Impact of Land Use Land Cover Changes on Streamflow

Impact of Land Use Land Cover Changes on Streamflow – A Study on Sope Creek Watershed, Georgia

INTRODUCTION

Water quantity and quality of any watershed are affected by land use and land cover (LULC) in that watershed. Continuous population growth and corresponding economic developments to accommodate the population demands impact hydrology of watersheds resulting in unexpected stream flows and impairment of streams in those watersheds. Anthropogenic actions like residential and commercial developments, pavement of roads, land management practices, and clearing forest lands to cater the human needs have significant and adverse impacts on stormwater runoff. LULC changes affect runoff characteristics of a watershed which ultimately alter hydrology and sediment transport of the basin (Defersha et al., 2012). LULC is also an important aspect in the energy balance within the hydrologic cycle due to the impact on evaporation, transpiration, and radiation interception (Tadese, 2014). Evaluation of LULC changes and corresponding effects on hydrology in any watershed is always challenging for researchers, engineers and policy makers (Ma et al., 2008). Developing hydrologic models is considered as a powerful prediction tool to study impacts of climate and LULC on hydrology (Fang et al., 2013). Several modeling approaches like EPA Storm Water Management Model (SWMM), Long Term Hydrologic Impact Analysis (L-THIA), Soil & Water Assessment Tool (SWAT), and Agricultural Non-Point Source Pollution (AGNPS) are developed to estimate the impact of LULC on streamflow and quality (Zhu et al., 2014). SWAT is a widely used small watershed modelling technique to evaluate land management practices, predict soil erosion potential and control non-point source pollution.

Get Help With Your Essay
If you need assistance with writing your essay, our professional essay writing service is here to help!
Essay Writing Service

In the current study, LULC change impact on the Sope Creek watershed in Cobb county, Georgia was studied using SWAT modelling. Cobb county, a suburban county is the third most populous county in the state of Georgia. According to the Cobb County Watersheds Authority, majority of the county is residential (54%) followed by municipalities (22%) and green space (11%). There is a significant growth in the county’s population between 1990 (450,812) and 2017 (755,754) as per the United States Census Bureau indicating a growth of 68%. Etowah and Chattahoochee are the two major watersheds in the 345 square miles Cobb county area. Several parameters like stream flows, sediments, chemical and biological pollutants in these watersheds are affected by stormwater runoffs from impervious areas due to heavy commercial and residential developments in the county. Chattahoochee river is the major potable and recreation source in the state and provides drinking water for more than 70% (300 million gallons/day) of the Atlanta area. Figure 1 shows the Chattahoochee watershed developed using ArcGIS module with digital elevation model (DEM) based stream definition. Nickajack Creek, Sope Creek, Sewell Mill Creek, Sweetwater Creek, Willeo Creek, and Rottenwood Creek are the major tributaries of the Chattahoochee river in the county. LULC maps from the years 1992, 2001, 2006, and 2011 were obtained from Multi-Resolution Land Characteristics Consortium (MRLC) as inputs for SWAT modelling to analyze the land cover changes and their impact on hydrology. A calibrated SWAT model was developed using 2001 LULC map and it was used to estimate streamflow in the Sope Creek in different validation periods corresponding to the available LULC maps. Similar study on several watersheds including Sope Creek in and around metro Atlanta area was conducted (Hill, 2016) to evaluate LULC changes and corresponding impacts on the measured streamflows in those streams. Developing models and comparing the model performance using statistical analysis at different LULC records would allow more accurate assessment of LULC effects irrespective of changes in climatic conditions. In the current study, LULC effects on the watershed are evaluated using a calibrated SWAT model instead of comparing measured streamflows only.

Figure 1. Chattahoochee watershed with DEM based stream definition and accumulation developed using ArcGIS.

