Assessing the suitability of the Soil Vulnerability Index ...

0 downloads 0 Views 4MB Size Report
The location of Tuckahoe Creek Watershed (TCW, left) and Greensboro Watershed (GW, ... (~70%) that belong to Hydrologic Soil Group (HSG) – A or B (Fig. 2),.
Catena 167 (2018) 1–12

Contents lists available at ScienceDirect

Catena journal homepage: www.elsevier.com/locate/catena

Assessing the suitability of the Soil Vulnerability Index (SVI) on identifying croplands vulnerable to nitrogen loss using the SWAT model

T



Sangchul Leea,b, , Ali M. Sadeghib, Gregory W. McCartyb, Claire Baffautc, Sapana Lohanid, Lisa F. Duriancike, Allen Thompsonf, In-Young Yeog, Carlington Wallaceh a

Department of Environmental Science & Technology, University of Maryland, College Park, MD 20742, United States USDA-ARS, Hydrology and Remote Sensing Laboratory, Beltsville, MD 20705, United States c USDA-ARS, Cropping Systems and Water Quality Research Unit, Columbia, MO 65211, United States d Department of Biology, University of Nevada, Reno, NV 89557, United States e USDA-NRCS Resource Assessment Division, Beltsville, MD 20705, United States f Department of Bioengineering, University of Missouri, Columbia, MO 65211, United States g School of Engineering, The University of Newcastle, Callaghan, NSW 2308, Australia h Interstate Commission on the Potomac River Basin, Rockville, MD 20850, United States b

A R T I C LE I N FO

A B S T R A C T

Keywords: Critical source areas (CSAs) Soil Vulnerability Index (SVI) Soil and water assessment tool (SWAT) Pollutant transport characteristics Site characteristics

Conservation practices are effective ways to mitigate non-point source pollutions, especially when implemented on critical source areas (CSAs) known as the areas contributing disproportionately high pollution loads. Although hydrologic models are promising tools to identify CSAs within agricultural landscapes, their application is limited to areas where data and modeling expertise are available. The Soil Vulnerability Index (SVI) developed by the USDA-Natural Resource Conservation Service (NRCS) Conservation Effects Assessment Project (CEAP) is regarded as a potentially powerful tool for supporting initial classification of inherent soil vulnerabilities at field scale and so could be useful for CSA identification. Its usefulness is being fully evaluated in this project and as part of a larger coordinated study. This particular study evaluated the suitability of the SVI classification scheme for identifying inherent vulnerability of cultivated soils to nitrate and organic N transport by surface runoff and nitrate leaching on two adjacent watersheds with contrasting soil drainage characteristics. We used simulated nitrate and organic N fluxes from the Soil and Water Assessment Tool (SWAT) as reference data. The results showed that the SVI runoff classification scheme was more suitable for organic N while the SVI leaching classification scheme was suited for nitrate due to pollutant transport characteristics. In addition, the SVI leaching classification scheme was more suitable for the poorly-drained croplands than the well-drained croplands. The SVI leaching classification scheme and SWAT output consistently classified nitrate leaching vulnerability based on soil drainage characteristics for the poorly-drained croplands, while the well-drained croplands were highly sensitive to a soil water content characteristic (i.e., gravitational water). Depending on the selection of reference data and test sites, however, the suitability of the SVI classification scheme could differ. Therefore, additional evaluation of the SVI using multiple validation data and sites is highly required to demonstrate its usefulness.

1. Introduction Over 50% of rivers and streams in the United States (US) are identified as “impaired waters”, mainly due to excessive non-point source (NPS) pollution (USEPA, 2017). Although controlling NPS pollution is an urgent problem to mitigate water quality degradation, it is a challenging task because the sources of NPS pollution are spatially diffuse over the landscape (Sharpley et al., 2011). In response to water quality degradation caused by NPS pollution from agricultural land,



implementing conservation practices has been recommended (Kurkalova, 2015; Her et al., 2016). The application of conservation practices on the entire landscape or watershed might lead to the best consequences, but in many cases it is not practical or feasible given the availability of resources. The need for identifying the areas with disproportionately high pollution loads, referred to as critical source areas (CSAs), drew keen attention from watershed managers as a cost-effective way of reducing NPS pollution using conservation practices (Osmond et al., 2012; Yeo et al., 2014; Lee et al., 2016).

Corresponding author. E-mail address: [email protected] (S. Lee).

https://doi.org/10.1016/j.catena.2018.04.021 Received 29 September 2017; Received in revised form 27 March 2018; Accepted 16 April 2018 0341-8162/ © 2018 Elsevier B.V. All rights reserved.

