The Extent of Soil Loss Across the US Corn Belt

The extent of soil loss across the US Corn Belt

Evan A. Thalera,1, Isaac J. Larsena, and Qian Yua

aDepartment of Geosciences, University of Massachusetts, Amherst, MA 01003

Edited by David Tilman, University of Minnesota System, St. Paul, MN, and approved December 29, 2020 (received for review December 20, 2019) Soil erosion in agricultural landscapes reduces crop yields, leads to remains a major gap in soil erosion research (19). Large-scale loss of ecosystem services, and influences the global carbon cycle. assessments of soil erosion are often based on model predictions Despite decades of soil erosion research, the magnitude of histor- (20–22) or qualitative information from soil surveys regarding ical soil loss remains poorly quantified across large agricultural the degree of soil degradation (23). In the United States, for ex- regions because preagricultural soil data are rare, and it is challeng- ample, nationwide soil loss trends (24) are simulated using water ing to extrapolate local-scale erosion observations across time and and wind erosion models that have been calibrated with erosion space. Here we focus on the Corn Belt of the midwestern United measurements made on small plots over a period of decades (21, States and use a remote-sensing method to map areas in agricul- 25). It has been debated whether upscaling such predictions to tural fields that have no remaining organic carbon-rich A-horizon. regional or national scales results in an accurate assessment of the We use satellite and LiDAR data to develop a relationship between current magnitude of soil loss in the United States (26, 27). A-horizon loss and topographic curvature and then use topographic Whereas such models are useful for assessing relative rates of data to scale-up soil loss predictions across 3.9 × 105 km2 of the Corn Belt. Our results indicate that 35 ± 11% of the cultivated area has erosion for soil conservation planning, the soil loss predictions do lost A-horizon soil and that prior estimates of soil degradation from not provide information regarding the cumulative soil loss that has soil survey-based methods have significantly underestimated occurred since the initiation of cultivation, and hence the overall A-horizon soil loss. Convex hilltops throughout the region are often magnitude of agricultural soil degradation. completely denuded of A-horizon soil. The association between soil To assess the degree of cumulative soil degradation, soil sur- loss and convex topography indicates that tillage-induced erosion is veys conducted by the US Department of Agriculture have an important driver of soil loss, yet tillage erosion is not simulated in assigned erosion classes to soils based on the percentage of the models used to assess nationwide soil loss trends in the United original A horizon that has been eroded (28). Because the A

States. We estimate that A-horizon loss decreases crop yields by horizon has the largest fraction of soil organic carbon within the ENVIRONMENTAL SCIENCES 6 ± 2%, causing $2.8 ± $0.9 billion in annual economic losses. Re- soil profile, it is a key component of water and nutrient retention gionally, we estimate 1.4 ± 0.5 Pg of carbon have been removed and soil productivity (29). Soils where 100% of the A-horizon from hillslopes by erosion of the A-horizon, much of which likely thickness has been removed are designated as Class 4 eroded remains buried in depositional areas within the fields. soils, and other classes represent lesser reductions in A-horizon thickness (<25%, 25 to 75%, and >75% for Classes 1, 2, and 3, soil erosion | remote sensing | soil organic carbon | agricultural productivity respectively). A major disadvantage of the use of erosion classes is that properly assigning classes based on the percentage of roductive agricultural soils are vital for producing food for a A-horizon loss requires accurate determination of the original Pgrowing global population (1–3). However, degradation of A-horizon thickness on all topographic positions (30). Hence, soil quality by erosion reduces crop yields, which can result in although soil erosion classes indicate soil degradation is wide- food insecurity, conflict (3), and the decline of civilizations (4). spread (23) we do not have a robust, quantitative understanding Degradation of soils leads not only to economic losses for farmers of how much soil has already been lost. but also a loss in ecosystem services (5), which alters the ability of soils to regulate hydrologic and biogeochemical cycles. Wide- Significance spread use of synthetic fertilizers to enhance the function of degraded soils increases food production costs (6) and impairs water Conventional agricultural practices erode carbon-rich soils that resources (7), which negatively impacts human health (8) and are the foundation of agriculture. However, the magnitude of aquatic ecosystems (9). A-horizon soil loss across agricultural regions is poorly con- Globally, the reservoir of carbon stored in soils is three times strained, hindering the ability to assess soil degradation. Using that in the atmosphere (10) and given the extent of agricultural a remote-sensing method for quantifying the absence of land use (11), understanding soil carbon dynamics in agricultural A-horizon soils and the relationship between soil loss and to- systems is critical to understanding the carbon cycle (12). Whether pography, we find that A-horizon soil has been eroded from soil erosion constitutes a net carbon sink or source depends on roughly one-third of the midwestern US Corn Belt, whereas both the depositional fate of the eroded carbon and the ability to prior estimates indicated none of the Corn Belt region has lost – replace carbon in degraded soils (13 15). If biological productivity A-horizon soils. The loss of A-horizon soil has removed 1.4 ± replaces eroded carbon, and decomposition of carbon stored in 0.5 Pg of carbon from hillslopes, reducing crop yields in the sedimentary deposits is halted or slowed, then soil erosion is a net study area by ∼6% and resulting in $2.8 ± $0.9 billion in annual sink of atmospheric carbon (14–17). However, if eroded carbon economic losses. rapidly decomposes and is not replaced in eroded soil horizons, then soil erosion constitutes a carbon source. Restoring carbon to Author contributions: E.A.T., I.J.L., and Q.Y. designed research; E.A.T. performed research; degraded soils therefore has potential to both reestablish soil E.A.T. analyzed data; and E.A.T., I.J.L., and Q.Y. wrote the paper. The authors declare no competing interest. function and sequester atmospheric CO2 (10). However, quantifying the impacts of soil degradation on agricultural productivity This article is a PNAS Direct Submission. and the carbon cycle first requires robust estimates of the mag- Published under the PNAS license. nitude of agriculturally induced soil loss (14, 16). 1To whom correspondence may be addressed. Email: [email protected]. Although thousands of soil erosion measurements have been This article contains supporting information online at https://www.pnas.org/lookup/suppl/ made globally (18), the lack of a robust and scalable method for doi:10.1073/pnas.1922375118/-/DCSupplemental. estimating the magnitude of erosion in agricultural landscapes Published February 15, 2021.

