Harvard ACE Project 2: Air Pollutant Mixtures in Eastern Massachusetts: Spatial Multi-resolution Analysis of Trends, Effects of Modifiable Factors, Climate, and Particle-induced Mortality Brent Coull, Petros Koutrakis, Joel Schwartz, Itai Kloog, Antonella Zanobetti, Joseph Antonelli, Ander Wilson, Jeremiah Zhu May 16, 2017 Harvard ACE Project 2 May 16, 2017 1 / 20
Objectives 1 Decompose high-resolution PM 2.5 mass and ground air temperature data into regional, sub-regional, and local spatial scales. 2 Conduct a spatiotemporal analysis of sub-regional and local variation in PM 2.5 mass and ground air temperature, and local PM 2.5 emissions. quantify the impact of modifiable factors identify locations of greatest impact identify lag times between implementation of a given control strategy and decreases in PM 2.5 mass and emissions Harvard ACE Project 2 May 16, 2017 2 / 20
Objectives (cont.) 3. Conduct spatial multi-resolution analysis of PM 2.5 mixtures. identify PM 2.5 elemental profiles that vary at regional, sub-regional, and local scales in Eastern Massachusetts identify modifiable factors driving urban background and local variability in PM 2.5 composition 4. Conduct an air pollution birthweight and mortality studies in Massachusetts using multi-resolution PM 2.5 mass and species data. Harvard ACE Project 2 May 16, 2017 3 / 20
Exposure Data Sources (2000-2016) 1 Satellite Remote Sensing Data on PM2.5 Mass 2 Ambient monitoring networks (AQS, Improve) 3 Ground Air Temperature Predictions 4 HSPH Boston Supersite 5 Indoor PM2.5 Samples (NAS, P01, COPD) 6 HEI-Funded Near Roadway Study 7 PM2.5 Species monitoring data Harvard ACE Project 2 May 16, 2017 4 / 20
Data Sources on Modifiable Factors 1 U.S Energy Information Administration (EIA) 2 National Transit Database 3 American Community Survey (ACS) 4 National Emissions Inventory 5 US Census Bureau 6 Climate 7 PEIRS PM2.5 emission estimates 8 Pollution Control Programs Harvard ACE Project 2 May 16, 2017 5 / 20
Published Papers Antonelli et al. Spatial multi-resolution analysis of the effect of PM 2.5 on birth weights. Annals of Applied Statistics (P). Wilson et al. Potential for Bias When Estimating Critical Windows to Air Pollution in Childrens Health. AJE (P). Wilson et al. Bayesian distributed lag models to identify perinatal windows of vulnerability in children s health. Biostatistics. Abu Awad et al. A spatio-temporal prediction model based on support vector machine regression: Ambient black carbon in three New England states. Environment International (P). Harvard ACE Project 2 May 16, 2017 6 / 20
Spatial multi-resolution analysis: Conceptual Framework Goal: Decompose daily pollution surfaces into different scales, which are representative of different sources of pollution Examine health effects, predictors, and trends at different scales Figure taken from HSPH class EH521 Notes (Annette Peters) Harvard ACE Project 2 May 16, 2017 7 / 20
Spatial Multi-resolution Analysis using Wavelets So far we have employed a two-dimensional wavelet decomposition: Have nice interpretation in terms of spatial scale Known to capture spikes in signals We extended two-dimensional wavelet methods for application to daily PM 2.5 surfaces Irregular grids and shapes Data that is not dyadic Harvard ACE Project 2 May 16, 2017 8 / 20
Applied to Satellite PM 2.5 Predictions All panels are averaged over days in 2006 Harvard ACE Project 2 May 16, 2017 9 / 20
PM 2.5 Associations with Birthweight in Massachusetts Live births from the Massachusetts Birth Registry from 2003 to 2008 Mother s residential location during pregnancy geocoded Excluded births with gestation period less than 37 weeks Total of 323,679 births Included temporal, low-frequency, high-frequency into health effect models Strongest associations with low frequency components for both pregnancy-wide and trimester-specific average exposures. Harvard ACE Project 2 May 16, 2017 10 / 20
Windows of Exposure During Pregnancy: TAE vs DLM Wilson et al. AJE 2017, in press. Harvard ACE Project 2 May 16, 2017 11 / 20
Seasonality and Confounding cor TAE 1 TAE 2 TAE 3 = 1.00 0.34 1.00 0.66 0.40 1.00 Harvard ACE Project 2 May 16, 2017 12 / 20
DLMs w/ Effect Heterogeneity Same within-window effect, β Different within-window effect, β j Different windows, w(t) Same window, wj (t) Wilson et al. Biostatistics 2017, in press. Harvard ACE Project 2 May 16, 2017 13 / 20
Prediction Models for Black Carbon: Support Vector Machines Abu Awad et al., Environmental International 2017. Harvard ACE Project 2 May 16, 2017 14 / 20
Ongoing Work and Immediate Next Steps Multi-scale PM 2.5 and mortality acute: case-crossover analysis chronic: difference in differences applied to annual rates Decompose OC predictions: Extension to curvelets Continue work on S-T models for PM 2.5 elements with new data Harvard ACE Project 2 May 16, 2017 15 / 20
Background: Wavelets Bases as Building Blocks Romeo, Horellou, and Bergh (Monthly Notices of the Royal Astronomical Society, 2004) Harvard ACE Project 2 May 16, 2017 16 / 20
Spatial Multi-resolution Analysis: Curvelets Harvard ACE Project 2 May 16, 2017 17 / 20
Spatio-temporal Modeling of PM 2.5 Elements 1 Proving difficult 2 Some element concentrations (e.g. V, Ni) are very low, noisy 3 Tried several machine learning algorithms (random forests, SVMs, etc.) 4 Novel google-based GIS covariates on building density 5 Current plan is to include large amount of new monitoring data Harvard ACE Project 2 May 16, 2017 18 / 20
Integration within the Center We propose to use P1 emissions data as inputs into our models. P2 S-T predictions and decomposition outputs to be used in P3 and P4 causal analyses. Initial discussions to use P5 models to generate data for testing performance of P2 decomposition methods. Harvard ACE Project 2 May 16, 2017 19 / 20
Reproducibility Irregular2dWavelets R package available at https://github.com/jantonelli111/irregular2dwavelets Regimes R package available at http://anderwilson.github.io/regimes/ Harvard ACE Project 2 May 16, 2017 20 / 20