Hydrogeological Study of the Glacial—Fluvioglacial Territory of Grandate (Como, Italy) and Stochastical Modeling of Groundwater Rising

applied sciences

Article Hydrogeological Study of the Glacial—Fluvioglacial Territory of Grandate (Como, Italy) and Stochastical Modeling of Groundwater Rising

Ivana La Licata * ID , Loris Colombo ID , Vincenzo Francani and Luca Alberti ID

Dipartimento di Ingegneria Civile e Ambientale, Politecnico di Milano, 20133 Milan, Italy; [email protected] (L.C.); [email protected] (V.F.); [email protected] (L.A.) * Correspondence: [email protected]; Tel.: +39-02-2399-6663

Received: 13 July 2018; Accepted: 21 August 2018; Published: 24 August 2018

Abstract: On November 2014, the Municipality of Grandate, near Lake Como, had to deal with a great emergency that was caused by the flooding of factory undergrounds. The authors realized a hydrogeological study to understand the causes of groundwater flooding and to prepare a pre-feasibility study concerning possible actions for groundwater control. The hydrogeological structure is rather complex and required time-consuming reconstruction of the conceptual site model. A transient numerical model was developed to analyse the system behaviour in different scenarios. The flow model was calibrated in a steady and unsteady-state using the automatic calibration code Model-Independent Parameter Estimation (PEST). The study demonstrated that the reason for floods was mainly due to the concurrence of three causes: (1) the hydrogeological structure of the area was recognized as a stagnation zone, (2) groundwater rising, and (3) extremely heavy rainfall in 2014. Through the PEST RandPar function, 100 random rainfall scenarios were generated starting from rainfall data for the last 20 years. The model was used to run 100 1-year long simulations considering the probability distribution of recharge related to the 100 randomly generated rainfall scenarios. Through collecting the piezometric heads that resulted from the simulations, monthly probability curves of groundwater exceeding a threshold level were obtained. The results provided an occurrence probability of groundwater level exceeding the underground structures level between 12% and 15%.

Keywords: stagnation areas; groundwater rising; mathematical modeling; stochastic analysis; MODFLOW

1. Introduction The development of industry in several major cities in Europe caused groundwater levels to drawdown until they were tens of meters below ground level. As the water tables were so deep, it was often assumed that they were stable and could be safely ignored. Over the last 10–20 years, a reduction in groundwater abstraction has led to a rise in water levels almost everywhere in Europe [1–4]. Then, since the end of the last century, many European cities have been affected by the groundwater rising phenomenon that is the result in many causes, such as the deindustrialization process, urbanization, and the variation of the aquifers recharge rate because of climate change. These different conditions have led to local rises of the water table, ranging from 3 m to 30–40 m with different consequences [5], with the ﬂooding of underground structures among them. The rising groundwater trends are registered especially in particularly exposed areas, such as the depressions that in the recent past have been the site of lakes, swamps, and peat bogs. On November 2014, the Municipality of Grandate, located 10 km southward of Lake Como, had to deal with a great emergency that was caused by the ﬂooding of numerous undergrounds of buildings and factories that

Appl. Sci. 2018, 8, 1456; doi:10.3390/app8091456 www.mdpi.com/journal/applsci Appl. Sci. 2018, 8, 1456 2 of 17 were located in that area. Flooding produced several problems in terms of direct and indirect damages to the economy and the territory. Grandate is located in the Prealps area and in a rather complex geological context. The hydrogeological asset of glacial amphitheatres in the Lombard Prealps has some properties that lead to genesis of stagnation areas [6–8], caused by the convergence of groundwater flow from the glacial hills (moraines) towards the morphological depressions lying between the morainic arches, that are filled by fluvioglacial and colluvial sediments. Groundwater stagnation areas represent a zero flux and equilibrium points where applied forces are balanced in all directions [9,10]. The convergence of the flow lines towards these areas makes their drainage difficult, lowering the seepage velocity and generating groundwater stagnation. For these reasons, along the morphological depressions, groundwater level often approaches the ground surface, even generating springs, lakes (Sartirana and Alserio lakes are an example in the Prealps area), and swamps. Analysis and studies about stagnation points and their possible practical applications have been conducted by many researchers [7,11–15], providing that, in order to avoid the risks for existing and projected buildings related to this phenomenon, a good knowledge of the system and its behavior is the first requirement. Through the hydrogeological study that was carried out during this work (November 2015–November 2016), the Grandate area has been characterized for the first time—no continuous monitoring data existed in this zone before 2015—and has been recognized as an example of stagnation area. The reconstructed hydrogeological conceptual model has been the base for the implementation of a 3D numerical one that has been run for complete knowledge of the system behavior. Thereafter, the model has been used in order to assess the hydrogeological hazard for the underground structures because of the water table rise. Different scenarios of groundwater system evolution have been defined based on stochastically generated rainfall scenarios. The recharge has been considered as probability distributions and Monte Carlo simulations [16] have been carried out to obtain the probability distributions of the groundwater levels. Therefore, the occurrence probability of groundwater exceeding a defined threshold level has been assessed. Uncertainty is inherent in any model-based risk assessment [17] and this uncertainty is particularly high when analyzing future impact [18,19]. Regardless, the current paper demonstrates that a hydrogeological analysis coupled with mathematical modelling and stochastic analysis can provide satisfying approximate forecasting of the behavior of the piezometric levels and can be a useful tool for the assessment of risk related to groundwater flooding.

2. Materials and Methods

2.1. Geology of the Study Area The Como Province is located in the North-Central area of the Lombardy Region (Figure1). This part of the region is a recharge zone for the aquifers of the exploited area of Milan and is recharged by the zones in contact with the pre-Alpine rising area. The hilly and high plain of the Como province, given the geological importance that the glacial deposits—here well preserved—acquire in this territory and the peculiarity of their distribution, has been at the center of the scientific interest of many researchers, Italians, and foreigners since the beginning of the nineteenth century. In particular, Penck and Brückner [20] recognize four glaciations in this area called, from the most recent, Würm, Riss, Mindel, and Günz which are separated by three interglacial ones called: Riss-Würm, Mindel-Riss, and Günz-Mindel. Ugolini and Orombelli [21] and Billard [22] at first emphasize the correlation between morainic and fluvioglacial deposits, examining in detail the characteristics of the main soils. In 1987, Bini [23] reviewed the model of the four glaciations of Penck and Brückner and, through a detailed survey, divided the sedimentary bodies into “glacial complexes”. The studies that have been undertaken [23–25] show that, since the late Pliocene, the Lake Como area was subjected to 13 glacial episodes during which the glaciers extended to the upper Po Appl. Sci. 2018, 8, 1456 3 of 17

Plain. However, given the application purposes of this work, the traditional classiﬁcation is used in

Appl.this study.Sci. 2018, 8, x FOR PEER REVIEW 3 of 17

Figure 1. Study area (red square) and the study area’s location; Grandate municipality is contoured Figure 1. Study area (red square) and the study area’s location; Grandate municipality is contoured in black. in black.