Catena 167 (2018) 1–12

S. Lee et al.

Fig. 1. The location of Tuckahoe Creek Watershed (TCW, left) and Greensboro Watershed (GW, right) (adapted from Lee et al. (2016)). Note: CBP: Chesapeake Bay Program, NCDC: National Climate Data Center, and USGS: US Geological Survey.

for identifying CSAs through a relatively simple overlay analysis using freely available geospatial data (i.e., Soil Survey Geographic Database (SSURGO) for soil drainage characteristics and K-factor, and National Elevation Dataset (NED) from the US Geological Survey (USGS) for slope). Therefore, its potential for broad application by conservationists and managers and in a majority of places, where robust data and expertise for models are often not present, makes it promising to assist with more focused conservation planning. It is important to note that because SVI is based on national databases, it does not include any effect of management on CSAs. In an initial evaluation of SVI, Chan et al. (2017) validated the accuracy of the SVI for the Conservation Effects Assessment Project (CEAP) Goodwater Creek Experimental Watershed in Missouri, based on professional judgement, a local index, and simulation results from a hydrologic model. The SVI indicated a high accuracy (86%) when compared with model simulation results (Chan et al., 2017). Although the SVI appeared to be easily applicable for CSA identification, it is still in need of a thorough validation in areas with different landscape characteristics to demonstrate its robustness and usefulness (Chan et al., 2017). The aim of this study was to evaluate the suitability of the SVI for identifying vulnerability to pollutant transport by surface runoff and

Spatially distributed process-based watershed-scale hydrologic models can identify CSAs because of their ability to quantify pollution loads from the multitude of sources in a watershed (White et al., 2009; Niraula et al., 2013; Liu et al., 2016). For example, Liu et al. (2016) found that CSAs (~20% of the total area) accounted for > 50% of total nitrogen (N) and phosphorus (P) loads using simulated outputs. In spite of their usability for CSA identification, the use of hydrologic models is not reliable in areas where geospatial, climate, and detailed monitoring data required for model simulation and calibration are not fully available (Orlikowski et al., 2011; Chan et al., 2017). In addition, modeling expertise is also required to obtain reliable outputs from complex hydrologic models. Hence, in situations as such, a simpler method with data and application flexibility for CSA identification becomes necessary. The soil vulnerability index (SVI) has recently been developed by the US Department of Agriculture (USDA)-Natural Resources and Conservation Service (NRCS) to classify the risk of pollutant transport from croplands by surface runoff or leaching (USDA-NRCS, 2012). Using soil and topographic characteristics (i.e., soil drainage characteristic, K-factor, and slope), the SVI classification criteria classifies the inherent risk for croplands of vulnerability for pollutant transport by surface runoff and leaching (Chan et al., 2017). This index is useful 2

Catena 167 (2018) 1–12

S. Lee et al.

leaching on two adjacent watersheds with contrasting soil drainage characteristics within the Coastal Plain of the Chesapeake Bay Watershed. This assessment uses simulated outputs from the Soil and Water Assessment Tool (SWAT) as reference because observational data are not available for this study site and this model has been successfully employed for CSA identification (White et al., 2009; Niraula et al., 2011 and 2013; Shen et al., 2015; Liu et al., 2016). The study selected two types of simulated N outputs (i.e., nitrate and organic N) for assessing the SVI runoff criteria because they have distinctive transport mechanisms. The SVI leaching criteria was assessed using simulated nitrate leaching and nitrate transport by lateral flow. We selected these two watersheds as test sites because water and nutrient cycles are distinctively different due to soil drainage characteristics (Lee et al., 2016; Sharifi et al., 2016). Therefore, this study can present an interesting case to examine how the two SVI classification schemes can be applied for CSA identification regarding pollutant and site characteristics. 2. Materials and methods 2.1. Study site The two adjacent watersheds, Tuckahoe Creek Watershed (TCW; 221 km2) and Greensboro Watershed (GW; 290 km2), are located in the upper region of the Choptank River Watershed within the Chesapeake Bay Watershed (Fig. 1). The Choptank River Watershed has been targeted for extensive research and monitoring during the past decade because (Ator et al., 2005) it is designated as “impaired waters” under the Clean Water Act, and (Beeson et al., 2014) it is a site for the NRCS CEAP and Agricultural Research Service (ARS) Long-Term Agroecosystem Research (LTAR) network (Duriancik et al., 2008 and USDAARS, 2016). The two watersheds are similar in the size of cropland areas (TCW: 119 km2 and GW 105 km2) as well as land management (e.g., crop rotations and fertilizer application). However, recent studies have shown nitrate loads from the TCW being approximately twice as much as those from the GW, although the majority of nitrate fluxes stems from croplands for the two watersheds (Lee et al., 2016 and Sharifi et al., 2016). This discrepancy originates from the contrasting soil drainage characteristics between the two watersheds (Fig. 2 and Table 1). The TCW cropland is dominated by the well-drained soils (~70%) that belong to Hydrologic Soil Group (HSG) – A or B (Fig. 2), which promote water percolation and nitrate leaching to groundwater. On the contrary nearly 67% of the GW croplands have poorly-drained soils (classified into HSG – C or D), which are characterized by high surface runoff potential and wet, anaerobic conditions that promote denitrification (Lee et al., 2016 and Sharifi et al., 2016). Therefore, the TCW croplands are susceptible to leaching while GW croplands are more susceptible to surface runoff. Note that cropland hydrologic soil classification is available in Table 1 for the whole watershed and for cropland only. Fig. 2. The physical characteristics of Tuckahoe Creek Watershed (left) and Greensboro Watershed (right): (a) land use, (b) hydrologic soil groups, and (c) elevation (adapted from Lee et al. (2016)). Note that HSGs are characterized as follows: A - well-drained soils with 7.6–11.4 mm/h water infiltration rate; B - moderately well-drained soils with 3.8–7.6 mm/h; C - moderately poorly-drained soils with 1.3–3.8 mm/h; and D – poorly-drained soils with 0–1.3 mm/h (Neitsch et al., 2011).

