Assessment of Climate Change Effects of Drought Conditions Using the Soil and Water Assessment Tool

Abstract

A combination of annual peak water demand due to seasonal population spikes along with small and shallow aquifers has prompted an assessment of the region’s watersheds as operating at a net water deficit. This study uses the Soil and Water Assessment Tool (SWAT) to simulate historical drought conditions in the Chipuxet watershed in Rhode Island, USA. The calibrated and validated model uses the Soil Moisture Deficit Index (SMDI) and Evapotranspiration Deficit Index (ETDI) as well as an Indicators of Hydrological Alteration (IHA) calculation to determine the frequency and severity of historical droughts and to simulate climate change conditions developed through a downscaled climate model selection. The output data for the historical and climate change scenarios were analyzed for drought frequency and severity. Results indicate that water stress will increase in both low-emission (RCP4.5) and high-emission (RCP8.5) scenarios. Additionally, the SMDI and ETDI show that RCP8.5 climate scenarios will have more severe deficits. Finally, IHA data indicate that zero-flow days and low-flow durations increase under all climate scenarios.

1. Introduction

Several rivers and watersheds in Rhode Island (RI) are stressed because of extensive reliance on groundwater sources, particularly in the southern part of RI, and the minimal storage capacity of local aquifers. Also, this region’s being a popular summer tourist destination further amplifies the water stress. In consequence, southern Rhode Island experiences seasonal peaks in water demand. The 2012 Rhode Island Water Resource Board (RIWRB) strategic plan reported inadequate water supply to meet the average and peak seasonal demand the region experiences annually [1]. With hydrological changes caused by climate change occurring in this century, watersheds all over New England will experience increasingly frequent and severe extreme flow events due to warmer, wetter winters, and drier summers [1,2,3,4,5,6]. The Soil Water Assessment Tool (SWAT) [7] analytical model can be used to estimate overall hydrologic responses to climate change with examples both in northeast United States watersheds [2,6,8] and abroad [9]. This study uses Global Climate data downscaled for New England from a study done by Shresta and Pradhanang in 2022 [10] as climate forcings input to SWAT to simulate changes in drought conditions in response to climate change.

Even though drought is a normal part of the climate cycle in all climate regimes, it is one of the more costly natural hazards, with significant and widespread impacts. Droughts can be characterized by their severity, location, duration, and timing [11]. The emergence of droughts can vary; they are caused by a range of hydrometeorological processes that suppress precipitation and/or limit surface water or groundwater availability, creating conditions drier than normal or otherwise limiting moisture availability to a potentially damaging extent [11,12]. These processes can be determined through indicators and indices that identify the severity, location, and duration of droughts. The socioeconomic contexts of the area where droughts occur can severely impact who and what is exposed. These contexts adversely affect agriculture, food, hydropower, health both human and animal, and livelihood security.

Tan et al. (2020) reviewed the application and performance of SWAT in terms of hydroclimatic extremes [13]. Out of over 4000 publications, only 111 studies of SWAT covered hydroclimatic extremes. Conversely, the number of studies that use SWAT as the preferred model to simulate these extremes has increased by a factor of three since 2017. Their final assessment indicated that provided the models are accurate, ecohydrological modeling could be used to simulate and understand the extremes associated with environmental impacts in regions without reliable hydrological observations. Yet, there are significant challenges when using SWAT to model these conditions. The main issue is the need to calibrate and validate SWAT using the time-continuous approach. In many cases, data that continuously display extreme conditions are few and far between [13].

The purpose of this paper is to calibrate and validate a hydrological model of the Chipuxet watershed in southern Rhode Island, USA to low-flow conditions. The validated model used the Soil Moisture Deficit (SMD) and Evapotranspiration Deficit (ETDI) drought indices and the Indicators of Hydrological Alteration (IHA) calculations to identify trends in drought frequency and severity. Finally, climate forcings were applied to the validated model to simulate future climate change impacts on drought conditions. Climate change simulations were analyzed using the indices and IHA to assess changes in patterns.

2. Materials and Methods

2.1. Study Area

The Chipuxet watershed (Figure 1) is located in the headwaters of the Pawcatuck River basin, and covers the towns of North Kingstown, South Kingstown, Richmond, Charlestown, and Exeter in RI, USA. The watershed characteristics are typical of suburban watersheds in the eastern United States, with agricultural, drinking water, recreational activity, and other daily demands affecting water quantity, quality, and recharge and runoff characteristics [1]. The watershed’s drainage area encompasses 65.65 sq. km, and the major river within the catchment is the Chipuxet River, which connects 4 major ponds (Worden, Thirty Acre, 100 Acre, and Larkin) [14,15]. The watershed provides most of the public water supply used for agriculture, recreational use, and drinking water. Furthermore, the Chipuxet River is primarily recharged by groundwater [1,16]. Geospatial statistical analysis of the watershed’s land use (Figure 1) shows that 41.5% of the land cover is urbanized and residential, 35.5% is forested, 11.0% is agriculture, and the rest is wetlands and open water created by the shallow water table.

