Quality It
Resume:
Environmental Monitoring and Assessment (****) ***: 95 117
c Springer 2006
DOI: 10.1007/s10661-006-6547-3
A METHODOLOGY FOR SPACE-TIME CLASSIFICATION
OF GROUNDWATER QUALITY
G. PASSARELLA and M. C. CAPUTO
Water Research Institute, IRSA CNR, via F. De Blasio, 5, Bari, Italy
( author for correspondence, e-mail: ********.**********@**.****.***.**)
(Received 9 November 2004; accepted 26 April 2005)
Abstract. Safeguarding groundwater from civil, agricultural and industrial contamination is matter
of great interest in water resource management. During recent years, much legislation has been
produced stating the importance of groundwater as a source for drinking water supplies, underlining
its vulnerability and de ning the required quality standards. Thus, schematic tools, able to characterise
the quality and quantity of groundwater systems, are of very great interest in any territorial planning
and/or water resource management activity.
This paper proposes a groundwater quality classi cation method which has been applied to a real
aquifer, starting from several studies published by the Italian National Hydrogeologic Catastrophe
Defence Group (GNDCI).
The methodology is based on the concentration values of several parameters used as indexes of the
natural hydro-chemical water condition and of potential man-induced modi cations of groundwater
quality. The resulting maps, although representative of the quality, do not include any information
on its evolution in time. In this paper, this stationary classi cation method has been improved by
crossing the quality classes with three indexes of temporal behaviour during recent years. It was then
applied to data from monitoring campaigns, performed in spring and autumn, from 1990 to 1996,
in the plain of Modena aquifer (central Italy). The results are reported in the form of space-time
classi cation table and maps.
Keywords: groundwater quality, water quality classi cation, space-time classi cation, water resource
management
1. Introduction
During the last two decades European Directives (2000/60/EC, 91/676/EEC,
91/271/EEC, 80/778/EEC, 80/68/EEC) and some Italian laws (DL152/99, L36/94,
DL132/92, DPR 236/88) regarding groundwater quality, have stated that ground-
water is a public resource and that the priority should be given to its use as drinking
water. In fact, groundwater systems often provide the most important source of
drinking water, above all in arid or semiarid countries. When correctly applied, this
legislation effectively safeguards such resources from contamination processes re-
lated to civil, agricultural and zootechnic activities. In particular, the 80/778/EEC,
and the following amendments, and the Italian law (DPR236/88) xed the maxi-
mum allowable concentration and guideline values for a large number of parameters
characterising water quality with respect to its suitability for drinking uses. In this
96 G. PASSARELLA AND M. C. CAPUTO
legislative frame, all authorities in charge of water resource programming, need
reliable, but at the same time, schematic tools able to determine water quality and
quantity for this speci c use.
In this study a methodology has been applied to classify and represent baseline
groundwater quality. Currently, many research projects related to the European
Water Framework Directive are in course, aimed at de ning technical tools for its
application. Many of the 25 countries of the EU are planning national legislation
in order to adopt the directive. In any case, the proposed methodology, being a
quick tool for assessing the qualitative status of groundwater, goes beyond a strict
implementation of the European WFD and it is applicable wherever such a practical
and simple method is required.
The proposed methodology processes the concentration values of several of the
most signi cant parameters typical of natural quality and of contamination induced
by human activities. Its main aim is to provide a simple tool capable of determining
groundwater quality quickly. Obviously, a complete and satisfactory evaluation of
water quality can be reached only by means of detailed, wide-spectrum, but at the
same time, expensive and time consuming, analytical measures. Furthermore, the
classi cation needs to be based on few representative analytical parameters, able to
give a sketched picture of the general qualitative status of groundwater. It focuses
mainly on the large scale, natural hydro-geo-chemical conditions, not taking into
consideration contamination due to substances produced by human activities on a
local scale.
