Abstract
Sustainable agriculture begins with reliable conservation management planning. Conservation management planning addresses soil loss or erosion control while coupling productivity and profitability on the farm together with environmental stewardship. Traditional row crop agriculture utilizes soil loss prediction technology to estimate the impact of different management strategies to minimize soil loss and maximize soil conservation. In this present study, two current USDA water erosion prediction technologies used in the United States to prescribe conservation management plans (RUSLE2 and WEPP) were examined. The technologies were compared first as designed and intended for management plan implementation (17,525 simulations) and second for equivalent climate-specific conditions (18,480 simulations), using similar input parameters for management, soils, terrain (slope length and gradient), and crop yields. Results of the simulations generally show similar soil loss trends for managements, terrain characteristics, and crop yields. Simulated soil loss estimates disagree particularly for clay-textured soils and no-till management. Both studies show these trends independent of climate inputs evaluated. Though the comparison results provide important insight into model tendencies, there are still open questions remaining regarding climate.
Introduction
Estimates of cropland soil loss worldwide indicate unsustainable levels of fertile topsoil loss in agricultural lands (Sartori et al. 2024). Current levels of annual global soil loss rates are expected to increase due to changes in precipitation patterns and sustained adoption of highly intensive growing practices in response to increasing food demand (García-Ruiz et al. 2015; Eekhout and de Vente 2022).
When describing erosion driven by water, there are many terms that can be used, such as erosion, soil loss, soil erosion, or water erosion. Among the multitude of available strategies to mitigate soil loss, water erosion prediction tools have long been recognized as technology critically necessary to support the development of conservation plans (Borrelli et al. 2021). In the United States, the Federal Water Pollution Control Act Amendments of 1972 (US Congress 1972) established that resources should be allocated to the assessment of the impact of cropland nonpoint source pollution to water quality. This government initiative led to the formation of interagency partnerships to develop science and technology to characterize existing conditions and evaluate alternative practices specifically designed to reduce long-term soil loss. Since then, significant advances have been made with an estimated reduction between 1992 and 2012 of 45% in cropland erosion magnitude (Hellerstein et al. 2019). These gains reflect the adoption of farming management practices by producers because of the work by the USDA Natural Resources Conservation Service (USDA NRCS) in developing conservation strategies throughout the United States aided by erosion prediction technology.
USDA NRCS works with agricultural producers to create conservation plans that address natural resource management objectives and assess and analyze issues related to soil, water, animals, plants, air, energy, and human interaction. Combining a producer’s farming and/or ranching skills with the science-based knowledge, skills, and tools of a USDA NRCS conservation planner helps to sustain the natural resources on the producer’s land. Having a conservation plan also helps producers qualify for USDA financial assistance programs that offset some of their investments. USDA NRCS uses a nine-step conservation planning process to (1) identify problems and opportunities, (2) determine objectives, (3) inventory resources, (4) analyze resource data, (5) formulate alternatives, (6) evaluate alternatives, (7) make decisions, (8) implement the plan, and (9) evaluate the plan (USDA NRCS 2023). Typically, USDA NRCS will evaluate different cropping management systems, tillage systems, and conservation practices on long-term average annual soil loss and compare the resulting soil loss values to a set tolerable soil loss value (T-value). This comparison can provide farmers and landowners with several management options to control soil loss, as well as maintain productivity and profitability from their farms. Every farm and its associated lands is unique and requires tailored management, and every decision-maker has different management concerns and needs. Between 2014 and 2016, US$1.9 billion was invested in the implementation of conservation practices (USDA NRCS 2023). With nearly 3,000 offices in communities nationwide, USDA NRCS provides the information, tools, and delivery systems necessary for producers to conserve, maintain, and improve their natural resources.
The USDA Agricultural Research Service (USDA ARS) is responsible for developing the science supporting USDA soil erosion technologies, such as the Revised Universal Soil Loss Equation, Version 2 (RUSLE2) (Foster 2008; USDA ARS 2013), maintained and supported by the National Sedimentation Laboratory (NSL), and the Water Erosion Prediction Project (WEPP) (Flanagan and Nearing 1995; Flanagan et al. 2001), maintained and supported by the National Soil Erosion Research Laboratory (NSERL). Currently, the USDA NRCS utilizes RUSLE2 technology within district conservationist (DC) offices for conservation management planning at the farm level. USDA NRCS has been working with the NSERL to develop WEPP interfaces and databases for use in their county offices. Despite being developed to address a national need in the United States, these two erosion prediction technologies have been extensively adopted and adapted to local conditions worldwide to support research studies and conservation projects in a wide range of conditions (Borrelli et al. 2021). Additionally, the RUSLE2 and WEPP model development teams have, in general, worked independently to develop and test science code for sustainable agriculture.
RUSLE2 is a hybrid soil erosion prediction model, based on the Universal Soil Loss Equation (USLE) (Wischmeier and Smith 1978), the Revised Universal Soil Loss Equation, Version 1 (RUSLE) (Renard et al. 1997), and 10,000 plot years of basic runoff and soil loss data (Renard et al. 1997). RUSLE2 computes sheet and rill erosion on a hillslope based on empirical equations driven by rainfall erosivity but uses process-based equations driven by runoff estimates to determine sediment transport capacity, deposition, and sediment enrichment in clay and organic matter (Foster 2008). RUSLE2 assumes that sheet and rill erosion are linearly related to rainfall erosivity, input as erosivity density that is defined as the amount of rainfall erosivity per unit of precipitation depth. The use of erosivity density lessens the impact of missing data on monthly means and shortens the period of record needed. A smoothly varying erosivity density surface for the entire nation was used to determine erosivity for each county or precipitation zone (USDA ARS 2013), ensuring geographically consistent erosion predictions that are needed for conservation planning. Management descriptions in RUSLE2 comprise combinations of field operations and vegetations. Vegetation descriptions specify temporal growth patterns; the canopy cover, height, and shape; harvest biomass relationships; hydraulic roughness properties; and associated residue characteristics for a single species or a mixture of species. RUSLE2 includes routines to estimate the amount of residue added during the growth of perennial vegetation and to accommodate the modeling of haying and grazing scenarios (Dabney and Yoder 2012; Dabney et al. 2014). Additional components were developed within RUSLE2 to account for ephemeral gully processes (Dabney et al. 2015) to estimate water erosion driven by runoff in addition to that associated with rainfall erosivity.
The WEPP model has been in development by USDA since 1985 and is based on simulations of fundamental physical processes involved in soil erosion by water (Flanagan and Nearing 1995; Flanagan et al. 2001, 2007). WEPP includes a climate generator and components that model surface hydrology, subsurface hydrology, plant growth, residue decomposition and management, tillage disturbance and soil consolidation, winter processes, overland flow hydraulics, soil detachment by raindrops and shallow flow, soil detachment by concentrated water flows, sediment transport, and sediment deposition (Flanagan and Nearing 1995). WEPP has been tested with experimental plot (Wang et al. 2023) and small watershed data, and it is used currently by the USDA Forest Service and many others for erosion and sediment yield predictions (Flanagan et al. 2007). Additional capabilities are included within WEPP to simulate watershed-scale processes, such as ephemeral gullies, channel erosion, and the role of impoundments in controlling sediment yield.