The connected groundwater aquifer is characterized by stratified drift. Interbedded lenses of sand and gravel make up most of the aquifer with areas of silt and silty sands from Pleistocene glacial streams dispersed throughout. Aquifer tests conducted by the USGS in 1984 suggest that the Chipuxet can support a maximum pumping capacity of 3 million gallons per day (MGD) [14,15]. Future projections of water demand conducted by the RIWRB suggest that the Chipuxet will have an Average Day Demand (ADD) of about 20 MGD. However, the groundwater system is limited in storage and is highly vulnerable to short- and long-term droughts [1]. The vulnerability is increased even further when considering that the aquifer supplies the towns of South Kingston and Narragansett as well as the University of Rhode Island [1,14,15,16,17].

The only stream gauge in the Chipuxet watershed is located halfway between Worden Pond and the river source at an intersection between the Chipuxet River and RI Route 138 (Lat 41.482, Long −71.551). The U.S. Geological Survey (USGS) manages and monitors stream height at the sub-daily time scale and water quality for nutrients and inorganics flowing downstream periodically. In this study, this stream gauge outflow was used to compare against the modeled results during calibration and validation of the model.

2.2. Analytical Modeling

Hydrological models effectively predict the effects of land use/land cover (LULC) and climate change on water resources. They are important tools used to simulate water quantity and quality through integrated approaches [16,17,18]. These models can serve as powerful decision support tools (DSTs), providing valuable information to apply to many questions related to natural resources and watershed management [19,20]. SWAT is a physically based, semi-distributed hydrologic model created by the United States Department of Agriculture (USDA) in 1998 along with Texas A&M University [7,21,22]. Volumes for surface water runoff and infiltration are calculated through the modified soil conservation service (SCS) 1984 curve number method [7,22]. The excess water available after accounting for abstractions and surface runoff, using the SCS curve number method, infiltrates into the soil. Then, to simulate flow through each soil layer, a storage routing technique is used. Within a layer, SWAT simulates saturated flow directly and assumes uniform distribution throughout a layer of soil. The flow between unsaturated layers is indirectly modeled using depth distribution functions for plant uptake and soil water evaporation. Saturated hydraulic conductivity governs the rate of downward flow, which only occurs when field capacity is exceeded and the soil layer below is unsaturated [7,12]. The estimation of evapotranspiration (ET) and potential evapotranspiration (PET) was calculated using the Penman–Monteith method [23], which is internally calculated in the model. Spatial data for creating the SWAT model was accessed through the Rhode Island Geographic Information System (RIGIS) database [24]. This database is publicly managed by the Rhode Island government and other private organizations. The setup of the SWAT model was done through ArcSWAT [7], run on ArcGIS an Environmental Systems Research Institute location software from Redlands, California, and the model inputs include topography, soil characteristics, land use/land cover, and meteorological data (precipitation and temperature). A Digital Elevation Model (DEM) was extracted from LIDAR data collected in 2011 with a 10 m resolution [25]. The land use/land cover dataset was collected from a 2020 LULC survey and projected to a 10 m resolution to match the DEM [26]. Soil data were collected from a soil survey conducted by the Rhode Island Soil Survey Program in partnership with the National Cooperative Soil Survey [27]. The characteristics were then cross-referenced with soil characteristics from the Natural Resource Conservation Service (NRCS) Soil Survey Geographic Database (SSURGO). Meteorological data were gathered from the National Oceanic and Atmospheric Administration’s (NOAA) Global Historical Climatology Network daily (GHCNd) [28].

ArcSWAT subdivides the overall watershed into sub-basins based on the topography and streamflow. These sub-basins are further subdivided into hydrologic response units (HRUs) based on unique land cover and soil characteristics [2,7,8,12,20]. For this study, the Chipuxet watershed was divided into 15 sub-basins and 120 HRUs to accurately capture the diverse land use and soil characteristics of the watershed. Initially, ArcSWAT divided the Chipuxet into 2093 HRUs. Then, thresholds were set on the land use, soil, and slope characteristics based on the heterogeneity of the watershed. To account for heterogeneity and reduce model calculation time, the thresholds of 10%, 10%, and 5% were assigned to land use, soils, and slope, respectively, and ultimately resulted in the definition of 120 HRUs.

