Climate change in the Upper Paraguay Basin and hydrological impacts on the Pantanal

Climate change caused by human activities has been a growing concern worldwide. These changes will make the management of water resources more critical as hydrological conditions will change in a very uncertain way in the future, where precipitation and temperature are expected to vary considerably from region to region and therefore, changes in patterns spatial and temporal dimensions of these variables can lead to significant changes in the climate that, in turn, end up affecting agriculture, industrial production, and urban development (Fahad et al. 2018). With an increasing trend in the frequency of extreme hydrological events, such as droughts or floods, the importance of studies on the impacts that these changes cause on water resources becomes evident (Hajihosseini et al. 2020).

Climate change can significantly impact the global water cycle by altering hydrological processes. High temperatures lead to higher levels of potential evaporation of surface water, thus increasing surface drying and increased air humidity (Chen & Chang 2021). Precipitation and water flow from rivers are considered common hydrological phenomena due to the periodicity in which they occur. However, many disastrous effects have been occurring due to floods and droughts in wetlands. Thus, they need to deepen their knowledge for a better understanding (Pinto et al. 1976).

The Pantanal is the largest of the five great wetlands in South America, drained into the South Atlantic Ocean via the Paraguay River (Hamilton 2002). A comprehensive classification for Brazilian wetlands was proposed, considering hydrological and botanical aspects (Junk et al. 2014). The Paraguay River and its connection to the south of the continent are exploited in various ways related to transporting people and goods. This river system is one of the oldest transport routes in the region. Used throughout history by populations along its axis, it recently became a strategic element when the Southern Common Market (Mercosul) was created in the early 1990s. Since then, proposals have been made to expand navigation systems to consolidate the Paraguay-Paraná Waterway (Sousa Júnior et al. 2019).

Simulations of system change through hydrological modeling have been widely used to support the decision-making and management of water resources in river basins worldwide (Ougahi et al. 2022). Hydrological modeling can be defined as an indispensable tool in the research, planning, and management of water resources, in addition to assessing the seasonal availability of water, thus helping in decision-making (Abdulkareem et al. 2018). Among the existing hydrological models is the Soil and Water Assessment Tool Plus – SWAT+. This model, on a hydrographic basin scale and of continuous time, has been developed to assist in managing water resources, impact assessment, management of supply water, and diffuse source pollution (Arnold et al. 1998; Neitsch et al. 2011).

The SWAT+ model, in addition to allowing the simulation of water flow, production, and transport of sediments and water quality, allows the analysis of components of the water balance and the analysis of the impact of future conditions of climate and environmental changes (Jha et al. 2006). In this context, and given the importance of the Pantanal region, a better understanding of the water balance and its components is essential, as well as environmental changes such as climate change and deforestation, and what these impact on the components of the water balance of this region. In this context, the Paraguay River watershed, in its upper portion, was considered for application in the present study. The first research hypothesis is that changes in the environment, such as climate change and the conversion of forests into pastures, would hurt the components of the water balance, with a potential impact on the flow series.

In addition, it is important to expand knowledge about meteorological models and projections in the region of influence of the Upper Paraguay Basin to analyze the vulnerability of water transport under climate change scenarios (Sousa Júnior et al. 2019). Therefore, it is important to understand how climate change affects hydrological behavior in the Upper Paraguay River basin. The objective of this study is to analyze the main changes in the hydrological regime that occur in the Pantanal region based on changes in land use and land cover in the upper portion of the Upper Paraguay Basin under climate change scenarios. The proposal for this work considers the use of hydrological modeling using SWAT+ on the Paraguay River, having as an outlet the point of the river in the city of Cáceres – MT, considering the starting point of the North Section of the Paraguay-Paraná Waterway.

The portion of the basin under study drains an area of 32,407,994 km². The main river, called the Paraguay River, rises in the Serra do São Jerônimo, in the municipality of Alto Paraguai in the state of Mato Grosso, at an altitude of approximately 1,200 m above sea level and runs in a North–South direction.

The municipalities covered by the hydrographic basin under study total 25, they are Alto Paraguai, Araputanga, Arenápolis, Barra do Burgres, Cáceres, Campo Novo do Parecis, Curvelândia, Denise, Diamantino, Jangada, Lambari D'Oeste, Mirassol D'Oeste, Nobres, Nortelândia, Nossa Senhora do Livramento, Nova Marilândia, Nova Olímpia, Porto Estrela, Reserva do Cabaçal, Rio Branco, Rosário Oeste, Salto do Céu, Santo Afonso, São José dos Quatro Marcos and Tangará da Serra.

