Glacier Creek Preserve (GCP) is located in eastern Nebraska 20 km northwest of Omaha, NE, United States (Table 1 and Figure 1A,B). The 4 km² preserve encompasses the entire first order Glacier Creek watershed and includes both restored tallgrass prairie and agricultural land use on the gently rolling topography (slopes < 15%). The site was used entirely for agriculture from approximately 1870 until 1970, when the southern 0.54 km² of the preserve were reseeded to native tallgrass prairie (Figure 1C). The prairie vegetation is managed with a 3 year fire return interval with one third of the area burned each year. The northern portion of the site (0.32 km²) continues to be used for minimal till corn (Zea mays L.) and soybean [Glycine max (L.)] agriculture; tilling practices switched from intensive to minimal tillage around 1970. In 2017 soybeans were grown in the study area and broadcast phosphorous and potassium fertilizer was applied the previous fall following a corn harvest.

Parent material at the site is dominated by Peoria loess in the upland areas (>10 m thick) and pre-Illinoian glacial till underlies the creek that runs through the center of the watershed. Soils on the ridgetop topographic position, where soil sensors and soil pore water were collected, are mapped as Contrary-Monona-Ida Complex (Soil Survey Staff, 2018). On the agricultural side of the watershed, soils are Contrary silt loam (Fine-silty, mixed, superactive, mesic Dystric Eutrudepts) while prairie soils are Monona silt loam (Fine-silty, mixed, superactive, mesic Typic Hapludolls). Mean annual temperature is 10°C and mean annual precipitation is 78 cm. In 2017, average air temperature was 11.8°C and precipitation in the watershed totaled 78 cm.

The Intensively Managed Landscapes Critical Zone Observatory (IML CZO) is located approximately 30 km northwest of Iowa City, IA (Table 1 and Figure 1A,B). The 270 km² HUC 10 watershed is drained by Clear Creek and land use is dominated by roughly 100 years of agricultural land use across the gently rolling topography (slopes < 18%) (Figure 1D). The agricultural study site is located on a farm in the northwest corner of the watershed where minimal tillage practices replaced more intensive chisel plow tillage in the 1990s (Papanicolaou et al., 2015b). In 2017 corn was grown in the study area with an application of broadcast phosphorous and potassium fertilizer and anhydrous ammonium following harvest of the soybean crop in 2016. Parent material at the site is Peoria loess (∼6 m thick) overlying Sangamon age glacial till. Soils are mapped as Tama silty clay loam (Fine-silty, mixed, superactive, mesic Typic Argiudolls) (Soil Survey Staff, 2018).

A small restored tallgrass prairie (0.06 km²) is located approximately 16 km to the east of the agricultural site in F.W. Kent Park. The prairie was originally forested prior to clearing for agriculture in the late 1800s; conventional corn-soybean row crop agriculture ceased in 1997. The site was left fallow until 2009 then reseeded with prairie grass mix; prairie vegetation is currently maintained through controlled burns every ∼5 years. Parent material at the site is thin Peoria loess (∼ 1.6 m thick) overlying Sangamon age glacial till; soils at the site are Fayette silt loam (Fine-silty, mixed, superactive, mesic Typic Hapludalfs) (Soil Survey Staff, 2018). Mean annual temperature in the Clear Creek watershed is 9°C and mean annual precipitation is 89 cm. In 2017, average air temperature was 12.1°C and 123 cm of precipitation was recorded near the agricultural study area.

At the Nebraska study site, two meteorological stations were installed on ridgetop topographic positions, one in the restored prairie and one in the agricultural field. Both stations provided hourly measurements of atmospheric and soil parameters at 10, 25, 50, and 100 cm depth (Figure 1C). The Campbell Scientific (Logan, UT, United States) meteorological stations recorded data on a CR1000 datalogger with the following sensors: anemometer (R.M. Young’s 03002 Wind Sentry Set), all season pyranometer (SP230), precipitation sensor (Texas Electronics Tipping Bucket Rain Gages), temperature and relative humidity sensor (HC2S3), barometer (CS106), and soil moisture, temperature and electrical conductivity (EC) sensors (CS655).