Materials and Methods

Study Area

The current SWAT modeling study was conducted on the Sope Creek watershed, located in the Chattahoochee watershed (HUC – 03130001). Sope Creek is approximately 12.0 mi long originating in east Marietta and one of the major tributaries of Chattahoochee river in the Cobb county area. The Sope Creek watershed is located at 33.95388890 latitude and -84.44333330 longitude with an elevation of 881.4 feet and a drainage area of 30.7 square miles. The stream bed is comprised of rocks, gravel, and silt. Several commercial establishments like Six Flags, White Water Park, TipTop Poultry, and Coca-Cola bottling unit are in the watershed. Majority of the watershed is residential (65%) followed by parks, recreational facilities and commercial developments. The USGS Georgia Water Science Center manages stream gage, “Sope Creek Near Marietta, GA” (USGS-02335870). Average slope of the watershed is around 7.0% with GA129 soil type. Cobb county watershed studies suggested significant deterioration in Sewell Mill Creek, a tributary of the Sope Creek due to sediment deposition. Delineated Sope Creek watershed with outlet close to the USGS-02335870 gaging station is shown in figure 2.

Figure 2. Delineated Sope Creek watershed with outlet near USGS-02335870 station.

SWAT Model and Inputs

SWAT is developed to simulate the impact of characteristics of the catchment area on surface runoff, groundwater hydrology, stream contamination, and sediment transport. SWAT uses spatial data (DEM, LULC map, and soil maps) and temporal data (hydrological and climate data) to develop a prediction model. SWAT divides the entire watershed into multiple subbasins depending on the spatial data and further divides into several homological Hydrological Response Units (HRU) based on the input thresholds for land use, slope and soil type. SWAT simulates routing from HRUs to the subbasin level followed by stream network to the outlet of the watershed (Shilling et al., 2007).

In the current study, SWAT modelling was done using the ArcGIS-ArcView interface, 2012. The DEM (30m x 30m resolution) required for the SWAT model for watershed delineation, sub-basins division and sub-basin parameters calculation was obtained from NHDPlus data, a geo-spatial hydrologic framework by EPA and USGS, developed by Horizon Systems. Corresponding hydrography and gauging stations datasets were also obtained from Horizon Systems. ArcSWAT’s inbuilt State Soil Geographic (STATSGO) database was used for soil classification inputs. The required climate data including daily precipitation, and daily maximum and minimum temperatures for the period of 1992 – 2018 were downloaded from National Oceanic and Atmospheric Administration (NOAA) from the closest weather station. The precipitation and temperature data were manually inputted to the SWAT model and the other required parameters including relative humidity, solar radiation and wind speed were automatically generated using the SWAT weather generator interface. The measured streamflow data at the gaging station USGS-02335870 required for calibration and comparisons was downloaded from USGS National Water Information System.

The LULC information was collected from MRLC consortium. The USGS along with some federal agencies released four NLCDs – 1992, 2001, 2006, and 2011 in the past two decades providing reliable land cover data over the nation (Yang et.al., 20018). The NLCD 1992 product was the first dataset at 30 – meter medium resolution. NLCD 2011 is an updated version of NLCD 2001 and NLCD 2006 only as the mapping methods changed significantly compared to those of NLCD 1992 (Wickham et al., 2014).

Modelling and Study Methodology

A SWAT model was developed to estimate streamflow in the Sope Creek with outlet close to the stream gaging station at Marietta using 2001 LULC data set. The model developed was calibrated by adjusting the most sensitive parameters to achieve the required statistical confidence in the period 2002 – 2005. The calibrated model was validated between 1982 – 2018 to study the applicability of the developed model in those selected test periods before and after 2001. Changes in land cover were also evaluated using ArcGIS platform and reported.

Model Performance and Calibration

Performance of the developed model was evaluated by quantitative statistical analysis. Three statistical indicators, Nash Sutcliffe Efficiency (NSE), Percent Bias (PBIAS) and ratio of root mean square to the standard deviation of measured data (RSR) were used in this study to determine the model performance. The SWAT Output Viewer tool was used to calculate NSE and PBIAS values indicators. The recommended statistics for a monthly time step to consider the performance of a developed SWAT model to be satisfactory are 0.5 < NSE < 0.65, 15 < PBIAS < 25%, and 0.60 < RSR < 0.70 (Moriasi et al., 2007). Description of each statistical indicator, calculation procedure and performance rating of the model for different values are reported by Moriasi et al., 2007.