PNAS 2021 Vol. 118 No. 8 e1922375118 https://doi.org/10.1073/pnas.1922375118 | 1of8 Downloaded by guest on September 27, 2021 Here we present results from a remote sensing method used to thick A-horizon soils (SI Appendix, Fig. S1 and Table S1). In the estimate the spatial extent of agriculturally induced loss of decades following European settlement, the prairie was plowed, A-horizon soil for a major global agricultural region, the Corn Belt and the landscape was rapidly and extensively converted to row- of the midwestern United States. Rather than simulate or measure crop agriculture. For example, in Iowa, Indiana, and Illinois, less short-term soil loss rates, we combine measurements of soil- than 0.1% of the original tallgrass prairie remains (32). The fertile surface reflectance in the visible spectrum (soil color) with high- soils and temperate climate make the Corn Belt one of the world’s resolution satellite imagery to directly measure the proportion of most agriculturally productive regions. The United States is the the agriculturally cultivated landscape that has completely lost its world’s largest producer of corn and soybeans (34), and 75% of original A horizon. Combining our spectral analysis with rela- the corn and 60% of the soybeans produced in the United States tionships between A-horizon soil loss and topography derived are grown in the Corn Belt (35). from high-resolution LiDAR topographic data allows us to predict Despite the importance of the region’s agricultural produc- A-horizon soil loss in areas where images are not available. We tivity, model predictions indicate the Corn Belt currently has the find that historical soil erosion has completely removed A-horizon highest soil erosion rates in the United States (24). The historical soil from approximately one-third of the Corn Belt. The spatial magnitude of A-horizon soil loss from the initiation of agricul- patterns of soil loss suggest that key erosion mechanisms are not ture to the present is unknown, but prior work in Iowa and simulated in nationwide assessments of soil erosion trends in the Minnesota noted that in some areas the magnitude of soil ero- United States and that soil survey data greatly underestimate the sion has been great enough to completely remove dark, carbon- extent of A-horizon loss. rich A-horizon soil, exposing light-colored B-horizon soil that is poor in organic carbon (36, 37). The Corn Belt Region of the Midwestern United States Our study focuses on the midwestern United States, on a A-Horizon Soil Loss in Individual Fields ∼ 2 390,000-km region that encompasses much of the area collo- We combined high-resolution satellite imagery, a newly devel- quially known as the Corn Belt (Fig. 1). The region was glaciated oped soil organic carbon index (38), and soil spectral data from repeatedly during the Pleistocene, with the exception of the the US Department of Agriculture’s Rapid Carbon Assessment Driftless Area (31). The most recent ice sheet advance deposited to develop a logistic regression to differentiate between A- and glacial till in the northern part of the Corn Belt, whereas older B-horizon soils exposed in plowed fields. The extent of historical glacial deposits to the south are mantled with loess (31). Prior to soil loss was measured by the absence of the A horizon, or in- European settlement in the mid to late 19th century, the vegeta- versely by the presence of B-horizon soils, which underlie the A tion was primarily tallgrass prairie with some savanna and wood- horizon. An example field where we applied our method is shown lands (32), and mollisols are the dominant soil order in the region in Fig. 2, where pixels of A- and B-horizon soil are distinguished (33). The native prairie vegetation fostered the accumulation of using the soil organic carbon index (Fig. 2A), and topographic curvature is calculated for each pixel in the field (Fig. 2B). Within the field, A-horizon soil has been completely removed from 34 ± 7% (±1 SD) of the area (Fig. 2C). The fraction of pixels classified as B-horizon soil is highest on convex slopes, indicating topography exerts a strong influence on the spatial distribution of A-horizon loss within the field (Fig. 2D). In 210 km2 of agricultural fields across 28 locations, the soil organic carbon index values indicate the mean extent of the agricultural land area with complete A-horizon loss is 34 ± 7% (Fig. 3A). These A-horizon soil loss values are minimum estimates because of the potential for soil moisture to cause mis- classification of soil horizons (SI Appendix). At all 28 sites, hillslope topography strongly controls the location of soil loss. B-horizon exposure occurs disproportionately on convex topography, which we quantitatively define as areas with topographic − curvature values <0m 1. Such areas make up only 50% of the area of the fields we analyzed but are the site of 68 ± 9% of the exposed B-horizon soil. On hillslopes with the most convex to- − pography (curvature <−0.02 m 1), 74 ± 8% of the land area has soil organic carbon index values indicative of exposed B-horizon soils. The proportion of the cultivated landscape with complete A-horizon loss decreases to 23 ± 5% for straight slopes (curvature = 0), and 39 ± 8% of the land area on concave topography − (curvature > 0m1) bears the spectral signature of B-horizon soils (Fig. 3B). The most convex and concave portions of the cultivated landscape tend to have the greatest proportion of soil loss, due to an association of steeper slopes (3.5 to 4.1°) with highly convex and concave topography (Fig. 3B). However, most of the cumulative A-horizon loss occurs on more modestly sloping topography; Fig. 1. Study area in the midwestern United States. The study area is defined 84 ± 2% of the B-horizon exposure occurs in areas with curvature − − by the spatial extent of LiDAR topographic data (gray hillshade). The squares values between −0.02 m 1 and 0.02 m 1, where the mean slope is show the locations of 28 sites where 210 km2 of imagery from plowed agricultural fields with bare soil was analyzed. The open square indicates the lo- 2.2°. These results indicate hillslope summit and shoulder po- cation of the field shown in Fig. 2. The glacial extent prior to the last glacial sitions are prominent locations of soil loss. The implications of maximum (LGM) and the extent during the LGM are shown as dashed and the topographic distribution of soil loss for inferring soil ero- solid lines, respectively. sion mechanisms are discussed below.