At the IML CZO in Iowa, a meteorological station recorded atmospheric parameters on a ridgetop topographic position at the agricultural study site (Figure 1D). The Meter (Pullman, WA, United States) meteorological station collected data with EM50 dataloggers including precipitation (ECRN-100 tipping bucket), wind speed and direction (Davis Cup anemometer), and relative humidity and temperature (VP-3) at 1 min. intervals. Soil parameters at 10, 20, and 60 cm depth were measured at both the agricultural and restored prairie sites every 15 min. Soil moisture and temperature were measured with a 5TM Moisture/Temperature sensor at the agricultural site and a GS3 Soil Moisture/Temperature/EC sensor at the restored prairie site.

General stream water quality parameters were collected in the field at the Nebraska site using Xylem (Rye Brook, NY, United States) ProDSS handheld sensors. The parameters included pH/oxidation reduction potential (ORP, 626904), temperature/conductivity (626902), turbidity (626901), and optical dissolved oxygen (626900). Measurements were taken approximately every 2 weeks at each of five sampling sites (Figure 1C). Each probe was calibrated with relevant calibration solutions on the day of sampling. Before taking a measurement, the probes were placed in the stream for several minutes to allow the probes to equilibrate and allow fresh water to displace any sediment disturbed by the placement of the probes.

A SonTek Xylem IQ in situ stream discharge sensor was placed in Glacier Creek at the Nebraska study site downstream of the GC2 sampling site in May 2017 (Figure 1C). The IQ system simultaneously measured the flow velocity and the stream channel dimensions using sound waves, which was then converted to volumetric flow. The IQ was programmed to log flow data every 15 min. In Iowa, several discharge sensors along Clear Creek recorded hourly discharge at the Homestead, Oxford and Coralville gaging stations¹ (Figure 1D).

Stream water samples were collected approximately monthly from all five sites in Glacier Creek in Nebraska beginning in April through December 2017. For data analysis purposes, four of the sampling locations were sorted based on physical proximity to either prairie or agricultural land uses. Locations GC3 and GC4 are physically closest to active agriculture and are located on the north fork of Glacier Creek, which drains predominantly agricultural land. Locations GC5 and GC6 are physically closest to the restored prairie and are located in the south fork of Glacier Creek, which drains the majority of the restored prairie. Site GC2 is located just below the convergence of the two forks, and was used for mass balance calculations described below. At each site, a grab sample was removed from the stream with a conditioned sampling beaker. The sample was filtered through a 0.45 μm PVDF syringe filter and then split. One split was left unacidified for anion and alkalinity analysis, the other split was acidified to pH < 2 with concentrated HCl. All samples were refrigerated at 4°C prior to analysis.

Clear Creek in the Iowa IML watershed predominantly drains agricultural land use in the headwaters and mixed urban and agricultural land use toward the mouth of the watershed, but does not have a stream draining predominantly restored prairie like in Nebraska. Grab samples from the Homestead gaging station (westernmost triangle in Figure 1D), which is located near the agriculture study site, were collected on two dates (2015-06-25 and 2016-12-01) and analyzed for cations and anions as described below.

Suction cup pore water samplers with a silicon carbide cup (SIC20) (UMS brand; Munich, Germany) were installed at four different depths in both restored prairie and agricultural land use at the Nebraska and Iowa study sites. Samplers were installed in spring – summer of 2016 to allow time for porous cups to equilibrate with soil conditions (Singh et al., 2018). Pore water samplers were inserted at an approximately 20° angle away from vertical into a hand-drilled hole using a custom gouge auger (TB-20 and TBE-100). In Nebraska, pore water samplers were installed at 10, 25, 50, and 100 cm depth; in Iowa, samplers were installed at 20, 60, 80, and 150 cm depth. Pore water samplers were connected to 500 ml glass sampling bottles (SF-500) and stored in an insulated plastic box (SF-box) buried in the ground. Sample bottles were evacuated down to ∼0.5 bar using a hand-operated vacuum floor pump (VPS-2) in Nebraska and an automated portable vacuum (VacuPorter) in Iowa. The vacuum was removed from all samplers and the site covered in straw from December to March to avoid frost damage to the pore water samplers.

Soil pore water was collected approximately every 2 weeks from May through November and the sample volume, pH, and electric conductivity were recorded in the field. Samples were filtered through a 0.45 μm PVDF syringe filter and then split. One split was left unpreserved for anion and alkalinity analysis, the other split was acidified to pH < 2 with concentrated nitric acid for cation analysis. All samples were refrigerated at 4°C prior to analysis. At both sites, insufficient sample volume limited some or all chemical analyses, especially during the driest parts of the year in late summer (Supplementary Tables 1–4).

