A water quantity-quality monitoring system to assess the impact of anthropogenic activities on urban rivers

Abstract

River monitoring plays a pivotal role in evaluating urban water quality status, detecting pollution events and their potential sources, and guiding the development of effective intervention strategies. In complex river networks, water quality degradation can stem from urban factors, such as inadequate sewer system management or runoff from contaminated surfaces during rainfall, as well as from external inputs linked to agricultural or industrial activities in surrounding areas. Accurately identifying the primary drivers of water quality deterioration is essential for designing targeted action plans and predicting their effectiveness. Although sporadic water sampling or visible indicators, such as foul odors during low-flow periods, can reveal quality impairment, more detailed information is required to differentiate pollution sources, thus, to implement appropriate restoration and mitigation measures. This study presents a monitoring system designed and tested to distinguish pollutants originating from urbanized areas from those associated with external activities (e.g., agriculture, industry), addressing the common challenge of prioritizing interventions for urban river improvement. The Sile River and the city of Treviso (Italy) were selected as a case study to validate the system and monitoring approach. Developed between 2021 and 2023, the system is characterized by six stations equipped with radar sensors for continuous water level and velocity measurements, three of which also feature multiparametric probes and refrigerated samplers. Two additional portable samplers allow for flexible sampling at other relevant locations. Both high-frequency (grab samples) and low-frequency (daily composite samples) monitoring strategies were implemented. Since 2023, the system collects data on the urban area’s impact during both dry and wet weather conditions and identifies temporal behaviour of physical and chemical species, providing a comprehensive overview of the Sile River’s current status. Data analysis developed through correlation, multilinear regression, principal component analysis, and load balance methods reveals which water quality parameters serve as indicators of specific pollution origins, ultimately supporting targeted improvement actions. Escherichia coli and P_tot resulted to be primarily associated with urban activities and indicative of sewer system contributions in both downtown and surrounding areas, whereas NO₃-N reflected agricultural inputs. Insights gained during the first two years of monitoring confirm the system’s effectiveness in assessing urban river quality and distinguishing between urban and agricultural contributions. Furthermore, the findings inform the design of monitoring programs for similar urban river contexts. After two years of operation, the study summarizes the strengths and limitations of the implemented system and evaluates its applicability to other case studies.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-51624-6.

Introduction

The progressive degradation of urban river water quality has been identified in the scientific literature as “urban river syndrome”[1]. The alteration of river water geochemistry and morphology, can result in temporary or chronic degradation, compromising habitats for riparian flora and fauna and reducing the recreational value of these systems and their surrounding areas for local communities (e.g., Thames¹ and Seine² rivers). Among the numerous factors affecting urban river water quality^3,4, alterations to the urban landscape driven by population growth play a major role⁵. These impacts are further exacerbated by the increasing frequency and intensity of rainfall events associated with climate change⁶. Urbanization processes lead to an expansion of impermeable surfaces, significantly altering the hydrological cycle by reducing rainfall infiltration and groundwater recharge, while simultaneously generating larger volumes of surface runoff that are conveyed to drainage systems in a shorter time^7–9. Most cities are served by combined sewer systems (CSS), which convey both domestic wastewater and stormwater runoff during rainfall events. These systems are often old, degraded, and, in some cases, not designed to accommodate the increased runoff volumes. Consequently, the frequency of surcharge events, including backflows through manholes and overflows from the CSS (CSO – combined sewer overflows –) to nearby water bodies has increased¹⁰. Because wastewater contain high concentration of pollutants, CSOs are widely recognized as one of the primary sources of pollution in urban rivers^11–14. Even in cities equipped with separated sewer systems (drainage system separated from the sanitary one), pollutants accumulated over the impermeable surfaces, such as heavy metals, hydrocarbons, and sediments¹⁵, can be mobilized during high intensity rainfall events (first-flush phenomenon¹⁶⁾ till to receiving water bodies.

In addition to more frequent intense rainfall events, climate change is also increasing the occurrence and duration of drought periods¹⁷, and more recent studies have focused on the adverse effects of drought on river discharge and dilution capacity^18–20. Finally, the growing presence of emerging contaminants, including pharmaceuticals, personal care products, and veterinary drugs²¹, along with microplastics²², represents an additional and increasingly significant challenge for urban river water quality management.

The growing awareness among both scientific and technical communities regarding the risks associated with diffuse and point-source pollution of water bodies⁴, has prompted international and national authorities to introduce standards and promote actions aimed at minimizing and ultimately eliminating sources of pollution, while protecting and restoring water bodies (e.g. the Water Framework Directive - WFD²³- and the Federal Water Pollution Act²⁴⁾. The WFD establishes a framework compelling EU countries to protect, improve, and restore water bodies to achieve good quality status, by implementing first a suitable Water Monitoring Program (WMP)²⁵. The problem of planning and/or optimizing a WMP for surface waters has been widely discussed in literature with the aim of addressing most of the important aspects, such as the selection of representative sampling locations, the definition of data collection frequency (high – HF – and low – LF), which is strictly related to the monitoring techniques (continuous sensors, grab or composite sampling), and of the number of significant parameters^26,27.

For both economic and technical reasons, the spatial density and temporal frequency of monitoring systems are usually limited, leading researchers to question whether the data adequately capture the complexity of the processes affecting river water quality. The debate was further prompted by the advent of in-situ technologies for continuous environmental monitoring, which have enabled a growing body of research to highlight the added value of HF monitoring strategies^28–32. Specifically, these studies have shown that HF monitoring allows to better characterize pollution sources as well as capture the highly dynamic features of rivers²⁹, and revealed significant discrepancies between HF-based assessments and conventional regulatory compliance testing. Weisner et al.[³³], through a refined stream monitoring approach, assessed that the pesticide risks in surface water may not be recognized by the WFD stream monitoring approach. Earlier, similar considerations were made by Skeffington et al.[³⁴] who investigated the accuracy and precision of WFD chemical status determination of groundwater-fed streams under specific sampling strategies, and by Bowes et al.[³⁵] who assessed the error introduced in phosphorous load estimation when monthly sampling is employed in lieu of weekly measurements. Jozic et al.,[³⁶] concluded that historical LF data on bathing water quality near coastal areas can introduce biases in planning bathing water quality monitoring and suggested the need of increasing the sampling frequency during the bathing season. Similarly, Jung et al.[³⁷] discussed how a limited temporal resolution for river monitoring can lead to imprecise and inaccurate classification especially when pollution sources, delivery, and dilution processes are dynamic. Reynolds et al.[³⁸] through a spatially distributed continuous real-time data discussed how nitrate concentration and flux estimation uncertainty is affected by the timing and frequency of monitoring strategy adopted, while Thompson & Jordan³⁹ highlighted how sampling methods affect the estimation of suspended solids loads. Again, few studies have also pointed out that LF data may not be suitable if the purpose of the WMP is providing input to models, and that HF data is needed to ensure their reliability^40–43. Conversely, Dai et al.[⁴⁴] demonstrated that when the temporal variability in water quality is negligible relative to the spatial differences recorded across monitoring stations distributed throughout the catchment, the number of the latter can be increased while reducing the sampling frequency at each location to offset the associated costs.

A review of the current literature reveals that every catchment has its own unique constraints, and it is generally accepted that a ‘one-size-fits-all’ WMP is not feasible²⁶. For this reason, sources apportionment techniques⁴⁵ have been widely applied in literature to help identifying and quantifying pollution sources at both the catchment^46,47 and urban⁴⁸ scales, and in optimizing monitoring networks by either reducing the number of stations, parameters or temporal frequency of samplings. Among others, factor analysis, positive matrix factorization models, cluster analysis and the principal component analysis (PCA) are the most used. Guerreiro et al.[⁴⁹], employed few of these statistical techniques to explore the optimization of a monitoring network aimed at evaluating the impact of urban effluents on a river within a small, largely artificial, catchment. The study focused on reducing stations, parameters, and temporal sampling frequency, relying however on LF dataset of standard physicochemical variables, including nitrogen and phosphorus. The effectiveness of these techniques in identifying pollution sources, whether instantaneous or continuous, point-source or diffuse, is strongly dependent on both spatial distribution and temporal frequency of the achieved data⁵⁰.

The design of a new monitoring network must be driven by the primary objectives of the monitoring program²⁵(e.g., compliance testing, identification of pollution sources, assessment of pollutant dynamics during both normal and extreme events), a preliminary analysis of the investigated areas (e.g. land cover and land use, pollution sources, river network spatial distribution, etc.), and the prevailing financial and technical constraints⁵¹.