2of8 | PNAS Thaler et al. https://doi.org/10.1073/pnas.1922375118 The extent of soil loss across the US Corn Belt Downloaded by guest on September 27, 2021 A B

Curvature (m-1) -0.045

0.05

100 m ENVIRONMENTAL SCIENCES C D

Fig. 2. Calculation of A-horizon loss and topographic curvature. (A) True-color image of a field near Clear Lake, IA (open square in Fig. 1); black polygons surround light-colored areas predicted to have exposed B-horizon soil. Gray polygons indicate ±1 SD of the predicted area of B-horizon exposure. (B) To- pographic curvature map where red pixels denote negative curvature (convex hillslopes) and blue pixels denote positive curvature (concave hillslopes). (C) Cumulative fraction of area with curvature values for the entire field (black line) and curvature values for the area with exposed B-horizon soil (gray line); 34 ± 7% of the area is exposed B horizon soil. Gray shading shows the ±1 SD prediction of B-horizon exposure. (D) Mean (±1 SD) of the fraction of pixels with a spectral signature of B-horizon soils versus topographic curvature, shaded by median topographic slope per curvature bin, which shows that exposureof B-horizon soil occurs predominantly on convex topography. (Inset) The probability density of topographic slope values for A-horizon pixels and for B-horizon pixels; gray shading shows the ±1 SD prediction of B-horizon exposure. Image credit: ©2013, DigitalGlobe; NextView License/Maxar, Inc.

The Region-Wide Extent of A-Horizon Soil Loss landscape (∼273,000 km2), and 68% of the area predicted to no 2 Based on the proportion of B-horizon exposure for a given value longer have A horizon (∼90,000 ± 33,000 km ) occurs on that of topographic curvature for the 210 km2 of fields with available convex topography, whereas the remaining 32% (∼42,000 ± satellite imagery (Fig. 3B)and3.9× 105 km2 of topographic 15,000 km2) occurs on concave topography. curvature data that span the region, we estimate that A-horizon The proportion of the study area within each Corn Belt state soil has been completely removed from 35 ± 11%, or 132,738 ± predicted to have exposed B-horizon soil ranges from 30 ± 10 to 46,849 km2, of the Corn Belt (Fig. 3C). Within the Corn Belt, 41 ± 13% (Fig. 4A), and county-level estimates range from 24 ± convex topography occupies roughly 70% of the cultivated 8to47± 14% (Fig. 4B). Glacial history influences the extent of

Thaler et al. PNAS | 3of8 The extent of soil loss across the US Corn Belt https://doi.org/10.1073/pnas.1922375118 Downloaded by guest on September 27, 2021 A soil loss, with greater loss predicted for older, now loess-covered glaciated areas where drainage networks and associated ridge and valley systems are more developed. The extent of A-horizon loss is lower in areas that were covered with ice during the last glaciation because drainage networks are more poorly developed and ridge–valley systems are less established (39). Our method predicts the land area with complete A-horizon loss, rather than soil erosion rates, so our results cannot be directly compared against regional soil erosion rates modeled using the Revised Universal Soil Loss Equation and the Wind Erosion Equation (24). However, the Class-4 erosion class cat- egory from soil surveys (28) is equivalent to the complete loss of A horizon measured by our analysis. US Department of Agri- culture soil survey data indicate that none (0%) of our study area has soils with Class-4 erosion or complete A-horizon loss (SI Appendix, Fig. S2) (28). However, we predict the A-horizon has been completely removed from 35 ± 11% of the cultivated area of the Corn Belt. Hence, our results suggest that prior assessment of soil degradation based on erosion classes may have B greatly underestimated the extent of A-horizon loss, and therefore the thickness or mass of soil that has been eroded from hillslopes in the Corn Belt. Soil Loss Mechanisms Our results indicate that A-horizon soil has been stripped from hilltops and hillslopes. Although our remote sensing method cannot detect soil deposition, prior work indicates that much of the eroded soil has accumulated in topographic concavities (40, 41), though some is ultimately transported out of fields by water erosion (42). Hence, topographic concavities tend to have thicker A horizons, higher soil organic carbon concentrations, and higher crop yields than eroded hilltop summits and shoulder positions (41). Although water erosion contributes to soil loss throughout the cultivated landscape, our observation of widespread loss of soil from low-gradient, convex hilltops and deposition in topographic concavities at the base of hillslopes suggests tillage erosion is also an important driver of soil transport in the Corn Belt. Tillage erosion is the net downslope movement of soil by re- C peated tillage operations, such as plowing (43). Soil transport by tillage causes diffusion-like evolution of topography, resulting in erosion of soil from topographic convexities and deposition in concavities (40). The effect of tillage on soil transport can be described with a diffusion-like coefficient that integrates tillage direction, depth, and soil physical properties (44), and measured − diffusion coefficients for tillage range from 0.03 to 0.52 m2·y 1 (44). Although contour plowing is a common strategy to mitigate soil erosion by water, measured diffusion coefficients for contour − plowing still range from 0.03 to 0.2 m2·y 1 (44). Such values are one to three orders of magnitude greater than diffusion coefficients measured in nonagricultural settings (45). Hence, order-of-magnitude increases in topographic diffusion due to plow- and tillage-based agriculture provides a mechanistic explanation for the extensive loss of soil from hilltops, especially where the lack of upslope flow accumulation area limits the potential for water erosion by overland flow. The observation that the fraction of a landscape without A-horizon soil increases with increasing topographic Fig. 3. Regional-scale A-horizon loss estimate for the Corn Belt. (A) Cu- convexity is also consistent with a diffusive, tillage erosion soil mulative distribution of topographic curvature for all (combined A- and transport process. B-horizon) soils (black line) and B-horizon soils (gray line) for the 210 km2 of fields that were analyzed, which indicates 34 ± 7% of the soil exposed in fields is B-horizon; gray shading shows the ±1 SD prediction of B-horizon exposure. (B) Mean (±1 SD) of the fraction of pixels with a spectral signature pixels and for B-horizon pixels; gray shading shows the ±1 SD prediction of of B-horizon soils versus topographic curvature for the 210 km2 of fields that B-horizon exposure. (C) Cumulative distribution of area as a function of were analyzed. Across all sites, convex topography (negative curvature) has topographic curvature for the entire Corn Belt study area (solid line) and experienced the highest proportion of B-horizon exposure and A-horizon for B-horizon soils (dashed line); gray shading shows the ±1 SD prediction of loss. Symbol color denotes the median slope angle for each curvature value. B-horizon exposure. Our calculation indicates 35 ± 11% of the landscape has (Inset) The probability density of topographic slope values for A-horizon completely lost A-horizon soil to erosion, exposing underlying B-horizon soil.