Tucci (2005) emphasizes that the quantity and representativeness of the data for regulating and verifying the hydrological models are essential to obtaining good results.

The different variables that the SWAT+ model requires are daily precipitation data, solar radiation, wind speed, relative air humidity, maximum and minimum air temperature, soil physical properties, topography and land use, and land cover of the area. Hydrographic basin under study (Neitsch et al. 2005), however, not always climatic data are found in a long series without failures on some days.

All the geographic data used were clipped for the study area and projected in a plane coordinate system, using SIRGAS 2000 as a geodetic reference system and the UTM (Universal Transverse Mercator) system as a projection. The corresponding time zone in the region is 21 South (EPSG: 31981). The data required by the model and the respective sources adopted in this work are detailed in Table 1.

In the initial process of hydrological modeling using SWAT + , inserting the relief map of the watershed under study and selecting the ten fluviometric stations as outlets was necessary. Then, the basin area was demarcated, resulting in 46 sub-basins.

After completing the delimitation of the sub-basins, maps of soil types, land use, and occupation were introduced. These maps were combined with the DEM that has the data to generate the percentage slope. In this way, 8,015 (eight thousand and fifteen) potential HRUs were generated. The presence of very small and location specific HRUs can slow down the model run without significantly contributing to the results.

HRUs were made to optimize the execution of the hydrological model. The methodology consists of identifying HRUs with an area of less than 10 ha and specific locations in the watershed. These HRUs were removed from the model, thus leaving a total of 5,295 (five thousand two hundred and ninety-five) HRUs without significantly affecting the results.

At this point, the model is configured with geospatial data. However, tabular data referring to climatic conditions, such as precipitation information, maximum and minimum temperature, solar radiation, wind speed, and relative humidity for the desired period were added to start the flow simulation.

Sensitivity is evaluated by analyzing the response of an output variable to a change in an input parameter. The more significant the change in the output response, the greater the sensitivity (White & Chaubey 2007). The advantage of sensitivity analysis lies in reducing the number of parameters used in calibration, which speeds up the process (Arnold et al. 2012).

Considering that the SWAT+ model is highly sensitive to a wide range of input variables related to vegetation, management, soil characteristics, climate, aquifers, channels, and reservoirs, it is impossible to accurately determine each one of them (Jha et al. 2006). For this reason, performing a sensitivity analysis is justified. In this study, the sensitivity analysis was conducted using the SWAT+ Toolbox software and the Sobol method for assessing global sensitivity.

Sobol's method involves a global sensitivity analysis based on variance using Total Sensitivity Indices. This method considers both the main effect of the variation and the interaction between the parameters (Sobol’ 1990). Furthermore, Sobol's method is widely used to evaluate the importance of parameters in complex models. It allows for identifying which parameters have a greater influence on the variability of results and which are less relevant. This enables a better understanding of the relationships between the parameters and the model output, helping to improve calibration and validation (Saltelli et al. 2000).

The sensitivity analysis of parameters in the SWAT+ Toolbox model was conducted comprehensively using two distinct methods. The substitution method was replacement for the parameters perco, surlag, slope, lat_ttime, esco, alpha, epco, flo_min, revap_co, and revap_min. This method replaced values within a predetermined range, allowing for a direct assessment of the impact of changes in each parameter on model outputs. Conversely, the parameters bd, awc, cn2, z, canmx, k, and ovn underwent sensitivity analysis using the percentage method. Here, values varied according to a pre-established percentage, enabling a broader sensitivity evaluation to parameter percentage changes. This combined approach provided a comprehensive understanding of the influence of each parameter on model behavior, allowing for precise and optimized adjustments to enhance prediction quality.

First-order sensitivity indices obtained using the Sobol method can be used to assess the linearity of a model. In the case of fully linear models, the sum of these first-order sensitivity indices is equal to one. In models with significant interaction between the parameters, the sum of these indices tends to approach one, while highly non-linear models have a sum close to zero of these first-order indices (Frey et al. 2004).

