Grassland-to-cropland conversion increased soil, nutrient, and carbon losses in the US Midwest between 2008 and 2016

In this study, we assessed the cumulative environmental impacts of grassland conversion occurring in the US Midwest between 2008 and 2016. Specifically, we chose 12 states in the US Midwest (figure 1) as our modeling area since they contained much of the grassland converted during this period (Lark et al 2020) and >80% of the total corn and soybean acres planted in the US (USDA-NASS 2020). Moreover, this highly productive agricultural area is a hotspot of cropland C sequestration in the US (West et al 2010), while excess erosion and fertilizer loss from croplands in the region are major causes of environmental problems, such as waterway siltation (Lal 1998), and harmful algal blooms and hypoxia in the Great Lakes (NOAA 2020) and the Gulf of Mexico (Rabalais et al 2002).

We used the Geospatial Agroecosystem Modeling System (GAMS) (Zhang et al 2010, 2015) to leverage high resolution geospatial datasets (e.g. soils, terrain, land use) to run the Environmental Policy Integrated Climate (EPIC) model (Williams et al 1989, Izaurralde et al 2006) at a 30 m resolution. EPIC is a process-based agroecosystem model capable of simulating key biophysical and biogeochemical processes, such as plant growth and development, water balance, (C) and nutrient cycling, soil erosion, and greenhouse-gas emissions. Detailed description of the GAMS, the input data for EPIC, and the evaluation of the model are provided in the supplementary information (SI; SI figures 1–4 and Methods 1–4) (available online at stacks.iop.org/ERL/16/054018/mmedia).

We conducted model simulations for two purposes: (a) estimating the environmental impacts of cropland expansion and abandonment during 2008–2016, and (b) understanding the effects of varying crop types and tillage intensity. In the cropland expansion assessment, we considered the conversion of grassland (i.e. two Cropland Data Layer (CDL) land use types: 37 Other Hay/Non-Alfalfa and 176 Grass/Pasture) into cropland. The pixels allowed to be converted to corn or corn/soybean were those identified as converted to cropland in Lark et al (2020). We selected two representative crop rotations, continuous corn and corn–soybean, as converted cropping systems to simplify the total number of simulations required. These rotations were selected for multiple reasons, including: (a) their prevalence, as corn and soybeans were the most dominant crops planted on newly converted land, as indicated above; and (b) because corn–grain ethanol and soy biodiesel account for most of the biofuel volumes produced to date (EPA 2018a). For each rotation, we also simulated two types of tillage, either conventional tillage or no-till.

For cropland abandonment, we employed land use change from different crops to the Other Hay/Non-Alfalfa and Grass/Pasture land use types. Due to the complexity and lack of historical CDL data to derive pre-abandonment crop rotations, we used a dominant crop rotation (i.e. corn–soybean) with conventional tillage to represent typical crop management in the cropland abandonment assessment. The pixels allowed to be abandoned were those identified in Lark et al (2020) as moving from cropland to non-cropland.

To investigate cropland expansion effects, we executed the EPIC model for a 30-year period (1979–2008) under the grassland scenario to initialize the model for state variables, such as soil organic matter. Then, we again ran EPIC using the initialized values to simulate each of the grassland and cropland scenarios. For each scenario, we ran EPIC for a 30 year period to derive the annual, long-term average environmental impacts of the land converted between 2008 and 2016. We used historical climate data (from 1979 to 2008), so our scenarios did not incorporate future climate change effects. The environmental impacts of cropland expansion were then calculated by subtracting the baseline scenario outputs from those of each cropland-tillage scenario. To assess cropland abandonment impacts, we initialized EPIC for the 30 year period under a single cropland scenario (i.e. tilled corn–soybean), before running EPIC using these initialized values to simulate the grassland and cropland scenarios. The environmental impact from cropland abandonment was estimated as the difference between the grassland and cropping scenarios.