4of8 | PNAS Thaler et al. https://doi.org/10.1073/pnas.1922375118 The extent of soil loss across the US Corn Belt Downloaded by guest on September 27, 2021 AB

CD ENVIRONMENTAL SCIENCES

Fig. 4. A-horizon loss and associated economic losses within the Corn Belt. (A) Aggregated percent A-horizon loss (black text) and annual economic losses expressed in millions of dollars (M) and as a percentage relative to uneroded soils (green text) for each state within the region (color scale is the sameasinB). (B) Percentage of A-horizon loss for each county. (C) Percent annual economic losses for each county relative to predictions for uneroded soils. (D) Predicted mean annual economic losses per farm in each county. The LGM and pre-LGM glacial extent, which define the distribution of glacial till- and loess-derived soils, are shown as solid and dashed lines, respectively.

Prior work has shown that whereas water and tillage both than 15% of the acreage of the upper Mississippi River water- contribute to soil transport, erosion on upland convex hilltops is shed, the heart of the Corn Belt, is farmed with no-till practices dominated by tillage, and erosion by water tends to be dominant in for at least three consecutive years (47). Similarly, nationwide, areas with steep, concave slopes (46). Our finding that B-horizon only 21% of corn, soybean, cotton, and wheat fields are contin- exposure increases with slope for concave topography is consistent uously farmed with no- or strip-till practices (48). Hence, wide- with previous work indicating that water erosion is dominant in spread adoption of no-till farming methods offers a strategy for such landscape positions. About 30% of the observed A-horizon preventing further soil loss. loss has occurred on concave topography where water erosion is expected to dominate soil loss. The remaining ∼70% of landscape Economic and Soil Organic Carbon Losses due to Erosion of with B-horizon exposure occurs on convex hilltops and slopes, A-Horizon Soil where tillage is expected to be a more important mechanism of Using county-level harvest data for corn and soybeans, and crop soil loss than erosion by water (46). Although it has been dem- yield reductions associated with severely eroded soils (37), we onstrated that models that do not simulate tillage erosion underestimate that loss of A-horizon soils decreases region-wide crop predict the total magnitude of soil loss (43), tillage erosion is not yields by 6 ± 2%, leading to $2.8 ± $0.9 billion in annual losses incorporated into nationwide assessments of soil erosion in the across the Corn Belt. Mean annual crop yield decreases, relative United States (24). Our analysis does not discount the need to to yields from undegraded soils, for each state in the region quantify and model soil erosion by water but highlights that tillage range from 3 ± 1% to 8 ± 3%, resulting in annual losses of $49 ± erosion is an important contributor to widespread removal of soil $16 million to $793 ± $262 million. The mean crop yield re- from hilltops in the Corn Belt that warrants greater recognition ductions for each county range from 2 ± 1% to 9 ± 3%, equating in soil erosion prediction and soil conservation efforts in the to annual economic losses of $0.1 ± 0.04 million to $32 ± $11 United States. million (Fig. 4C). The average county-level yield reductions per Adoption of no-till agriculture greatly reduces soil erosion farm range from 2 ± 1% to 9 ± 3%, leading to losses of $300 ± rates (2) and effectively eliminates tillage erosion. However, less $100 to $40,000 ± $14,000 (Fig. 4D), and varies as a function of

