Sediment retention by natural landscapes in the conterminous United States

Abstract

Soils provide vital ecosystem services, from sequestering carbon to providing food and moderating floods. Soil erosion threatens the provisioning of these services and degrades downstream water quality. Vegetation plays an important role in soil retention: by holding it in place, soil can continue to provide ecosystem goods and services and protect water resources. The aims of this study were to: (1) develop a 30-meter resolution map of erosion in the conterminous United States, and (2) quantify the soil retention service of natural vegetation. Using the Revised Universal Soil Loss Equation and physiographic and remote sensing datasets, we estimated sheet and rill erosion. We also developed a map of sediment delivery ratio to connect erosion to downstream delivery using hydrologic connectivity. The estimated sheet and rill erosion in the conterminous United States was 1.55 Pg yr⁻¹, of which 0.52 Pg yr⁻¹ reached waterbodies. Natural land cover prevents 12.3 Pg yr⁻¹ of sheet and rill erosion and 5.1 Pg yr⁻¹ in delivery to waterbodies. The value of natural land cover in retaining sediment is a function of the land cover, physiographic characteristics, and spatial context. This study has implications for spatial prioritization of natural land cover preservation and agricultural land management to minimize sediment erosion and delivery.

1. Introduction

Goods and services provided by soil extend well beyond the obvious and intuitive – an irreplaceable medium needed to meet humanity’s nutritional requirements. The global soil resource is the third largest pool of planetary carbon, with substantial potential to serve as a carbon sink (Lal et al., 2015). Soils moderate floods by storing and slowly releasing water to streams. Soils are recognized for their high biological diversity (Barrios, 2007). Biomass, a correlate of species richness and diversity is very high over very small areas (Pimentel, 2006; Mulder et al., 2011). The fungi, bacteria, and invertebrates that comprise soil biota are critical to soil ecological processes (Barrios, 2007). Soil invertebrate activity contributes to soil structure, and hence facilitates root growth and water holding capacity (Lavelle et al., 2006). Perhaps overlooked (except by soil scientists) are the cultural services supplied by soil (Dominati et al., 2010). Most recreational experiences occur on a foundation of soil.

Not surprisingly, a recast of soil science into an ecosystem goods and services framework followed the release of the Millennium Ecosystem Assessment, which evaluated the consequences of anthropogenic activities on ecosystems and the need for sustainable and conservation-focused efforts (MEA, 2005). Adhikari and Hartemink (2016) report a rapid expansion in the number of publications on framing soil and associated processes as ecosystem goods and services. Some ecosystem goods and services frameworks for soils place greater emphasis on the natural capital of soils (Dominati et al., 2010, 2014). That is, the natural capital (soil) must exist for humans to derive benefits from it. A “dis-service” occurs when rates of soil erosion (loss) exceed rates of pedogenesis (soil formation) (Montgomery, 2007; Pimentel and Burgess, 2013). This emphasis on natural capital links ecosystem goods and services to the concept of sustainability more seamlessly and more intuitively.

Soil erosion is considered a serious worldwide problem that can be viewed as an environmental crisis when considered in the context of feeding an ever-expanding global population (Pimentel, 2006; Powlson et al., 2011; Pimentel and Burgess, 2013). In the United States, soil erosion rates are much lower than in other parts of the world, but still exceed estimated rates of pedogenesis (Pimentel and Burgess, 2013). The United States Department of Agriculture (USDA) Natural Resource Conservation Service (NRCS) reported a ~35% decline in soil erosion rates between 1982 and 2007 (Pimentel and Burgess, 2013). The decline may be, in part, attributable to The Food Security Act of 1985 (Sullivan et al., 2004). The Conservation Reserve Program (CRP), which came into being with passage of the 1985 Food Security Act, was instituted by the USDA to pay farmers to remove agricultural land from production on highly erodible soil. The CRP was USDA’s first program designed to motivate farmers to establish practices that foster soil sustainability; there had been other (earlier) programs that paid farmers for other actions, but none that were designed to promote sustainability.

There have been several efforts to map soil erosion at continental or global extents at high resolution. Panagos et al. (2015c) used the Revised Universal Soil Loss Equation (RUSLE) (Renard et al., 1997) to estimate soil loss in the European Union at 100-meter resolution, with a focus on policy scenarios that resulted in land-use changes and establishment of soil conservation practices. More recently, Borrelli et al. (2017) produced a global map of sheet and rill soil erosion at a 250-meter resolution using RUSLE, integrating land use change between 2001 and 2012, and potential erosion offsets by conservation practices. Mapping soil loss also has the potential to be coupled with other models to help quantify other ecosystem services, such as carbon storage (Doetterl et al., 2012; Lugato et al., 2018), and economic impact (Sartori et al., 2019). However, these studies do not consider connectivity and the ecosystem service value of natural land cover in preventing soil export to waterbodies.

