Seawater intrusion and infiltration modelling coupled to digital tools to avoid high saline concentrations in reclaimed water: application in coastal central Italy Open Access

Industrial activities have been traditionally based on a linear approach where water and other resources are extracted from the natural environment, used in the technosphere, and then disposed back as waste. Water is a highly valuable resource and plays a critical role in human well-being, socio-economic development, and the sustainability of ecosystems. Water also contains valuable materials such as nutrients and is viewed as a carrier of energy (Nika et al. 2020). Excessive water extraction, pollution, and waste generation lead to environmental degradation and increase the risk of scarcity, especially in those regions that are more vulnerable to climate change (Sgroi et al. 2018). In this respect, by the year 2030, half of the river basins of the European Union are expected to be affected by water scarcity (WRE 2022). Plenty of regions worldwide face competition for water supply between economic sectors, such as industry, agriculture, tourism, livestock, and household activities, which can create lots of socio-economic conflicts and political instability (FAO 2014). Moreover, in coastal areas, high pressures on groundwater sources commonly imply saline intrusion issues with the subsequent chlorine contamination of water sources and hydro-morphological changes (Hussain et al. 2019). This is becoming an increasing concern in overexploited coastal aquifers as those of many Mediterranean countries, such as Italy, Spain, Croatia, Greece, Albania, Turkey, and Morocco (Foglia et al. 2020; Telahigue et al. 2020). Saline intrusion affects the quality of water for multiple applications such as domestic and agricultural consumption, as well as for industrial processes that require low-salinity water such as cooling towers and others (Asgari et al. 2023).

A transition from linear to circular economy practices is thus imminent (Foglia et al. 2023). In this respect, it is highly relevant to transform the interlinkages between the water and industrial sectors. Currently, industrial activities are highly water-demanding, being the second largest water-consuming sector after agriculture. In addition, water scarcity implies higher energy consumption to treat the water and achieve the high-quality standards required for industrial applications. In the European framework, the EU has set ambitious environmental-related objectives such as the EU Green Deal and the Zero-Pollution Action Plan for the European industry to decrease its environmental footprint. It is thus essential to search for industrial solutions that are framed within the circular economy paradigm and aimed at sustainable development, climate resilience, and/or mitigation of the effects of water scarcity (Pitkänen et al. 2016). In this respect, two issues need to be addressed: the reduction of water losses (efficient use of water) and the search for alternative water sources (water reuse). These aspects can be approached by implementing valuable and strategic solutions that fall within the Water Smart Industrial Symbiosis (WSIS) approach. Industrial symbiosis promotes a circular economy by connecting water and waste flows among industrial actors and turning them into resources (Shi 2020; Henriques et al. 2021). Within this approach, wastewater is reused as an alternative water source in industries. It, therefore, enables threduction of energy consumption and environmental impacts of the current water management sector and the dependency on potable water abstractions so that they can be future-proofed towards both climatic and market stresses and avoid competition with other water-consuming sectors such as agriculture and domestic activities.

As an example of WSIS, a public–private partnership (PPP) has been established in Central Italy between an urban water utility, an industry, and a technology provider. This PPP was created to treat municipal wastewater from two municipalities for industrial reuse, thus optimizing water management at the regional level by reducing the industrial consumption of high-quality groundwater. For this, a water reclamation facility (WRF) was constructed to substitute the main source of the water supply of the industry (high-quality groundwater) with fit-for-purpose-treated municipal wastewater for industrial reuse (reclaimed water). This way, the water utility can ensure long-term water supply to the local population by freeing up groundwater sources for drinking water supply, generating added value for the local environment and society. Besides, the industry can secure a long-term water supply to fulfil its industrial needs, avoiding any possible restriction in case of severe drought, which could cause catastrophic consequences with potential definitive closure of the site. A maximum of 3.8 Mm³/year of treated municipal wastewater is required by the industry nowadays (120 L·s⁻¹ (LPS) on average), but the industrial plant has expanded in terms of both production and variety of products, thus intensifying its water requirements with a high risk of increasing the already high hydric pressure to which this coastal region belongs. In fact, due to excessive groundwater extraction, the catchments of the municipalities considered are impacted by unpredicted and relevant seawater intrusion that increases the salt (and subsequently chlorine) concentrations in groundwater. This saline intrusion reaches the local sewage systems. Consequently, conductivity in sewage shows peaks of up to 7,000 μS·cm⁻¹ (according to the information provided by the local water utility).

