Sulfur isotopes reveal agricultural changes to the modern sulfur cycle

We measured δ³⁴S–SO₄ ²⁻ and [SO₄ ²⁻] in soil leachate and surface water samples collected throughout the Napa River Watershed, California, US. This watershed is 1103 km² and is dominated by two contrasting land use/land covers (LULCs): wine grapes are grown nearly exclusively throughout the Napa Valley (∼11% of the watershed area) and are surrounded by hillsides of forests (26%) and woodlands (37%; figure 1). The Watershed encompasses the traditional and contemporary territories of the Lake Miwok, Coast Miwok, Southern Pomo, Wappo, and Patwin peoples. The Napa River drains into extensive wetlands in San Pablo Bay, connecting to the greater San Francisco Bay Estuary.

The region's Mediterranean climate strongly influences seasonal hydrology and agricultural management practices. Wine grapes grow during the dry season (April through September) and farm workers spray elemental S (S⁰) fungicide weekly to biweekly to prevent powdery mildew infection. Tributaries to the Napa River are largely dry during this period. During the wet and dormant crop season (October through March), nearly all annual precipitation falls as rain. There is a gradient in precipitation from north to south in the watershed: 931 mm in St. Helena to 518 mm in Napa (Arguez et al 2012). Rains periodically saturate vineyard soils to ⩾0.5 m depth, mobilizing S below the vine rooting zone and affecting soil porewater δ³⁴S–SO₄ ²⁻ values (Hinckley et al 2008, Hinckley and Matson 2011). Rains also activate tributary flows. Although redox profiles were not collected in this study, observations of surface water ponding throughout wet seasons suggests the potential for dynamic redox conditions in vineyard soils.

We established sampling locations within the predominant LULCs in the watershed and to incorporate the precipitation gradient and variability in underlying geologies (figure 1; supplementary table 1). We collected samples during 11 field campaigns, focusing efforts during the wet seasons of water years 2018–2020 (supplementary figure 1).

We sampled soil leachate from six vineyards, one forested area, and one grassland area. At each sampling site, we installed four tension lysimeters (SoilMoisture, Inc.) across one vineyard management block or equivalent area (for the non-vineyard sites; ∼1–2 ha) to capture spatial heterogeneity. Lysimeters were installed to 0.5–0.6 m depth, except at four steep sites, where lysimeter depths targeted shallow (0.2–0.3 m) and deep (0.4–0.6 m) flow paths. We purged lysimeters once before collecting samples into polycarbonate VacLok 60 ml syringes (Merit Medical Systems) or high density polyethylene bottles under vacuum. The lysimeters did not appear to affect δ³⁴S–SO₄ ²⁻ values (supplementary figure 2).

We also collected water samples from rain, irrigation lines, culvert outflows, tributaries, and the Napa River. We pre-rinsed HDPE bottles three times with sample water before filling completely to minimize headspace. Rainwater was collected into aluminum trays pre-rinsed with deionized water. We filtered all water samples through sequential 0.8 µm (Pall Acrodisc, sterile Supor polyethersulfone) and 0.45 µm (VWR, sterile polyethersulfone) 25 or 47 mm filters, stored and shipped samples on ice, and then we froze samples (∼−20 °C) until processing and analyses. We analyzed six S⁰ fungicide samples provided by winegrower collaborators.

