Modification of the Soil Vulnerability Index to account for increased erosion risk from winter precipitation in the southern United States

Results and Discussion

Classification of Data Sets. The regression between average annual monthly R-factor and monthly precipitation shows a different behavior at the sites located in Vermont, Ohio, Maryland, Pennsylvania, Iowa, and Missouri compared to Georgia, Mississippi, Arkansas, and Texas, hereafter referred to as “North” and “South,” respectively. Labeling the North and South groups is a convention limited to this article. It refers to the geographic location of these sites relative to the set of 11 sites and does not imply that all sites in the North or South would behave similarly. In the North, greater summer precipitation (April through September) coincides with greater rain intensity and greater R-factor. These sites experience less rainfall-equivalent precipitation amount and intensity in the winter. As a result, the regression between monthly R-factor and precipitation is strong (i.e., r² > 0.5) (table 1). In the South, summer and winter precipitation amounts are similar (table 2), but summer and winter intensities and R-factors differ. As a result, the strongest correlation between monthly precipitation and monthly R-factor at these southern sites has an r² of 0.24. In addition, rain events are more intense in the South than they are in the North for both seasons: winter and summer average R-factors in the South are greater than in the North (table 2). In summary, winter precipitation in the North is less intense than in summer in terms of erosive power; it consists of lighter rainfall events as well as snowfall events. Winter precipitation amounts in the South are similar to summer precipitation, and include large, erosive events.

Vulnerability of Cropland to Precipitation. The average monthly simulated sediment loss showed a more pronounced correlation with average monthly precipitation amounts (tables 3 to 5) than with erosivity, indicated by the monthly R-factors (table 6). Under fallow conditions, the sediment yields were higher of course, and the correlation of sediment yields versus precipitation amounts or R-factors were more pronounced in the summer, but not always in the winter (table 7). Some regression slopes of sediment yield versus R-factors were negative, which did not happen under growing crops. Since SVI was based on sediment yields obtained with growing crops but no conservation practice, we used those conditions to present the analysis and highlight the results under fallow conditions when necessary. Figure 3 shows the simulated average monthly sediment yields as a function of average monthly precipitation amount for the No-Practice scenario in each watershed for the 0% to 2% slope. We show the results for this slope class because it is a common slope class for cropland, and one found in each of the modeled watersheds. Charts for the three other slope classes are shown in supplemental figures S2 to S4.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Charts of average monthly sediment yields for a No-Practice management scenario versus average annual monthly precipitation for the 0% to 2% slope class in the (a) Goodwater Creek Experimental Watershed (GCEW), (b) Maumee River watershed (Maumee), and (c) WE-38 Experimental watershed (WE-38). Summer includes April through September months; winter includes October through March.

The results confirm that the erosion risk changes with management and with slope. Sediment yields are greater in both GCEW (corn–soybean rotation) and Maumee (continuous soybean) than they are in WE-38 (continuous corn) for the <2% slope class (figure 3). For slopes >2% (supplemental figures S2 to S4), Maumee (continuous soybean) sees greater sediment yields than WE-38 (continuous corn) or GCEW (corn–soybean). Although multiple factors may explain the results, those are consistent with the good ground protection provided by corn residues contrasted with the poor protection provided by soybean residues, and the enhanced benefit of these residues for greater slopes.

For evaluating the risk of seasonal precipitation in one region relative to other seasons or regions, we want to consider the regression slopes of sediment yield versus precipitation amount and R-factor, and significant differences that exist between them. For each of the watersheds and each slope class, the regression slopes for the summer precipitation data sets were similar whether in the North or the South (e.g., 0.008 and 0.01 for GCEW, <2% slope, respectively) (table 3). However, the regression slopes for the winter southern precipitation data sets (figure 3, dash line) were always significantly greater (p = 0.05) from the slopes of the winter northern data sets (figure 3, dot-dash lines). For GCEW and Maumee, they were also greater than the regression slopes for the summer data sets. Thus, winter precipitation in the South poses a greater risk of erosion and its associated contaminant transport.