High salinity impacts municipal wastewater treatment plants (WWTPs) in several ways. It can interfere with chemical reactions of processes such as coagulation-flocculation, affecting the mixed liquor of the activated sludge microbiome by increasing the osmotic pressure of their cells, ending up with a decrease in the efficiency of pollutant removal (Egea-Corbacho et al. 2021). Moreover, as urban WWTPs are normally not designed to remove these ions, those salinity peaks can hinder their reuse as they will make surpass the acceptable limit of conductivity accepted for industrial reuse, i.e., 2,000 μS·cm⁻¹ in the case of this study. Irregular and unpredicted peaks of other pollutants contained in the sewage (e.g., surfactants, chemical oxygen demand (COD), and hardness) can also exceed the quality standard required for industrial reuse. Under these undesirable circumstances, the water cannot be provided to the industry, and the WRF is forced to either discard the treated water directly to the sea or post-treat the effluent water to improve its quality, with the subsequent increase in treatment costs. For example, reverse osmosis (RO), which is typical technology in desalination processes, requires around 0.869–1.5 kWh·m⁻³ (Melgarejo et al. 2016; Akhoundi & Nazif 2018), whereas the consumption of conventional treatment is usually in the range of 0.35–0.8 kWh·m⁻³ (Jiménez-Benítez et al. 2020; Acién et al. 2023). In addition to the problems caused by saline intrusion, during heavy rain events in combined sewerage systems where sewage and stormwater are collected in unified networks (such as those in Italy), runoff volume may exceed the capacity of WWTPs, causing plant failures (Abbasi et al. 2021). To avoid this, a relief mechanism, known as combined sewer overflow, is inserted in the sewage system, but these CSOs can be a relevant source of environmental pollution if runoff peaks are often emitted through them (Botturi et al. 2020). Some mismanagement situations often occur due to the problems related to saline intrusion and heavy rain periods, e.g.: (i) excessive water losses (discharges) due to a lack of prevention of saline intrusion control; (ii) increasing water treatment costs to unsustainable levels; and (iii) a lack of treatment and reuse of the wastewater discharged by CSOs. These issues are challenging, but recent advances in digitalization have allowed the implementation of predictive models for estimating variations in water quality prior to their occurrence. In this respect, early warning systems (EWS) can allow risk minimization and rapid reactions if hazardous events are detected. This enables us to avoid delays in decision-making due to the lag time for data acquisition from grab sampling, measuring, and analysis, thus reducing inefficiencies in water management (Radini et al. 2023).

This study, which is based on a WSIS approach implemented in the ULTIMATE H2020 project, aims to analyze the effects that saline intrusion has on the quality of treated water received by a water reclamation plant (WRP), to maximize the industrial reuse of wastewater. To do so, an innovative method was proposed for identifying seawater intrusion (in terms of quantity and quality of water) in complex sewage systems. Seawater intrusion and infiltration models were developed, calibrated, and validated for the sewage networks of two municipalities from a coastal area of Central Italy. This kind of modelling can help understand the hydrodynamics and variations in the wastewater quantity and quality during seawater intrusion and/or precipitation events (Salvadore et al. 2015; Janga Reddy & Nagesh Kumar 2020), which can help support monitoring and address the limitations of analytical methods to some extent. However, to achieve this target and obtain a reliable assessment, the integrated models should be properly calibrated and validated. The data produced by these models will be used by a smart-equalization (SEQ) system and matchmaking platform (MMP) to distribute treated water to different reuse applications (or post-treatment processes) according to their quality, following a fit-for-purpose approach (Radini et al. 2023). This aims to maximize reuse possibilities, avoiding (as much as possible) direct discharges to the sea and the need to use intensive post-treatment processes to obtain the reused water with appropriate quality. This study shows the preliminary results of the models and digital tools implemented in this ULTIMATE case study, which defines a flexible and replicable WSIS approach.