Comparative studies between these two technologies have focused on specific scenarios and conditions, including (1) small-scale soil armoring investigations (Cochrane et al. 2019); (2) literature assemblage, evaluation, and comparison for model selection and perception (Alewell et al. 2019; Borrelli et al. 2021); (3) event erosion on bare fallow plots with different calibration approaches (Kinnell 2017); and (4) specialized site studies (Stolpe 2005; Wade et al. 2012). There is a need for studies depicting the utilization of these two technologies as approved and adopted versions by the USDA NRCS for application in all 3,143 counties (municipalities) using standard and available input databases for long-term erosion estimation without calibration to support planning of conservation alternatives. This knowledge gap has been recognized by USDA NRCS, and therefore, it required comparison testing and evaluation to provide confidence to interested stakeholders concerned with conservation management.
To support the USDA NRCS mission, the study performed a quantitative comparison of RUSLE2 and WEPP soil loss prediction technologies to evaluate their applicability as conservation management planning tools and to provide confidence to interested stakeholders concerned with natural resources conservation management (Moriasi et al. 2007). Herein, results are presented of an extensive, collaborative testing program in which the effects of topography, cover management, crop yield, soil type, and climate on average annual soil loss computed by RUSLE2 and WEPP are examined. The model comparison studies were conducted at the request of NRCS to evaluate how the two USDA ARS erosion prediction models perform under similar input conditions. The testing program comprised two studies: the Iowa Study and the Vermilion Study. The Iowa Study represents the typical application of a conservation management planning scenario by an NRCS county office, and the Vermilion Study explores the impact on soil loss estimations when both technologies are provided with identical climate inputs.
Materials and Methods
Experimental Design. Two simulation studies are reported herein: Iowa (conducted in 2018 to 2019; figures presented with black axis coloring) and Vermilion (conducted in 2019 to 2020; figures presented with red axis coloring). In both the Iowa Study and the Vermilion Study, soil loss estimates were calculated using RUSLE2 version 2.6.11.1 (April 4, 2018) and WEPP version 2018.6. The corresponding model databases were NRCS RUSLE2 database (dated April 7, 2016) and NRCS WEPP CRLMOD4 desktop version with 2015 climate inputs (dated April 12, 2019). The simulations were executed in automated batches to facilitate the large number of simulations required to evaluate model sensitivity to input parameters critical to soil loss.
RUSLE2 and WEPP are fundamentally different models with unique input requirements. The soil loss algorithms are different, the information that is input as climate is different (figure 1 and supplemental tables S1 and S2), and some of the management and operations parameters are different. However, despite their differences, the two technologies should tend toward a similar solution when identical or near identical inputs are provided. Large differences in model predictions would indicate that at least one and possibly both technologies and/or their inputs are erroneous. Small differences do not indicate that either or both technologies and their inputs are correct.
Climate input comparison for the five Iowa counties represented in the Iowa Study (grayscale) and one Illinois county in the Vermilion Study (red): (a) average annual precipitation and (b) average annual erosivity. DSI-3260 refers to the nearest NOAA 15-minute precipitation gauges to the six county centers (1970 to 2013). RUSLE2 refers to the six RUSLE2 county climate inputs (Foster 2008). WEPP-CLIGEN refers to the nearest PRISM-adjusted CLIGEN inputs for the six county centers (Srivastava et al. 2019).
Iowa Study. The Iowa Study was designed to evaluate RUSLE2 and WEPP conservation planning technologies as intended for use by USDA NRCS county office personnel. Simulations consisted of multiple hillslope profiles, soils, and management configurations that were designed to examine soil loss estimates in five Iowa counties: Adair, Des Moines, Hardin, Jackson, and Plymouth.
Three groups of simulations (17,525 total simulations) were conducted, herein referred to as Group 1 (G1), Group 2 (G2), and Group 3 (G3). G1 simulations (13,230 simulations for each model) consisted of all possible combinations of 7 soils (full textural range; figure S1; table 1), 6 slope gradients, 7 slope lengths, and 9 management systems (continuous corn [Zea mays L.], continuous soybeans [Glycine max {L.} Merr.] and corn–soybean rotations; table 1). G2 simulations (4,200 simulations for each model) consisted of 8 crop yield combinations for corn and soybeans using 7 soils (same as G1; figure S1; table 1) and 15 management systems (table 1), while holding slope length (150 ft [45.7 m]) and slope gradient (6%) constant. G3 simulations (95 simulations for each model) consisted of 19 soils (table 1) using tilled-fallow operations, while holding slope length (72.6 ft [22.1 m]) and slope gradient (9%) constant.
Iowa Study model input.
In the Iowa Study, RUSLE2 climate data were provided by the existing NRCS RUSLE2 database, and WEPP climate data were generated by WEPP CLIGEN (Nicks et al. 1995; Srivastava et al. 2019); both climate data sets were prepared following standard conventions as intended for NRCS operations (figure 2). See also supplementary material table S1 for specifics related to the Iowa Study climate, the section “Climate Considerations” within the Methods and Materials section for content relevant to both Iowa and Vermillion studies, and the section “Climate and Hydrology” within the Results and Discussion section for content relevant to the results.
Comparison of soil loss simulated by RUSLE2 and WEPP: (a) Iowa-G1 Study (13,230 runs), (b) Iowa-G2 Study (4,200 runs), (c) Iowa-G3 Study (95 runs), and (d) Vermilion Study (18,480 runs). Soil loss estimates are grouped by soil texture. The solid line is the one-to-one relation.
Vermilion Study. The goal of the Vermilion Study was to test whether the average annual soil loss values predicted by RUSLE2 and WEPP are comparable when climate inputs are based on similar precipitation characteristics (i.e., rainfall amount, intensity, and erosivity) derived from the identical climate input. See also supplementary material table S2 for specifics related to the Vermilion Study climate.
In the Vermilion Study, 42 years of observed 15-minute precipitation data from the National Weather Service station in Hoopeston, Vermilion County, Illinois (No. 11419800), were used to prepare climate inputs to both RUSLE2 and WEPP. This station was selected for the study because it contained the lowest percentage of missing (i.e., gaps within the record) and accumulation (i.e., annotated beginning and end of event, expressed as total precipitation) data from the US Midwest region. The WEPP Climate File Formatter (WEPPCLIFF) (McGehee et al. 2020) was used to prepare both WEPP and RUSLE2 precipitation inputs. No corrections were applied to the raw 15-minute precipitation record to derive the RUSLE2 erosivity index (EI) and erosivity density inputs. This methodology differed from the standard RUSLE2 R-factor determination as the input here included (1) small storms (i.e., storms with precipitation less than 0.5 in [13 mm]) to calculate EI and (2) storm events of return periods exceeding 50 years. The deviation from the standard RUSLE2 procedure was necessary because WEPP does not remove such storms from its climate analysis. WEPP CLIGEN was used to derive the climate data for WEPP (Nicks et al. 1995; Srivastava et al. 2019).
To assess model agreement, 18,480 simulations combining 21 soils (same soils as the Iowa Study with the addition of Portageville clay and Aholt clay; figure S1), 4 slope gradients, 4 slope lengths, 7 management systems (created with assistance of NRCS staff), and 9 crop yield level groups (101.25, 135, and 168.75 bu ac−1 [6.36, 8.48, and 10.6 Mg ha−1] corn yield levels; 31.5, 42, and 52.5 bu ac−1 [2.12, 2.82, and 3.53 Mg ha−1] soybean yield levels) were examined (table 2). The baseline crop yields for corn (135 bu ac−1 [8.48 Mg ha−1]) and soybeans (42 bu ac−1 [2.82 Mg ha−1]) were based on the long-term (1971 to 2013) crop yield data from the USDA National Agricultural Statistics Service (NASS) website (https://www.nass.usda.gov/). The selected cover crop was cereal rye (Secale cereale L.).
