Validation of the simulation of human activity and pollutant exposure (SHAPE) model using paired days from the Denver, CO, carbon monoxide field study

Abstract

Listen

Translate article

The U.S. Environmental Protection Agency (EPA) developed the Simulation of Human Activity and Pollutant Exposure (SHAPE) model to estimate the frequency distribution of population exposures to carbon monoxide (CO) by computer simulation of microenvironmental concentrations and human activity patterns. To validate the SHAPE model, measured personal CO exposures from an EPA study in Denver, CO, in the winter of 1982–83 were compared with estimates generated by the model. Microenvironmental CO concentrations for the model were generated by Monte Carlo simulation based on the Denver, microenvironmental data, but the activity simulation portions of the model were modified to accommodate real activity patterns from Denver. Observed and predicted population exposure frequency distributions then were compared. A total of 899 24-h responses from Denver yielded 772 usable profiles after invalid responses were eliminated, giving 336 paired days of observations (CO exposure profiles from two successive days for the same respondent). From these data, 22 microenvironments were identified with at least 10 measurements on each of the two days. Microenvironmental CO concentrations were calculated by subtracting hourly ambient background CO concentrations. Ambient background CO concentrations were estimated by three different approaches. All three yielded similar results, with the average value from all fixed monitoring sites performing slightly better than the nearest fixed monitoring site. For nearly every microenvironment, the study found negligible differences between the microenvironmental CO frequency distributions on the 2 days. The microenvironmental CO frequency distributions for Day 1 provided the basis for SHAPE model estimates of Day 2 exposure profiles, and the activity patterns were based on the Denver diaries for Day 2 (the observed times at which people entered and left each microenvironment). CO exposure profiles were calculated using Monte Carlo sampling from the Day 1 microenvironmental CO concentration distributions and adding the estimated ambient background components. The predicted frequency distributions of the 1- and 8-h maximum average CO concentrations agreed reasonably well with the observed frequency distributions. Mean values were quite similar, but the variability in the observed values exceeded the variability in the predicted values, which may be attributable to serial dependencies in a person's activities during a 24-h period and autocorrelation of microenvironmental concentrations and the finite nature of the distributions from which microenvironmental concentrations were sampled.