Using EPIC, we simulated soil erosion, N and P loss (lost from fields through erosion, runoff, and leaching) and SOC loss (lost from fields through emissions to the atmosphere and through runoff, erosion, and leaching). For each environmental variable, we calculated total values for the region and values per unit area.

Across the 12 US Midwestern states in this study, approximately 2.05 and 0.34 million ha of cropland expansion and abandonment occurred, respectively, between 2008 and 2016 according to the data from Lark et al (2020). Grasslands were the most common land cover converted to cropland, with the area of grassland-to-crop conversion in the US Midwest roughly equal to the size of New Jersey, or 3.3% of the total grassland area in the modeled region. Grasslands represented approximately 94% and 90% of total cropland expansion and abandonment area, respectively, in the US Midwest. The remaining 6% of cropland expansion and 10% of cropland abandonment were mainly from- or to- forests, wetlands, and shrublands (Lark et al 2020). Given the complexity and large uncertainty in applying process-based models to simulate other land use transitions, here we focused on the conversion between grassland and cropland, and not the other land use transitions.

The spatial patterns of grassland-to-cropland conversion generally followed the distribution of grasslands in the region (figures 2(a) and (b)). Most of the grassland is located in the central and western areas of the US Midwest, including the Dakotas, Nebraska, Kansas, Minnesota, southern Iowa and western Missouri (figure 2(a)), and most of the conversion to cropland occurred in places like southern Iowa and the Dakotas (figure 2(b)). The percent grassland converted in each county ranged from 0% to 31%. Cropland abandonment often co-located with cropland expansion in the western Midwest, but also occurred in more easterly states including Wisconsin, Michigan, and Ohio (figure 2(c)).

Zoom In Zoom Out Reset image size

Land capability classes (LCC) is a system by which lands are grouped by ability to produce crop yields (Klingebiel and Montgomery 1961). Values range between 1 and 8, with larger values indicating less productive soils for crop production. For the 12 state area simulated, the average LCC was 3.2 for the grassland areas converted to cropland, higher than the average for existing cultivated cropland, ca. 2.8 (Zhang et al 2015). Abandoned cropland had an LCC of 3.0. Thus, the most productive soils are continuously being used for crops, while land converted to- and from- cropland mainly occurred on more marginal soils (Lark et al 2015).

Our modeling results found clear negative environmental impacts of cropland expansion on a per area basis (figure 3) and in total across the region (table 1), with the magnitude of the impacts varying greatly depending on crop type and tillage. These results represent the annual, long-term average environmental impacts of the land converted between 2008 and 2016, and barring additional changes in land use or management, would be expected to continue to accrue each year into the foreseeable future.

Zoom In Zoom Out Reset image size

Tilled corn and corn–soybean exhibited the highest amount of soil erosion, and N, P, and SOC loss; the grassland baseline was the lowest for all metrics; and no-till corn and corn–soybean were intermediate (figure 3). For soil erosion, tilled corn and corn–soybean caused 4–6 times more loss than the baseline (figure 3). Conversion to no-till corn–soybean and no-till corn rotations increased erosion above the grassland baseline, but this was substantially lower than the tilled crops. These differences reflect in part that perennial grasslands provide year-round protection from soil erosion relative to corn and soybeans. Similarly, tillage mixes surface residue with upper soil layers, reducing the amount of residue protecting the soil and thus increasing susceptibility to erosion.

In general, the spatial patterns of soil erosion reflected the spatial distribution of grasslands converted (figure 2), with hotspots in the eastern Dakotas, southern Iowa, and Missouri (SI figure 5). Despite a larger area of cropland expansion in the Dakotas, however, the amount of soil erosion was higher in Iowa and Missouri. Newly converted cropland in Iowa and Missouri had a higher slope than the grasslands converted in the eastern Dakotas, likely explaining this finding (SI figure 6(a)).