We analyzed samples for [SO₄ ²⁻] and the S stable isotope composition of SO₄ ²⁻ or S⁰. We measured [SO₄ ²⁻] using an ion chromatograph (Dionex; detection limit 0.2 mg S l⁻¹, relative percent difference between sample duplicates ⩽5%). To prepare aqueous samples for S stable isotope analysis, we thawed samples overnight and then precipitated BaSO₄ (Hinckley et al 2008). Briefly, we acidified samples to within a pH of 2–4 with hydrochloric acid (Fisher Chemical, trace-metal grade), brought samples to a boil, and added 10% (w/w) BaCl₂ solution (MilliporeSigma, ACS grade) in excess. We collected BaSO₄ precipitate on 0.45 µm Whatman mixed cellulose ester filters, which we then dried overnight at 60 °C. We analyzed solid BaSO₄ and S⁰ samples for δ³⁴S on a Flash IRMS elemental analyzer coupled with a Delta V Plus isotope ratio mass spectrometer (Thermo Fisher Scientific EA IsoLink). We report stable isotope values in conventional δ-notation in parts per 1000 (‰; Sharp 2017) and relative to the international standard Vienna-Canyon Diablo Troilite. Long-term analytical precision for isotope analysis is ±0.2‰ based on using internal reference standards calibrated annually against International Atomic Energy Agency-certified reference materials.

Statistical analyses were conducted in R software (v.4.0.4; R Core Team 2021). We selected non-parametric analyses, because the data violated the parametric assumptions of normality and sample independence. We tested the null hypothesis that median δ³⁴S–SO₄ ²⁻ values and [SO₄ ²⁻] were equal across LULC groups using a non-parametric Kruskal–Wallis test (Kruskal and Wallis 1952), followed by post-hoc Dunn's test (Dunn 1964) with a Holm's p-adjustment (Holm 1979). We chose a p-value < 0.05 to determine statistical significance and, throughout, values are reported as the median ± interquartile range.

Despite differences in the quantity of S applied, underlying geology, regional climatology, and topography (supplementary table 1), the six vineyards sampled followed a general pattern showing an increase in median δ³⁴S–SO₄ ²⁻ values from lower to higher S concentrations (figure 2; Hermes and Hinckley, 2021). Vineyard δ³⁴S–SO₄ ²⁻ values were 7.2 ± 5.2‰ below 22 mg SO₄ ²⁻–S l⁻¹ (n = 35; range = 13.7‰) and shifted to 12.8 ± 3.7‰ (n = 20, range = 11.5‰) above 22 mg SO₄ ²⁻–S l⁻¹. We compared our measurements to prior data collected intensively in an additional vineyard (Hinckley et al 2008) and found that the overall pattern was remarkably consistent, suggesting that the vineyard S fingerprint is detectable within and across vineyards and over time (supplementary figure 2).

Zoom In Zoom Out Reset image size

The S patterns from vineyard agriculture and non-agricultural areas of the watershed were strikingly different (figure 2). Non-agricultural soil leachate and surface water had δ³⁴S–SO₄ ²⁻ values of −0.28 ± 5.7‰ (n = 30, range = 14.4‰), depleted by ∼10‰ relative to vineyards (9.9 ± 5.9‰; n = 55, p = 1.2 × 10⁻¹⁰). Surface waters from non-agricultural areas also had roughly three-fold lower [SO₄ ²⁻] than vineyard samples (5.0 ± 5.5 and 13.6 ± 26.6 mg SO₄ ²⁻–S l⁻¹, respectively; p = 1.1 × 10⁻⁴), although we note that a number of vineyard waters had similar [SO₄ ²⁻] to those from non-agricultural areas (∼1–15 mg SO₄ ²⁻–S l⁻¹, n = 29, figure 2). Within individual sub-catchments, vineyard agriculture δ³⁴S–SO₄ ²⁻ values were enriched by 17.8–20.5‰ relative to adjacent non-agricultural areas (supplementary figure 3). Our results clearly indicate that vineyard agriculture has a consistent S biogeochemical fingerprint that is distinct from non-agricultural areas.