Because of its fluvio-glacial fluvio-glacial origin, the the territory of of the Como province discloses singular morphological characteristics and a rather complex st structure.ructure. In the the late late Miocene, Miocene, the the pre-Alpine pre-Alpine arc arc was subjectedsubjected toto a a strong strong erosive erosive stage stage with with the the consequent consequent formation formation of canyons of canyons at existing at existing pre-alpine pre- alpinelakes. Duringlakes. During the Pliocene the Pliocene highstand highstand sea level sea stage, level stage, clay (Villafranchiano) clay (Villafranchiano) filled thefilled low the areas low inareas the inpre-Alps. the pre-Alps. From From late Pliocene—lower late Pliocene—lower Pleistocene Pleistocene there there was awas low a standlow stand sea level sea level with with glaciers glaciers that thatmoved moved southwards southwards and occupiedand occupied the pre-alps the pre-alps zone, zone, and the and area the was area covered was covered by fluvioglacial by fluvioglacial coarse coarsesediments sediments that subsequently that subsequent lithifiedly (Ceppo lithified Lombardo). (Ceppo Lombardo). During the Quaternary,During the some Quaternary, glacial-eustatic some glacial-eustaticphases occurred phases and the occurred bedrock and at the the foot bedrock of the at Como the foot pre-Alps of the was Como deeply pre-Alps etched was by deeply glacial massesetched bycoming glacial from masses Canton coming Ticino from (glacier-Faloppio, Canton Ticino Italy) (glacier-Faloppio, and Lario (Como Italy) side). and The Lario enormous (Como amountside). The of enormoussediments thatamount were of transported sediments bythat glaciers were transported and meltwater by filledglaciers the lowestand meltwater areas and filled formed the thelowest first areasmorainic and amphitheaters. formed the first Later, morainic due to the amphitheaters. decrease in the glacialLater, phenomena,due to the internaldecrease and in less the extensive glacial phenomena,moraines were internal formed. and Generally, less extensive the most moraines recent sedimentswere formed. of the Generally, morainic the amphitheaters most recent aresediments located ofat the a lower morainic altitude amphitheaters than the oldest are located ones. Theat a hydrographiclower altitude networkthan the oldest here is ones. mainly The developed hydrographic in a networkNorth-South here direction.is mainly developed in a North-South direction. From the hydrogeological point point of of view, the area has marginally been taken into consideration by the Consorzio Acque Potabili (CAP) publication [26] [26] and and by by some some subsequent subsequent studies studies [27] [27] where where authors examine the southern limits of the provinceprovince bordering the Milan one. A A summary of the hydrogeological aspects aspects of of the the Como Como area area is is presen presentedted by by Francani Francani et et al. al. [28,29] [28,29] and and Beretta Beretta [30]. [30]. These studies recognize three main aquifers in in th thee area and the presence of major water resources within paleo channels. A similar similar geological geological and and hydrogeological hydrogeological structure structure is is that that of of the the main main Lombard Lombard morainic morainic arches: arches: Lambro river, Molgora stream in Merate and Sartirana, Oglio river, and river Mincio. In all of these cases, in in the valley between a a morainic arch arch and the other, ponds, swamps, and even lakes can be noticed [31,32]. [31,32]. The The presence presence of of pi piezometricezometric depressions depressions in in correspondence with lacustrine basins is frequentlyfrequently reportedreported in in detailed detailed geotechnical geotechnical studies, studies, like like that bythat CEPAV by CEPAV regarding regarding the glacial the deposits glacial depositsof Garda of Lake Garda [33 ]Lake and the[33] study and the by study Arpa Lombardiaby Arpa Lombardia [34]. [34]. The Grandate municipality is is located located less less than than 10 10 km km south of Lake Como and its territory presents the characteristics that have already been desc described.ribed. From Figure 22a,a, inin fact,fact, itit isis possiblepossible toto recognize two morainic arcs that enclose the less topographically elevated zone of the Grandate plain. Seveso River crosses the plain from North to South. In the area, groundwater flows generally from NNW to SSE. For this study, a bigger area has been considered in order to implement the mathematical model (red square in Figure 2). Appl. Sci. 2018, 8, 1456 4 of 17

recognize two morainic arcs that enclose the less topographically elevated zone of the Grandate plain. Seveso River crosses the plain from North to South. In the area, groundwater ﬂows generally from NNW to SSE. For this study, a bigger area has been considered in order to implement the mathematical

Appl.model Sci. (red2018, square8, x FORin PEER Figure REVIEW2). 4 of 17

Figure 2. ((a)) Digital Elevation Model Model of of the study area, principal hydrographic network, and and model domain (red (red square); ( b)) stratigraphic logs logs (yellow (yellow dots), dots), and and realized realized cross-section cross-section traces traces (in (in black black SectionSection 4 and C). Section C).

2.2. Conceptual Site Model Reconstruction 2.2. Conceptual Site Model Reconstruction The reconstruction of the conceptual site model started from the geometries of aquifers that were The reconstruction of the conceptual site model started from the geometries of aquifers that provided by previous works and were improved through the analysis of 110 stratigraphic logs, which were provided by previous works and were improved through the analysis of 110 stratigraphic logs, have been used to obtain an accurate representation of the hydrogeological structure of the study which have been used to obtain an accurate representation of the hydrogeological structure of the study area (Figure 2b). The collected stratigraphic information has been compared with data from previous area (Figure2b). The collected stratigraphic information has been compared with data from previous studies [30,35], with the topographic data of the Digital Terrain Model (DTM) [36,37] and with the studies [30,35], with the topographic data of the Digital Terrain Model (DTM) [36,37] and with the geological map that was elaborated for the CARGproject (Cartografia Geologica e geotematica) by geological map that was elaborated for the CARGproject (Cartografia Geologica e geotematica) by the Istituto Superiore per la Protezione e la Ricerca Ambientale (ISPRA) [32]. For this study, eleven the Istituto Superiore per la Protezione e la Ricerca Ambientale (ISPRA) [32]. For this study, eleven cross-sections have been prepared, six with a NS direction (1–6) and five are WE oriented (A–E), as cross-sections have been prepared, six with a NS direction (1–6) and five are WE oriented (A–E), shown in Figure 2b. Section 4 and Section C are shown in Figure 3. as shown in Figure2b. Section 4 and Section C are shown in Figure3. What is possible to understand from the hydrogeological sections is that the Grandate area is a valley that is carved in the rocky substratum of Würm glaciers into which the glacial and alluvial deposits have accumulated. This formation is generally considered semipermeable because there are very few wells with screens in this unit and they are only in the most superficial and fractured part of the rock. Moving southward, the bedrock deepens, covered by two aquifers: the first, generally phreatic, corresponds to the gravelly-sandy complex, while the second, confined, is composed by the fluvioglacial Riss/Mindel deposits and the Villafranchian ones and is found more than 100 m below ground level. The first aquifer is separated from the deepest one by the glacial Riss/Mindel deposits that are about 30 m below the ground level. In the southern part of Grandate, the rocky substratum and the Villafranchian deposits rise again. The valley in the substratum tends in fact to close in this zone where the separation level between the first and the second aquifer also disappears. The aquifers here are thus in contact and reduce their thickness, making the outflow section southward thinner. From Section 4 and Section C (Figure3), it is possible to notice what was previously explained: the Grandate plain is like a big bucket, surrounded by morainic hills, where water flows from the topographically more elevated zones, infiltrates in the plain, and hardly discharges southward. In fact, for the uplift of the glacial Riss/Mindel deposits and the approaching of the morainic arcs, the outflow section shrinks in the southern zone of the plain, making it difficult for groundwater to flow downgradient.

Figure 3. Cross Section 4 and C passing through the Grandate plain. Appl. Sci. 2018, 8, x FOR PEER REVIEW 4 of 17

Figure 2. (a) Digital Elevation Model of the study area, principal hydrographic network, and model domain (red square); (b) stratigraphic logs (yellow dots), and realized cross-section traces (in black Section 4 and C).

2.2. Conceptual Site Model Reconstruction The reconstruction of the conceptual site model started from the geometries of aquifers that were provided by previous works and were improved through the analysis of 110 stratigraphic logs, which have been used to obtain an accurate representation of the hydrogeological structure of the study area (Figure 2b). The collected stratigraphic information has been compared with data from previous studies [30,35], with the topographic data of the Digital Terrain Model (DTM) [36,37] and with the geological map that was elaborated for the CARGproject (Cartografia Geologica e geotematica) by the Istituto Superiore per la Protezione e la Ricerca Ambientale (ISPRA) [32]. For this study, eleven cross-sectionsAppl. Sci. 2018, 8, 1456have been prepared, six with a NS direction (1–6) and five are WE oriented (A–E),5 of as 17 shown in Figure 2b. Section 4 and Section C are shown in Figure 3.

FigureFigure 3. Cross 3. Cross Section Section 4 and 4 and Section C passing C passing through through the Grandate the Grandate plain. plain.