Most studies available in literature focus on medium- to large-scale catchments⁵², from hundreds to thousands of square kilometers, which are characterized by heterogeneous land uses with multiple and diverse pollution sources, and affected by the hydrologic regime of rivers^53,54. Pollution originating from agricultural areas is commonly addressed, analyzing the temporal evolution of mainly nitrogen and phosphorous^30,55,56. Halliday et al.[⁵⁷] were among the first to investigate an ‘urbanized’ medium scale catchment using HF data, even though half of the study area is devoted to agricultural activities. Results from sub-sampling of HF monitoring data showed that the influence of frequency on status classification is parameter-specific and strongly dependent on the temporal variability of the monitored constituents.

As in many other papers, quality data monitored in one station is associated with discharge data achieved in a different location, mainly due to different technical and economic installation constraints. Although the importance of coupling water quality and discharge measurements is not widely discussed in the literature, discharge measurements are essential not only to acknowledge the hydrological processes occurring in the watershed⁵⁸, but also to evaluate pollutant loads affecting waterbodies⁵⁹. This coupling helps in planning mitigation measures and targeting water pollution prevention actions⁶⁰.

Given the widespread need to prioritize interventions aimed at improving urban river quality, an approach that can adequately account for both urban small-scale processes and external contributions is required.

The present study applies a local WMP⁶¹ to an area that has a limited spatial extent (about 130 ha), crossed by a lowland spring-fed river, where 95% of the surface is characterized by artificial land cover, while the surrounding areas consist predominantly of agricultural lands. Thus, the co-existence of multiple pollution sources (urban/agricultural) operating across a wide range of scales, considering spatially and temporally constrained processes within the investigated area as well as broader spatial and temporal dynamics beyond its boundaries, necessitated the design of a specifically tailored^11,62,63 WMP. This was aimed at achieving comprehensive control of inflows and outflows within the boundaries of an urban area, and at developing a robust water quantity-quality balance to support future local interventions. To this end, advanced monitoring stations were installed at strategic locations, each equipped with a set of sensors to measure both discharge and water quality parameters. At each station, more than one monitoring technique was employed, to account for the spatial distribution and temporal evolution of pollutant sources. The simultaneous deployment of multiple monitoring sensors and techniques at the same locations represents a combination that, to the authors’ best knowledge, has rarely been applied in its entirety within a single study. Within the technical and economic constraints, additional monitoring sites were established to measure either discharge or water quality parameters, with the aim of accurately defining hydraulic boundary conditions and assessing water quality at specific locations within the study area. In addition to HF monitoring of few common physicochemical, periodic analyses of pollutants associated with urban and rural effluents were investigated.

Following a detailed discussion on the development of the WMP, detailing the considerations that informed its design, and the challenges encountered during implementation as well as the strategies adopted to overcome them, an analysis of the data collected over the first two years of operation is presented.

The main objectives of this study are summarized as follows. The first objective is to design, develop, and test a monitoring system capable of distinguishing pollutants originating from urbanized areas from those associated with external sources, namely agricultural activities. The second objective is to collect HF data at small spatial scale on the water quality status of urban rivers and channels within a small urban area. Finally, the study aims to derive insights to support the design and implementation of targeted monitoring programs applicable to other similar urban contexts. This is particularly relevant considering that the downtown areas of most European cities are limited in size, and rivers crossing these cities are affected by both internal and external pollution sources. Experience from the present work may therefore prove useful in addressing the common challenge of prioritizing interventions for urban river improvement.

Materials and methods

Study area

The Sile River is a lowland spring river crossing the Venetian Plain (Veneto Region, Italy) from northwest to southeast, flowing towards the Adriatic Sea. With a total length of about 90 km and a watershed area of approximately 750 km² (Fig. 1a), it is the longest lowland spring river in Europe, part of the Natural Protection Park of the River Sile and of the Natura 2000 network defined by the European Council⁶⁴.

Fig. 1.

Overview of the study area: (a) Sile river watershed with the two water intakes along the Piave River, (b) longitudinal section A-A of the system of aquifers characterizing the watershed, (c) map of the river network crossing the upper part of the Sile River watershed subdivided into 3 sub-basins according to the ARPAV monitoring stations (66–79 – 330) and land use classification for each sub-basin, (d) blow-up of the area of Treviso with three ARPAV stations surrounding downtown as closing sections for the three sub-basins (d).

It originates as groundwater emerges to the surface along the border between two distinct geological units: one composed of coarse-grained alluvial deposits to the north, and the other characterized by a sequence of fine sediment layers to the south (Fig. 1b). The watershed is mainly exploited for agricultural purposes (Fig. 1c and d), and it is characterized by a dense network of artificial channels that are fed by two water intakes (Fener – Belluno and Nervesa della Battaglia – Treviso) along the Piave River and discharging into the Sile River. This network, together with few other spring rivers, serves for irrigation purposes during the dry periods (April to September) and as drainage system in the rainy seasons, receiving contributions from the agricultural areas nearby.

Most of the cited rivers eventually cross the municipality of Treviso (Fig. 2), where the urban planning developed over the years in connection with the river network. The city currently accounts for 85’000 inhabitants, 8000 of which live in the downtown (130 ha) inside the historical walls. The downtown sewer system is not connected to a Wastewater Treatment Plant (WWTP); sewage is collected from private houses, and, after primary treatment (e.g. Imhoff tanks), the effluent is discharged into the waterbodies through the CSS^65,66. In Fig. 2 the sanitary sewer system serving the areas outside the historical walls and connected to the WWTP is shown together with the CSS serving the downtown area and discharging to the watercourses.

Fig. 2.

The downtown area of Treviso and locations of the monitoring stations along the Sile River and its main tributary, the Botteniga channel. The WWTP, the metereological station and both sanitary and combined sewer system are also reported. (Figure created with QGIS v.3.28.4-Firenze https://qgis.org/, basemap source: OpenStreetMap contributors https://www.openstreetmap.org/)

This practice led to a gradual deterioration of water quality, partially mitigated by numerous hydraulic drops and relatively high flow velocities. However, periodic decreases in flow rates, e.g. for maintenance of channels, have highlighted ongoing water quality issues in the city’s waterways.

According to the WFD, a WMP consisting of 13 water quality stations was developed by the Regional Environmental Protection Agency (ARPAV) from 2010 to assess the quality status of the main rivers in the catchment. A total number of 13 water quality stations were established in the upper part of the Sile River watershed between 2010 and 2021. These stations are unevenly distributed across the upper part of the Sile River watershed (approximately 400 km²), with three stations specifically surrounding the downtown area of Treviso (Fig. 1c and d). Each monitoring station collects four samples annually for chemical and physical analysis. In the uppermost reach of the Sile River, about 12 km upstream from Treviso, a water level monitoring station was installed in 2019. Apart from this station only some sporadic measurements were carried out in the past.

To overcome the previously cited problems related to the sewer system serving the downtown area, a comprehensive initiative was launched by the local Integrated Water Service (Alto Trevigiano Servizi spa – ATS spa) to transition the entire municipality from a combined to a separated sewer system. Indeed, despite the absence of CSOs, the water courses are influenced by CSS effluents and by discharges from extensive agricultural areas. Additionally, pollutants that accumulate on urbanized surfaces during dry periods are washed into water bodies through the drainage network during rainfall events, further complicating the water quality landscape⁶⁷.

To explore the potential improvement in water quality, it is crucial to differentiate between the impacts of the CSS and other pollution sources affecting the physical and chemical parameters of the Sile River and its tributaries adopting a new monitoring system.

The monitoring system

Figure 2 depicts the dense network of rivers and channels crossing the downtown area, along with the monitoring positions that characterize the WMP. The number and spatial arrangement of monitoring stations were determined to differentiate between contributions from downtown activities and external sources, as well as to potentially identify the spatial locations of main pollution sources within the downtown area. The number and locations of monitoring stations along the Sile River and Botteniga were selected to develop a water quantity and quality balance around the downtown and to evaluate seasonal contributions related to agricultural practices and other urban areas outside the downtown. The final placement was conditioned by technical constraints related to sensor installation (e.g. availability of a straight river reach and of electrical supply in public areas, see Section 1Technical details regarding the monitoring station set-upof the Supplementary Materials) Introduction, operational requirements (e.g. accessibility and safety) and environmental constraints (e.g. recreational use of the Sile River).

Based on these criteria, positions A and B along the Sile River (west and est to downtown) and position E along the main tributary the Botteniga channel (north to downtown) were selected for the installation of both quantity and quality instruments, considering canal inputs A and E and the output B within downtown to develop a global balance. The water level and velocity in two additional small tributaries were monitored (positions C and F), while measurements at station D, downstream station B along the Sile River, provides a global discharge control at the downstream boundary condition. Moreover, two mobile quality stations were then employed to detail the quality status of the small channels (P1, P2, P3, P4) inside downtown.

