A Geostatistically Based Probabilistic Risk Assessment Approach - Evolving Application Domains of Data Warehousing and Mining - page 290

Figure 3. a) Map of the krigged estimates of arsenic concentration in groundwater b) Homogeneous
areas according to land use information (modif. from Castrignanò et al, 2004)

the study area in the past, distant areas that were
similar for activity have been pooled, arriving at
the definition of 5 homogeneous areas (Figure
3b) defined in such a way:

in Figure 5b and shows that the probability of
underestimation is mostly smaller than 30%, with
the exclusion of a small area westbound where
the probability is 60%. The impact of the risk α
can be easily evaluated because it is linked to
the cost of an unnecessary remediation, but it
does not directly involve other considerations of
non-monetary type, such as human health, that
are more difficult to be estimated and are instead
involved in the risk β.

• Area 1: deposit;

• Area 2: oil storage tank;

• Area 3: free area;

• Area 4: external area;

• Area 5: water treatment and implantations.

Applying the indicator variable (co)kriging,
two types of maps have been obtained:
Probability map of arsenic exceeding the

concluSIon

•

compulsory threshold (Figure 4a);
Marginal probability map of arsenic ex- • In this paragraph an approach has been described

where the available information, both hard and soft
under the form of prior probability distribution,
is coded as a set of indicator variables. The non-
parametric method yielded in two specific areas
(area 4 and area 5) a high probability of exceed-
ing the critical threshold for arsenic pollution in
groundwater. The presence of this pollutant is not
linked with the past activities of the manufacturing
factory, because from information given by the
control authorities of the district, high concentra-

ceeding the compulsory threshold relative
to each area (Figure 4b).

The risk of As pollution, expressed as prob-
ability of exceeding, is greater than 80% in two
specific areas, the area 4 and the area 5 along the
coast line. The risk α of a false positive is con-
toured in Figure 5a and shows that the probability
of overestimation is mostly smaller than 40%.
The risk β, or of a false negative, is contoured