Both municipalities have around 30,000 inhabitants and their sewage networks are gravity and pressure systems. Wastewater coming from both sewage networks accounts for a total of 6 Mm³·year⁻¹ and is forwarded to WWTP-1 and WWTP2 (further information in Section 2.3). These depuration plants are located in very similar areas (coastal towns and a few kms from the sea). Effluents from both urban WWTPs are directed to a WRP. In this context, WWTP-1 channels treated water to a storage tank, which is connected to the WRF via an underground pipeline approximately 11 km in length. In contrast, WWTP-2 is situated close to the WRF, with a pipeline that operates by gravity.

On this pipeline, there is an automatic valve that allows the partial conveyance of the effluent from the WWTP-2 to the equalization tank at the head of the WRF (Section 2.4). Moreover, depending on the quality of water produced by the WRF, the reclaimed water will be used for industrial or agricultural purposes by MMP (Section 2.5).

The Environment Protect Agency's storm water management model (SWMM) was used to model the quantity and quality of sewage in the catchments of the two municipalities, including wastewater production by domestic users and intrusion by leaking sewers. The water flow in the sewer systems was modelled using a dynamic wave approximation (Rossman 2010). The flow resistance is described by the Manning equation. The curve number function was used to predict intrusion into the ground during rainfall events. An inflow function was used to predict these contributions. The flow rate of wastewater from domestic and industrial services was determined on the basis of potable water consumption, reduced by the sewer flow discharge coefficient α (usually about 0.8–0.9) and divided by categories (domestic, non-residential domestic, and industrial). In addition, water collected from wells was also considered, applying the same flow coefficient α, primarily for agricultural and non-residential domestic, which is common in the area of the case study. The developed algorithm includes the following steps:

• (a) Determination of sub-catchments and virtual points with aggregated intrusion.

• (b) Development of simulators of inflow (quantity, quality of wastewater) to WWTPs for t = 0–24 h using the multilayer perceptron (MLP) method. This method was chosen for its ability to model complex, nonlinear relationships and provide accurate predictions in hydrological modelling contexts (Giberti et al. 2024). Compared to other machine learning methods such as random forests, boosted trees, extreme gradient boosting, the predicted relationships of MLP (input–output) can be selected by the model builder according to the process. The work was guided by these considerations and by the fact that the MLP model has been used many times to predict flows in sewer networks and can be considered, to some extent, as a reference solution (Jamali et al. 2018; Jato-Espino et al. 2018).

• (c) Optimization of the quantity and quality (measured as the concentration of COD, total suspended solids (TSS), NH4-N, Cl) of intrusion at virtual nodes (calibration of SWMM models for dry and wet conditions).

Further information on the proposed methodology can be found in Section S1.

Based on the methodology developed at the stage of creating the catchment models, sub-catchments were separated (Figures S1, S2) where virtual nodes were introduced with aggregated seawater intrusion and infiltration values, i.e., quantity and quality of total wastewater streams (infiltrated water + sewage). A total of 8 sub-catchments were separated for Mu-1, including the main side channels, and separate typical areas (Figure S1). For sub-catchments 1, 6, 7, and 8, the points where aggregate seawater intrusion was introduced were located in cross-sections representing the connection of side channels. For sub-catchments 2, 3, 4, and 5, the points, where aggregate intrusion was introduced, were placed in the centre of the sub-catchment. The catchment model Mu-1 includes 9 pumping stations, 224 junctions, 227 sub-catchments, 3 storm overflows, and 10 outlets, including one outlet to WWTP-1. During rainfall events, stormwater runoff from a total area of 520 hectares, with impervious surfaces covering 13–69% of the area, flows into the sewer system.

In the case of Mu-2, 9 sub-catchments, which include the main side channels, were separated (Figure S2). The catchment includes eight pumping stations, whose characteristics (level on–off pump, discharge curve, cross-sectional area in plan) were obtained from the water utility. The model represents 230 nodes, 250 pipes, and one outlet (e.g., the inflow to WWTP-2). For sub-catchments 1, 2, 5, and 6, the points, where aggregate intrusion was introduced, were also in cross-sections representing side channels. Point 4 was located on the main channel to separate aggregate intrusion in the Sub1 sub-catchment. Points 7, 8, and 9 were positioned on the main channel. For sub-catchments 2, 3, 4, and 5, the points, where aggregate intrusion was introduced, were included in the centre of the sub-catchment. The stormwater contribution was not calculated in this case as their relevance to the final water quality was much lower than in Mu-1 (Section S7).

