Implications of Different DEMs on Watershed Runoffs Estimations

Watershed modelling tools like ArcSWAT, an ArcGIS extension of Soil and Water Assessment tool (SWAT), are useful to watershed managers in many ways. One particular use is analyzing model outputs for decision making related to waterway restoration and mitigation, which is often undertaken to improve water quality in streams. The present study evaluates the use of digital elevation model (DEM) at 10 meter, 30 meter, and 100 meter pixel size on non-point runoff predictions for three sub-watersheds in Raritan River Basin in New Jersey. These three watersheds include: Bound Brook, Lamington River, and Lawrence Brook watersheds. ArcSWAT is utilized to investigate the difference due to DEM variation in predicting monthly estimates of pollutant loads including ammonium (NH4), nitrite (NO2) and sediment transported with water out of a watershed. Using land use/cover, slope and soil data for 2012, monthly pollutant loads are calculated for each sub-basin in the watershed over a 10-year simulation period (2012-2022) in ArcSWAT. Overall statistical and spatial results show that ArcSWAT results are sensitive to changes in DEM pixel size for watershed modeling. The results show that total sum of monthly runoffs including NH4, NO2 and sediment differ among the three different DEMs. Moreover, the spatial pattern of input (in sub-catchments) also changes among the three DEMs for most watersheds. This indicates that watershed managers need to supplement model predictions with field measurements before making substantial investments in stream restoration programs.

SWAT is developed by the USDA Agricultural Research Service, is a continuous, distributed parameter, daily time step model used to assess the effects of land management practices on the hydrology, nutrients, sediments, and non-point source pollutant transport in watersheds under various slope, soils, and land use/cover conditions in a continuous-time framework [1] [4].One of the functionalities of the SWAT model is to divide a watershed into sub-basins and then further divide each sub-basin into hydrological response units (HRUs).SWAT represents a realistic projection given specific biophysical features such as land use/cover, soil, topography, hydrology, climate, and policy effects at sub-watershed area [5].
In predicting surface runoff, the hydrological process requires determination of topographic characteristics [6].The spatial patterns in such systems are heavily based upon the attributes such as slope and the area per slope length.DEMs are used as digital raster based map of the land surface area [6] [7].DEMs are implemented as a topographic representation in ArcSWAT and serve as a crucial data layer to define physical parameters such as area, slope and slope length for each sub-basin within the watershed.The quality of the ArcSWAT model performance in predicting future scenarios depends upon how well the model inputs represent the relevant characteristics of the watershed.In general, there is a tradeoff between the DEM resolution and fine scale details for simulation, the accuracy of the data and computing speed [6].Due to this tradeoff, users often select coarse resolution of DEM to speed up the simulations.It has been suggested that lower resolution in spatial input data results in segmented watersheds while higher resolution allows better delineation of flat surfaces [6] [8].
The quality of spatial input data is crucial for model development and accuracy [9].Several studies analyzed the significance of scale effects on the quality of natural system processes and predictions [6] [8] [10] [11] [12].Lin [13], analyzed the impact of different resolutions of DEMs.These DEMs are collected from different data sources to evaluate the sensitivity of SWAT output for three runoffs: sediment, total phosphorous (TP) and total nitrogen (TN).The results of their study suggest that SWAT is sensitive to the grid size effects due to the variations in DEM.The use of small grid size improves the model outcomes for 90 meter and 30 meter grid size but does not improve the results for 5 meter DEM.Another study investigated that difference in scales are minimal in small watershed, however in large scale watersheds there is an increased amount of uncertainty in stream flow outputs due to scale variation [11].
In ArcSWAT, multiple HRUs are calculated and generated based upon topography of the landscape.The HRUs derived through the use of DEM help in investigating the spatial variation in input, output, and flow of water pollutants in catchments [14] [15].In order to capture the changes in watershed management on water quality outcomes, the model must reflect the quality and accuracy of such input data in the model [15].Still, watershed managers use various DEMs in analyzing the watershed health under various stream restoration programs.