In November 2015, a piezometric survey over 31 wells/piezometers and 5 hydrometers along Seveso River was carried out. The data that has been collected are not numerous because unfortunately in the past, the area has never been characterized from the hydrogeological point of view and there are not many monitoring points. The piezometric map (Figure4) confirms the conceptual site model reconstruction: the first aquifer, though presenting a moderate hydraulic conductivity, hosts a groundwater body that is sustained by finer sediments, which confine the second aquifer. Part of this groundwater flows toward the lake (as happens in some similar systems [38]) and part thereof flows southward through the Seveso River depression aquifer. Most of the groundwater, however, flows from the borders of the morainic hills that surround the plain toward the center of the plain and, with difficulty, discharges southward where the flow section shrinks. The low gradient (<0.5%) that is registered in the plain (Figure4) is a consequence of what was previously explained, because this hydrogeological structure has been recognized in many cases as a stagnation area. In Figure5, the groundwater levels that were monitored from March 2016 to present in two piezometers that are located in the flooded zone are represented. One of the two piezometers is monitored continuously through a data logger, while in the other, the groundwater level is manually recorded approximately every 15 days. The graph clearly shows a very fast response of the groundwater to no single rain events—especially when these exceed 20 mm—which is probably due to a very small effective porosity of the sediments hosting the groundwater body, but is undoubtedly caused by the fact that the zone is a stagnation area where the outflow is reduced due to the hydrogeological setting. Single events like the heavy rains in November and December of 2017 do not create great piezometric level variation as in the other cases, however an interruption in the downward trend can be observed. Appl. Sci. 2018, 8, 1456 6 of 17 Appl. Sci. 2018, 8, x FOR PEER REVIEW 6 of 17 Appl. Sci. 2018, 8, x FOR PEER REVIEW 6 of 17

Figure 4. Piezometric map of the first aquifer and measured points with their groundwater level (head Figure 4. PiezometricPiezometric map of of the the first ﬁrst aquifer aquifer and and measur measureded points points with with their their groundwater groundwater level (head expressed in m a.s.l.). Encircled in red is the zone that was flooded in 2014. Piezometric lines 305 m expressed in m a.s.l.). Encircled in red is the zonezone that was ﬂoodedflooded in 2014. Piezometric lines 305 m a.s.l. and 295 m a.s.l. mark the low gradient zone (stagnation area). a.s.l. and 295 m a.s.l. mark the lo loww gradient zone (stagnation area).

300.50 Real time monitored point Manually monitored point Rain Esternal temperature 300.50 Real time monitored point Manually monitored point Rain Esternal temperature 140.00 140.00 300.00 300.00 120.00 120.00 299.50 and Temperature [mm] Rain [C°] Rain [mm] and and Temperature [mm] Rain [C°] 299.50 100.00 100.00 299.00 299.00 80.00 80.00

298.50 298.50 60.00 60.00

298.00 40.00

Gorundwater Level [m LevelGorundwater a.s.l.] 298.00 40.00 Gorundwater Level [m LevelGorundwater a.s.l.]

297.50 20.00 297.50 20.00

297.00 0.00 297.00 0.00

Figure 5. Monitoring data for the two points located in the flooded zone. Figure 5. Monitoring data for the two poin pointsts located in the floodedflooded zone. In this study, the piezometric map, which was constructed by interpolating the November 2015 In this study, the piezometric map, which was constructed by interpolating the November 2015 surveyGenerally, data (Figure after 4), finding shows that out a and stagnation outlining area the would stagnation probably zones be present based onin the the Grandate flow network plain. survey data (Figure 4), shows that a stagnation area would probably be present in the Grandate plain. reconstruction,For this reason, it is essential the case toof understandGrandate—which what thehas physical a hydrogeological phenomena setting are that that control is similar the to origin that For this reason, the case of Grandate—which has a hydrogeological setting that is similar to that andof an presence extensive of part zero-velocity of the Lombard areas. territory, This knowledge as indicated is useful in the already for correctly mentioned interpreting studies the [31–34]— model of an extensive part of the Lombard territory, as indicated in the already mentioned studies [31–34]— resultshas been that examined are related in detail to the through behavior the of implementa the aquifer andtion theof a effects numerical of potential model. drainageThe implementation systems on has been examined in detail through the implementation of a numerical model. The implementation groundwaterphase was carried levels out (Figure taking6). into In fact, account in order the hypoth to achieveesis optimalof the stagnation results in area different existence pluviometrical which was phase was carried out taking into account the hypothesis of the stagnation area existence which was andsupported hydrological by the conditions, piezometric the map design and of the drainage geological interventions reconstruction in these of kindsthe area. of areas The mathematical requires more supported by the piezometric map and the geological reconstruction of the area. The mathematical model then allowed a better understanding of the factors that affect groundwater flow and helped to model then allowed a better understanding of the factors that affect groundwater flow and helped to evaluate the most suitable interventions for avoiding future groundwater flooding. evaluate the most suitable interventions for avoiding future groundwater flooding. Appl. Sci. 2018, 8, 1456 7 of 17 than the usual preliminary hydrogeological analysis of the most suitable water well location and a Appl.correct Sci. evaluation2018, 8, x FOR of PEER the REVIEW withdrawal and the water level lowering. 7 of 17

Figure 6. Example of a stagnation area (arrows represen representt flow ﬂow lines, the light-blue line represents the phreatic surface, and the black lines representrepresent thethe verticalvertical headhead contours).contours).

2.3. Mathematical Model In this study, the piezometric map, which was constructed by interpolating the November 2015 surveyThe data conceptual (Figure4 ),site shows model that has a stagnationbeen implemented area would in Modflow2005 probably be present [39] throughout in the Grandate an 80 × plain. 80 m grid Forand thiswas reason, refined the up case to of40 Grandate—which× 40 m in the area has where a hydrogeological flooding occurred setting in that 2014. is similar In the to vertical that of direction,an extensive the part model of the has Lombard been extended territory, based as indicated on the stratigraphic in the already characteristics mentioned studies and the [31 –geometry34]—has ofbeen the examinedaquifers in in the detail area. through Four layers the have implementation been identified of a according numerical to model. the defined The hydrogeological implementation phasesections was (Figure carried 7): out the taking first intorepresents account the the sha hypothesisllow aquifer of the that stagnation is constituted area existence by the whichpermeable was depositssupported of bythe the glacial piezometric and fluvioglacial map and the Würm geological phase, reconstruction the second follows of the area.the impermeable The mathematical layer, wheremodel present, then allowed the third a better and understandingfourth layers instead of the factorsrepresent that the affect deepest groundwater part of the flow hydrogeological and helped to systemevaluate (fluvioglacial the most suitable Riss/Mindel interventions and Villafranchian, for avoiding future respectively groundwater). The flooding. bottom of the model is represented by where the rocky formation where present and was set at a distance of 70 m from the layer2.3. Mathematical above where Model the identification of the substratum was not possible. The grid thus is composed by 87The rows, conceptual 76 columns, site modeland 4 layers has been (26, implemented448 active cells), in Modflow2005 representing [a39 40] throughout km2 area. General an 80 × Head80 m Boundarygrid and was conditions refined (GHB) up to 40have× 40been m inset the over area the where borders flooding of the model occurred domain in 2014. based In theon the vertical data thatdirection, was collected the model during has been the piezometric extended based survey on the(November stratigraphic 2015) characteristics which covered and a larger the geometry area than of the aquifersmodel. GHB in the have area. been Four chosen layers because have been of the identified absence according of geological to the or definedhydrogeological hydrogeological natural limitssections in (Figurethe proximity7): the of first the represents modelled thearea shallow and in aquiferorder to that make is constitutedthe influence by of the the permeable boundary conditionsdeposits of on the the glacial target and zone fluvioglacial as negligible Würm as phase,possible. the The second River follows condition the impermeable (RIV package) layer, was where used present,to represent the thirdthe Seveso and fourth River layers (Figure instead 7) that represent is a perennial the deepest river partoriginating of the hydrogeological about 10 km North system of (fluvioglacialGrandate and Riss/Mindel presents, in andthis Villafranchian,zone, a small respectively).river stage. The The Seveso bottom River of the has model little is influence represented on groundwater,by where the rockyas is possible formation to wherenotice presentfrom Figure and was4. The set River at a distance condition of 70was m set from on thethe layerbase aboveof the wheremeasured the identificationlevels (5 points of thealong substratum the river was path) not and, possible. for the The river grid thusbed, isconsidered composed a by hydraulic 87 rows, conductivity76 columns, andvalue 4 layersthat was (26,448 similar active to that cells), of the representing surrounding a 40sediments km2 area. [40]. General Head Boundary conditions (GHB) have been set over the borders of the model domain based on the data that was collected during the piezometric survey (November 2015) which covered a larger area than the model. GHB have been chosen because of the absence of geological or hydrogeological natural limits in the proximity of the modelled area and in order to make the influence of the boundary conditions on the target zone as negligible as possible. The River condition (RIV package) was used to represent the Seveso River (Figure7) that is a perennial river originating about 10 km North of Grandate and presents, in this zone, a small river stage. The Seveso River has little influence on groundwater, as is possible to notice from Figure4. The River condition was set on the base of the measured levels (5 points along the river path) and, for the river bed, considered a hydraulic conductivity value that was similar to that of the surrounding sediments [40].