Vermilion Study model input.
Results from these model simulations are close to the soil loss that would be calculated by NRCS office personnel, but they will deviate slightly since the climate files were different from the current NRCS database climate files. A critical point for readers to understand related to this is that there was no attempt to benchmark erosivities in the Vermilion Study, which means that while identical, the inputs used in the Vermilion Study were somewhat less erosive than should be expected in reality. This issue was minimized by selecting a precipitation gauge with less than 2% combined missing gaps and accumulated data according to the NOAA metadata definition and the processing procedure used in McGehee and Srivastava (2018). However, this still does not account for precipitation undermeasurement when the gauge is operating properly and the interactive effects of intensity dampening and gauge precision (McGehee et al. 2021, 2022).
Climate Considerations. While it is possible to prepare and drive both models with the exact same climate inputs (e.g., nearly identical precipitation amount, erosivity of that precipitation, and other precipitation characteristics; table S2), neither model was designed to correctly predict erosion based on any arbitrary precipitation interval in its respective climate inputs. Breakpoint precipitation data are ideal for driving both technologies because they preserve precipitation characteristics to within the accuracy and precision limitations of a given gauge (McGehee et al. 2021), and they were the data originally used to develop the USLE (Wischmeier 1959; Wischmeier and Smith 1958, 1965, 1978). WEPP’s infiltration and runoff predictions are computed at the precipitation interval, so the use of breakpoint and/or fixed-interval inputs can have substantial impact on runoff and erosion predictions (Flanagan and Nearing 1995; Flanagan et al. 2020). The only way that both models can be forced with truly identical climate inputs and with inputs that satisfy both RUSLE2 and WEPP design requirements is to use the same breakpoint precipitation data to create the climate inputs for a given comparison study. Unfortunately, this kind of data is relatively rare, especially for periods long enough to compute reliable erosivity estimates for USLE-type models.
In the Vermilion Study, the agreed best option for providing similar climate inputs was to use the same 15-minute precipitation data to derive climate inputs for each model, generally following each of the respective models’ climate preparation procedures with two exceptions: the RUSLE2 input was prepared without omitting small storms (0.5 in [<13 mm]) and without omitting storm events of greater than a 50-year return period. The Hollinger et al. (2002) study recommendation of 4% intensity dampening correction for maximum 30-minute intensity was not applied, nor were any gap-filling or other corrections applied. Ultimately, the RUSLE2 and WEPP climate inputs used in the Vermilion Study were much more similar in terms of average annual precipitation, erosivity, and erosivity density with relative differences of −0.44%, −0.30%, and 0.00%, respectively (computed as [{WEPP – RUSLE2} / RUSLE2] × 100%) (table S2).
The US NOAA National Climatic Data Center (NCDC) inputs from Hoopeston, Illinois, were not corrected for several issues known to impact 15-minute, fixed-interval precipitation gauges. Therefore, uncorrected NCDC inputs should be expected to be less intense than breakpoint precipitation data from which both RUSLE2 and WEPP were developed. While both models’ simulations and general behavior are impacted by these input limitations, WEPP may be more sensitive to dampening effects of suboptimal climate data than RUSLE2 (Flanagan et al. 2020).
Simulation Metrics. In presenting the results and calculating statistical measures, we used the RUSLE2 estimated soil loss as the reference value; that is, it is plotted along the abscissa and represents the observed value in the calculation of statistical measures. We strongly emphasize that this is simply to facilitate the presentation of the results.
Several methods exist to compare soil loss estimates. Herein, the soil loss estimate difference (ED) (equation 1),
1
was used, which maintains the units of soil loss (t ac−1 yr−1 [Mg ha−1 y−1]) and does not distort or magnify small values (i.e., typical with no-till differences). Positive ED means that WEPP soil loss is higher than RUSLE2 soil loss, and negative ED means that RUSLE2 soil loss is higher than WEPP soil loss. Given the large number of simulations, similar factors (i.e., slope length, slope gradient, soil, management, crop yield, and climate) were grouped, and group means are presented within figures and tables.
RUSLE2 and WEPP at the factor level (both grouped and nongrouped) agreement was assessed using the following statistical measures: adjusted coefficient of determination (Adj. R2), Nash-Sutcliffe model efficiency (NSE) (Nash and Sutcliffe 1970), and percent bias (PBIAS).
The adjusted coefficient of determination is calculated as equation 2:
2
where N is the number of simulations and k is the number of independent variables. The Adj. R2 is an unbiased estimator of the goodness of a linear fit (f) between the RUSLE2 and WEPP model results. The coefficient of determination is calculated as equation 3:
3
where the overbar indicates the mean soil loss. Adj. R2 is negative when the residual sum of squares (numerator of the above fraction) exceeds the total sum of squares (denominator), which indicates that the linear model has a worse prediction than one that always predicts the mean. Adj. R2 values greater than 0.5 typically indicate an acceptable linear relationship between the compared data sets. Note: Adj. R2 indicates if RUSLE2 and WEPP soil losses are correlated; however, it does not indicate if the soil losses agree.
Nash-Sutcliffe model efficiency is calculated as equation 4:
4
and ranges from −∞ to 1. An NSE value of 1 indicates that the soil loss values simulated by the models for a factor compare perfectly, while anything lower indicates that model responses become less similar. Typically, NSE is widely used in model validation studies to compare model predictions to observed measurements. In such studies, an NSE value exceeding 0 is assumed minimally acceptable (Gupta et al. 1999), which indicates the model performance is equivalent to the mean of observations. Again, in model validation studies, NSE values should exceed 0.5 to indicate satisfactory agreement (Moriasi et al. 2007). This study is not a model validation study but a model comparison study. There are no “observed” data to which either model can be compared, so this application of NSE arbitrarily selects RUSLE2 as the “observed” data set for the calculation. NSE, as defined here, is scaled by N times the RUSLE2 soil loss variance (i.e., 
). Given the same ED, standard deviations can be either (a) RUSLE2 ≈ WEPP, (b) RUSLE2 > WEPPm, or (c) RUSLE2 < WEPP. When (a) is true, there will be no substantial difference in the reported NSE value if either model were used as “observed.” When either (b) or (c) is true, the NSE value reported in this study will be (b) greater or (c) lesser than if WEPP had been used for the “observed” value set in the equation.
The percentage bias is calculated as equation 5:
5
and represents the average tendency of one model’s results with respect to the other model’s results. For nonnegative comparison values (i.e., average annual soil loss) and this definition, PBIAS ranges from −∞ to +100%. The optimal value of PBIAS is 0.0%, with low-magnitude values (|PBIAS| < 25%) indicating acceptable model comparison (Gupta et al. 1999). Positive values indicate RUSLE2 predictions were greater than WEPP predictions, and negative values indicate WEPP predictions were greater than RUSLE2 predictions.
Given the substantial variability in soil loss estimates over the full range of evaluated factors, model agreement was deemed unsatisfactory for values of the statistical measures that were lower than those typically found in the literature (Moriasi et al. 2015). The authors deemed model agreement to be unsatisfactory when any of the following conditions were true: Adj. R2 < 0.25, NSE < 0.25, PBIAS < −100%, or PBIAS > 50%. These measures reflect model agreement/ disagreement for correlation (Adj. R2), goodness of fit (NSE), and bias (PBIAS). These values help characterize the nature of model agreement/disagreement, and no one value should be considered in isolation of the others when interpreting this comparison study.