Thaler et al. PNAS | 5of8 The extent of soil loss across the US Corn Belt https://doi.org/10.1073/pnas.1922375118 Downloaded by guest on September 27, 2021 regional differences in soil parent material, crop yields, and the reflectance was measured. Using the laboratory hyperspectral reflectance average farm size per county, which ranges from 10 to 718 ha. measurements, we calculated the SOCI at 478 nm (blue), 546 nm (green), and Because our analysis only identifies areas where the A horizon 659 nm (red) for each of the RaCA samples. The RaCA-derived SOCI values has been completely removed and not areas where the A horizon are offset from the satellite-derived SOCI values due to atmospheric effects and imperfect radiometric calibration. Hence, the RaCA SOCI values were has been thinned, which also reduces crop yields (37), our esti- scaled to the same range as based on a regression relationship between the mates of economic losses are minimum values. Fertilizer is satellite-derived and laboratory-measured SOCI values (38). widely applied to degraded soils in the Corn Belt, though it does We evaluated the extent of A- and B-horizon exposure in each image by not restore crop yields to levels measured in noneroded soils calculating the probability that a pixel has a SOCI signature of B-horizon soil. (29). Our analysis of economic losses does not account for the For each study site, we performed a bootstrapped logistic regression with cost of fertilizer inputs required to raise crop yields in degraded 500 iterations using the sci-kit-learn module in Python 3.6 to determine the soils, but others have indicated overfertilization of low-yielding probability that a pixel from a satellite image with a given SOCI value has a areas in the Midwest alone costs nearly $0.5 billion a year (49). B-horizon spectral signature. The soil horizon and SOCI data used in the site- Global-scale assessments of the influence of soil erosion on specific logistic regressions were from samples in the RaCA database that the carbon cycle are commonly based on modeled predictions of were collected within a 50-km radius of individual study sites. We trained the logistic regression model using 20% of the SOCI values for the A- and the extent of soil degradation, which have considerable uncer- B-horizon soils. We then tested the power of the logistic regression by using tainty (50). Our remote sensing method provides a means for the derived probability function to predict the soil horizon for the remaining quantifying the land area of degraded soil. Our method relies on 80% of soil samples based on the SOCI value of a sample. To quantify the the strong color contrast between the A and B horizons of error in the predicted B-horizon exposure probability values, we generated a mollisols but in principle can be applied to other agricultural probability density function of B-horizon exposure using results from each regions with spectrally distinct soil horizons. In the Corn Belt, we of the 500 iterations of the logistic regression fit, each based on a different estimate that the regional extent of A-horizon erosion has re- random selection of the 20% of the data used for training (SI Appendix, Fig. − moved 1.4 ± 0.5 Pg C (10 ± 2GgC·km 2) from hilltops and S7). We further assessed the validity of the logistic regression model by hillslopes. Within fields (40, 41, 51) and fluvial systems (52, 53) plotting the receiver operator characteristic curve and calculating the area in the Corn Belt there is evidence of widespread storage of the under the curve (AUC) for each site. An AUC value of 0.5 indicates the logistic regression cannot distinguish between classes, and an AUC value of 1.0 soils that have eroded since European settlement. Due to the indicates the regression perfectly distinguishes between classes. The AUC greater land area, carbon preservation potential is higher in the values for the classification performed in this study range from 0.52 ± 0.15 to hummocky topography of the recently glaciated portion of the 0.96 ± 0.04, and the mean true-positive classification of samples is 88 ± 8%, Corn Belt, where we estimate erosion of A-horizon soils has indicating a high true-positive classification of soil horizon based on the removed 0.80 ± 0.3 Pg C, whereas 0.6 ± 0.2 Pg C is predicted to SOCI values (SI Appendix, Fig. S8). The probability functions derived from the have been lost from the area glaciated earlier in the Pleistocene. bootstrapped logistic regressions were applied to the SOCI raster calculated Burial of these carbon-rich sediments can act as a carbon sink on for each image, and the mean percentage ±1 SD of exposed B-horizon was ≥ timescales of decades to centuries (13), and because the initia- calculated from the fraction of pixels with 50% probability of classification tion of soil erosion in midwestern United States was relatively as B horizon. recent there is a high potential for the landscape to preserve Effect of Moisture on the SOCI and Soil Horizon Classification. Soil moisture carbon in young sedimentary deposits (54). Hence, changes in causes soils to appear darker and can obscure the spectral signature of soil land use (55, 56) or adoption of farming practices (57, 58) that organic carbon (62). We performed two analyses to assess the potential increase soil organic carbon concentrations in areas that have lost impact of soil moisture on our estimation of A-horizon soil loss. We evalu- A-horizon soils may generate a net sink for atmospheric CO2. ated whether soil moisture influenced the spectral reflectance of soils in the images that we used, then we estimated the potential magnitude of a soil Methods moisture impact on our analyses. Study Area and Topographic Data. The availability of LiDAR-derived digital The soil surface, which is imaged by satellites, has been shown to become elevation models (DEMs) dictated the specific extent of the area of the completely dry after 3 to 4 d after rainfall (63). To assess the surface soil glaciated, former tallgrass prairie region we analyzed. The LiDAR data were moisture condition when the images we used were acquired, we analyzed clipped to the Herbaceous Agricultural Vegetation layer from the US Geo- precipitation data from National Oceanic and Atmospheric Administration logical Survey Gap Analysis Program (59), so that the analysis excluded areas weather stations nearest to each of our 28 sites. We found that the mini- with nonagricultural land use. Topographic slope and curvature were cal- mum time between image acquisition and prior rainfall was 20 h, with a culated as the first and second derivatives of elevation, respectively, using a mean of 73 h, and the mean magnitude of precipitation for all events was 4-m-resolution DEM (SI Appendix, Fig. S3). Details of the topographic anal- 10 mm (SI Appendix, Fig. S9). These results suggest that the soil surface ysis are described in SI Appendix. would have been dry when the images were acquired. A more detailed description of the precipitation analysis is in SI Appendix. Differentiating between A- and B-Horizon Soils Using Satellite Imagery. We To determine the influence of soil moisture on the SOCI and the potential used the National Geospatial-Intelligence Agency catalog (60) to identify impact on the differentiation of soil horizons from spectral data we con- 28 GeoEye-1, Quickbird-2, WorldView-2, and WorldView-3 satellite images ducted a laboratory experiment to measure the spectral reflectance of 26 soil showing plowed fields with exposed bare soil (SI Appendix, Fig. S4). We samples with a range of soil organic carbon concentrations collected from the analyzed 759 individual cropland fields with a total area of 210 km2.An Corn Belt. The reflectance was measured when the samples were dry and example of a study site, which is made up of multiple fields, is shown in SI again when the samples were saturated with moisture. When moisture is Appendix, Fig. S5. The spatial distribution of those fields within the study added to the soils, the largest increase in the SOCI occurs for samples with the area was primarily dictated by the availability of imagery with plowed fields, greatest soil organic carbon concentration (SI Appendix,Fig.S10). We used as soil horizons could not be distinguished in fields with no-till or conser- the relationship between soil organic carbon and the maximum change in vation tillage practices that left organic carbon-bearing crop residue ex- SOCI due to moisture saturation to simulate the effect of soil moisture on posed at the ground surface. Details of the image preprocessing are the threshold distinguishing A- and B-horizon SOCI values. We scaled the described in SI Appendix. For each of the 28 images, we calculated the soil RaCA SOCI values by the percent change in the SOCI due to moisture satu-

organic carbon index [SOCI = ρBlue/(ρGreen·ρRed)], where ρ is spectral reflec- ration. For each of the 28 sites, our analysis indicates that any addition of tance in the blue, green, and red bands, respectively (38). We demonstrated moisture to the RaCA samples increases the threshold SOCI value that dis- the validity of the SOCI by examining the relationship between the SOCI and tinguishes between A and B horizons (SI Appendix, Fig. S11). When the measured soil organic carbon values at five sites (four within the study area), threshold that accounts for SOCI changes due to addition of soil moisture is and the R2 for the correlation ranges from 0.63 to 0.72 (SI Appendix, Fig. S6). applied to the image where the SOCI values have not been adjusted for The Rapid Carbon Assessment (RaCA), undertaken by the Soil Science moisture, a higher fraction of pixels are classified as B horizon, relative to Division of the US Department of Agriculture National Resource Conservation the threshold based on dry calibration samples (SI Appendix,Fig.S12). Be- Service, collected soil samples at 6,148 sites in the conterminous United States cause the simulated effect of soil moisture consistently results in an increase (61). For each sample, the soil horizon was designated and hyperspectral in the fraction the pixels predicted to have B-horizon soils, our estimate of