With soils, sustainable use underpins the provision of ecosystem goods and services. Our aim is to produce a national map of soil retention, and illustrate how it can quantify the role of natural vegetation in soil retention considering both soil loss and connectivity to waterbodies. We hypothesize that value of natural vegetation in soil retention is a function of its physiographic and spatial context, and thus, variable across the conterminous United States (CONUS). The resulting map serves as an ecosystem service indicator for the U.S. EPA EnviroAtlas (https://www.epa.gov/enviroatlas). The EnviroAtlas provides geographic data and tools (models) that stakeholders can use to evaluate provision of ecosystem goods and services and sustainable use of the environment for their region of interest (Pickard et al., 2015).

2. Materials and Methods

Our long-term average annual soil retention estimates are based on RUSLE and linked to a landscape-based sediment delivery ratio (SDR) model that accounts for hydrologic connectivity (Borselli et al., 2008) to model sediment yield to waterbodies. These methods are derived from those used in the InVEST (Integrated Valuation of Environmental Services and Tradeoffs) sediment retention model (Hamel et al., 2015; Sharp, 2018) and extended to the conterminous United States.

We used RUSLE (Renard et al., 1997) to develop a 30-meter resolution estimate of sheet and rill erosion by water in the CONUS (Equation 1).

$$A=R\times K\times LS\times C\times P$$ (1)

where A is the average annual soil loss (Mg ha⁻¹ yr⁻¹), R is the annual rainfall-runoff erosivity factor (MJ mm ha⁻¹ hr⁻¹ yr⁻¹), K is the soil erodibility factor (Mg ha hr ha⁻¹ MJ⁻¹ mm⁻¹), LS is the slope length and steepness factor (dimensionless), C is the cover-management factor (dimensionless), and P is the support practice factor (dimensionless). The equation does not predict gully, ephemeral gully, or bank erosion.

In addition, the RUSLE does not estimate the eroded soil delivered beyond a field, i.e. the sediment yield (Renard et al., 1997). To estimate sediment yield (Mg ha⁻¹ yr⁻¹), a sediment delivery ratio (SDR, dimensionless) was calculated for each pixel and multiplied by the average annual soil loss (A) for each pixel. The following sections describe derivation of each RUSLE factor and the SDR for calculation of sediment yield and Table 1 contains the data sources used to derive each factor.

Table 1.

RUSLE factors and data sources

Factor	Input data sources
R	PRISM Climate Group precipitation (Daly and Taylor, 2002)
K	SSURGO/STATSGO2
LS	USGS 3D Elevation Program
C	MODIS NDVI; 2011 National Land Cover Database (NLCD)
P	Not applicable
SDR	USGS 3D Elevation Program; National Hydrography Dataset Plus Version 2 (NHDPlusV2)

2.1. Rainfall-runoff erosivity factor

The rainfall-runoff erosivity factor (R) accounts for the effect of raindrop impact and the amount and rate of runoff associated with the rainfall, represented by total storm energy and maximum 30-minute intensity in Equation 2 (Renard et al., 1997).

$$R=\sum E_{storm}{I}_{30}$$ (2)

Where R is the rainfall erosivity factor (MJ mm ha⁻¹ h⁻¹ yr⁻¹), E_storm is the total storm energy (MJ ha⁻¹), and I₃₀ is the maximum 30-minute intensity (mm h⁻¹). Mean annual R for 1971–2000 at 4 km resolution was produced by the PRISM Climate group using station-based 15-minute temporal resolution rain gauge data and the PRISM spatial regression method (Daly and Taylor, 2002; Hollinger et al., 2002). The R factor is generally highest in the eastern half of the US, with values reaching 11,000 MJ mm ha⁻¹ h⁻¹ yr⁻¹ along the Gulf of Mexico, although values of this magnitude also occur in the Cascades and Sierra Nevada mountains. The western US is characterized by low R values consistently less than 1000 MJ mm ha⁻¹ h⁻¹ yr⁻¹ and a minimum of 34 MJ mm ha⁻¹ h⁻¹ yr⁻¹.

2.2. Soil erodibility factor

The soil erodibility factor (K) represents susceptibility to erosion by runoff and raindrop impact. This value (kffact) was extracted from the Soil Survey Geographic Database (SSURGO), with gaps in SSURGO being filled with the State Soil Geographic Database (STATSGO2). Values in the SSURGO database are presented in U.S. customary units (ton ac h 100⁻¹ ac⁻¹ ft⁻¹ tonf⁻¹ in⁻¹), ranging from 0 to 0.64, with higher values reflecting greater erodibility. In SI units, K ranges from 0 to 0.0842 Mg ha h ha⁻¹ MJ⁻¹ mm⁻¹.

