Air pollution has a number of detrimental effects. It causes: -Problems to human health (or at least exacerbates existing problems) -Reduction of agricultural.

Health Impact Models

Air pollution has a number of detrimental effects. It causes:- Problems to human health (or at least exacerbates existing

problems)- Reduction of agricultural yield (particularly, ozone)- Damage to structures (particularly, PM)- Acidification of water bodies- …

We will examine only the impact on human health (HIA – health impact assessment).

This is normally measured in terms of- Mortality (YOLL – year of life lost in a population)- Morbidity (number of cases of different pathologies in different

population groups)

The overall impact on a population is considered to be the product of two main factors: 1) some indicator of air quality 2) some indicator of the population exposure.

The Impact Pathway Ap-proach

Since the conclusion of the EU EXTERNE Project (www.externe.info) in the ’90s, the impact pathway approach has become the standard method to evaluate health impact. It consists of the following steps:

• Emission identification

• Dispersion modelling

• Health impact evaluation

• Monetization

Dose-response functions

They are based on extensive studies of population living in North America cities.

They may refer to long-term effects and consist in following a population or a cohort for tens of years in different pollution conditions.Health effects may be measured in number of additional cases (in a given population) or in increment of Relative Risk (RR).

Dose-response functions

They may refer to short-term (few days after a severe air pollution episode) and are generally studied by counting the hospitalization for different causes immediately after the episode.Most problem refer to respiratory and cardiovascular diseases.

Dose-response functions - 2

They are based on extensive studies of population living in North America cities under the assumption that the effect of each pollutant can be distinguished.(DALYs = disability-adjusted life years)

Dose-response functions - 3

The standard assumptions are:

- The number cij of new cases of disease i per unit increment of the pollution indicator Pk of pollutant k (e.g. yearly PM10 average) is CONSTANT (i.e. linear)

- Only the RESIDENT population is considered (no special assumption on the exposition, no consideration for commuting,…)

- The population can be divided into classes Cj (children, elderly, asthmatic, all) for each of which the number of cases is known.

Note that these impact factors have been measured in CURRENT pollution conditions and that they may not be valid in new situations.Thus, the total impact (number of cases in a population) is:

The overall impact on a territory (region, state) is the sum of those computed for the component units.

j

jkijik CPcI

Why residents?Average time spent in each microenvironment

However, commuting is not negligible in some areas (e.g. Lombardy)

Dose-response functions - 4

Example of the cij coefficients for PM10 (Newext, 2006)

Impact of a single pollutant

Census data (NUTS 3) coefficients

X

unità

BU-Aa 0,163 casoC-Aa 0,335 day

LRS-Aa 0,061 day

BU-Ca 0,078 casoC-Ca 0,267 day

LRS-Ca 0,103 day

CHF-Oa 1,85E-05 caso

CC-Cna 2,07E-03 episodio

RAD-Ana 0,025 dayCB-Ana 4,9E-05 caso

CM 0,26%RHA 2,07E-06 casoCVA 5,04E-06 caso

YOLL 4,00E-04 YOLL

30+/TOT 0,6991

ricoveri per problemi cardiocircolatori PM10

NON ASMATICI

INTERA POPOLAZIONE

[unità/y*p*(μg/mc)]

giorni di attività ridotta per PM10bronchiti croniche per PM10

mortalità cronica per PM10ricoveri per problemi respiratori PM10

BAMBINI

ADULTI

uso di broncodilatatore per PM10tosse per PM10

sintomi di affaticamento nella respirazione PM10OLTRE I 65 ANNI

anni di vita persi

coefficiente per il calcolo degli YOLL complessivi

ASMATICIADULTI

uso di broncodilatatore per PM10tosse per PM10

sintomi di affaticamento nella respirazione PM10BAMBINI

tosse cronica per PM10

infarto per PM10

=

IMPACTS

The multi-pollutant prob-lemGeneral formulation still missing.