Not surprisingly, grasslands experienced much lower nutrient (N and P) loss compared with the cropping systems (table 1), mainly caused by the application of synthetic fertilizer to cropland. Notably, nutrient loss from grasslands is not zero, since grassland soils still contain N and P, some of which can be lost via leaching, runoff, or erosion. Tillage generally increases nutrient loss through higher soil erosion and soil organic matter decomposition. Converting grassland to no-till corn and no-till corn–soybean increased N loss by a factor of ca. 3–6 times, while grassland conversion to tilled corn and tilled corn–soybean led to an increase by 6–10 times. Similarly, grassland to cropland conversion increased P loss, but with less magnitude, mainly because the P application rates were lower than N application rates. The spatial patterns of N and P losses largely coincided with the spatial distribution of grassland conversions to cropland (SI figures 7 and 8).

We also found that converting grassland into crop production decreased SOC stocks (table 1). Without tillage, land use conversion led to moderate SOC loss in the soil profile with an average depth of 1.6 m. With tillage, the amount of SOC loss exceeded 800 Gg C yr⁻¹ for both rotations (table 1). The hotspots of SOC loss coincided with the spatial pattern of soil erosion and nutrient loss. Despite the overall loss of SOC from newly expanded cropland, the conversion resulted in slightly higher amounts of SOC in certain areas, particularly for no-till corn and no-till corn–soybean (SI figure 9). This indicates that the SOC impacts vary from location to location, and with site specific conditions (such as climate, terrain, soil properties, and management).

All environmental impacts varied substantially across counties on per unit area basis. Figure 4 shows the spatial variability of soil erosion on expanded cropland. In general, the grassland–cropland conversion in the northern Midwest (including Dakotas, Wisconsin, Minnesota, and Michigan) resulted in less soil erosion per ha than in the southern portion (including Nebraska, Kansas, Iowa, Missouri, Illinois, Indiana, and Ohio). In addition, the spatial patterns of N, P, and SOC losses per ha (SI figures 10–12) were very similar to that of soil erosion. Soil erosion per ha was modestly correlated with slope (R² of 0.35 under the grassland scenario), suggesting this is a partial driver of the soil erosion pattern.

Zoom In Zoom Out Reset image size

In contrast to cropland expansion, cropland abandonment produced positive environmental outcomes (table 1). Compared with a tilled corn–soybean rotation, cropland abandonment to grassland reduced soil erosion and nutrient loss by a factor of 4–5, while increasing SOC stocks by 460 kg C ha⁻¹ yr⁻¹ on average (calculated as the difference between grassland and tilled corn–soybean; see table 1). Like that of grassland conversion, the spatial pattern of environmental benefits corresponded with the distribution of areas abandoned (figure 2(c) and SI figure 13). In addition, slope (SI figure 6(b)) and other site-specific factors can alter the pattern of the four environmental variables. Environmental benefits of cropland abandonment varied greatly between sites, as indicated by the large standard deviations in outcomes (figure 5) and spatial variability of environmental impacts on a per unit area basis (SI figure 14).

Zoom In Zoom Out Reset image size

Given the opposite effects of cropland expansion and abandonment, we estimated the net effects from both expansion and abandonment across the 12 states in the US Midwest. We calculated the net impacts of the tilled corn–soybean scenario, where we simulated grassland conversion to corn–soybean and abandonment from corn–soybean to grassland. In aggregate across the region, the negative effects of grassland conversion outweighed the benefits of cropland abandonment, such that the net outcome was greater erosion, and N, P, and SOC loss (table 2; figure 6). The net results are driven mainly by the cropland expansion, as it accounts for over 80% of the total converted land considered in this study.

Zoom In Zoom Out Reset image size

To put our findings into context, we compared our results to impacts from current cropland and the benefits of US CRP land (table 2). Compared to the reported soil erosion, and N and SOC loss from cultivated cropland in previous literature (West et al 2008, Zhang et al 2015), the net impacts of cropland expansion and abandonment resulted in 7.9%, 3.7%, and 5.6% increases in soil erosion, N loss, and SOC loss, respectively, from US Midwestern cropland. Comparing directly for context, our findings also suggest the net land use change in these 12 states could counter many CRP benefits, with the magnitude, for instance, to offset up to 18.6% of the N retention benefits for the entire US (table 2).