The methodology combines legislative acts and technical knowledge in order not
only to represent the actual qualitative status of groundwater, but also to propose
reliable solutions for improving the water quality up to the standards required
for drinking water, which is the quality limit considered in the paper. The worst
qualitative condition considered is Unsuitable for drinking purposes, with some
limitations for other uses . All the environmental laws in the EU exclude any use of
water when it is contaminated by organic compounds, pesticides or solvents. In this
case, intensive, speci c and, probably, expensive and time consuming treatments
are required in order to make the water usable for any use except drinking, which is
excluded a priori. In this framework, it is not useful to consider groundwater systems
contaminated by such substances, at least for the general purposes of the study.
In its original form (Civita et al., 1993; Civita and Oliviero, 1993; Francani,
1993; Zavatti, 1993; Giuliano et al., 1995), the method allowed groundwater re-
sources to be classi ed according to a scale of nine quality levels. However, the
resulting cartographic representations were often not simple to decipher because
of the fragmentation of areas with different characteristics. In a previous paper
(Caputo et al., 1999) the original number of the classes was reduced to ve, based
on a criterion that assigned a greater weighting to several undesirable parameters
since their removal from water requires expensive intervention. Both the original
and the shortened classi cation were widely applied in Italy but only in station-
ary aquifer conditions. These applications always considered data from a single
SPACE-TIME CLASSIFICATION OF GROUNDWATER QUALITY 97
sampling campaign. In this study a substantial improvement of the methodology
has been tested, which consists in crossing the classes of stationary quality with
three indexes of temporal behaviour during the last years.
Data sets related to fourteen consecutive sampling seasons have been used with
the purpose of improving the methodology in order to make it capable of represent-
ing, at the same time, both the spatial characteristics and the temporal evolution of
groundwater quality.
The raw data were organised and ltered with the purpose of eliminating coarse
inaccuracies. In the following phase the data were statistically elaborated to deter-
mine their global and average characteristics (Vajani, 1987; Ott, 1995). Through
structural analysis or variography (Krajewski and Gibbs, 1993), the parameters
characteristic of the spatial behaviour of each substance, were determined in order
to estimate their values in the non-sampled nodes of a regular grid, using the or-
dinary kriging method (Delhomme, 1978; Jurnel and Huijbregts, 1978; Isaaks and
Srivastava, 1989). Using these estimations, it was possible to draw quality maps for
each of the selected parameters; overlaying these maps, using GIS tools (Keckler,
1994; ESRI, 1996), seasonal quality maps were obtained applying the ve class
classi cation. Finally, by means of the space-time classi cation table, the informa-
tion contained in all the maps was merged, intersecting two indexes, the rst of
which represents the most recent qualitative status of the considered aquifer and
the second, its temporal evolution during recent years.
2. Study Area
This study considered an area located in the Modena plain (central Italy) which
covers an area of about 580 km2 (Figure 1). It is bound on the southern side by
the Tosco-Emiliano Apennine, and the Secchia and Panaro Rivers cross it. Other
small watercourses include the torrents Tiepido and Cerca. The Modena plain has
been extensively studied from a hydrogeological standpoint in order to tackle the
problem of drinking water supply. Important hydrogeological studies have been
produced from the 1930s to date (Visentini, 1935; Colombetti et al., 1980; Barelli
et al., 1990; Gelmini et al., 1990a; Muratori, 1990; Paltrinieri and Pellegrini, 1990;
Vicari and Zavatti, 1990).
From a geologic standpoint, the high and middle areas of the Modena plain are
basically characterised by a system of primary piedmont fans belonging to the main
Rivers (Secchia and Panaro), intersected by secondary ones connected to the small
torrents.
A simplifying hypothesis can be made in order to schematise the groundwater
system. Even though the actual aquifer consists of several overlapping layers, the
horizontal and vertical continuity of certain hydrodynamic parameters, such as
piezometric head and electric conductivity (Idroser, 1978), in fact, allow, the aquifer
to be considered as a one-layer system.
98 G. PASSARELLA AND M. C. CAPUTO
Figure 1. Study area.
In the upper part of the plain, the aquifer is uncon ned and it is recharged by
rainfall and in ltration from the river, while towards the distal part of the alluvial
fans, the groundwater system is con ned.