Finally, the rate of soil loss change with respect to slope length (tn ac−1 yr−1 ft−1) and the rate of soil loss change with respect to slope steepness (tn ac−1 yr−1 %−1) were compared. These values were calculated individually for climate, soil, and management factors. For instance, consider soil loss values from the G1 simulations in the Iowa Study. For a given climate, soil, and management, 42 simulations from each model (six gradients and seven lengths) would be used for the calculation. For each gradient, a mean soil loss for all seven slope lengths was calculated. This generated one mean soil loss for each of the six gradients. The slope of a regression line for the six mean soil losses (across the seven lengths) was obtained. The same would be obtained for slope length (across the six gradients). This was repeated for both models and all combinations of climate, soil, and management factor levels. These regression slopes indicate local model sensitivity to topographic characteristics.
Results and Discussion
To assist the reader in distinguishing one study from another in the same figure, the graphs for the Iowa Study are shown in black and the graphs for the Vermilion Study are shown in red (unless specifically noted otherwise within a legend or caption).
Scatter plots comparing soil loss estimates by RUSLE2 and WEPP of all simulations from the Iowa (figures 2a to 2c) and Vermilion (figure 2d) studies illustrate agreement and departure from the one-to-one line. Table 3 lists the statistical measures for the four sets of simulations and the overall study. In general, the overall soil loss estimates by RUSLE2 and WEPP compared fairly well for the Iowa-G3 Study simulations and well for the Vermilion Study simulations, whereas the overall soil loss estimates by RUSLE2 and WEPP for the Iowa-G1 and Iowa-G2 Study simulations did not compare well.
Statistical measures of the comparison of RUSLE2 and WEPP soil loss estimates for all simulations of the Iowa (G1, G2, and G3) and Vermilion (Illinois) studies.
Figure 3 presents histograms of ED from the Iowa-G1, Iowa-G2, Iowa-G3, and Vermilion studies. Figures 1 and 3 show a wide range of results that do not offer insight to model differences; however, for example, figure 3 does indicate that 56% of the results from the Iowa-G1 Study were within ±3 tn ac−1 yr−1 (7.41 Mg ha−1 y−1), and 83% of the results from the Vermilion Study were within ±3 tn ac−1 yr−1. Generally, RUSLE2 and WEPP soil loss estimates agreed for loamy soils, slope lengths <100 ft (30 m), slope gradients <6% to 9%, and management practices with higher tillage intensities (e.g., tilled-fallow and conventional tillage). Differences between RUSLE2 and WEPP soil loss estimates were greatest for soils containing clay and sand, for large slope lengths and gradients, and for almost all management practices in the Iowa studies and cover crops in the Vermilion Study (tables S3 through S6). The following sections present in more detail the influence of each factor on soil loss estimates in order of climate, soils, topography, and management (including crop yield adjustments).
Histogram of soil loss estimate difference (ED; WEPP minus RUSLE2) from the (a) Iowa-G1 Study (13,230 runs), (b) Iowa-G2 Study (4,200 runs), (c) Iowa-G3 Study (95 runs), and (d) Vermilion Study (18,480 runs). Fifty percent of the simulations fall within the dark gray shaded area, while 75% of the simulations fall within the light gray shaded area. X-axis limits were arbitrarily set to plus and minus 20 tn ac−1 yr−1, actual ED values are smaller or larger than the x-axis limits.
Climate and Hydrology. Within the RUSLE2 technology, soil loss is motivated by a combination of the R-factor (rainfall-runoff erosivity factor; units are MJ-mm ha−1 h−1 y−1) and the EI10 (single storm erosivity with 10 y return frequency; units are MJ-mm ha−1 h−1) (table 4). Therefore, the expected soil loss predictions from RUSLE2 are generally assumed to be (1) smaller in Plymouth County, Iowa, because of its smaller R-factor than the other four counties in the Iowa Study; and (2) larger in Vermilion County, Illinois, because of its larger R-factor compared to the counties in Iowa. RUSLE2 was designed to perform in this manner, and that was observed in this study. In table 4, average soil loss is from 19 soils, tilled-fallow management, and unit plot dimensions (72.6 ft [22.1 m] length, 9% slope), that is, for the unit plot conditions of the Iowa-G3 Study. The Vermilion average soil loss was for a 75 ft (22.9 m) simulation slope length and 21 soils, while all other characteristics are the same as for the Iowa-G3 Study.
RUSLE2 climate variables and average annual soil loss for unit plot conditions (~75 ft slope length, 9% slope gradient).
Within the WEPP technology, the computed average annual soil loss is the summation of all the individual storm soil loss predictions through time, divided by the years of simulation. For an individual storm event, the WEPP-predicted soil loss is a function of the storm rainfall intensity, peak runoff rate, durations of rainfall and runoff, and the daily adjusted erodibility parameter values. WEPP simulates these dynamics at the climate input interval, dependent upon the active storm characteristics, and is an important feature of its design. Therefore, no single climate or hydrologic factor can always be attributed to greater soil loss predictions in WEPP. WEPP was designed to perform in this manner, and that was observed in this study. Average annual soil loss predictions for the Iowa-G3 Study simulations did generally increase with runoff magnitude, but this generalization includes detachment-limited cases, cases where greater runoff does not always result in greater soil loss, and cases resulting from reduced soil loss due to ponding, hydraulic conductivity of the soil, and the rate and timing of rainfall (figure 4). Figure 4 further shows the effects of soil texture on runoff, where very fine-grained soils such as clay with limited infiltration have increased runoff, while sands with high hydraulic conductivity have reduced runoff. Soil loss is higher for silt loam soils and decreases for loam and clay loam soils relative to the general soil loss trend between annual soil loss and annual runoff.
General trend between runoff and soil loss for WEPP from the Iowa-G3 Study: (a) all soils for each county and (b) average of the texture groups for each county. The broken line is the linear regression.
In terms of differences between RUSLE2 and WEPP, to evaluate the effects of climate on soil loss, figure 5 compares the RUSLE2 and WEPP soil loss estimates by county for unit plot conditions (Iowa-G3 Study and Vermilion Study limited to tilled-fallow management, 9% slope gradient, and 75 ft [22.9 m] slope length). The soil loss estimates shown were averaged over all soils: 19 soils for the Iowa-G3 Study and 21 soils for the Vermilion Study. The magnitude of RUSLE2 and WEPP simulated soil losses followed a different order by county, that is climate. RUSLE2 soil loss followed the climate intensity (R-factor) ranking (low to high): Plymouth, Iowa; Jackson, Iowa; Hardin, Iowa; Des Moines, Iowa; Adair, Iowa; and Vermilion, Illinois (table 4). For WEPP, climate intensity can potentially be ranked by runoff or soil loss. If runoff is used, the climate intensity ranking is (low to high) Plymouth, Iowa; Adair, Iowa; Jackson, Iowa; Des Moines, Iowa; and Hardin, Iowa (Vermilion runoff not provided). If soil loss is used, the climate intensity ranking is (low to high) Plymouth, Iowa; Jackson, Iowa; Vermilion, Illinois; Adair, Iowa; Des Moines, Iowa; and Hardin, Iowa. WEPP-estimated soil loss generally followed the trend provided by CLIGEN for these counties. The soil loss pattern by county shown in figure 5 was also observed at the individual soil level (figure S3). The statistical measures for the data plotted in figure 5 indicate that there is no agreement between RUSLE2 and WEPP soil loss estimates at the climate level (six observations): Adj. R2 = 0.61, NSE = −1.88, and PBIAS = −22.2. However, as shown in table 3 and figure 3, the Vermilion Study simulations by themselves showed agreement between RUSLE2 and WEPP soil loss estimates when evaluated across all factors.