Moving beyond agricultural source areas (fields), culvert outflows and tributaries to the mainstem of the Napa River reflected a combination of vineyard and non-agricultural sources. We found that vineyard culvert outflows had δ³⁴S–SO₄ ²⁻ values and [SO₄ ²⁻] similar to soil leachate (figure 2), suggesting that the S fingerprint found in vineyard soil leachate is carried by drainage outflows into the broader watershed. Tributaries draining mixed LULC areas had δ³⁴S–SO₄ ²⁻ values that spanned the entire range measured, from −7.4 to 17.6‰, and [SO₄ ²⁻] of 13.0 ± 7.1 mg SO₄ ²⁻–S l⁻¹ (n = 17), which were intermediate to vineyard agriculture and non-agricultural endmembers. Two tributaries with 12% and 23% vineyard land cover, respectively, had δ³⁴S–SO₄ ²⁻ values (∼4–18‰) that were more similar to vineyard soil leachate and surface waters than the predominant non-agricultural areas in those sub-catchments (supplementary table 1). However, it is worth noting that their δ³⁴S–SO₄ ²⁻ values changed from ∼4–6‰ to 10–18‰ between early-season and late-season sampling campaigns (supplementary figure 3). We hypothesize that these changes in δ³⁴S–SO₄ ²⁻ values reflect within-season shifts in the contributions of vineyard and non-agricultural S sources depending on source area hydrologic activation and connectivity to tributary channels. Overall, our ability to detect the influence of the vineyard S fingerprint within tributaries indicates that vineyard S is detectable at sub-watershed scales.

To capture the watershed scale, we measured S fingerprints in surface water from the mainstem of the Napa River, which fell within the range of vineyard measurements (p = 0.9) and varied spatially and temporally. Napa River samples had δ³⁴S–SO₄ ²⁻ values of 10.5 ± 7.0‰ and 8.5 ± 6.1 mg SO₄ ²⁻–S l⁻¹ (n = 13; figure 2). To examine spatial changes to the S fingerprint along the Napa River, we sampled a transect from the headwaters to just below the tidal extent (figure 1(a); supplementary table 1). The headwaters drain non-agricultural (primarily forested) areas and had similar S chemistry to that of non-agricultural tributaries (4.3‰ and 4.3 mg SO₄ ²⁻–S l⁻¹; point 1, figure 2). Moving downstream, transect samples increased in [SO₄ ²⁻] and δ³⁴S–SO₄ ²⁻, consistent with the increase from 0% to 5%–15% vineyard agriculture in contributing areas (points 2–6, figure 2; supplementary table 1). Overall, transect samples had δ³⁴S–SO₄ ²⁻ values of 4.3–13.4‰ and 4.3–20.2 mg SO₄ ²⁻–S l⁻¹. We captured temporal variability through repeat sampling of the Napa River above the city of Napa, CA and found that δ³⁴S–SO₄ ²⁻ values ranged from 4.9 to 18.3‰ with 6.9–16.1 mg SO₄ ²⁻–S l⁻¹ (n = 7; figure 1(a); point 4, figure 2). These results show that the Napa River S fingerprint varies temporally by as much as it does spatially along the transect (figure 2). Similar to the shifts in S chemistry observed in tributaries, we suggest that the shift in δ³⁴S–SO₄ ²⁻ and [SO₄ ²⁻] at the repeat sampling location could arise from changes in source water contributions over the course of the wet season. Linking hydrology and S chemistry at catchment-to-watershed scales is an outstanding, but critical, research direction. Nevertheless, the overall enriched S stable isotope signal and elevated [SO₄ ²⁻] within the Napa River indicate that the vineyard S fingerprint remains predominant at the watershed scale.