Calibration of physical based semi-distributed model like SWAT requires optimization of multiple parameters simultaneously to consider correlation between multiple processes. Uncertainty in inputs, parameters and model structure results in the uncertainty of the watershed model (Pai et al., 2013). SWAT modelling tool considers several parameters and mathematical relationship between those parameters to simulate the hydrology in a watershed. Manual calibration by altering single parameter at a time may not be suitable for several large-scale applications. Manual calibration or auto calibration becomes more complex and laborious process when several parameters are selected for calibration. By conducting sensitivity analysis on the parameters to determine the most sensitive parameters reduces the calibration effort significantly. SWAT – Calibration and Uncertainty Program (SWAT – CUP) is a widely used calibration technique for sensitivity analysis and auto-calibration of hydrologic models.

In the current study, SWAT – CUP’s Sequential Uncertainty Fitting (SUFI2) algorithm was used for sensitivity analysis and best parameter estimation. SUFI2 computes overall uncertainty in the output for each parameter using 95 percent prediction uncertainty (95PPU) and outputs 95PPU and dotty plots for each parameter selected. Two statistical values, p-factor (percentage data bracketed by 95PPU) and r-factor (average width of the 95PPU band divided by standard deviation of the measured variable). High p-factor (close to 1.0) and low t-factor (close to 0.0) indicates efficient calibration performance.

Results and Discussions

LULC Changes

The LULC maps from 1992, 2001, 2006, and 2011 were used to study the composition of the watershed after the delineation and subbasin parameters calculation process. The land cover from the four selected periods and distribution are shown in figure 3 and figure 4 respectively. Compared to the 1992 map, there was a significant increase in residential area in 2001. Low, medium and high density residential areas increased by 11.3%, 36.7%, and 37.5% respectively, indicating a considerable increase in the population and shift in the distribution. Around 3,000 acres of mixed forest area reported during 1992 was converted into residential area in 2001. Around 50% reduction in industrial/commercial was observed in 2001 compared to the 1992 land cover map. A small agricultural area (2.4 acres) was reported in the 1992 land cover map.

Figure 3. LULC changes in Sope Creek watershed in the 1992, 2001, 2006 and 2011 LULC maps.

s

Figure 4. Land use distribution in Sope Creek watershed from 1992, 2001, 2006 and 2011 LULC maps.

Noticeable difference in the land cover was not observed between the maps in the years 2001, 2006, and 2011. Compared to 2001, the low and medium density residential areas remained constant (<2% change) but high density residential areas increased by 12.1% and 24.4% in the 2006 and 2011 maps respectively. Moderate increase in the industrial areas was reported in the years 2006 (6.9%) and 2011 (13.0%) compared to 2001. Increase in the high density residential areas indicate concentration of the increased population in few areas of the watershed only. Around 7.0 – 15.0% reduction in forest areas was observed in these two land cover maps compared to the 2001 map. Detailed distribution of different types of land use is all the four available land cover maps is shown in the table 1.

Table 1. Land use distribution in Sope Creek watershed in acres

Area (acres)

Land Use Type

1992

2001

2006

2011

Water

58.0

97.6

94.3

94.3

Residential-Low Density

6,685.4

7,537.6

7,591.4

7,527.8

Residential-Medium Density

3,180.7

5,026.3

5,096.6

5,079.7

Residential-High Density

844.4

1,350.2

1,513.4

1,679.7

Industrial

1,253.6

838.0

896.0

947.2

Southwestern US (Arid) Range

9.8

28.7

9.6

4.4

Forest-Deciduous

1,545.9

1,815.4

1,689.5

1,658.4

Forest-Evergreen

2,862.4

2,616.2

2,435.2

2,333.4

Forest-Mixed

3,235.8

81.6

74.9

71.4

Range-Brush

–

2.4

2.4

18.5

Range-Grasses

–

42.5

50.0

42.0

Hay

–

141.4

125.4

122.1

Wetlands-Forested

–

100.5

99.6

99.6

Agricultural Land-Row Crops

2.4

–

–

–

SWAT Modelling, Sensitivity Analysis and Calibration

The preliminary modelling was conducted on the watershed using 2001 LULC map during the period of 2002 – 2005 using the default parameters in SWAT. A threshold area of 250 ha was selected for DEM based flow stream definition. By selecting the outlet close to the gaging station, the watershed was delineated into 13 subbasins and further divided into 113 HRUs based on the inputted data and thresholds selected. Comparison of the simulated average monthly streamflows and measured values at the USGS-02335870 gaging station indicated satisfactory performance of the model. The NSE, PBIAS and RSR values of the model were 0.574, 24% and 0.653 respectively. Average monthly precipitation (mm), surface runoff (SURQ, mm), simulated streamflow (cms) and observed streamflows are shown in figure 5.