Comparison of RUSLE2 and WEPP soil loss estimates (averaged over all soils) by county (that is, different climates) for unit plot conditions (Iowa-G3 Study and Vermilion Study limited to tilled-fallow, 75 ft slope length, 9% slope gradient). The line connects soil loss by county, ordered by RUSLE2 soil loss (low to high). The solid line is the one-to-one relation.
Soils. The effects of soils on soil loss estimates from RUSLE2 and WEPP are compared for unit plot conditions (Iowa-G3 and Vermilion studies). Unit plot conditions are unique to USLE-based technology and relate to soil loss resulting from tilled-fallow management with 72.6 ft (22.1 m) slope length at 9% slope gradient.
Overall, figure 6 shows a comparison of average annual soil loss simulated by each technology using 19 soils grouped into seven textures from the Iowa-G3 Study and 21 soils grouped into seven textures from the Vermilion Study (note that unit plot conditions in the Vermilion Study were 75 ft [22.9 m] slope length at 9% slope gradient under tilled-fallow management). Table 5 lists the corresponding statistical measures by soil texture.
Unit plot soil loss comparison in the Iowa-G3 and Vermilion studies. Soil loss estimates are colored by soil texture. The solid line is the one-to-one relation.
Statistical measures by soil texture of the comparison of RUSLE2 and WEPP soil loss estimates for unit plot conditions for Iowa-G3 and Vermilion studies. Note: The simulated Vermilion soil losses were limited to 75 ft slope length, 9% slope gradient, and tilled-fallow management. Simulation groups found in agreement according to the statistical interpretation outlined earlier (Adj. R2 > 0.25, NSE > 0.25, and 100% < PBIAS < 50%) are bolded.
Although direct comparisons of RUSLE2 and WEPP erodibility parameters were hindered by the different erodibility approaches incorporated within each model, trends of soil erodibility and average annual soil loss were evaluated for both models separately to explore possible reasons for differences using the standard unit plot simulations. Figure S3 shows trends from the Vermilion Study of soil erodibility along with predicted average annual soil loss for RUSLE2 and WEPP. Figure S3A shows that RUSLE2-predicted average annual soil loss exhibited a linear trend with its soil erodibility factor. This result is not surprising since soil erodibility (i.e., K-factor) was defined for RUSLE2 using unit plot conditions. The silt loam soil contains the highest erodibility and results in the highest RUSLE2 average annual soil loss for that soil. The sandy soils had the lowest soil erodibility factor and resulted in lower RUSLE2 average annual soil loss. The soil erodibility factors for loam, clay loam, Lamoni clay, and silty clay soils were similar, hence RUSLE2 showed similar predicted average annual soil loss for these soils.
In contrast to the RUSLE2-predicted average annual soil loss results using a single soil erodibility factor, WEPP primarily represents soil erodibility using three time-varying parameters: interrill erodibility, rill erodibility, and critical shear stress. Figures S3B-D show WEPP-predicted average annual soil loss with each baseline parameter separately, although the resulting soil loss was based on the interaction of all three parameters.
Figure 7 shows box plots of ED in average annual soil loss predictions by texture and by soil for the Iowa G3 and Vermilion studies. Though magnitudes of the ED distributions between the Iowa-G3 and Vermilion studies are different (the interquartile ranges of ED for the Vermilion Study are approximately centered about or closer to zero), the patterns between soil textures and between soils within a texture were consistent and therefore independent of climate. RUSLE2 predicted greater average annual soil loss for loamy soils than WEPP, and WEPP predicted greater average annual soil loss for clayey and sandy soils than RUSLE2. Both models predicted the greatest average annual soil loss for silt loams and the least average annual soil loss for sands. Besides the general differences in ED according to texture (figures 7a and 7c), it should be noted that there were differences in average annual soil loss estimates from the two models, even for soils within the same texture. Specifically, the Lamoni clay, Webster clay loam, Clarion loam, and Ida silt loam behave differently than the other two soils within their respective textural classes.
Average annual soil loss estimate difference (ED) for the Iowa-G3 and Vermilion studies: (a) ED by soil texture and (b) ED by soil for the Iowa-G3 Study; and (c) ED by soil texture and (d) ED by soil for the Vermilion Study. The × symbols in the figure indicate the mean values ED. Note: For presentation purposes the plotted ED range is limited to (−20, 40) tn ac−1 yr−1. ED for the clays in the Vermilion Study was as large as 150 tn ac−1 yr−1, while ED for the Ida silt loam was as small as −30 tn ac−1 yr−1 in the Vermilion Study.
Topography and Management. Both models responded as expected to increasing slope lengths and slope gradients with increasing soil loss predictions (table S3 and S6). Irrespective of climate, the average annual soil loss results were generally lower for WEPP than RUSLE2 at smaller slope lengths and gradients, and higher for WEPP than RUSLE2 at larger slope lengths and gradients (figure 8). However, for some very low slopes and high residue conditions, there were cases where WEPP predicted slightly lower average annual soil loss and soil loss rate at increasing slope lengths. The hydrologic processes in WEPP are sensitive to profile length, and longer slope lengths can slightly decrease predicted runoff.
Estimate difference (ED) as a function of (a) slope length and (b) slope gradient for all simulations of the Iowa-G1 Study simulations, and ED as a function of (c) slope length and (d) slope gradient using the Vermillion Study simulations.
Figure 8 lumps multiple soils, cropping systems, and crop yields. Another method of analyzing these data considers the rate of soil loss with respect to slope length or slope gradient. These rates are independent to and diagnostic of each model and can be viewed from multiple climate, soil, and management combinations. For the Iowa-G1 Study, figure 9a compares the simulated soil loss rates with respect to slope length for all management practices by soil, and figure 9b represents all managements with averaged soil texture. For the Vermilion Study, figure 9c compares the simulated soil loss rates with respect to slope length for all management practices by soil, while figure 9d represents all managements with averaged soil texture. Figure 9 shows WEPP soil loss rates to be larger than those from RUSLE2, which corroborates the increase in soil loss ED for increasing slope length and slope gradient shown in figure 8.
Soil loss rate with respect to slope length for RUSLE2 and WEPP simulations from the Iowa-G1 Study using (a) all managements, all soils, averaged topography, and climate; (b) all managements, averaged soils, averaged topography, and climate; and from the Vermilion Study using (c) all managements, all soils, and averaged topography; and (d) all managements, averaged soils, and averaged topography. The solid line is the one-to-one relation. Warmer colors indicate higher tillage intensity. Management acronyms are defined in table 1.
Several model differences and similarities can be observed when assessing soil loss rates. First, the data set from Vermilion was used to examine the trends of each soil, individually, using four managements (no-till, conservation tillage, conventional tillage, and tilled-fallow). An example is the soil loss rate with respect to slope length for the Kenyon loam (see supplementary material figure S5 for all Vermilion Study soils). Both RUSLE2 and WEPP simulate increasing soil loss as the tillage intensification increases. Within the 21 soils examined in the Vermilion Study, several stand out (i.e., Ida silt loam, Webster clay loam, all three silty clay soils, Perks loamy sand, and all clay soils), but the common thread is clay texture. When all soils are shown for a particular management, very similar patterns are observed. For example, all soils with clay texture were plotted in a different color, while the other soils share the same gray color (figure 10).
