Spatial Analysis to Retrieve SWAT Model Reservoir Parameters for Water Quality and Quantity Assessment

Abstract

Owing to their capacity to conserve water and regulate streamflow, small reservoirs are useful for agriculture, domestic water supply, energy production, industry, flood control, recreation, fisheries, and ecosystem conservation. The presence of these small reservoirs often affects the natural water pathways, but the use of a hydrological model such as the Soil and Water Assessment Tool (SWAT) can help to better apprehend these effects at the watershed scale. Indeed, the SWAT model allows modelers to represent and operate reservoirs by inputting the related parameters while setting the model. However, these reservoir parameters are not automatically generated by the SWAT model algorithms. Subsequently, SWAT users are left alone and must sort out the adequate approach to separately obtain or determine the reservoir parameters. Traditionally, reservoir parameters such as the volumes and surface areas are obtained through in situ hydrographic surveys which are costly and labor demanding. To help SWAT modelers retrieve the input parameters needed for modeling small reservoirs, this paper explicitly presents a spatial analysis procedure using the case study of a small watershed reservoir. In this procedure, the digital elevation model of the watershed is transformed into a triangulated irregular network and turned into contour lines which are used to identify the reservoir surface and volume at the principal and emergency spillways. The retrieved parameters were successfully used to calibrate and validate SWAT simulations of the watershed hydrological behavior. The spatial analysis procedure reported here is a cost-effective alternative to traditional in situ hydrographic surveys and it is useful for addressing watersheds with small reservoirs. The procedure eases the inclusion of reservoirs in SWAT and reduces the risk of model overfitting. Furthermore, the procedure could be useful for developing reservoir elevation–capacity–area curves.

1. Introduction

At the watershed scale, the water cycle is governed by multiple and complex bio-physical factors including climatic, topographic, edaphic, biotic, and anthropologic. With advances in computer science, hydrological models were developed to reproduce the circulation of water and quantify its environmental and societal ramifications. Today, the Soil and Water Assessment Tool (SWAT) model is considered the leading computer-based hydrological model as it has been widely used across the globe and has shown relevant applications in water resources, agricultural, environmental, and climate change fields [1,2]. Basic applications of the SWAT model for reproducing natural flow require watershed scale data including geospatial (i.e., soil, land cover, topographic), climate, discharge quantity and quality [3,4]. However, natural flows in many watersheds are being affected by small reservoirs often constructed for various purposes. Indeed, small reservoirs are useful for agriculture, domestic water supply, energy production, industry, flood control, recreation, fishery, and ecosystem conservation [5]. Nonetheless, the presence of these small reservoirs often causes disturbances to the hydrologic regime. Explicitly, the presence of small reservoirs in a watershed can have a significant influence on sediments, nutrients, and water transport [6,7]. Yet these small reservoirs are often overlooked in hydrological modeling due in part to their small sizes. Because reservoirs are multipurpose structures, they can be man-operated or function under a free-flow regime. With the SWAT model, a reservoir module based on a target release method allows modelers to input target reservoir storage for each month especially for man-operated reservoirs [8]. Nevertheless, Wang et al. [6] noted poor performance of this module and reported a generalized module that offers more flexibility to grasp the range of flow fluctuation. However, small reservoirs operating under free-flow regimes do not require the entry of target monthly reservoir storage and their simulations accuracy is therefore highly dependent on the SWAT reservoir parameters [7,9]. Even though the interface of the SWAT model allows modelers to include reservoirs and ponds, the required input parameters are not automatically generated by the model algorithms. In practice, data on the ponding capacity, spillways, and surface areas are not often available for these small reservoirs because of their small sizes, multiplicity, and diverse ownership. As a result, SWAT users must sort out the adequate approach to obtain these reservoir parameters to achieve acceptable watershed simulations. Traditionally, in situ hydrographic surveys can be used to determine reservoir parameters such as the volumes and the pond surface area [10,11]. However, in situ hydrographic surveys are costly and labor-demanding. Particularly, when a watershed contains numerous small reservoirs, it is impractical to expect using in situ surveys. Furthermore, reservoirs are often situated in inaccessible areas suggesting the importance of developing remote sensing approaches for estimating their storage volumes [12]. Hence, this paper reports on a spatial analysis procedure that enables users to remotely retrieve small reservoir parameters for the SWAT model application.