2.3. Calibration and Validation of Model

The SWAT Calibration and Uncertainty Program (SWAT-CUP) was used to calibrate, validate, and conduct sensitivity analyses on the model [22]. The calibration and validation algorithm, the Sequential Uncertainty Fitting Version 2 (SUFI-2) [29], was used on daily stream discharge from the only hydrograph in the Chipuxet watershed. Measurement of the performance was determined using three criteria: the coefficient of determination (R²), Nash–Sutcliffe Efficiency (NSE), and percent bias (PBIAS). NSE is a normalized statistic that determines the relative magnitude of the residual variance compared to the observed data variance [30]. Values for NSE range from −∞ to 1, with values closer to 1 signifying better model performance. Typically, a model is satisfactory if its NSE value is at or above 0.50 [2,6,7,22,29,31,32,33], with models estimating daily data having an acceptable range between 0.50 and 0.70 and monthly data between 0.60 and 0.80. PBIAS is the relative percentage difference between the averaged simulated and observed data time series over the total time steps. Minimizing the value of PBIAS is the objective for this performance indicator, and values below 15% are typically accepted.

The initial model spanned 30 years of historical meteorological data ranging from the beginning of 1990 to the end of 2019. Since a significant warm-up time is necessary to start the calibration, a period of 2 years where low streamflow occurred was necessary. Historically, there have been many low-flow periods in the Chipuxet, such as 2002, 2012, and 2014, but the 2015–2016 years showed consistently moderate to severe drought conditions during the summer months [32,34]. Calibration results are shown in Figure 2a,b, and

Table 1. Calibration statistical result values for the Chipuxet produced by SWAT-CUP.

Timestep	R2	PBIAS	NSE
Daily	0.61	11.65	0.60
Monthly	0.69	11.86	0.66

. To validate the model, the extremely dry years of 2019 and 2020 were selected. Validation Results are shown in Figure 3a,b, and

Table 2. Validation statistical result values for the Chipuxet produced by SWAT-CUP.

Timestep	R2	PBIAS	NSE
Daily	0.58	12.50	0.56
Monthly	0.73	12.53	0.69

.

Three hydrologic parameter types were most sensitive when calibrating the daily streamflow. Most were related to soil characteristics and overland flow, with a few groundwater parameters and one snow parameter, as shown in

Table 3. The ten most sensitive parameters during daily streamflow calibration using SWAT-CUP for the Chipuxet watershed. Values are listed from the most sensitive to the least sensitive and showing relative change based on initial values assigned by SWAT.

Parameter	Definition	Value Range	Units
GW_DELAY.gw	Groundwater delay	3.0–5.0	days
CN2.mgt	SCS runoff curve number	−0.58–−0.47	-
SOL_Z.sol	Depth from soil surface to bottom of layer	−0.15–−0.01	mm
SOL_K.sol	Saturated hydraulic conductivity	−0.03–0.1	mm/h
SFTMP.bsn	Snowfall temperature	−1.5–−0.07	°C
OV_N.hru	Manning’s “n” value for overland flow 1	0.93–1.1	-
SMTMP.bsn	Snow melt temperature	1.0–2.0	°C
GW_REVAP.gw	Groundwater “revap” coefficient 2	0.45–0.61	-
GW_SPYLD.gw	Specific yield of the shallow aquifer	0.08–0.16	m3/m3
CH_K2.rte	Effective hydraulic conductivity in main channel alluvium	−0.3–−0.2	mm/h

¹ “n” value is the roughness coefficient for overland flow, ² “revap” is defined as the water loss from evaporation or diffusion upwards in the capillary fringe.

. GW_DELAY is the most sensitive parameter, which dictates the delay the groundwater will show after a precipitation event in days. SOL_K and SOL_Z are soil characteristics in unsaturated flow conditions. SOL_K refers to the hydraulic conductivity of the soil in a specific layer. In this instance, all four soil layers were calibrated using relative change with the same range of values. SOL_Z refers to the depth of each soil layer in millimeters. Like the hydraulic conductivity, all soil layers were adjusted simultaneously using relative change using the same range of values. SFTMP refers to the temperature in degrees Celsius that would permit snow to accumulate, and SMTMP refers to the temperature in degrees Celsius that would cause snow on the ground to melt.