3. Preliminary Data Analysis
The groundwater system considered in this study has been extensively monitored
by means of samplings in about 180 wells belonging to two different monitoring
networks, one of about thirty wells and the other of 150. This study was performed
using only the data collected in the second network, even though the number of the
wells actually considered was never higher than 90 in any given season.
The proposed methodology allowed us to generate groundwater quality maps us-
ing the concentration values of eight easily available parameters, usually considered
representative of groundwater qualitative status. The eight parameters were split
into two groups. The chemical and physical parameters group includes hardness,
electric conductivity, sulphates and chlorides, which often indicate the natural status
of the groundwater. Nitrates, iron, manganese and ammonia, which may suggest
human-induced pollution, belong instead to the undesirable substances group.
High values of the parameters of the second group often require restrictive mea-
sures for drinking water use. A ninth parameter, the piezometric head was also
considered, since it is an important indicator of the hydrodynamic behaviour of
groundwater.
SPACE-TIME CLASSIFICATION OF GROUNDWATER QUALITY 99
Before applying the proposed classi cation, the collected data was passed
through several important steps. During the rst step they were ltered, organised
and statistically analysed in order to remove macroscopic anomalies and describe
the mean behaviour of each parameter. During this phase, fourteen sets of nine
parameters were processed, one for each of two seasons considered from 1990 to
1996.
The data sets were all incomplete since a number of sampling points resulted
missing in each season, but the number of the values was, generally, suf ciently
meaningful, with the exception of a couple of seasons before 1992, and a few pa-
rameters. This phase of the analysis outlined the presence of anomalous values
(outliers), but they were adjusted when any explanation was recognised for these
anomalies (measurement error, error of transcription, measures out of scale, etc.).
Nevertheless, sometimes, they could not be explained as trivial errors and, conse-
quently, they were kept in the data sets, postponing an explanation until the spatial
analysis phase (probable small scale perturbations).
The statistical distributions were rarely comparable to the normal distribution
and often showed remarkable asymmetry. In some cases, the statistical distributions
showed the possible presence of two different distributions (bimodal distributions).
This behaviour could be due either to the presence of two different phenomena,
producing the same pollution, or to different sampling depths.
Unfortunately, although the computed statistics produced information on the
mean behaviour of the variables, they did not give any idea of their spatial behaviour.
Most of the data sets related to earth sciences, contain a spatial continuity for each
variable. In general, the difference between the values sampled in two different,
but close locations, would be expected to be very small and should increase with
the distance between the locations. There are statistic tools that can be used to
describe the relationships between the values of a variable sampled in two different
points (Journel and Huijbregts, 1978). Geostatistics uses a probabilistic approach
to provide a suitable and reliable tool capable of producing spatial estimations of
a variable, sampled in a limited number of points. The variogram is one of the
fundamental geostatistical tools for achieving this objective. In fact, to estimate
a value in a point not sampled, a model able to describe the spatial behaviour of
the considered phenomenon is required. The goal of variography is, therefore, to
create a variogram model that gives a representation of the real spatial behaviour
of the considered variable. To achieve this objective, accurate physical knowledge
of the phenomenon is necessary. There are various models for variograms; thus
theoretical regularity of methodology can be assured by choosing among a set of
available models.
The second preliminary step consisted in implementing variographic techniques
(Isaaks and Srivastava, 1989; Krajewski and Gibbs, 1993). Variography is a tool
provided by geostatistics which may be de ned as the statistical study of spatial
correlation. The variography aspect of geostatistics assesses in aggregate how the
contaminant concentrations at any two given locations varies as a function of the
100 G. PASSARELLA AND M. C. CAPUTO
distance separating those two locations. Numerically, a variogram function is cal-
culated in units of distance and, consequently, a variogram value is plotted against
distance. An experimental variogram is tted through plotted points to assess spatial
correlation. Experimental variogram functions are characterised by three parame-
ters: nugget, sill, and range. The range is the distance from a measurement point
(known sample) to the point where the variance stops increasing with distance
from the sample point. The value at which the variogram model attains the range
is known as the sill, meaning the change in variance no longer increases with in-
creasing distance from the sample point. The nugget or nugget effect, is created
by measurement errors or spatial sources of variation at distances smaller than
the sampling interval and is the value of semivariance when the distance from the
sample point equals zero (Main et al., 2004). In this work, the spatial behaviour of
every variable was determined by calculating experimental variograms and inter-
polating them with some standard analytical models. This allowed us to evaluate
the characteristic variogram parameters (i.e. nugget, range and sill). In particu-
lar the gaussian, exponential, spherical and linear models were used in this study.