It should be noted that the previously mentioned ARPAV stations surrounding the downtown area (66, 79 and 330 in Fig. 1d) are located near, for station A (0.8 km from station 66) or at the same point, for station B and E stations. Although ARPAV samples are usually collected with different time lag and techniques (one grab sample every three months collected at the river surface), as well as different time instants along the day, historical records of water quality measurements have been used in choosing the most proper parameters to be monitored.

Meteorological data were collected from the nearest regional station, located immediately northwest of the city center. Hourly observations of the year 2024 were analyzed to characterize air temperature and precipitation patterns. Mean air temperature ranged from 4 °C in January to 26.8 °C in August, with an average of 14.7 °C. Precipitation intensity averaged 0.2 mm h⁻¹, while the maximum (54.8 mm h⁻¹) was recorded on the evening of 23 November (Fig. 3).

Fig. 3.

Climatic data: total precipitation (mm), maximum intensity (mm/h) and mean air temperature (°C) for each month of the year 2024 recorded at the ARPAV metereological station in Treviso.

Each station is equipped with two non-contact radar sensors: one for water level (OTT RLS by Corr-Tek Idrometria Srl^®) and another for surface velocity (SVR100 by Corr-Tek Idrometria Srl^®), to obtain discharge evaluation. This technology offers several benefits for outdoor applications⁶⁸; however, specific installation requirements must be met. Among the most critical are selecting a straight river reach with a regular cross-section and positioning the sensors at the centre of the river to accurately capture surface velocity in the most active flow area. Whenever possible, installations should take advantage of existing bridges.

Because historical water-level and discharge data for these rivers are sporadic, a high-frequency data resolution (5 min) was selected to investigate the hydrologic response of the catchment at the monitoring sections during dry, wet, and agricultural seasons. This high temporal resolution also informed the selection of the monitoring frequency for water-quality data.

From the water surface elevation h and the surface velocity v measurements, the mean velocity and the wetted area , where is the constant value that express the linear relationship between the surface velocity and the mean velocity , can be computed obtaining as result the discharge . Bathymetry and velocity surveys have been developed through Acoustic Doppler Current Profiler (ADCP SonTek M9) to define the cross-section geometry and to calibrate the parameter (Fig. 4). Several ADCP surveys during different river hydraulic conditions were developed to obtain a robust stage-discharge relationship considering also the uncertainty related to algae growth. In each station, surveys have been developed for discharge values up to about one half of the peak recorded during flood events. This is because the Sile is a spring river with a limited catchment surface so that the discharge variation is limited too. Being the water level controlled by two hydraulic power plants in the main river and several control facilities, it manifests negligible variations, and this sometimes leads to uncertainty in the definition of a common stage-discharge relationship as discussed in the following Section Monitoring requirements and operational considerations.

Fig. 4.

Discharge measurements obtained through water level and surface velocity data using radar sensorstogether with bathimetry and velocity surveys through ADCP.

Regarding water quality assessment, a total of 16 parameters were monitored (Table 1). The selection of both number and type of parameters represented a compromise between the need to identify and, if combined with discharge data, potentially quantify pollution sources, and the laboratory analysis costs. Following a preliminary assessment of site characteristics (e.g. land use, anthropogenic activities, etc.) and literature review, the parameters most representative for distinguishing among pollution sources were selected.

Table 1.

For each monitored parameter: name, instruments (P/S), monitoring program (I/II), number of samples collected, percentage of sample values below the LOD (and LOD value), standard operating procedure (SOP) of laboratory.

	Parameters	Probe/sampling	I/II Program	No of samples A,B,E/P1-P4	% samples < LOD A,B,E/P1-P4 (LOD value)	SOP reference
1	Electrical conductivity (EC)	P/S	I/II	99/24	0/0	APHA standard methods for examination of water and wastewater 2510B (2021)
2	pH	P/S	I/II	99/24	0/0	APHA standard methods for examination of water and wastewater 4500-H B (2021)
3	Dissolved Oxygen (DO)	P/S	I/II	99/24	0/0	APHA standard methods for examination of water and wastewater 4500-0 G (2021)
4	Turbidity	P	I/II	99/24	0/0	APAT CNR IRSA 2110 Man 29 2003 (par. 7.2.2.A)
5	Temperature	P/S	I/II	87/18	0/0	APHA standard methods for examination of water and wastewater 2550 (2021)
6	Total Suspended Solids (TSS)	P/S	I/II	99/24	73/83 (< 10)	APAT CNR IRSA 2090B Man 29 2003
7	Ammonia (NH4)	P/S	I/II	99/24	100/100 (< 0.5)	APAT CNR IRSA 4030 B Man 29 2003
8	Nitrate (NO3)	P/S	I/II	99/24	0/0	APAT CNR IRSA 4020 Man 29 2003
9	Total nitrogen (Ntot)	S	I/II	99/24	0/0	UNI EN ISO 20236:2022
10	Total phosphorous (Ptot)	S	I/II	99/24	14/33 (< 0.01)	EPA 200.2 1994 + EPA 6020B 2014
11	Iron (Fe)	S	I/II	99/24	4/17 (< 0.001)	EPA 200.2 1994 + EPA 6020B 2014
12	Escherichia Coli	S	I/II	99/24	0/0	APAT CNR IRSA 7030 F Man 29 2003
13	Fecal streptococci & Enterococcus	S	I/II	99/24	0/0	APAT CNR IRSA 7030 F Man 29 2003
14	Metolachlor ESA	S	I/II	99/24	56/79 (< 0.01)	APAT CNR IRSA 5060 Man 29 2003
15	AMPA	S	I/II	99/24	48/42 (< 0.01)	Rapporti ISTISAN 2019/07 pag 162 Met ISS CBC 001 < 0,01rev. 00
16	Desethyl-atrazine	S	I/II	99/24	52/79 (< 0.01)	EPA 536 2007

Probe data (P) has been achieved with a frequency of 30 min at station A, B and E. Sampling (S) has been done with a frequency of: 12 h during the I monitoring program (grab sample), 1 month during the II monitoring program (mean daily sample).

The impact of the sewer system serving downtown was investigated through the analysis of pathogens concentration. These, which originate from warm-blooded animals and humans, may derive from both diffuse sources in catchments, such as livestock runoff, and point sources, like sewage networks^69,70. In this study, Escherichia coli and Enterococcus were selected.

Agricultural activity’s impact was evaluated by monitoring plant macronutrients, primarily nitrogen and phosphorus, as well as plant protection products, including selected herbicides such as Metolachlor ESA, AMPA and Desethyl-atrazine. Additional plant protection products were also monitored; however, their values were consistently below the Limit of Detection (LOD) and were therefore exluded from subsequent analysis. Given the wide variety of agricultural products available on the market and their potential threats to human and environmental health, an assessment of the most used agricultural products within the river-catchment should indeed be made. This assessment, combined with the analysis of historical data from previous monitoring campaigns when available, supports the selection of those products most likely to reach and/or impact the river system.

While previous studies^15,71,72 have identified evident relationships between heavy metals, hydrocarbons, aromatic organic solvents concentration, and the wash-out process of pollutants accumulated on urban surfaces, a similar analysis is not presented here, as a specific monitoring of rainfall events was not yet developed.

The temporal resolution of water quality monitoring was designed to capture contaminant evolution according to pollution source releases characterized by different temporal durations. Monitoring frequencies ranging from continuous to sub-daily to monthly intervals were adopted to capture both short-term and seasonal processes. While different kind of sampling programs can be adopted for sub-daily and monthly frequency, multiparametric probes only allow for continuous monitoring.

In this study, multiparametric probes (Manta + 35 by Eureka^®) were installed in A, B and E. In A and B a floating system (buoys) was employed to monitor water quality in the active portion of the river cross section (Fig. S1a of Supplementary Materials). These probes monitor seven parameters: temperature (T), pH, electrical conductivity (EC), turbidity, dissolved oxygen (DO), ammonium-nitrogen (NH₄-N), and nitrate-nitrogen (NO₃-N). Despite several checks have been developed adopting higher frequencies, probes acquire data every 30 min and the monitoring can be assumed continuous with respect to the contaminant release duration. The probes were calibrated in the laboratory using one-point or two-point calibration according to the manufacturer’s guidelines and standard solutions for pH, EC, DO, turbidity, NH₄-N, and NO₃-N.