Figure 5. SWAT model performance with default values – average monthly precipitation, surface runoff, streamflow default and measured.

SWAT’s manual calibration helper was used to select parameters for sensitivity analysis and SUFI2/SWAT – CUP auto calibration. Some parameters were manually replaced using the manual calibration helper interface and effect on the model performance was observed. Base flow alpha factor (ALPHA_BF, days), threshold depth of water in the shallow aquifer required for return flow to occur (GWQMN, mm), groundwater delay (GW_DELAY, days), runoff curve number (CN2), available water capacity of the soil layer (SOIL_AWC), saturated hydraulic conductivity (SOL_K), average slope length (SLSUBBSN), average slope steepness (HRU_SLP), soil evaporation compensation factor (ESCO), plant uptake compensation factor (EPCO), and surface runoff lag time (SURLAG) were selected for auto calibration. SWAT – CUP’s SUFI2 auto calibration to improve NSE value was done by incorporating the selected parameters in their respective ranges. The most sensitive parameters reported by the auto calibration after 500 simulation runs were SOL_AWC and SOL_K. The best fitted values for the selected parameters from auto calibration are shown in table 2. The p-factor for the calibration was 0.92 indicating a successful sensitivity analysis and calibration. The 95PPU plot and dotty plots from the auto calibration step are shown in figure 6.

Table 2. Default values and fitted values of the SWAT model using auto calibration

SI. No.

Parameter

Default

Fitted Value

Minimum

Maximum

1

CN2.mgt

66

50.26

40

60

2

ALPHA_BF.gw

0.048

0.281

0

1

3

GW_DELAY.gw

31

232.9

30

450

4

GWQMN.gw

1000

925

0

5000

5

SOL_AWC(..).sol

0.1

0.385

0

1

6

SOL_K(..).sol

400

518

0

2000

7

SLSUBBSN.hru

91.46

98.55

0

150

8

SURLAG.bsn

2

18.34

0

20

9

HRU_SLP.hru

0.0452

0.183

0

1

10

ESCO.hru

0.95

0.421

0

1

11

EPCO.hru

1

0.775

0

1

Figure 6. SWAT – CUP SUFI2 auto-calibration: 95 PPU plot and dotty plots.

Manually replacing the values from auto calibration and re-running the model significantly improved NSE value. The statistical indicators NSE, PBIAS and RSR for the calibrated model were 0.776, 12% and 0.47 respectively indicating a very good performance rating (Moriasi et al., 2007). Streamflows from default and calibrated simulation runs along with the measured are shown in figure 7. Water balances for the default model and calibrated model are determined using the SWAT Check and are shown in figure 8.

Figure 7. Calibrated SWAT model– average monthly precipitation, surface runoff, streamflow – default, calibrated and measured.

 

Figure 8. Water balance of default (left) and calibrated model (right).

Model Validation

Three scenarios were established to study the impact of LULC on the Sope Creek streamflow. Using the 2001 LULC map and the calibrated model for the period 2002 – 2005, streamflows were predicted using temporal data from three different periods 1993 – 2000, 2007 – 2010, and 2012 – 2016 where land cover composition was different from that of the 2002 – 2005 period. The validation periods were selected based on LULC maps release years (1992, 2001, 2006, and 2011) so that the effect of LULC changes on the streamflow can be predicted by correlating with performance of the calibrated model in different test periods. The start year for the validation period is selected close to the LULC map release year. The statistical identifiers, NSE, RSR and PBIAS were calculated for the three validation periods manually.

