A continental-scale predictive simulation tool for accidental spills in major South American water courses

Early warning systems (EWSs) integrate a set of actions aimed at notifying decision-makers about sudden events or incidents. When efficiently implemented and align with social needs, they can significantly enhance risk management (Hou et al. 2013; Tiyasha Tung et al. 2020). EWSs have been developed around the world in response to sudden water contamination accidents. Notable examples include the organic detention system ORSANCO (Ohio River/US) and the IKSR system in Rhine River (Germany, Netherlands, and Switzerland), a multinational EWS that extensively utilizes biomonitors (Grayman et al. 2001).

Many EWS of water pollution have evolved from water quality models. An example of this is the ICWater (Incident Command Tool for Protecting Drinking Water), an incident command tool developed from the RiverSpill water quality model (Samuels et al. 2006). ICWater tracks and models the transport of contaminants in water bodies that supply drinking water to the North American population (Samuels et al. 2015).

Mathematical modeling of accidental contaminant releases into watercourses plays a crucial role in EWSs by predicting travel time and concentration using water quality models. According to Wang et al. (2013), over 100 water quality models have been developed worldwide. Numerous reviews have examined different types of models (Moriasi et al. 2015; Yuan et al. 2015; Hrachowitz et al. 2016; Tiyasha Tung et al. 2020; Bai et al. 2022), including those designed for large basins (Fu et al. 2019).

Several of the currently available water quality models can track the transport of contaminants from large accidental releases, which are sources of water pollution and frequently disrupt municipal water supplies. Machine learning and other advanced statistical techniques offer additional tools for water quality simulations. However, the accuracy and effectiveness of these models largely depend on the availability and quality of data, posing challenges, particularly in the case of large-scale spills.

This study focuses on developing solutions to address challenges in South American watersheds. Over the past few decades, South America has experienced numerous accidental pollutant releases into vital water bodies, impacting the population that depends on them. In 2003, about 1.2 billion liters of paper industry waste contaminated the Paraíba do Sul River following the rupture of a waste dam in Cataguases, Brazil (Mady et al. 2018). In 2015, the rupture of the Fundão dam in Mariana/BR led to the release of 32 million m³ of mining waste in the Doce River. That same year, operational failures at the Veladero mine, in San Juan, Argentina, resulted in the released of about a 1,000 m³ of a cyanide-containing solution into the Las Taguas and Blanco rivers (Peressotti 2017; Halim & Naidu 2024).

Natural disasters have also contributed to water contamination. In 2017, hydrometeorological events in Rolante, Brazil, triggered a series of landslides that caused debris flows to enter the region's waterways, shutting off water supplying for more than eight cities (Guirro & Michel 2023) due to the increased sediment concentrations. In 2019, another mining dam failure occurred in the state of Minas Gerais, Brazil. The rupture of the Córrego do Feijão dam in Brumadinho released about 12 million m³ of iron ore rejects into the Paraopeba River (Gomes et al. 2020; Thompson et al. 2020).

Governmental agencies are essential in managing responses to water contamination incidents. In Brazil, the National Center for Monitoring and Early Warning of Natural Disasters (CEMADEN) is a key example. CEMADEN has been instrumental in advancing scientific knowledge, implementing monitoring systems, and issuing timely warnings for various disasters, including landslides, hydrological events, and meteorological phenomena (Marengo et al. 2023). The significance of such agencies extends beyond national borders, as water contamination events can propagate through interconnected river networks, affecting multiple countries, particularly in South America.

In this context, developing and implementing models capable of operating at national or continental scales could be invaluable for disaster management agencies. These models would enhance their ability to issue effective warnings and coordinate responses across broad geographic areas. However, challenges such as limited data availability and language barriers hinder the rapid response required for water quality emergencies.

Continental-scale modeling of water resources and water quality itself is a relatively new entrepreneur in science and technology. Abbaspour et al. (2015) developed and calibrated a high-resolution, large-scale SWAT model for Europe, simulating water resources, crop yields, and nitrate leaching. Their model addresses groundwater quality challenges and has practical applications for water management. Similarly, Voß et al. (2012) introduced a continental-scale river water quality model for Europe that simulates biochemical oxygen demand (BOD₅) and total dissolved solids (TDS) based on anthropogenic loadings and flow dilution. However, none of those cases specifically proposed water quality technologies for EWS applications and spill modeling.