An AS950 refrigerated sampler (HACH^®), equipped with 24 bottles, was used for automatic water sampling at user-defined frequencies. Two sampling programs (I and II in Tab. 1) were implemented to capture parameters affected by physical processes at different temporal scales. The first program aimed to assess the impact of the downtown sewer system by monitoring the evolution over time of parameters associated with wastewater discharges occurring at the sub-daily scale. Two grab samples (morning and evening) were collected, every three days for ten times, at A, B and E (60 samples in total). The program started in December 2023 and ended in February 2024. Sampling days were varied throughout the week to minimize bias. Grab samples provide instantaneous snapshots of river quality and are suitable for detecting sub-daily variations linked to domestic wastewater releases. The second monitoring program began in March 2024 and involved the collection of sequential composite samples obtained from the mixing of 24 grab samples (1/hour). Composite samples were collected monthly at sites A, B, E and bimonthly at sites P1, P2, P3, P4, to characterize overall rivers quality status. This approach captures cumulative contributions and delayed effects from urban and rural sources over time scales exceeding one day. Results from both programs were compared with data obtained from multiparametric probes to ensure the accuracy of the latter. During the first program samples were stored in the refrigerated sampler until collection the following morning (shortly after 10 AM, approximately 12 h after the first grab sample) and then transported to the laboratory. During the second program, the oldest grab sample used in the composite, was preserved for a maximum of 24 h, with collection occurring shortly after the final hour sample.

Data collected from radar sensors and multiparametric probes in A, B and E, is stored by a datalogger and transmitted with a time lag of 1 h. This allows for nearly real time check of the status of the sensors and probes and allows for prompt intervention, limiting the loss of data.

Analysis of collected data

Overview of collected data availability

In Fig. 5, data collected from the end of May 2022, when the first radar sensors for discharge monitoring were installed, are shown. Probe sensors were activated between February and June 2023, while sampling activities began in mid-December 2023. ADCP surveys have been carried out to initialize and/or improve the calibration of radar sensors. In the figure, NH₄-N and NO₃-N Ion Selective Electrode (ISE) parameters are distinguished from the other multiparametric sensors (pH, T, DO, EC, and turbidity), due to the significant difference in the performances of the former with respect to the latter. Indeed, despite many sensor calibration interventions, drift phenomenon in ISE measurements was consistent and here represented by decreasing intensity color.

Fig. 5.

Data availability for each monitoring station: discharge data (radar), multiparametric probe data (probe, ISE NH₄-N, ISE NO₃-N) and samples (sampler). ADCP surveys and probe calibrations are indicated with triangles and rhombuses, respectively.

After one year (December 2023–January 2025), 32 and 6 samples have been collected for stations A-B-E and P1-P2-P3-P4, respectively, according to the two previously described sampling programs.

Overview of water courses’ status

In Fig. 6, correlations between discharge Q and qualitative parameters (T_water, pH, EC, DO, and turbidity) recorded by the probes, as well as climatic conditions (rainfall h and air temperature T_air) recorded at the meteorological station in Treviso, are reported in Fig. 6 considering in the calculation three different time periods: June 2022 – February 2025 (panel a); winter season December 2023–March 2024 (panel b) and summer season June 2024 – September 2024 (panel c). This was done to determine the overall correlation between parameters, and to assess possible differences in correlation values due to seasonality at the three sites. Mean daily data, obtained as integral mean over the 24 h, are used for the correlation analysis. Pearson’s linear correlation coefficient has been employed for this purpose.

Fig. 6.

Correlation values between quantity and quality parameters recorded at stations (A) (left) (B) (middle) and (E) (right), using mean daily data for three time intervals: (a) June 2022 – February 2025; (b) winter season December 2023–March 2024; (c) summer season June 2024 – September 2024.

Both the Sile River and the Botteniga channel are water level-controlled rivers, meaning that variations in Q primarily result in changes in v. A correlation analysis developed considering water level (y), velocity (v) and discharge (Q) in each station, confirms that Q is primarily correlated with velocity (view Section 2y-v-Q correlation analysisof the Supplementary Materials) Material and methods. Therefore, since Q integrates all the information from the radar sensors, it was used to assess correlations between quantity and quality parameters.

A positive correlation between T_w and Q is observed at both monitoring stations along the Sile River (A and B in panel a). This relationship, however, displays seasonal variability. During summer, the correlation becomes weakly negative (− 0.14 for A and − 0.30 for B, panel b), while it is negligible in winter (panel c). The positive correlation in A and B displayed in panel a is primarily driven by the strong positive correlation during the autumn period (view Section 3Linear trendlineslope of the Supplementary Materials) Analysis of collected data, which significantly influences the overall correlation value. As previously described, the Sile River is primarily fed by springs. During the summer season, it receives additional contributions from the irrigation channels, which ultimately recharge both the aquifers and the Sile River baseflow. Indeed, after the river’s discharge reaches its annual minimum between February and March, also due to the operational maintenance of main irrigation channels requiring 15 days of closure, it increases in spring in response to precipitation events first. Then, as from April onward, water diverted from the Piave River to supply irrigation channels, contributes to aquifer recharge, increasing the Sile River baseflow. After reaching the maximum value between July and September, discharge progressively decreases as water derived for irrigation purposes gradually decreases.

In contrast to the Sile River trends, little correlation between T_w and Q is observed along the Botteniga channel (Fig. 6a). This channel is primarily fed by a nearly constant inflow from the Piavesella channel (from Nervesa water intake, see Figs. 1 and 2), which conveys about 6 m³s⁻¹ continuously diverted from the Piave River. A minor contribution (about 3 m³s⁻¹) is coming from springs.

In Fig. 6a correlation between T_w and DO is mostly evident in station E (-0.91) and disappearing in A, while a not negligible value (-0.64) is recognized in B. A marked seasonal variability appears at station A and B, where DO-T_w correlation significantly varies from winter (-0.71 for A and − 0.89 for B, in panel b) to summer (+ 0.5 for A and 0 for B, in panel c), while in E it remains always negative with a limited decrease during the winter season. As pointed out also in previous studies^73,74 decreasing T_w the DO saturation capacity increases. This suggests that DO values in station E are soundly controlled by T_w, whereas in A and B other processes occurring along the Sile River smooth the inverse relationship between the two variables.

In the Sile River DO has an inverse trend with respect to Q even if a correlation decrease can be observed during the winter, most remarkable at station B. Also, in this case, the Botteniga channel exhibits a markedly different pattern with respect to the Sile River, that somehow affects the DO-Q correlation at B, being the latter fed also by Q from E, where little to no DO-Q correlation is observed. This behavior is attributed to the hydraulic characteristics of the Botteniga channel, which is characterized by shallow water levels (generally less than 1 m). Under these conditions, DO concentrations are primarily controlled by T_w rather than Q, and this confirms the negative correlation between T_w and DO (-0.91, Fig. 6a).

The consideration developed about Q, T_w, and DO can be deepened through the analysis of Fig. 7 that shows data collected by the probes (represented by continuous lines) at location A, B and E from June 2023 to January 2025, along with discharge behaviour. Dots and triangles in the figure represent DO and T_w values, respectively, registered on the day of sampling. For each station two blowups (5 days during the winter and 5 days during the summer) are also reported. Panels a and b reveal that during autumn, both T_w and Q exhibit a concurrent decreasing trend that is not as clearly discernible during other seasons, supporting the previously developed correlation analysis.

Fig. 7.

Temporal evolution of Q measured by radar sensors and of T_w and DO measured by the probes, together with sampling values (triangles for T_w, dots for DO) for stations: (a) A, (b) B and (c) E. All time series of data on the left, a 5 days focus during the summer and the winter on the right.

For the Sile River, the behavior of DO and Q, antithetic with respect those of T_w and Q, in Fig. 7a and b, support the obtained correlation values, indicating that increases in Q are associated with decreases in DO.

Strong daily fluctuations during the summer period are manifested for both T_w and DO in all three stations (compare blow-ups of Fig. 7), with the latter increasing during the day together with T_w, suggesting photosynthetic activity and respiration cycles processes^75,76. Moreover, values of DO are rarely below 6 mg/L in all the three stations and this may be due to the presence of numerous sudden water drops related to the previously mentioned hydraulic power plants, that grant the reoxygenation of water.

A decrease in EC values during high flows (negative Q-EC correlation) has been observed in literature suggesting that this inverse relation is attributable to dilution process^77,78. In the present case study, since the river network is mainly groundwater-fed, a weak negative correlation between EC and Q appears at stations A and E, while no correlation is observed at B. Moreover, while for station A a fairly constant negative correlation value is always observed, in station B and E seasonal variations occur. This may reflect the flashy contributions of surface runoff in urban areas (Treviso downtown and urban areas north to Treviso along the Botteniga channel). The amount of rainfall events and their intensity vary throughout the seasons, and those characterized by less relevant events may result in low or negligible correlation between Q and EC. Moreover, the effect of dilution due to contributions from rainfall events may be less marked in B and E since pollutants washed off from urban areas may in turn increase EC⁷⁹. The fact that during rainfall events EC decreases because of dilution seems to be confirmed by the negative correlation of EC with turbidity for station A and E for all seasons. During rainfall events Q increases allowing for dilution, at the same time sediments are resuspended due to turbulence causing turbidity increase.