Several variogram parameters proved to be extremely stationary over a period of
time, since, on average, they did not change during the observed period. On the
contrary, other parameters varied greatly over time and, sometimes, the variogram
model changed during the observed period. The data available were processed us-
ing the ordinary kriging method (OK) (Keckler, 1994) to estimate the unknown
values of each parameter in the 14700 nodes (105 rows and 140 columns) of a
regular grid, having a regular mesh of 200 200 meters, overlapping the study
area.
4. Groundwater Quality Classi cation
The objective of the proposed methodology is to describe the qualitative status of
a groundwater system, starting from a number of values sampled in the wells of a
monitoring network during a number of consecutive seasons.
As generally happens when a classi cation is performed, synthesis is inversely
proportional to detail, i.e. the more synthetic the classi cation system is, the more
detail is lost. However, this loss is counterbalanced by a simpli cation of the clas-
si cation, which can make it more understandable. The validity of a classi cation
depends, therefore, on the equilibrium that is reached between synthesis and detail.
Before starting to describe the different phases of the classi cation, it is useful
to summarise shortly the spatial and temporal behaviour of the piezometric head
that is a naturally binding parameter for transport and dispersion of pollutants in
groundwater. Contour maps of piezometric levels were plotted. These represent the
features of the phreatic surface, in the southern part of the study area, where the
aquifer is uncon ned, and that of the piezometric head in the middle plain, where
the aquifer is con ned.
SPACE-TIME CLASSIFICATION OF GROUNDWATER QUALITY 101
On the whole, these maps shows a modest temporal variability of the piezometric
head on a regional scale. Changes of the piezometric surface on a local scale are
clearly visible. This was probably due to anthropogenic phenomena bounded both
in space and in time.
A remarkable difference of the mean gradient can be noted between the uncon-
ned aquifer, which was steeper, and the con ned one, which was at. However,
the maps allow the effect of the recharge from the rivers Secchia and Panaro, as
well as that from the torrent Tiepido, to be underlined.
As mentioned above, the parameters considered in this paper were those pro-
posed by the CNR-GNDCI in a study of 1993. This method was based on two xed
threshold values of the eight chosen parameters: the guideline value (GV) and the
maximum allowable concentration (MAC). The parameters were also clustered in
two groups whose quality class was automatically de ned by the worst parameter
within it (Table I).
The threshold values allowed groundwater quality to be classi ed, in a generic
location of the study area, with regard to each of the eight parameters in three
different classes:
Class A (optimal): this is the case when the parameter value is lower than the GV.
Groundwater is suitable for drinking water purposes with no treatment, and is
thus acceptable for the majority of industrial and agricultural uses;
Class B (acceptable): this is the case when the parameter value is lower than the
MAC but larger than the GV. Groundwater is suitable for drinking water purposes
with no treatment, with some limitations for industrial and agricultural uses.
Class C (poor): this is the case when the parameter value is larger than the MAC.
Groundwater is unsuitable for drinking water purposes, with some limitations for
other uses. In particular, if the parameter belongs to Group 1, speci c treatment is
required, while, if it belongs to Group 2, a simple or advanced oxidation treatment
is needed. Even if only one of the four parameters of a group exceeds the MAC the
quality related to that group is considered poor, while, if none of the parameters
exceeds the GV the groundwater quality related to that group is set to optimal.
The three levels of quality (A-Optimal, B-Acceptable, C-Poor) reported in
Table Ia and explained in Table Ib, need to be assessed for each of the groups of
Table Ia. Consequently, we can de ne a level of quality with respect to each of
the groups. For instance, a water sample can be classi ed A1-C2 if it is Optimal
(A) for Group 1 and Poor (C) for Group 2. Optimal for Group 1 means that none
of the measured values of the parameters of the rst group is beyond the limits for
Class A.