As noted by Strokal et al. (2019), global multi-pollutant water quality modeling presents numerous scientific and practical challenges. These include integrating diverse and often sparse data sources across different scales, accurately calibrating and validating models with limited observational data, and developing methodologies for large-scale scenario analyses to predict future impacts. Addressing these challenges requires advanced modeling techniques and enhanced computational frameworks to support effective and scalable water quality management.

The primary challenge in developing large-domain water quality models in contamination events is the issue of scale and data acquisition. However, studies, such as Siqueira et al. (2018), show the possibility of applying hydrological–hydrodynamic across all of South America. Similarly, Fan et al. (2015) present large basin-scale water quality modeling tools that operate effectively with limited data availability, low computational cost, and broad spatial coverage. A tool that integrates continental- and large-scale modeling approaches in both hydrology and water quality could enhance early warning capabilities for accidental releases enabling more effective and timely management responses at national or continental scales.

Given this context, the present study aims to develop and evaluate a solution to address the lack of large- to continental-scale EWS for water quality emergencies. The objectives of this study are: (1) to address the absence of early warning tools for water contamination accidents in South America by developing a simulation tool called SPLACH-AS (Sistema de Previsão de Lançamentos Acidentais em Cursos Hídricos – acronym for Predictive System of Accidental Spills in Watercourses in Portuguese), designed for integration into an EWS and (2) to assess the tool's effectiveness in predicting travel times and peak concentrations during accidental release events. The tool combines a water quality model with hydraulic–hydrological simulations, providing a novel framework for large-scale contamination spill scenarios. Additionally, we aim to enhance the underlying water quality model by incorporating reservoir and water temperature modules to improve its predictive accuracy.

Figure 1 also illustrates the dense grid of reservoirs across Brazilian rivers, highlighting their significant presence within the modeled region. These reservoirs regulate river flow and water storage but also affect the transport and distribution of pollutants throughout the river network. The alteration of natural flow regimes by these reservoirs can affect sediment transport, nutrient cycling, and the dispersion of contaminants, potentially leading to localized impacts on water quality and aquatic ecosystems. A suitable tool to predict contaminant transport would benefit of the inclusion of these reservoirs in its stream network.

Figure 1 also presents the three study sites used for tool validation. Two sites, Brumadinho and Mariana, are in Minas Gerais, a southeastern state of Brazil, while the third, Rolante, is in Rio Grande do Sul, the country's southernmost state. Pollutant dispersion from accidental release events can extend over vast distances, potentially crossing state borders and even reaching the ocean, as observed in the Mariana case along the Doce River. The extensive coverage of the simulation allows the tool to capture the hydrological dynamics across various regions of South America. By modeling such a broad area, the tool offers a platform for analyzing water resource management, ecological impacts, and potential responses to environmental changes. Numerous large rivers traverse vast areas of the continent, highlighting the interconnected nature of these water systems. This underscores the importance of having a simulation tool capable of operating at this scale. Therefore, the development and application of this simulation tool provide valuable insights, reinforcing the significance of your study in advancing continental water management efforts in South America.

As input for the SPLACH-AS simulation tool, we used spatial, hydraulic, and hydrological data from the hydrological and hydrodynamic simulations developed by Siqueira et al. (2018) for the entire South American continent. This simulation utilized the MGB hydrodynamic–hydrological model (Modelo de Grandes Bacias – acronym for Large Basins Model in Portuguese), first published by Collischonn et al. (2007) and further improved by Pontes et al. (2017), Lopes et al. (2018), and Fleischmann et al. (2021).