N. N. Arbab et al. Journal of Water Resource and Protection
To date, no such study existed for the Raritan River watershed to help watershed management to identify the suitable and efficient scale for predicting watershed quality outcomes.The Raritan River watershed's uniquely diverse activities and physiography make water resource management a complex issue to address and prove to be a challenge to state and local regulators working to maintain its integrity.This complex socio-ecological system consists of abiotic, biotic, and anthropogenic entities that provide a range of ecosystem services.In this regard, this paper investigates the impact of the three different cell sizes in the digital elevation model on simulated NH 4 , NO 2 and sediment outputs of three watersheds: Bound Brook, Lamington River and Lawrence Brook watersheds of Raritan River Basin in New Jersey by using the Soil and Water Assessment Tool (SWAT) in ArcGIS.The present study analyzed the sensitivity of ArcSWAT outputs on three DEM resolutions: 10 m, 30 m and 100 m.The suitability and selection of these resolutions is based upon the results from several studies [9] [13] [16] [17] [18].The relationship between resolution and runoff is important to understand the specific scale that is useful to achieve optimal results in simulation [19].The literature of Raritan River Basin currently lacks methods for systematically analyzing the effect of grid size on statistical and spatial characterization of the land surface and associated hydrological response in terms of watershed quality parameters.The relationship between the hydrological spatial data input and associated hydrological response at different scales is not well understood.In this regard, the novelty of this paper is to provide the knowledge on scale assessment of elevation data in hydrological simulation.
The primary goal of this study is to demonstrate the sensitivity of the ArcSWAT model due to a change in topographic parameter and to provide a better understanding to assess the impacts of land surface variation due to flow direction changes with changing DEMs on surface water quality.Accordingly, the objective of this research is not only to project surface water quality outcome, but to provide an answer to a research question: if the change in resolution of surface patterns affecting water quality output in water quality modeling for small watersheds.
The following goals are met in order to achieve the primary objective of this study.
1) Use spatially determined surface changes to simulate impacts on the transport of NH 4 , NO 2 and sediment in the selected sub-watersheds in Raritan River Watershed.
2) Analyze and compare the outcomes and relative importance of DEM of different sizes on surface water quality predictions of ArcSWAT.

Study Area
The Raritan River Basin intersects three of New Jersey's physiographic regions-Journal of Water Resource and Protection the Highlands, Piedmont, and Coastal Plain.Each of these regions has distinct geologic and soil characteristics.These characteristics affect soil drainage, surface water runoff, ground water recharge, and land use/cover development patterns throughout the Basin.The Raritan River Basin is a 1105 square mile drainage area and serves as an important source of drinking water for the central portion of New Jersey.Municipalities within the Raritan River Basin have been developing at a rapid pace, increasing the amount of impervious surface, reducing riverine buffers, and increasing storm water loads in rivers and streams which makes this a study system recognizable across the globe.Three sub watersheds as shown in Figure 1 were selected for analysis within the Raritan Basin: Bound Brook (Highly Urbanized), Lamington River (Mostly Forested), and Lawrence Brook (Moderately Urbanized and Forested).

Input Data for ArcSWAT
All the spatial input files including raster and polygon files are projected in

Soil
The Soil Survey Geographic (SSURGO) database is used in ArcSWAT in defining the HRUs.The SSURGO data is collected from USDA, NRCS, and New Jersey Office of Geographic Information Systems.SSURGO data provides smaller polygons (soil map units) and higher resolution with fine details.Each soil map unit represents a soil type in each watershed (Figures A1-A3 in Appendix).

Slope
In ArcSWAT multiple slope classes in percentage using 100 meter, 30 meter, and 10 meter DEMs are used to define the HRUs.The slope is broken down into four classes (1%, 5%, 25% and more than 25%) to represent the variation in topography of the Raritan River watershed (Figures A1-A3 in Appendix).

Weather Data
Weather data are obtained from weather database from first order stations in ArcSWAT, which has the weather data from the year 1960 to 2010.The data on rainfall, temperature, relative humidity, solar radiation and wind speed are simulated using the weather generator function in ArcSWAT for each subbasin.