The input parameters needed to represent a small reservoir in the SWAT model can be obtained through in situ topographic and hydrological surveys. However, such in situ surveys are labor and resources demanding and the physical accessibility to these small reservoirs could also represent a major limitation. Alternatively, the spatial analysis procedure reported in this paper enables a remote estimate of small reservoir parameters for use in SWAT modeling. The procedure is cost-effective, and it would help SWAT modelers retrieve and easily incorporate small reservoir parameters. The representation of the small reservoirs is critical for modeling the hydrologic behavior of the watershed and capturing the influence of the reservoirs on the flow regime [7]. In addition, it prevents model overfitting during the calibration stage. Especially for the SWAT model, the inclusion of reservoir parameters is needed for reasonable simulations of hydrological processes [7]. To ease replicability, this paper reports on the small reservoir parameter retrieval procedure using a case study where the SWAT model was applied to simulate the monthly discharge of a small watershed in North Carolina. The content of this paper complements recent SWAT modeling studies reported by Sohoulande et al. [3,13]. Explicitly, this paper reports all the steps of the procedure which may ease modelers’ assessment of watersheds characterized by the presence of small reservoirs.

2. Data and Model

2.1. Data

In this paper, the Herrings Marsh Run watershed is used to exemplify the spatial analysis procedure for retrieving small reservoir parameters. Herrings Marsh Run is a small watershed in North Carolina (Figure 1) referenced with the United States Geological Survey (USGS) hydrological unit code 30300070206. Water from this watershed discharges through the stream-gage USGS 0210783240 (Latitude 35.10°, Longitude −77.93°, Elevation 28.96 m) which flow records are used for calibrating and validating SWAT in the case study. Explicitly, the monthly flow records of the time slices 1993–1996 and 1997–1999 were used for SWAT calibration and validation, respectively. To achieve the flow modeling, daily climate data including precipitation, and maximum and minimum air temperature were obtained from the National Oceanic and Atmospheric Administration stations USC00311881 (latitude 35.02, longitude −78.28), USC00314684 (latitude 35.19, longitude −77.54), and USW00013713 (latitude 35.34 and longitude −77.96). In addition, the study used geospatial data including a 10 m resolution digital elevation model (DEM), a 10 m resolution soil map, and a 30 m resolution of land use land cover obtained from the USGS database, the United States Department of Agriculture’s Natural Resources Conservation Service database, and the Multi-Resolution Land Characteristics Consortium database, respectively [14]. These data sources have been widely used in scientific studies for their data quality and their public availability. Additional details on this swat modeling study are reported by Sohoulande et al. [3,13]. The studied watershed drains approximately an area of 4.09 km² and it includes a small reservoir that delays the free flow toward the watershed’s outlet (Figure 1c,d).

2.2. SWAT Model

The practical use of the SWAT model in hydrological simulation is sustained by a wide range of studies [1]. However, the achievement of acceptable model simulations depends significantly on the quality of input data [15]. Particularly with the SWAT model, a comprehensive representation of the watershed is needed to achieve realistic simulations and avoid model overfitting scenarios. In the case of watersheds containing small reservoirs, such as the Herrings Marsh Run (Figure 1c), the input of SWAT reservoir parameters enables the representation of the hydrological processes related to the presence of the small reservoirs in the watershed. As per SWAT manual guidelines [16], the basic input reservoir parameters for the SWAT model include the reservoir surface area and volume when filled into the emergency spillway, the surface area and volume when filled into the principal spillway, and the initial volume. Unfortunately, these input parameters are not automatically generated by the model algorithms and the modeler must separately obtain and input the reservoir parameters. An explicit list of the SWAT reservoir parameters and the values determined in this case study of the Herrings Marsh Run are reported in the section describing the spatial procedure. Once the reservoir parameters were input into SWAT, the model was calibrated and validated for monthly discharge simulation. As reported by Sohoulande et al. [3], the SWAT’s calibration and uncertainty programs [17] were first used to analyze parameter sensitivity, then an iterative manual calibration was employed to refine the parameter values that yielded reasonable flow simulations. As reported by Sohoulande et al. [3], the parameter sensitivity analysis was conducted using auto-calibration algorithm SUFI-2 of the SWAT’s calibration and uncertainty programs (SWAT-CUP). The final parameter values used for the watershed flow calibration are presented in

Table 1. SWAT model parameter values used for the Herrings Marsh Run’s flow simulation. Flow parameter values are differently set for the hydrologic response units (HRUs). Details are reported by Sohoulande et al. [3].

Parameter	Description	Default Range	Range Values Used at Calibration and Validation
CN2	Soil Conservation Service runoff curve number for moisture condition II	35–98	25–92
SOL_AWC	Soil available water capacity	0–1	0.05–0.46
SOL_K	Saturated hydraulic conductivity	0–2000	51–1155

. The overall model performance was evaluated using indicators such as Nash–Sutcliffe’s Efficiency (NSE) (Equation (1)), the index of agreement (d1) (Equation (2)), the coefficient of determination (R2), and the Root Mean Squared Error (RMSE) (Equation (3)) [18,19].

$$NSE=1-{\displaystyle \frac{{\displaystyle \sum _{i=1}^n{(Q_i-q_i)}^2}}{{\displaystyle \sum _{i=1}^n{(Q_i-\overline{Q})}^2}}}$$