2.2. Description of the SWAT model The SWAT is a process-based, semi-distributed hydrologic and water quality model. The model can predict land management impacts on the transport of water, sediment, nutrient, and pesticide from lands to water bodies in an agricultural landscape (Neitsch et al., 2011). The watershed is partitioned into sub-watersheds and then further into Hydrologic Response Units (HRUs) based on a unique combination of land use, soil, and slope. The water and nutrient cycles are simulated at individual HRUs, and all the hydrologic variables are aggregated at the sub-watershed level and then transported to the watershed outlet through routing processes (Neitsch et al., 2011). Therefore, the spatial distribution of high pollution sources can be identified at the HRU level. In this study, a total of 542 and 760 HRUs were generated, and 312 and 431 HRUs were classified as croplands (referred to as cropland HRUs, hereafter) for the TCW and GW, respectively.

The generation of surface runoff in the SWAT model is calculated using the Soil Conservation Service (SCS) Curve Number (CN) method (Neitsch et al., 2011). The baseline CN value is determined by the user, based on the land use type and HSGs, assuming average soil moisture condition and a slope value of 5%. Everything else being the same, poorly-drained soils (HSG – C or D) have higher CN values than welldrained soils (HSG – A or B). A high CN value has a high surface runoff generation potential with low water retention potential on the land surface, and vice versa. In addition, the SWAT model adjusts the CN 3

Catena 167 (2018) 1–12

S. Lee et al.

factor is based on the composition of the soil, saturated hydraulic conductivity, and soil structure. The K-factor and HSG were derived from SSURGO, freely available from the Geospatial Data Gateway (https://datagateway.nrcs.usda.gov/). The original SVI definition used the SSURGO representative slope value for each soil type (Chan et al., 2017). However, Chan et al. (2017) found that using a DEM-driven slope increased the discriminating power of SVI. In this study, a 10-m resolution DEM derived slope was used (Beeson et al., 2014).

Table 1 Soil properties and land use distribution of Tuckahoe Creek Watershed (TCW) and Greensboro Watershed (GW) (adapted from Lee et al. (2016)). Land use

TCW

GW

Agriculture Forest Pasture Urban Water body

54.0% 32.8% 8.4% 4.2% 0.6%

36.1% 48.3% 9.3% 5.6% 0.7%

Hydrologic Soil Groups (HSGs) A 0.3% [0.0%] B 55.8% [69.5%] C 2.2% [3.4%] D 41.7% [27.2%]

