Temporal changes in metal concentrations in a polluted urban basin: the contribution of multivariate techniques to the evaluation of their recovery

In urban rivers in many fast-growing cities, heavy metals pollution is one of the major quality issues. Quantifying its recovery success seems to be a dicult but necessary challenge, in part, because a complex database is generally needed to assess water quality. The overall goal of this work was to evaluate temporal changes in metal concentrations in a polluted basin applying the AEM method and to determine the physicochemical and meteorological parameters associated with these changes using the RDA. We analyzed temporal changes of 15 sites located in the Matanza Riachuelo River Basin, one of the most polluted basins in the word. For this, we collected data of metal concentrations in surface water, physicochemical parameters of water and meteorological factors of eight years (2008 to 2015) in each site. The results of this work allowed make evident temporal patterns (at different scales) in metal concentrations and several factors associated with these patterns. Also, we found that the effects of physicochemical and meteorological factors in metal concentrations were dependent of each site, possibly related to the presence of different sources of pollution or characteristics of the river. Our study showed that the combination of AEM and RDA multivariate techniques is a useful tool for both detecting temporal trends in the concentrations of environmental pollutants, which are not easily detectable in highly degraded environments, and for evaluating factors associated with these changes. These techniques could be applied to different scenarios (aquatic and terrestrial) affected by the continuous advance of human activities.


Introduction
Urbanization is a global process that has great negative impacts on the hydrology and water quality of watersheds. It is considered that today there is not almost a complete natural river in the world (Wang et al. 2012). In urban rivers in many fast-growing cities heavy metals contamination is one of the major quality issues because of the practice of discharging of untreated domestic and industries wastes (Khadse et al. 2008; Reza and Singh 2010;Venugopal et al. 2009). The impacts of metals in the environment are severe and long term because they are di cult to eliminate (Ding et  Rivers and streams of fresh water are important for food, services and recreation, and this is why their recovery has become a world-wide requirement (Chen 2017;Palmer et al. 2005). However, quantifying recovery success seems to be a di cult but necessary challenge. A study carried out by Jones and Schmitz (2009) in which the recovery of 240 aquatic systems was analyzed, showed that in 54% of the sites the monitoring program have not been conducted over a long time enough to detect recovery and in the 5% of them the ecosystems were irreversibly entrained into alternative states, thereby precluding recovery. Also, a complex database is generally needed for assessment of water quality as it is common the measurement of several parameters, taken from different monitoring stations and at different times (Chapman 1992; Wunderlin et al. 2001). Moreover, it is di cult to distinguish treatment effects from seasonal variation, episodic events and long-term climatic changes (Clements et al. 2010). Thus, to make a proper assessment of the success of recovery actions is essential determine which parameters are the most important to explain these temporal variation (Wunderlin et al. 2001).
Temporal changes in water quality parameters is commonly analyzed used univariate analyses, as multiple linear regression to evaluate changes of one parameter, or generating a water quality indices to analyze multiple parameters (Arora and Reddy 2013; Mainali and Chang 2018; Pesce and Wunderlin 2000). However, to quantify recovery success, the use of these analyses could make di cult to determine which are the pollutants that describe the changes over time and the factors associated. The multivariate techniques are appropriate for this purpose as they allow simultaneous evaluation of all variables in a complex dataset (Haase and Ellis 1987;Tripodi et al. 2021). Among them, the Asymmetric Eigenvectors Maps (AEM) analysis is a multivariate technique that was originally developed to model multivariate spatial distributions generated by a directional physical process and had usually been applied to study communities of living organisms (Legendre and Gauthier 2014). This analysis along with the redundancy analysis (RDA) are two multivariate techniques that have been useful for determining spatial patterns in metal contamination (Tripodi et al. 2021). As the processes associated with time are directional, these two techniques should also be suitable for analyzing time series of pollutants (Legendre and Gauthier 2014). sanitation guidelines. Among the focus of these guidelines were to generate actions to improve the life quality of the inhabitants, the recompositing of the environment in all its components (water, air and soil) and the prevention of damage (ACUMAR 2016). The monitoring of the surface water of the MRR is an action that has been quarterly carried out since 2008 to the present. This supervision includes tracking in 38 sites from the MRR basin of more than 50 representative parameters of water quality, including the measurement of physicochemical parameters, metals, organic compounds and hydrocarbons (ACUMAR 2016).
Although descriptive analyses of the metal concentration in water have been conducted in the MRR, there are no studies that focus on the temporal changes in metal concentration to evaluate the effectiveness of the actions carried out. The overall goal of this work was to evaluate temporal changes in metal concentrations in a polluted basin applying the AEM method and to determine the physicochemical and meteorological parameters associated with these changes using the RDA.