2.4. Climate Change Scenarios

Five climate model runs were chosen based on downscaled model selection for the study area [10]. These models were obtained from the downscaling of Coupled Model Intercomparison Project Phase 5 (CMIP5) climate models from a 100–200 km to 4 km scale. Typical climate models show four different representative concentration pathways (RCPs):

• Mitigation scenario (RCP2.6);
• Medium stabilization scenarios (RCP4.5/RCP6.0);
• Very high baseline scenario (RCP8.5).

The mitigation scenario RCP2.6 is a scenario that requires a decline in carbon dioxide (CO₂) emissions starting in 2020, resulting in zero emissions by 2100. Since this scenario is relatively unrealistic, it was not used in this study. The lower medium stabilization scenario RCP4.5 represents a probable baseline scenario. The higher medium stabilization scenario RCP6.0 shows an emissions peak in 2080 before declining. The high baseline scenario of RCP8.5 is the most aggressive scenario of high emissions through 2100. Because the scenarios end around the year 2100, the scenario time spans were broken up into three 30-year scenarios between the years 2020 and 2100. The short-term scenario starts in 2020 and ends in 2049. The medium-term scenario goes from 2041 to 2070. And the long-term scenario starts at 2071 and ends at 2100 if the data support it. This matches the time span of the 1990 to 2019 30-year historical model run.

Table 4. CMIP model runs including the predicted projection of how the extremes would change due to the increased CO₂ in each RCP.

RCP	CMIP Model	Projection
RCP4.5	GISS-E2-R_r6i1p3 1	Cold
	CanESM2_r4i1p1 2	Warm
RCP8.5	CSIRO-Mk3-6-0_r5i1p1 3	Cold
	IPSL-CM5A-MR_r1i1p1 4	Warm
	HadGEM2-ES_r4i1p1 5	Warm

¹ Goddard Institute for Space Studies climate model [35]. ² Canadian Earth System model 2nd generation [36]. ³ Commonwealth Scientific and Industrial Research Organisation model Mk 3.6 [37]. ⁴ Institut Pierre-Simon Laplace [38]. ⁵ Met Office Hadley Centre earth system model [39].

summarizes the climate change models chosen and their projections of average changes in climatic extremes for the target area.

2.5. Drought Indication and Indices

A drought occurrence can vary depending on where in the hydrologic cycle it is being observed. Therefore, there can be many factors that can indicate its onset. To assess the severity of a drought, federal and state government agencies use drought indices. These use various meteorological and hydrological parameters such as precipitation, evapotranspiration, and other factors combined into one number to create a comprehensive view of conditions as they evolve for quick decision making [12]. Since forests and agriculture dominate the Chipuxet land use, there are three drought indicators that are relevant to the surrounding watershed. The first drought indicator is based on the deficit between evapotranspiration (ET) and potential evapotranspiration (PET) known as the Evapotranspiration Deficit Index developed by Narasimhan and Srinivasan (2005). Based on ET and PET, a water stress (WS) ratio is calculated (Equation (1)):

$$WS=\frac{PET-ET}{PET}$$

(1)

PET and ET values are taken from the daily output from SWAT in the output file generated. WS values range from 0 to 1, with values closer to 1 indicating no evapotranspiration and values closer to 0 showing ET occurring at the same rate as PET. The maximum (max.WS) and minimum WS (min.WS) values were then taken weekly, as well as the median (MWS), to calculate the weekly water stress anomaly (WSA) in the following equations:

$$\begin{gathered} WS{A}_{i,j}=\frac{MW{S}_j-W{S}_{i,j}}{MW{S}_j-minW{S}_j}\times 100,ifW{S}_{i,j}=MW{S}_j; \\ WS{A}_{i,j}=\frac{MW{S}_j-W{S}_{i,j}}{\max W{S}_j-MW{S}_j}\times 100,ifW{S}_{i,j}>MW{S}_j \end{gathered}$$

(2)

where i is the year being calculated and j is the week being calculated. This equation eliminates seasonality and values range from −100 to 100, showing very dry to very wet conditions, respectively, with reference to evapotranspiration. ETDI is then calculated through the following equations to show the severity of the drought in any given week:

$$ETD{I}_1=\frac{WS{A}_1}{50},$$

(3)

$$ETD{I}_j=0.5ETD{I}_{j-1}+\frac{WS{A}_j}{50},$$

(4)

ETDI ranges between −4 and 4, indicating dry to wet conditions respectively, in reference to ET and PET.

