2
D.2.4.1 Report on multimedia fate and exposure model with various spatial resolutions at the European level
Front page for deliverables
Project no. 003956
Project acronym NOMIRACLE
Project title Novel Methods for Integrated Risk Assessment of Cumulative Stressors in Europe
Instrument IP
Thematic Priority 1.1.6.3, ‘Global Change and Ecosystems’
Topic VII.1.1.a, ‘Development of risk assessment methodologies’
Deliverable reference number and title:
D.2.4.1 Report on multimedia fate and exposure model with various spatial resolutions at the European level
Due date of deliverable: november 01 Actual submission date: november 01, 2005
Start date of project: 1 November 2004 Duration: 5 years
Organisation name of lead contractor for this deliverable: JRC
Revision [draft, 1, 2, …]: 2
Project co-funded by the European Commission within the Sixth Framework Programme (2002-2006)Dissemination Level
PU / Public / X
PP / Restricted to other programme participants (including the Commission Services)
RE / Restricted to a group specified by the consortium (including the Commission Services)
CO / Confidential, only for members of the consortium (including the Commission Services)
Authors and their organisation:
Alberto Pistocchi
EC - Joint Research Centre
Ispra, Italy /
Deliverable no:
D.2.4.1 / Nature:
Report / Dissemination level: PU / Date of delivery:
november 01- 2005
Status: Finished – modified version after review / Date of publishing:
november 01- 2005
Reviewed by:
M.Huijbregts, 31-10-2005
J.Armitage, 25-10-2005
I.Cousins, 24-10-2005
D.2.4.1 Report on multimedia fate and exposure model with various spatial resolutions at the European level
Alberto Pistocchi, PhD
Introduction 6
The proposed modeling strategy 11
Soil 13
Water 15
Inland waters 15
Oceans 15
Atmosphere 16
The algorithms of the model 16
Degradation rates 17
Partitioning of the chemical between phases 18
Volatilization from water in the stream network, lakes, ocean and soil water 20
Gas absorption from the atmosphere 21
The Soil water budget 21
Total discharge 24
Potential Evapotranspiration 24
Precipitation 24
Correction of precipitation deficit 25
Soil texture 25
Runoff 26
Actual evapotranspiration 30
Plant transpiration 33
Back calculation of snow fall 33
Soil moisture accounting and infiltration 33
Soil erosion 34
Soil chemical budget 34
Inland waters physical parametrization 36
Freshwater chemical budget 39
Ocean physical parameterization 40
Ocean chemical budget 41
Atmosphere physical parameterization 44
Atmosphere chemical budget 45
Data requirements to run the model 48
Physico-Chemical properties of substances 48
Release estimation and the issue of monitoring data 49
Modes of entry of the chemicals into the environment 50
Spatial patterns of emissions 50
Temporal disaggregation of the emissions 52
Generic mapping of chemical distribution favourability 52
Landscape parameters 54
Model application and expected results in WP 4.4 55
Future research 56
Atmosphere component 58
Suspended sediments 58
Groundwater 58
Ocean 58
Acknowledgements 59
References 59
Summary
The report describes a modeling strategy for multimedia fate and transport of chemicals at the continental scale. The strategy grounds on available data with varied resolution and is suitable for discretization of 1 km in space and monthly time steps. All landscape and climate parameters, when applicable, are given as monthly “climatological” means, i.e. averages over the available observation period.
The computations are done using built-in Geographic Information System (GIS) functions such as flow length and flow accumulation, cost distance, zonal statistics and map algebra. The model is currently implemented in ESRI ArcGIS 9.0 at the prototypal level.
The model includes compartments of air, soil, fresh and sea water, while sediments and aerosols are not treated as separate compartment but considered through equilibrium relations with water and air respectively. Vegetation and biota are also not considered as compartments but will be considered only at the exposure level.
The multimedia calculation is decoupled according to the findings of Margni et al., 2004, which make the model computationally simpler and faster. However, this is expected not to affect the results appreciably.
The report outlines future research to be developed based on the proposed approach, and indicates typical application domains where modeling could benefit from distributed assessment with respect to current lumped or low resolution models.
