Groundwater modelling quality in the cold region of the Athabasca River Basin

Groundwater is a vital resource for human welfare. However, due to various factors, groundwater pollution is one of the main environmental concerns facing. Yet, it is challenging to simulate groundwater quality dynamics due to the insufficient representation of nutrient percolation processes in the soil and Water Assessment Tool model. The objectives of this study were extending the SWAT module to predict groundwater quality. The results proved a linear relationship between observed and calculated groundwater quality considering No 3 and TDS with R 2 , NSE and PBIAS values in the satisfied ranges, albeit underestimation and overestimation were observed due to limited data availability. These results highlight that nitrate and TDS concentrations and variability in groundwater may used as a tool in surface water quality that have to be assumed for designing adaptive management scenarios. Hence, extended SWAT model could be a powerful tool for future regional to global scale modelling of nutrient loads supporting effective surface and groundwater management.


Introduction
Groundwater is a vital resource for sustainable social and economic development around the world [1,2]. Stored groundwater is generally purified and filtered during infiltration through natural soils and sediments [3,4]. Therefore, the quality and quantity of groundwater storage tends to be more stable than surface water. Thus, majority of people in the world primarily depend on groundwater for drinking water [5]. Groundwater offers drinking water for > 1.5b people and supports approximately 40% of agriculture in the form of irrigation [6]. Therefore, protecting groundwater resources, including efficient use and conservation measures, are an important strategy of water resources plan in both developing and developed countries.
However, natural processes, anthropogenic activities, and climate change significantly influence the quality and quantity of groundwater. In various area of the world watersheds, lakes, rivers, wetlands and the associated ecosystems have experienced impacts; thus, the vitality, availability, quality and quantity of these water resources face serious threats. Elevated concentrations of chemical elements and biological constituents exist in the environment, and depending on geo-environmental backgrounds, water pollutants may exhibit spatial and temporal variations [7,8,9]. Anthropogenic processes, such as discharge of untreated sewage water to water environment. [30,31,32] coupled the SWAT model with the three-dimensional groundwater flow model (MODFLOW) to represent groundwater flow. The main drawback of the coupled model is that the users must establish numerously formatted variables that properly fit [33]. Although these coupled SWAT-MODFLOW models enable simulation of groundwater recharge, aquifer evapotranspiration and groundwater levels, it is a big challenge to find a way to dynamically simulate water quality when coupling SWAT source code with MODFLOW code because the two models may have different definitions, formats and arrays of water quality variables. Therefore, such a coupled approach of the two different models requires modification of core codes, such as definitions of variables and its arrays, to be appropriate for simulating ground water quality [34].
Although a plethora of models have been developed in the past few decades, most of the modelling experiences have focused mainly on understanding on the spatio-temporal groundwater storage.
Appraisal of distribution and mitigation of chemical elements in the groundwater using SWAT model is still lacking [5].
Economic activity and human settlement in the Athabasca river basin (ARB) are being varied, and the basin is culturally vibrant and diverse as the homes for more than 150,000 residents with 13% of Aboriginal peoples [35]. However, increasing the growth and intensification of urban development, agriculture, recreation, forestry, conventional and in situ oil and gas, as well as mining can negatively affect the health of the river basin. Developing reliable groundwater quality model as a tool could provide a very useful insight on the transport and potential mitigation of nonpoint source pollution into groundwater in river basins, which are especially important in assessing the pervasive high nutrients loadings from fertilization and manure application. The main objectives of this study are to: (i) integrate a groundwater nutrient module for nitrate (NO3) and total dissolved solid (TDS) in the SWAT model for the Athabasca River Basin; (ii) assess the effect to NO3 and TDS on groundwater quality status; and (iii) evaluate SWAT model sensitivity and uncertainty analysis to understand the possible limitations, and recommend forthcoming directions in model formulation efforts.