A total of 72 iterations were performed to investigate the relevance of the parameters considered in the analysis. After completing these iterations, the corresponding results were obtained and are presented in Table 2, together with the sensitivity and order of the parameters: perco, bd, awc, surlag, cn2, z, slope, canmx, k, lat_ttime, esco, alpha, epco, flo_min, revap_co, ovn, and revap_min. Observing Table 2, we can see that the sensitivity of the parameters varies considerably. Some parameters have a higher sensitivity, indicating that small changes in their values can significantly impact model outputs. On the other hand, other parameters demonstrate a lower sensitivity, suggesting that their variations have a less expressive effect on the model's responses. The percolation coefficient is a parameter that indicates the rate of water infiltration into the soil, and its sensitivity is because it directly influences the amount of water that is absorbed by the soil and the amount that becomes runoff. The apparent density parameter of the soil affects its capacity to store water, and changes in this parameter can impact the retention capacity and release of water by the soil.

The depth parameter of the soil layers is also relevant, as it influences the storage and release of water by the soil profile. The vegetation absorption compensation factor is a parameter responsible for adjusting plant evapotranspiration, directly affecting soil water availability and river flow. The available water capacity in the soil has also proved to be a fundamental parameter since it defines the maximum amount of water the soil can retain and supply to plants and watercourses. Finally, the curve parameter number for condition 2 is an important indicator to represent water infiltration in areas where the soil is covered by denser vegetation. It is important to emphasize that the order of relevance may vary depending on the context and specific objectives of the study. By understanding the relative importance of each parameter, we can optimize model calibration and validation and direct our efforts toward obtaining more accurate and reliable data for the most influential parameters.

After concluding the sensitivity analysis, the input parameters that exert the most significant influence on the model output were identified. Based on this information, starting the model calibration step is possible using the Dynamically Sized Search (DDS) proposed by Tolson & Shoemaker (2007). This step aims to adjust the values of each input parameter considered sensitive in the modeling.

According to Tolson & Shoemaker (2007), parameter calibration can be performed in three different ways: (a) Substitution: In this method, the value of the parameter in question is changed throughout the study area to the new informed value. (b) Relative Value: The parameter is modified from the original value according to the increase or decrease provided in this approach. (c) Percentage: This strategy adjusts the parameter value according to the informed percentage, allowing a proportional variation.

According to Abbaspour (2015), it is essential to carefully consider the selection of each parameter change method since each option has a unique effect on the resulting model. For example, the ‘Replacement’ approach, which replaces the parameter value across the entire study area, is not recommended for spatial parameters such as available soil water capacity, curve number for condition 2, and depth of soil layers, which is intrinsically related to the type of land use. According to KLEMES (1986), it is recommended to divide the complete data series into two parts: one for calibration and the other for validation. In the initial phase, input parameter values are updated through multiple iterations until the model response is consistent with actual flow measurements. The calibration period adopted in this study covered ten years, from January 1, 2005, to December 31, 2014. To obtain more accurate results, it was decided to consider the years 2003 and 2004 as a warm-up period, thus allowing the filling of information in the model without generating significant results.

It is essential to point out that the choice of these non-sensitive parameters in the sensitivity analysis is due to their lesser influence on the model results. Although these parameters can play a significant role in hydrological processes, the analysis has shown that their variations have a relatively low impact on the model output. Therefore, they were kept with fixed values since changing them would not significantly affect the accuracy of the simulations performed. Although the alpha, esco, and flo_min parameters had a relatively low influence on the sensitivity analysis results, they still play an important role in hydrological processes. Their consideration in the calibration is essential to guarantee the accuracy of the simulations. In the SWAT+ Toolbox Software, modifications were made to seven parameters of the model, which are essential for calculating the hydrological processes in the basin. These parameters are alpha factor for deep aquifer recession curve, soil available water capacity, curve number for condition 2, soil evaporation compensation factor, threshold depth of water in shallow aquifer required for return flow, percolation coefficient, and depth of soil layers.

Here, we have: Q is a flow variable, m represents the mean value, and s represents the simulated value.

Table 3 presents the values obtained for the parameters after the calibration process. These values represent the optimized parameter setting to ensure a better match between the SWAT+ model simulations and the observed data.

The results of the SWAT+ model performance evaluation in the flow estimates for the basin under study, considering the calibration period and the monthly time steps. The results presented in the table indicate that the SWAT+ model obtained a performance considered ‘very good’ according to the classification proposed by Moriasi et al. (2007) for all three coefficients analyzed in estimating monthly flows for the basin under study, RSR equal to 0.405, NSE equal to 0.834 and PBias equal to −2.542.

These results demonstrate the ability of the SWAT+ model to reproduce the hydrological behavior of the basin under study. However, it is important to emphasize that the evaluation of the model's performance must also consider other factors, such as the consistency of the input data, the quality of the parameters used, and the adequate representation of the hydrological processes in the basin.