6of8 | PNAS Thaler et al. https://doi.org/10.1073/pnas.1922375118 The extent of soil loss across the US Corn Belt Downloaded by guest on September 27, 2021 A-horizon loss, which is based on dry calibration samples, is a minimum. A soybean yields due to complete A-horizon loss for each parent material, pre- detailed description of the moisture experiment and sensitivity analysis is dicted areas of A-horizon loss (mean loss ±1 SD), hectares planted of corn and given in SI Appendix. soybeans (SI Appendix,Fig.S16), and the average corn and soybean prices from 2012 to 2017, we estimate mean economic losses for each county. The Upscaling Soil Loss Estimates to the Corn Belt Area Using Topographic uncertainties we report are based on the 1 SD uncertainties in the percentage Curvature. Because high-resolution satellite imagery for plowed fields in of A-horizon soil loss for each county. We estimate total loss with the glacial the Midwest is limited by both spatial coverage and by seasonal crop, snow, till-derived soils to be $1,683 ± $575 million and losses from loess-derived soils and cloud cover, region-wide estimates of B-horizon exposure and to be $762 ± $246 million. The mean per-farm economic loss for glacial till- A-horizon loss based solely on high-resolution satellite imagery is not pos- derived soils is $15,200 ± $5,200 and is $5,900 ± $1,900 for loess-derived soils. sible. However, analysis at four sites throughout the Corn Belt (SI Appendix, Figs. S13 and S14) indicates that satellite-derived SOCI is related to topo- Estimation of Soil Organic Carbon Erosion. To estimate the magnitude of soil graphic curvature, where pixels with low SOCI values are observed on to- organic carbon erosion, we multiplied the mean value of carbon stocks (10.5 pographic convexities and high SOCI values are located in topographic −2 billion g C·km ) in the upper 30 cm of soil samples collected on native concavities (SI Appendix, Fig. S14). Hence, we use the relationship between 2 prairie hillslopes (50) from locations with convex topography (curvature B-horizon exposure and topographic curvature from the 210 km of ana- − − values between −0.01 1 and −0.09 m 1) by the area of A-horizon loss. We lyzed fields to upscale our estimate of soil loss to the entire Corn Belt region. also compiled published measurements of A-horizon thickness for native We extracted the SOCI and topographic curvature values from colocated tallgrass prairies within our study area (SI Appendix, Table S1). The mean pixels within each of the 28 study sites. Using data from all 210 km2 of fields, A-horizon thickness was 37 cm, indicating a carbon stock value measured to we calculated the fraction of area with SOCI values diagnostic of exposed B-horizon soil. We treated each study site (each made up of 6 to 109 fields) 30 cm depth reasonably approximates the A-horizon carbon stock. The loss as an individual measurement and calculated the mean and one SD of of 30 cm of soil is also consistent with our spectral measurements of soils on B-horizon exposure as a function of curvature for the 28 sites. The rela- convex hilltops in a prairie and adjacent field in Iowa (SI Appendix, Fig. S1). tionship between soil loss and curvature (Fig. 3B), which includes uncertainty in B-horizon pixel classification from the bootstrapped logistic regression, Data Availability. Data are cataloged at the Oak Ridge National Laboratory was used to calculate the mean and SD of the land area with curvature values Distributed Active Archive Center (https://daac.ornl.gov/cgi-bin/dsviewer.pl? indicative of B-horizon soil exposure. Pixels classified as B-horizon soil are ds_id=1774) (65). The archive includes spatial raster data for topographic − disproportionately located where curvature values are <−0.02 m 1.Compared metrics (elevation, slope, and curvature), soil organic carbon index values, to the analyzed fields, the full Corn Belt study area has a slightly larger fraction and the probability of B-horizon soil. Spatial vector data and tabular data − − of curvature values between −0.02 m 1 and 0.02 m 1 (SI Appendix,Fig.S15), with county-, state-, and farm-level erosion and economic loss values are also where most erosion is predicted to occur (Fig. 3A). Hence, the estimated archived. The soil organic carbon index values derived from the RaCA sam- percentage of the Corn Belt with B-horizon exposure is slightly higher than the ples and used to develop the logistic regression and receiver operator

B-horizonexposuredeterminedfortheanalyzedfields. characteristic curve for each site (such as those shown in SI Appendix, Fig. S8) ENVIRONMENTAL SCIENCES are also archived. Calculation of Economic Losses. Our estimate of the magnitude of annual economic losses incurred from the loss of the A-horizon relies on the simi- ACKNOWLEDGMENTS. We thank Elizabeth Hoy and Jaime Nickeson of larity between yield reductions reported for corn and soybeans in severely NASA and Paul Morin of the Polar Geospatial Center for assistance accessing eroded soils (64). Corn yields were previously evaluated at 569 sites in 44 images. DigitalGlobe/Maxar data were provided by the Commercial Archive counties in Iowa on both glacial till and loess soil parent materials, where Data for NASA investigators (https://cad4nasa.gsfc.nasa.gov) under the Na- categorical measurements of soil erosion were also evaluated (37). In se- tional Geospatial-Intelligence Agency’s NextView license agreement. Polar verely eroded soils, where the A-horizon was completely removed, corn Geospatial Center support was from NSF Office of Polar Programs awards yields decreased by 137,300 kg·km−2 in soils derived from glacial till, whereas 1043681 and 1559691. We also thank Skye Wills of the US Department of − yields decreased by 67,100 kg·km 2 in loess-derived soils. We estimated the Agriculture for assistance with access to the Rapid Carbon Assessment database and for sharing soil sample data; Wells Hively and Xia Li for sharing soil total area of loess-derived soils by assuming that all soils south of the last sample data; Rebecca McCulley and Jonathan Sanderman for sharing the glacial maximum (LGM) ice limit are derived from loess, and the areas north native prairie soil carbon databases; Yong Tian for providing a spectroradi- of the LGM ice limit have soils formed from glacial till; soils in the Driftless Area 2 ometer; Jeffrey Kwang, Brendan Quirk, and Caroline Lauth for assistance were classified as loess-derived. We calculate that there are 235,632 km and collecting field samples; Paul Willis and Todd and Jane Gruis for permission 2 154,775 km of glacial till- and loess-derived soils in our study area, respec- to collect soil samples; Oliver Korup for suggesting statistical approaches; Da- tively. The area of till-derived soils that no longer has A-horizon is estimated to vid Montgomery for comments on a prior draft; and two reviewers for con- be 80,114 ± 28,275 km2 and the area of loess-derived soils with no A-horizon is structive feedback. The research was supported by NASA (80NSSC18K0747 estimated to be 52,623 ± 18,573 km2. Based on the reductions in corn and P0004) and NSF (1653191) grants to I.J.L.