An empirical rule:- PM2.5 = 70% of the impact of PM10- Nitrates = 50% of the impact of PM10- Sulfates = 100% of the impact of PM10- Ozone = different effects, but high respiratory hospital

admissions

For more than one pollutant, two extreme cases can be assumed:- Since the relation is linear, the effect are considered to

be additive,- “winner-takes-all”, only the most relevant effect is

considered.

In principle, the presence of more than one pollutant may exacerbate all the impact in a nonlinear way (more than the sum), but this hasn’t been demonstrated.

The multi-pollutant prob-lem - 2

It is important to understand that all the number of cases computed with the linear formulation (or even a nonlinear one) must not be interpreted as an actual impact of air pollution. E.g. one can measure exactly the number of respiratory hospital admissions, but cannot be sure that they are due to air pollution.

What has been proved is that there is a (strong) correlation between some pollution indicators and some heath effect.

So, all health related impacts must be considered as different INDICATORS of air pollution. This is particularly true for mortality: there is a definite correlation between PM10 concentration and mortality, but also between temperature and mortality and temperature and PM10 concentrations.

Actual deaths cannot be attributed to air pollution (can be attributed to cancer, heart disease,…which in turn could be caused by air pollution). This is why a different way of measuring mortality (YOLL) is preferable.

Monetization

To sum the overall burden to society (all diseases + mortality) of a single pollutant,one may try to transform all the impacts in (external) costs.

These do not represent actual market values nor a real exchange of money, rather the “value to the society” of avoiding a certain impact.

They can be measured in different ways:- Expressed preferences (a sample of the populations is asked

about its “willingness-to-pay” to avoid a certain impact)- Revealed preferences (the “value to the society” is indirectly

estimated by looking at how much is spent to “buy” a similar impact). E.g. the value of a certain air pollution can be derived from the price difference of two similar houses in areas with different pollution values (hedonic prices).

In both cases, a part of the value may remain unexpressed (the value of a cough to me is higher that what I am willing to pay or I actually spend to cure it).

Monetization - 2

The ExternE project determined a set of “reference” values for the monetization of diseases and mortality.

Monetization - 3

The overall monetary impact Mk of a pollutant k on a population is:

where mkij is the monetary value of impact i on population group j.

When all external costs are expressed in monetary units, it appears natural to sum all those related to different pollutants (but remember what already said about the multi-pollutant case).

k j i

kijjkijk mCPcM

Again, these monetary values must be interpreted as indicators: they do not correspond to ANY real exchange of money and may serve only to compare different situations.

Integrated models - GAINS

GAINS(GHG-Air

IIASA’s GAINSoptimization model

PM

pollution INteractions and Synergies )

GHGsNOx VOCNH3

SO2

Health Eutrophication Acidification Ozone Policytarget on

GHGemissions

Policy targets on air quality

GAINS - 2

GAINS - 4

Regional Integrated Assessment Tool (RIAT)

It is more efficient to address local problems than aggravating general European regulations

Results of the GAINS model (IIASA, 2013) over Europe (at regional scale) show that there are areas where compliance is difficult to achieve

Develop regional/local tools

RIAT - 2

CTM selection

Local data

CTM calibration

Design of experiments

CTM simulations

Optimization problem definition

Surrogate model

selection

Surrogate model

training

Optimization problem solution

Determine the trade-offs between costs (industrial+external) and air quality

RIAT - CTM

Discretization on a regional domain 5x5 km2

Calcultion of selected general Air Quality/ Pollution Indexes

RIAT – Surrogate modelling

Air Quality Index Neighborhood Pollutants

PM10, PM2.5 4 cells NH3, PM10, PM2.5, SO2, NOx, VOC

AOT40, SOMO35, NO2, MAX8H

16 cells NOx, VOC

ANNs inputs:

quadrant precursor emissions

ANNs output:

AQI

RIAT – Design of Experiem-nts

•B (business-as-usual) CLE + 10% (2010)•L (low): emissions (CLE+MFR)/2 ( 2015) •H (high): emissions MFR -10%( 2020)