Comparison of soil loss rate with respect to slope length for all simulations in the Vermilion Study and classified by four management practices: (a) no-till, (b) conservation tillage, (c) conventional tillage, and (d) tilled-fallow. The clay-textured soils are highlighted in color, while all other textured soils are shown in gray. The solid line is the one-to-one relation. Note the soil loss rate ranges within the graphs.
The no-till soil loss rates with respect to slope length appear to compare well (figure 10a) with 12 values below and 9 values above the line of perfect agreement. However, the managements with increasing tillage activities demonstrate a different trend (figures 10b through 10d). In all three tillage managements, WEPP soil loss rates for clay-textured soils increase more rapidly than those for RUSLE2. As the slope length becomes longer, WEPP produces a higher rate of soil loss per unit increase in slope length than RUSLE2.
Similarly, for the analysis with slope gradient, multiple patterns emerge (figure 11). The no-till soil loss rates with respect to slope gradient seem to be clustered around 0.06 tn ac−1 yr−1 %−1 (0.15 Mg ha−1 y−1 %−1) for WEPP (figure 11a), as RUSLE2 soil loss rates increase. As slope increases, WEPP results do not change for each percentage increase in slope and RUSLE2 results increase from 0.01 to 0.3 tn ac−1 yr−1 %−1 (0.02 to 0.74 Mg ha−1 y−1 %−1) as slopes increase. In the managements with tillage (figures 11b through 11d), there is a similar trend as reported for slope length.
Comparison of soil loss rate with respect to slope gradient for all simulations in the Vermilion Study and classified by four managements: (a) no-till, (b) conservation tillage, (c) conventional tillage, and (d) tilled-fallow. The clay-textured soils are highlighted in color, while all other textured soils are shown in gray. The solid line is the one-to-one relation. Note the soil loss rate ranges within the graphs.
Discussion. Herein, collaboration between two national USDA ARS laboratories has resulted in a plethora of information regarding soil loss prediction where a significant number of factors were tested. Under no circumstances should either RUSLE2 or WEPP erosion technology be considered as “correct” in this study, since the outcomes are long-term averages, given a certain set of soils, topography, management, crop yields, and climate input. No end user should assume that observed differences between the two technologies specifically makes one technology “right” and the other “wrong.”
In the Iowa Study, G1 simulations were designed to test fundamental factors (i.e., slope length, slope gradient, soil, and management). G2 simulations were designed to test yields of common crops (i.e., corn and soybeans) using simple operations (i.e., chisel and moldboard plow, no-till). G3 simulations were designed to test “unit plot” conditions (i.e., tilled-fallow management, 72.6 ft [22.1 m] slope length, 9% slope gradient) and effects of climate and soils. In the Vermilion Study, every effort was made to provide both models with an identical climate input with the same rainfall erosivity.
The goal of this study was to evaluate and compare how two models respond to similar sets of inputs, first in the Iowa Study, then with more carefully parameterized climate in the Vermilion Study. Overall, there was a substantial improvement in model agreement in the Vermilion Study as a direct result of this effort and we evaluated fewer managements, slope lengths, and slope gradients. Still, there were points of disagreement between the two models in both studies, which are detailed in the following sections topically and briefly summarized in the conclusion.
Climate and Hydrology. A major motivation behind the design and execution of the Vermilion Study stemmed from climate differences in the Iowa Study (figure 1, tables S1 and S2). Climate in the Iowa Study followed the standard procedures and databases for conservation management planning implemented for use at the NRCS county office level; however, across the five Iowa counties, WEPP climate input erosivities were 47% to 80% greater than their RUSLE2 counterparts (table S1). Figures 2 and 3 show these results visually on all simulations in the Iowa and Vermilion studies. In Figure 7, all box and whiskers and associated interquartile ranges (IQR) shrink substantially in the Vermilion Study as compared to the Iowa Study, while maintaining a similar trend. The ED in figure 8 decreases by a factor of 5 and the slope flattens. The soil loss rates per unit length in the Vermilion Study were roughly half that of the Iowa Study for similar managements (figure 9). However, limited managements, slope lengths, and slope gradients also contributed to observed differences in the Vermilion Study (figures 8 and 9). Therefore, not all the observed differences may be attributed to climate alone, though it was a substantial factor.
The climate input for the Vermilion Study was checked for similarity in the following three metrics: precipitation, erosivity, and erosivity density (table S2). These metrics were within 0.5% of each other for the two model inputs, and that is why the Vermilion Study resulted in much greater agreement. The model agreement for the Vermilion Study was much improved compared to the Iowa Study (table 3 and figure 3d), though only one climate factor level was analyzed for the Vermilion Study. There was significant improvement in NSE and PBIAS statistical measures. The Iowa Study had NSE values that were low or negative (maximum NSE was 0.31) and PBIAS ranging between −78 and −24 (indicating significantly greater predictions from WEPP as compared to RUSLE2). In the Vermilion Study, the NSE increased to 0.54 and the PBIAS magnitude reduced to 4.23 (indicating slightly greater predictions from RUSLE2 as compared to WEPP). This agreement was not only observed across the whole of simulations but also on an individual factor level (table S6). In short, model agreement in the Vermilion Study was better, but there were a few notable disagreements, which are examined in the following sections.
Table 4 and figure 4 show that both RUSLE2- and WEPP-predicted soil losses increase with increasing climate intensity and runoff magnitude. There was a monotonically increasing relation between RUSLE2 soil loss and R-factor (table 4), but not a monotonically increasing relation between WEPP runoff and soil loss (see scatter in figure 4). Further, the order of WEPP soil loss by county (i.e., climate) differed from that of RUSLE2 (figure 5). When runoff is plotted with average soil loss by county (figures 4 and 5), a smaller runoff prediction can correspond to a larger soil loss prediction (Adair County) when all nonclimate factors are equal. When the Jackson County, Iowa, results are removed (one of two acceptable climates in the Iowa-G1 and -G3 simulations), there exists a consistent trend of higher soil loss for increased runoff due to the different erosivity ranking of Jackson County between the two models (figure S1). The trend shown in figure 5 is consistent for all soils in the Iowa-G3 Study (also see figure S2). The Haig silt loam and Flagler sandy loam have higher infiltration capacity than their textural counterparts, and the Fayette silt loam and the Dickinson sandy loam have higher soil loss than their counterparts (figure S4). This result demonstrates that WEPP’s soil loss predictions can be highly dynamic and do not exhibit a linear response to runoff predictions. The loam, clay loam, and silty clay soils all have similar results when Jackson County, Iowa, results are removed from the analysis.
Patterns of soil loss as impacted by climate are more pronounced when limiting the simulation complexity to tilled-fallow management. In the Iowa-G3 Study, 19 soils were used under unit plot conditions. Previously, simulations indicated that RUSLE2 and WEPP have different climate ordering for the five counties in Iowa (figure 5). The Iowa-G3 Study simulation results were also examined by sorting the simulations by soil, then sorting by RUSLE2 soil loss (small to large), and finally by plotting similar soil textures (figure S4).
Figure S4A shows, when isolated by county, the Dickinson sandy loam soil has the largest soil loss, followed by Flagler and then Zenor. Also, both RUSLE2 and WEPP respond similarly, with soil loss increasing for RUSLE2 in order by county: Plymouth, Jackson, Hardin, Des Moines, and Adair; and for WEPP in order by county: Plymouth, Jackson, Adair, Des Moines, and Hardin.