Description of the study area
The MRR basin is located within the Province of Buenos Aires (90%) and the city of Buenos Aires (10%), Argentina. This is a plain river of short length (80 km), low ow (8 m 3 /s) and a not very steep slope (0.35%) (Faggi and Breuste 2015). The MRR basin is topographically divided into three sub-areas: the upper, middle and lower basins. In the upper basin, which is mainly surrounded by agricultural and cattle breeding areas (ACUMAR 2009), the water comes from plenty of streams, although only three of them are main streams. In the middle basin, which is less urbanized and industrialized than the lower basin and thus has lower pollution levels (ACUMAR 2009), these streams join to form the main watercourse, the Matanza River (Faggi and Breuste 2015). Finally, in the lower basin, which shows the highest population and industrial densities of the entire basin, the Matanza River is renamed Riachuelo about15 km before discharging into the Río de La Plata River (Armengol et al. 2017). The matrix of this area is characterized by shanty towns and industries, with parks and open green areas forming patches. The main industries located here are tanneries, cold storage warehouses, and metallurgical and chemical industries (Nápoli 2009). In this subarea, the river has been channelized and recti ed and thus lost its natural characteristics. The natural functioning of the basin is also affected by road and railway embankments arranged across the water courses.

Physicochemical parameters of water dataset
To assess the relationship between temporal changes in metal concentrations and physicochemical water parameters in each site, we used concentrations of Electric conductivity (EC), Biological oxygen demand (BOD), Chemical oxygen demand (COD), Dissolved oxygen (DO), pH, Total Suspended solid (TSS) and Water temperature (TW) as explanatory dataset. These parameters were obtained from the same fteen sampling sites used for the metal concentration datasets between 2008 and 2015 from ACUMAR surface water monitoring campaigns (ACUMAR 2018a).

Meteorological dataset
To assess the relationship between temporal changes in metal concentrations and meteorological variables, a dataset with the daily temperature and rainfall records from each site were provided by the National Meteorological Service of Argentina (Ezeiza station, period 2008-2016). The accumulated rainfall and the average of medium, minimum and maximum temperature 3, 7, 30, 90 and 180 days prior to each ACUMAR sampling were also analyzed.

Data analyses
To explore the relationships between metal concentrations (response dataset) and each explanatory dataset (physicochemical parameters of water, meteorological factors or temporal variations), three Redundancy Analysis (RDA) were carried out in each site (S1 to S15). In each case, to reduce the number of explanatory variables, a forward selection procedure was used to select variables for a parsimonious model (Blanchet et al. 2008). Variation partitioning was performed through partial RDA to quantify the variations in the metal concentration explained individually and jointly by all explanatory datasets (AEM, physicochemical parameters of water and meteorological factors) (Borcard et al. 1992;Meot et al. 1998).

Results
Metal concentrations, physicochemical parameters of water and meteorological variables for each site in the MRR basin between 2008 and 2015.
Although there were some differences among sites, metal concentrations showed few obvious temporal trends in each site (see supplementary material). Sites in the lower basin had, on average, a greater dispersion in their metal concentrations (trace of the metal matrix values between 0.0596 mg.L −1 in S1 and 4.1303 mg.L −1 in S4) than sites of the middle and upper basins (trace of the metal matrix values between 0.0281 mg.L −1 in S13 and S15 and 0.0898 mg.L −1 in S11) and the highest value was found in site 5. Also, this site showed the highest concentrations of chromium, BOD, COD and TSS. In turn, this site had one of the lowest concentrations of dissolved oxygen in water. A general pattern of the variables analyzed and the location of the sites within the basin was not observed.
Regarding the meteorological variables in the study period, the average annual temperature ranged from 17.0°C to 17.7°C and the average annual precipitation varied between 7.1 mm and 10.8 mm. In winter, the average temperature was 11.9 ºC (range -5.3 ºC -34.3 ºC) and the average of rainfall was 6.6mm (range 0 mm -72.3 mm). In summer, the average temperature was 23.6 ºC (range 2.5ºC-39.0ºC) and the average of rainfall was 10.4 mm (range 0 mm -92.0 mm).