ArcSWAT Model
The ArcSWAT model processes overview is shown in Figure 2 where Sed is the sediment yield on a given day (metric tons), surf Q is the surface runoff volume (mm H 2 O/ha), peak q is the peak runoff rate (m The amount of sediment released out of the watershed on a given day is regarded as a function of final concentration in ArcSWAT. The sediment Outflow in ArcSWAT is calculated as Equation (2) [21]: ( ) where sed is the amount of sediment discharged to the main channel on a given day (metric tons), sed' is the amount of sediment load generated in the HRU on a given day (metric tons), sed stor,i-1 is the sediment stored or lagged from the

Statistical Analysis
Once the level of monthly NH 4 , NO 2 , and sediment are calculated for each watershed, the relative performance of 100 m, 30 m and 10 m DEMs is compared considering sum outputs, the minimum, maximum, mean, median, and standard deviation of pollutant runoff yields for each watershed.By using sum outputs of NH 4 , NO 2 , and sediment, difference and percentage difference are calculated for 100 m, 30 m and 10 m DEMs.By using three DEMs, descriptive statistics and spatial maps of runoffs for each watershed scenario are compared.As suggested in [22], the coefficients of variation (CV) are calculated to compare model efficiency using three DEM grid size.CV also serves as a sensitivity analysis of the model for comparison.CV is calculated as Equation ( 5): where σ is the standard deviation and μ is a mean.
Estimation for monthly runoff of NH 4 , NO 2 and sediment using the 30 m and 10 m DEMs are compared to a reference 100 m DEM for quality assessment of DEM results.Statistical assessment as suggested in [9] is performed to investigate the sensitivity of the DEM in the model.This has been achieved using various statistical parameters including the mean difference (MD), the mean absolute difference (MAD), and the root mean standard difference (RMSD) between Journal of Water Resource and Protection 30 m and 10 m DEMs, and 100 m as a base DEM (Equations ( 6)-( 8)).
( ) Here, MAD, MD and RMSD decrease show increasing model accuracy across all DEMS.The RMSD represents the degree to which the value of 10 m and 30 m differs from the reference 100 m DEM value.

Results
The ArcSWAT estimated the monthly yield of NH 4 , NO 2 and sediment over 10 year time period generated from the different resolutions of DEM combined with land use/cover and soil maps (Table 4).The total sum yield of sediment (tons/hectare) increased with 100 m for all watersheds.NH 4 (kg/hectare) has mixed results showing increase for Lawrence Brook with 10 m, Lamington River with 100 m and Bound Brook with 30 m. Similarly NO 2 (kg/hectare) has mixed results showing slight increase for Lawrence Brook with 30 m, Lamington River with 10 m and Bound Brook with 100 m (Table 4).Overall, significant percentage difference was shown between 30 m and 100 m across all pollutants except NH 4 for Bound Brook.This difference resulted due to different numbers of sub-basins with variation in DEMs for most sub-watersheds (Table 4).The impacts of DEM resolution on model efficiency are investigated using coefficient of variation (CV) (Tables 5-7).With three DEMs, fixed scales of land use/cover and soil maps (1:250,000) are used in ArcSWAT analysis.CV indicates how sensitive the model is to the DEM pixel size on which the runoffs are simulated in ArcSWAT.The results show the changes in value of CV.This indicates that the model is sensitive to the scale variation (Tables 5-7).

Conclusions
Simulation results over a 10-year time period showed that runoff predictions

NAD_1983_StatePlane_New_Jersey_FIPS_2900_Feet and 1 :
250,000 scale format is used.DEM DEM data for 100 m, 30 m, and 10 m are collected for each watershed in the study area to compare runoff results for each grid size.100 m and 30m data are collected from National Elevation Dataset (NED) of U.S. Geological Survey (USGS) and 10 meter is collected from New Jersey Department of Environmental Protection (NJDEP).
. The ArcSWAT model is set up using data on the three watersheds including: Bound Brook, Lamington River, and Lawrence Brook in Raritan River watershed, New Jersey.For each watershed 100, 30, and 10 meter resolution digital elevation model (DEM), land use/cover for 2012, soil type, and local meteorological conditions are used.A DEM is the input used to delineate the watershed sub-basins using topography, such as overland slope and slope length (in meters) to analyze the drainage patterns of the landscape and define the area of the sub-basin in the watershed.ArcSWAT delineated the physical characteristics of the watershed such as size, boundaries, and stream network based upon the digital elevation model (DEM), and divided the watershed into hydrologically and spatially connected sub-basins.Using ArcSWAT, each of the three sub-watersheds is partitioned into sub-basins using sub-basin outlet locations.This division spatially connects each sub-basin Journal of Water Resource and Protection

Figure 2 .
Figure 2. ArcSWAT model procedure for the Raritan River Watershed.