The loam soils (Clarion, Kenyon, and Nicollet; figure S4B) produced the same pattern concerning the order of counties by climate intensity for the two models as the sandy loam soils. However, as observed most clearly with the Jackson County results discussed earlier, the two models do not respond similarly to these loam soils. WEPP presents decreasing soil loss in the order of Clarion, Nicollet, and Kenyon; whereas RUSLE2 presents increasing soil loss in the order of Clarion, Nicollet, and Kenyon.
Soils. The statistics for the Iowa-G3 and Vermilion studies indicated reasonably agreeable RUSLE2 and WEPP soil loss estimates when comparing all simulations (table 3), although the statistical measures indicate rather poor agreement at the individual texture level and mean of all soil texture levels for unit plot conditions (table 5; see also Kinnell [2017]). Interestingly, the mean of all texture levels was more agreeable in the Iowa-G3 Study than in the Vermilion Study where the climate erosivity was nearly identical. However, when two soils were removed from this analysis (the Aholt and Portageville clay soils), the agreement between the two models substantially improved (NSE = 0.57 as opposed to NSE = −0.39). Therefore, the agreement for unit plot conditions was better for the other 19 soils in the Vermilion Study, and these 2 soils were significant outliers. In both the Iowa and Vermilion studies, the average annual soil losses predicted by WEPP for several soils (Ida silt loam, Webster clay loam, Clarion loam, Perks loamy sand, Sparta sand, Fruitfield coarse sand, all silty clay, and all clay) trend differently than those predicted by RUSLE2. There are several points to make concerning the effects of soils on simulated soil loss.
Generally, soil loss predictions from both models followed the one-to-one relation (figure 6; Iowa-G3). Considering the CL texture in the tilled-fallow results from the Iowa-G3 Study (figures 7a and 7b), the mean ED is 3.3 tn ac−1 yr−1 (8.15 Mg ha−1 y−1), though the ED for the Webster clay loam was 115% higher than the Canisteo and 141% higher than the Nicollet clay loams, while RUSLE2 estimates are very similar to one another (i.e., between 40.99 and 41.05 tn ac−1 yr−1 [101.25 and 101.39 Mg ha−1 y−1] when averaged over all counties; figure S4C), since the K-factor for each of these soils is 0.28 (all K-factor units are tn-ac-hr/hundreds of ac-ft-tn-in). The ED difference is linked to the disparity in WEPP-generated soil loss estimates for the Webster clay loam (59.63 tn ac−1 yr−1 [147.29 Mg ha−1 y−1]), Canisteo clay loam (38.60 tn ac−1 yr−1 [95.34 Mg ha−1 y−1]), and Nicollet clay loam (34.20 tn ac−1 yr−1 [84.47 Mg ha−1 y−1]). Potential for detachment and transport with the Webster clay loam soil within WEPP can be explained by lower relative baseline critical shear stress (figure S3B) and increased relative baseline rill erodibility (figure S3D), compared to Canisteo and Nicollet clay loam soils.
Similar observations are evident for silty clay, loam, and silt loam textures in both the Iowa Study (figure 7b) and the Vermilion Study (figure 7d). Both studies show the same trends with the soils. Figure S4D shows that silty clay results from WEPP are all roughly 74% higher than their RUSLE2 counterparts and ordered slightly differently (i.e., within the silty clay soils, RUSLE2 predicted the greatest soil loss from Onawa silty clay while WEPP predicted the least for the same soil). For example, the soil loss ED for the Ida silt loam (Iowa Study) departs from the silt loam texture trend (figure 7b). The departure arises from soil erodibility values, where the RUSLE2 K-factor for the first two (Haig and Fayette silt loam) is 0.37 and 0.43, and it is 0.49 for the Ida silt loam. The WEPP soil loss estimate for the Fayette is higher than the WEPP soil loss estimate for the Ida (i.e., difference of 5.6 tn ac−1 yr−1 [13.83 Mg ha−1 y−1]). The ED for the Fayette and the Haig is very similar (6.2 and 6.1 tn ac−1 yr−1 [15.31 and 15.07 Mg ha−1 y−1]); however, the ED for the Ida is −8.2 tn ac−1 yr−1 (−20.25 Mg ha−1 y−1); therefore, a change in soil loss ED by 14.5 tn ac−1 yr−1 (35.82 Mg ha−1 y−1) for a soil that is seemingly more erodible (K = 0.49) creates uncertainty concerning the soil loss calculation.
Considering both Iowa and Vermilion studies, Vermilion ED values were generally lower magnitude while exhibiting similar ED trends for both soil textures and individual soils. This was observed for the clay (Lamoni), loam (Clarion), silt loam (Ida), and silty clay (Onawa) soils. Usually, clay soils are highly transportable but are very often hard to detach (Ellison 1948). Generally, RUSLE2 estimates are based on K-factor, and these estimates rank from high to low soil loss as the following: SiL, CL, C, L, SiC, SL, S; whereas WEPP estimates rank from high to low soil loss as the following: C, SiL, SiC, CL, SL, L, S. The differences in soil responses with clay textures seem to be associated with the ability of the rainfall and runoff (figure S7) to detach and transport the materials; however, investigations into program-specific soil detachment and transport methodologies are needed.
Topography. The EDs between RUSLE2 and WEPP were compounded within complex factors. For the Iowa Study simulations, the soil loss differences between the two technologies were underpinned by a greater sensitivity of WEPP predictions to topography inputs. ED values were greater in the Iowa Study due to differences in climate input erosivity. These ED values decreased in the Vermilion Study due to a study design that minimized differences in climate input erosivities (figures 2 and 7). This is why the ED intersection changed from 50 ft to 100 ft (15.2 to 30.5 m) and the ED inflection point changed from 3% to 5% between the Iowa and Vermilion studies, respectively (figure 8). These points of intersection and inflection would be expected to change depending on the specific study design (e.g., input levels).
In both the Iowa and Vermilion studies, the linear regressions of the tillage managements soil loss rates with respect to length (figures 9a and 9b) were on slopes of 1.08 and 1.57, respectively, and the intercepts were 0.084 and 0.021, respectively. In the Iowa Study, the linear regression of the no-till managements soil loss rates with respect to length yielded a slope of 0.41 and an intercept of 0.002 (figure 9b). The analysis could not be performed on the single no-till point from the Vermilion Study (figure 9d). In the Vermilion Study, if the tilled-fallow point is included in the linear regression, the slope of tillage managements with tilled-fallow soil loss rates with respect to length had a slope of 1.56 and an intercept of 0.022 (figure 9d).
Management. Both models responded as expected concerning crop yields, where increasing crop yields resulted in decreasing average annual soil loss levels due to increased canopy cover and residue biomass.
Soil loss ED in the Iowa-G1 and Iowa-G2 studies followed similar general trends with management. Chisel operation ED was larger than moldboard plow or no-till operations. Generally, spring operations had higher soil loss ED and soybeans had higher soil loss ED under tillage operations. In the Vermilion Study when the clay-textured data were removed (eight soils removed) from the comparison (figure S5) and the managements combined, the model agreement improved (figure 12). The slope of linear regressions of RUSLE2- and WEPP-simulated soil loss rate per unit length for each management increased from 0.33 to 0.49 for no-till, 0.70 to 1.22 for conservation tillage, 0.48 to 0.97 for conventional tillage, and 0.49 to 1.02 for tilled-fallow. The slope of each management should approximate one. Seven of the nine clay-textured soils responded differently than the loam, silt loam, and sandy loam soils. Very similar to the slope length analysis, when certain disagreeing soils were removed (figure S6), the result was agreeable (figure 13). In figure 13, when 11 soils were removed, in all cases except for no-till management, the slope of linear regressions fitted through the WEPP versus RUSLE2 simulated soil loss rate per unit gradient for each management improved. The regression slope of the remaining soils changed from 0.12 to 0.07 for no-till, 0.51 to 0.70 for conservation tillage, 0.36 to 0.60 for conventional tillage, and 0.42 to 0.92 for tilled-fallow.