2.4. HRU classification by the SVI criteria and SWAT outputs 3.1% [2.5%] 22.4% [30.3%] 4.2% [10.5%] 70.3% [56.7%]

We classified the HRUs using the SVI criteria and SWAT outputs, separately, and then compared the two classified HRU groups to evaluate the suitability of the SVI. This comparison considered only the cropland HRUs because they are the major N source areas for the study sites. The SVI is also currently developed for cropland uses only. Based on the unique combination of HSG, slope value, and K-factor for individual HRUs, cropland HRUs for TCW and GW were classified by the SVI criteria for surface runoff (Table 2) and leaching (Table 3) vulnerability. The HSG and K-factor for each HRU were estimated using soil input data obtained from SSURGO data. The slope value for each HRU was calculated using input Digital Elevation Model (DEM) via the HRU analysis process of ArcSWAT (Winchell et al., 2011). We used the SWAT model that was calibrated and validated successfully for our study sites against monthly observations of stream flow and nitrate loads for 14 years (2001–2014; Lee et al., 2018). The parameters known to be sensitive for this region through previous SWAT model studies were selected and adjusted based on our judgement using a semi-automated process. Uncertainty analysis was also conducted using 95% prediction uncertainty, which represents the interval between the top and bottom 2.5% of the > 1000 simulations generated during the calibration processes. Model simulation results indicated acceptable performances (Table S1) based on the model criteria for monthly streamflow and nitrate loads suggested by Moriasi et al. (2007). Lee et al. (2018) present further details about the input data, calibration and validation of the SWAT model used in this study. To avoid the spatial variability of transported nitrate fluxes by different fertilizer applications for different crops, we assumed the same crop rotations (i.e., corn-soybean rotation) for all cropland HRUs. We then ran the SWAT with calibrated parameters and this modified land management, and calculated 14-year averages of annual nitrate and organic N fluxes from cropland HRUs. Two SWAT outputs, simulated nitrate (SNIT) and organic N (SORN) fluxes transported by surface runoff, were considered for evaluating the SVI runoff criteria because the criteria was developed to identify the area inherently vulnerable to pollutant transport by surface runoff (Chan et al., 2017). Nitrate transport by lateral flow was considered a “leaching” process (communication with the SVI developer, Lee Norfleet). Therefore, the sum of simulated nitrate leaching and nitrate transport by lateral flow (SLEA) was used for evaluating the SVI leaching criteria. Although the SWAT model was not calibrated and validated for organic N owing to unavailability of observation, organic N was used as reference data. The transport of this type of pollutant is highly affected by topographic characteristics while nitrate transport is less influenced by those characteristics. Hence, this comparative analysis can show which type of pollutant is more suitable for the SVI runoff criteria. Following the study by Chan et al. (2017), we used the Jenks natural break method (Jenks, 1967) to divide cropland HRUs into four classes (high, moderately high, moderate, and low) based on the SWAT outputs. This method determines intervals by reducing the variance within each group while maximizing the variance between groups (Jenks, 1967). Contrary to the SVI classification criteria, there was no absolute threshold values or criteria to define each class for nutrients (e.g., N and P). It is not easy to transfer threshold values from one watershed to another because weather, crops, and land management differ by sites. However, the classification method is transferable, and Chan et al.

Note: Value in brackets, [], denote the proportion of agricultural land in each HSG.

value on a daily basis depending on soil water conditions. After adjustment of the CN value, surface runoff and infiltration are calculated. Detailed equations for calculating surface runoff using the CN value are available in Neitsch et al. (2011). Water infiltrated into the soil layers reaches the stream through lateral flow or percolates to groundwater. When the water content of the soil layer exceeds its field capacity, percolation takes places. The amount of water percolated into the underlying layer increases as the drainable volume of water increases and the travel time for percolation decreases. When the saturated soil hydraulic conductivity increases or gravitational water (i.e., the amount of water held by the soil between saturation and field capacity) decreases, the amount of water percolation increases (Neitsch et al., 2011). Water stored in groundwater is re-distributed in three ways: transported to the stream through groundwater flow, recharged into deep groundwater, or discharged to soil profiles when the water in soil layers is removed from the capillary fringe by evaporation. In the SWAT model, two nitrogen (N) forms (nitrate and organic N) are transported to the streams (Neitsch et al., 2011). Nitrate is added to soil layers by nitrification, mineralization, and fertilization, and it is lost by denitrification, plant uptake, volatilization, leaching, and transported water, such as surface runoff and lateral flow. Organic N is added to the system via fertilization. Thus, fertilizer application is a major N source for croplands. In this study, detailed agricultural practices (e.g., the amount and type of fertilizer) developed based on discussion with a local expert were manually set by the user (Lee et al., 2016). Representative crop rotations for individual croplands were identified using the USDA-National Agriculture Statistics Service (NASS) Cropland Data Layers with 30-m spatial resolution that show representative crop types for each cropland over the US continent (Lee et al., 2016). As a soluble N form, nitrate transport is highly dependent on the water cycle and is delivered to the stream via surface runoff, lateral and groundwater flow. Nitrate is not attached to soil particles due to its negative charge, and therefore it is easily leached into groundwater with percolating water. Organic N moves only with surface runoff as it is attached to sediment particles. The SWAT model calculates sediment loads using the modified universal soil loss equation (MUSLE, Neitsch et al., 2011).

2.3. The Soil Vulnerability Index (SVI) criteria The SVI classification criteria for surface runoff and leaching vulnerability were developed by the USDA-NRCS as a part of CEAP cropland study (USDA-NRCS, 2012) based on HSG, K-factor, slope, and coarse fragment content of the soil (Tables 2 and 3). The K-factor refers to the soil erodibility factor (K) found in the Universal Soil Loss Equation (USLE). It is a relative index of susceptibility of bare and cultivated soil to particle detachment and transport by rainfall. The K4

Catena 167 (2018) 1–12

S. Lee et al.

Table 2 The SVI criteria for surface runoff vulnerability (adapted from USDA-NRCS, 2012). SVI classes

Hydrologic Soil Group (HSG) A

B

C

D

Low

All acres

Slope < 4

Slope < 2

Moderate

None

Moderately high

None

High

None

4 ≤ Slope ≤ 6 K-factor < 0.32 4 ≤ Slope ≤ 6 K-factor ≥ 0.32 Slope > 6

2 ≤ Slope ≤ 6 K-factor < 0.28 2 ≤ Slope ≤ 6 K-factor ≥ 0.28 Slope > 6

Slope < 2 K-factor < 0.28 Slope < 2 K-factor ≥ 0.28 2 ≤ Slope ≤ 4 Slope > 4

Note: The unit of a slope is %. Table 3 The SVI criteria for leaching vulnerability (adapted from USDA-NRCS, 2012). SVI classes

Low

Table 4 Classified cropland HRUs by the SVI runoff criteria and SWAT outputs.