A scientific theory should be as simple as possible, but no simpler.
Albert Einstein
Introduction
Multimedia models have often been limited to oversimplifications of the environmental processes governing the transport of chemicals, mainly because of the large uncertainty on the physical and chemical properties of substances, which discourage any refined computing exercise.
A brief discussion on the existing modeling approaches of multimedia fate and exposure to chemicals allows focusing on the main limitations that restrict applications in continental scale spatially explicit evaluations. One of the best-known multimedia models is the Equilibrium Criterion (EQC) model (Mackay et al., 1996). This model represents the direct translation into software of the main concepts illustrated in the classic work of Mackay (1991, 2001).
Based on similar concepts, the European Commission endorsed the use of the model SimpleBox (Brandes et al., 1996) within the decision support system EUSES, for the evaluation of existing and new chemicals in Europe (EC, 2004).
In other contexts, the CalTox model (McKone and Enoch, 2002) has been widely used for similar purposes, and proves to be largely equivalent to the SimpleBox (Huijbregts et al., 2005).
The common feature of these (and many other similar) models is to describe the environment as a set of compartments, each of which is analogous to a continuous stirred tank reactor (CSTR), and to simplistically take advection terms into account providing input of chemical from outside the area of interest.
Although for many purposes using a non spatial model is satisfactory, there are cases where neglecting spatial distribution of emissions and the advection effects is not sufficiently precautionary (Pennington et al., 2005).
One can observe that the single box models, which neglect or simplistically treat advection from nearby regions, underestimate range; moreover, although it is likely that in many cases the mass predicted with average parameters and a larger single box is comparable with the central value or mean of the range, the single boxes always underestimate peak mass.
Single box models prove to be quite satisfactory when providing estimates of mass distribution from a region which is the only source of emission in space; this is the case e.g. for emissions generated only on a continent, provided that the emission is rather uniform across the continent and that no other significant source exists that can be advected to the continent.
This case is quite important: for instance, one might want to evaluate the effect of a certain pesticide used in Europe, in terms of average concentration on the continent. In this case, although the contribution from outside might be relevant, what is of interest is the only quota related to emissions that European policy makers could control.
Another situation of the like occurs when one wants to evaluate the average regional concentration of a chemical emitted by a point source.
Although mass distribution near the emission is strongly affected by factors not described by these models (e.g. emission stack height, plume dispersion), at a certain distance it is satisfactory to use the average “box” concentration as an estimate of average mass distribution. In addition, it must be said that often the effects of advection are negligible when compared to the huge uncertainty on other terms of the problem. In most cases, the order of magnitude of concentrations is the only parameter one can estimate. These types of considerations allow an understanding of why simplified models have also been used so far for relatively advanced decision support.
Consideration of advection from nearby regions can be introduced through both spatially explicit and nested models. A well-known example of a nested model is SimpleBox, referenced above, which considers a single box within a larger system of regions at different spatial scales (from regional to the global). SimpleBox can be regarded as a spatially implicit model, in that it does not allow identification of which geographical areas are affected by a certain emission of the chemicals, but treats them in a generic way.
On the other hand, spatially explicit models allow estimation of average concentrations in geographically defined areas.
In the last years, the dramatic increase in available geographic data for both landscape and climate parameters and chemical emissions has fostered the development of spatially resolved models.
When spatial resolution of computations is low, usually variations in environmental characteristics tend to average out, and adoption of roughly selected representative or characteristic values allows the depiction of the correct orders of magnitude of outputs (Mackay, 2001).
Thus models sometimes classified as oligo-zonal (Bachmann, 2004) can be run with no effort at characterizing geographic variation of parameters in depth. Examples of such models are the ones presented in Woodfine et al., 2002; Wania and Mackay, 1995; Toose et al., 2004; Prevendouros et al., 2004)
Generally, these models rely on the classic method of solving a system of ordinary differential equations (ODEs) of first order, that represent a set of interconnected CSTRs although, in some cases, parts of the model (typically the atmosphere component) are treated according to a trajectory-based approach (e.g Bachmann, 2004).
