Chapter 4 1. MESA Air Modeling Regions
4.1 H. Ozone Modeling Regions
4.2 I. Performance Statistics for Exposure Models
Table 27. Cross-validation measures of predictive accuracy for site means at monitoring locations for likelihood-based exposure models of NO2, NOx, PM2.5, and LAC. Leave-one-out cross-validation was used for AQS and fixed sites and ten-fold cross-validation was used for home sites. New York models include NYCCAS reference sites with AQS and fixed sites and NYCCAS distributed sites with home sites. Units for RMSE are ppb (NO2 and NOX), µg/m3 (PM2.5), and 10-5m-1 (LAC). R2reg represents the regression R2. Table taken from Keller et al.21
Table 28. Number of monitors and cross-validated measures of predictive accuracy for exposure models of NO2, NOx, PM2.5, and O3 for SPIROMICS Air. Leave-one-out cross-validation was used for AQS and fixed sites and ten-fold cross-validation was used for home sites.
Table 29. Model Performance using Cross Validation at AQS+Fixed Sites for O3 Spatio-Temporal Models.
Table 29. Number of monitors and cross-validated measures of long-term predictive accuracy for exposure models of NO2, NOx, PM2.5, and O3 for the “omnibus” models combining MESA Air and SPIROMICS Air. Leave-one-out cross-validation was used for AQS and fixed sites and ten-fold cross-validation was used for home sites. City Modeling region radius (km) Number of monitors AQS and fixed sites Home sites
H/O C/S
NOx Baltimore
Los Angeles
San Francisco
New York
Salt Lake City
Ann Arbor
Winston-Salem
Chicago
St. Paul
NO2 Baltimore
New York
San Francisco
Los Angeles
Salt Lake City
Winston-Salem
Ann Arbor
Chicago
St. Paul
O3 Baltimore
New York
San Francisco
Los Angeles
Salt Lake City
Winston-Salem
Ann Arbor
Chicago
St. Paul
PM2.5 Baltimore
New York
San Francisco
Los Angeles
Salt Lake City
Winston-Salem
Ann Arbor
Chicago
St. Paul
A – AQS, F – Fixed, H/O – Home outdoor, C/S – Community Snapshot, – Regression-based R-squared, – MSE-based R-squared
Table 30. 10-fold cross-validated R2 and RMSE by year for national NO2 model. All metrics are on the square root scale (sqrt(ppb)). Year
R2
Year
R2
Year
R2
Table 31. 10-fold cross-validated R2 and RMSE by year for national satellite-based NO2 model. All metrics are on the square root scale (sqrt(ppb)).
Table 32. 10-fold cross-validated R2 by year for national PM10 model. All metrics are on the square root scale (sqrt(µg/m3)).
Table 33. 10-fold cross-validated R2 by year for national PM2.5 model. All metrics are on the square root scale (sqrt(µg/m3)). Year 2000 published in Sampson et al, Atmospheric Env, 2013.22
Table 34. Cross-validated R2 and RMSE for EC, OC, Si, and S PM2.5 Components. All metrics are on the square root scale. Published in Bergen et al, EHP 2013.23
Table 35. Cross-validation statistics of the historical PM2.5 models for 1999-2010 by year and region.
Table 36. 10-fold cross-validated R2 on native and model scale for As, Du, Ni, SO4, So2, NO3, V, and Cr. * All components except for Cr were developed using covariates from Rev 5. Cr was built using Rev 4 covariates.
Table 37. Model performance (10-fold cross validated R2 and RMSE) for PM10-2.5 mass (µg/m3) and species concentrations (ng/m3) using land use regression (LUR). Table adapted from Zhang et al, Under review at EHP, 2013.24
Table 38. Leave-one-out cross validation RMSE of PM2.5 using “Pragmatic Model” at all sites and at fixed sites on native scale (ug/m3). * N is the number of sites used for modeling. Modified from Table 2 in Sampson et al, Atmostpheric Environment, 2011. 25
Table 39. 10-fold cross-validation of National Spatiotemporal PM2.5 Model. These validation statistics are produced automatically as an output of this model. See code for additional details (step4_national_CV.R)
Table 40. Cross-validation measures of predictive accuracy for site means at monitoring locations for likelihood-based exposure models of PM2.5 in the Puget Sound. Leave-one-out cross-validation was used for AQS and Historical Nephelometer sites and ten-fold cross-validation was used for RAD monitor sites. R2reg represents the regression R2.
4.3 J. National Spatiotemporal PM2.5 Modeling Regions
Figure. Participant locations and modeling region (left), Monitoring locations (right)
4.4 K. Database location tables
Preferred table naming conventions:
For batches of participant locations: <study>_<yyyymmdd>_location_tbl
Table 41. Location table names with native id patterns
4.5 L. ACT Spatiotemporal PM2.5 Modeling Regions and Temporal Monitoring Coverage
Figure. Participant locations and modeling region (left), Monitoring locations (right)
Figure. Temporal Coverage of ACT PM2.5 ST Model Monitoring Data
4.6 M. SPIROMICS Indoor Exposure Modeling Predictions
Table 42. Indoor exposure modeling participants, dates, and cities
Table 43. Census data filled via ACS 5-year estimates
References for Census data:
Eberwein, Kris. 2019. blscrapeR: An API Wrapper for the Bureau of Labor Statistics (BLS) [R package, version 3.2.0]. https://CRAN.R-project.org/package=blscrapeR
U.S. Bureau of Labor Statistics. Consumer Price Index - All Urban Consumers (Current Series). Retrieved from https://download.bls.gov/pub/time.series/cu/
U.S. Census Bureau. Selected housing characteristics, 2009-2013 American Community Survey 5-year estimates. Retrieved from https://data.census.gov/cedsci/table?q=United%20States&tid=ACSDP5Y2018.DP05&hidePreview=false
Table 44. Additional sources for building ages by web domain