Finally, combining the qualitative water status with regard to each group we
obtained nine different combinations which are those reported in Table II.
In some cases, a nine class classi cation is scarcely meaningful because of
heavy fragmentation of the considered area as a result of the large number of
102 G. PASSARELLA AND M. C. CAPUTO
TABLE I(a)
Classi cation criteria for groundwater quality
Group 1 Group 2
(chemical and physical parameters) (undesirable substances)
TH El.Cond. SO4 Cl NO3 Fe Mn NH4
( F) ( g/l) ( g/l) ( g/l)
(mS/cm) (mg/l) (mg/l) (mg/l)
Parameters
Class A
15a 30 50 >2 >250 >200 >50 >200 >50 >500
Poor
a
Minimum recommended value.
Intermediate value between the Maximum Allowable Concentration (MAC) and the Guideline
Value (GV) (DPR 236/88).
GV doubled.
TABLE I(b)
Class judgment related to water use
Class A: Suitable for drinking water purposes with no treatment, acceptable for the
majority of industrial and agricultural uses.
Class B: Suitable for drinking water purposes with no treatment, with some
limitations for industrial and agricultural uses.
Class C: Unsuitable for drinking water purposes, with some limitations for other uses.
C Group 1: requiring some speci c treatment.
C Group 2: requiring simple or advanced oxidation treatment.
TABLE II
Former methodology
Group 1 Group 2 Quality
A1 A2 Best
B1 A2
C1 A2
A1 B2
B1 B2
C1 B2
A1 C2
B1 C2
C1 C2 Worst
SPACE-TIME CLASSIFICATION OF GROUNDWATER QUALITY 103
TABLE III
Scheme of simpli ed general classi cation
Original Simpli ed
General classi cation
A1 A2 Optimal
B1 A2
C1 A2 Good
A1 B2 Acceptable
B1 B2
C1 B2
A1 C2 Poor
B1 C2 Very poor
C1 C2
Civita et al., 1993.
classes. Consequently, in a previous study (Caputo et al., 1999), several classes
were joined (Table III) giving a signi cant weighting to the undesirable substances
(Group 2), since, in general, more expensive intervention is required to reduce
the concentration of these substances in water. This simpli ed classi cation was
based on ve quality classes.
In the nal stage of the methodology, starting from the simpli ed classi cation
system described above, the temporal information was considered combining the
quality maps of fourteen succeeding sampling times, according to de ned rules.
The sampling stages started in 1990 and continued, in spring and autumn each year,
until 1996.
The methodology proposed here was developed through a succession of phases,
which represented intermediate steps of classi cation, each of which provided
precise information on the water quality. This information was synthesised more
and more, at every level of classi cation, until a nal, representative, map of space
time groundwater quality was achieved.
The four main levels of classi cation are brie y described below:
1. Classi cation by parameter: the rst level of classi cation provides a map of
groundwater quality for each of the eight parameters de ned in Table I, based
on the three classes reported in the same table.
2. Classi cation by groups: the second level of classi cation provides one map
for each group of Table I. These maps were achieved assigning to the cells of
each map the worst classi cation grade of the four parameters of the group.
These two maps are also characterised by the three quality levels reported in
Table I.
3. Simpli ed general classi cation: the third level of classi cation provides one
map of water quality referred to the ve classes of Table III. This map was
104 G. PASSARELLA AND M. C. CAPUTO
achieved coupling, for each cell, the classes of the two groups as reported in
Table III ( rst column) and grouping them as reported in Table III (second
column).
The rst three phases were repeated for each sampling season so that a series
of spatial water quality maps (simpli ed classi cation) were produced.
4. Space-time classi cation: nally, the ve classes de ned in the simpli ed gen-
eral classi cation, were processed again, with the purpose of determining two
new indexes of classi cation able to describe, respectively, the middle level of
quality at a considered time period and the correspondent temporal trend.