Siqueira et al. (2018) extended the hydrological model to cover the entire South American continent (MGB-SA), comparing the results obtained with observed data on daily river flow, water levels, storage, and evapotranspiration estimates. Additionally, the study compared daily flows with results from global models, achieving satisfactory results. This continental model was subsequently applied to other studies, including those by Fagundes et al. (2021), Siqueira et al. (2021), Petry et al. (2023), and Kolling Neto et al. (2023).

The spatial and hydraulic data provided by the MGB-SA include the rivers' drainage network and detailed information about each river section, such as its catchment area, slope, length, order, width, and depth. Additional hydraulic information, such as the estimated Manning coefficient for each river section, is also provided. To visualize the spatial data, maps of the drainage network, catchments, and watershed boundaries on the continent were also obtained from the study by Siqueira et al. (2018).

The hydrological data provided by the MGB-SA, serving as input for the SPLACH-AS simulation tool, encompasses both reference (Q90, Q70, Q50, Q30, and Q10) and daily flows for each river section spanning from 01 January 1990 to 31 December 2010 across the entire continent. Here, the reference flow Qx denotes the flow with a probability of exceeding x%. These flows provide a wide range of scenarios that can be leveraged for real-time applications, enabling the simulation of water quality for any stream within the South America domain. By incorporating multiple scenarios, users can also generate uncertainty bands, offering a robust framework for decision-making under varying conditions. To enhance real-time applicability, historical daily data can be employed to calibrate the physical parameters of specific rivers or watersheds more precisely. These calibrated parameters can then serve as a foundation for predicting pollutant propagation during potential future accidents or sudden events, such as extreme rainfall, droughts, or other unanticipated occurrences. Furthermore, the tool integrates additional parameters, including flow velocity and daily water levels.

Pollutant release data was also collected to validate the SPLACH-AS simulation tool. We focused on three major events in Brazil. The data came from the affected cities' water supply companies and national agencies, including ANA (Agência Nacional de Águas – acronym for National Water Agency in Portuguese), CPRM (Companhia de Pesquisa de Recursos Minerais – acronym for Brazilian Mineral Resources Service in Portuguese), IGAM (Instituto Mineiro de Gestão das Águas – acronym for Minas Gerais Water Management Institute in Portuguese), and IBAMA (Instituto Brasileiro do Meio Ambiente e dos Recursos Naturais Renováveis – acronym for Brazilian Institute for the Environment and Renewable Natural Resources in Portuguese).

The pollutant propagation throughout a large-scale river network in this study is conceptually based on the water quality model introduced by Fan et al. (2015). The modeling framework simulates the propagation of pollutants from point sources, such as industrial discharges, in extensive water bodies. The processes of advection, decay, and dispersion are considered in the pollutant propagation.

To calculate the propagation of the plume in section iii, the challenge is that Equation (2) requires a mass of pollutant as input, while the available data is the concentration over time. Consequently, the pollutograph convolution scheme employs a discretization method, transforming the concentration over time into small increments of mass. These increments are propagated through section iii using the advection-dispersion-reaction (2) equation and are summed at the end of section iii, at confluence b, as shown in Panel (4), following the superposition principle. A more detailed explanation can be found in Fan et al. (2015).

Additional kinetic and physical parameters are available for calibration, including dispersion (C1), temperature (C2), decay (C3), and water velocity (C4) coefficients. These parameters can multiply the original values, enabling the model to account for the specific characteristics of each river.

Additional improvements have been made to the original water quality model (Fan et al. 2015) algorithms while integrating them into the SPLACH-AS simulation tool. These include adding modules for computing the effects of water temperature on pollutant transport and another module for accounting for the retention effects of large reservoirs on pollutants. Haag & Luce (2008) emphasize that water temperature serves as a fundamental parameter for assessing water quality, directly affecting ecological processes, chemical reactions, and the toxicity of pollutants.

The modifications made help enhance the accuracy of the model, particularly in continental-scale modeling. South America's extensive climatic diversity, spanning from equatorial zones to tundra regions, leads to significant variations in river temperatures. Moreover, the considerable altitudinal differences across various regions also impact freshwater temperatures.