2.3. Slope length and steepness factor

The slope length and steepness (LS factor) represents the influence of slope length (L) and slope steepness (S) on erosion. We derived the LS factor from a 30-meter digital elevation model (DEM) from the U.S. Geological Survey 3D Elevation Program. The Desmet and Govers (1996) LS factor method was implemented using the System for Automated Geoscientific Analysis (SAGA) LS factor, Field Based module (Conrad et al., 2015). This method uses a multiple flow direction algorithm that is a function of the unit-contributing area, along with pixel aspect direction in calculating LS. It has been successfully applied at 25-meter resolution for Europe (Panagos et al., 2015a). The LS factor was limited to a maximum of 72.15 as specified by RUSLE (Renard et al., 1997).

2.4. Cover-management factor

The cover-management factor (C) typically represents the effects of agricultural cropping and management practices on erosion rates (Renard et al., 1997). Lower values indicate more cover and lower erosion potential. The C factor is often developed as a weighted average of the soil loss rate (SLR) over the year, which itself is a function of prior land use, canopy cover, surface cover, surface roughness, and soil moisture. A constant SLR (and therefore C factor) is applicable for land cover classes that are typically not managed.

Deriving C factors typically requires extensive knowledge of a study area’s cover characteristics and management practices, which makes it difficult to characterize at scales larger than a field or farm (Benavidez et al., 2018). Multiple studies have used the Normalized Difference Vegetation Index (NDVI) to quantify the C factor by assuming that as greenness decreases, erosion potential increases (Van der Knijff et al., 2000; de Asis and Omasa, 2007; Suriyaprasit and Shrestha, 2008; Ma et al., 2010; Durigon et al., 2014; Li et al., 2014). The advantage of using NDVI for large-scale studies is that it can account for spatial and temporal variations in cover, rather than using literature-based C values designed for specific conditions or locations (Benavidez et al., 2018).

We used MODIS (Moderate Resolution Imaging Spectroradiometer) NDVI values at 250 m resolution for 16-day intervals between 2000–2014 to calculate a mean annual NDVI. The NDVI was downscaled to 30 m resolution using the 2011 National Land Cover Database (NLCD) (Homer et al., 2015) as follows: (1) median NDVI was calculated for all NLCD land cover classes in each of 73 USDA crop management zones (CMZs) (USDA, 2019), (2) the median NLCD-level NDVI was applied to corresponding NLCD pixels in that land cover class/CMZ combination. To relate NDVI to C, we used the generalized equation developed by Van der Knijff et al. (2000) for the European Union (Equation 3), where α=2, β=1, and γ=1.

$$C={\gamma e}^{\left(-\alpha \frac{NDVI}{\beta -NDVI}\right)}$$ (3)

The α, β, and γ parameters were used to calibrate RUSLE average annual soil loss to the USDA 2015 National Resources Inventory (NRI) state-level cropland erosion by water estimates. Rather than using a single parameter set, calibration was performed spatially for the 73 CMZs. A total of 200,000 parameter sets were created using Latin hypercube sampling, with each set containing 219 parameters (a unique α, β, and γ for each CMZ). Parameter ranges were set to: α = [1, 8], β = [1, 2], and γ = [0.2, 1] for cropland and α = [1, 8], β = [1, 2], and γ = [0.01, 0.7] for pasture. Spatially stratified samples of R, K, LS, and NDVI were drawn from each state’s pasture/hay and cultivated crops NLCD classes, with a minimum of 1000 samples per state and class (number of samples were proportional to amount of pasture/hay and cultivated crops in the state). The RUSLE predicted average annual soil loss (Mg ha⁻¹) calculated using each parameter set were compared to state-level NRI sheet and rill erosion by water estimates using the mean absolute error (MAE). Rather than using a single best parameter set, an ensemble average of parameter sets was used, in which the number of ensemble members was determined by minimizing the MAE. This process was completed separately for cultivated crops and pasture/hay NLCD classes, resulting in unique α, β, and γ for each class in each CMZ. State level ensemble mean performance and comparisons with NRI are presented in Figs. 1a and 1b. The optimal number of simulations (lowest MAE) to include in calculation of the ensemble mean C factor was 29 for cropland (MAE = 1.13 Mg ha⁻¹) and 25 for pasture (MAE = 0.40 Mg ha⁻¹) in Figs. 1c and 1d.

Fig 1.

State-level comparison of RUSLE versus NRI using ensemble mean for (a) cropland and (b) pasture, minimizing MAE for (c) cropland and (d) pasture. Note the differing scales in a vs. b and c vs. d. Black dot in (c) and (d) represents minimum ensemble mean MAE.