Data on channel retention for the separated sub-catchments are given in Table S2.

The calibration was performed for dry weather, identifying the typical inflow from each customer group and intrusion (according to Section S6) on a daily cycle (t = 0–24 h). The calibration of the models for dry weather was performed identically for both municipalities, but the model in Mu-1 also included surface runoff during rainfall events. The rainfall–runoff process is modelled as nonlinear reservoirs that include intrusion, evaporation, and surface runoff. Assumptions for the calibration during rainfall are given in Section S7.

Qualitative calibration was performed by entering in SWMM the pollutant concentrations measured during sampling campaigns in the selected sampling points (Section S2), and the optimized intrusion (inflow flow rate and concentration of Cl, COD, NH4-N, and TSS). Model outputs were then checked by comparing them with the laboratory analysis values (Section S7). Model evaluation was performed using the determination coefficient (R²) and per cent bias (PBIAS) (Section S3).

From the developed models, the quantity and quality of the wastewater entering both WWTPs in dry conditions were simulated during 2 consecutive days (springtime) with a 5-min resolution.

The water line of WWTP-1 consists of a pre-treatment system composed of screening and sand removal units, primary sedimentation, biological treatment with nitrification–denitrification and aerobic treatment, secondary sedimentation, and final disinfection. WWTP-2, after the pre-treatment consisting of screening and sand removal, has two biological treatment lines: one with conventional denitrification and nitrification/oxidation stages in different reacting zones, and another one that alternates biological treatment cycles in the same reactor.

It must be noted that the measurement of macro-pollutants in urban WWTPs is generally not monitored online but measured in grab or composite samples using lab analyses (Ruano et al. 2009). Consequently, the characteristics of the effluents of WWTP-1 and WWTP-2 were not simulated but obtained from data measured in the plants during a 5-year period (Table 1).

Even though data were under legal limits for discharge, some of the parameters did not accomplish the requirements for industrial reuse (i.e., TSS < 2 mg·L⁻¹; COD < 10 mg·L⁻¹). This is addressed after post-treatment at the WRP. Its process consists of several units. Following an equalization unit, the flow is divided into two treatment lines and goes to coagulation-flocculation units. The water is then submitted to lamella clarification, sand, and biological filtration to remove the remaining solids. Disinfection is achieved through ultraviolet (UV) lamps placed downstream (Figure S4). The produced sludge is collected, treated, and disposed of together with the waste-activated sludge of WWTP-2. However, this post-treatment is not able to significantly reduce the electrical conductivity. In fact, at the WRF effluent, electrical conductivity remained at average values of 2,320 ± 483 μS·cm⁻¹ during the measurement period, which is generally over the limit required for the industry, i.e., 2,000 μS·cm⁻¹. For this reason, the post-treatment process is planned to be implemented by an RO unit to reduce salinity to maximize water reuse for industrial purposes. The RO unit is not available currently, but it will be considered in the MMP developed in this study (Section 2.5).

SEQ is the algorithm used to control the valves at the effluent of both WWTPs that connect to the WRF. It aims to minimize the introduction of salts to the reclaimed water facility by discharging the water streams that present high-salinity peaks.

If from phase 1, the conductivity of WWTP-1 is lower than that of WWTP-2, the first pump that draws water from WWTP-1 is turned on. If this flow rate is not enough to fill the equalization tank, the second pump in WWTP-1 is also turned on and, if it is still not enough, the valve of WWTP-2 is opened as well. All these processes, including the opening level of the valves, are regulated by the setpoint at the equalization tank. The amount of flow that cannot be stored by the tank (from WWTP-1 or WWTP-2, depending on the criteria above) is directly discharged to the sea (i.e., not sent for reclamation). The functioning of the implemented SEQ of this study was simulated during 2 days by assuming the conductivity and flow entering in both WWTPs was the same as those simulated for their influent streams, i.e., water losses and salinity removal were considered negligible during their primary and secondary treatments at both WWTPs.