Another significant advancement in the SPLACH-AS simulation tool is its integration of large reservoirs into the Brazilian river network (Figure 1). The reservoir data was obtained from Passaia et al. (2020), who utilized the continental version of the MGB-AS model to incorporate observed outflows from these reservoirs. Their study served as a preliminary investigation for potentially explicitly representing these effects in the hydrological model MGB. Approximately 109 reservoirs have been integrated into the SPLACH-AS tool.

Due to computational constraints, the memory allocation for the entire continent's original input data (approximately 33,000 river reaches) would only support simulations lasting around 80 h or less. However, large contamination events, for which the tool was developed, commonly have a propagation of pollutants over a longer duration. For instance, the rupture of the Fundão dam in Mariana, Brazil, in 2015, where the rejects took almost 15 days to reach the Atlantic Ocean. To address this, a preprocessing program was developed to extract only the user's area of interest from the original input files. The preprocessing program was developed using Microsoft Visual Studio 2019 software in the VB.Net language, utilizing the .NET Framework. It extracts from the original files of the entire South American continent only the river sections affected by pollutants until they reach the ocean.

The development of the novel GUI for the SPLACH-AS simulation tool was also undertaken using the Microsoft Visual Studio 2019 software, employing the VB.Net language.

Validating a tool like SPLACH-AS requires observed data, which is often a significant challenge in such studies. Unfortunately, in recent years, a series of environmental disasters have occurred in South America and particularly in Brazil. These events not only provide accidental pollutant data but also highlight the importance of developing such a tool that can help predict pollutant propagation on a large scale.

This way, after developing the SPLACH-AS simulation tool (Figure 2), a series of simulations were conducted for three significant water contamination events (Figure 1). The first simulated event occurred in 2017 along the Sinos River, where landslides north of the city of Rolante caused a substantial influx of sediment into the river. The second simulated event modeled the 2015 release of tailings from the Fundão dam failure in Mariana, Minas Gerais. The third simulation focused on the 2019 iron waste contamination caused by the B1 dam failure at the Córrego do Feijão mine in Brumadinho, also in Minas Gerais.

For each contamination event, we ran four different simulations (numbered 1 to 4 in Figure 2). The first two simulations were made with the developed SPLACH-AS tool, one with calibration (1) and one without (2). Simulation number (3) was conducted with the basin-scale benchmark SIAQUA-IPH model (Fan et al. 2015) for comparison and validation purposes. The last simulation (4) was made with a simplified model, applying the equation of advection-dispersion-reaction at each measurement site and comparing it with the observations.

For simulation (1), we calibrate the parameters ⁠, k, and (velocity corrector coefficient) at the SPLACH-AS tool. This calibrated approach aimed to fine-tune the model's parameters to achieve better accuracy in representing pollutant transport. For simulation (2) (no calibration), we used Equation (3) (Kashefipour & Falconer 2002) to automatic estimate ⁠, with a fixed value of and a reference flow Q90. This non-calibrated approach provided a baseline comparison for the model's performance under default parameter settings.

For run number (3), the SIAQUA-IPH model was used with calibration of the parameters (longitudinal dispersion corrector coefficient), k, and (velocity corrector coefficient). This served as an alternative model to compare against the performance of the SPLACH-AS system, exploring different modeling methodologies.

For run number (4) utilizing a simplified model based on the advection-dispersion-reaction equation (Equation (1)), we used the distance from monitoring stations or affected cities and average flows, velocities, and areas. The parameters and k were calibrated for all the stations/cities. This simplified approach was intended to mimic a model that could be used in case SPLACH-AS was not available, providing a quick, practical, and simplistic solution for emergency scenarios involving accidental releases in South America.

The combination of these simulations allowed us to assess the capabilities of SPLACH-AS, understand the impact of different parameter settings, and establish a simplified contingency model for critical situations when time and resources are limited.

The application of the SPLACH-AS simulation tool consists of six stages (Figure 4). The first stage (1. Watershed Data) involves loading the hydraulic and hydrological data into the tool (Figure 4(a)). This stage also includes selecting the time frame for the simulation. After loading the data, the tool verifies if the data were loaded correctly and provides a summary of the loaded data.