RIAT - Surrogate models (NN)

AQI Neurons Hidden Layer

FunctionHidden layer

FunctionOutput layer Radius [Km] Precursors

PM1020 Logsig Tansig 24 NOX,VOC,pPM10,pPM25, SO2,NH3

PM2520 Tansig Tansig 24 NOX,VOC,pPM10,pPM25, SO2,NH3

NO220 Logsig Tansig 84 NOX,VOC

AOT4020 Tansig Purelin 84 NOX,VOC

SOMO3520 Logsig Purelin 84 NOX,VOC

MAX8H20 Logsig Purelin 84 NOX,VOC

Es. YEARLY AVG. PM10

Current LEgislation - 2010

Error Map

Yearly avg. PM10 over domain cells

SURROGATE MODEL VALIDATION

Es. Ozone

RIAT - Optimization

Decision variables = Application rates (how much each end-of-pipe technology is adopted in each specific activity) which determine emissions

Objectives = Minimize total annual costs (industrial + external: morbidity and mortality)

Minimize an air pollution index

Constraints = reduction feasibility

Method = constraint method (fix a value for air quality and minimize costs, then iterate)

The result is a Pareto boundary

RIAT - Implementation

Analysis of a current AQ planThe air quality plan of Lombardy Region (PRIA, formally approved in September 2013, www.reti.regione.lombardia.it/) will be used as an example.

The plan consists in 66 measures to be first undertaken, subdivided in:

• 26 concerning road transport and mobility;

• 27 concerning point emission and efficient energy uses;

• 13 concerning agriculture and forestry.

Another possible classification is:• Scenario measures (they can just be implement or not,

without graduality; ex. building a new metro line)• Efficiency measure (also called “non-technical” measures,

they modify some emitting activity)• End-of-pipe (also called “technical” measures, they decrease

the emission of pollutant , (almost) without changing the correspondent activity).

Emission calculation

To analyze the plan, let’s consider the expression already presented for the emission Ep(z1,z2,z3,u), of a given pollutant p, i.e.

where - Ai is some measure of the activity going on in position z1,z2,z3

at time t, - efip is the “unabated” emission factor of pollutant p by activity

i- rip(u) is the fraction of emission reduced by actuating decision

u.

Efficiency measures mean a modification of Ai

End-of-pipe measures mean a modification of rip Scenario measures simply require two different evaluations of

the rest of the plan (yes, no).

u1 )u,,,,()u,,,( pp321321p iii

i reftzzzAzzzE

The presence, in the emission definition, of terms depending on efficiency measures ueff and end-of-pipe measures uend shows that the same emission (and thus the same air quality) of pollutant p can be obtained with different combinations of actions.

How to evaluate the plan? Cost-Benefit analysisCosts refer to the actual deployment of all the measures (decisions).Benefits are of two types:- Direct, derived from the reduced used of fuels (ueff)- Indirect, derived from the reduction (uend ) of external costs

due to an improved air quality.

The following analysis concentrates on only one pollutant (PM10) and only external costs due to human health.

The evaluation scheme

i

endiiiieffi urefAefuAE pppp

The evaluation scheme - 2

Scenario definition

Energy efficiency measures

End-of-pipe measures

Computation of air quality index(es)(surrogate model)

Computation of health impact and

cost reduction Computation of measure

implementation costs

Computation of direct benefits

(energy savings)

CTM

Emission calculation

dir

ect

ca

lcula

tion

Air quality model

Health impact model

Sample actions

Scenario measures (yearly values)

Ex. New metro line 5 (effects on in Milan)Þ Annual cost (discounted over 25 years at 5%) 3,96

M€Þ 25 Mkm reduction (transportation model) Þ 87,500 GJ energy reduction Þ 7,13 M€ cost reductionÞ Emission reductions (t): 1 PM10, 9 NOx, 2 VOCÞ PM10 concentration reduction: negligibleÞ Reduction of health impact: negligible

