Where is environmental statistics going? Peter Guttorp University of Washington NRCSE
Thanks Noel Cressie David Fox Mark Kaiser Doug Nychka Eric Smith Michael Stein Jim Zidek Colleagues at NRCSE
Outline Space-time (global) processes Deterministic/stochastic models Health effects Emissions modelling Statisticians in environmental decision-making
Topics I will not talk about Focus on air quality–very similar issues in water and soil quality More work needed particularly in water quality issues Ecology Social aspects of the environment standards social ecology environmental justice environmental accounting
Outline Space-time (global) processes Deterministic/stochastic models Health effects Emissions modelling Statisticians in environmental decision-making
Space-time models X(z,t) = (z,t) + Y(z,t) + E(z,t) mean + smooth + error Simplifying assumptions: isotropy stationarity (in time and/or space) separability z R 2 Richard Smith’s talk tomorrow Nonstationary spatial models Need non-separable, spatially and temporally heterogeneous processes on the globe, taking into account height Often processes operate on different scales
Vertical distribution of ozone
French precipitation data Altitude-adjusted 10-day aggregated rainfall data Nov-Dec for 39 sites from Languedoc-Rousillon region of France.
California ozone Spatial correlation structure depends on hour of the day (non-separable):
Global temperature Global Historical Climatology Network 7280 stations with at least 10 years of data. Subset with 839 stations with data selected.
Global correlations isotropic nonstationary
Outline Space-time (global) processes Deterministic/stochastic models Health effects Emissions modelling Statisticians in environmental decision-making
The issue(s) Air quality model components: Emissions Atmospheric transport Atmospheric chemistry Deposition Visualization Data from deposition monitoring Important model use: scenarios Compare model output and data Combine model with stochastic components
Model assessment Geostatistical approach Temporally and spatially data rich Bayesian melding approach Temporally rich, spatially poor Requires many model runs Approximation approach Statistical modelling of model output TIES session Thursday Needed: assessment of uncertainty about model structure
Combining models and data Mark Berliner’s talk tomorrow: Bayesian hierarchical models Data assimilation Stochastic downscaling Stochastic partial differential equations dc(t) = (q(t) – d(t)c(t))dt + c(t)dB(t) source deposition concentration
Outline Space-time (global) processes Deterministic/stochastic models Health effects Emissions modelling Statisticians in environmental decision-making
Exposure issues for particulate matter (PM) Personal exposures vs. outdoor and central measurements Composition of PM (size and sources) PM vs. co-pollutants (gases/vapors) Susceptible vs. general population
2 years, day sessions A total of 167 subjects: 56 COPD subjects 40 CHD subjects 38 healthy subjects (over 65 years old, non-smokers) 33 asthmatic kids A total of 108 residences: 55 private homes 23 private apartments 30 group homes Seattle health effects study
pDR PUF HPEM Ogawa sampler
HI Ogawa sampler T/RH logger Nephelometer Quiet Pump Box CO 2 monitor CAT
PM 2.5 measurements
Where do the subjects spend their time? Asthmatic kids: – 66% at home – 21% indoors away from home – 4% in transit – 6% outdoors Healthy (CHD, COPD) adults: – 83% (86,88) at home – 8% (7,6) indoors away from home – 4% (4,3) in transit – 3% (2,2) outdoors
Panel results Asthmatic children not on anti- inflammatory medication: decrease in lung function related to indoor and to outdoor PM 2.5, not to personal exposure Adults with CV or COPD: increase in blood pressure and heart rate related to indoor and personal PM 2.5
Modeling approach Estimate space-time field from monitoring data Estimate individual paths from population data Estimate ambient exposure from path integral over space-time field and house type infiltration estimate Estimate non-ambient exposure from predictor variables UNCERTAINTY!
Difficulties with health effects studies Most studies deal with acute effects Chronic effects potentially more serious Opportunistic studies limit power Very small health effects Model uncertainty/model selection
Outline Space-time (global) processes Deterministic/stochastic models Health effects Emissions modelling Statisticians in environmental decision-making
Emissions data in US Point sources most data from industry self-reporting daily or hourly data created from annual reports only worst offenders are required to monitor allowable emissions can depend on weather Diffuse sources traffic data heating–no data collected
Source-receptor models Mass balance equation m t = P a t m vector of mass of different components P matrix of emissions signatures each row corresponds to a source a vector of relative source contributions Observe with additive errors
Identifiability problems Need to choose chemicals so that –source profiles are distinctive –little or no chemical change in atmosphere
Generalizations Space-time dependence multiple receptors errors spatially dependent source profiles time dependent source contributions spatiotemporally dependent use air quality models to evaluate chemistry in the air advection and deposition Back-trajectories to estimate actual emissions
A different approach Time series of proportions (ignoring total mass) Allows estimation of both P and a
Outline Space-time (global) processes Deterministic/stochastic models Health effects Emissions modelling Statisticians in environmental decision-making
Multi-disciplinary research projects Are statisticians good at herding cats? Modern statistics focuses on collaboration more than consulting Collaborative research centers
How to tell the prime minister the facts Need tools to describe uncertainty concisely Need to teach decision-makers to want two numbers
International projects United Nations Environment Program Intergovernmental Panel on Climate Change 30 co-chairs and vice-chairs in three working groups. No statisticians. Global Environment Facility Scientific and Technological Advisory Panel 15 members. No statisticians STAP Roster of Experts 430 scientists. No statisticians