Precipitation was collected every 2 weeks at the Iowa agriculture and prairie sites and at the Nebraska prairie site. The collection vessel included a plastic funnel covered by wire mesh connected by zip ties to a 250 ml plastic separatory funnel. A thin layer of mineral oil was added to prevent evaporation. A subsample was collected and filtered through a 0.45 μm PVDF syringe filter and then split and preserved in the same manner as the pore water samples. On some occasions, the volume of rainwater exceeded the capacity of the collection vessel or the addition of bird fecal matter contaminated precipitation samples and precluded their inclusion in the dataset (Supplementary Table 5).

For all water samples, bulk ions (Na⁺, K⁺, Mg²⁺, Ca²⁺, Cl^-, NO₂^-, NO₃^-, SO₄^2-, and PO₄^3-) were determined using a Dionex Aquion ion chromatograph using CS12A analytical and guard columns for cation analysis, and AS23 analytical and guard columns for anion analysis (Supplementary Tables 1–10). Trace metal analysis (Al³⁺ and Mn²⁺) was completed on a Varian inductively coupled plasma mass spectrometer (ICP-MS) (Walnut Creek, CA, United States). Total dissolved iron (Fe²⁺ and Fe³⁺) was determined through complexation with phenanthroline on an Agilent Technologies Cary 8454 UV-Vis spectrophotometer (Santa Clara, CA, United States). For the instrumental methods, initial method validation was completed using sample spikes, dilutions, and replicate samples. For each analytical run, calibration standards and check standards were analyzed to ensure data quality.

Stream sample alkalinity determinations were completed within 24 h of sampling. Potentiometric alkalinity titrations were completed on 50 mL aliquots of stream samples using a Hannah Instruments (Woodsocket, RI, United States) autotitrator with standardized HCl. When possible, titrations were completed on pore water samples. Due to sample volume limitations, only 7 mL of sample were used.

Since stream flow rate was only directly measured after the south fork (prairie) and north fork (agriculture) of Glacier Creek had converged (Figure 1C), mass balance calculations using conservative ions were used to split the total flow on the date of sampling into flow rates for each fork of the creek. Concentrations of several different ion pairs (Na/Cl, K/Cl, and Ca/Cl) from sampling positions just above and just below fork convergence were used to estimate the two fork flows (Davis and Cornwell, 1998). Once the flow for each fork was known, the flow was multiplied by the stream concentration and normalized to drainage area at the most downgradient sampling point on both the south fork and north fork to determine the elemental flux emanating from the prairie and agricultural land uses, respectively. During the driest parts of the year (fall/winter) total stream flow was frequently too low to be reliably measured. For months where the IQ sensor was not accurately capturing the total flow rate (September, October, and December), average monthly flow for each fork was estimated by averaging July, August, and November monthly discharges, which were considered to represent baseflow conditions at the site (i.e., not influenced by spring flush) (Supplementary Table 11).

Similarly, soil pore water fluxes were determined for each sample by multiplying elemental concentration by the collected volume and dividing by the area over which each pore water sampler suction cup captures water (25 cm²) and the sampling period (∼14 days) (Supplementary Tables 12–14) (Weihermuller et al., 2007). Soil pore water fluxes were averaged over all depths for the year at each site to determine an average flux by site and land use.

All speciation calculations were completed using PHREEQC (Parkhurst and Appelo, 2013). For stream samples, where all major ions had been directly measured for every sample, the ion concentrations and basic water quality parameters (pH, dissolved oxygen, temperature) were used as input to calculate the error in the charge balance, saturation indices for relevant mineral species, and the partial pressure of CO₂ (P_CO2). Each stream water sample was speciated individually, and the annualized results from the prairie or agricultural stream locations were averaged for analysis. For the pore water samples, volume limitations precluded direct determination of all ion concentrations in every sample. Some samples had either anion or cation analysis but not both, many were missing alkalinity values, and no samples had trace metal analysis (Fe, Mn, or Al). Since each sample could not be speciated individually, the annual average concentration for each measured ion was used instead. Pore water samples were averaged by land use type in Nebraska and Iowa; individual ion concentrations from all prairie related samples were averaged and individual ion concentrations from all agricultural samples were also averaged. To remove any outliers from consideration, concentrations larger than two standard deviations above the mean were not included. Non-detectable ion concentrations were also omitted from averaging. Non-detectable concentrations were common for NO₂^- and PO₄^3-, while the other major ions were more consistently detected.