As the reader may have noted above, the terms classi cation table and quality
map have been used indiscriminately. The reason is that maps are the natural and
most useful graphic representation of such classi cation tables.
4.1. LEVEL 1 CLASSIFICATION BY PARAMETERS
The rst level of classi cation provides indications on the groundwater quality with
respect to each of the eight parameters de ned in Table I. In fact, once the values of
the eight considered parameters were estimated all over the study area, it was split
into different quality zones, optimal, acceptable and poor, as reported in Table I.
This level of classi cation produced four maps of classi cation related to the
chemical and physical parameters (Group 1) and four related to the undesirable
substances (Group 2). It is necessary to note that, because of the lack of manganese
samplings from autumn 1990 to autumn 1991, it was not possible to classify ground-
water with respect to this parameter during the years mentioned.
4.2. L EVEL 2 CLASSIFICATION BY GROUPS
The second level of classi cation was developed de ning rules that allowed the four
parameters of each group to be overlaid to produce just one map of the chemical
and physical parameters and one of the undesirable substances. These two maps
were characterised again by three quality levels (optimal, acceptable and poor).
The rules have been de ned to guarantee the greatest safety. In other terms, even
if only one of the four parameters of a group exceeds the MAC, the quality related
to the whole group will be considered poor. Anyway, if none of the parameters
exceeds the GV, the groundwater quality related to that group will be optimal.
Applying these rules, fourteen maps of groundwater quality by groups were
plotted, one for each season. Actually, there were only eleven maps in the second
group, due to the lack of manganese values for three seasons.
In the following paragraphs some spatial and temporal considerations are re-
ported which appear from the joint observation of these maps.
SPACE-TIME CLASSIFICATION OF GROUNDWATER QUALITY 105
4.2.1. Chemical and Physical Parameter Maps
The chemical-physical parameter maps show, in general, a slight predominance of
groundwater areas qualitatively classi ed acceptable, or rather, in the worse case,
characterised by values of the four considered parameters between the correspond-
ing GV and MAC. One zone, qualitatively better, but limited in extension, was
constantly present in the NE part of the study area. A large portion of the ground-
water system characterised by poorly valued water was present around the town
of Sassuolo and it often extended as far as including the city of Modena. Other
poor-quality zones, even if characterised by a lesser extension, were also noted
in the eastern part of the towns of Castelnuovo Rangone and San Cesario sul Pa-
naro. In the last years sampled the general qualitative situation seems to have been
worsening. In particular it seems that the poor-quality classi ed areas extended to
become larger and to merge. The main cause of the presence of these areas and
their worsening trend, can be explained mainly by the presence of elevated hardness
values.
4.2.2. Maps of the Undesirable Substances
The undesirable substance maps clearly show very bad qualitative status for most
of the studied groundwater system, which was constantly classi ed poor.
The only portions of it characterised by a slightly better qualitative level, enough
to allow it to be classi ed as acceptable, were those in correspondence with the
recharge areas from the two main rivers. For all these parameters a positive in-
uence of fresh water from the two rivers was found. Because of the lack of the
manganese samplings from autumn 1990 to autumn 1991, it was not possible to
classify the groundwater quality for these three seasons. Obviously, this lack re-
ected negatively on the subsequent levels of classi cation.
4.3. L EVEL 3 SIMPLIFIED GENERAL CLASSIFICATION
The further overlaying of the chemical-physical parameters with those of the un-
desirable substances produced a classi cation table characterised by nine quality
classes (Table II), which was not easy to interpret in map form because of widespread
fragmentation. Consequently, some of the nine classes were merged together re-
ducing the number to ve, thus simplifying the classi cation table and making
the related maps more comprehensible. This classi cation level is indicated as the
simpli ed general classi cation (Table III).
In merging the nine original classes, a greater weighting was given to the param-
eters belonging to the second group (undesirable substances), since, in general, the
presence of such substances in water requires more expensive intervention, both
from a structural and a managerial point of view, in order to make it drinkable.