Comparison of soil loss rate with respect to slope length by management for selected simulations (a) with all soils shown and (b) with 13 of 21 soils (clay-textured soils were omitted) in the Vermilion Study. The solid line is the one-to-one relation.
Comparison of soil loss rate with respect to slope gradient by management for selected simulations (a) with all soils shown and (b) with 10 of 21 soils (Ida silt loam, Webster clay loam, Kenyon loam, all silty clay soils, Sparta sand, Fruitfield coarse sand, and all clay soils were omitted) in the Vermilion Study. The solid line is the one-to-one relation.
In the Iowa Study, soil loss reductions implied from conversion of tilled managements (FP, SP, FC, and SC) to no-till management are different between the two technologies: RUSLE2 has 92% to 96% soil loss reduction due to no-till corn, 69% to 81% soil loss reduction due to no-till corn–soybean rotations, and 48% to 65% soil loss reduction due to no-till soybeans; while WEPP provides 94% to 95% soil loss reduction due to no-till corn, 94% to 95% soil loss reduction due to no-till corn–soybean rotations, and 92% to 93% soil loss reduction due to no-till soybeans. The differences in soil loss due to conversion to no-till results in increased soil loss reduction of 0.4% for continuous corn, 20% for corn–soybean rotations, and 36% for continuous soybean when implementing WEPP. Also, the difference equates to an overall soil loss reduction that is 2.5 times higher than currently incorporated in RUSLE2. Similar trends were found in a study by Stolpe (2005). Although no directly comparable analysis was possible in the Vermilion Study, WEPP predicted an increased soil conservation due to no-till management relative to RUSLE2 in part because WEPP’s rate of soil loss per unit gradient was less sensitive than RUSLE2 by a factor of seven (i.e., slope of linear regression for “All Other” soil textures was 0.14; figure 11a).
The cropping management systems used in the WEPP and RUSLE2 simulations were developed to be as similar as possible using the default vegetation, operation, and residue databases. WEPP used a calibration step to target the crop yields set by each study that was implemented in RUSLE2 simulations. Beyond matching the yields between the two models, other management database parameters were not evaluated to determine their agreement. Some examples of important parameters that impact erosion predictions include the amount of biomass left on the field after harvest, the decomposition rates of the various crop residues, canopy cover through time, senescence, and aspects of soil disturbance related to the burying and resurfacing of residue. None of these critical factors were evaluated for uniformity across the two databases used in the Iowa or Vermilion studies. The degree to which adjusting these management parameters for agreement and their effect on the soil loss comparisons was not evaluated in either the Iowa or Vermilion studies.
WEPP soil loss predictions in the Vermilion Study for almost all no-till management simulations (20 of 21 soils) and several conservation management simulations (8 of 21 soils) at a 2% slope gradient were higher at a 50 ft (15.2 m) slope length than longer slope lengths (75 to 150 ft [22.9 to 45.7 m]). WEPP soil loss was 0.58, 0.57, 0.56, and 0.54 tn ac−1 yr−1 (1.43, 1.41, 1.38, and 1,33 Mg ha−1 y−1) (average of 20 of the 21 soils simulated) for increasing slope lengths of 50, 75, 100, and 150 ft [15.2, 22.9, 30.5, and 45.7 m] for 2% no-till management. In the 2% conservation management simulations, impacted soils were two clay loams (Canisteo and Nicollet), all three loams soils (Clarion, Kenyon, and Nicollet), and the sandy soils (Sparta, Fruitfield, and Perks). WEPP-estimated soil loss was 1.07, 1.05, 1.05, and 1.04 tn ac−1 yr−1 (2.64, 2.59, 2.59, and 2.57 Mg ha−1 y−1) (average of 8 of the 21 soils simulated) for increasing slope lengths of 50, 75, 100, and 150 ft for 2% conservation management. Similar effects were seen with the Nicollet clay loam for 2% conventional management, the sandy soils (Sparta, Fruitfield, and Perks) for 2% tilled-fallow management, and the sandy soils (Sparta, Fruitfield, and Perks) for 5% no-till management.
The decrease in soil loss with increasing length is due to a slight decrease in predicted runoff with increasing slope lengths in WEPP, resulting in decreased sediment transport capacity. In mild slopes and high residue conditions, sediment transport capacity typically controls soil loss (at least in WEPP simulations), and the predicted soil loss is, therefore, slightly decreased. The same trend of decreasing runoff with longer slopes is also exhibited for steeper slopes and lower residue conditions; however, because transport capacity is much greater in these cases, average annual soil loss increases with increasing slope lengths.
Summary and Conclusions
The Iowa and Vermilion studies examined the effects of topography, soil type, management, crop yield, and climate on soil loss simulated by RUSLE2 and WEPP regarding conservation planning. The Iowa Study investigations showed selective agreement within the G1, G2, and G3 simulations, while the Vermilion Study found broader consensus when forced with identically erosive climate inputs for the two erosion prediction technologies. Under no circumstances should either RUSLE2 or WEPP erosion technology be considered as “correct” in this study, since the outcomes are long-term averages, given a certain set of soils, topography, management, crop yields, and climate input. No end user should assume that observed differences between the two technologies specifically makes one technology “right” and the other “wrong.”
Regarding specific disagreements, there were differences in the Iowa Study involving soil loss rankings by county and the magnitude of erosivity in the two model climate databases. Differences related to erosivity magnitude were resolved in the Vermilion Study by providing identically erosive climate inputs to the two models. Although, within both studies and irrespective of the climate input, there were differences in the simulated soil loss between the two technologies that were attributed to differences in soil erodibility behavior across textures, varying sensitivity to topographical inputs, and distinct outcomes for no-till agriculture practices. Future work should focus on reconciling climate databases, soil erodibility, and no-till managements.
Better insight into the differences between the models was developed by selecting many combinations of typical slopes, soils, and crop managements along with climate data for the comparison study rather than using a smaller set of observed data such as the USLE database. An outcome of this study is the raised awareness and illustrated need for actual observations and historical data sets in a statistically sufficient quantity and broad diversity of experimental field conditions to (1) inform a more objective model comparison, (2) invite and address concerns of skeptics of either model, (3) guide and expand continued collaborative model development, and (4) attract a discussion and broader involvement of the governmental and academic scientific community in all these aspects.
Supplementary Material
The supplementary material for this article is available online at https://doi.org/10.2489/jswc.2024.00072.
Acknowledgements
The authors acknowledge the contributions to this project made by Dr. Chi-hua Huang, retired soil scientist with the USDA Agricultural Research Service at the National Soil Erosion Research Laboratory. We also would like to thank Dr. Marlen Eve, deputy administrator for the Natural Resources and Sustainable Agricultural Systems (NRSAS) program area at the USDA Agricultural Research Service, Office of National Programs, for his suggestions and guidance.
- Received August 29, 2023.
- Revision received April 12, 2024.
- Accepted June 5, 2024.
- © 2024 by the Soil and Water Conservation Society
References
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.