Hydrologic Soil Group (HSG)

Classes

A

B

C

D

None

None

None

Nonorga1nic soils None

Moderate

None

Moderately high

Slope > 12

High

Slope ≤ 12 or Organic soils

Slope ≤ 12 and K-factor ≥ 0.24 or Slope > 12 Slope ≥ 3 and Slope ≤ 12 and K-factor < 0.24 Slope < 3 and Kfactor < 0.24 or Organic soils

Nonorganic soils None

Organic soils

Tuckahoe Creek Watershed

Greensboro Watershed

Area (ha)

Proportion (%)

Area (ha)

Proportion (%)

10,398 1273 302 5

86.8 10.6 2.5 0.0

9967 171 283 1

95.6 1.6 0.3 0.0

SWAT Nitrate (SNIT) Low 8324 Moderate 3397 Moderately high 230 High 26

69.5 28.4 1.9 0.2

3356 1381 5685 0.03

32.2 13.2 54.6 0

SWAT Organic N (SORN) Low 7058 Moderate 4609 Moderately high 291 High 20

58.9 38.5 2.4 0.2

4534 5178 709 1

43.5 49.7 6.8 0.0

SVI Low Moderate Moderately high High

None

Organic soils

Note: The unit of a slope is %.

(2017) also successfully tested the Jenks natural break method. Therefore, this method is a reliable approach for classifying nutrient vulnerability. Contingency tables based on the areal proportion of cropland in each SVI class and class of SWAT outputs show the extent of consistency or inconsistency between the two vulnerability assessment tools. They are the basis for the comparison of the two tools in each watershed.

the two watersheds likely resulted in the different SNIT classification results. For the two watersheds, the SNIT SWAT results led to “Low”, and “Moderately high” classes assigned to the well-drained (HSG – A or B) and the poorly-drained (HSG – C or D) soils, respectively. The SWAT results showed greater nitrate fluxes transported by surface runoff for the poorly-drained soils with a high CN value than for the well-drained soils with a low CN value, in spite of low slopes. Hence, the TCW croplands dominated by the well-drained soils indicated a greater proportion of “Low” class when assessed with SNIT, followed by the “Moderate” class. In contrast, the GW croplands dominated by the poorly-drained soils indicated a majority of “Moderately high” class (Table 4), followed by the “Low” and “Moderate” classes. When classified by SORN, “Low” and “Moderate” classes accounted for > 90% of the two watersheds (Table 4). Relative to SNIT classification, contribution of well-drained soils in the TCW croplands to “Moderate” class increased while there was a decrease in its contribution to “Low” class. In case of the GW croplands, contribution of poorlydrained soils to “Low” and “Moderate” classes substantially increased. This was because the SWAT considers soil drainage as well as topographic and bio-physical characteristics (i.e., slope and K-factor) for sediment transport, and thus for SORN calculation. The well-drained soils in the TCW croplands indicated higher erosion than the poorlydrained soils due to higher values of topographic characteristics (Table S2), which led to increased proportion of “Moderate” class in the welldrained soils despite their low surface runoff potential. The poorlydrained soils in the GW croplands have lower values of K- or topographic factors than the well-drained soils (Table S2) and therefore lower erosion, which limited organic N transport. As a result, increased proportions of “Low” and “Moderate” classes were observed on the poorly-drained soils in the GW croplands.

3. Results 3.1. The SVI runoff criteria vs. SWAT outputs (SNIT and SORN) 3.1.1. Classification results Table 4 and Figs. 3 & 4 show the cropland HRUs classified by the SVI runoff criteria and SWAT outputs (i.e., SNIT and SORN). When classified by the SVI runoff criteria, a large majority of cropland HRUs were in the “Low” class for the two watersheds (Table 4), due to the low topographic characteristics of these watersheds. The SVI runoff criteria assigned “Low” class for the areas with slope < 2% for soils with HSG-C or D, and < 4% for HSG-B (Table 2). Typically, the Eastern shore of the Chesapeake region, where the study sites are situated, is characterized by a flat plain (NOAA, 1978). The median slope values of cropland HRUs are 1.2% for the TCW and 1.1% for the GW. Therefore, “Low” class was prevalent for the two watersheds: ~87% in the TCW croplands and ~96% in the GW croplands, a greater percentage for the GW croplands than the TCW croplands due to lower slope values (Table 4). The classification results based on SNIT significantly differed between the two watersheds (Table 4 and Figs. 3 & 4). Nearly 70% of the TCW croplands was classified as “Low” class while only 32% of the GW croplands was designated as “Low” class. These percentages correspond to the amounts of the well-drained cropland in TCW and GW croplands (Table 1). Interestingly, “Moderately high” (54.6%) was the major class for the GW croplands. Contrasting soil drainage characteristics between 5