For station A, since there is a negative correlation between DO and Q being the correlation between EC and Q negative, even though negligible, a positive correlation between EC and DO is observed. For station B and E, although a negative correlation between DO and Q is observed, since EC-Q correlation values vary across seasons according to the amount and intensity of rainfall events, the correlation between EC and DO at B and E can be appreciated only for the season characterized by numerous or intense rainfall events.

Figure S4 in the Supplementary Materials shows correlation values between Q and T_w (panel a), Q and DO (panel b), Q and EC (panel c), DO and EC (panel d) for stations A, B and E are reported together with total rainfall (mm) and maximum hourly rainfall intensity (mm/h) for each month.

Diel variation in rivers of DO-pH and T_w-pH also emerges in this case study. A positive, for DO, and negative, for T_w, correlation with pH is observed in E, as observed also in other studies⁸⁰, whereas no correlation was observed in A and B. The difference between minimum and maximum values of pH is very limited for all three stations, and this is coherent with the high alkalinity levels (> 300 mg/L, measured by ARPAV between 2010 and 2023) of both the Sile River and Botteniga channel, indicating that both rivers are able to maintain a fairly stable pH level.

Impact of urban discharges and agricultural activities

In the downtown area of Treviso numerous septic and Imhoff tanks intercept wastewater before discharging the effluent to the nearby water courses. Wastewater primarily contains pathogens, phosphorous and nitrogen, the presence of the latter being revealed by ammonium (NH₄) and nitrate (NO₃) levels as reported in other previous studies^81,82. The NH₄-N data from probes at all stations (A, B and E) exhibited drift phenomenon, and this issue could not be resolved through sample analysis, as the returned values are consistently below the laboratory detection limit (0.5 mg/L). Therefore, to assess the impact of the CSS and to differentiate this impact from potential contributions from agricultural areas, the temporal evolution of Escherichia Coli (E. coli), Enterococcus, P_tot and NO₃-N, among the parameters monitored, was analyzed.

Figure 8 presents E. coli and Enterococcus concentrations measured at each sampling position (A, B, E, and the mobile P1, P2, P3, P4) along with hourly rainfall data h from a regional meteorological station (panel a) and c). The monitored period encompasses both the first sampling program (characterized by 2 samples every 3 days) and second sampling program (1 sample every month). Panels b) and d) provide a detailed view of the first program covering the period December 15th, 2023 to February 1st, 2024, for both parameters.

Fig. 8.

Escherichia coli and Enterococcus concentration measured between: (a) and (c) December 2023 and February 2025 in all stations; (b) and (d) December 2023 – February 2024 (first sampling period) in stations A, B and E. Horizontal dashed lines represent the bathing limit for both parameters and discharge limit for E. coli.

At stations B and E, E. coli concentrations predominantly exceeded 1’000 UFC/100mL (Colonis Forming Units per 100mL), the bathing limit per Italian Regulation⁸³, while at station A values were mostly below this limit. Although bathing in the Sile River should be allowed only upstream of downtown Treviso, the river sees recreational use also downstream, especially rowing. In some instances, concentrations at B and E approached or exceeded 5’000 UFC/100mL, the WWTP discharge limit according to the Italian Regulation⁸⁴. The mean of concentration values measured during sampling programs at B and E, are 3 and 5 times higher than A, respectively, and this highlights downtown Treviso’s significant impact on the Sile River, as well as the influence of urban areas located north along the Botteniga channel. It is important to note that measurements at B reflect both contributions, combining concentrations from downtown with those from the Botteniga channel, along with the negligible input from station A located upstream downtown along the Sile River.

Despite the limited number of samples collected along the small channels crossing downtown (stations P1, P2, P3, P4), concentration values are notably high in P2 and P3. These stations proved to be the most affected by discharges from the sewer system, with concentration values frequently exceeding 5’000 UFC/100mL. This is likely due to the limited discharge of these channels compared to the wastewater outflowing from both sides. In P4, where the channel appears to have sufficient discharge for dilution and in P1, outside the historical center, concentration remains consistently low. As expected, concentrations in small channels exceeded those at A, B and E due to the lack in dilution processes.

During the first sampling program (Fig. 8b), E. coli concentrations were deduced from grab samples. The short time interval (12 h) between subsequent collections allows for a more in-depth analysis of the physical processes occurring and discharge peaks occurring at A, B and E, between the 17th and 18th of January, can be recognized as a consequence of a low-intensity rainfall event (about 4 mm/h).

By assuming a limited lifetime of pathogens in surface water (around 2 days^85–87, the peak of E. coli concentration at E on January 21st is unlikely related to the rainfall event, but maybe due to maintenance actions on the network draining to the Botteniga channel. Similar considerations can be developed for the event recorded on May 18th (panel a). Both events show higher concentrations at E than B, indicating that the contribution mainly comes from Botteniga, which is then diluted in the Sile River as confirmed by the negligible concentrations at A. The response in terms of E. coli concentration to January 18th event can be appreciated from the value measured in E. Instead, the peak in concentration observed at B on the 18th of January at 6 am can be assumed as consequence of a washout process following the rainfall event. Concentration rapidly decreased until January 21st, revealing residual discharges from downtown. The subsequent increase on January 22nd is likely due to the Botteniga channel’s contribution unrelated to the rainfall event as previously discussed.

Although it is not easy to recognize it from Fig. 8b, concentrations measured in the morning (9–10 AM) do not always prevail on the ones measured in the evening (9–10 PM), and this reflects the typical behavior of drinking water consumption of Treviso area. Being unavailable the hourly consumption of the downtown, in Fig. 9 the temporal pattern of water consumption of nearby areas similar in terms of inhabitants, urban organization and activities, is shown. In both cases, the pattern is characterized by two peaks, one around 8 AM and one around 8 PM.

Fig. 9.

Temporal pattern of water consumption of towns near Treviso, similar in terms of inhabitants, urban organization and activities to downtown Treviso. E. coli concentrations for A (triangles), B (circles) and E (squares) obtained during the first sampling program are also shown.

The sampling time was initially chosen under the hypothesis that the collection of samples between 9 and 10 PM was done during low water consumption, to grab a sample when lower releases from the sewer system occur. This is not always true as manifest from Fig. 9, and sampling values agree with this behavior. Moreover, the results may be at some extent jumbled due to the complex downtown sewer system previously described. Obviously, by assuming a different sampling frequency, e.g., every 3–4 h, a more detailed temporal behavior could have been identified, but the potential information gain must be anyway carefully evaluated against increased costs.

According to literature⁸⁸, the presence of E. coli suggests recent pollution phenomenon, while Enterococcus presence is related to older discharges, since the latter is considered to be more resistant to external pressures with respect to the former. While there is a direct correspondence between the of E. coli concentrations and sewer releases both characterized by a high variability, due to the greater lifetime of Enterococcus this effect is smoothed. In these monitoring campaigns, Enterococcus concentrations were mostly below the bathing limit (500 UFC/100mL) according to the Italian Regulation. High values were recorded at B, E and P3 shortly after the rainfall event on May 18th, with values at B and E increasing up to 4 and 6 times, respectively, compared to the pre-event value. It should be noted that the same increase was not observed during subsequent rainfall events that are less relevant in terms of intensity (82 mm in 31 h for May 18th vs., e.g., 67 mm in 70 h for June 13th ). Other studies present in literature highlighted concentrations downstream urban areas comparable⁸⁹ or even higher³ to the ones monitored in this study.

As previously described, the monitored area is not only impacted by the sewer system but also by agricultural activities. During the first monitoring program, the impact of downtown on the Sile River water quality can also be observed by comparing P_tot values measured at B with the ones measured at A and E. Even if the presence of P_tot and NO₃-N in rivers and channels may be related to contributions from agricultural activities⁹⁰, that is from the northern and western areas, the limited concentration of P_tot values measured in A and E seem to exclude this circumstance (Fig. 10).

Fig. 10.

P_tot concentration measured between: (a) December 2023 and February 2025 in all stations together with rainfall data; (b) December 2023 – February 2024 (first sampling period) in A, B and E together with discharge data recorded at each of the three stations.

During the second monitoring program, P_tot concentration values measured at B are comparable to the ones at A and E. The first monitoring program reveals concentration of a spot sample while each concentration value obtained during the second monitoring program is deduced from a mix of 24 spot samples regularly collected over the day. As previously observed, the sampling time during the first monitoring program corresponds to peaks in water consumption, thus to sewer system releases. Often, concentration values measured at B during the first monitoring program are greater in the evening than in the morning and are higher than concentration values obtained during the second monitoring program (Fig. 10b), mean over one day that manifest contributions from downtown (at B) like those recorded at A and E.