This corresponds to previous findings about the accuracy of the SVI in Beasley Lake watershed; the SVI ratings were too low for what the data showed in terms of sediment yield and erosion (Lohani et al. 2020; Yasarer et al. 2020). Soil loss data in these regions confirm that sediment yields are greater during the winter than when crops are growing. Kuhnle et al. (1996, 2008) showed greater winter precipitation and runoff in the Goodwin Creek Experimental Watershed in Mississippi with similar fine sediment concentrations, thus resulting in greater winter sediment loads. Harmel et al. (2006) showed large winter sediment loss from the Riesel, Texas, watersheds and concluded that small grains and cover crops are important conservation practices against soil loss in the region. Aryal et al. (2018) and Reba et al. (2020) found greater discharge and losses of suspended sediment, total phosphorus (P), and total nitrogen (N) during the nongrowing season (November to April) than the growing season in the Mississippi Delta region of eastern Arkansas. Along with cover crops, they also mentioned flooding fields in winter for waterfowl and groundwater recharge as a conservation practice that targets winter losses. Reasons for these high winter losses include winter precipitation amounts similar to those in the summer, low ground cover at that time, and winter R-factors only slightly smaller than during the summer (table 2).

Regressions of average annual monthly sediment loss with average annual monthly R-factor were inconclusive. Table 6 shows detailed results for GCEW; results for Maumee and WE-38 were very similar (table 7). Substantial scatter was observed around the regression lines, indicating a weak relationship, with correlation coefficients frequently below 0.1 (table 6). Notably, the regression slopes were negative for the winter months in GCEW, which is contrary to expectation and our understanding of the processes that control erosion and sediment transport by water. There was no significant difference between the regression slopes of any season/region combinations for any of the modeled watersheds. This result was undoubtedly surprising as we expected that the R-factor would be a likely candidate variable because it characterizes rainfall in ways that are relevant to erosion and nutrient losses. Even under fallow conditions, which eliminate the impact of management on soil erosion, there was still substantial scatter, some negative regression slopes, and no significant difference between the regression slopes for different season/region combinations (table 7). A lack of consistency between the 1960 to 1989 rainfall data (and up to 1999 in several cases) used to calculate the RUSLE2 R-factors (USDA ARS 2013) and the 1985 to 2014 precipitation inputs used in this study may explain this unexpected result. Given the changes in precipitation characteristics and seasonality from 1960 to 2014 (Giesling et al. 2022; Tobin et al. 2015; Phung et al. 2022), there may be a discrepancy between the two. Indeed, Dabney et al. (2012) recognized the need to update the R-factors to account for changes in precipitation characteristics. Yet, another explanation may be that R-factor calculations use subdaily data but exclude both small events with depths less than 12 mm (0.5 in) (USDA ARS 2013) and very large events with a return period of 50 years or greater (USDA ARS 2013). It excludes small events because their erosive power is small. It excludes very large events because extreme precipitation would affect the average erosive power of annual rainfall at one site, which is the basis for conservation needs assessment and what the R-factor describes (USDA ARS 2013). The thinking in policy design was that average conditions, not extreme events, should drive the amount of conservation needed in crop fields. If extreme events cause damage, rebuilding the practice may be cheaper than designing for greater than average erosive power. The four precipitation data sets that show poor correlation between precipitation and R-factors are all affected by hurricanes. If the erosivity calculations ignore these events, the correlation between R-factor and precipitation will necessarily be lower or even lost. McGehee et al. show increases in erosivity values of up to 20% when small and extreme events are included compared to erosivity values when those are omitted (McGehee et al. 2022).

Effect of Southern and Northern Precipitation Data Sets on Average Annual Sediment Yields. Figures 4 to 6 show how annual and seasonal monthly sediment yields with growing crops obtained with each precipitation data set for each slope class compare to the thresholds defined by Phung et al. (2023) for each watershed. In the GCEW, SVI classified most cropland (90% of cropland area) in the Moderate vulnerability class despite the low slopes because of the hydrologic group D and the High erodibility of all the cropland soils (Phung et al. 2023). Some cropland was in the Moderately High (10%) and High (<1%) categories. Figure 4 shows that only cropland with less than 2% slope and with northern precipitation data sets met the upper average annual threshold of 4.8 Mg ha⁻¹ for the Moderate class defined by Phung et al. (2023). Cropland with <2% slope met the SVI classification for Moderate when northern precipitation data sets were used as inputs, whereas it exceeded the 4.8 Mg ha⁻¹ threshold when southern precipitation data sets were applied (figure 4). With southern precipitation, cropland vulnerability is classified as Moderately High; that is, it increased one vulnerability class (figure 4). For cropland with greater slopes, the sediment-based classification is Moderately High or High. The 2% to 4% slope cropland vulnerability class did not increase, but the average annual sediment yield almost doubled from just above the Low threshold value (5.6 relative to 4.8 Mg ha⁻¹ y⁻¹) for Moderately High to just under the upper threshold value (10.1 relative to 10.3 Mg ha⁻¹ y⁻¹). The 4% to 6% slope cropland did not increase by one category even though sediment yield doubled, because it was already in the highest vulnerability class.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Average (av.) annual monthly and total values of sediment yield for the Goodwater Creek Experimental Watershed, by season and region of the precipitation data sets, classified according to Phung et al. (2023) thresholds of 0.8, 4.8, and 10.3 Mg ha⁻¹ y⁻¹, and according to Soil Vulnerability Index (SVI) classification. Each box represents the maximum, mean, and minimum monthly sediment yield.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5