A similar approach was used to calculate Soil Moisture Deficit Index, which is based on the deficit between actual and mean soil moisture. Daily soil moisture is taken from SWAT output and is the amount of water in all layers of soil throughout the whole watershed. The mean soil water (SW), maximum (max.SW), and minimum (min.SW) were then extrapolated by week. The soil moisture deficit (SD) for each week was calculated by (Equation (4)):

$$\begin{gathered} S{D}_{i,j}=\frac{S{W}_{i,j}-MS{W}_j}{MS{W}_j-\min S{W}_j}\times 100,ifS{W}_{i,j}=MS{W}_j; \\ S{D}_{i,j}=\frac{S{W}_{i,j}-MS{W}_j}{\max S{W}_j-MS{W}_j}\times 100,ifS{W}_{i,j}>MS{W}_j \end{gathered}$$

(5)

Equation (4) eliminates the seasonality and provides the soil moisture deficit (SD) for a given week. The SD ranges from −100 and 100, showing dry to wet conditions, respectively. To calculate the Soil Moisture Deficit Index (SMDI), the following equations were used:

$$SMD{I}_1=\frac{S{D}_1}{50},$$

(6)

$$SMD{I}_j=0.5SMD{I}_{j-1}+\frac{S{D}_j}{50},$$

(7)

The Soil Moisture Deficit Index (SMDI) has a range between −4 to 4, showing dry to wet conditions, respectively, for the entire soil profile.

Both the ETDI and SMDI were determined for all climate change runs in 30-year increments to show potential drought severity with respect to each index. The average deficit index and number of weeks below 0 and −3 observed in each index were also calculated, indicating the variation of each climate change scenario and future drought trends.

The final indicator was calculated with outflow from the entire modeled sub-basin using an altered Indicators of Hydrologic Alteration (IHA) spreadsheet developed by the Nature Conservancy [40]. Each of the modeled runs, including climate change outflow, was used as the input into the altered spreadsheets to show the statistical alterations that occur in the river. Extreme baseflow conditions, which include the 1-, 3-, 7-, 30-, and 90-day minimum and maximum flows, are taken from moving averages at the calculated time length for every possible time period that is within the water year. Additionally, the frequency and duration of extreme events of baseflow were determined. This will show potential shifts in baseflow during each model run.

3. Results

3.1. Historical Simulations

Drought conditions simulated in historical runs between 1990 and 2019 show average annual precipitation of 726.0 mm with corresponding annual ET simulated at 386.0 mm and PET simulated at 1429.5 mm. This resulted in surface runoff averaging to 40.7 mm. Groundwater flow was simulated at 330.9 mm and 31.2 mm in shallow groundwater and deep aquifer outflow, respectively. Overall historical water stress averages to about 69.5 days annually with annual temperature stress days of 72.5 days. The average weekly SMDI calculation is around −0.11. There were 766 weekly instances wherein the SMDI falls below 0, which is 49% of the 30-year period, and 259 instances below −3, which is 17% of the modeled period. The ETDI weekly average is −0.23, with 856 instances below 0, which is 55% of the modeled period, and 5 instances below −3, which is 0.3% of the modeled period.

IHA calculations of minimums are shown in

Table 5. Calculated IHA minimum flows for the historical model run in cubic meters per second with the corresponding coefficient of variation (CV).

Parameter	Flow (m3/s)	CV (%)
1-day	0.1	103.5
3-day	0.1	103.0
7-day	0.1	99.7
30-day	0.2	87.4
90-day	0.3	66.4

. There were zero-flow days calculated for the historical run. The IHA also calculated the mean number of historical low pulses as 8.6 with a mean duration of 11.1 days.

3.2. Future Climate Change Projections

3.2.1. RCP4.5

The GISS-E2-R simulations (

Table 6. Modeled data from GISS-E2-R RCP4.5 climate projection run. Data presented show average annual for each modeled period.

Period	Precipitation (mm)	ET (mm)	PET (mm)	Runoff (mm)	Shallow AQ (mm)	Deep AQ (mm)	Water Stress (d)
Short-Term	793.1	381.9	1443.8	46.5	386.6	36.3	62.4
Med-Term	812.1	387.5	1473.6	49.0	398.1	37.3	41.3
Long-Term	798.1	381.9	1512.1	50.7	389.2	36.5	54.4

and

Table 7. Calculated IHA minimum flow values and low pulses for modeled GISS-E2-R RCP4.5 climate projections.

Period	7-Day (m3/s)	30-Day (m3/s)	90-Day (m3/s)	Zero Flow (d)	Low Pulses	Low-Pulse Duration
Short-Term	0.1	0.2	0.4	0.2	6.8	14.9
Med-Term	0.1	0.2	0.4	0.2	6.6	14.9
Long-Term	0.1	0.1	0.3	0.1	6.3	15.4

, Figure 4) project that the atmosphere‘s cold extremes will increase with the introduction of more CO₂. The most relevant data for this simulation are presented in the tables below.