Metal concentrations and their temporal distribution in each site
In each site, the AEM explained between 10 and 57% of the variation in metal concentrations. All sites showed temporal changes in metal concentrations in the period studied. Each site had among one and seven signi cant eigenvectors and the rst eigenvector (vector accounting for monotonic trends) was statistically signi cant in all models (Table 1). These eigenvectors were associated with the RDA axes and form different temporal pattern. In total, each site had 1 or 2 different temporal patterns of change in metal concentrations and the rst RDA axis indicated the trend of temporal variation in the metal concentrations in the MRR. Between 2008 and 2012, thirteen sites (Sites 1, 2, 3, 5, 6, 7, 8, 9, 10, 11, 12, 13 and 14) showed a decrease in all metal concentrations except arsenic, whereas site 4 showed the same pattern but only for concentrations of cadmium, nickel and lead. After 2012, several sites (lower basin site 4; middle basin: sites 11 and 12; upper basin sites13, 14 and 15) exhibited a reversal in this decrease, even returning to the original values of metals concentrations. The last site, site 15, was the only that did not follow the trend and showed an increase in As, Cr and Pb concentrations. Also, ten sites (1, 4, 5 6, 8, 10, 11, 12 13 and 14) showed cycles (meso and micro-scale variations) in metal concentrations in super cial water (Fig. 2). In the lower basin, changes in metal concentrations were positively associated with COD (S1, S3, S5, S7 and S8), TSS (S1, S3, S6 and S7), BOD (S1, S2, S3 and S7) and pH (S4); and negatively associated with EC (S1, S4 and S5), and DO (S3). In the middle basin, BOD (S9, S10 and S12) and TSS (S11) were signi cant variables associated positively with metal concentration. Finally, in the upper basin the only variable positively associated with changes in metal concentrations was TSS (S13 and s14) and the signi cant variables negatively associated were pH and WT (both in S13); and BOD and COD (both in S15).

Relationships between metal concentrations and Meteorological variables
The meteorological variables explained a maximum of 16% of the changes in metal concentrations. In 6 to 15 sites, the signi cant meteorological variables were related with the rainfall. In only two sites, the signi cant meteorological variables were related with the temperature. Metal concentrations was positively associated with accumulated rainfall in 4 of the 6 sites (Sites 1, 10, 12 and 13) and negatively associated with the remaining two sites (Sites 2 and 7). Also, in two sites (S4 and S12) metal concentrations was negatively associated with temperature. Table 1 shows all the meteorological signi cant variables in each site.

Comparison of the in uences of physicochemical variables in water, meteorological variables and temporal patterns on the variation in metal concentrations
The amount of variation in metal concentrations explained by all datasets analyzed (temporal variation, physicochemical parameters of water and meteorological factors) differed among sites (Fig. 3). The total variation in metal concentrations of each site explained by these datasets in conjunction varied between 18 and 62%. Individually, in 8 of 15 sites (sites 2, 4, 6, 8, 12, 13, 14 and 15) temporal dataset explained the highest amount of variance (11-39%). In six sites (sites 1, 3, 5, 9, 10 and 11) the combined effect of temporal variation and physicochemical parameters of water was the dataset that explained the greatest amount of variance (between 9% and 30%). In the remained site (S7), the largest amount of variation in metal concentrations (15%) was explained by the physicochemical parameters of water individually. The meteorological factors, individually or combined, explained at most 16% of total variance (Table 1 and Fig. 3).