(1)

$$d_1=1-{\displaystyle \frac{{\displaystyle \sum _{i=1}^n\left|q_i-Q_i\right|}}{{\displaystyle \sum _{i=1}^n\left(\left|q_i-\overline{Q}\right|+\left|Q_i-\overline{Q}\right|\right)}}}$$

(2)

$$RMSE={\left[n^{-1}{\displaystyle \sum _{i=1}^n{\left(Q_i-q_i\right)}^2}\right]}^{0.5}$$

(3)

With n representing the number of months during the calibration or validation time slice, Qi and qi represent the observed and simulated discharge during the ith month 1 ≤ i ≤ n, respectively, and $\overline{Q}$ is the average monthly discharge for the given time slice.

3. Spatial Analysis Procedure

In this study, the reported spatial analysis procedure uses geospatial data to estimate the needed reservoir parameters for SWAT modeling. These parameters include the water pond surface area and volume, and they are computed using the ArcGIS 10.7 software. Explicitly, the watershed’s DEM and the reservoir geo-locations are utilized to estimate the pond’s volume and surface area when the reservoir is filled at principal and emergency spillways.

3.1. Step1: Estimating Reservoir Surface Area

The geospatial data including the DEM of the watershed and the geolocation (i.e., latitude, longitude, and elevation) of the reservoir spillways are needed. In this case study, a 10 m resolution DEM of the Herring Marsh Run watershed was used. To determine the geolocation of the reservoir spillways, open access tools such as Google Earth Imageries can be helpful. Indeed, Google Earth compiles high-resolution georeferenced imageries, allowing users to visually identify features extend and edges in the landscapes [20,21]. Hence, Google Earth Imageries were used to geolocate the reservoir spillways in the Herrings Marsh Run watershed. Using the 10 m DEM of the Herrings Marsh Run watershed, 0.5 m gradient contour lines were generated (Figure 2a). The created contour lines were used along with the spillways’ geo-reference to extract two specific contours. The first contour line corresponds to the reservoir surface edge when filled to the principal spillway and the second contour line corresponds to the reservoir surface edge when filled to the emergency spillway (Figure 2b). Using these two contour lines as benchmarked area edges, geo-referenced polygons of the reservoir surface at principal and emergency spillways were generated. Using ArcGIS computation tools, the resulting polygons were separately utilized to estimate the reservoir surface areas at principal and emergency spillways.

3.2. Step 2: Estimating Reservoir Volume

The computation of a pond volume using ARGIS requires two inputs including the digital polygon of the pond surface area and the digital land surface curvature of the terrain containing the pond [22,23]. In general, the triangulated irregular network (TIN) DEM of the watershed is used as the digital land surface curvature of the terrain. The TIN uses adjacent triangular facets and vertices to represent continuous land surfaces [18]. In the present case study, the 10 m resolution DEM of the Herring Marsh Run watershed (Figure 3a) was transformed into a TIN’s DEM with an altitude tolerance of 0.5 m (Figure 3b). Finally, the Herrings Marsh Run watershed’s TIN and the digital reservoir surface polygons were inputted into ArcGIS to calculate the reservoir volume when filled at principal and emergency spillways. Figure 4a,b highlight features of the reservoir in the Herrings Marsh Run’s landscape.

Table 2. Reservoir and lake water quantity/quality inputs used to represent the reservoir in the SWAT model. Details are reported by Sohoulande et al. [3,9].

Input Parameters for Reservoir (.res)	Definition/Description	Input Value
RES_ESA	surface area when filled to emergency spillway (ha)	4.52
RES_EVOL	volume when filled to emergency spillway (104 m3)	2.54
RES_PSA	surface area when filled to principal spillway (ha)	3.04
RES_PVOL	volume when filled to principal spillway (104 m3)	0.6
RES_VOL	Initial volume (104 m3)	0.6
RES_K	hydraulic conductivity (mm/hr)	8
IRESCO	Outflow simulation code (0 = uncontrolled reservoir)	0
RES_RR	Average daily principal spillway release rate (m3/s)	0.06
EVRSV	evaporation coefficient	0.6
WURESN	Average amount of water withdrawn each month for consumptive use (104 m3)	0

reports the estimated reservoir parameters setting used for SWAT modeling of the Herrings Marsh Run watershed.

4. Results and Discussion

Using the procedure, the reservoir parameters were successfully estimated and utilized for setting the SWAT model. The model was calibrated and validated for stream discharge. Figure 5 compares the simulated flow to the observed flow during the calibration and validation time slices. In addition,

Table 3. SWAT model performance evaluation during calibration and validation periods. Details are reported by Sohoulande et al. [3,9].