The study area
The research has performed in ARB, which is found in the central part of the province of Alberta as shown in Fig. 1. The area is a cold climate region. The river basin in general have substantial economic contribution for the area by providing reliable water supply to the people as well as for various industries, such as oil sands mining and pulp mills [36]. The main land cover type of the ARB is forest, which shared 82% of the total land; agriculture share took 9.5%, which is located in the central portion of the watershed (i.e., Pembina, Lesse Slave, McLeod and upper parts). Overall forest, agriculture, traditional oil and gas extraction, oil sand mining, and coal mining are the major industries of ARB [35]. Mean annual precipitation of the area ranges from 300 mm in the lower portion of the river basin to > 1000 mm at the headwaters, while the mean temperature 1.8°C -5.1°C [37]. The following major aquifer types, viz., near-surface sands, buried channels and valleys, Paskapoo aquifers, and bedrock aquifers characterize the basin with primary and/or secondary porosity [36]. Groundwater contributions to stream flow in the Athabasca River are also mentioned in reference to Instream Flow Needs assessments [38]. In-situ oil sands operations typically occur where soil sand deposits are present at the depth, which is greater that 150m deeper, and the main types of operations exists in the southern Athabasca Oil Sands region. In the area, groundwater continues within unconsolidated surficial deposit consist of a mixture of preglacial and glacial sand and gravel aquifer.

Soil and Water Assessment Tool (SWAT) description
The SWAT model is a river basin scale model, which is developed in order to quantify the impact of land management practices in large, complex watersheds that operates on a continuous daily time step [39,40]. It is employed to simulate the hydrological process, nutrient concentration, runoff generation, and sediment yield in the watershed under various landuse as well as soil scenarios [39,41,42]. Additionally, it is used for simulating the effects of climate change, landuse and management practices on the quantity and quality of water [43].

Soil and Water Assessment Tool Application
SWAT model can simulate the N2 cycle of soil profile and in the aquifer at shallow depth [44,45]. In the water and soil, the condition of N2 is very exists in a various form of dynamics. It could be added to the soil in the form of manure, bacteriological fixation, residue, and rainfalls. It Letter moved out of the shallow aquifer with water moving into the soil zone during water deficiencies and hence recharge into deep aquifer. The amount of nitrate carried by the water is obtained by multiplying the concentration of nitrate in the mobile water by the volume of water moving in each route. The discharge of NO3into the tiles is depend on NO3concentration in the soil-water near the tiles. The first level of nitrate, which has the initial form of N2 that uptake by plants, have inverse exponential relationship with the depth [40]. 3 , where NO3conc, z refers to initial nitrate concentration (mg/kg) at the depth of z (mm). Only the parts of NO3is mobile and therefore it available for the discharge through the tiles. To calculate the concentration of nitrate in the mobile water fraction, the following equation has been adopted from [46,47].
where ConcNO3, mobile refers to NO3concentration in the movable water at a given layer (kg N/ha), Wmobile represent the amount of water lost by runoff, percolation or side flow in the layer (mm H2O), e  represent fraction of porosity from which anions are excluded, SAT1y is saturated water content in the soil layer (mm H2O). To obtain the transport of nitrate with runoff, percolation, and lateral flow, the following generic equation has been adopted form [48].
where, NO3 represent the NO3removed by the physical transport mechanisms consider (lateral flow, runoff, percolation)/kg ha-1; βNO3 is the concentration of NO3in the mobile water for the top 10mm of soil/kg ha-1 considering both surface and subsurface lateral flow in the top layer; and Qx is the physical transport mechanism considered (i.e. Qsurf, Qlat, Qperc). NO3percolating to the shallow aquifer from the soil profile may store in the aquifer or move with groundwater flow into the main channel or to the shallow aquifer and recharge the deep aquifer. Organic transport of N with sediment is obtained as a concentration function proposed by [49] and letter employed by [9] for the application of separated runoff events. Estimation of the daily organic N runoff loss is on the basis of concentration of organic N in the topsoil layer, the sediment yield as well as the enrichment ratio: that of the organic nitrogen in sediment to organic N in soil layer [44] (see Fig.   2). In the SWAT model, water quality procedures integrate essential interactions and relationships used in the QUAL2E model [50], which includes the major interactive factors such as the nutrient cycles, benthic oxygen demand, and algae production. The conceptual framework demonstrating nitrate occurrence in the groundwater (adopted from [51].
TDS shows all dissolved chemicals in the water, which is expressed as mg/L, can be obtained by adding all the concentrations from chemical analysis or can be measured as weight of the residue after a volume of water has been evaporated to dryness. Typically, the concentration of TDS increases with the sample depth that groundwater has traveled from the recharge area to the sample sites. High nutrient concentration in the groundwater shows contamination form different sources of pollution [52].