While using NDVI to calibrate C factor for cropland is advantageous in simulating spatial and temporal differences in cover, Equation 3 typically resulted in unrealistically high values for forest, grassland, and wetlands, which was also noted by Van der Knijff et al. (2000). This was also the case for developed NLCD classes when compared to literature values. Therefore, C factors for non-agricultural NLCD classes were derived from literature, with values and sources presented in Table 2. The NDVI for NLCD barren land was used to calculate the C factor, using Equation 3 parameters for cultivated crops.

Table 2.

Final C factors per NLCD class, and source if derived from literature. Ranges of mean C factor values for each CMZ and area-weighted means across all CMZs are presented for Barren Land, Pasture/Hay, and Cultivated Crops, derived using Equation 3.

Class/Value	Classification	C factor	Source
11	Open Water	0	(Linard et al., 2014; Panagos et al., 2015b)
12	Perennial Ice/Snow	0	(Linard et al., 2014; Panagos et al., 2015b)
21	Developed, Open Space	0.03	(Fernandez et al., 2003; Linard et al., 2014)
22	Developed, Low Intensity	0.03	(Fernandez et al., 2003; Linard et al., 2014)
23	Developed, Medium Intensity	0.03	(Fernandez et al., 2003; Linard et al., 2014)
24	Developed, High Intensity	0.03	(Fernandez et al., 2003; Linard et al., 2014)
31	Barren Land	0.29 (0.02–0.52)	Calibrated Equation 3
41	Deciduous Forest	0.002	(Linard et al., 2014)
42	Evergreen Forest	0.002	(Linard et al., 2014)
43	Mixed Forest	0.002	(Linard et al., 2014)
52	Shrub/Scrub	0.04	(Linard et al., 2014)
71	Grassland/Herbaceous	0.01	(Linard et al., 2014)
81	Pasture/Hay	0.07 (0.008–0.21)	Calibrated Equation 3
82	Cultivated Crops	0.17 (0.02–0.31)	Calibrated Equation 3
90	Woody Wetlands	0.001	(Linard et al., 2014)
95	Emergent Herbaceous Wetlands	0.001	(Linard et al., 2014)

2.5. Support practice factor

The support practice factor (P) reflects agricultural management practices that reduce soil erosion. Although the P factor is not explicitly used here given the lack of a national database, we expect that the impact of some support practices (e.g., winter cover crops, no-till, etc.) will be reflected in the calibrated C-factor using NDVI on cultivated cropland and pasture.

2.6. RUSLE validation

We compared our soil loss results with those from the Daily Erosion Project (DEP) (Gelder et al., 2018). The DEP estimates daily hillslope erosion for HUC-12s in Iowa and parts of Minnesota, Nebraska, Kansas, and Missouri. The DEP uses the Water Erosion Prediction Project (WEPP) hillslope model with high temporal resolution Next-Generation Weather RADAR (NEXRAD) precipitation, and crop specific parameters such as C and P factors obtained from the confidential NRI database (Gelder et al., 2018).

The comparison between RUSLE and DEP was made for HUC-12s with greater than 75% agricultural land cover (Fig. 2). This threshold was used because DEP only models agricultural erosion, while our HUC-12 estimates include all land cover types. The mean absolute difference between the two methods is 2.77 Mg ha⁻¹ yr⁻¹. However, these estimates are not a one-to-one comparison. Beyond the differences in agriculture/non-agriculture, DEP weather data used for comparison was from 2010–2019 on an event basis, while R is a long-term annual factor derived from 1971–2000 PRISM data.

Fig. 2.

Comparison of average annual soil loss using RUSLE (1971–2000) versus the Daily Erosion Project (DEP, 2010–2019) for HUC-12s in Iowa, Kansas, Minnesota, Missouri, and Nebraska.

2.7. Sediment delivery ratio (SDR)

RUSLE only predicts soil loss and does not account for off-site sediment yield. Therefore, the SDR was applied to the soil loss estimates from RUSLE to calculate estimated net sediment yield to downstream waterways. The SDR was calculated for each pixel in the CONUS using flow path connectivity (Borselli et al., 2008). The methods used here are adapted from previous work by Hamel et al. (2015, 2017).