Following the reasoning developed for P_tot concentrations, the NO₃-N contribution from the agricultural areas prevails over the downtown sewer system one. The link between agricultural activity and NO₃-N concentration is confirmed by the higher concentration measured at A (5.5–6 mg/L), where the watershed extension is 235 km² with an estimated percentage of agricultural areas of 65%, compared to 2–2.5 mg/L at E, where the watershed extension is 145 km² and 55% the estimated percentage of agricultural areas (Fig. 1). Concentration at B is the consequence of the mixing of water coming from A and E. Due to the difference in the discharge (about twice the value measured at A with respect to E, as shown in Fig. 11b), concentration at B lies between A and E values, but is closer to the one measured at A. No significant trends related to seasonality, mainly related to agricultural activities, can be observed from the second program’s data. The sampling technique does not affect NO₃-N concentration values measured over the entire monitoring period, as shown in Fig. 11a.

Fig. 11.

NO₃-N concentration from sample analysis between: (a) December 2023 and February 2025 in all stations together with rainfall data; (b) June 2024 – November 2024 in A, B and E together with Q, h and continuous water quality data recorded at each of the three stations.

Advantages from continuous measurements are manifest from the analysis of NO₃-N behaviour between June and November 2024 (Fig. 11b). Although NO₃-N data from ISE probes is often uncertain, in this time interval they are substantially confirmed by the lab analysis of samples in all three stations. From continuous measurement, an exceedance of the limit for river water quality classification can be recognized, which instead can be unrevealable if discrete measurements are used. This is the case of, e.g., discrete concentration values measured at B, always below 4.8 mg/L, with respect to continuous data that show clearly the exceedance of this limit in between sampling when a rainfall event and/or a sudden discharge increase takes place. Again, the good agreement between measurements developed by ISE probes and sample analysis results grant, in this case, the overall reliability of the continuous data.

The correspondence of few rainfall events with NO₃-N peaks (e.g. September 2024) may suggest a link related to washout processes. To explore this possibility, a correlation analysis was developed between NO₃-N measured in B with NO₃-N measured in A, NO₃-N measured in E, Q in B and rainfall h. Being the acquisition frequency different for the variables involved (30 min and 1 h depending on parameters and measurement technique), the correlation analysis was developed setting the lowest frequency as time lag.

In Fig. 12 the results of the analysis are reported considering all the data collected between June-September 2024 (panel a) and a subset of data collected during the rainfall event of June 25–27 (panel b) and the rainfall event of September 5–8 (panel c). This to understand if the overall trend is confirmed also during a single event, or, in other words, to check if the link between parameters during a single rainfall event is shadowed when a long time series of data is considered.

Fig. 12.

Correlation among instantaneous values: (a) using all available data; (b) using data acquired during the rainfall event occurred between 25–27 July 2024; (c) using data acquired during the rainfall event occurred between and 18–20 September 2024.

From all plots (panel a to c), no correlation is manifest between NO₃-N measured in B with rainfall intensity recorded near downtown, while an evident link with NO₃-N at A and Q at B clearly emerges. It is also evident the positive correlation between nitrates measured in A with the ones measured in E. Although these two stations show the contributions of two different watersheds, the positive correlation suggest a common source related to agricultural activities that seems to characterize the presence of NO₃-N in the watercourses crossing downtown. The analysis developed on the available data indicates that the presence of NO₃-N is mainly related to agricultural activities and not to sewer system contributions. At the same time, despite the appearance in Fig. 11, the analysis developed on the available data denies the correlation of NO₃-N at all three stations with rainfall events. A positive correlation between rainfall and discharge may or may not appear using data collected during specific rainfall events (panels b and c), resulting in a weak positive correlation when all available data is used (panel a). The rainfall-runoff process in the lowland spring Sile river is cumbersome thus the response may differ from one event to another, especially when the latter is not relevant. To investigate the effect of the time delay due to the movement of flow and contaminants in the subsurface, the correlation analysis has been repeated considering a shift of 6, 12 and 18 h between rainfall (h) and the other parameters (NO₃-N at B, NO₃-N at A, NO₃-N at E, Q at B, ). Results have been reported in Fig. S5 of the Supplementary Materials.

A stepwise multi-linear regression analysis^91,92 was used to generate predictive relationships between the dependent variable ( ) and the independent ones (; ; ; ). The results, reported in Tab. S2 of Supplementary Materials, shows that NO₃-N in B are primarily dependent on NO₃-N in A being however the dependence on Q in B not negligible. Adding NO₃-N measured at E and h as independent variable does not improve at all the linear model. Collinearity between independent variables has been checked through the variance inflation factor which resulted to be always lower than 2.

NO₃-N concentrations measured along three small downtown channels (at P2, P3 and P4) were mostly below 2.4 mg/L, similar to values measured at E, confirming the fact that the presence of NO₃-N is not related to downtown sewer system discharges. Concentration values recorded along the small channel on the west (at P1) are close to the ones monitored at A, suggesting that this channel receives contributions from the agricultural areas outside downtown.

To investigate potential seasonal trends and develop a mass balance in the downtown area, the temporal behaviour of Escherichia coli, Enterococci, P_tot and NO₃-N loads was computed. The load was calculated as the product of the mean monthly discharge and mean monthly concentration. The former was computed as the mean value of continuous data from radar sensors, while the latter was assumed to be represented by the mean daily concentration of one sample per month. Formally speaking, daily fluctuations of concentrations and discharges may originate a contribution that increases the previously defined load. Since only one mean daily concentration is available for each month, it is not possible to quantify this contribution. However, the daily fluctuations of discharge at stations A, B, and E over each month are negligible, being characterized by a mean coefficient of variation of about 8% for all stations. Moreover, daily variations of concentrations, evaluated using data from the first monitoring program, are finite and comparable to the monthly variations. This leads to a reasonable assumption that the monthly expected value of the product of daily concentration and discharge fluctuations is negligible compared to the product of the mean daily values. This allows defining the balance between the load entering and exiting downtown Treviso as:

being the contribute of downtown if positive. When approaches zero or becomes negative, also considering uncertainties in monthly load estimates or unaccounted travel times associated to grab samples, the downtown contribution is negligible. In Fig. 13 (panels from a to d), this contribution has been assessed for all concentration values obtained from both monitoring programs: using grab samples and instantaneous discharge measurements for the first program, and monthly values, as previously defined, for the second monitoring program.

Fig. 13.

Temporal behaviour of loads (dashed line and triangles) during the first (left) and second (right) monitoring program for: (a) Escherichia coli, (b) Enterococci, (c) P_tot and (d) NO₃-N. In each plot, continuous lines represent discharge values recorded at station A, B and E.

The same diagram also displays the temporal pattern of discharge. A seasonal trend is evident only for the mean monthly load in panel a (E. coli), which appears to follow the seasonal rainfall pattern shown in Fig. 3. In this case, it is important to highlight that the contribution from downtown is consistently positive in both the monthly and daily analyses, suggesting that the main source of E. coli at station B is linked to discharges from this area. Conversely, the monthly analyses shown in panels b, c, and d indicate loads close to zero, suggesting that downtown does not contribute significantly to Enterococci, P_tot, and NO₃-N at station B. However, the hourly analysis developed during the first monitoring program, shows for P_tot (panel c) high load values, indicating that mean daily and grab samples can lead to different concentration estimates and, consequently, different load values. These discrepancies between grab and mean daily samples are observed only for P_tot, while for the other parameters the correspondence between the two sampling approaches is generally good.

To assess the dominant pollution sources at each station, Principal Components Analysis (PCA), a method used for analysing co-variation among chemical variables and examine relationships between monitored parameters, was used. PCA was applied to data collected during both monitoring campaigns, including grab samples (identified as A1, B1, E1) and mean daily samples (identified as A2, B2, E2). The PCA was performed first using only samples were parameters listed in Table 1 reveal concentrations above the LOD. Thus, to ensure statistical robustness, only parameters (9 out of 15) with a limited proportion of censored values were retained for analysis, considering 85 out of the total 99 samples in the PCA. From the analysis, the primary (PC1), secondary (PC2) and tertiary (PC3) components explain 67% of the data variability. In Fig. 14a the biplot of the primary (PC1) and secondary (PC2) components (explaining more than 51% of the data variability) is reported. Turbidity, Fe, and P_tot emerged as the main contributors to the primary component PC1 (loadings 0.44, 0.45, 0.44 respectively). This finding suggests that both Fe and P_tot are predominantly associated with suspended solids which are mobilized through urban wash-off and resuspension processes, as reported in literature⁹³. N_tot and NO₃ resulted to be the main contributors to the secondary component PC2 (loadings 0.47 and 0.46 respectively), whereas DO and Enterococci were the main ones for the tertiary component PC3 (loadings 0.67 and 0.41 respectively). In the same figure it is possible to notice that samples from stations A and E cluster closely, whereas those from station B exhibit greater dispersion. Furthermore, grab samples at stations A and E (A1, E1) tend to form tighter clusters compared to mean daily samples (A2, E2), while the opposite pattern is observed at station B, where mean samples (B2) are mostly placed around the centre, while grab samples (B1) are scattered. The more the samples are clustered, the more their chemical composition is similar. This suggests that daily composite sampling at station B may mask short-term variability in contaminant concentrations, likely driven by dynamic urban activities (see Fig. 8).