Catena 167 (2018) 1–12

S. Lee et al.

Fig. 3. Spatial distribution of cropland HRUs classified by the (a) SVI runoff criteria, (b) SWAT nitrate (SNIT), and (c) SWAT organic N (SORN) for the Tuckahoe Creek Watershed.

classified these well-drained soils in “Low” class due to a low CN value (i.e., low surface runoff potential). As shown in Tables 7 and 8, SNIT “Low” class was almost entirely derived from the well-drained soils. Therefore, the results based on the SVI runoff criteria and SNIT were consistent for the well-drained soils, resulting in a high consistency rate for the TCW croplands dominated by well-drained soils. On the contrary, the classification results for the poorly-drained soils greatly differed between the SVI runoff criteria and SNIT. The SVI runoff criteria classified the poorly-drained soils with a slope value < 2% as “Low” or “Moderate” and those poorly-drained soils with a slope value > 2% as “Moderately high” class (Fig. S1). Due to low topographic characteristics of the two watersheds, the SVI classified greater portions of the cropland in the “Low (TCW: 57% and GW: 93%)” risk class than in the “Moderate (TCW: 35% and GW: 3%)” and “Moderately high (TCW: 8% and GW: 4%)” classes for the poorly-drained soils. However, the poorly-drained soils were classified as “Moderate” or “Moderately High” by SNIT regardless of slope values due to a high surface runoff potential (Tables 7 and 8, Fig. S1). As a result, the GW

3.1.2. Comparison results To evaluate the equivalence between the SVI runoff criteria and the classification based on SWAT outputs, we created the contingency tables for areal proportions (Tables 5 and 6). The classification consistency (sum of all the diagonal percentages in the table) of the SVI runoff criteria was about 81% (SNIT) and 70% (SORN) for the TCW croplands but 34% (SNIT) and 46% (SORN) for the GW croplands. The major discrepancy was observed for the cropland HRUs classified as “Low” class by the SVI runoff criteria and “Moderate” or “Moderately High” classes by the SWAT outputs (Tables 5 and 6). 3.1.2.1. Comparison of the SVI runoff criteria and SNIT. The consistency rate between the SVI runoff criteria and SNIT was greater in the TCW croplands (81%) than in the GW croplands (34%), likely due to contrasting soil drainage characteristic. Overall, the well-drained soils were classified in the “Low” risk class by the SVI and SNIT. Due to low topographic relief, “Low” class characterized the well-drained soils when classified by the SVI runoff criteria. The SNIT classification also

Fig. 4. Spatial distribution of cropland HRUs classified by the (a) SVI runoff criteria, (b) SWAT nitrate (SNIT), and (c) SWAT organic N (SORN) transport by surface runoff for the Greensboro Watershed. 6

Catena 167 (2018) 1–12

S. Lee et al.

croplands dominated by the well-drained soils indicated a higher consistency than the GW croplands mainly covered by the poorly-drained soils.

Table 5 Contingency table for areal proportions between the SVI runoff and SWAT classification methods for the Tuckahoe Creek Watershed (TCW) croplands. SVI runoff Low Moderate Moderately high High

Low 69.5 0.0 0.0 0.0

Low Moderate Moderately high High

58.5 0.1 0.3 0.0

Nitrate (SNIT) Moderate Moderately High 15.8 1.3 10.6 0.0 1.9 0.6 0 0 Organic N (SORN) 27.7 0.5 9.8 0.7 0.9 1.2 0.0 0.0