The Index of Connectivity (IC), defined here as a given pixel’s connectivity to the nearest waterbody, was calculated using the ArcGIS Connectivity Index Toolbox (Cavalli et al., 2014, 2013) based on the methods of Borselli et al. (2008). The IC is based on the upslope (D_up) and downslope (D_down) characteristics from pixel i, as defined in equation 4.

$${IC}_i={\log }_{10}\left(\frac{D_{up}}{D_{down}}\right)={\mathrm{l}\mathrm{o}\mathrm{g}}_{10}\left(\frac{\overline{CS}\sqrt{A}}{\sum _i\frac{d_i}{C_i{S}_i}}\right)$$ (4)

Where for the ith pixel, $\stackrel{-}{C}$ is the average C factor of the upslope contributing area, $\stackrel{-}{S}$ (m/m) is the average slope of the upslope contributing area, A (m²) is the upslope contributing area, d_i (m) is the length of the flow path along the ith pixel in the steepest downslope direction, C_i is the C factor of the ith pixel, and S_i is the slope gradient of the ith pixel. D-infinity flow direction was used to calculate upslope contributing area in TauDEM (Tarboton, 1997). The downslope flow path was calculated using D8 flow direction along the steepest downward slope.

The C factor for developed land cover was adjusted for Equation 4. Adjustment was based on the disparity of low sediment generation by impervious surfaces (reflected in low C factor, Table 2), versus the greater potential for connectivity due to those impervious surfaces. Therefore, the NLCD Urban Imperviousness 2011 dataset (Xian et al., 2011) was used to represent developed land cover C factors in Equation 4. Imperviousness in the original dataset ranges from 0–1; these values were used directly for the developed land cover C factor in Equation 4, allowing them to vary from pixel to pixel. The mean C factor values increased with increasing density: 0.05 (developed, open space), 0.3 (developed, low intensity), 0.61 (developed, medium intensity), 0.88 (developed, high intensity).

Sinks (delivery endpoints) were defined using the National Hydrography Dataset Plus Version 2 (NHDPlusV2) streams and waterbodies. Waterbodies not identified as sinks (i.e. not included in the NHDPlusV2), such as smaller wetlands and streams, still have high estimates of connectivity based on the definition of IC, which accounts for the ratio of the square root of upstream area to distance to a downstream sink. The SDR for pixel i is then calculated using Equation 5.

$${SDR}_i=\frac{{SDR}_{max,i}}{1+exp\left(\frac{{IC}_0-{IC}_i}{k_b}\right)}$$ (5)

Where SDR_max,i is the maximum possible delivery ratio for the ith pixel, defined as the fraction of topsoil particles finer than coarse sand (less than 1 mm in diameter) (Vigiak et al., 2012). The SDR_max,i values were derived from SSURGO and STATSGO2 data. In rare cases where data were missing we set SDR_max,i = 0.8 (Vigiak et al. 2012). IC₀ and k_b are calibration parameters. Both Borselli et al. (2008) and Vigiak et al. (2012) found a best fit with IC₀ equal to 0.5 across differing study areas, while Hamel et al. (2015) noted its insensitivity. We set k_b equal to 2 (Vigiak et al. (2012). Sediment yield from each pixel was then calculated as the product of RUSLE’s average annual sheet and rill erosion (A) and the SDR.

By including all the components of calculating RUSLE and SDR, it is expected that areas with some combination of steep slopes, fine soil particles, low soil cover, and high rainfall intensity will experience the greatest sheet and rill erosion. Areas with steep slopes, lower soil cover, and close proximity to streams will have high connectivity and as a result high SDR.

2.8. Estimating the sediment retention benefits of natural land cover

To determine how much sediment is retained by natural land cover, the analysis with existing vegetative cover was compared to an alternative scenario where current natural vegetative cover was “removed.” Alternative vegetative covers were modeled by replacing the C factors for forest (NLCD 41, 42, 43), shrub (NLCD 52), grassland/herbaceous (NLCD 71), and wetland (NLCD 90, 95) classes with the maximum C factor from barren land, applied uniquely for each CMZ. Maximum barren land C factors ranged from 0.26 to 0.38. Sediment delivery is also impacted by the alternative cover, because SDR is a function of the C factor (see Equations 4 and 5). The difference between the existing vegetative cover and the alternative vegetative cover gives an estimate of average annual sediment retention by natural vegetative cover. The outcome represents a worst-case scenario of soil loss and sediment yield – removal of natural land cover everywhere.

3. Results

3.1. Sediment erosion and yield

About one-third of the estimated soil loss was delivered to water bodies. The estimated average annual rate of soil erosion by water was 2.14 Mg ha⁻¹ yr⁻¹, totaling 1.55 Pg yr⁻¹. When accounting for delivery to waterbodies using the SDR, the estimated sediment yield was 0.72 Mg ha⁻¹ yr⁻¹, yielding 0.52 Pg yr⁻¹ delivered to waterbodies. These estimates exclude lands that are considered non-erodible (NLCD open water and perennial ice/snow). On average, soil loss and sediment yield were greatest in the agricultural Corn Belt and in areas with complex topography along the Pacific coast (Fig. 3).

Fig. 3.

(a) estimated average annual soil loss (Mg ha⁻¹ yr⁻¹), (b) estimated average annual sediment yield (Mg ha⁻¹ yr⁻¹). Values in both panels are aggregated by HUC-12 watersheds.