Fig. 14.

Principal Component Analysis considering: (a) only parameters (9) in which no or few samples values were below the LOD, samples below the LOD were not considered; (b) all monitored parameters (12), values below the LOD were substitute with a mean value between 0 and the LOD.

Samples collected in A gather near the N_tot and NO₃ arrows, suggesting that this station is mostly affected by agricultural discharges, even though the P_tot contribute is less relevant as suggested by the versus of the corresponding arrow. This result agrees with considerations developed on the analysis of temporal behaviour of P_tot and NO₃-N concentrations reported in Figs. 10 and 11, respectively.

Samples of station B are mainly spread along the turbidity, Fe, and P_tot arrows. Resuspension processes related to powerplants, and control sections drops in Treviso, and the limited contribute of both Fe and P_tot at station A that reveals the external contributes, agree with the hypothesis that this station is mostly affected by urban discharges.

The Enterococci and E. coli role is difficult to establish from the PCA results, being the corresponding arrows located between samples collected at B and E. Even though at station E the contribute of pH is manifest, contributes coming from Enterococci and E. coli are not negligible, and this confirm as pollution from sewer release are coming also from the area north to downtown, that collect not only agricultural but also peri urban areas. In any case, the PCA analysis seems to exclude the occurrence of sewer pollution in the samples collected at A.

To better distinguish between urban and agricultural contributions, a second PCA was performed by incorporating Metolachlor ESA, AMPA, and Desethylatrazine data, even though half of the concentration values for each compound were below the LOD (0.01 µg/L). Excluding all samples where any of the compounds was below the LOD would have resulted in a dataset too limited to ensure robust PCA outcomes (about 40 out of 99). To overcome this issue, values below the LOD were imputed arbitrarily setting the concentration to the mean between 0 and the LOD for each parameter, that is to the most probable value.

Figure 14b presents the biplot of the first two components (PC1 and PC2). The inclusion of the three plant protection products modifies the definition of PC1, which is now primarily influenced by parameters associated with agricultural activities (N_tot, NO₃, Metolachlor ESA, and Desethylatrazine). Samples from station A remain clustered near these variables along PC1, whereas samples from station B continue to align predominantly with turbidity, Fe, and P_tot. In both the analyses (panel a and b), the collocation of A and B samples consistently indicates that P_tot is more indicative of urban signatures than agricultural ones.

Discussion

Insights on urban and rural contributions

After two years of experience of the new monitoring system applied to investigate water quality of the river network crossing the city of Treviso, the following summary thoughts can be made. The system has been developed with the main aim of distinguishing between contributions coming from the sewer system of downtown from the ones coming outside. In three positions, grab samples were integrated with continuous monitoring from multiparametric probes, and in the same and other three more positions, radar sensors allow monitoring of water level and velocity. The latter values, validated through sporadic on-site measurements (ADCP surveys), grant a continuous overview of discharge values flowing in the river network.

Two approaches were applied for water quality monitoring. Continuous monitoring sensors provide the evolution of investigated parameters but are generally limited to recognizing only a few of them and can sometimes suffer from drift phenomena. Laboratory analyses of samples potentially allow monitoring of any physico-chemical parameter, yet the associated cost limits sampling frequency. Coupling these two techniques enables controlling for potential drift and failures affecting continuous monitoring signals through laboratory sample analysis and retaining temporal behavior between samplings. The two monitoring programs, implemented at different spatial and temporal scales, reveal physical processes that cannot be otherwise identified, highlighting the relevance of considering both sub daily and monthly frequencies.

Probes performance was not always reliable. For ion‑selective electrodes (ISEs), despite careful installation and regular calibrations, NH₄‑N sensors exhibited drift at all stations, while NO₃‑N electrodes frequently failed, resulting in only six months of reliable NO₃‑N data over a two‑year period. These limitations raise concerns about the practical benefits of their deployment. In contrast, continuous measurements of T_w, pH, DO, EC and turbidity, were consistently accurate as confirmed by discrete sampling. These data were used to develop a correlation analysis based on daily average values along summer season, winter season and considering all available data (June 2022 – February 2025), so that differences in hydraulic regimes between Sile River and Botteniga channel could be witnessed. Along the Sile River, Q-T_w correlation is strongly season-dependent. The correlation during the summer is weakly negative and negligible in winter, even though the overall correlation is positive. This is due to the strong positive correlation manifest in the autumn season, when both parameters decrease simultaneously. These patterns reflect the river’s spring-fed regime and the seasonal inputs it receives from irrigation channels, raising the discharge through summer before declining as irrigation withdrawals reduce.

In contrast, for the Botteniga discharge, which is mainly fed by constant water from the Piavesella channel, minimal Q-T_w correlation is observed. The hydraulic characteristics of the Botteniga channel at station E also affects the correlation values between T_w and DO which is persistently negative, indicating that DO is strongly affected by T_w. In contrast, along the Sile River, correlation between the two parameters varies seasonally, becoming positive in station A (upstream downtown) and uncorrelated in B during the summer, while being negative at both stations during the winter. This suggests additional biogeochemical (e.g. photosynthesis and respiration) and hydrological processes, masking the expected inverse correlation between T_w and DO. Moreover, the effect of the presence of hydropower plants and hydraulic structures along both rivers, which grant reoxygenation of water, is evident as DO values never reduce below than 7–8 mg/L at stations A, B and E.

Even though the domestic contribution from the downtown area of Treviso has been estimated in 20–30 Ls¯¹ (as mean annual value), and this value is negligible with respect to the Sile (30 m³s¯¹ at B) discharge, the impact of the downtown CSS is evidenced by the presence of E. coli, Enterococcus and P_tot. Laboratory analyses of grab samples collected twice every three days during the first monitoring program show concentration patterns consistent with daily water‑consumption cycles, with pronounced peaks corresponding to periods of increased use. This behaviour was not captured during the second monitoring program, in which daily composite samples were obtained by mixing 24 hourly grab samples. The peaks of the E. coli in watercourses that are unrevealed when considering mean values, suggests that their origin is related to urban wastewater system and thus can be recognized only by instantaneous sampling. This interpretation is further supported by spatial differences in concentration data. Higher values were consistently higher at station B (downstream downtown) compared to the ones at station A (upstream), while concentrations at E were comparable to those at B, albeit lower. Given that discharges at E are one third of that at B, the observed concentrations at this latter station cannot be explained without accounting for downtown inputs. Elevated E. coli values at E also suggest that urban areas north to the city are responsible of the Sile River contamination, whereas low and rainfall-independent values measured at A indicate a limited contribution from areas located west do downtown. Load balance for E. coli leads always positive values, further supporting the previous considerations. Within the downtown area, relevant concentration of E. coli were consistently observed for samples collected at P2, P3, and occasionally at P4, with some differences that may suggest a prioritization of possible interventions on the sewer system network. Limit discharges into the small channels fed by the downtown sewer system, which appears to be their principal contribution, justify the higher concentrations respect to the ones measured at E. In contrast, low concentrations recorded at P1, similar to the ones recorded at A, suggest a limited contribution of the sewer system on that watercourse.

The collected data supports the need for intervention, as E. coli concentrations downstream downtown (at B) are 3 to 5 times higher than the ones measured upstream (at A), frequently exceeding the bathing-water quality limits. Given the recreational use of the Sile River, mainly rowing, mitigation measures are required both within downtown and upstream along the Botteniga channel to reduce untreated effluent discharges.

Contributions related to activities developed outside downtown, mainly agricultural, were recognized through NO₃-N analysis. Although in literature NO₃-N concentration in rivers has been associated with sewer system contributions, in this case study this link was not revealed. As discussed previously, NO₃-N levels at B result from the mixing of contributions from A and E. Both A and P1 are located on watercourses that are predominantly groundwater-fed, whereas additional inputs are relevant for the Botteniga channel at E. The absence of E. Coli, Enterococcus and P_tot at A and P1 further indicated that NO₃-N concentrations in this case study are mainly associated with agricultural sources instead of wastewater discharges. Increases of NO₃-N concentrations in watercourses during and after relevant rainfall events are attributed to groundwater table rising that leads to the washing of NO₃-N from surficial soil layers. Continuous NO₃-N data collected between June and September 2024 is consistent with values from sample analysis. Correlation data between NO₃-N at B, NO₃-N at A, NO₃-N at E, Q at B and h measured near downtown, shows that the primary contribution of NO₃-N at B derives from the watershed that drains at station A. This is confirmed by the weak correlation observed with both NO₃-N at E and h, which in turn suggests that concentrations measured downstream Treviso mainly originate from agricultural areas in the north-western watershed, with limited contribution from rainfall-driven washout processes. These findings are corroborated by the PCA results, in which parameters commonly associated with agricultural activities (NO₃-N, N_tot, and plant protection products) explain the variability of the second component, with data collected in station A clustered near the arrows for these parameters. The concentration of nitrates measured at P2, P3, P4, consistent with the ones measured upstream downtown at E, confirms that the contribution in NO₃-N of downtown is negligible. Moreover, PCA shows that P_tot and turbidity explain the variability of the first component, suggesting that P_tot concentration is mainly driven by resuspension processes, which appear to be most pronounced at station B, downstream Treviso.