The modeled data show annual averages for each data type. Precipitation increases in the med-term, which results in increased outflow and a decrease in water stress. The average of each index also increases in the med-term, with percentages of drought detected to be decreasing as well. Long-term GISS-E2-R projections show a decrease in precipitation, yet there is an increase in runoff flow. The long-term index average increases, showing a decrease in the SMDI below zero but an increase in the SMDI below negative three.

The CanESM simulation used projects that the atmosphere‘s warm extremes will increase with the addition of CO₂. The most relevant data for this simulation are presented in

Table 8. Modeled data from CanESM RCP4.5 climate projection run. Data presented show average annual for each modeled period.

Period	Precipitation (mm)	ET (mm)	PET (mm)	Runoff (mm)	Shallow AQ (mm)	Deep AQ (mm)	Water Stress (d)
Short-Term	1553.2	325.6	1060.5	56.2	1099.2	98.8	99.3
Med-Term	1704.6	332.2	1059.5	63.6	1219.9	109.4	101.4
Long-Term	1899.1	340.5	1063.9	73.0	1375.0	123.0	103.1

and

Table 9. Calculated IHA minimum flow values and low pulses for modeled CanESM RCP4.5 climate projections.

Period	7-Day (m3/s)	30-Day (m3/s)	90-Day (m3/s)	Zero Flow (d)	Low Pulses	Low-Pulse Duration
Short-Term	0.1	0.1	0.5	1.6	2.4	40.5
Med-Term	0.1	0.1	0.7	0.3	2.1	44.7
Long-Term	0.2	0.2	1.0	0.5	2.4	41.1

as well as Figure 5 below.

The modeled data show annual averages for each data type. In this simulation, precipitation continually increases throughout the modeled runs with very little change in ET and PET. Water yield also increases but the water stress days also see a significant increase. There is an overall decrease in all calculated drought indices, in the SMDI and ETDI, which can be attributed to the overall increase in precipitation. On the other hand, IHA calculation shows very long low-pulse durations, with them being low for over 40 days.

3.2.2. RCP8.5

The CSIRO-Mk3-6-0 simulation projects an extreme addition of CO₂ causing an increase in the atmosphere’s colder extremes. The most relevant data for this simulation are presented in

Table 10. Modeled data from CSIRO-Mk3-6-0 RCP8.5 climate projection run. Data presented show average annual for each modeled period.

Period	Precipitation (mm)	ET (mm)	PET (mm)	Runoff (mm)	Shallow AQ (mm)	Deep AQ (mm)	Water Stress (d)
Short-Term	3289.3	238.1	768.4	168.9	2583.7	230.4	82.4
Med-Term	3520.2	247.6	792.7	182.3	2766.0	246.5	86.8
Long-Term	3892.5	261.7	831.0	203.7	3061.4	272.5	93.3

and

Table 11. Calculated IHA minimum flow values and low pulses for modeled CSRIO Mk3-6-0 RCP8.5 climate projections.

Period	7-Day (m3/s)	30-Day (m3/s)	90-Day (m3/s)	Zero Flow (d)	Low Pulses	Low Pulse-Duration
Short-Term	1.3	1.7	3.5	0.1	5.2	19.7
Med-Term	1.6	2.2	3.9	0.3	5.6	18.6
Long-Term	2.3	3.0	4.3	0.0	5.7	18.3

as well as Figure 6 below.

The modeled data show annual averages for each data type. This simulation shows high precipitation with less ET and PET. Most of the outflow seems to go through shallow aquifer output with significant runoff. Water stress days are still above 80 annually, showing extreme conditions. Drought indices show that the SMDI, on average, is below zero and the ETDI averages around 0, with over 80% of the modeled time periods showing drought conditions.

The IPSL-CM5A-MR scenario simulates an atmosphere with significant increase to warm extremes with the addition of CO₂ on the RCP8.5 curve. The most relevant data for this simulation are presented in

Table 12. Modeled data from IPSL-CM5A-MR RCP8.5 climate projection run. Data presented show average annual for each modeled period.

Period	Precipitation (mm)	ET (mm)	PET (mm)	Runoff (mm)	Shallow AQ (mm)	Deep AQ (mm)	Water Stress (d)
Short-Term	3176.0	230.2	743.7	156.2	2501.7	222.7	67.2
Med-Term	3589.1	246.6	787.2	182.1	2827.5	251.7	72.8
Long-Term	4009.5	259.2	815.4	211.8	3160.1	281.0	79.54

and

Table 13. Calculated IHA minimum flow values and low pulses for modeled IPSL-CM5A-MR RCP8.5 climate projections.