Soil loss varies greatly by land cover (Fig. 4), with barren land soil loss (20.6 Mg ha⁻¹ yr⁻¹) and cultivated crops (5.6 Mg ha⁻¹ yr⁻¹) being the greatest. In contrast, the two wetlands classes (woody and emergent herbaceous) combine for a soil loss rate of only 0.03 Mg ha⁻¹ yr⁻¹.

Fig. 4.

Average annual soil loss and sediment yield by NLCD class

Soil loss rates varied spatially as a function of regional and local physiographic and climatic conditions, which was reflected in differences in cropland and pasture erosion rates across the US (Fig. 1), as well as the high values that appear within the agricultural Midwest (Fig. 3). Cultivated crops and shrub/scrub contribute the most to soil loss, at 0.68 Pg yr⁻¹ and 0.37 Pg yr⁻¹, respectively, due to their large land areas and high C factors representing relatively exposed soil. However, most detached soil recompacts or is deposited on the landscape and does not reach waterbodies. By including connectivity, only one-third of soil loss from cultivated crops (0.24 Pg yr⁻¹) and shrub/scrub (0.12 Pg yr⁻¹) is estimated to reach waterbodies.

3.2. Sediment retention by vegetation

The sediment retention value of natural land cover was determined by replacing the C factors for NLCD natural vegetation classes (forest, shrubland, and wetland) with the barren land C factor for each CMZ. The total sediment retention benefit of natural land cover was 12.3 Pg yr⁻¹, which yields 5.1 Pg yr⁻¹ in avoided delivery to waterbodies. These values are almost ten times greater than the total amount of sheet and rill erosion that occurs annually. As expected, the spatial variation in avoided sediment loss and delivery follows the spatial pattern of natural vegetation; the greatest sediment loss and delivery avoidance occurs where natural vegetation dominates (Fig. 5). The forested land cover classes provide the greatest value for soil loss and sediment yield, while wetland land cover provides the least value (Fig. 6). Evergreen forest (1.94 Pg yr⁻¹ avoided) and deciduous forest (1.16 Pg yr⁻¹ avoided) prevent the most sediment yield entering waterbodies, compared to emergent herbaceous wetlands preventing 0.01 Pg yr⁻¹.

Fig. 5.

(a) Avoided annual soil loss (Mg ha⁻¹ yr⁻¹) and (b) sediment yield (Mg ha⁻¹ yr⁻¹) by natural vegetation. Values in both panels are aggregated by HUC-12 watersheds.

Fig. 6.

Avoided annual soil loss and sediment yield by natural land cover.

4. Discussion

4.1. Valuing natural land cover in sediment retention

The value of natural land cover in sediment retention is spatially variable, and a function of the physiographic context (topography, soil, climate), spatial arrangement of land cover, and proximity to waterbodies. There are regional trends and physiographic conditions that dictate where natural land cover holds the greatest value in sediment retention. These areas have relatively large LS factors (complex topography and high slopes) and/or large R factors (those areas that experience intense rainfall). This is supported by Doetterl et al. (2012), who identified slope steepness (S) and rainfall erosivity (R) as the primary controls on continental scale erosion rates in the US. Across these landscapes, barren land has high erosion rates, and thus, natural land cover holds a higher value in what it is able to retain.

Forested lands provide the greatest value in sediment retention among all the natural land cover types. While wetlands and forests have similar C factors, wetlands are typically located in topographically flat areas with low LS factors that do not generate much soil loss. In the scenario we model, when wetlands are replaced by barren land cover, the sediment erosion is still low due to the topographic conditions. The value only reflects a pixel-by-pixel replacement of the natural land cover with barren land and does not consider how wetlands retain sediment (Phillips, 1989; Craft and Casey, 2000; Cohen et al., 2016) and reduce downstream river bed and bank erosion, among many other ecosystem services. This is the case in the Midwest USA, where draining of wetland depressions for row-crop agriculture with artificial drainage created more erosive rivers, increasing the contribution of non-field sediment (Schottler et al., 2014), a phenomenon not captured by the scenario. However, the low soil loss and sediment yield of wetlands (Fig. 4) demonstrate how wetlands reduce sediment connectivity and delivery from upstream areas. It is important to note that the results here only account for soil loss, and neither sinks nor net sediment yield (pixel-scale deposition vs. loss) are quantified.

Forests occupy a variety of complex topographies with varying levels of erosion susceptibility, leading to high sediment retention when they are located on steep areas with high LS factors or low delivery in riparian areas. The differences in sediment retention within forest classes as modeled here are due to geographic location and topography. Evergreen forest has greater value than deciduous and mixed forest because it is located on steep slopes in the western US and in some areas of the southeastern US. Riparian forests also hold relatively high value because of their proximity to waterbodies; these locations experience increases in SDR when they are replaced. This is particularly apparent in headwater areas with relatively large LS and R factors, such as the southeastern US.