1. D. Tilman, K. G. Cassman, P. A. Matson, R. Naylor, S. Polasky, Agricultural sustain- 14. K. Van Oost et al., The impact of agricultural soil erosion on the global carbon cycle. ability and intensive production practices. Nature 418, 671–677 (2002). Science 318, 626–629 (2007). 2. D. R. Montgomery, Soil erosion and agricultural sustainability. Proc. Natl. Acad. Sci. 15. R. Lal, Soil erosion and carbon dynamics. Soil Tillage Res. 81, 137–142 (2005). U.S.A. 104, 13268–13272 (2007). 16. A. A. Berhe, J. Harte, J. W. Harden, M. S. Torn, The significance of the erosion-induced 3. R. Amundson et al., Soil science. Soil and human security in the 21st century. Science terrestrial carbon sink. Bioscience 57, 337–346 (2007). 348, 1261071 (2015). 17. Z. Wang et al., Human-induced erosion has offset one-third of carbon emissions from 4. D. R. Montgomery, Dirt: The Erosion of Civilizations (University of California Press, land cover change. Nat. Clim. Chang. 7, 345–349 (2017). Berkeley, CA, 2007). 18. J. M. García-Ruiz et al., A meta-analysis of soil erosion rates across the world. Geo- 5. G. C. Daily, P. A. Matson, P. M. Vitousek, Nature Services: Societal Dependence on morphology 239, 160–173 (2015). Natural Ecosystems (Island, Washington, DC, 1997). 19. J. Poesen, Soil erosion in the Anthropocene: Research needs. Earth Surf. Process. 6. D. Pimentel et al., Environmental and economic costs of soil erosion and conservation Landf. 43,64–84 (2018). benefits. Science 267, 1117–1123 (1995). 20. M. A. Nearing, G. R. Foster, L. J. Lane, S. C. Finkner, A process-based soil erosion model 7. K. E. Schilling, J. Spooner, Effects of watershed-scale land use change on stream ni- for USDA-water erosion prediction project technology. Trans. ASAE 32, 1587–1593 trate concentrations. J. Environ. Qual. 35, 2132–2145 (2006). (1989). 8. P. J. Weyer et al., Municipal drinking water nitrate level and cancer risk in older 21. K. G. Renard, G. R. Foster, G. A. Weesies, D. K. McCool, D. C. Yoder, Predicting Soil women: The Iowa Women’s Health Study. Epidemiology 12, 327–338 (2001). Erosion by Water: A Guide to Conservation Planning with the Revised Universal Soil 9. R. E. Turner, N. N. Rabalais, Coastal eutrophication near the Mississippi River delta. Loss Equation (RUSLE) (US Department of Agriculture, Washington, DC, 1997), vol. Nature 368, 619 (1994). 703. 10. R. Lal, Soil carbon sequestration impacts on global climate change and food security. 22. R. Morgan et al., The European Soil Erosion Model (EUROSEM): A dynamic approach Science 304, 1623–1627 (2004). for predicting sediment transport from fields and small catchments. Earth Surf. Pro- 11. N. Ramankutty, J. A. Foley, Characterizing patterns of global land use: An analysis of cesses Landforms J. B. Geomorphol. Group 23, 527–544 (1998). global croplands data. Global Biogeochem. Cycles 12, 667–685 (1998). 23. N. A. Jelinski, K. Yoo, The distribution and genesis of eroded phase soils in the con- 12. L. B. Guo, R. M. Gifford, Soil carbon stocks and land use change: A meta analysis. Glob. terminous United States. Geoderma 279, 149–164 (2016). Change Biol. 8, 345–360 (2002). 24. National Resources Conservation Service, “National resources inventory, 2015 annual 13. R. F. Stallard, Terrestrial sedimentation and the carbon cycle: Coupling weathering NRI, soil erosion” (National Resources Conservation Service, US Department of Agri- and erosion to carbon burial. Global Biogeochem. Cycles 12, 231–257 (1998). culture, Washington, DC, 2018).