In the validation step, the procedure is like the calibration step. Input data are provided to the model, and the model response is compared to the actual measurement data. However, in this step, the second portion of the data series is used, and the values of each parameter are kept fixed, being determined from the calibration iterations that presented the best results.

For the validation of the model, the climatic values and flows from January 1, 2015, to December 31, 2019, were considered, totaling five years. It is important to note that 2013 and 2014 were designated as a warm-up phase, a period for adjusting and filling in information in the model without generating significant results. The results demonstrate a performance considered ‘very good’ according to the classification proposed by Moriasi et al. (2007). In addition, the model performed even better than the calibration analysis results for the three analyzed coefficients, with the RSR equal to 0.366, NSE equal to 0.864, and PBias equal to −6.606.

These results indicate that the SWAT+ model could accurately reproduce the observed flow patterns, validating its effectiveness in hydrological simulation. The ‘very good’ performance of the model is indicative of its ability to provide reliable flow estimates. Therefore, analysis of the performance of the SWAT+ model during calibration and validation provides a solid basis for confidence in the flow estimates generated by the model. These results are essential to support decision-making related to managing and planning water resources in the Paraguay River basin, contributing to a more efficient and sustainable management of water resources.

At the Cáceres monitoring post (DNPVN), the annual time series of the flow observed during the period from January 2005 to December 2019 showed an average value of 544.21 m³/s, a maximum monthly average value of 1,059 m³/s in March and a minimum average value of 231.78 m³/s in September. While in the simulated flow series for January 2005–December 2019, the annual average value was 565.37 m³/s, with a maximum monthly average of 983.06 m³/s in February and a minimum average of 145.38 m³/s in September. We registered higher flow values from December to May, while the months from June to November presented lower flow values.

In the flow simulation using the MIROC 4.5 model, an annual average value of 340.94 m³/s was found, a maximum of 653.70 m³/s in March, and a minimum value of 99.9 m³/s in September. Again, the months from December to May registered higher flow values, while the months from June to November presented lower values. Thus, the simulated results revealed an average monthly reduction in flow of 37.35%. Reduction of 38.27% in the average of the month with the highest flow and reduction of 43.48% in the month with the lowest average.

According to previous studies in the region, according to UFPR/ITTI (2015), floods in Cáceres occur mainly between January and April, with the most recurrent peak in March. Droughts are observed from July to November, with lower values identified in September. Due to its extension and low slopes, the Pantanal floodplain can dampen and delay the flows, resulting in a delay of three to 4 months in the peak of the flood in Corumbá. Therefore, the simulations carried out in this study align with previous studies in this region.

This information is of paramount importance to understand the hydrological behavior of the Paraguay River basin under different climate change scenarios, allowing a comprehensive assessment of possible variations in flow patterns over the years. Identifying the months with the highest and lowest average flow and their respective reductions is essential to support management strategies and decision-making related to the sustainable use of water resources in this basin.

The impacts arising from decreases in simulated flows according to different climate scenarios can be considerable for the Paraguay-Paraná Waterway. These reductions may affect the navigability of the waterway, hampering the transport of goods and impacting associated economic activities. In addition, such changes may have socio-environmental impacts, including reducing water availability for supply, agricultural activities, and hydroelectric generation. Given this context, it is essential to consider the effects of climate scenarios and land use on management and decision-making related to the Paraguay-Paraná Waterway. Adopting adaptation strategies and mitigation measures becomes essential to minimize the negative impacts of reductions in simulated flows.

The use of geomatics allowed a detailed understanding of the hydrological processes in the basin, but the lack of data integration still poses a challenge. Building a unified database is crucial for more efficient management of water resources and informed decision-making. The SWAT+ model proved appropriate and well-calibrated for the Upper Paraguay Basin in the Pantanal, providing reliable results in scenario simulations. The projection of future water availability based on different climate scenarios offers important subsidies for managing water resources and the operation of the Paraguay-Paraná Waterway. The reduction in the flow of the Paraguay River in unfavorable scenarios can impact navigation on the waterway and harm ecosystems and populations dependent on water resources. Adaptive measures and integrated management are needed to minimize the negative impacts of climate change. The uncertainty inherent in scenarios highlights the importance of considering multiple future possibilities and exploring other approaches to improving hydrological analyses. The results presented are relevant for decision-making and formulating water resources management policies in similar regions, focusing on sustainability and facing climate challenges.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.