The second stage (2. Flow) involves the flow definition process (Figure 4(b)). In this phase, the user has the option to choose between using reference flows or daily flows for the simulation. If the user selects reference flows, they can further choose between Q90, Q70, Q50, Q30, or Q10 flows. If the user opts for daily flows, they need to input the desired date for the simulation and load the corresponding daily flow and daily level data.

In the next step (3. Parameters), the user can calibrate the physical parameters of the basin (Figure 4(c)). This involves modifying the longitudinal dispersion coefficient, the environmental decay coefficient, and adjustment coefficients for temperature and velocity. There is also an option to save the chosen parameters in an external file or open previously saved parameter sets.

In the next stage (4. Release), the release data is uploaded to the model. The pollutant mass and the location of the release need to be specified. As in the previous stage, this information can be saved or opened from external files. There is an additional feature that allows the user to calculate the length of the not completely mixed zone, indicating from where the results from the model can be considered valid.

Afterward, in the next step of the simulation tab (5. Simulation), the user can check a summary of the input data, choose to activate the reservoir module or not, and finally, run the simulation. During the simulation, a progress bar is displayed to track its progress. When the simulation is complete, the model automatically proceeds to the next and final step (Figure 4(d)).

The final stage (5. Results) presents different ways to check the simulation results. The user can specify a location downstream the release in the river and generate a graph of the pollutant propagation in this section. The user can also save the results in text or binary files.

In 2017, after heavy rains in the hillside regions to the north of the city of Rolante, Brazil (Figure 5), near the headwaters of the Sinos River basin, a large quantity of sediments reached the Sinos River and traveled for many days. This plume was identified as the cause of the interruption of many municipal water supplies (Guirro & Michel 2023). We obtained the approximate arrival time of the sediment plume from the reported event, along with records of water supply interruptions due to sediment load during this period, from direct contact with the public supply companies of the cities located downstream of the release.

To conduct our simulations, a sediment release was introduced into the river section closest to the city of Rolante in each model (SPLACH-AS, SIAQUA-IPH, and the simplified model). The mass was estimated based on the number of scars from the landslides. However, the concentration data are not relevant in this context, as no observed concentration data are available for comparison. Our analysis will focus on sediment travel time. Figure 5 shows the drainage networks of both the SPLACH-AS simulation tool and the SIAQUA-IPH model. The river network in the SIAQUA-IPH model is more detailed compared with SPLACH-AS, which features a coarser representation.

The water supply system of Campo Bom is interconnected with the systems of the cities of Estância Velha, Sapiranga, and Portão. Additionally, the system of the city of Esteio is linked to Sapucaia do Sul (Figure 5). These interconnected systems should be taken into account when analyzing the pollutographs and interpreting the results.

Table 1 provides a summary of the performance metrics for each of the simulated models. The water supply interruption time corresponds to the observed peak for each city. Time 0 indicates the start of the plume travel north of Rolante.

Considering Table 1 and Figure 6, the SPLACH-AS tool with calibration and the SIAQUA-IPH model exhibited similar and satisfactory performances. The maximum errors in terms of total simulation time (72 h) for these models were 5 h for the SIAQUA-IPH model and 2.2 h for the calibrated SPLACH-AS. In contrast, the SPLACH-AS without calibration and the simplified model displayed larger errors. However, even the SPLACH-AS model without calibration demonstrated fewer errors compared with the simplified model.

These results demonstrate that the calibrated SPLACH-AS tool and the SIAQUA-IPH water quality model captured the dynamics of sediment transport within the Sinos River. These findings underscore the potential of these models in simulating complex environmental processes and are consistent with previous studies that have utilized advection-dispersion-reaction equation models to simulate sediment transport dynamics. For example, Andika & Julien (2021) focused on simulating sediment transport originating from a point source of mud, highlighting the versatility and applicability of such modeling approaches across various scenarios and geographical contexts.