High 0.2 0 0 0

3.2. The SVI leaching criteria vs. SWAT output (SLEA) 3.2.1. Classification result The cropland HRUs classified by the SVI leaching criteria and SWAT output (i.e., SLEA) are presented in Table 9 and Figs. 5 & 6. The classification results from the SVI leaching criteria indicated a higher leaching vulnerability in the TCW croplands compared to the GW croplands. All HSG-C and D were classified as “Moderate” and “Low” classes by the SVI leaching criteria, respectively, since all soils are nonorganic in our study site (Table 3). Therefore, 85% of the GW croplands dominated by HSG-C and D was classified as “Low (57%)” or “Moderate (28%)” classes while 51% of the TCW croplands dominated by HSG-B and C were classified as “High (34%)” or “Moderately high (17%)” classes, according to the SVI leaching criteria (Table 9). Overall, the classifications based on the SVI leaching criteria were highly affected by soil drainage characteristics, which likely led to high leaching vulnerability in TCW croplands relative to the GW croplands. Similar to the results from the SVI leaching criteria, SLEA classification also indicated a higher leaching vulnerability in the TCW croplands, compared to the GW croplands (Table 9). The SWAT model promotes nitrate leaching on the soils with a higher saturated hydraulic conductivity and therefore SLEA vulnerability was greater on the croplands with a larger percentage of well-drained soils. SLEA classification represented substantial increases in “Moderate” and “Low” classes for the TCW and GW croplands, respectively, compared to the SVI leaching criteria. This was because the SWAT model also considers gravitational water for estimating nitrate leaching. As gravitational water decreases, nitrate leaching is promoted by reduction of the travel time for percolation in the SWAT model (Neitsch et al., 2011). Thus, 79% and 21% of HSG-B in the TCW croplands were classified as “Moderate” and “Moderately High” classes, respectively, by SLEA, depending on gravitational water (Table S4). In addition, 64% and 28% of HSG-D in the TCW croplands represented “Moderate” and “Moderately High” classes, respectively, due to low gravitational water (Table S4). In the GW croplands, high gravitational water led the majority of HSGC and D to have “Low” class when classified by SLEA (Table S4). Therefore, SLEA classification was different from the SVI leaching classification, due to consideration of soil drainage characteristics and gravitational water.

0.0 0.0 0.2 0.0

Note: the sum of numbers in gray boxes indicates accuracy rate.

croplands dominated by the poorly-drained soils indicated a low consistency. 3.1.2.2. Comparison of the SVI runoff criteria and SORN. The consistency rate between the SVI and SORN was 70% and 46% for the TCW and GW croplands, respectively (Tables 5 and 6). Compared to SNIT, the consistency rate increased by 12% for the GW croplands and decreased by 11% for the TCW croplands. This was because organic N transport is affected by soil drainage as well as topographic (e.g., slope and k-factor) characteristics. All well-drained soils were classified as “Low” by the SVI runoff criteria when a slope value is < 4% (Table 2). However, SORN further classified these well-drained soils depending on a slope or K-factor. For example, the well-drained soils with a low slope (1.1%) were classified as “Low” class and those welldrained soils with a high slope (2.9 or 3.0%) as “Moderate” or “Moderately High” classes (Table S3). 60% and 40% of well-drained soils were classified as “Low” and “Moderate” classes, respectively, by SORN, while 100% of those well-drained soils as “Low” by the SVI and SNIT (Tables 7 and 8). Thus, the consistency rate between SVI and SORN (70%) was lower compared to SNIT (81%) for the TCW croplands. Due to overall low slope and K-factor in the poorly-drained soil of the GW croplands (Table S2), SORN classified 35% and 56% of those poorly-drained soils as “Low” and “Moderate”, respectively, while 19% and 81% of those poorly-drained soils as “Moderate” and “Moderately high” by SNIT (Table 8). Therefore, an increased consistency was observed on SORN (46%) compared to SNIT (34%) in the GW croplands because “Low” class characterized the majority of poorly-drained soils by the SVI runoff criteria. Moreover, there was a great disagreement (i.e., 56% of the poorly-drained soils classified as “Moderate” by SORN and 94% of the poorly-drained soils as “Low” by SVI, Table 8) for the GW croplands while a great agreement (i.e., 64% of the well-drained soils classified as “Low” by SORN and 100% of the well-drained soils as “Low” by SVI, Table 7) for the TCW croplands. Therefore, the TCW

3.2.2. Comparison results of the SVI leaching criteria and SLEA The classification consistency of the SVI leaching criteria was 14.4% for the TCW croplands and 71.4% for the GW croplands (Tables 10 and 11). The consistency rate was substantially low in the TCW croplands. The majority of well-drained soils in the TCW croplands indicated a high leaching vulnerability (“High”) when classified by the SVI leaching criteria while low vulnerability (“Moderate”) according to SLEA classification (Table 12). In addition, poorly-drained soils in the TCW croplands represented “Low” class when classified by the SVI leaching criteria, but SLEA classified the majority of poorly-drained soils as “Moderate” or Moderately High” classes (Table 12). Due to 79% of welldrained soils with high gravitational water, overall nitrate leaching vulnerability was reduced for the TCW croplands dominated by welldrained soils when classified by SLEA while nitrate leaching vulnerability in poorly-drained soils increased due to low gravitational water (Table S4). These two changes collectively resulted in a low consistency between the SVI and SLEA for the TCW croplands. In contrast, gravitational water was high in the poorly-drained soils of GW croplands (Table S4) and therefore the majority of poorly-drained soils in the GW croplands were consistently classified as “Low” class by the SVI leaching criteria and SLEA (Table 13). The well-drained soils in the GW croplands with a low slope and high K-factor were mainly classified as