Period	7-Day (m3/s)	30-Day (m3/s)	90-Day (m3/s)	Zero Flow (d)	Low Pulses	Low Pulse-Duration
Short-Term	1.4	1.6	2.4	0.0	3.8	28.3
Med-Term	1.8	2.1	3.1	0.2	4.1	29.0
Long-Term	2.3	2.7	3.8	0.1	4.6	24.9

as well as Figure 7 below.

The modeled data show annual averages for each data type. Like the previous RCP8.5 simulation, this run shows very high precipitation. Even with the high precipitation, the water stress in days is only about 20 days lower than the cold atmosphere condition. Drought indices are showing a very large difference, with over 90% of the modeled time showing a below-zero SMDI and an increase to around 20% of severe drought with the SMDI below negative three. IHA calculations show that the minimum flow is not zero, yet low-pulse durations increase in comparison to the cold projection simulation.

The final scenario, HadGEM2-ES, projects a much warmer temperature than the IPSL projection. The most relevant data for this simulation are presented in

Table 14. Modeled data from HadGEM2-ES RCP8.5 climate projection run. Data presented show average annual for each modeled period.

Period	Precipitation (mm)	ET (mm)	PET (mm)	Runoff (mm)	Shallow AQ (mm)	Deep AQ (mm)	Water Stress (d)
Short-Term	135.9	224.3	1463.5	2.7	0.2	0.3	182.9
Med-Term	159.9	253.4	1576.4	3.6	0.6	0.8	72.7
Long-Term	155.9	258.8	1708.3	3.3	0.2	0.4	60.5

and

Table 15. Calculated IHA minimum flow values and low pulses for modeled HadGEM2-ES RCP8.5 climate projections.

Period	7-Day (m3/s)	30-Day (m3/s)	90-Day (m3/s)	Zero Flow (d)	Low Pulses	Low-Pulse Duration
Short-Term	0.0	0.0	0.0	316.9	15.7	21.7
Med-Term	0.0	0.0	0.0	290.3	16.9	18.1
Long-Term	0.0	0.0	0.0	313.3	18.0	18.8

and Figure 8 below.

The modeled data show annual averages for each data type. This simulation is an extreme example of RCP8.5 with very little precipitation and very high ET and PET. The repercussions of this can be seen in the low Q values in surface water flow and both groundwater flows. This also resulted in a very high amount of water stress days annually. The SMDI calculations show a very large amount of the modeled timeline below negative three, indicating severe drought. The ETDI calculations also show that a majority of the modeled timeline is below zero. IHA calculations show that all minimum flows are at zero and that there are a significant number of zero-flow days throughout the modeled timeline.

4. Discussion

Key differences can be inferred from the comparison of modeled runs between historical and climate change scenarios. Annual precipitation increases with respect to the historical run versus all climate runs except for the extreme RCP8.5 scenario from HadGEM2-ES. Additionally, the RCP4.5 scenarios had less precipitation than the RCP8.5 scenarios. Consequently, all instances of runoff and groundwater outflow increase because of the increased precipitation; this could be corroborated by the studies done in the New York City Water Supply System [6]. It is interesting to note that the cold scenario of the GISS-E2-R RCP4.5 scenario shows fewer water stress days annually than the historical model, possibly due to the decrease in ET and PET and an increase in precipitation. All other climate change scenarios show an increase in water stress annually, with a few exceptions: the short-term warm IPSL RCP8.5, and the medium- and long-term of the extreme HadGEM2 RCP8.5.