where flowout sed is the amount
of sediment released out of the water with outflow (metric tons), sed,f conc is the final sediment concentration (Mg/m 3 ), and flowout V is the volume of outflow from the impoundment (m 3 of water).Having the sediment yield calculation, ArcSWAT calculates the amount of sediment released to the main channel as Equation (3) [21]: Spatial patterns of pollutant loads for each sub-basin in the watershed over a 10 year simulation period also show variation in the distribution of pollutant runoffs among three sub-watersheds (Figures3-5).DEM is the major spatial data input to assess the prediction of runoff and sediments in the watershed.Results show that spatial concentration pattern of NH 4 , NO 2 and sediment have changed across DEMs in most watersheds (Figures 3-5).The spatial distribution of watershed pollutants across different sub basins shows DEMs influenced spatial patterns of pollutants by the changing DEM resolution effects.Root mean square deviation focuses on overall relative measure of similarity of two maps of each DEM grid for pollutants in each watershed, allowing to assess the differences in spatial distribution.Selected results show with 100 m DEM the runoff of NH 4 and NO 2 is stretched out in the central-lower part of the Bound Brook watershed (Figure 3) with smooth topography, compared to the northern area exhibited relatively steeper slopes (Figure 3 & Figure A1 in Appendix).Having estimations of runoff of sediment, NH 4 , and NO 2 over 2012-2022, ArcSWAT is used here as a decision tool to test the effect of the DEM on the mean monthly runoff of sediment, NH 4 and NO 2 .Most of the results show some variation in runoff due to difference in DEMs.When 100m DEM used as a reference, mean difference (MD), mean absolute difference (MAD), and especially root mean square difference (RMSD) revealed a significant change in terms of model sensitivity in predicting sediment.The results show no variation in overall relative sensitivity with NO 2 with change in DEMs (Tables8-10).

Figure 5 .
Figure 5. NH 4 , NO 2 and sediment yields from each sub-basin as estimated by ArcSWAT.(a) Lawrence Brook 10 meter DEM (b) Lawrence Brook 30 meter DEM (c) Lawrence Brook 100 meter DEM.

Table 1 .
Land use/covers classes and their description ((NJDEP, 2012).Areas characterized by streams and canals, natural lakes, artificial lakes, estuaries & other tidal waters, tidal rivers, inland bays and other tidal waters, open tidal bays, dredged lagoon and Atlantic ocean.

Table 2 .
Land use/covers classes.ArcSWAT provides the weather database from local stations.ArcSWAT is run monthly over a 10 year time period.The pollutant loading data are extracted from model results.The selection criteria for runoff is the pollutant load releasing out of the watershed rather than the final pollutant load flowing into the watershed.Three pollutants: NH 4 , NO 2 and sediment are selected from ArcSWAT output defined in Table3.

Table 3 .
Variables and definitions of pollutant loads in ArcSWAT.
[21]imp is the function of the total area that is impervious, and the β variables are regression coefficients.The conversion factors to implement metric units in equations are used: 25.4 mm/inch, 2.59 km 2 /mi 2 , and 2.205 lb/kg.ArcSWAT assigns the annual precipitation to each sub-basin by aggregating the monthly precipitation from the weather generator data[21].

Table 4 .
Total sum, difference, and % difference in sediment, NH 4 , and NO 2 in 10 m, 30 m, and 100 m DEMs for Lawrence Brook, Lamington River, and Bound Brook.

Table 5 .
Descriptive statistics of average monthly runoff for Bound Brook 10 meter, 30 meter and 100 meter DEM.

Table 6 .
Descriptive statistics of average monthly runoff for Lamington River 10 meter, 30 meter, and 100 meter DEM.

Table 7 .
Descriptive statistics of average monthly runoff for Lawrence Brook 10 meter, 30 meter, and 100 meter DEM.

Table 8 .
Effect of the 10 meter and 30 meter DEM on the prediction quality of sediment, NH 4 and NO 2 in Bound Brook watershed.

Table 9 .
Effect of the 10 meter and 30 meter DEM on the prediction quality of sediment, NH 4 and NO 2 in Lamington River Watershed.

Table 10 .
Effect of the 10 meter and 30 meter DEM on the prediction quality of sediment, NH 4 and NO 2 in Lawrence Brook Watershed.