N.B. in Lombardy the total car mileage is 20 Gkm/year

Sample actions - 2

Efficiency measures (yearly values)

Ex. Detailed heat accounting for all buildings (effects on all the region in urban areas)

Þ Annual cost 0.84 €/GJÞ Planned (2020) 46M GJ reductionÞ Total cost (2020) 39 M€Þ Direct benefit: 2380 M€ Þ Emission reductions (t): 39 PM10, 2700 NOx, 255 VOCÞ Reduction of health impact: 6.7 M€ (morbidity), 12.2 M€

(mortality)

Note that the full application of the measure (46M GJ in 2020) may not be the most efficient choice.

Sample actions - 3

End-of-pipe measures (yearly values)

Ex. Substitution of commercial vehicles < EURO 3 with EURO 6

Þ Unit cost: depends on the vehicle type (17 types considered)

Þ Planned (2020) 246,000 substitutions 4 GkmÞ Total cost (2020) 5,65 M€Þ Direct benefit: 0 M€ Þ Emission reductions (t): 234 PM10, 6385 NOx, 433 VOCÞ Reduction of health impact: 13.7 M€ (morbidity), 25 M€

(mortality)

Note the estimated cost is based on the pure evaluation of the technology difference, it may not represent a correct estimation for the process of scrapping old vehicles and buying new ones.A full implementation of the plan may not be the most efficient choice.

C-B analysis

  2020 (M€/anno)Measure costs

Direct benefits

Health benefits

Private transportation 208 966 57

Public and alternative

transportation846 1454 <0,01

Electric energy 2590 2058 <0,01

Thermal energy 3781 8688 37

TOTAL 7425 13166 80

• Direct benefits seem to exceed costs in a remarkable way,

• Most benefits are in the thermal energy sector

• Health benefits are much less significant (limited reduction of the average PM10 concentration)

• Actions in electric energy do not impact health (high and remote emissions)The implementation of the 2020 plan will imply a substantial

(20%) reduction of energy consumption (see: 2nd MOD).

Trade-off between pollu-tants

If one considers more than one pollutant, the problem becomes more complex:- Investing on the reduction of one pollutant prevents investing on

the other- Certain emission reduction measures act on more than one

pollutant at the same time and in a fixed proportion (they do not allow to act on both pollutant separately) With a moderate

investment, one may find a set of solutions dominating the CLE. Increasing the investment, the advantage becomes less evident and the choice between improving ozone or PM2.5 is more difficult.

Trade-off between pollu-tants - 2

One way to solve the trade-off is to look at the external costs (in comparison to CLE) along one cost curve.

They strongly depend on the geographical distribution of actions, effects, and population

Open problems

Cost and benefit of some measures pertain to many different sectors (typically, transportation, agricultural production, …) how can we attribute the correct share to air quality plans?

Uncertainty about activities, emissions, models, and impact is very large. Actions (response) must be evaluated in terms of risk of taking the wrong decision rather than in terms of finding the optimal ones (see 2nd MOD).

The cost of adopting a new technology, dismissing an old one, cannot be measured by the pure difference of the two costs (it’s a problem of economic estimation).

Efficiency measures often imply some change of habits (learning) and thus their adoption may be very long.

Efficiency measures often imply a different service (ex. bicycle vs car). How can we compare them, except for the difference in emissions?

Thus…

Drivers, technologies, external conditions are permanently evolving, thus a traditional plan, foreseen the actuation of a set of measures within a given horizon (5-10 years), will never materialize as planned. Some (technology driven) change are very fast.

Ex. the rapid diffusion of LED lights (the power of lights in a flat may be reduced up to 10 times);

car and bike sharing are substantially decreasing the number of cars.

It is necessary to move from traditional planning to a continuous control policy, that fosters the evolution in the planned direction, but taking into account the current situation.

Such a control policy can be based on what is known as “receding horizon” in control theory, and is being developed by sociologists and planners as “transition management”.

Thanks for your attention