For further context, we also estimated the economic effects of net erosion and N loss. Based on 11.8 Tg yr⁻¹ of soil erosion (table 2) and assuming a value of $8 USD per metric ton of soil in 1992 (or $14.3 in 2018 after considering inflation) (Pimentel et al 1995), the economic cost (including both on-site crop productivity and off-site siltation and water quality damage) of increased soil erosion in the Midwest amounts to $169 million USD per year. Thus, an estimated cost of $5.1 billion USD (not factoring inflation) would be expected over 30 years, if net soil erosion costs from grassland conversion (i.e. expansion and abandonment) were internalized. Similarly, Compton et al (2011) estimated the cost of N pollution ranged between $2.20 and $56.00 USD per kg N due to its impact on factors, such as biodiversity, recreation, and clean water. Based on that estimate, net cropland expansion and abandonment would cost ca. $10–246 million USD per year (or $30 million–7.4 billion USD over 30 years, not factoring inflation) Thus, even though there are large uncertainties associated with the estimates using the above approaches, increased erosion and N loss exert notable economic cost.

To our knowledge, this is the first study to estimate the cumulative environmental quality impacts of observed grassland conversion to row crops across the US Midwest during this time period, apart from estimates of SOC loss. Despite this, our results can be compared in both directionality and on a per area basis to previous estimates in the literature. Our erosion simulations are consistent with previous findings that soil erosion is low for land in perennial cover (Nearing et al 2017), and conversion to corn or soybeans will generally increase soil erosion (Yasarer et al 2016). In addition, our results are consistent with the scientific literature showing tillage intensity is a key factor influencing soil erosion (Chung et al 1999, Hao et al 2001, Wang et al 2008). Typically, less tillage, such as conservation tillage and no-till management, and greater soil cover by crop residues will help minimize soil erosion (Seitz et al 2019).

On a per unit area basis, the simulated N and P losses here (table 1; figure 3) also generally fall within the range of values reported in previous studies. For example, USDA's Natural Resources Conservation Service estimated an average total N loss between 20 and 60 kg N ha⁻¹ yr⁻¹, while P losses were between 1 and 6 kg P ha⁻¹ yr⁻¹ from cropland across the US (USDA-NRCS 2017a, 2017b). Combining both field observation and modeling results, Syswerda et al (2012) showed nitrate losses of 62.3 ± 9.5 and 41.3 ± 3.0 kg N ha⁻¹ yr⁻¹ respectively, for conventional and no-till row crops in Michigan. Our estimates of 17 to almost 50 kg N ha⁻¹ yr⁻¹ and 2.5–5.2 kg P ha⁻¹ yr⁻¹ fall within a reasonable range of these results.

In line with previous studies (Schierhorn et al 2013, Kämpf et al 2016, Yu et al 2019), our simulations show that cropland abandonment increases SOC, while grassland conversion to cropland decreases it. In general, our estimate of ca. 0.5 Mg C ha⁻¹ yr⁻¹ (calculated as the difference between SOC under the grassland and tilled corn–soybean scenarios; see table 1) is close to the previously reported SOC sequestration benefits from cropland abandonment of ca. 0.7 Mg C ha⁻¹ yr⁻¹ (Schierhorn et al 2013, Kämpf et al 2016). For grassland conversion to cropland, estimates of SOC loss can vary greatly, often depending upon tillage practices and the method of estimation. Our estimates of SOC loss of ca. 3.6–12 Mg C ha⁻¹ for grassland to no-till corn or tilled corn–soybean, respectively (see table 1 for per area values, multiplied by 30 years), are generally lower than in empirical studies (Gelfand et al 2011, Spawn et al 2019). For instance, Spawn et al (2019) estimated a release of 55.0 Mg C ha⁻¹ (±39.9 Mg C ha⁻¹) for the entire contiguous US, and 47.2 Mg C ha⁻¹ for the SOC component of converted grasslands within our 12 state region. Yet, for comparison purposes, they also calculated a much lower grassland-to-cropland SOC loss of 7.9 Mg C ha⁻¹ from the National Greenhouse Gas Inventory, which uses a process-based model (Spawn et al 2019). Differences between empirical and process-based methods—the latter of which was employed in this study—have been widely acknowledged (Zhang et al 2015), as they use different input data and mechanistic processes to represent the ecosystem responses. Overall, the SOC estimates for grassland conversion to cropland we provide here likely falls on the low end of the possible impacts, while others may estimate the high end.