To conduct the simulations, a sediment release was introduced at the river section nearest to the Candonga reservoir. This decision was based on the understanding that approximately 90% of the tailings from the dam rupture were deposited in the plains between the Candonga reservoir and the Fundão dam, involving more complex processes than sediment transport in rivers. Therefore, only 10% of the total mass released during the dam break was used in the simulation, representing the volume of tailings that flowed into the basin's channels downstream of the Candonga dam.

As observed in Figure 8 and Table 2, the simulations conducted with the calibrated SPLACH-AS tool and the SIAQUA-IPH model effectively represented the propagation of tailings, showing minimal errors in peak time. Regarding peak concentration, the calibrated SPLACH-AS tool exhibited errors of up to 7% at 9 out of 10 monitoring stations, with larger errors observed at the last station.

The simulations using the uncalibrated SPLACH-AS tool, and the simplified method exhibited considerably larger errors. Regarding peak times, the simplified method outperformed the uncalibrated SPLACH-AS tool only at the first four monitoring stations compared to simulations with the SPLACH-AS tool without calibration. In terms of peak concentrations, both the SPLACH-AS tool without calibration and the simplified model showed substantial errors. Figure 8 displays the concentration on the secondary vertical axis for the simplified model and SPLACH-AS without calibration.

The observed trend indicates that the curves simulated without the presence of reservoirs exhibit higher concentrations compared to the scenario with the reservoir module activated. Therefore, the retention effect caused by these structures in sediment retention within the river can be analyzed.

The Baguari reservoir, situated downstream of the second monitoring station, presents a retention of approximately 12% of sediment, effectively slowing down the sediment transport process. Similarly, the Aimorés and Mascarenhas reservoirs, located between monitoring stations 5 and 6, exert a significant influence by contributing to a 56% reduction in sediment transport downstream. Together, all reservoirs examined in this study contribute to an overall sediment retention of approximately 68%. These findings align with observations made by Palu & Julien (2019), who also highlighted the significant role played by reservoirs in trapping sediments following dam break events, underscoring their consistent impact on sediment dynamics in river systems.

On January 25, 2019, approximately 3 years after the Mariana disaster in the Doce River, the B1 dam at the Córrego do Feijão mine ruptured near the city of Brumadinho, Minas Gerais, Brazil. The dam contained approximately 12 Mm³ of iron ore mining rejects (Costa et al. 2024). Following the rupture, approximately 75% of the mine's rejects, equivalent to 9.9 Mm³, were released into the Ribeirão Ferro Carvão and subsequently reached the Paraopeba River, a significant tributary of the São Francisco River (Robertson et al. 2019).

The monitored data revealed multiple peaks of sediment concentration following the release of rejects. The initial peak was directly linked to the dam failure on January 25th, while subsequent peaks occurred after rain events on January 30th, February 4th, February 7th, and February 16th. These precipitation events likely triggered the remobilization of rejects that had previously settled (Fonseca et al. 2022).

Based on the analysis of the observed data, four simulation scenarios were developed. These scenarios incorporated the pollutant release due to the dam rupture (scenario 1) and pollutant remobilization from the identified precipitation events (scenarios 2, 3, and 4). Scenario 3 presents two releases due to two precipitation events that occurred very close to each other in time.

It is noted that for the SPLACH-AS tool simulations, the monitoring stations were located in the same river section. In contrast, for the SIAQUA-IPH model simulations, the stations were located in two different river sections, resulting in pollutographs from two river sections. Additionally, for the simplified model, two distances from the monitoring stations were adopted, resulting in two simulated pollutographs.

A potential caveat of this third case study was the spatial distribution of the monitoring stations, particularly for the SPLACH-AS tool and the SIAQUA-IPH model simulations. The discretization of the models proved to be too coarse for the scale of the measurement sites in this event. Various monitoring stations were positioned in the same river section. Consequently, it was difficult to calibrate a single river section when there were two different observed values of pollutant concentration for the same section.

From Figure 11 and Table 3, it can be observed that for scenarios 1, 2, and 3, the pollutographs from the simulations using the SIAQUA-IPH model and the calibrated SPLACH-AS tool resembled the observed data, particularly in terms of matching peak times rather than peak concentrations. The SIAQUA-IPH model results showed better performance compared with the SPLACH-AS results. In these scenarios, the simplified model displayed slightly better metrics compared with the uncalibrated SPLACH-AS tool, but both exhibited substantial and comparable errors.