Pennington et al., 2005, for instance, developed a model at European continental scale, which in principle allows the spatial variability of sources of pollution, heterogeneity of soil, water, plant, atmosphere properties, and intra-media advective transport to be accounted for. This is accomplished by structuring the problem in a linear system of equations, which describe the mass balance in a set of interconnected boxes, each representing the control volume of a particular medium (soil, water, air etc….) at a given spatial location. As the authors state, “Matrix algebra provides a straightforward solution to readily solve the n simultaneous differential equations for a multi-compartment environmental model (see supporting information for further discussion). Adopting the common assumption of steady state (dM/dt = 0), the mass distribution is given by inversion of a matrix of transport rate coefficients,
(1)
where (kg/day) is the vector of the emission rate Si in each compartment i, (kg)is the vector of bulk chemical masses with elements Mi, and is the bulk rate coefficient matrix (day-1).”
This formulation is very elegant, easy to implement and has a number of advantages, including the absence of error-prone iterative procedures (Pennington et al., 2005), and the possibility to quickly compute the bulk chemical masses vector for a number of source distribution scenarios; this can be done while keeping unvaried the bulk rate coefficient matrix, which only depends on the chemical considered.
The main disadvantages of the approach are twofold. First of all, advection terms require processing of flow fields in air, water and oceans in order to provide throughflow rates for each box considered in the mass balance system of equations; also for a relatively small number of boxes, this involves a number of tedious and error-prone computations that necessarily require pre-processing codes in order to automate them. Such codes need to be tailored to the specific geographic information system (GIS) used, and on the kind of flow field information available. This makes the underlying landscape data virtually unchangeable, as changing data would imply a processing comparable to model development. Second, there are limitations on the number of spatial locations that can be considered, due to computing resources requirements. In fact, the matrix dimension N is given by the number of spatial locations times the number of media considered at each spatial location.
So for a continental area of e.g. 5000 km x 5000 km, a spatial resolution of 5 km (corresponding to 106 spatial locations to be considered) and an order of 10 media would require a matrix with N = 107. Assuming that the computing time for the inversion of the matrix is approximately linear with matrix dimension N, the inversion time would be 10,000 times higher than the one so far experienced with smaller systems such as the ones considered in Pennington et al., 2004, where N is approximately 103. Also, assuming that each element of the matrix requires 12 bytes of memory (8 for the value and 4 for the location within the array), a matrix with N= 107 with 102 non-zero elements per row would require 12 GB of memory. On the basis of such considerations, N = 106 can be assumed as a reasonable upper extreme for the use of such an approach without adopting more advanced mathematical solutions or model variations. This would allow coping with a spatial resolution of the order of 30 km over the continental scale, although systems to handle large sparse matrixes exist and are used in other fields of environmental modeling (e.g. Tomei, 2005; Pistocchi et al., 2005).
Both implicit and oligo-zonal explicit spatial models can include advection but always underestimate the peaks of concentration or mass of a chemical (hot spots) within a region. Underestimation of peak mass is higher at coarser resolutions, and tends to decrease with decreasing spatial discretization unit size. For many purposes the existing spatially explicit models with a limited number of boxes interlinked by advection might be sufficient; in some cases, however, a more detailed spatially resolved model might be necessary. This might be considered under the following conditions:
- When emissions follow a clear spatial pattern, so that a uniform distribution in space is not appropriate
- When the inter-box transport processes are likely to signiciantly affect the overall distribution of mass (the fraction of mass transferred from one box to the others is not negligible with respect to the emission in the receiving box)
- When landscape and climate parameters are likely to vary greatly in space and can consequently affect the distribution of the chemical.
The choice of geographic resolution is influenced also by the consideration of the physical processes involved. As an example, having overly large computational cells does not allow the simulation of in-stream processes at the river network level, or coastal zone/estuary processes, which might be of high importance.
Research has been starting to cope with spatially explicit models of fate and transport with increasing resolution, and now a few models with resolution from a few tens of km up to 1 km are available for calculations at the continental scale (Bachmann, 2004; Pennington et al., 2005; Suzuki et al., 2004 ). However, in any case the computational effort associated with this modeling strategy is generally quite high and limits routine applications when a large number of chemicals need to be evaluated.