Figure 7. Horizontal (a) and vertical (b) discretization (in the plant, red lines indicate the sections traces, the River condition is blue, while light blue represents GHB conditions contouring the model). Appl. Sci. 2018, 8, x FOR PEER REVIEW 7 of 17

Figure 6. Example of a stagnation area (arrows represent flow lines, the light-blue line represents the phreatic surface, and the black lines represent the vertical head contours).

2.3. Mathematical Model The conceptual site model has been implemented in Modflow2005 [39] throughout an 80 × 80 m grid and was refined up to 40 × 40 m in the area where flooding occurred in 2014. In the vertical direction, the model has been extended based on the stratigraphic characteristics and the geometry of the aquifers in the area. Four layers have been identified according to the defined hydrogeological sections (Figure 7): the first represents the shallow aquifer that is constituted by the permeable deposits of the glacial and fluvioglacial Würm phase, the second follows the impermeable layer, where present, the third and fourth layers instead represent the deepest part of the hydrogeological system (fluvioglacial Riss/Mindel and Villafranchian, respectively). The bottom of the model is represented by where the rocky formation where present and was set at a distance of 70 m from the layer above where the identification of the substratum was not possible. The grid thus is composed by 87 rows, 76 columns, and 4 layers (26,448 active cells), representing a 40 km2 area. General Head Boundary conditions (GHB) have been set over the borders of the model domain based on the data that was collected during the piezometric survey (November 2015) which covered a larger area than the model. GHB have been chosen because of the absence of geological or hydrogeological natural limits in the proximity of the modelled area and in order to make the influence of the boundary conditions on the target zone as negligible as possible. The River condition (RIV package) was used to represent the Seveso River (Figure 7) that is a perennial river originating about 10 km North of Grandate and presents, in this zone, a small river stage. The Seveso River has little influence on groundwater, as is possible to notice from Figure 4. The River condition was set on the base of the measuredAppl. Sci. 2018 levels, 8, 1456 (5 points along the river path) and, for the river bed, considered a hydraulic8 of 17 conductivity value that was similar to that of the surrounding sediments [40].

FigureFigure 7. HorizontalHorizontal ( (aa)) and and vertical vertical ( (bb)) discretization discretization (in (in the the plant, plant, red red lines lines indicate the the sections traces,traces, the the River River condition condition is is blue, blue, while while light light blue blue represents represents GHB GHB conditions conditions contouring contouring the the model). model).

Appl. Sci. 2018, 8, x FOR PEER REVIEW 8 of 17 The morainic arc that is located in the right part of the Grandate plain (Figure8) has been modelled with a lowerThe valuemorainic of hydraulic arc that is conductivity located in the compared right part to of the the surrounding Grandate plain zones (Figure and a8) recharge has been value equalmodelled to half of with the a one lower that value was of assigned hydraulic to conductivity the green areas. compared The correctto the surrounding representation zones of and this a zone is fundamentalrecharge value for equal modelling to half theof the system one that behavior was assigned given to thatthe green part ofareas. the The water correct that representation ﬂows toward the plainof arrives this zone from is fundamental the surrounding for modelling hills. Figure the system9 shows behavior the monthly given that rainfall part of the that water was that registered flows in the 2010–2016toward the period plain arrives in Como, from from the surrounding which the inﬁltration hills. Figure in 9 non-urbanizedshows the monthly areas rainfall (8·10 that−9 m/s)was has −9 beenregistered evaluated in using the 2010–2016 the Thornthwaite period in Como, method from and which assuming the infiltration a 30% in runoff. non-urbanized For the urbanized areas (8∙10 areas, m/s) has been evaluated using the Thornthwaite method and assuming a 30% runoff. For the the recharge has been set over each municipality as 15% of the public withdrawal, then taking into urbanized areas, the recharge has been set over each municipality as 15% of the public withdrawal, accountthen the taking network into account losses [ 41the]. Wherenetwork the losses withdrawal [41]. Where data the was withdrawal not available data (Luisagowas not available and Casnate con Bernate(Luisago municipalities), and Casnate con a usageBernate of municipalities), 160 L/year per a personusage of has 160 been L/year considered per person and has then been 15% of this valueconsidered was usedand then for the15% evaluation of this value of was recharge. used fo Urbanizedr the evaluation and of non-urbanized recharge. Urbanized areas and are shownnon- in Figureurbanized8. areas are shown in Figure 8.

FigureFigure 8. Urbanized 8. Urbanized (colored (colored zones) zones) and and non-urbanizednon-urbanized (dotted (dotted zones) zones) areas areas in the in themodel model domain. domain. The morainic arc in the right part of Grandate is contoured in black. The morainic arc in the right part of Grandate is contoured in black.

Figure 9. Monthly rainfall that was registered over 7 years in a meteorological station in Como. Appl. Sci. 2018, 8, x FOR PEER REVIEW 8 of 17

The morainic arc that is located in the right part of the Grandate plain (Figure 8) has been modelled with a lower value of hydraulic conductivity compared to the surrounding zones and a recharge value equal to half of the one that was assigned to the green areas. The correct representation of this zone is fundamental for modelling the system behavior given that part of the water that flows toward the plain arrives from the surrounding hills. Figure 9 shows the monthly rainfall that was registered in the 2010–2016 period in Como, from which the infiltration in non-urbanized areas (8∙10−9 m/s) has been evaluated using the Thornthwaite method and assuming a 30% runoff. For the urbanized areas, the recharge has been set over each municipality as 15% of the public withdrawal, then taking into account the network losses [41]. Where the withdrawal data was not available (Luisago and Casnate con Bernate municipalities), a usage of 160 L/year per person has been considered and then 15% of this value was used for the evaluation of recharge. Urbanized and non- urbanized areas are shown in Figure 8.

Appl. Sci. 2018, 8, 1456 9 of 17 Figure 8. Urbanized (colored zones) and non-urbanized (dotted zones) areas in the model domain. The morainic arc in the right part of Grandate is contoured in black.

Figure 9. Monthly rainfall that was registered over 7 years in a meteorological station in Como. Figure 9. Monthly rainfall that was registered over 7 years in a meteorological station in Como.