Table 6 Contingency table for areal proportions for between the SVI runoff and SWAT classification methods the Greensboro Watershed (GW) croplands. SVI runoff Low Moderate Moderately high High

Low 32.2 0 0 0

Low Moderate Moderately high High

43.5 0.0 0.0 0.0

Nitrate (SNIT) Moderate Moderately High 12.4 51.1 0.2 1.5 0.7 2.0 0 0 Organic N (SORN) 48.6 3.5 0.2 1.4 0.8 1.9 0.0 0.0

High 0 0 0 0 0.0 0.0 0.0 0.0

Note: the sum of numbers in gray boxes indicates accuracy rate. 7

Catena 167 (2018) 1–12

S. Lee et al.

Table 7 Soil type and the two N fluxes transported by surface runoff for classes defined by the SVI runoff criteria and SWAT outputs for the TCW croplands. Soil type (%)

Nitrate (kg/ha)

Organic N (kg/ha)

Well-drained

Poorly-drained

Median

Min

Max

Median

Min

Max

SVI runoff Low Moderate Moderately high High

69.46 (100) 0.00 (0) 0.00 (0) 0.00 (0)

17.35 (57) 10.63 (35) 2.52 (8) 0.04 (0)

0.72 1.42 1.49 0.58

0.23 1.20 1.05 0.44

2.04 1.46 1.96 0.72

10.0 22.47 27.52 38.72

1.14 13.88 10.81 32.14

31.98 30.40 45.90 45.31

SWAT Nitrate (SNIT) Low Moderate Moderately high High

69.46 (100) 0.00 (0) 0.00 (0) 0.00 (0)

0.04 (0) 28.36 (93) 1.92 (6) 0.22 (1)

0.69 1.45 1.96 2.00

0.2 0.74 1.59 2.00

0.73 1.54 2.00 2.00

– – – –

– – – –

– – – –

SWAT Organic N (SORN) Low Moderate Moderately high High

44.49 (64) 24.44 (35) 0.54 (1) 0.00 (0)

14.44 (47) 14.04 (46) 1.90 (6) 0.17 (1)

– – – –

– – – –

– – – –

8.39 20.67 31.66 45.90

1.14 14.20 27.6 45.90

14.10 27.52 45.82 45.90

Note: A value outside and within a parenthesis in soil type indicates the relative proportion to the total area and each soil type, respectively. There was only one HRU classified as “High” class based on SWAT nitrate and organic N. Table 8 Soil type and the two N fluxes transported by surface runoff for classes defined by the SVI runoff criteria and SWAT outputs for the GW croplands. Soil type (%)

Nitrate (kg/ha)

Organic N (kg/ha)

Well-drained

Poorly-drained

Median

Min

Max

Median

Min

Max

SVI runoff Low Moderate Moderately high High

32.77 (100) 0.00 (0) 0.00 (0) 0.01 (0)

62.86 (94) 1.64 (2) 2.72 (4) 0.00 (0)

2.92 3.96 6.79 2.08

0.85 3.60 4.26 1.57

9.81 6.86 9.79 2.10

21.40 32.23 43.26 44.58

1.32 30.20 10.32 42.21

60.66 44.25 76.32 50.90

SWAT Nitrate (SNIT) Low Moderate Moderately high High

32.20 (98) 0.58 (2) 0.00 (0) 0.00 (0)

0.00 (0) 12.67 (19) 54.55 (81) 0.01 < (0)

2.36 4.31 7.29 9.81

0.85 3.28 5.91 9.81

3.10 5.86 9.79 9.81

– – – –

– – – –

– – – –

SWAT Organic N (SORN) Low Moderate Moderately high High

20.29 (62) 11.90 (36) 0.58 (2) 0.00 (0)

23.21 (35) 37.78 (56) 6.22 (9) 0.01 < (0)

– – – –

– – – –

– – – –

16.08 30.31 46.24 76.32

1.32 23.07 40.15 76.32

23.00 40.06 69.75 76.32

Note: A value outside and within a parenthesis in soil type indicates the areal proportion to the total area and each soil type, respectively. There was only one HRU classified as “High” class based on SWAT nitrate and organic N. The HRUs on well-drained soils were rated as “High” by the SVI, due to extremely high slope values (> 7).

“Moderate” by the SVI leaching criteria (Table 3) and SLEA also classified the majority of those well-drained soils as “Moderate” due to high gravitational water (Tables 13 and S4), which also contributed to a high consistency between the SVI leaching criteria and SLEA for the GW croplands.

Table 9 Classified cropland HRUs by the SVI leaching criteria and SWAT leaching output. Classes

Tuckahoe Creek Watershed

Greensboro Watershed

Area (ha)

Proportion (%)

Area (ha)

Proportion (%)

SVI Low Moderate Moderately high High

3255 2642 1984 4098

27 22 17 34

5910 2961 278 1274

57 28 3 12

SWAT (SLEA) Low Moderate Moderately high High

260 9024 2694 0*