Groundwater module in the SWAT model
Existing SWAT model dividers underground saturated aquifer into shallow as well as deep aquifer. As per [40,53], the balance of water the shallow aquifer can be obtained as follows: 1 ,, , The volume of recharge entering to the shallow aquifer on the day i is obtained using the following i rg sh dp rg In order to represent better simulation results of groundwater modelling, we adopted equation 6 as follows: , ,

SWAT model setup
The study are delineation processes were done in ArcGIS. Based on soil types and landuse classes the hydrological response unit was defined. Therefore, based on similar soil, similar landuse, and slope types, about 1370 HRU were identified for the basins. Weather data viz., temperature, rainfall, wind speed, humidity, and radiation observation were obtained from Alberta Environment and Parks. Groundwater well and water quality data (nitrogen species and TDS) for 1,300 different monitoring stations was obtained from Alberta Environment and Parks as shown in Fig. 1. Data availability is one of the most prominent factors, which affect the model accuracy.

Model performance Metrix analysis
Measured daily groundwater quality data (NO3and TDS) obtained from Alberta, Environment and Parks and were used for the model calibration and validation. Model simulation performances were compared in relation to observed data. In order to perform parameter sensitivity and uncertainty analysis, we used Sequential Uncertainty Fitting2 (SUFI-2) algorithm in the SWAT-CUP [54]. In SWAT-CUP, sensitivity analysis has been used for model sensitivity analysis. To examine the effect of NO3and TDS on groundwater quality, parameters have been chosen and considered during calibration and validation. To evaluate model performance, R 2 , PBIAS and NSE were used [55]. The descriptions of NSE, R 2 , and PBIAS can be found in [56].
The overall model performance evaluations obtained as follows: where Oi refers to the i th observed value; Oavr is the average observed value; Pi is the i th simulated value; Pavr is average simulated value; n is the number of time steps (days in our case).

Performance of the modified bacteria module in SWAT model
To check our modified groundwater module, the SWAT model should be calibrated and validated for the stream flow, temperature, and sediment load. These have been done in our previous work. The interested readers can check for references, such as stream flow [38], temperature [57], and sediment [43]. In this paper, we, therefore, focus on the calibration and validation of groundwater quality modelling.

Sensitivity and uncertainty analysis
In order to perform performance assessment, the entire dataset was divided into the periods

Model Calibration and Validation performance
Model calibration and validation are done after the sensitivity analysis. Here, we calibrated and validated the model automatically in SWATCUP, which was recommended by [58], using groundwater quality data obtained from groundwater monitoring stations at two locations.

Calibration of Nitrate and assessment
Nitrate calibration processes happened in a nitrogen percolation coefficient of 0.5, hence a denitrification threshold of water content and exponential rate of denitrification coefficient of 2.5 (Table 1) daily NO3 concentration (Fig. 3), the concentrations of NO3 were acceptable range as it follows similar trend as the observed values. The percentage BIAS obtained from the observed and calculated daily loads ranked as acceptable to very good for all the selected stations ( Fig. 3 and Table 2).