The calibration was initially performed in steady state using the inverse procedure through PEST [42] with reference to the piezometric survey that was carried out in November 2015. The calibration parameter was the hydraulic conductivity, the initial value of which was derived from a few hydraulic tests that were performed in some of the wells that were located in the model domain during the drilling phase. The available data have been interpolated in order to generate the initial hydraulic conductivity distribution for the four layers. The spatially distributed parameter was then the object of the automatic calibration process. Once the model was calibrated, the attention was focused on the area where flooding occurred in 2014 for the evaluation of the feasibility of groundwater level control through pumping wells. Some wells were then implemented in the model and were optimized, first using PEST to minimize their number and pumping rate, which was then done manually in order to set their position. The goal of the optimization was to maintain the groundwater depth around 9 m below the ground (level constraint), in steady state condition. This level was defined considering a safety distance from the underground structures, which in this zone generally reaches a depth of 4.5–5 m from the ground. A global objective function was constructed in order to take into account the level constraint and the minimization of the pumping rate. This second objective was obtained allowing PEST to set zero as the minimum possible flow rate per well and thereby, at the same time, to optimize the number of wells (which was constrained to a maximum of 10). The position of the wells was then manually arranged according to the area that was effectively available for their installation. The mathematical model was then calibrated in transient mode as well, using the hydraulic heads that were registered in the monitoring points, and was finally used to forecast the system behaviour in three different scenarios, namely Scenario1 (2014 rainfall and no interventions), Scenario2 (2014 rainfall and continuous extraction), and Scenario3 (2014 rainfall and three months extraction). The unsteady calibration was performed taking into account 1 year piezometric data that was collected during the project development (November 2015–November 2016) and has been carried out to define the effective porosity value and the recharge term for the first layer. During this second calibration phase, the still calibrated hydraulic conductivity distribution was not modified.

2.4. Stochastic Analysis The year 2014 has been recognized everywhere in Italy as a very exceptional year in terms of rainfall (100 years return time), as is possible to notice from Figure8. Anyway, during this study, the numerical model was also used to perform a stochastic analysis with the aim of assessing the probability that ﬂooding, such as what occurred in 2014, could happen again in the future. For this analysis, the available rainfall data in a meteorological station that was located close to Grandate was collected (20 years). It is possible to see in Figure 10a that the years 2014 and 2010 represent two peaks Appl. Sci. 2018, 8, 1456 10 of 17 in the graph, with an annual cumulative precipitation that exceeds 2000 mm. It is also interesting to note the great difference between these years and the following ones (2011 and 2015, respectively), which are among the lowest of the entire twenty-year period in terms of rainfall amount. However, in 2010, the problems that occurred in 2014 did not happen. This could be explained by looking at the average monthly rainfall trend (Figure9), showing that most of the rainfall that was observed in 2010 was recorded in warm or relatively hot months, such as May, August, and September when the evapotranspiration is high and, consequently, there is low inﬁltration of rainwater in the ground. Secondly, it is important to consider the process of deindustrialization that the area has faced: in 2010, just two years after the start of the 2008 economic crisis, the number of companies and production activities were higher than in 2014, the extraction from the aquifer was accordingly more intense, and the groundwater was at a greater depth. The probability distribution of the collected rainfall data has been compared with the theoretical one [43] in order to verify whether, as expected, the data could be represented by a Gaussian distribution (Figure 10b). The slight disagreement between the data and theoretical curve occurs just in the tails of the distribution (representative only of the extreme values) and could be considered acceptableAppl. forSci. 2018 the, 8 purposes, x FOR PEER ofREVIEW this analysis. 10 of 17

Figure 10. 20 years of available rainfall data (a) and a comparison between the rainfall distribution Figure 10. 20 years of available rainfall data (a) and a comparison between the rainfall distribution and and the Gaussian one (b). the Gaussian one (b). Starting from the monthly mean and standard deviation of the normally distributed rainfall Startingdata, 100 from random the monthly rainfall scenarios mean and were standard generated deviation through the of PEST the normally RANDPAR distributed utility [42]. rainfall This data, 100 randomutility rainfallallows the scenarios generating were of a generated series of random through values the PEST(rainfall RANDPAR values in this utility case) [ 42that]. Thisare utility distributed according to a probability distribution that is characterized by a determined mean and allows thestandard generating deviation. of In a seriesthis case, of randomthe 100 rainfa valuesll values (rainfall therefore values prove in this to be case) independent that are distributedand accordingnormally to a distributed. probability The distribution rainfall data that that was is thus characterized generated have by been a determined randomly sampled mean in and order standard deviation.to create In this100 annual case, rainfall the 100 scenarios rainfall from values which thereforethe recharge prove to be assigned to be independent to the stress periods and of normally distributed.100 transient The rainfall model runs data has that been was evaluated. thus generatedAmong the 100 have scenarios, been randomlythere are three sampled hypothetical in order to create 100years annual that referred rainfall to scenariosvery high rainfall, from which strong thedrought, recharge and medium to be assigned conditions. to theThese stress extreme periods of scenarios have been created by collecting for each month the 10th, 50th, and 90th percentile over the 100 transient model runs has been evaluated. Among the 100 scenarios, there are three hypothetical 100 generated rainfall data. Thereafter, the model was used to run 100 1-year long transient years thatsimulations referred considering to very high the rainfall,probability strong distribution drought, of recharge and medium related conditions.to the 100 randomly These extreme scenariosgenerated have beenrainfall created scenarios by (including collecting the three for each extreme month scenarios). the 10th, The transient 50th, and simulations 90th percentile (1 year) over the 100resulted generated in different rainfall piezometric data. Thereafter, level maps. the By modelcollecting was the usedpiezometric to run head 100 distributions 1-year long that transient simulationsresulted considering from the simulations, the probability monthly distribution probability of curv rechargees of groundwater related to theexceeding 100 randomly a threshold generated rainfalllevel scenarios have been (including obtained. the The three threshold extreme level scenarios). is the one of Thethe underground transient simulations buildings in (1 the year) flooded resulted in area (about 5 m below the ground). The initial head distribution that was used in these simulations different piezometric level maps. By collecting the piezometric head distributions that resulted from is the one that is related to the November 2015 monitoring campaign that can be considered as the simulations,representative monthly of the present probability piezom curvesetric setting. of groundwater As it is possible exceeding to see from a thresholdFigure 5, the level current have been obtained.average The thresholdgroundwater level level is in the the oneflooded of thearea underground is about 299 m a.s.l., buildings which inis similar the flooded to the value area that (about 5 m below thewas ground). measured Thein November initial head 2015. distribution The analysis thatwas wasthus usedintended in theseto assess simulations what the probability is the one that is relatedcould to the be November for groundwater 2015 monitoringto exceed the campaigndefined threshold that can starting be considered from the current as representative piezometric of the setting. present piezometric setting. As it is possible to see from Figure5, the current average groundwater level in the flooded3. Results area is about 299 m a.s.l., which is similar to the value that was measured in November 2015. The analysis was thus intended to assess what the probability could be for groundwater to exceed3.1. the Numerical defined thresholdModel starting from the current piezometric setting. The model was calibrated at the beginning in steady state and the general statistics are reported in Table 1. The model results are shown in Figure 11a where the residual values of hydraulic head are also indicated. Figure 11b shows the scatterplot of the observed vs. simulated heads. Calibration was performed using the piezometric levels of the November 2015 survey assuming that they could be representative of the average piezometric distribution in the area since no other useful data were available. The automatic calibration process with Pilot Point technique (PEST, [44]) was applied to the hydraulic conductivity parameter and resulted in a hydraulic conductivity distribution for each layer. Appl. Sci. 2018, 8, 1456 11 of 17

3. Results

3.1. Numerical Model The model was calibrated at the beginning in steady state and the general statistics are reported in Table1. The model results are shown in Figure 11a where the residual values of hydraulic head are also indicated. Figure 11b shows the scatterplot of the observed vs. simulated heads. Calibration was performed using the piezometric levels of the November 2015 survey assuming that they could be representative of the average piezometric distribution in the area since no other useful data were available. The automatic calibration process with Pilot Point technique (PEST, [44]) was applied to the hydraulic conductivity parameter and resulted in a hydraulic conductivity distribution for each layer.

Table 1. General statistics resulting from the model.