Average (av.) annual monthly and total values of sediment yield for the Maumee River watershed, by season and region of the precipitation data sets, classified according to the Phung et al. (2023) thresholds of 5.8, 15.6, and 25.4 Mg ha⁻¹ y⁻¹, and according to Soil Vulnerability Index (SVI) classification. The 4% to 6% cropland is not featured because this slope class represented a small, nonrepresentative fraction of the cropland. Each box represents the maximum, mean, and minimum monthly sediment yield

• Download figure
• Open in new tab
• Download powerpoint

Figure 6

Average (av.) annual monthly and total values of sediment yield for the WE-38 Experimental watershed, by season and region of the precipitation data sets, classified according to the Phung et al. (2023) thresholds of 3.3, 10.9, and 18.6 Mg ha⁻¹ y⁻¹, and according to Soil Vulnerability Index (SVI) classification. Each box represents the maximum, mean, and minimum monthly sediment yield.

The Maumee River watershed was more complex because slope was not the major factor that affected vulnerability. While SVI classified nearly all (98.98%) cropland with <2% slope in the Low vulnerability class, it also classified some land with greater slopes (including >6%) in the Low vulnerability class because of the hydrologic soil group. Inversely, all the land that SVI classified in the High vulnerability class had slopes >6%. The 2% to 4% cropland increased by one category when the southern precipitation data sets were applied (figure 5). The 0% to 2% slope did not increase in vulnerability class, but the average annual sediment yield obtained with southern precipitation data sets (5.4 Mg ha⁻¹ y⁻¹) was very close to the Low/Moderate threshold of 5.8 Mg ha⁻¹ y⁻¹ for that watershed. Similar results were obtained with the WE-38 watershed model (figure 6). The 2% to 4% and 4% to 6% cropland increased by one vulnerability class, but the 0% to 2% sloped land did not. However, the average annual sediment yield almost doubled for the 0% to 2% land and came very close to the Low/Moderate threshold of 3.3 Mg ha⁻¹ y⁻¹ for that watershed. In both watersheds, the cropland with slopes >6% did not change vulnerability class because it was already at the highest level under northern precipitation data sets. Given these results, increasing vulnerability by one class in locations characterized by significant rainfall in winter would reflect the need for increased conservation to protect the soil during winter as well as during summer. Conservation practices that address winter conditions include continuous ground cover, which cover crops, winter cash crops (wheat [Triticum aestivum L.]), or perennial grass can provide; no-till; avoiding fall tillage if tillage is necessary; and being careful with winter grazing of silage or residues. Of course, other conservation practices that address soil loss remain useful (grassed waterways, terraces, upland buffers, and others).

Inherent Vulnerability Relationship with Rainfall. These results suggest that the inherent vulnerability (i.e., the ability of cropland to withstand storm events without excessive soil or nutrient losses) does vary with rainfall characteristics and especially with the rainfall winter characteristics. Two land tracks with a similar SVI rating and with the same number of ACT practices, but one located in the North and one in the South, do not have the same conservation requirements and should not have the same priority for conservation investment. Thus, these results contradict the SVI underlying assumption that it is independent of rainfall characteristics.

This might explain why the second CEAP I to CEAP II Comparison report (USDA NRCS 2022) shows that out of all the cropland area that exceeds the threshold of 4.5 Mg ha⁻¹, 40% were receiving >1,143 mm of rainfall (table 8). Out of those, 27% were classified in the Low SVI category, which is defined by a sediment loss under No-Practice scenario of <4.5 Mg ha⁻¹. Clearly, the Low SVI rating is not appropriate for these 27%. Thus, inherent soil vulnerability cannot be totally independent of rainfall. Indeed, high rainfall on land with insufficient ground or canopy cover will start a vicious cycle of erosion, which leads to eroded land that is more vulnerable, more erosion, and ultimately completely eroded land when not addressed with appropriate conservation practices. It is therefore justified to modify the runoff component of SVI and increase vulnerability by one class in regions where winter conditions augment the erosion risk. In this study, we have not investigated the two subsurface loss components of the SVI, but there is no reason to think that rainfall would not affect the subsurface movement of contaminants in similar ways (Williams et al. 2018).