Discussion
Quantifying recovery success in polluted ecosystems is an urgent challenge that faces numerous di culties due to the complexity of the datasets necessary for its useful evaluation. The results of this work showed that AEM and RDA multivariate techniques are helpful for this purpose, because they allowed make evident temporal patterns (at different scales) in metal concentrations and several factors associated with these patterns.
In 13 of the 15 sites analyzed in the MRR basin, all metals analyzed had the same trend between 2008 and 2016, except arsenic. Possibly this is because the arsenic is mainly associated with natural contamination of water layers, the use of certain fertilizers, herbicides and pesticides; unlike the other metals analyzed that are associated with sources of industrial pollution (ATSDR 2004(ATSDR , 2005(ATSDR , 2012 after rainfall events, nding that although most of the dissolved metals decreased their concentrations due to the dilution effect to the increase in water ow and precipitation processes; concentrations of metals (such as As, Pb and Cr) bound to particles increased due to the mobilization of sediments. In the Matanza-Riachuelo, previous studies have found the absence of the dilution effect of precipitation on the concentration of metals, mainly due to the characteristics of the sediments of the Riachuelo, with high levels of sulfur, and to the little dissolved oxygen in the water, as well as to the discharge to the sewage channel without previous treatment (Bargiela and de Iorio 2013). Similar results were observed in streams in the Reconquista River Basin (Argentina) (Arreghini 2005). The results of this work showed a positive association between accumulated precipitation days previous to sampling and changes in the concentration of metals in 4 sites, but negative association among these variables in other 2 sites. These differences between sites in the effect of precipitation on the concentration of metals could be explained by differences in the composition of the sediments of the river bed. Sites with sediments more contaminated could lead to further resuspension of particles with metals.
On the other hand, we found a negative relationship between metals and temperature. Korfali and Davies (2003) suggested that this phenomenon could be a deposition process of metals, because the decrease in temperature could be lowered both the carbonate mineral solubility and the oxidation of organic constituents (lower oxygen content in water), and consequently lowering concentration of metals in the water. However, Hazarika and Kalita (2020) showed a positive relationship between different heavy metals and temperature in River Brahmaputra, then, possibly the effect of the ambient temperature is also affected by the composition of the sediments and the physicochemical characteristics of the water. This is consistent with the fact that meteorological variables alone explained less than 5% of the changes in According to ACUMAR (2021b), the 42.28% of the industries that were considered polluting had adapted their processes in order to avoid environmental contamination outside the admissible limits. In addition, more than 3770 tons of solid waste per year was removed from the waters of the MRR (ACUMAR 2021a).
These actions, as mentioned above, possibly reduced the levels of organic contaminants in water, as well as total suspended solids, and also metals. This may the reason of why in most of the sites (14 out of 15) throughout the entire basin, the change of metal concentrations were associated with at least one these three physicochemical variables. However, Casares and de Cabo (2018) found an increase in the chemical oxygen demand (COD) in three monitoring points along the lower basin of the MRR from 2008 to 2016. These results, unlike those obtained in our study, indicated a greater discharge to the river of organic compounds whose biodegradation requires more time (chlorides, nitrites, metals, sul des, ammonia, soluble phosphorus, suspended solids), implying a possible increase in the toxicity for the species of the aquatic biota. Therefore, although it was not the objective of our study, more studies would be necessary to evaluate the temporal changes of this variable.
On the other hand, the results of our study indicated that DO and pH had signi cant associations with metal concentration at only one and two sites (respectively), indicating that do not in uence markedly the MRR basin stretch under consideration. Organic matter dissolved in higher concentrations consumes large amounts of oxygen, thus reducing DO. In turn, due to the lack of DO, anaerobic fermentation processes are produced that release organic acids. The pH of the water decreases possibly due to the hydrolysis of these organic acids (Chounlamany et al. 2017;Wang et al. 2013). Therefore, these two physicochemical variables would also be related to the amount of organic matter in the water.
In the restoration of aquatic systems contaminated with metals, once the toxic input has ceased, and except for speci c actions such as liming, everything else can be left to natural processes (Bradshaw 1996). An important factor that can in uence the variation of the levels of physicochemical parameters and the total concentration of heavy metals is the river ow (Chapman 1992). The Matanza-Riachuelo is a plain river, with a low slope and low ow (Faggi and Breuste 2015). Rainwater ows into the main course only through 4 major tributaries (one piped), despite having hundreds of tributary streams, and all of them are highly polluted. Thus, the characteristics of the MRR signi cantly affect its capacity for selfpuri cation and the restoration processes would be more extensive in time, unlike other larger and faster rivers that have a greater dilution capacity. This highlights the importance of long-term temporary studies and adequate monitoring and analysis strategies to assess changes over time.
Conclusions Wohl et al. (2005) suggested that self-sustaining, ecologically successful recovery efforts must be designed in relation to broad spatial (watershed) and temporal contexts. However, long-term monitoring programs that evaluate recovery of degraded ecosystems following the removal of a stressor are relatively rare (Clements et al. 2010). Therefore, the effectiveness of recovery programs is hard to asses (Bernhardt 2005; Clements et al. 2010; Jones and Schmitz 2009). The analysis proposed in this study is a useful tool for detecting temporal trends in the concentrations of environmental pollutants, which are not easily detectable in highly degraded environments and evaluating factors associated with these changes. This method of analysis could be applied to different scenarios (aquatic and terrestrial) affected by the continuous advance of human activities that generate pollution and require, therefore, a rigorous evaluation to avoid harmful consequences for human health and the ecosystem. For example, temporal variations of compounds derived from the mining industry in the waters near the exploitation site, or from the agrochemical compounds (fertilizers, pesticides, etc.) in both the water and the soil of areas dedicated to agriculture, among many other, could be analyzed as well as their associations with different environmental variables.

Statements And Declarations
Funding: This work was supported by Universidad de Buenos Aires, Consejo Nacional de Investigaciones Cientí cas y Técnicas (CONICET, Argentina) and Gobierno de la Ciudad de Buenos Aires.
Availability of data and material: Data is available in the supplementary material Code availability: Not applicable Authors' contributions: All authors contributed to the study conception and design. Data collection and analysis were performed by Mariel Tripodi and Gerardo Cueto. The rst draft of the manuscript was written by Mariel Tripodi and all authors commented on previous versions of the manuscript. All authors read and approved the nal manuscript.

Con ict of interest/competing interests: Not applicable
Ethics approval: Not applicable.
Consent to participate: Not applicable.
. WHO (2010) World Health Organization. Exposure to cadmiun: A major public health concern. Geneva, Switzerland

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. Dataandmaterial.csv Supplementarymaterial.docx