Differences between agriculture and restored prairie solute concentrations, soil moisture and soil EC were analyzed for significant differences using a two-sample student t-test (α = 0.05) using OriginLab software (Wellesley Hills, MA, United States). The statistical test compared the agriculture and restored prairie data at each site individually (e.g., Nebraska prairie vs. Nebraska agriculture, Iowa prairie vs. Iowa agriculture).

In the agricultural soils, pore water annual average cation concentrations were significantly higher (p < 0.05) for Ca²⁺ and Mg²⁺, but not Na⁺ and K⁺, in both Nebraska and Iowa (Figure 4A,B). As with the concentrations, annual average Ca²⁺ and Mg²⁺ elemental fluxes in agricultural soil pore waters were higher at both Nebraska and Iowa, but the difference is significant only in Iowa (Figure 4C,D). At all measured depths, soil pore water cation concentrations were lower in prairie than agricultural soils (Supplementary Figure 1).

For stream water concentrations in Nebraska, all of the major cations except for Na⁺ were significantly higher (p < 0.05) in stream locations draining restored prairie compared to stream locations draining agricultural fields (Table 2 and Figure 5). Conversely, Na⁺ was significantly lower in streams draining prairie land use. However, the cation fluxes in the stream were higher in water draining agricultural fields (Table 3). The concentrations of redox active species (Fe/Mn) were also significantly higher in the prairie locations. Aluminum concentrations were consistently just above the detection limit for the ICP-MS (0.7 ppb), and were not different between the prairie and agriculture associated sections of the stream. Stream fluxes for the major cations associated with agriculture were higher than the elemental fluxes from prairie (Figure 5).

Soil pore water Cl^- and NO₃^- concentrations were higher overall in Nebraska compared to Iowa, but only NO₃^- concentrations were significantly higher (p < 0.05) in agricultural soils compared to prairie soils in Nebraska; anion concentrations were not significantly different amongst soil pore waters in Iowa (Figure 6). Very little NO₃^- (<1 mg L^-1) was detected in either agriculture or prairie soil pore waters in Iowa. Masarik et al. (2014) measured mean annual NO₃^--N leachate concentrations of 12.2 mg L^-1 under no till corn crops and <0.1 mg L^-1 under restored prairie vegetation in silty loess soils with a similar climate to our study. The Iowa site has much lower NO₃^- concentrations under agriculture than the Masarik et al. (2014) data, while the average under Nebraska agriculture is much higher (42 ± 51 mg L^-1), although highly variable. The concentrations under the prairie land use at both locations are within an order of magnitude of Masarik et al. (2014). At this point it is not clear if the variability in concentrations are related to fertilizer application rates or varied biogeochemical cycling rate of nitrogen at the different sites.

Similar to the cations, major stream anion concentrations (Cl^- and SO₄^2-) were significantly higher (p < 0.05) in the prairie locations than they were in the agricultural stream (Figure 6). The minor anions (NO₂^-, PO₄^3-) were frequently non-detectable. Nitrate concentrations were more reliably detected, but were not significantly different between streams draining prairie and agriculture in Nebraska. When detectable in Nebraska stream samples, NO₃^- concentrations were typically < 0.4 mg L^-1. In both the prairie and agricultural stream stretches, there was a pulse (>10 mg L^-1) of NO₃^- in early spring (March for agriculture, and April for prairie) (Supplementary Tables 6–10). The juxtaposition in time implies a spring flush of NO₃^- from both systems, but longer time scales are needed to confirm the large episodic NO₃^- exports. Nitrite concentrations were typically too low to be detected (<11 ppb), but they generally followed the NO₃^- concentrations. Phosphate was also generally too low to be detected (<19 ppb). When it was detected, the SI for hydroxyapatite was generally undersaturated (SI range: -4.8 to -0.08) implying a non-mineral control on PO₄^3- concentrations.

The alkalinity was significantly higher in the stream waters associated with the prairie (Tables 4, 5). Unlike other major ions, the pore water data was not consistent between agriculture and prairie between the two study sites. In Nebraska, the pore waters had slightly higher alkalinity under the prairie, but in Iowa the alkalinity was noticeably lower under the prairie.