Adjustment Outcomes in the Southern Watersheds. The Beasley Lake watershed in Mississippi, Riesel watersheds in Texas, Little River Experimental watershed in Georgia, and the Delta sites in Arkansas are the CEAP watersheds located in these southern states. It is therefore useful to consider the implications of the proposed change in SVI ratings in these watersheds. In Beasley, most of the crop fields had a Low or Moderate vulnerability rating due to slopes less than 2%; however, most of the watershed soils are hydrologic group C or D with high K-factors and high risk of soil loss (Lohani et al. 2020). With the original SVI rating criteria, soils in hydrologic group D would never be ranked above Moderate unless they had slopes higher than 2%, and soils in group C would never be ranked as High without a slope greater than 6%. The proposed change in SVI ranking considering precipitation effects would increase Beasley soil ratings to Moderate or Moderately High, which would better match model-simulated sediment yield results and measured runoff and suspended sediment values from cropland (Yasarer et al. 2020; Locke et al. 2020).

At the Riesel watersheds, many of the fields already had a Moderately High runoff vulnerability rating. Thus, the change would move them into the High category, which is consistent with the average annual sediment yields from 1939 to 1947 under conventional tillage and no conservation practices (36 to 42 Mg ha⁻¹) (Harmel et al. 2006). In the Little River Experimental watershed in Georgia, most of the cropland had a Low runoff vulnerability rating, which would change to a Moderate one. A quarter of the cropland was already in a Moderate category and would switch to a Moderately High rating. Finally, 10% of the watershed had a Moderately High or High rating and would fall in the High category. Those are areas with slopes >2%, erodible soils (soil erodibility >0.28), and C or D hydrologic group; they happen to be close to streams (supplemental figure S5). Settimi et al. (2010) also identified these areas as being at high risk of erosion and water quality degradation, independently of the SVI technology. Whether the rating should be Moderately High or High is difficult to say. The number of implemented practices did not correlate well with the identified risk (Settimi et al. 2010), and other factors (availability of funding, soil health concerns, or subfield concerns, for example) could affect the decision. The reference crop fields used to assess SVI at the Delta sites in Arkansas had a Low or Moderate vulnerability rating (Lohani et al. 2020). The change would increase the vulnerability to Moderate and Moderately High. Thus, one (when classified as Moderate) or two (when classified as Moderately High) ACT conservation practices may be needed to attain an acceptable risk according to the CCBI. No known modeling results exist for these Arkansas sites, and monitoring data include only a few years of data, with most events missed because of equipment malfunction or flooding (Aryal et al. 2018). Thus, they provide unreliable annual sediment yields. However, one site, for which soil erodibility was very high (0.49) for most of the field, had an average event sediment yield of 0.6 Mg ha⁻¹ with 14 events being measured over two years and a maximum event load of 128 Mg ha⁻¹. Ongoing monitoring at these sites may provide confirmation or question the increase of vulnerability class.

Regions Where These Conditions Exist. In this limited study, we separated the 11 precipitation data sets based on the correlation between measured average annual monthly precipitation and average monthly R-factor from the RUSLE2 database. This resulted in two groups: data sets with a r² > 0.5, which were the most northern data sets among the 11 used in the study, and the others, which were all located in the South (table 1). However, these were only 11 specific precipitation data sets, which is not sufficient to define if and where these conditions exist regionally. We asked the CEAP Cropland modeling team to extract from their APEX runs the average annual monthly erosivity calculated by APEX and the 1986 to 2015 average annual monthly rainfall amounts given as input. These runs use the precipitation record at one single station per county (C. Lester, Natural Resources Specialist, NRCS, personal communication). Note that APEX calculates rainfall erosivity in a different way than for the RUSLE2 database (Williams 1995). One major difference is that it includes the events that are <12 mm and those that have a return period >50 years. In addition, APEX uses daily rainfall and cannot calculate a peak rainfall intensity from subdaily data. To remedy this limitation, it uses the average monthly maximum half-hour intensities provided as input to APEX, makes assumptions about the rainfall distribution and the peak rainfall rate during each event, and introduces a random factor (Williams 1995). The spatial distributions of average annual monthly precipitation and erosivity (supplemental figure S1) match the spatial patterns presented by McGehee et al. (2022). From these, we calculated and mapped the coefficients of determination (r²) between monthly erosivity and monthly precipitation for all the counties covered by the CEAP 2 Survey (USDA NRCS 2022) (figure 7).