Agriculturally-dominant areas have relatively low sediment retention values, because much of the land is not natural vegetation (Mississippi Delta, Central Valley in California, and the Corn Belt). These areas currently generate a majority of the sheet and rill erosion and sediment yield in the United States. We expect that the sediment retention value would be high if natural vegetation was restored on cultivated cropland; Nearing et al. (2017) noted that the U.S. CRP, which takes cropland out of production in favor of perennial plant cover, reduces average annual erosion rates to about 1 Mg ha⁻¹ yr⁻¹.

The SDR also drives the value of natural land cover in preventing sediment yield to waterbodies. In locations with a high SDR, characterized by high slopes and close proximity to waterbodies, the value of natural land cover is higher as the change in C factor modulates the sediment yield. The IC also changes with the mean C factor of the upslope contributing area; as natural land cover is lost upstream, it increases downstream connectivity and subsequent SDR. The more natural land cover that is lost in the watershed, the greater the increase in SDR throughout that watershed. Given these characteristics of IC and SDR, the value of natural land cover is greatest in areas where SDR is largest due to high slopes, high upstream and downstream C factors, and proximity to receiving waterbody—leading to conditions where the percentage of eroded soil delivered to a waterbody is high. This is consistent with Chaplin-Kramer et al. (2016), demonstrating that landscape configuration, particularly proximity to streams, can result in order of magnitude differences in erosion potential and sediment export. Understanding the conditions that reflect high connectivity and sediment delivery can be used at a screening level to identify natural areas that provide the greatest benefit in preservation or in targeting high erosion areas to be taken out of agricultural production.

The value of natural land cover in sediment retention does not consider spatial suitability, vulnerability, or likelihood of transition away from natural land cover. Rather, it is a broad estimate of what sheet and rill erosion would be if natural land cover is lost. For example, low productivity marginal land with constraints on agricultural cultivation (Kang et al., 2013), where marginal land is often on steep slopes and erosive soil, resulting in variable spatial likelihood that that natural land cover would be lost to agriculture. In wildfire prone areas of the western US, variation in soil moisture and vegetation as a function of hillslope aspect influences fuel density and subsequent fire intensity and erosion (Abrahams et al., 2018). Coupling likelihood of natural vegetation loss with value in sediment retention would estimate the risk of the loss of the service provided across land cover types. Including likelihood of natural land cover loss would place greater risk and value to areas more suitable for transition, such as wetlands in the midwestern US rather than forests on steep slopes in the western US.

4.2. Model performance, uncertainty, and limitations

Using RUSLE and the SDR across large spatial extents presents challenges regarding uncertainty in estimating sediment erosion and delivery to waterbodies. RUSLE does not quantify gully or stream erosion, and it was initially developed for use on only cropland and pasture at plot scales (Renard et al., 1997), but has been since applied at broad scales and across varying land cover, use, management, climatic, and topographic conditions with satisfactory results. The RUSLE and SDR input parameters are inherently uncertain and in some respects are a function of the spatial resolution of the data (Hamel et al., 2015). Vigiak et al. (2012) notes that parameter sensitivity in the SDR to the DEM is low because the IC is a ratio, but the DEM resolution does affect model predictions and contribute to uncertainty (Hamel et al., 2017).

To mitigate some uncertainty in the RUSLE parameters, we used previously reported relationships between C factor and NDVI (Van der Knijff et al., 2000) as a calibration parameter to fit our RUSLE cropland and pasture/hay predictions to published NRI data. While this improves predictions, it does not reduce uncertainty in soil loss for natural landscapes, for which there was not a comprehensive source of erosion estimates. This necessitated the use of literature values for natural land C factors. In terms of uncertainty and sensitivity, literature-based C factors for forest, shrub, and wetlands are orders of magnitude lower than those of barren land and cultivated crops. The sensitivity analysis performed in Hamel et al. (2015) demonstrates that as baseline C factors decrease, so does the potential change in sediment export. This means that sediment export is more sensitive to larger C factors (i.e. agriculture and barren). A global study of soil erosion found that uncertainty is generally lower in forested locations and higher in those areas that are subject to higher erosion rates, including barren land and agricultural areas of the United States (Borrelli et al., 2017). Across studies included in a soil erosion measurement meta-analysis, it was found that almost any erosion rate was possible regardless of the slope, climate, scale, and land use, demonstrating the high variability and uncertainty apparent even in measurement (García-Ruiz et al., 2015).