The notable contrast between S stable isotope values derived from vineyard and non-agricultural areas yields insights into S sources and dominant processes within fields and sub-catchments. To examine S sources, we compared laboratory-based grassland (non-agricultural) and vineyard soil leachates (Hermes et al 2021) to S inputs (figure 2). Soil leachate δ³⁴S–SO₄ ²⁻ values in non-agricultural areas were 4.9 ± 5.0‰ (n = 14), similar to precipitation water (5.5‰; n = 1; Hinckley et al 2008) and surface waters from non-agricultural areas (−0.3 ± 5.7‰; n = 27). The isotopically-depleted soil leachate, culvert, and stream water in non-agricultural areas likely reflect a mixture of precipitation and geologic S sources enriched in ³²S, as occurs with the oxidation of reduced S species from geologic weathering or springs (Grasby et al 1997, Mayer et al 2010). In contrast, repeated laboratory-based measurements of soil leachate from one vineyard had δ³⁴S–SO₄ ²⁻ values (n = 38) that were enriched by ∼15.3‰ relative to S fungicide (3.1 ± 1.8‰, n = 6) and by ∼13‰ relative to irrigation and precipitation water (5.7 ± 0.5‰; n = 10; figure 2). Field-based vineyard soil leachate δ³⁴S–SO₄ ²⁻ values fell in between the laboratory leachates and S sources—similar to the mixing of soil waters and sources reported by Hinckley et al (2008). Although the enriched vineyard soil leachate δ³⁴S–SO₄ ²⁻ values could reflect additional S sources to vineyards such as gypsum—a soil conditioner—not all vineyards we sampled apply gypsum (supplementary figure 4). Rather, the discrepancy between soil leachate δ³⁴S–SO₄ ²⁻ and S sources in vineyards implies that additional S fractionation processes occur.

We hypothesize that a number of microbially-mediated S transformations that fractionate S within soils may affect the isotopically enriched vineyard S fingerprint, as summarized in a conceptual model (figure 3). Within vineyards, rapid oxidation of S⁰ fungicide following application to soils during the dry, growing season results in accumulation of SO₄ ²⁻ and immobilization into organic S species (Germida and Janzen 1993, Yang et al 2010, Hinckley et al 2011), both processes with minimal S stable isotope fractionation (Mitchell et al 1998, Chalk et al 2017). We hypothesize that the enriched vineyard soil leachate δ³⁴S–SO₄ ²⁻ values relative to S inputs could be derived from MSR, a process that strongly fractionates ³⁴S/³²S and enriches the residual SO₄ ²⁻ pool (Kaplan and Rittenberg 1964, Bradley et al 2015). Although typically found in low oxygen environments (Barton and Fauque 2009), MSR could occur during intermittent-to-sustained soil saturation during irrigation events, the wet season, and/or in soil anoxic microsites within oxygenated soils (Hansel et al 2008, Santana et al 2021), similar to the discovery that oxic soils host high rates of methanogenesis (Angle et al 2017). Since sulfur isotope fractionation is inversely related to sulfate reduction rate (Kaplan and Rittenberg 1964), even slow rates of MSR could impart a strong effect on vineyard soil δ³⁴S–SO₄ ²⁻ values. In contrast to S biogeochemistry within vineyards, depleted δ³⁴S–SO₄ ²⁻ values in forested/grassland areas compared to vineyards likely reflect mixing of atmospheric and geologic S, enriched in ³²S (Mayer et al 2010). We next discuss additional processes that may prevent 'runaway' enrichment of the soil SO₄ ²⁻ pool within vineyards.

Zoom In Zoom Out Reset image size

We hypothesize that three additional processes may act to constrain the agricultural S fingerprint. First, there has been little research into how soil wetting and drying cycles control the balance of S oxidation and reduction during irrigation events or the wet season. Sulfur oxidation can result in ∼1‰ depletion of the SO₄ ²⁻ pool (Wainwright 1984); this process could act to counter the enrichment effects of MSR on δ³⁴S–SO₄ ²⁻ values (figure 3). Highly managed irrigation practices and water routing through tile drains and ditches within and surrounding vineyards likely influence soil redox state as well as S transport to streams by controlling water residence times. Alternatively, although sulfide produced by MSR is rarely measured or studied in upland agricultural soils (Wainwright 1984), if it is indeed produced, it could be abiotically scavenged by the predominant organic S pool (Sleighter et al 2014, Poulin et al 2017). Upon organic S remineralization, the newly formed SO₄ ²⁻ would carry the depleted ³⁴S signature from MSR, limiting further enrichment of δ³⁴S–SO₄ ²⁻ values. Secondary SO₄ ²⁻ production from organic S remineralization is an additional source of SO₄ ²⁻ in forested ecosystems (Mayer et al 1995, Marty et al 2019), and more research is needed to understand the effects of cycling between organic and inorganic S on δ³⁴S–SO₄ ²⁻ values and [SO₄ ²⁻]. Finally, reactive S intermediate species may complicate studying agricultural S transformations within soils, known as the 'cryptic' S cycle (Hansel et al 2015). A key next step in examining the S cascade is to conduct studies that constrain the enriched, asymptotic agricultural S fingerprint, including measurements of soil redox conditions alongside S biogeochemical processes and rates and S-isotope fingerprinting.

