You are here:
Metamodels for Ozone - Comparison of Two Techniques
Citation:
Hogrefe, C., R. Mathur, S. Porter, E. Gego, AND S. Rao. Metamodels for Ozone - Comparison of Two Techniques. 34th International Technical Meeting on Air Pollution Modelling and its Application, Montpellier, FRANCE, May 04 - 05, 2015.
Impact/Purpose:
The National Exposure Research Laboratory’s Atmospheric Modeling Division (AMAD) conducts research in support of EPA’s mission to protect human health and the environment. AMAD’s research program is engaged in developing and evaluating predictive atmospheric models on all spatial and temporal scales for forecasting the Nation’s air quality and for assessing changes in air quality and air pollutant exposures, as affected by changes in ecosystem management and regulatory decisions. AMAD is responsible for providing a sound scientific and technical basis for regulatory policies based on air quality models to improve ambient air quality. The models developed by AMAD are being used by EPA, NOAA, and the air pollution community in understanding and forecasting not only the magnitude of the air pollution problem, but also in developing emission control policies and regulations for air quality improvements.
Description:
A metamodel is a mathematical relationship between the inputs and outputs of a simulation experiment, permitting calculation of outputs for scenarios of interest without having to run new (presumably costly) experiments. Ozone metamodels are typically designed to capture a particular range of emissions for a fixed set of meteorological conditions. Here we describe the development of two metamodels based on a single 18-year simulation experiment for the Northeastern US (NEUS). The long time span of the experiment covers a wide range of emission as well as meteorological scenarios that are considered most likely for the model domain over the 18 years (i.e., they are not statistical samples). The metamodels were fit with monthly mean ozone, meteorology and emissions and validated using leave-one-month-out cross-validation. All models estimate monthly ozone for each CMAQ grid cell in the northeast US domain.The two types of metamodels investigated include projection onto latent structures (PLS) and stochastic kriging (SK). PLS is regression driven by principal components of the original predictors. The PLS model used gridded meteorology and emissions and in theory permits connection of meteorology and emissions to ozone at each CMAQ grid. SK is a spatial interpolation technique in which emissions and meteorology play the role of location coordinates. The model finds ozone values closest to a given set of emission and meteorological conditions. SK was driven with gridded meteorology and domain-wide mean emissions. Metamodel performance is described using spatial and temporal images of traditional metrics (R2, bias, absolute bias, and root-mean-squared error). Metamodel response to changes in a single emission source or meteorological variable (leaving others at their mean value) is also estimated.