The results of this classi cation phase were then used to plot eleven maps,
which illustrate the groundwater quality according to the classi cation table de-
scribed above. Unfortunately, once more, the lack of three classi cation maps for
106 G. PASSARELLA AND M. C. CAPUTO
undesirable substances, reduced the number of classi able seasons from fourteen
to eleven.
Each of the resulting maps is, therefore, representative of a single sampling
season and does not contain any information relating to the temporal evolution of
the water quality. Considered as a sequence, however, these instantaneous maps
represent an interesting movie, which describes water quality evolution exactly.
4.3.1. Simpli ed General Classi cation Maps
The criterion used for de ning this level of classi cation was that of giving greater
weighting in each cell to the worse quality class of the two determined at the pre-
vious level of classi cation. Remembering that classi cation based on undesirable
substances had outlined prevalently very poor groundwater quality, it can be in-
ferred that the in uence of this weighting on the simpli ed general classi cation
map was strongly conditioning. In fact, the nal result consisted in cartographic
representations where the best quality classes are almost always absent.
Examining the cartographic representations for the different seasons, it can be
seen that there was little water quality variability over time in most of the aquifer.
In particular, a large part of the study area is classi ed as having very poor quality
water.
A portion of the aquifer, near the town of Nonantola in the NW side of the study
area, characterised by poor classi ed water and the cleaning effect of freshwater
from the main rivers, con rms the in uence of the undesirable substances on the
general classi cation.
4.4. L EVEL 4 SPACE - TIME CLASSIFICATION
The last level of classi cation aims to insert the temporal information into the
groundwater quality representation. In particular, the ve classes de ned in the
simpli ed general classi cation, were processed again, to determine two new clas-
si cation indexes able to describe, respectively, the middle level of quality at a
considered time period and the correspondent temporal trend.
Different proposals of classi cation were tested, all based on the intersection of
the described indexes but characterised by different classi cation interval widths.
Before entering into the details of the different proposals, the following para-
graphs provide a general scheme of the procedure followed to achieve this last
classi cation step.
(a) Numerical values, ranging from 1 (optimal) to 5 (very poor), were assigned to
each of the ve quality classes de ned at the previous level of classi cation
(Table III);
(b) a position index was de ned classifying the arithmetic mean of the seasonal
series of the numerical values de ned above (mean cell value = MCV) on a
ve point scale;
SPACE-TIME CLASSIFICATION OF GROUNDWATER QUALITY 107
(c) a trend index was de ned computing the slope of the linear regression line
that ts all the seasonal numerical values de ned at point a) and classifying this
value on a three point scale: positive slope (i.e.: increasing trend qualitative
worsening); negative slope (i.e.: decreasing trend qualitative improvement);
null slope (i.e.: no trend stationary quality). The slope was expressed in
degrees and ranged between 90 and +90 . Linear regression was chosen,
rather than other methods, to evaluate the index of water quality trend, due
to its conceptual simplicity and the immediateness. Other more accurate and
reliable methods could be used, but, considering that the calculation had to
be performed about ten thousand times (one for each cell of the estimation
grid) and that the methodology had to be as simple as possible, the choice was
considered acceptable.
The fteen classes, resulting from the intersection of the ve levels of the position
index with the three of the trend index, were grouped together according to the
scheme reported in Table IV. As Tables IVa shows, and Table IVb describes, the
nal space-time classi cation refers to ve classes, which derive from crossed
evaluation of the average groundwater qualitative status and its observed evolution
during several past years.
Actually, the ve classes, from BB to PP, do not represent a unique level of
quality but rather labels for groups of different qualitative levels arbitrarily merged
by the authors on the basis of observed behaviour over time (qualitative trend)
to form one single class. For example, as shown in Table IV, class B represents
areas where the mean quality level is optimal but worsening, or acceptable and
improving.
4.4.1. Maps of Simpli ed General Classi cation
Following the scheme proposed in Table IV, different classi cation trials were
tested varying, case by case, the range of the regression line slope to be considered
TABLE IV(a)
Space-time classi cation
Trend indexa
Position index
Value Label Worsening Constant Improving
1.0 MCV
Contact this candidate