Forecasting High Air Pollution
Episodes by Classification Trees
Francesca Bruno, Daniela Cocchi,
and Carlo Trivisano
The paper proposes classification
trees (CART) as a suitable technique for forecasting daily exceeding standard
concentration of pollutants established by Italian law. The motivations
of such a proposal can be found in the following two points:
a) from a statistical point of view,
a classification rule based on classification trees allows to treat the
so called problem of curse of dimensionality, when the number of possible
predictors is very high, and to find a solution to the problem of non homogeneity,
since the conditions under which high air pollution episodes occur are
often different;
b) from a practical point of view,
classification trees suggest decision rules which can be easily applied
by public authorities.
The pollution data are characterised
by an impressive discrepancy in the dimension of the two classes of events
of interest. The first is characterised by days with an observed value
exceeding the standard of interest, the second is formed by days without
exceedance events. Due to this peculiarity, model selection by cross validation
pruning usually brings to a tree which classifies any value in the non-exceeding
class. We take account of this drawback by an iterative technique which
attributes, at each iteration, a weight increasing with the order of the
iteration itself.
A model has been built for predicting,
2 days ahead, the most probable class for daily urban ozone concentrations
in the city of Bologna. The standard considered is the so called "attention
level" (180 mg/m3). Meteorological forecasted variables have been
considered as predictors. Taking into account the poor quality of the predictors
and the non negligible measurement error of the monitoring network, the
results are surprisingly good. This result can be further appreciated whenever
one realises that, in managing high pollution episodes, a main goal is
the indication that a standard threshold can be exceeded.
FRANCESCA BRUNO
Dipartimento di Scienze Statistiche
Università di Bologna
Via Belle Arti 41
Bologna 40126, Italy
bruno@stat.unibo.it
Reducing the Environmental Impact
of Industrial Areas Through Energy-Sharing Policies
V.G. Dovi, C.Solisio, and A.
Del Borghi
The industrial site examined includes
plants and processes of several different companies. Due to the fact that
the level of pollution cannot be reliably apportioned to each single plant,
the enforcing authority has accepted a strategy agreed upon on a voluntary
basis by all the companies involved. This strategy foresees the reduction
of the production level for all the processes, whenever the environmental
conditions (typically the atmospheric parameters) are likely to give rise
to unacceptable values of air and/or water quality.
Thus, we are presently considering
the possibility of modifying the traditional process integration techniques
by extending the integration procedure to all the companies present in
the site. Due to the presence of different goals and of (at least potentially)
conflicting interests, we have to consider a different objective function
and a different optimization criterion. The modifications to the objective
function arise from the necessity of considering the amount of
working days in which the level of production is reduced (due to environmental
constraints). The modifications to the optimization procedure are connected
with the necessity of adopting a multiobjective (Pareto) optimality criterion.
The basic idea is to investigate
the possibility of a company accepting a sub-optimal level of integration
for its own process (which might be necessary for the attainment of an
overall optimal integration and consequently of a lower global amount of
pollutants emissions), if this is offset by a lower number of days with
a reduced production level.
The consensus to a similar strategy
can be expected if there exists a Pareto optimum that is more convenient,
for all the companies, than the present situation.
VINCENZO G. DOVI
DICheP
Universita di Genova
Via Opera Pia 15
Genoa, 16145, Italy
Tel.&Fax: +39-010-3532921
V.G.Dovi@iol.it
Applicability of Exponential Smoothing
Models for the Prediction of Road Traffic Noise
Krishan Kumar and V.K. Jain
Measurement of A-weighted instantaneous sound pressure levels were made at site in the vicinity of a busy road carrying vehicular traffic typical of the city of Delhi. The resultant time series has been analyzed and modelled using Winter's exponential smoothing method. The model fit is found to be good. The suitability of the model for the purpose of forecasting on short time scales is established.
KRISHAN KUMAR
Deaprtment of Environmental Science
and Engineering
Guru Jambheshwar University
Hisar-125 001, Haryana, India
kkksd@hotmail.com