A summary table outlining the contribution of each monitoring element towards separating urban versus agricultural signatures has been provided in the Supplementary Materials (see Tab. S3).

Monitoring requirements and operational considerations

From the results it emerges that high-frequency sampling is essential to track the highly variable temporal patterns of E. coli concentrations associated with sewer system discharges, as well as of P_tot concentrations influenced by resuspension processes. In contrast, comparison of the two sampling programs suggests that the temporal behaviour of Enterococci and NO₃-N can be captured using mean daily sampling conducted once per month. This is reasonable considering that literature indicates a longer persistence of Enterococci compared to E. coli one, and that, being the Sile River spring-fed, NO₃-N concentrations are primarily controlled by groundwater fluxes and agricultural inputs. A repetition of high-frequency sampling should however be carried out also during the summer season, when the mean daily loads decrease (Fig. 13a) as downtown population diminishes due to vacations.

It is important to note that neither sampling programs account for contributions associated with heavy rainfall events, which would require a dedicated high-frequency approach able to capture concentration dynamics throughout the runoff event. Multiparametric probes can follow the dynamics of a flood event, but the retrieved information are limited to few macro-parameters. The automatic samplers can monitor any parameter by collecting grab samples with a minimum time lag of 3 min, however the total monitoring duration is limited by the number of bottles available (24). When all the bottles are filled, sampling must be suspended to allow collection and substitution with a new set of empty bottles. The operational sequence of grab sampling to capture the pollution dynamics affecting the investigated area during heavy rainfall events should be defined based on the typical heavy rainfall events and the estimated time delay in the hydrologic response of the catchment.

A minimum number of monitoring positions, where both quantity and quality data are achieved, must be installed to grant the mass balance for the investigated area. More discharge measurements allow the development of a hydraulic model of the river network from which it is possible to derive information about discharge at ungauged locations. When a calibrated model is defined, so that the discharge in all sections of the river network is known, the load can be assessed also where unsystematic grab sampling is done.

The same system can be applied to other case studies, provided that specific installation requirements must be met. For sites in which grid power is unavailable, battery-supplied solutions with solar power recharge, can be employed; however environmental conditions (e.g. temperature and bad weather duration) could limit their functionality and increase maintenance interventions, which in turn increases the related costs. To limit the energy consumption, an adaptive time frequency monitoring as well as the use of a memory buffer to limit radio connection to the server, are advisable. If due to the installation requirements, both water level and velocity radar sensors cannot be installed in a prescribed river cross section, the nearest available section can be monitored if the presence of external inflows can be excluded, otherwise these must be quantified. Ultimately, if radar monitoring station cannot be installed for both site and economic constraints, discrete measurements can be done (e.g. by ADCP) to define a stage-discharge relationship. However, a water level monitoring sensor is needed. When the measurement continuity cannot be granted (e.g. spot measurements), the monitoring frequency is to be defined according to the hydrological regime of the river and the purposes for collecting the data.

As in the presented case study, if a structure along the river section, useful to support the multiparametric probe for monitoring in the active portion of a river, is not available at a prescribed cross section, a buoy can be employed. However, in the latter case both installation and maintenance costs increase. As for quantitative data, discrete measurements frequency is to be defined according to the purpose of data collection. If the sensor cannot be installed in the active portion of the river but only at the bankside, a preliminary analysis assessing differences between the two sampling positions should be done.

The sampler position is limited by the so-called vertical lift that is the elevation with respect to the minimum surface water level plus the head losses in the intake tubing. Consequently, sampling in the active portion cannot always be performed at any desired section. If the installation requirements are not met, the nearest section that satisfies the vertical lift constraint must be selected, while ensuring that external inflows along the reach are avoided.

Refrigerated samplers require a grid-power supply. Alternatively, non-refrigerated transportable samplers can be used, though the vertical lift limitation still applies. Moreover, samples must be collected promptly to prevent degradation.

In the case study here presented, samplings and probe continuous monitoring have been developed at the same position, together with discharge measurements. This allowed both to check the reliability of probe data, to acquire the temporal evolution of water quality data between samples, and to develop load balances. If this optimal condition cannot be achieved, different locations for quality and quantity measurements must be chosen according to the considerations discussed previously.

Conclusion

This study investigated the effectiveness of a monitoring system in tackling issues usually affecting urban rivers, and the possibility of distinguishing between pollution sources considering different parameters each characterized by a proper spatial and temporal scale. The Sile River flowing through the city of Treviso was considered as case study. After designing a quali-quantitative monitoring network, a two-year monitoring program was developed to acknowledge the pollutants driving the river quality status at sub-daily and monthly timescales. Through correlation, principal component analysis, regression analysis and load balances, it was possible to identify different pollution sources and assess which parameters are most indicative of each source. The analysis showed that E. coli, Enterococci, and to some extent also P_tot, are indicative parameters of the impact of urban area, with high concentrations of the latter linked to turbidity and thus to resuspension processes. Agricultural contributions were revealed by NO₃-N and plant protection products.

Data collected at different temporal scales (daily grab sampling vs. monthly composite sampling) showed that only through instantaneous sampling, developed according to the temporal evolution of water consumption, can the actual impact of the sewer system serving the city be detected. In contrast, a coarse monitoring frequency, such as monthly composite samples, can adequately capture seasonal trends of E. coli. However, this consideration emerged only when concentration values were coupled with discharge data to develop a load balance, thanks to the strategic positioning of monitoring stations (upstream and downstream downtown). Indeed, such seasonal trends could not be identified from concentration data alone, thus highlighting the importance of integrating water quality monitoring programs with discharge monitoring. The strategic positioning of monitoring stations, also allowed for distinguishing between internal and external contributions to the urban area. Despite some instrumentation issues, continuous NO₃-N data allowed for the development of correlation and regression analyses, which highlighted that agricultural activities are the main source of this pollutant and identified the area that contributes most to NO₃-N loads in the Sile River.

From the experience gained, it emerges that the spatio-temporal scale of the monitoring system, the integration between continuous measurements and sampling, and the sampling methodology itself (grab or composite), must be commensurate with the area targeted for intervention.

An analysis of the features and requirements of the monitoring system has been developed to explore the applicability of similar approaches to other urban contexts where precise identification of the main pollutants and their sources is required to effectively guide and plan remediation actions on urban rivers. This is particularly relevant for many cities that seek to restore urban watercourses for recreational purposes.

Electronic Supplementary Material

Below is the link to the electronic supplementary material.

Abbreviations

CSS

Combined sewer system

CSO

Combined sewer overflow

HF,LF

High-frequency, low-frequency

WWTP

Wastewater treatment plant

ARPAV

Agenzia regionale protezione ambiente veneto

WFD

Water framework directive

WMP

Water monitoring program

ADCP

Acoustic Doppler current profiler

ISE

Ion selective electrodes

Q

discharge

y

water level

v

water velocity

T_w

water temperature

pH

potential of hydrogen

EC

Electrical conductivity

DO

Dissolved oxygen

Turb

Turbidity

h

rainfall height

T_A

air temperature

UFC

Colony-forming units

Author contributions

Conceptualization , Giulia Mazzarotto and Paolo Salandin; methodology , Giulia Mazzarotto; writing—original draft preparation , Giulia Mazzarotto; writing—review and editing , Giulia Mazzarotto and Paolo Salandin; supervision , Paolo Salandin. All authors have read and agreed to the published version of the manuscript.

Funding

Funding for this research was provided by the water utility company Alto Trevigiano Servizi Spa through the 2019 project “Studio dell’interazione fra scarichi fognari e corpi idrici ricettori nella città di Treviso (Study of the Interaction between Sewage Discharges and Receiving Water Bodies in the City of Treviso)” and the 2021–2023 PhD Research Project “Impact of Urban Drainage and Sewerage Systems on the Quality of Water Bodies and Mitigation Strategies”.

Data availability

The use of data and information reported in this study is limited to the scientific support of the present manuscript and cannot be reproduced without the formal approval of the authors and the water company (ATS Spa) managers. A direct request for these materials may be made to the authors or to the provider as indicated in the Acknowledgments.

Declarations

Competing interests

The authors declare no competing interests.