To put our measurements from the Napa River Watershed into perspective, we compared our δ³⁴S–SO₄ ²⁻ values and [SO₄ ²⁻] to those from rivers around the world, compiled by Burke et al (2018). Our δ³⁴S–SO₄ ²⁻ measurements collected throughout the Napa River Watershed were strikingly similar to the pattern of values from rivers (figure 4). Burke et al (2018) calculated a modern flux-weighted global riverine δ³⁴S value of 4.4 ± 4.5‰ (1 s.d.). Our overall average δ³⁴S–SO₄ ²⁻ value of 6.13‰ was slightly enriched from this mean, but within one standard deviation. Globally, river δ³⁴S–SO₄ ²⁻ values ranged from −13.4 to 21.7‰, and, remarkably, our samples from within the Napa River Watershed alone accounted for much of this variability (−7.4–18.3‰). The spatial and temporal variability we found along the Napa River is similar to Burke et al's (2018) note that δ³⁴S values from a single river can range widely, driven by varying tributary contributions with different S sources (Burke et al 2018).

Zoom In Zoom Out Reset image size

Comparing our measurements to the global dataset (Burke et al 2018) also reinforced the importance of considering both S sources and processes in interpreting δ³⁴S–SO₄ ²⁻ values. The lowest global δ³⁴S–SO₄ ²⁻ values measured (−8.5 to −13.4‰) were from the Santa Clara River in Southern California—reflecting oxidation of pyrite and organic S from organic-rich shales and sandstones of the Monterey Formation. Some of our depleted δ³⁴S–SO₄ ²⁻ values from non-agricultural tributaries drain the Great Valley Complex, a similar shale/sandstone bedrock. However, our non-agricultural values may be less depleted than the Santa Clara River overall due to the highly heterogeneous geology of the Napa Watershed. The highest global δ³⁴S–SO₄ ²⁻ values measured (∼14–22‰) were from the Lena and Yenisei rivers draining the Siberian Platform, a source of enriched evaporite S. While evaporite S as an additional agricultural input to vineyards could contribute to our enriched S values, it could not fully explain the pattern we observed across multiple vineyards (supplementary figure 4), further reinforcing that MSR could account for additional enrichment. Changes to δ³⁴S–SO₄ ²⁻ values from MSR could result in an overestimation of evaporite S contributions to the global S cycle if MSR affects δ³⁴S values more broadly. Overall, comparing our measurements to global values reinforces that intensive agricultural S additions can significantly alter the δ³⁴S–SO₄ ²⁻ signature of tributaries and rivers, affecting S source flux attributions and revealing the importance of delving into microbial dynamics when interpreting the global river δ³⁴S pattern.

This study provides the first evidence that intensive agricultural S applications change the biogeochemical fingerprint of S at field-to-watershed scales in an upland, mixed LULC watershed. The dramatic difference between δ³⁴S–SO₄ ²⁻ values from vineyard agriculture and non-agricultural areas (9.9 ± 5.9‰, n = 55, vs −0.28 ± 5.7‰, n = 30, respectively; figure 2) provides clear and compelling evidence for an altered S cycle in agricultural areas. Furthermore, the persistence of the agricultural S fingerprint in the Napa River—very similar to that found in soil leachate from fields—suggests that intensive S applications alter the S cycle at watershed scales, despite their input to a minor proportion of the watershed area. Ultimately, this study demonstrates the potential to understand the modern S cascade in agricultural systems, which is critical to documenting and developing solutions to human manipulation of the S cycle more broadly.