The MMP is a decision support tool that aims to identify the most suitable application for the reuse of the water treated in the WRP. It is especially useful in cases where industrial reuse is not possible due to high conductivity or excess of reclaimed water is produced (reclaimed water production > industry requirements). In particular, the platform considers two macro-categories: (i) industrial reuse and (ii) agricultural reuse. Each category is, in turn, divided into subcategories with an associated threshold value of conductivity. For industrial reuse, the tool always tries to satisfy the water demand of the company, i.e., a maximum of 483 m³·h⁻¹. Only the flow in excess goes to agricultural reuse.

The functioning of the implemented MMP of this study was simulated for 2 days assuming that 93.4% of the water treated by RO is recovered, whereas the rest is lost as brine (Figure 3). With respect to agricultural reuse, its reusable capacity will depend on the type of crop present in the area and its ability to tolerate water with a certain conductivity (Table S7). For this springtime simulation study, the selected crops and their respective areas were tall wheat grass (500 ha), red grass (500 ha), spinach (100 ha), and lettuce (100 ha), totalling 1,200 ha. This area represents a hypothetical allocation within the agricultural land available in the local area and surrounding zones, specifically chosen to focus the simulation on selected crops. Depending on the final conductivity of the reclaimed water used in agriculture, the platform will choose the most appropriate crop to be irrigated and indicate the percentage of tolerance of the selected crop(s). With respect to soils, the specific salinity tolerance of each soil type has not been considered in the current algorithm to maintain a simplified and flexible approach. The focus on crop tolerance provides practical guidance for irrigation strategies compatible with general soil conditions.

For the average daily inflow to WWTP-1 (V = 5,700 m³·day⁻¹), seawater intrusion was determined at virtual points (Qzb, Szb) for sub-catchments Sub1–Sub8 (Table 2). It must be noted that intrusion was considered constant throughout the time for both rainy and dry weather. The reliability of the results is confirmed by fitting the measurement data at the inflow to the WWTP and at the measurement points for the calibration (Figure S5) and verification phase (Figure S6) of the obtained model. Moreover, for the calibration step, it was found that for 80% of the measurement points for Cl, COD, NH₄-N, and TSS, the PBIAS value was not greater than 8.4, 12.2, 14.2, 12.02%, and for verification 7.2, 12.6, 15.7, 12.4%, respectively. The greatest intrusion was shown for sub-catchment Sub7 (Qzb = 169.41 m³·day⁻¹, Szb = 2,241 mg·L⁻¹) and the lowest for Sub1 (Qzb = 8.64 m³·day⁻¹, Szb = 1,000 mg·L⁻¹). These intrusion flows only accounted for 0.1–2.9% of the total influent flow to WWTP-1. However, the chloride concentration on these streams (1,000–2,365 mg·L⁻¹) was significantly higher than that of the domestic wastewater, i.e., 399 ± 10 mg·L⁻¹, so the conductivity supplied by this seawater intrusion was relevant input to the system. On the other hand, the contribution of other pollutants from seawater intrusion was negligible as the values of COD, NH_4, and TSS were significantly lower than those of the domestic water.

Regarding rainy weather, the default and post-calibration values of the parameters describing surface runoff in the SWMM model are given in Table S5. The prediction error of surface runoff volume for the calibration data was for all sub-catchments less than 10% (Table S6). Moreover, there was a high correlation between measured and predicted data using SWMM for COD (R² = 0.87), NH₄-N (R² = 0.95), and Cl (R² = 0.96). In addition, it was found that for 80% of the measurement data, the PBIAS value does not exceed 16.6% (COD), 17.1% (NH₄-N), 18.1% (TSS), and 14.7% (Cl). More details can be found in Section S7.

For the average inflow of wastewater to WWTP-2 (V = 9,500 m3·day⁻¹) and measured wastewater quality, the intrusion was determined at the measurement points (Qzb and Szb) for the virtual points (Table 3). The reliability of the obtained data (Qzb and Szb) was confirmed by the high goodness of fit of the flows at the measurement points (R² = 0.948–0.976, PBIAS = 4.2–6.7%).