Several points should be considered when assessing the results of this study. Foremost, our results suggest a directionality and magnitude of effects, but these are modeling estimates of reality, not actual measurements. Our model verification, for example, indicates that EPIC may overestimate soil erosion by approximately 14% (SI figure 4). Moreover, we did not model the exact crop and management-type for each converted parcel of land since this would have been computationally too demanding. Instead, we modeled the effects of conversion only to corn or corn–soybean. Corn (29.3%) and soybeans (26.7%) were the dominant crops planted on converted grasslands nationwide between 2008 and 2016, with wheat (22.6%) a close third (Lark et al 2020). Alfalfa was the next closest with 3.6%. Corn and soybeans were even more prevalent across the 12 states we modeled—each planted on 36% of the land converted to crop production. We did not simulate the effects of wheat because this crop is not as common in the eastern portion of the US Midwest. The planting of wheat on grasslands may have comparatively lowered N and P loss, while also increasing SOC losses relative to corn or corn–soybean rotations. Corn requires more N and P fertilizer than wheat on average (USDA-ERS 2019), and replacing wheat with corn typically increases SOC (Qin et al 2016).

Likewise, we could not apply an exact tillage type to each parcel of converted land since this information at this level is not available. In this study, we bracketed the spectrum of tillage by applying both no-till and conventional tillage scenarios, representing the two extremes in tillage practices, with actual effects likely in between these endpoints. In addition, the converted grasslands were not necessarily undisturbed or unmanaged, but instead represented a spectrum of grassland and management types, including pasture, lands managed for hay, and CRP grasslands. This likely most affected the SOC findings. We used a 30 year spin-up time for the grasslands before converting to crops, yet some of these grasslands may have only been out of production for 6–10 years. Hence, this may have resulted in an overestimation of SOC stocks and subsequent losses in some locations when converted. However, as noted above, our estimates of effects appear to be in line, or even on the low-end, compared to other estimates of C loss.

We also could not follow the temporal dynamics of impacts through time due to computer-storage limitations and simulating so many parcels. Instead, we present annual, long-term average effects in this study, simulated over a 30 year period and consistent with the central aim of the study. It is likely that effects were not uniform across this entire period. For instance, grassland conversion to tilled cropland typically causes a rapid loss of soil C in the first few years of cultivation due, in part, to the mixing of soil and decomposition (Davidson and Ackerman 1993). By contrast, the accumulation of C after cropland abandonment can take much longer as biomass inputs increase (Schierhorn et al 2013). In a similar fashion, annual rates of erosion and nutrient loss were likely not uniform across the time period simulated.

Beyond the estimated effects on soil erosion, N, P, and SOC, there are additional environmental impacts of grassland-to-cropland conversion that we did not model here, including those on air and water quality. For example, recent cropland expansion is associated with an increase in airborne dust-generation in the Midwest, with implications for human health, downwind ecosystems, and visibility (Lambert et al 2020). In addition, our estimates represent edge-of-field soil and nutrient losses, providing only the potential for water quality degredation. Hydrologic modeling is a logical next step, needed to translate these results to sediment and nutrient loading to waterways.