Statistical Indicators Value Target Number 28 Observation Range (m) 68.39 Minimum Residual (m) −1.09 Maximum Residual (m) 2.27 Residual Mean [M] (m) −0.06 Root Mean Square error [RMS] (m) 0.63 Mean Absolute error [MA] (m) 0.38 Mean Absolute error/Range (calibration ﬁt) 0.6% Appl. Sci. 2018, 8, x FOR PEER REVIEW 11 of 17

Figure 11. (a) Model simulation result and observed-simulated values—zoom on Grandate, (b) Figureobserved/simulated 11. (a) Model scatterplot simulation before result and after and the observed-simulated calibration procedure. values—zoom on Grandate, (b) observed/simulated scatterplot before and after the calibration procedure. The hydrogeological study and the model runs demonstrated that the reason for the 2014 flooding was mainly due to the concurrence of three causes: (1) the hydrogeological structure of the The hydrogeological study and the model runs demonstrated that the reason for the 2014 flooding area recognized as a stagnation zone, (2) the groundwater rising—which has been happening for was mainlyseveral due years to in the this concurrence area and generally of three in Lomb causes:ardy (1) [43,45]—and the hydrogeological (3) the abnormal structure quantity of of the area recognizedrainfall as that a stagnation occurred in zone, 2014 (100-years (2) the groundwater return period). rising—which has been happening for several years in thisOnce area the and model generally was calibrated, in Lombardy the attention [43,45 was]—and focused (3) on the the abnormal area where quantity flooding occurred of rainfall that occurredin in2014 2014 to (100-years assess the returnfeasibility period). of groundwater level control through pumping wells. The Onceoptimization the model process was calibrated,result shows the that attention 8 wells are was necessary focused to achieve on the the area goal, where with floodingpumping rates occurred in varying between 3 and 6 L/s for a total of about 40 L/s in the area (Figure 12). The action of pumping 2014 towells assess close the to feasibility the areas ofthat groundwater mainly experienced level controlthe flooding through problem pumping could thus wells. be an The effective optimization processsystem result to shows keep groundwater that 8 wells levels are under necessary control toand achieve to avoid the repetition goal, with of such pumping a phenomenon. rates varying between 3 and 6 L/s for a total of about 40 L/s in the area (Figure 12). The action of pumping wells close to the areas that mainly experienced the flooding problem could thus be an effective system to keep groundwater levels under control and to avoid the repetition of such a phenomenon.

Figure 12. (a) The area that was subjected to the flooding in 2014 (red grid) and pumping wells (blue circles) inserted into the model; (b) line of 9 m groundwater depth that was created by the action of the 8 wells in the November 2015 recharge condition. The line outlines the internal cells of the grid where the groundwater depth is 9 m or even bigger.

Appl. Sci. 2018, 8, x FOR PEER REVIEW 11 of 17

Figure 11. (a) Model simulation result and observed-simulated values—zoom on Grandate, (b) observed/simulated scatterplot before and after the calibration procedure.

The hydrogeological study and the model runs demonstrated that the reason for the 2014 flooding was mainly due to the concurrence of three causes: (1) the hydrogeological structure of the area recognized as a stagnation zone, (2) the groundwater rising—which has been happening for several years in this area and generally in Lombardy [43,45]—and (3) the abnormal quantity of rainfall that occurred in 2014 (100-years return period). Once the model was calibrated, the attention was focused on the area where flooding occurred in 2014 to assess the feasibility of groundwater level control through pumping wells. The optimization process result shows that 8 wells are necessary to achieve the goal, with pumping rates varying between 3 and 6 L/s for a total of about 40 L/s in the area (Figure 12). The action of pumping Appl. Sci. 2018wells, 8, 1456close to the areas that mainly experienced the flooding problem could thus be an effective 12 of 17 system to keep groundwater levels under control and to avoid the repetition of such a phenomenon.

Appl. Sci. 2018, 8, x FOR PEER REVIEW 12 of 17

Table 1. General statistics resulting from the model. Figure 12.Figure(a) The 12. area(a) The that area was thatStatistical was subjected subjected Indicators to to the the floodingfloodi ng in in2014 2014 (red (red grid)Value grid)and pumping and pumping wells (blue wells (blue circles) inserted into the model;Target (b) lineNumber of 9 m groundwater depth that was 28 created by the action of circles) inserted into the model; (b) line of 9 m groundwater depth that was created by the action of the the 8 wells in the NovemberObservation 2015 recharge Range condition. (m) The line outlines the 68.39 inte rnal cells of the grid 8 wells in thewhere November the groundwater 2015Minimum depth recharge is 9 m Residual condition. or even bigger. (m) The line outlines the−1.09 internal cells of the grid where the groundwater depth is 9 mMaximum or even bigger.Residual (m) 2.27 Residual Mean [M] (m) −0.06 3.2. Stochastic Analisys Root Mean Square error [RMS] (m) 0.63 Mean Absolute error [MA] (m) 0.38 Through collectingMean the Absolute head distributions error/Range (calibration that resulted fit) from the 0.6% 100 simulations that were run during the stochastic analysis that was carried out, a monthly probability curve of groundwater 3.2. Stochastic Analisys exceeding a threshold level was obtained. Figure 13 shows the probability distribution of the hydraulic Through collecting the head distributions that resulted from the 100 simulations that were run head for someduring months the stochastic of the analysis simulated that was year carried compared out, a tomonthly the elevation probability of curve the undergroundof groundwater buildings in order toexceeding point out a thethreshold hazard level for was the obtained. underground Figure buildings13 shows the (i.e., probability parking distribution area, cellars, of andthe building foundations)hydraulic to be head flooded. for some The months curves of the resulting simulated fromyear compared the stochastic to the elevation analysis of the show underground that, in the area, there is anbuildings occurrence in order probability to point out of the exceeding hazard for the underground groundwater buildings level (i.e., of the parking underground area, cellars, structures and building foundations) to be flooded. The curves resulting from the stochastic analysis show that, between 12%in theand area, 15%,there is respectively an occurrence probability in November of exceeding and the December. groundwater This level meansof the underground that underground flooding isstructures likely to between occur once 12% approximatelyand 15%, respectively every in 10 November years. Thus, and December. even if 2014 This was means a very that abnormal year in termsunderground of rainfall, flooding the is underground likely to occur once buildings approximately are exposed every 10 years. to anot Thus, negligible even if 2014 (equal was a to about 0.12–0.15)very occurrence abnormal probability year in terms to of berainfall, flooded the underground in the future buildings (considering are exposed no changingto a not negligible of groundwater (equal to about 0.12–0.15) occurrence probability to be flooded in the future (considering no changing extractionof conditions groundwater compared extraction conditions to the simulated compared situation).to the simulated situation).

Figure 13. Probability distribution of groundwater levels for some months of the simulated year. The Figure 13. Probability distribution of groundwater levels for some months of the simulated year. black line represents the buildings’ underground level (5 m below the ground) in the flooded zone. The black line represents the buildings’ underground level (5 m below the ground) in the ﬂooded zone. 4. Discussion The flooding event that occurred in Grandate in November 2014 pointed out the need to understand the hydrogeological system in this area and to conceive drainage interventions to lower and control groundwater levels. Appl. Sci. 2018, 8, 1456 13 of 17

4. Discussion The flooding event that occurred in Grandate in November 2014 pointed out the need to understand the hydrogeological system in this area and to conceive drainage interventions to lower andAppl. control Sci. 2018 groundwater, 8, x FOR PEER levels. REVIEW 13 of 17 The mathematical model, which was calibrated in an unsteady state as well as using the levels The mathematical model, which was calibrated in an unsteady state as well as using the levels that were registered in the piezometer that was located in the flooded area since March 2016, was used that were registered in the piezometer that was located in the flooded area since March 2016, was to forecast the system behavior in three different scenarios: used to forecast the system behavior in three different scenarios: Scenario 1 (Business As Usual: 2014 rainfall, no interventions are simulated)—Starting from the Scenario 1 (Business As Usual: 2014 rainfall, no interventions are simulated)—Starting from the calibrated situation, the recharge values of 2014 were set in the model and the year was simulated. calibrated situation, the recharge values of 2014 were set in the model and the year was simulated. Thus,Thus, the the model model was was used used to analyzeto analyze the the behaviour behaviour of of the the system system in in case case a verya very rainy rainy year, year, such such as as the yearthe 2014—listed year 2014—listed among among the first the 10 first rainiest 10 rainiest years year of thes of lastthe last century—should century—shou occurld occur again. again. Figure Figure 14 a shows14a theshows model the model results results at the endat the of end the of month the month of November of November in terms in terms of flooded of flooded cells cells at ground at ground level, 4.5level, m depth 4.5 m and depth 8 m and depth. 8 m depth.