For the period 2007 – 2010, the model performance rating was good with NSE and RSR values of 0.684 and 0.562 respectively. PBIAS value indicated over estimation of the streamflow by 14%. During the validation period 2012 – 2016, the model overestimated the flow by 36%. NSE value decreased significantly and indicated unsatisfactory performance. Increase in dense population areas and conversion of forest lands into residential and industrial areas impacted the model performance significantly. The model performance was unsatisfactory during the validation period of 1993 – 2000 based on the NSE and RSR values but comparatively performance rating was better. The model underestimated the flow by 5%. A trend in prediction of streamflow by the model was observed considering the PBIAS values for the periods before and after the 2002 – 2005 calibration period. The model underpredicted the flow before the calibration period whereas it overpredicted after. NSE, RSR and PBIAS values for all the simulated periods are shown in figure 9. Figures 10, 11 and 12 show comparison plots of simulated and measured streamflows for the three validation periods – 1993 – 2000, 2007 – 2010, and 2012 – 2016 respectively.

Figure 9. Statistical Identifiers NSE, RSR and PBIAS for all the periods simulated.

Figure 10. Simulated and measured streamflows at the Sope Creek watershed for the period 1993 – 2000.

 

 

 

Figure 11. Simulated and measured streamflows at the Sope Creek watershed for the period 2007 – 2010.

 

 

 

 

Figure 12. Simulated and measured streamflows at the Sope Creek watershed for the period 2012 – 2016.

 

References

Defersha, M. B., and Melesse, A. M. (2012). Field-scale investigation of the effect of land use on sediment yield and runoff using runoff plot data and models in the Mara River basin, Kenya. Catena 89: 54-64.

Fang, X., Ren. L., Q., Zhu, Q., Shi, P., and Zhu, Y. (2013). Hydrologic response to land use and land cover changes within the context of catchment-scale spatial information. J. Hydrol. Eng. 18(11): 1539-1548

Chee Hill, T. (2017). Land cover changes impact on multidecal streamflow in metropolitan Atlanta, GA, USA. MS thesis. Athens, Georgia: Georgia State University, Department of Geosciences.

Ma, Z., Kang, S., Zhang, L., Tong, L., and Su, X. (2008). Analysis of impacts of climate variability and human activity on streamflows for a river basin in arid region of northwest China. J. Hydrol. 352(3-4): 239-249.

Moriasi, D. N., Arnold, J. G., Van Liew, M. W., Bingner, R. L., Harmel, R. D., and Vieth, T. L. (2007). Model evaluation guidelines for systematic quantification of accuracy in watershed simulations. Trans. ASABE 50(3): 885-900.

Pai, N., and Saraswat, D. (2013). Impact of land use and land cover categorical uncertainty on SWAT hydrologic modeling. Trans. ASABE 56(4): 1387-1397.

Shilling, K. E., Manoj, K. J., You-Kuan, Z., Philip, W. G., and Calvin, F. W. (2008). Impact of land use and land cover change on the water balance of a large agricultural watershed: Historical effects and future directions. Water Resources Research 44: W00A09.

Tadese, W. (2014). Assessing the impact of land use land cover change on stream water and sediment yields at watershed level using SWAT. Open Journal of Modern Hydrology 5: 68-85.

Wickham, J. D., Collin, G. H., James, E. V., Alexa, M., Rick, M., Nate, H., and John, C. (2014). The Multi-Resolution Land Characteristics (MRLC) Consortium: 20 years of development and integration of USA national land cover data. Remote Sens. 6(8): 7424-7441.

Yang, L., Suming, J., Patrick, D., Collin, H., Leila, G., Stacie, M. B., Adam, C., Catherine, C., Jon, D., Joyce, F., Michelle, F., Brian, G., Greg, C. L., Matthew, R., and George, X. (2018). A new generation of the United States national land cover database: Requirements, research priorities, design, and implementation strategies. ISPRS Journal of Photogrammetry and Remote Sensing 146: 108-123.

Zhu, C., and Yingkui, L. (2014). Long-term hydrological impacts of land use/land cover change from 1984 to 2010 in the Little River Watershed, Tennessee. International Soil and Water Conservation Research 2(2): 11-22.