Prior Projects

About this project

Creation of an adaptive multimedia model determining intake fractions at local (1km grid around the emission source), regional (200km grid within the continent of emission) and continental (world divided in continents) levels. Special emphasis is given to further develop modeling of exposure in the food chain linked to highest intake fractions and high level of uncertainties. This research builds up on the Impact 2002 model, a spatial multimedia (air, water, soil, sediment) model, enabling to calculate the intake fraction, to analyze the spatial variations in toxic impacts over Europe.

Principal Investigator

Olivier Jolliet

Co-Principal Investigator

Elisabete Salay

Additional Personnel

Cedric Wannaz

Project summary

1. Introduction
   1. Emissions, fate and exposure
   2. TCDD, PCBs, DDT, PBDEs, VOCs and many other chemicals are part of our current pollution "landscape", and are increasingly grabbing the media's attention. These toxic pollutants were emitted in huge quantities during the last century by industries, power plants, vehicles, and many other sources, without necessarily considering their impact on the biosphere. Many concerns have been raised in the past few decades about the (sometimes exponentially) increasing concentrations of pollutants in the various environmental media, even in places very far from the emission points. These pollutants also enter the food chain, were they tend to biomagnify and reach huge concentrations at the top : eagles, big fish, humans and many more. These alarming facts lead to many questions about:
   • The emissions of pollutants : when are or were they emitted, where and how much?
   • The fate of these pollutants : where do they go, how do they degrade, what are the involved media, and what are the time frames?
   • The impact on the ecosystems, and at larger scale on the biosphere.
   • The exposure of the population to these pollutants : how much is inhaled, ingested, etc.?
   • The overall impact on health.
   • All these points are part of a "modeling chain", as show in figure 1.
   • This page introduces a new consistant and powerful framework, extremely well suited to address these questions : the multiscale multimedia modeling.
   1. Multimedia modeling
   2. A practical approach for studying the fate of pollutants - linking emissions and quantities (masses of pollutants) in the environment - is to build multimedia models. These consist of a set of compartments distributed within each media (air, water, soil, sediments, etc.) and a set of interaction models describing the pollutant exchanges between them. The interaction models are expressed through a system of linear differential equations called mass balance equations.

1. Fig. 2:Modeling of the exchanges between the different medias (Margni, M. (2003). Source to Intake Modeling in Life Cycle Impact Assessment. Unpublished PhD. Thesis N° 2773, Swiss Federal Institute of Technology (EPFL), Lausanne.
2. To give an idea of the complexity of these operations, the air compartments alone (we are currently using several datasets of wind speeds from GEOS-Chem1) are built on a 2° by 2.5° virtual air grid over the whole globe, with 55 vertical layers. It makes 91·144·55 = 720'000 compartments2. This dataset appears as if it would be large enough to ensure a certain accuracy, but in the Great Lakes area for example, these boxes would be too big to study the interactions between a big city and its vicinity (in this case, the city would essentially be diluted in by its compartment).
3. Whereas all the compartments can be imagined to be rectangular at a theoretical level, this over-simplification is almost never possible when studying practical problems. This leads to a particularly difficult problem : each grid in each media can potentially have a different geometry. To model the hydrology for example, the hydrological compartments should be chosen to reflect the watersheds' irregular contours, which makes the study of their interactions with rectangular air compartments totally non-trivial.
4. Without even addressing the compartments in the other media, it becomes obvious - for whom is used to dealing with mathematics and programming techniques - that we cannot analytically solve problems in all the dimensions involved3. But why should we model the whole world when, for example, only the Great Lakes region is our area of interest ? In fact, more and more studies tend to show that long range transport plays a key role in pollutant distributions. For the PBDEs, for example, it has been shown that some congeners4 can be transported over large distances, whereas others cannot. However, all the cogeners appear to be present in remote locations and mechanisms other than long range transport (through jet streams, for example) were proposed, like the Gras- shopper effect due to forest-filtering or accumulation in the snowpack, ands spring pulse effects. This example shows that we have to be able to model accurately the regions of emission, the phenomenom of long range transport, and the region(s) of concern, but also focus on local phenomena repeated over long ranges, which can act as seasonal/periodic sinks or re-emitters.
5. In addition to the spatial scales (air grid/compartments, watershed compartments,..), timescales are also important in multimedia models. For examples, the wind speed at one point averaged over one day this year is unlikely to be very representative of (correlated to) the mean wind speed on the same day last year. However, the mean wind speeds over this whole year are far more likely to be correlated to last year's mean wind speeds. Although this would imply that the mean wind speeds over one year are pretty stable, the mean wind speeds over a particular year are pretty unlikely to be very correlated to the mean wind speeds 100 years later. These considerations about three different timescales in our particular example of wind speeds, suggest that there is a time range for each phenomenom where things are likely to be correlated. For the wind it ranges probably from some weeks or months to some years or decades (the time needed to observe major climate changes, at least at our epoch). These aspects should be taken into account in our models (to estimate their validity in time), and the different time scales involved are obviously dependent on the different medias and phenomena.
6. All these considerations point out the benefits of using multiscale models. Spatial multiscale models would allow us to combine huge boxes with tiny boxes in order to find the right balance between complexity and computing power needed. The "time" multiscale would allow us to deal properly with correlation in time and the temporal validity of our predictions.
7. Multiscale multimedia modeling
8. Multiscale multimedia models are models dealing with multiple media, as well as multiple spatial and temporal scales. Taking the example of the 2° by 2.5° air grid with 55 vertical layers, the practical "spatial challenge" is to generate a new grid5 with bigger compartments in the irrelevant regions (for the problem we are studying) and smaller compartments in some strategical regions. Bigger compartments are synonymous of rough precision (but acceptable in the less relevant regions), and small compartments lead to the fine precision we need in our regions of interest. Two rescalings are then possible : one consists of increasing the compartments size (by aggregating smaller compartments for example), which means that we have a multitude of data in one compartment that we use to calculate an overall mean (or some other function) ; and the second option is to decrease the compartments size (by splitting some compartments for example), which results in new compartments that are smaller than the original dataset grid with values estimated by interpolation.
9. Multiscale modeling is the cornerstone for solving the practical problems generated by multimedia "box"-models at a global scale (interconinental exchanges, dynamics at a world level). These problems translate into numerical mathematics through systems of differential equations in hugedimensional mathematical spaces, and need to be cleverly re-expressed in some more appropriate (lower dimension) spaces (regarding to the "limited" computing power we have). This is precisely the purpose of the multiscale and its underlying theories.

Footnotes

1. GEOS-Chem is a "global 3-D model of atmospheric composition driven by assimilated meteorological observations from the Goddard Earth Observing System (GEOS) of the NASA Global Modeling and Assimilation Office".
2. This dataset contains wind speeds over the whole globe each 6h for the whole year. To be efficient enough however, we should use a 1° by 1.25° grid with 3h wind speeds (which exists but seems to be less easily available).
3. To solve these problems, we have to calculate inverse matrices and/or eigen values/vectors, which becomes tough when the matrices' dimension exceeds ten thousands.
4. Congeners of a substance are derivative or an element in the same group of the periodic table.
5. And there is potentially an interesting problem here : we can generate a new grid and then see what happens (which means calculate everything and obtain results with erros), or we can try to find the optimal grid - under some constraints - which minimizes the error.