FigureFigure 14. 14.Results Results of of Scenarios Scenarios 1 1 (a (),a), 2 2 ( b(b),), and and 3 3 ( c(c);); wellswells (circles)(circles) and flood ﬂoodeded zones zones (grey (grey cells) cells) at at differentdifferent depths. depths.

Scenario 2 (2014 rainfall and continuous extraction)—The same situation of Scenario 1 is simulated, however now the 8 wells are implemented in the model, pumping out continuously throughout the simulation time. The result is set out in Figure 14b, which shows the ﬂooded cells at ground level, 4.5 m depth and 8 m depth at the end of the month of November. As expected, the well pumping keeps groundwater level below the 4.5 m from ground level and the area is just partially ﬂooded at 8 m depth. It should be recalled that a very peculiar year in terms of rainfall (and then recharge) is simulated. Scenario 3 (2014 rainfall and 3 months extraction)—Scenario 2 has been replicated, but here the wells do not work continuously: their activation was presumed in August, about three months before the most critical period in terms of rainfall which, in these zones, is generally November. In fact, from the perspective of the groundwater control systems, to optimize the extraction and to carry it out close to the most critical period might prove convenient in terms of energy consumption and extracted groundwater disposal. The result of this last simulation is shown in Figure 14c, setting out the model response at the end of the 11th month of the simulation (November). It is still possible to notice the effect of the pumping, which makes it possible to maintain groundwater level below 4.5 m from ground level. The wells’ effect is obviously reduced at an 8 m depth from the ground compared to Scenario 2, given that the wells were active for just three months. However, as has already been said, a very exceptional year in terms of rainfall was simulated.

5. Conclusions In 2014, the Grandate plain experienced the problem of groundwater flooding of several underground factory buildings. The groundwater table rise can interfere with the underground structures, with obvious inconveniences and high economic damages. This happens especially in particularly exposed areas, such as the depressions that in the recent past have been at the site of lakes, swamps, and peat bogs. Grandate is an example of such areas: its hydrogeological setting, similar to that of extensive parts of the Lombard territory, makes this zone representative of a large number of areas that are affected by similar problems. The hydrogeological characterization that was carried out in Grandate after the flooding event and the numerical model that was implemented have demonstrated that the reason for the floods was the concurrence of three main causes: (1) the hydrogeological setting of the area was recognized as a stagnation zone, (2) the groundwater rising due to the process of deindustrialization that was happening almost everywhere in Lombardy, and (3) the abnormal quantity of rainfall that occurred in 2014 (100-years return period). The geological setting of the Grandate plain, representative of the Lombard morainic amphitheatres, makes the area particularly susceptible to groundwater flooding. The Grandate area has in fact been recognized as a stagnation area and its conformation and surroundings cause water to accumulate in the subsoil and flow with difficulty southward. Because of this geological conformation, a strong correlation between rainfall events and groundwater level increase has been observed, especially when intense rainfall occurrences happen, to the extent of those that were registered in 2014. Considering the variability of rainfall and, consequently, the infiltration rate influencing the groundwater table, a stochastic analysis was carried out in order to define probability exceeding curves in the critical area. Results confirm that the probability for groundwater to exceed the underground levels in the current piezometric setting is not negligible (12–15%), and the numerical model proves to be essential to quantify the effects of groundwater drainage that are operated by pumping wells in such a complex area. In order to control and forecast the system behavior, the monitoring of groundwater levels in real time by remote sensors is one of the first requirements. Moreover, a WebGis tool could be suitable for keeping the groundwater level variation under control and giving indications about the groundwater approaching an exceeded threshold level, whereupon a pre-alarm could be given and/or the pumping wells system could be activated. This kind of control of groundwater could be adequate for avoiding flooding occurrences in the future. Appl. Sci. 2018, 8, 1456 15 of 17

Author Contributions: The authors have contributed to this study by carrying out different activities including project organization and development, characterization surveys, model implementation, and calibration. All of the authors conceived and designed the methodology and discussed the simulation results. L.A. (scientific coordinator of the project) and I.L.L., as experts in the hydrogeology area and in groundwater numerical modeling through MODFLOW/SEAWAT codes, conceived and designed the project for Grandate. I.L.L. performed the survey activities on Grandate and the surrounding area, analyzed the data, and carried out the data organization. V.F., a retired engineering geology full professor, developed the conceptual site model of the study area. L.A. and I.L.L. implemented the 3D numerical model and I.L.L. and L.C., as experts in groundwater numerical modeling and stochastic analysis, performed the stochastic analysis. I.L.L. carried out the calibration process. All of the authors discussed the model results. Therefore, the manuscript was written by the four authors in equal parts. Funding: This research received no external funding. Acknowledgments: This study was supported by the Grandate Municipality. The authors did not receive funds for covering the costs to publish it in open access. The authors thank the Grandate mayor, Monica Luraschi, and Carlo Mancuso of the technical office for enabling the study and the field activities. Conflicts of Interest: The authors declare no conflict of interest.

References

1. Wilkinson, B. Rising groundwater levels in London and possible effects on engineering structures. In Proceedings of the 18th Congress of the International Association of Hydrogeologists, Cambridge, UK, 8–13 September 1985; pp. 145–156. 2. Lamé, A. Modélisation Hydrogéologique des Aquifères de Paris et Impacts des Aménagements du Sous-Sol sur les Écoulements Souterrains. Ph.D. Thesis, Ecole Nationale Supérieure des Mines de Paris, Paris, France, 2013; pp. 10–11. 3. Vasquez-Sune, E.; Sanchez-Vila, X.; Carrera, J.; Marizza, M.; Arandes, R.; Gutierrez, L.A.; Chilton, J. Rising groundwater levels in Barcellona: Evolution and effects on urban structures. In Groundwater in the Urban Environment: Problems, Process and Management; Balkema: Rotterdam, The Netherlands, 1997; pp. 267–271. 4. Beretta, G.P.; Avanzini, M.; Pagotto, A. Managing groundwater rise: Experimental results and modelling of water pumping from a quarry lake in Milan urban area (Italy). Environ. Geol. 2004, 45, 600–608. [CrossRef] 5. Dean, J.; Sholley, M. Groundwater basin recovery in urban areas and implications for engineering projects. In Engineering Geology for Tomorrow’s Cities; IAEG: Dublin, Ireland, 2006. 6. Kasenow, M. Applied Ground-Water Hydrology and Well Hydraulics, 3rd ed.; Water Resources Pubblication: Littleton, CO, USA, 2010; ISBN 1-887201-28-9. 7. Winter, T. Numerical Simulation Analysis of the Interaction of Lakes and Groundwater; Professional Paper 1001; U.S. Government Printing Office: Washington, DC, USA, 1976. 8. Anderson, M.P.; Cheng, X. Long- and short-term transience in a groundwater/lake system in Wisconsin, USA. J. Hydrol. 1993, 145, 1–18. [CrossRef] 9. Colombo, L.; Francani, V. Analytical method for stagnation point calculation: Theoretical developments and application to a hydraulic barrier design (Sicily, Italy). Q. J. Eng. Geol. Hydrogeol. 2014, 47.[CrossRef] 10. Colombo, L.; Cantone, M.; Alberti, L.; Francani, V. Analytical solutions for multiwell hydraulic barrier capture zone defining. Ital. J. Eng. Geol. Environ. 2012, 2.[CrossRef] 11. The Flow of Homogeneous Fluids through Porous Media. Available online: https://catalog.hathitrust.org/ Record/009073808 (accessed on 13 July 2018). 12. Guymon, G.L. Hydraulics of groundwater. Adv. Water Resour. 1980, 3, 193. [CrossRef] 13. Strack, O.D.L. Groundwater Mechanics; Prentice Hall: Englewood Cliffs, NJ, USA, 1989; ISBN 0-13-365412-5. 14. Christ, J.A.; Goltz, M.N. Hydraulic containment: Analytical and semi-analytical models for capture zone curve delineation. J. Hydrol. 2002, 262, 224–244. [CrossRef] 15. Christ, J.A.; Goltz, M.N. Containment of groundwater contamination plumes: Minimizing drawdown by aligning capture wells parallel to regional flow. J. Hydrol. 2004, 286, 52–68. [CrossRef] 16. Raychaudhuri, S. Introduction to Monte Carlo simulation. In Proceedings of the 2008 Winter Simulation Conference, Miami, FL, USA, 7–10 December 2008; pp. 17–21. [CrossRef] 17. Walker, W.E.; Harremoës, P.; Rotmans, J.; van der Sluijs, J.P.; van Asselt, M.B.A.; Janssen, P.; Krayer von Krauss, M.P. Defining Uncertainty: A Conceptual Basis for Uncertainty Management in Model-Based Decision Support. Integr. Assess. 2003, 4, 5–17. [CrossRef] Appl. Sci. 2018, 8, 1456 16 of 17