Total dissolved solids (TDS)
The development of TDS module for groundwater quality detection using the daily TDS concentration from the selected stations at ARB. The calibrated TDS parameters with specific ranges are presented in Table 1. Daily dissolved solid loads were calibrated by correcting the parameters settling to the groundwater. Some of the considered parameters are found to be HRU scale parameters and while the other parameters are considered to basin and scale parameters, which control groundwater TDS concentration. As per the report by Morias et al., (2015) the model performance evaluation criteria for the daily nutrient simulation were used as guideline to evaluate the model performance for the daily TDS loads. As shown in Fig. 4, for the whole periods of simulation were found to be good to very good. While during the validation satisfactory to good during calibration based on R 2 , NSE and PBIAS values as indicated using Table 2. Generally, as per the overall model performance evaluation for the daily TDS loads simulations in the ARB shows the model were capable to capture the loads. On the basis comparison between the calculated and observed daily concentrations (Fig. 4), the TDS loads were acceptable as both shows similar trend. Yet, local inconsistencies were noticed through the river basin. observed in [62]. The other possible sources of overestimation and underestimation of the model simulations were subjected to uncertainties of the input data and observed data. overestimations were observed at some points for ABG0215416 monitoring stations (Fig. 3), while underestimations were found in ABG0215416 and ABG0220481 monitoring stations. The probable reason for model overestimation is the data uncertainty such rainfall and snow distribution, in which the rainfall and snow distribution caused in nutrient concentrations. The Underestimation probably subjected to the uncertainty of the input data and measured data.
Furthermore, the observed NO3 data is not sufficient for continuous daily NO3concentration for the entirely considered period for this study. Therefore, observed data could be sources for the errors. In fact, soil denitrification caused excessive NO3simulation could improved by varying the river basin parameters. However, the existing SWAT model could not represent perfectly the seasonal difference of nitrate. After carefully extending the model, the findings of this study highlight the essential to improve spatial representation of nitrification in the groundwater where its dependability is restricted by setting appropriate parameters at watershed level. Noticeable differences were observed after extending the SWAT model, albeit overestimations and underestimations (Fig. 3) were observed at some points with limited available data.

Uses of SWAT model predictions and future applications
The SWAT predictions and limitations analysis is associated with data availability and structure of the model, able to offers valuable insights related to groundwater condition and nutrient processes at the various spatio-temporal scale at different time scale (i.e. daily, monthly and annually). This "state of the art" groundwater quality modelling at ARB recognized as a vital spot where further in-depth studies may be required. For examples, to better support, river basin management in the nutrient monitoring network could be strengthened and harmonized monitoring in concentrations of nutrient should be enhanced in ARB. Besides recognizing the processes and pathways accountable for groundwater pollution as well as pinpointing the problematic areas of actions could be targeted, hence, the modelling framework presented in our research could support the scenario analysis, as internal responses between groundwater and its determinant parameters are considered. Therefore, SWAT model results could support the development of indicators of water quality parameters with groundwater in a nexus thinking approach with reference to the SDGs of the UNs [62,67]. A limitation of this specific study was limited availability of observed input data, which could assist with better representing the distribution, which resulted for some sort of errors during modeling process in the river basin. However, a reliable sate of the art information of groundwater and nutrient fluxes in the ARB were therefore provided to be used for evaluating the effect of best management practices and used to support to the implementation better management practices.

Consideration about groundwater nutrient pollution
Athabasca River Basin is providing numerous services including industrial, agriculture, fisheries, and drinking water supply. Such intensive water uses put several burdens on groundwater quality influencing the ecosystem and polluting huge parts of the Athabasca River Basin. Our study of groundwater quality modelling allowed in detecting the impact of diffused and point and nonpoint sources of pollutants. Percolation of elevated sources of NO3 -, and TDS into the groundwater and diffuse in the form of lateral flow hence affecting the quality of groundwater.
Particularly, the use of fertilizers, manure, and industrial wastes significantly contributes to the pollution of groundwater.

Conclusions
Groundwater is the precious natural resource for the existence of life on earth. However, various factors, viz., various soil properties, crop development, industrial wastes, and agriculture influenced its quality. It is difficult to predict groundwater quality using the existing SWAT model due to inadequate representation of complex interaction of surface water and groundwater. In our study, we extended the existed SWAT model to better represent groundwater quality through considering water carrying capacity of quality parameters (i.e. NO3and TDS). A systematic calibration and validation techniques of SWAT model has been performed to compare the observed groundwater NO3and TDS fluxes in the ARB. The simulated results illustrate, the modified SWAT model performed well, albeit at some point the concentrations were underestimate and overestimate. As per the model performance comparison, the results reviled that the new groundwater quality model in the SWAT is able to capture the daily nutrient concentrations of groundwater. Thus, the process-based hydrologic groundwater quality model is an effective tool in simulating the groundwater quality dynamics (NO3and TDS) for sustainable groundwater and surface water management in the river basin.

Conflict of interest
The authors confirmed there is not conflict of interest.