Population served by community water systems with no reported violations Probability Coastal water quality index National average blood lead levels Other Reported toxic chemicals in wastes released, treated, recycled, or recovered for energy use Ambient concentrations of particulate matter Fish faunal intactness
Contaminated groundwater movement at cleanup sites Type inventory Sources of uncertainty On-site measurements Determination of status Completeness Documentation
Water systems with no reported violations Type inventory Sources of uncertainty Detection of violations (measurement and sampling error) Reporting of violations
Blood lead levels Type Probability Sources of uncertainty Sampling error Collection error Measurement error Data entry and management error
Coastal water quality index Type Probability Sources of uncertainty Sampling error Measurement error Calibration error Data entry and management error Aggregation error
Ambient PM concentrations Type Other Sources of uncertainty Measurement error Missing values (days and years) Calibration error Data entry and management error Representativeness
Toxic wastes released, recycled, etc. Type Other (model-based) Sources of uncertainty Filing decisions Estimations of wastes released, etc. Data entry and record-keeping errors Elephants
Fish faunal intactness Type Other Sources of uncertainty Completeness of sample data (both site and region) Taxon identification Representativeness of species distribution
Root cause analysis: Community Water Systems Indicator Origin of Indicator Data: CWSs monitor for drinking water contaminants. CWSs report contaminant levels that exceed the maximum contamination level (MCL) and/or surface water treatment rule treatment technique (SWTR TT) violations of health-based standards to State or Tribal drinking water programs, which report the violation(s) to EPA. At a CWS: CWS samples water on a daily basis. CWS sends water samples to a certified laboratory for testing. CWS reports MCL and/or SWTR TT violation(s) to State or Tribal drinking water programs. At State or Tribal Drinking Water Program: State or Tribal drinking water program prepares a violation report. State or Tribal drinking water program submits a violation report to EPA (via SDWIS/FED). At US EPA HQ: EPA collects violation reports from states and compiles them in SDWIS/FED. EPA prompts states to resubmit inaccurate or incomplete violation reports. EPA performs triennial audits to assess the accuracy and completeness of violation reports in SDWIS/FED. To Prepare the ROE Indicator: ROE selects the data fields appropriate to include in the Indicator. ROE sums selected data fields to produce Indicator data cuts for presentation. Indicator Presented: “Populations Served by Community Water Systems with No Reported Health-Based Violations“ At a Certified Laboratory: Laboratory tests water sample to determine if any contaminants exceed the MCL. Laboratory reports results to CWS (in some cases it will bypass the CWS and report results directly to the State or Tribal drinking water programs). DATA FLOW ROE may make data processing errors. Water samples not collected using normal protocol. Water samples mistakenly not sent for testing. CWS incorrectly identifies a violation(s) has or has not occurred. State or Tribal drinking water program does not submit violation report within the 60-day window to report a violation(s) at the end of each quarter. Violation form is inaccurate. Violation form is not complete. States do not report certain types of violations on a consistent basis (i.e., SWTR TT violations). Violation reported inaccurately estimates actual population affected. Underreporting of monitor and reporting (M/R) violations, could mask MCL and SWTR TT violations. Compliance determination error. Discrepancy between state violation data and data in SDWIS/FED. Data entry problems. SDWIS/FED software limitations. Low rate of violation report resubmission by State and Tribal drinking water programs (i.e., violation report(s) not resubmitted within the 30-day window to verify and correct the violation report(s)). Color Key: Data Origin(s) Data Flow in Root Data System(s) Processing For ROE Final Indicator Uncertainty Elements Laboratory receives false positive or false negative test results. Laboratory mistakenly does not report results to CWS, or State or Tribal drinking water programs. UNCERTAINTY
Uncertainty pilot questions Are all of the major sources of uncertainty and variability identified? Can the major sources of uncertainty in current status be quantified with the data currently available? Can the major sources of uncertainty and variability in trends be quantified with the data currently available? Is the uncertainty small enough to detect change in the indicator values over time?
Uncertainty summaries Sources Status Trends Sensitivity Inventory (31) Groundwater Yes* Yes*/? Drinking water Yes ? Probability sampling (28) Coastal WQ Blood lead Other designs (26) Toxic releases No Particulates Fish fauna
Pilot Conclusions (tentative) Most sources of uncertainty in indicators can be identified It may not be possible to quantify uncertainty in status or trends of indicators not based on inventories or probability sampling The practical effects of uncertainty on change or trend detection remains largely unexplored.
A final caveat – the importance of indicator scale National trends may mask important regional, state, and local variation Smaller sample size means additional uncertainty at regional, provincial, and local scales Indicators or performance measures may require time and space scale that are “just right.”
Effect of sampling error on ability to detect a trend or achieve a target