The estimate for cropland erosion rate in this study matched the NRI 2015 sheet and rill erosion on cropland (6.1±0.1 Mg ha⁻¹ yr⁻¹) and U.S. estimates by Borrelli et al (2017) using RUSLE at 250-meter for the globe. Calibrating our RUSLE soil loss estimates with existing state-level NRI data achieved an acceptable level of model performance for cropland and pasture. Access to the complete NRI database, rather than aggregated state-level data, would likely improve our spatial estimates of cropland and pasture soil loss and sediment yield. The values of non-cropland erosion (excluding barren land) are consistent with the Nearing et al. (2017) estimate of less than 2 Mg ha⁻¹ yr⁻¹ for natural non-cropped conditions. The lack of model sensitivity to natural land cover C factors reduces uncertainty in our predictions and is consistent with other studies (Hamel et al., 2015; Borrelli et al., 2017). Minor discrepancies in forest type have little influence on soil loss predictions (Fraser et al., 1995). Assigning small differences in C factors across forest types would have minor impacts on soil erosion due to the greater influence of R and LS.

4.3. Future opportunities

There are opportunities to increase the spatial and temporal resolution of soil loss and sediment yield while still using only publicly-available datasets. Emerging datasets and methods would (1) improve the spatial resolution and accuracy while reducing uncertainty of RUSLE and SDR input variables, and (2) introduce a temporal aspect beyond average annual estimates.

R factor:

Enhanced temporal resolution would better reflect climate seasonality impacts on soil loss and sediment yield, especially in agricultural fields that are more prone to erosion during certain times of the year. In addition, climate change will likely impact erosivity. Large areas of the United States will likely experience greater erosivity with an accompanied increase in annual variability, although there is uncertainty in the locations of change, methods to calculate future erosivity, and in the spread across climate simulations (Segura et al., 2014; Biasutti and Seager, 2015).

K factor:

soil erodibility was obtained from SSURGO/STATSGO2 databases, which have uncertainties due to scale, discontinuities across county boundaries, and map units with varying numbers of spatially unassigned soil components (Ramcharan et al., 2018). New probabilistic soil series maps developed at 30-meter (Chaney et al., 2016, 2019) and 100-meter (Ramcharan et al., 2018) resolutions may address these shortcomings and could be used to reduce uncertainty in the RUSLE soil loss estimates.

C and P factors:

Literature values of C factors vary widely by land cover type (Benavidez et al., 2018), and local management context is important. Emerging remote sensing products may provide opportunities for improvement. Remote sensing products with greater spatial and temporal resolution provide opportunities to develop more precise C factors that vary over short time spans. Including seasonality in the NDVI to C factor relationship can capture peak agricultural soil loss and implicitly represent the impact of management practices (P factor), such as planting cover crops. Remotely-sensed tillage and crop residue indices such as the Normalized Difference Tillage Index would further improve representation of crop cover and management. These types of indices can quantify crop residues and soil tillage intensity across varying moisture conditions (Beeson et al., 2016; Quemada and Daughtry, 2016; Sonmez and Slater, 2016; Quemada et al., 2018).

Applications:

the estimates of soil loss, sediment yield, and sediment retention by natural vegetation in this study can be used to address environmental change, ecosystem services, resilience, and natural disasters. For example: (1) modeling potential soil loss and sediment yield post-wildfire; (2) accounting for annual soil carbon loss due to erosion and avoided carbon loss by natural vegetation by linking estimates in this study with a carbon model; (3) downstream impacts of soil erosion on economic costs of drinking water treatment or reservoir dredging; (4) coupling the methods with land change models to understand the impacts of future land use change on soil loss and sediment delivery; and (5) quantifying the value of natural land restoration in locations with high soil erosion and sediment delivery ratios.

4.4. Conclusions

RUSLE and sediment delivery ratio based on hydrologic connectivity were scaled to 30-meter spatial resolution, providing the first estimate of sediment retention ecosystem services across the United States. The value of natural land cover as a sediment retention ecosystem service is unevenly distributed across the United States and driven by underlying topographic and climatic factors. Forested lands hold the most value in preventing soil delivery, especially those in areas with high slopes, erodible soils, intense rains, and proximity to waterbodies. This has implications for the decision-making process in guiding where natural land cover should be preserved, or agricultural land be taken out of production to minimize sediment delivery to waterbodies.

The estimates of soil loss represent sheet and rill erosion, while sediment delivery models the percentage of that eroded soil delivered to a waterbody. Other components of the sediment budget, such as gully erosion, streambed and bank erosion, and deposition are not considered here. Natural land cover, or its loss, also has value beyond the hillslope; it ultimately exerts some level of influence on erosive flows, transport, and deposition. Connecting the landscape value of natural land cover presented here to the remainder of the sediment budget is the next step in developing a total value of natural land cover with respect to total sediment flux.