• Download figure
• Open in new tab
• Download powerpoint

Figure 7

Coefficients of determination between average annual monthly erosivity index calculated by APEX and precipitation for the CEAP National Resources Inventory points, by county.

Figure 7 and supplemental table S2 show that the r² remained high at the seven study sites where the correlation was high (r² > 0.5; table 1). Additionally, all the sites that showed poor correlation in table 1 still show poor correlation: Turner County in southwest Georgia (r² ~ 0), Sunflower County in Mississippi (r² = 0.03), Falls County in Texas (r² = 0.30), and Mississippi County in Arkansas (r² = 0.02). This independent confirmation that precipitation and erosivity are poorly correlated in these areas also defines a greater region where this is happening. This region includes a large area in the southeast, which could be initially defined as the area influenced by storms that originate in the Gulf of Mexico and moved northeast in a corridor along the Mississippi and Tennessee River. Once in Tennessee, the r² are again >0.5. Defining a more precise r² threshold and identifying the causes of this poor correlation and possible seasonality factors is beyond the scope of this analysis.

Figure 7 also shows an area in the Pacific Northwest with poor correlation between erosivity index and precipitation. Crop production in this region can be rainfed (the Palouse area) but can also be heavily supplemented by irrigation, for example in the Upper Snake/Rock Watershed in southern Idaho for which both the flow regime and the water quality of the Upper Snake River have been heavily influenced by irrigation withdrawals and irrigation returns from irrigated land (Bjorneberg et al. 2008; Lentz and Lehrsch 2010). The risks to stream water quality are therefore different from those caused by agriculture in the eastern half of the United States where rainfall remains the major driver of sediment and nutrient loss. This study did not include watersheds or precipitation data sets from the western United States.

Vulnerability under a Changing Climate. This study demonstrates the importance of rainfall in assessing vulnerability. This has consequences when thinking about vulnerability under changing climate conditions. A drier climate may decrease vulnerability; a wetter one may increase it. However, these results cannot be translated directly to future climate because temperatures and management remained constant in the model. Under future climate conditions, temperatures would likely be higher, and the same crops may not be planted at the same time or in the same place as represented in the model, which might affect sediment loss. However, given the impact of rainfall on vulnerability, there is a need to assess vulnerability under a changing climate so that conservationists, managers, and policymakers have that information when defining vulnerability.

Limitations and Further Studies. In this study, we have a limited number of sites and precipitation data sets. Specifically, we did not have a calibrated model for a watershed located in the South, nor did we have one for a watershed located in the western United States. Given the importance of winter precipitation in the South and the poor correlation between monthly precipitation and erosivity in the West and in a corridor from the Gulf of Mexico to Tennessee, it would be useful to confirm these results with additional model sites.

The results of this study led us to define the South as the region where the coefficient of determination between average annual monthly precipitation and average annual monthly erosivity was <0.5 (figure 7). However, this was defined based on erosivity values extracted from the APEX runs for the CEAP cropland study. APEX uses a different method and uses assumptions about the rainfall distribution and intensity distribution to calculate the rainfall erosivity. There is a need to further define the region characterized as the South, including recalculating R-factors across the United States from modern, observed, and subdaily data; studying the seasonality of the regressions between precipitation and R-factor; and identifying the cause for the poor correlation.

Nutrients, which were not considered in this study, do pose a water quality risk to receiving water bodies, which can be independent from soil erosion. For example, winter nutrient loss when manure is applied on frozen soil is a known source of vulnerability that is specific to winter conditions in the North. The suitability of SVI and the modification we propose has not been evaluated for nutrient losses and remains to be done. Similarly, the determination of a possible modification remains necessary for the leaching component of SVI. However, there is currently no plan to complete such studies.

Finally, as mentioned above, the R-factor values were not calculated with the precipitation data sets we used for sediment loss simulation. Assessing vulnerability with more modern rainfall data and calculating R-factors consistent with these data, and without excluding any events, will clarify the relationship (or lack thereof) between sediment yield and R-factors. A new study has been initiated to accomplish this.