18. Gattinoni, P.; Francani, V. Depletion risk assessment of the Nossana Spring (Bergamo, Italy) based on the stochastic modeling of recharge. Hydrogeol. J. 2010, 18, 325–337. [CrossRef] 19. Troldborg, M.; Nowak, W.; Tuxen, N.; Bjerg, P.L.; Helmig, R.; Binning, P.J. Uncertainty evaluation of mass discharge estimates from a contaminated site using a fully Bayesian framework. Water Resour. Res. 2010, 46. [CrossRef] 20. Penck, A.; Brückner, E. Die Alpen im Eiszeitalter; Tauchnitz: Leipzig, Germany, 1909. 21. Ugolini, F.; Orombelli, G. Notizie preliminary sulle caratteristiche dei depositi glaciali e fluvioglaciali tra l’Adda e l’Olona in Lombardia. Rend. Ist. Lomb. Sc. Lett. 1968, 102. 22. Billard, A. Paléosols quaternaires de l’Alta Pianura de Milan, interprétation stratigraphique et paléoclimatique. Quaternaire 1973, 10, 267–286. [CrossRef] 23. Bini, A. L’apparato Glaciale Wurmiano di Como. Ph.D. Thesis, University of Milan, Milan, Italy, 1987. 24. Bini, A.; Felber, M.; Pomicino, N.; Zuccoli, L. La massima estensione dei ghiacciai (MEG) nel territorio compreso tra il Lago di Como, il Lago Maggiore e le rispettive zone di anfiteatro. Geol. Insubrica 1996, 1, 65–77. 25. Bini, A. Stratigraphy, chronology and palaeogeography of Quaternary deposits of the area between the Ticino and Olona rivers (Italy-Switzerland). Geol. Insubrica 1997, 2, 21–46. 26. CAP. Consorzio per l’acqua potabile ai comuni della provincia di Milano. In Depauperamento Delle Risorse Idriche Sotterranee nel Comprensorio Della Brianza a Nord del Canale Villoresi; CAP: Milano, Italy, 1972. 27. Cavallin, A.; Francani, V.; Mazzarella, S. Studio idrogeologico della pianura compresa fra Adda e Ticino. Costruzioni 1983, 326, 39. 28. Francani, V.; Beretta, G.P.; Scesi, L. Caratteri idrogeologici della parte meridionale della Provincia di Como. Le Strade 1981, 1199, 1–20. 29. Francani, V.; Beretta, G.P.; Scesi, L. Studio idrogeologico del lago Alserio (Provincia di Como). Costruzioni 1983, 334–335, 1–27. 30. Beretta, G.P. Studio idrogeologico del territorio canturino (CO). Costruzioni 1984, 344. 31. Brenna, S. Suoli e Paesaggi Della Provincia di Brescia; Ersaf: Milan, Italy, 2004. 32. Bini, A.; Sciunnach, D.; Bersezio, R.; Scardia, G.; Tomasi, F. Note Illustrative Della Carta Geologica D’Italia Alla Scala 1:50’000–096 Foglio Seregno; Regione Lombardia: Milano, Italy, 2005. 33. Infrastrutture Ferroviarie Strategiche Definite Dalla Legge Obiettivo n. 443/01 Linea a.v./a.c. Torino—Venezia Tratta Milano—Verona Lotto Funzionale Brescia-Verona Progetto Definitivo-Studio Geologico-Idrogeologico di Dettaglio. Available online: https://www.google.com.tw/url?sa=t&rct=j& q=&esrc=s&source=web&cd=1&ved=2ahUKEwiHrNHvyoTdAhVCfd4KHXWtB7EQFjAAegQIARAC& url=http%3A%2F%2Fwww.va.minambiente.it%2FFile%2FDocumento%2F163751&usg= AOvVaw0TA50q8jqc688RSoblIlHI (accessed on 13 July 2018). 34. ARPA Lombardia. Stato Delle Acque Sotterranee, Area Idrogeologica Oglio-Mincio. Rapporto Annuale 2014; ARPA Lombardia: Milan, Italy, 2015. 35. Alba, M. Studio Idrogeologico Della Parte Meridionale Della Provincia di Como; Politecnico di Milano: Milan, Italy, 2004. 36. Tarquini, S.; Isola, I.; Favalli, M.; Mazzarini, F.; Bisson, M.; Pareschi, M.T.; Boschi, E. TINITALY/01: A new Triangular Irregular Network of Italy. Ann. Geophys. 2007, 50, 407–425. [CrossRef] 37. Tarquini, S.; Vinci, S.; Favalli, M.; Doumaz, F.; Fornaciai, A.; Nannipieri, L. Release of a 10-m-resolution DEM for the Italian territory: Comparison with global-coverage DEMs and anaglyph-mode exploration via the web. Comput. Geosci. 2012, 38, 168–170. [CrossRef] 38. Anderson, M.P.; Munter, J.A. Seasonal reversals of groundwater flow around lakes and the relevance to stagnation points and lake budgets. Water Resour. Res. 1981, 17, 1139–1150. [CrossRef] 39. Harbaugh, A.W. MODFLOW-2005, The U.S. Geological Survey Modular Ground-Water Model—The Ground-Water Flow Process; U.S. Geological Survey Techniques and Methods 6-A16; U.S. Department of the Interior, U.S. Geological Survey: Reston, VA, USA, 2005. 40. Genoni, P. Corsi D’acqua Principali Della Provincia di Milano; ARPA Lombardia: Milano, Italy, 2004. 41. Alberti, L.; Cantone, M.; Colombo, L.; Lombi, S.; Piana, A. Numerical modeling of regional groundwater flow in the Adda-Ticino Basin: Advances and new results. Rend. Online Soc. Geol. Ital. 2016, 41, 10–13. [CrossRef] Appl. Sci. 2018, 8, 1456 17 of 17

42. Doherty, J.E. Calibration and Uncertainty Analysis for Complex Environmental Models—PEST: Complete Theory and What it Means for Modelling the Real World; Watermark Numerical Computing: Brisbane, Australia, 2015; ISBN 978-0-9943786-0-6. 43. Colombo, L.; Gattinoni, P.; Scesi, L. Stochastic modelling of groundwater ﬂow for hazard assessment along the underground infrastructures in Milan (northern Italy). Tunn. Undergr. Space Technol. 2018, 79, 110–120. [CrossRef] 44. Doherty, J.E. PEST, Model-Independent Parameter Estimation, User Manunal, 5th ed.; Watermark Numerical Computing: Brisbane, Australia, 1989. 45. Alberti, L.; Francani, V. Studio idrogeologico sulle cause del sollevamento della falda nell’area milanese. Geoing. Ambient. Min. 2001, 104, 257–264.

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).