In scenario 4 (Figure 12), the SIAQUA-IPH model simulations demonstrated a better visual representation of the reject plume propagation, particularly concerning peak time. In the pollutographs simulated by the calibrated SPLACH-AS tool, the second river section simulated exhibited relatively large errors in both time and peak concentration, while the other river sections displayed more accurate propagation results. In both simulations, peak time errors were small, less than 15%, whereas peak concentration errors varied significantly, reaching up to 100% in SPLACH-AS and 56% in SIAQUA-IPH.

Regarding this fourth scenario, the errors in the uncalibrated SPLACH-AS tool simulations exceeded those of the calibrated scenarios. However, these errors were generally smaller, particularly in peak time, compared with the errors produced by the simplified model.

The authors have developed a simulation tool (SPLACH-AS), specifically designed for predicting accidental pollutant releases in watercourses across all the South American continent. The tool was developed by coupling spatial, hydraulic, and hydrological data derived from comprehensive hydrological simulations covering the entire continent with pollutant propagation routines from an improved water quality model. This framework enables large-scale contaminant simulations adapted to data-scarce environments.

Among the improvements are the incorporation of daily flow modeling and the inclusion of reservoir and water temperature modules. These enhancements significantly expand the application of the system, particularly considering the extensive presence of reservoirs and the significant climatic variations experienced across regions and seasons throughout the South American continent. The inclusion of daily flows also enabled more accurate historical simulations, which can be used to calibrate parameters for future simulations. Validation tests were conducted to assess the tool's performance in representing accidental pollutant releases in large water bodies across South America in recent years. The case studies include accidental spills in the Sinos River, Doce River, and Paraopeba River (Brazil). The latter two represent the most significant environmental disasters in water quality that occurred in Brazil in the last decade: the rupture of the Fundão dam in Minas Gerais and the failure of the B1 dam at the Córrego do Feijão mine in Brumadinho, also in Minas Gerais. These incidents released millions of tons of mining waste into the river network.

The tests were performed on the SPLACH-AS tool with and without calibration. These tests aimed to address scenarios when calibration data is available and when it is not. For comparison, the tests were also conducted using benchmarks (SIAQUA-IPH water quality model and a simplified model).

In comparison with the observed data and in most simulations, the calibrated SPLACH-AS tool achieved adequate outcomes, sharing similar performances with the SIAQUA-IPH model, which benefits from a more detailed spatial discretization. The uncalibrated SPLACH-AS model displayed significant errors, though smaller than the results from the simplified model in most simulations. The simplified model exhibited the most substantial errors, often causing the pollutants' propagation to be ahead of time. The primary advantage of the SPLACH-AS simulation tool is its readiness for immediate simulation. This capability makes it particularly valuable in scenarios where a rapid response is crucial.

The case studies presented in this study are based on the limited data available following major environmental spills in South America. This scarcity of data reflects the challenges in obtaining comprehensive datasets in the aftermath of such incidents. Despite this limitation, our findings underscore the potential value of developing a prediction tool that can model environmental impacts in similar scenarios. Such a tool would be invaluable for effective response strategies, aiding in the mitigation of adverse environmental and ecological consequences in future spill events.

The SPLACH-AS tool has the potential to be an asset in EWSs for South American national water agencies or cross-boundary institutions. The tool can facilitate the development of a comprehensive set of parameters (e.g., longitudinal dispersion coefficient) for the entire South American continent, which could also be applied in simulations performed by other water quality models. This capability would be advantageous, given the wide array of available water quality models and the limited data typically available for conducting water quality simulations.

The first author would like to acknowledge the Brazilian National Council for Scientific Research (CNPq) for the scholarships provided during the research development and for the anonymous reviewers for the suggestions that helped to improve the quality of this work.

All relevant data are available from an online repository at https://www.ufrgs.br/hge/modelos-e-outros-produtos/splach/.

The authors declare there is no conflict.