Efficiency Indicator	Calibration Stage	Validation Stage
NSE	0.69	0.83
d1	0.70	0.67
R2	0.71	0.85
RMSE	0.03	0.05

reports the SWAT model performances in simulating the Herring Marsh Run watershed’s hydrological behavior. As highlighted by Sohoulande et al. [3,13] the efficiency indicators values of NSE ≥ 0.69, d1 ≥ 0.67, R2 ≥ 0.71, and RMSE ≤ 0.05 m³/s (Table 3) suggest acceptable model simulations during the calibration and validation stages as per the guidance of Moriasi et al. [18]. Likewise, Sohoulande et al. [3] reported reasonable water quality simulation with d1 = 0.76, R2 = 0.52, and RMSE = 17.4 kg P for soluble phosphorus (P) outflow validation. Figure 6 highlights SWAT simulation of total nitrogen (TN) and total P (TP) when the reservoir parameters are included or not over the period 2001 to 2020. The graphs in Figure 6 highlight the delaying effect of the reservoir on TN and TP outflows outside the watershed. Sohoulande et al. [3,13] asserted the positive influence of this flow delay on water quality downstream. The inclusion of the reservoir parameters is essential to achieve a realistic watershed simulation given the critical role of the reservoir in delaying the discharge [13]. Indeed, there is a risk of overfitting the model when critical physical components such as reservoirs and ponds are not taken into consideration. In this case, the reservoir was represented and the effects on the streamflow and even the nutrient flow were successfully simulated as reported by Sohoulande et al. [3,13]. Note that the estimated volume of the reservoir when filled at the principal spillway is closer to the 29,000 m³ reported a decade ago by Novak et al. [24]. The performance of the model in simulating the Herrings Marsh Run watershed’s hydrological behavior sustains the relevance of the spatial procedure used in this study. Hence, the procedure could be recommended as a cost-effective alternative to traditional in situ hydrographic surveys used for estimating reservoir volumes and areas [10,11].

In many cases, the site locations of small reservoirs are inaccessible, making their in situ hydrographic surveys impractical [12]. In such a condition, this spatial procedure can be useful for estimating or monitoring the volume and surface area of reservoirs. Even though the primary goal of this manuscript is to support reservoir parameters retrieval for SWAT modeling, it is evident that the procedure can find applications beyond the scope of the SWAT model. Indeed, Li et al. [25] and Ouma [26] reported DEM-based approaches for estimating reservoir storage capacities. Even though their approaches used DEM, the concept and goal are quite different from the present spatial procedure whose aim is to retrieve needed reservoir parameters for SWAT modeling. However, with additional data on the reservoir water depth or the water surface altitude, the spatial procedure can be used to estimate the reservoir volume and surface area for different values of water depth. In practice, the reported procedure can be useful during a reservoir design phase to develop elevation–capacity–area curves that would help to identify the right spillway elevation. In such a situation, the procedure could be applied to different values of spillway elevation to determine the corresponding pond volumes and areas. These elevation values and pond volume–area estimates will be used to generate reservoir elevation–capacity–area curves. Information from such curves is useful for the reservoir designer to assess watershed upstream areas that could be flooded during the reservoir operation as well as the storage capacity [25,27]. The outcomes can henceforth be used to monitor the reservoir water capacity and flooding areas resulting from the rise in water levels. Such potential application of this spatial procedure aligns with Fassoni-Andrade et al. [28] and Gao et al. [10], who highlighted the use of remote sensed-based approaches for quantifying reservoir storage capacities and monitoring flooding events. Ultimately, the spatial analysis procedure reported in this paper would be helpful for SWAT modelers and useful for reconstructing the influence of small reservoir on water quantity and quality changes at the watershed scale [3,5,29]. However, the reported procedure also presents some limitations as it relies on the availability of terrain DEM products. Indeed, the procedure application using fine-resolution DEM products would enhance the accuracy of reservoir volume and area estimates. For instance, a reservoir volume estimate using a 1 m resolution DEM would be closer to the actual reservoir volume compared to an estimate using a 30 m resolution DEM. However, fine-resolution DEM products are only available for limited areas across the globe [30,31].

5. Conclusions

This paper explicitly reports a spatial procedure useful for retrieving reservoir parameters for SWAT modeling. The procedure was applied to the Herrings Marsh Run, a small watershed containing a reservoir. Estimates of the reservoir volume and surface area at principal and emergency spillways contributed to a successful SWAT simulation of the watershed hydrological behavior as reported in previous studies. Ultimately, the spatial procedure is shown to be relevant for the retrieval of needed reservoir parameters for SWAT modeling and a potential cost-effective strategy for quantifying reservoir storage capacities. However, future research projects are encouraged to quantify the influence of DEM resolutions and reservoir sizes on the parameter estimates and the SWAT model performance.