Thaler et al. PNAS | 7of8 The extent of soil loss across the US Corn Belt https://doi.org/10.1073/pnas.1922375118 Downloaded by guest on September 27, 2021 25. N. P. Woodruff, F. H. Siddoway, A wind erosion equation. Soil Sci. Soc. Am. J. 29, 47. J. K. Horowitz, R. M. Ebel, K. Ueda, “No-till farming is a growing practice” (Economic 602–608 (1965). Information Bulletin 70, USDA Economic Research Service, Washington, DC, 2010). 26. S. W. Trimble, P. Crosson, U.S. soil erosion rates—Myth and reality. Science 289, 48. US Department of Agriculture, “Tillage intensity and conservation cropping in the 248–250 (2000). United States” (EIB-197, USDA, Washington, DC, 2015). 27. M. A. Nearing et al., Measurements and models of soil loss rates. Science 290, 49. B. Basso, G. Shuai, J. Zhang, G. P. Robertson, Yield stability analysis reveals sources of 1300–1301 (2000). large-scale nitrogen loss from the US Midwest. Sci. Rep. 9, 5774 (2019). 28. Soil Science Division Staff, Soil Survey Manual (USDA Handbook No. 18, US Depart- 50. J. Sanderman, T. Hengl, G. J. Fiske, Soil carbon debt of 12,000 years of human land ment of Agriculture, Washington, DC, 2017). use. Proc. Natl. Acad. Sci. U.S.A. 114, 9575–9580 (2017). 29. R. Lal, Enhancing crop yields in the developing countries through restoration of the 51. J. C. Ritchie, G. W. McCarty, E. R. Venteris, T. C. Kaspar, Soil and soil organic carbon – soil organic carbon pool in agricultural lands. Land Degrad. Dev. 17, 197 209 (2005). redistribution on the landscape. Geomorphology 89, 163–171 (2007). 30. K. R. Olson, L. D. Norton, T. E. Fenton, R. Lal, Quantification of soil loss from eroded 52. J. C. Knox, Human impacts on Wisconsin stream channels. Ann. Assoc. Am. Geogr. 67, soil phases. J. Soil Water Conserv. 49, 591–596 (1994). 323–342 (1977). 31. E. A. Bettis III, D. R. Muhs, H. M. Roberts, A. G. Wintle, Last glacial loess in the con- 53. P. Kumar et al., Critical transition in critical zone of intensively managed landscapes. terminous USA. Quat. Sci. Rev. 22, 1907–1946 (2003). Anthropocene 22,10–19 (2018). 32. F. Sampson, F. Knopf, Prairie conservation in North America. Bioscience 44, 418–421 54. S. Doetterl et al., Erosion, deposition and soil carbon: A review of process-level con- (1994). trols, experimental tools and models to address C cycling in dynamic landscapes. Earth 33. Soil Survey Staff, A Basic System of Soil Classification for Making and Interpreting Soil Sci. Rev. 154, 102–122 (2016). Surveys (US Department of Agriculture Handbook 436, National Resource Conserva- 55. C. Li et al., Soil carbon sequestration potential in semi-arid grasslands in the conser- tion Service, ed. 2, 1999). – 34. Food and Agriculture Organization of the United Nations, FAOSTAT statistics data- vation reserve program. Geoderma 294,80 90 (2017). base (2017). www.fao.org/faostat/en/#home. 56. D. L. Hernández, E. H. Esch, C. J. Alster, M. J. McKone, P. Camill, Rapid accumulation of 35. US Department of Agriculture, “Crop production” (2017). https://www.nass.usda.gov/ soil carbon and nitrogen in a prairie restoration chronosequence. Soil Sci. Soc. Am. J. Publications/Todays_Reports/reports/crop1117.pdf. 77, 2029–2038 (2013). 36. M. J. Lindstrom, W. W. Nelson, T. E. Schumacher, G. D. Lemme, Soil movement by 57. R. Lal, M. Griffin, J. Apt, L. Lave, M. G. Morgan, Ecology. Managing soil carbon. Sci- tillage as affected by slope. Soil Tillage Res. 17, 255–264 (1990). ence 304, 393 (2004). 37. T. E. Fenton, M. Kazemi, M. A. Lauterbach-Barrett, Erosional impact on organic 58. D. R. Montgomery, Growing a Revolution: Bringing Our Soil Back to Life (WW Norton matter content and productivity of selected Iowa soils. Soil Tillage Res. 81, 163–171 & Company, 2017). (2005). 59. US Geological Survey Gap Analysis Program, “20160513, GAP/LANDFIRE national 38. E. A. Thaler, I. J. Larsen, Q. Yu, A new index for remote sensing of soil organic carbon terrestrial ecosystems” (US Geological Survey, 2011). https://doi.org/10.5066/ based solely on visible wavelengths. Soil Sci. Soc. Am. J. 83, 1443–1450 (2019). F7ZS2TM0. 39. J. Lai, A. M. Anders, Modeled postglacial landscape evolution at the southern margin 60. C. S. Neigh, J. G. Masek, J. E. Nickeson, High-resolution satellite data open for gov- of the Laurentide ice sheet: Hydrological connection of uplands controls the pace and ernment research. Eos 94, 121–123 (2013). style of fluvial network expansion. J. Geophys. Res. Earth Surf. 123, 967–984 (2018). 61. S. Willis et al., “Overview of the US rapid carbon assessment project: Sampling design, 40. X. Li, G. W. McCarty, D. L. Karlen, C. A. Cambardella, Topographic metric predictions initial summary and uncertainty estimates” in Soil Carbon, A. E. Hartemink, K. – of soil redistribution and organic carbon in Iowa cropland fields. Catena 160, 222 232 McSweeney, Eds. (Springer International Publishing, Cham, Switzerland, 2014), pp. (2018). 95–104. 41. S. K. Papiernik et al., Soil properties and productivity as affected by topsoil movement 62. M. Nocita, A. Stevens, C. Noon, B. van Wesemael, Prediction of soil organic carbon for within an eroded landform. Soil Tillage Res. 102,67–77 (2009). different levels of soil moisture using vis-NIR spectroscopy. Geoderma 199,37–42 42. H. Li, R. M. Cruse, R. L. Bingner, K. R. Gesch, X. Zhang, Evaluating ephemeral gully (2013). erosion impact on Zea mays L. yield and economics using AnnAGNPS. Soil Tillage Res. 63. W. J. Capehart, T. N. Carlson, Decoupling of surface and near‐surface soil water 155, 157–165 (2016). content: A remote sensing perspective. Water Resour. Res. 33, 1383–1395 (1997). 43. G. Govers, T. A. Quine, P. J. Desmet, D. E. Walling, The relative contribution of soil 64. G. W. Langdale, W. D. Shrader, “Soil erosion effects on soil productivity of cultivated tillage and overland flow erosion to soil redistribution on agricultural land. Earth ” Surf. Process. Landf. 21, 929–946 (1996). cropland 1 in Determinants of Soil Loss Tolerance, B. L. Schmidt, R. R. Allmaras, J. V. 44. K. Van Oost, G. Govers, S. De Alba, T. A. Quine, Tillage erosion: A review of con- Mannering, R. I. Papendick, Eds. (ASA Special Publication 45, American Society of trolling factors and implications for soil quality. Prog. Phys. Geogr. 30, 443–466 (2006). Agronomy, 1982), pp. 41–51. 45. N. F. Fernandes, W. E. Dietrich, Hillslope evolution by diffusive processes: The time- 65. E. A. Thaler, I. J. Larsen, Q. Yu, Remote sensing derived topsoil and agricultural scale for equilibrium adjustments. Water Resour. Res. 33, 1307–1318 (1997). economic losses, midwestern USA. ORNL DAAC (Oak Ridge National Laboratory Dis- 46. J. A. Schumacher et al., Identifying spatial patterns of erosion for use in precision tributed Active Archive Center). https://daac.ornl.gov/cgi-bin/dsviewer.pl?ds_id=1774. conservation. J. Soil Water Conserv. 60, 355–362 (2005). Deposited 19 December 2019.

8of8 | PNAS Thaler et al. https://doi.org/10.1073/pnas.1922375118 The extent of soil loss across the US Corn Belt Downloaded by guest on September 27, 2021