Drought index calculations can show if a drought is occurring and how extreme the drought is [12]. The range of both the SMDI and ETDI is from 4 to −4, with the drought being more extreme as it approaches −4. The average SMDI and ETDI were calculated for each 30-year timeline. Historical SMDI-indicated droughts, compared to the projected cold RCP4.5 scenario, occurred less frequently, both in instances below zero and below negative three, by 5–6%. Furthermore, the average value increased, indicating, on average, fewer droughts. On the other hand, the ETDI shows an increase in instances, even though the average decreases slightly. This shows that the ET deficit will increase while the soil moisture deficit will decrease with enough mitigation and a cooling atmosphere [12]. Comparison between the historical to the projected warm RCP4.5 shows that the SMDI and ETDI, on average, will increase. The frequency of both SMDI- and ETDI-indicated instances below zero decreases, yet the frequency of instances below negative three increases, indicating an increase in the severity of the soil moisture deficit. Historical comparison to the cold projected RCP8.5 shows a decrease in the SMDI average and an increase in the ETDI average. Frequency calculations show an increase in instances below zero by 40% and a decrease in instances below negative three. This indicates a more frequent soil moisture deficit that is not as severe. Conversely, the frequency of the ETDI being below zero decreases, and below negative three it increases, suggesting more frequent, severe ET deficits [41]. Warm projected RCP8.5 comparisons to the historical show a similar decrease in the SMDI average and an increase in the ETDI average. SMDI frequency calculations show that instances of soil moisture falling below zero increase by about 50% and a slightly smaller increase of around 4–5% is seen in instances below negative three. The extreme RCP8.5 simulation yielded very interesting deficit calculations compared to the historical scenario. Both instances of below zero and below negative three significantly increased for both soil moisture and ET deficits. Though the average SMDI and ETDI are lower than in the historical scenario, they are not as low as the previous two RCP8.5 projections. Overall, the average indices are the lowest for RCP8.5, followed by the historical numbers, then by the RCP4.5-modeled runs. Additionally, RCP8.5 cold and warm projections experienced the greatest deficits out of all.

Comparisons of IHA calculations show the changes in baseflow throughout the modeled timelines [6,40]. In all climate change simulations, the minimum baseflow increases, with the exception of the extreme RCP8.5 simulation. This could be explained by the increased precipitation seen in each of the climate change simulations. Furthermore, zero-flow days increase in frequency through all climate change scenarios. Comparing RCP4.5 to RCP8.5 simulations, except for the extreme scenario, minimum flows increase as well, caused by the increased precipitation. Low-pulse frequency and duration are variable and depend on the climate change simulation. Cold RCP4.5 simulations show a decrease in low pulses and an increase in low-pulse duration, while the warm RCP4.5 simulation is similar, with the low-pulse duration lasting 0 days longer than the cold simulations. Cold and warm RCP8.5 simulations have very similar numbers of low pulses and their duration, with both decreases in the number of pulses but increased low-pulse duration. The extreme RCP8.5 simulation is the only simulation that shows an increased number and duration of low pulses in all climate change simulations. Further, recent studies have also indicated that changes in pulses are more likely to cause flow regime shifts [42], which is corroborated by the study done by Wolfe et al. [43] highlighting impacts of climate change and increases in both the magnitude and frequency of pulse events.

5. Conclusions

Overall trends can be deduced to potentially predict the response of the Chipuxet watershed to climate change conditions. The overview of these predictions shows water stress increasing in all simulations, as predicted with any climate change scenario. Furthermore, water stress is still increasing, even with the high-precipitation climate change scenarios. Both deficit index calculations, the SMDI and ETDI, show that RCP8.5 scenarios have a more severe deficit than RCP4.5 scenarios, as predicted, with more intense climate change. The manner of the severity varies, with some scenarios showing a prolonged mild deficit with little severe deficit, and some showing more spikes of severe deficit with little mild deficit. Overall prediction scenarios show that deficits will increase, which prompts the need to be aware of which hydrological system is most important in the watershed.

IHA calculations are important indicators of baseflow reactions in a watershed. Overall analysis of the calculations shows that RCP8.5 scenarios will experience more zero-flow days than any other scenario. Additionally, the Chipuxet will experience these zero-flow days more frequently regardless of the scenario compared to historical records. Minimum flow shows an increase in all scenarios except the severe RCP8.5, which can be explained by the major increase in yearly precipitation predicted in the climate change scenarios. In conclusion, these results incentivize the need for the mitigation of climate change. At the very least, management practices will need to change to lessen the effects of the drying climate around the Chipuxet watershed.

SMDI calculations and predictions impact agriculture the most. This project can inform policy makers in the Chipuxet and surrounding area for decisions regarding future practices. Potential shifts in water use regarding agriculture may be needed as the results of climate projections show major impacts on the amount of water in the soil. ETDI calculations and predictions show drought effects on water stress, which can translate to ecological stress. Understanding future trends may inform policy around conservation activity in the Chipuxet and its effects on the larger watershed that the Chipuxet drains into.

Future work that stems from this research should include more investigations on water supply and the effects of drought on the underlying aquifer. Combining the now-calibrated SWAT model with other models such as MODFLOW can give a more accurate depiction of the watershed’s changing conditions. Additionally, climate change runs could be integrated into the MODFLOW model for the region to create a much better picture of how the aquifer will respond to the changing climate. Incorporating the results of this model into decision support tools such as WMOST, developed by the EPA, can help calculate the deficit of water the aquifer will experience in the future with climate change [44]. A combination of models may help with the mitigation of droughts and management of the watershed with future climate change scenarios.