Multi-scale, multimedia modeling to compare local and global life cycle impacts on human health Cédric Wannaz 1, Peter Fantke 2, Olivier Jolliet 1 1 School of public Health, University of Michigan (U.S.) 2 Institute of Energy Economics and the Rational Use of Energy, University of Stuttgart (Germany)
[ SPH, University of Michigan | IER, University of Stuttgart ] “Box” type Multimedia Models Main assumption: instantaneous homogeneity
[ SPH, University of Michigan | IER, University of Stuttgart ] Spatial Differentiation - Fixed number of grid cells - Months of work to parameterize - Higher resolution when reduced extend but still not “high resolution” - Need for global => low resolution
[ SPH, University of Michigan | IER, University of Stuttgart ] 12,960 fixed grid cells (2° x 2.5°) Global, high resolution, but how? Issue : “The number of grid cells grows faster than the resolution”
[ SPH, University of Michigan | IER, University of Stuttgart ] Drawbacks of Large Grid Cells (artifacts) Artificial dilution Assuming body of water with large residence time
[ SPH, University of Michigan | IER, University of Stuttgart ] Need for Multi-scale Grid Need for high resolution where it matters Need for multi-scale grid 5,127 multiscale grid cells
[ SPH, University of Michigan | IER, University of Stuttgart ] Potential for Grid Refinement Background grid (static) Multiscale grid (iterative refinement) Potential for refinement normalized i Δ i Df D ; or ; or any normalized ii (a +b D ) f i=1 i c+ 6 7 n : spatial dataset (raster) #i. Each raster pixel indicates a local weight for refinement (0=no to 1=max) : scalars associated with D i, that allow offset + rescale : scalar, offset normalized i D f ({D i }) a i, b i c
[ SPH, University of Michigan | IER, University of Stuttgart ] Example: Potential «North America» : high interest for refinement : no interest for refinement (prevented) (A) Two polygons (countries are super-imposed): Black polygon (drawn by hand): covering North America White background covering rest of the globe
[ SPH, University of Michigan | IER, University of Stuttgart ] Selection of power plants: Power plant Example: Potential «Plant Proximity» Power plants (B) GIS operation : multiple ring buffers around plants
[ SPH, University of Michigan | IER, University of Stuttgart ] Example: Potential «Population Count» Number of capita per raster cell: (C) This potential is not hand-made, but comes directly from a dataset (raster) of population counts.
[ SPH, University of Michigan | IER, University of Stuttgart ] Example: Total Potential Total potential = 0 + (0 + 1 * raster North America) * ( * raster proximity) * (0 + 1 * raster population) Targets for refinement: North American regions with large population and close to (a selection of) power plants.
[ SPH, University of Michigan | IER, University of Stuttgart ] Resulting Multiscale Grid Step 1: Creation of a user-defined background grid
[ SPH, University of Michigan | IER, University of Stuttgart ] Resulting Multi-scale Grid Step 2: Iterative grid refinement according to potential
[ SPH, University of Michigan | IER, University of Stuttgart ] Resulting Multi-scale Grid Step 2: Iterative grid refinement according to potential zoom in to the U.S.
[ SPH, University of Michigan | IER, University of Stuttgart ] Air Concentration [kg/m³] Example: emission from a power plant near Houston: 1,2-Dichlorobenzene (CAS: , half life in air: 21.1 [days]) Kg/m 3 Cities > 1mio Power plants
[ SPH, University of Michigan | IER, University of Stuttgart ] Local Studies Intake at Different Scales LC(I)A studies
[ SPH, University of Michigan | IER, University of Stuttgart ] Global modeling with high resolution at specific places Example: compare intake in vicinity of emission source with global intake some % of intake in emission cell local study misses most of impacts global study misses adequate resolution Grid adjustable to data availability, user interests, etc. Evaluation of grid characteristics via sensitivity study Conclusions for Environmental Scientists
[ SPH, University of Michigan | IER, University of Stuttgart ] Conclusions for SGM 2010 Potential for refinement (PfR) is a very flexible solution for both GIS specialists and non-specialists to define the characteristics of the desired refined grid. A PfR is a combination of multiple contributions that can be based on any dataset => unlimited possibilities. Synergistic and antagonistic contributions can be used: some contributions can oppose to refinement. Absolute constraints can be defined => possible to limit refinement according to dataset native resolution/availability. The full modeling chain includes coded procedures (Python+ Geoprocessor) for projecting data into the grids (scalar and vector fields), and then building the mathematical objects that describe the compartmental system => possible to perform sensitivity studies towards grid variations!
[ SPH, University of Michigan | IER, University of Stuttgart ] A F.W. N.L. A.L. S A F.W. N.L. A.L. S Appendix – K matrices 1779x1779, nnz = x38521, nnz = Our basic example A more elaborate example
[ SPH, University of Michigan | IER, University of Stuttgart ] Appendix – Gridded water network WWDRII gridded water network, 0.5°x0.5°
[ SPH, University of Michigan | IER, University of Stuttgart ] Appendix – Clustering
[ SPH, University of Michigan | IER, University of Stuttgart ] Appendix – Clusters Composition
[ SPH, University of Michigan | IER, University of Stuttgart ]
Intake Fraction [kg/kg] kg/kg Population intake: Global iF = 2.98 E-5 96.6% outside of local area Local iF = 1.01 E-6 3.4% of total intake (but highest individual intake)