The predictions for high-resolution flow measurements (Figure S9) and effluent quality for the learning set (Figure S10) and test set (Figure S11) with respect to COD accounted for R² = 0.951–0.978, PBIAS = 7.44–9.81%. For the other pollutants, they were NH4-N (R² = 0.935–0.954, PBIAS = 5.50–8.54%), Cl (R² = 0.923–0.959, PBIAS = 4.70–6.21%), and TSS (R² = 0.961–0.970, PBIAS = 6.30–7.36%). The greatest intrusion was obtained at points P1, P6, and P7, which fluctuates between Qzb = 302.4–328.32 m³·day⁻¹, corresponding to loading rate of chloride (LCl) = 311.90–453.95 kg·day⁻¹. The lowest intrusion was found for point P5, where Qzb = 129.60 m³·day⁻¹ and LCl = 129.60 kg·day⁻¹. It could be noticed that the intrusion in the sewage network of Mu-2 was considerably higher than that of Mu-1, i.e., five-fold higher considering the total average intrusion. As a consequence, the supply of contaminants to the network was also higher in this case although chloride was the only relevant pollutant affecting the influent wastewater to WWTP-2.

The infiltration in the sewage networks can be described by a linear combination of aggregate infiltration (Qzb) at the different points selected for each catchment, i.e., points P1–P8 to determine the daily inflow to WWTP-1, and P1–P9 to describe the daily inflow to WWTP-2. Both models were able to predict daily inflow quite accurately as a high goodness of fit was obtained (R² = 0.992–0.997). Detailed information on the optimization of these models can be found in Section S8.

It must be pointed out that despite the high goodness of fit of the result obtained, the SWMM programme is of limited use in this context. Analyses aimed at integrating the SWMM model and machine learning models in the context of salinity error predictions would be recommended for future works.

On the other hand, higher peaks of conductivity (>2,600 μS·cm⁻¹) will complicate the industrial reuse, so that they generally coincide with the higher peaks of discharge.

These results give some indications of the usefulness of the SEQ. If this tool was not installed, two extreme situations would occur:

• (i) If only the 2,000-μS·cm⁻¹-conductivity threshold had been considered, all the wastewater produced during the whole 2D period (i.e., 29,636 m³) would have been discarded, thus forcing the industry to satisfy their requirements for those days (i.e., 120 LPS) with groundwater.

• (ii) If no wastewater had been discarded, a total amount of 5,568 m³ of high-conductivity water would have been introduced to the system during these 2 days. This would have implied a 23% increase in the average conductivity of the water arriving at the WRF, which would entail higher post-treatment costs.

To reduce the adverse effects of saline intrusion in this area, this study proposes a digital solution that combines a dynamic simulation model that predicts seawater infiltration and runoff with digital tools, i.e., SEQ (control algorithm) and MMP (decision support system). The proposed solution aims to predict the periods where significant peaks of salinity enter the sewage network of two municipal WWTPs and one WRF. The tools aim to distribute the wastewater and reclaimed water streams to diverse applications (industrial, agricultural) and/or treatments (conventional treatment, RO) to maximize the amount of wastewater reused in an efficient and sustainable way.

The SWWM models developed in this study were able to predict with relatively high resolution both the quantity and quality of the seawater infiltrated into the sewage networks of the municipalities evaluated. In both municipalities, seawater infiltration seemed to be a relevant factor affecting the sewage entering the urban WWTPs. Hence, this model could be used to implement an EWS that would enable the WWTPs to detect in advance peaks of salinity (and therefore conductivity) that could affect the downstream treatment processes and distribution of wastewater. In fact, the 2D simulation carried out in this study showed wastewater conductivity remained in the range of 2,100–2,700 μS·cm⁻¹, which is over the established limit for industrial reuse. To avoid excessive conductivity entering the WRF, the SEQ system developed in this study enabled a discharge of 19% of the total to reduce salinity loads by 23%, therefore improving the quality of reclaimed water. Of the total reclaimed water produced, 87% was sent to the industry and the remaining 13% would be used for irrigation, thus maximizing the amount of water reused locally. The approach implemented in this study would be very useful to be replicated in coastal areas where saline intrusion is relevant.

This research work was supported by the H2020 Programme which funded ‘ULTIMATE’ project under the grant agreement 869318. The authors would also like to acknowledge C. Bruni and M. Ciampechini for their support in the data collection.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.