Anion Exchange Capacity Explains Deep Soil Nitrate Accumulation in Brazilian Amazon Croplands

Deep tropical soils with net anion exchange capacity can adsorb nitrate and might delay the eutrophication of surface waters that is often associated with many temperate croplands. We investigated anion exchange capacity and soil nitrate pools in deep soils in the Southern Brazilian Amazon, where conversion of tropical forest and Cerrado to intensive fertilized soybean and soybean-maize cropping expanded rapidly in the 2000s. We found that mean soil nitrate pools in the top 8 m increased from 143 kg N ha−1 in forest to 1,052 in soybean and 1,161 kg N ha−1 in soybean-maize croplands. This nitrate accumulation in croplands aligned with the estimated N surpluses in the croplands. Soil anion exchange capacity explained the magnitude of nitrate accumulation. High nitrate retention in soils was consistent with current low levels of streamwater nitrate exported from croplands. Soil exchange sites were far from saturation, which suggests that nitrate accumulation can continue for longer under current cropping practices, although mechanisms such as competition with other anions and preferential water flowpaths that bypass exchange sites could reduce the time to saturation.


INTRODUCTION
Increased use of synthetic nitrogen (N) fertilizer, expanded global cropland area, and increased planting of N-fixing soybeans have dramatically increased world crop production since the middle of the twentieth century (Smil 1999). But expanded and intensified cropping that increasingly relies on application of reactive N now causes a cascading series of problems associated with excess releases of N into the environment (Galloway and others 2003). Across the most intensively-farmed regions of the Earth's temperate zones in Europe, China, and North America, N that leaches from cropland soils into ground and surface waters now contributes to the eutrophication of surface fresh waters and drives the formation of large coastal marine hypoxic zones (Rabalais 2002;Galloway and others 2003). Mechanisms that reduce or prevent N leaching from intensively cropped soils are of great interest because they can help to mitigate the environmental consequences of N applications to croplands.
In recent decades, higher population and food demand, limits to the potential for expanded cropland area in temperate regions, and increased global trade have expanded and intensified cropping in tropical regions (Laurance and others 2014). One of the largest of these regions is in the Cerrado and Amazon biomes of Brazil. From 2003 to 2013, the area of cropland in the Cerrado and Amazon region of Brazil more than doubled from 1.2 to 2.5 million ha, and cropping practices over most of the region shifted from single-cropping of soybeans to double cropping of soybeans and maize in the same year (Spera and others 2016). These changes to cropping have increased N fertilizer use; because while the soybean crop required little or no N fertilizer, N fertilizer is needed for maize double cropping. Between 2000 and 2018, Brazil's consumption of N fertilizer tripled from 1.7 million to 5.1 million tons (http://faostat.fao.org/). Expanded and intensified cropping in the Cerrado and Amazon regions fueled Brazil's rise to become the world's leading exporter of soybeans and secondlargest exporter of maize by 2018 (http://faostat.fa o.org/).
The soils that underlie the Cerrado and Amazon regions of Brazil differ in important ways from the soils underlying most intensive crop agriculture in temperate regions. These differences have the potential to influence N leaching from soils and produce different environmental consequences of N compared with those from temperate croplands. Large areas of temperate cropland soils contain predominantly 2:1 clays that create permanent negative soil charge and cation exchange capacity that retains cations but allows rapid NO 3 movement. In the Cerrado and Amazon regions of Brazil, most new and intensified croplands lie on highlyweathered Oxisols (Figure 1) made up largely of variable-charge kaolinite and iron and aluminum sesquioxide clays that have predominantly positive charge (net anion exchange capacity) in the subsoil (Uehara and Gillman 1985;West and others 2010). This anion exchange capacity can hold large amounts of NO 3 per volume of soil and prevent or delay NO 3 leaching (Uehara and Gillman 1985;Wong and others 1990;Rasiah and Armour 2001;Lehmann and others 2006;Warren and Kihanda 2006;Feldpausch and others 2009). The great depths of weathered tropical Oxisols (Sanchez 2019) can enhance this effect.
There is new evidence that N leaching from intensive soybean and maize cropping on Brazilian Oxisols is minimal. Studies of the fate of N from intensive croplands in Mato Grosso, Brazil found low rates of inorganic N export to streams of 2 kg N ha -1 y -1 (Riskin and others 2017) despite an N surplus from the cropping systems of 20-40 kg N ha -1 y -1 (Figueira and others 2016; Jankowski and others 2018). A small portion of this imbalance may be explained by gas losses, but a large majority of added N is stored between 2 and 6 m in soil profiles, and soil NO 3 accumulation accounts for nearly all of the N surplus even at N fertilizer application rates up to 200 kg N ha -1 y -1 Figure 1. Map of soybean and maize cropping and Oxisol soil distribution in Brazil. Data on soybean single-cropping and soybean-maize double cropping are from (Spera and others 2016). Oxisols data come from (Dias 2015), derived from the equivalent Brazilian soil classification, Latossolos. The red outline is our study site, Tanguro Ranch. (Jankowski and others 2018). Large pools of soil NO 3 raise questions about the storage mechanism and the ultimate storage capacity in croplands.
We quantified the storage of NO 3 -, NH 4 + , total carbon (C), and total N to 8 m across multiple soybean-only croplands, soybean-maize croplands, and native forests at Tanguro Ranch, a large working soybean farm in Mato Grosso, Brazil. We tested the hypothesis that anion exchange capacity and the large volume of soil explain the pattern of soil NO 3 storage in a way that is consistent with low observed NO 3 leaching. We then addressed how rapidly NO 3 is accumulating in each cropping system and for how long NO 3 might continue to accumulate.

Study Site
We conducted this study at Tanguro Ranch, a commercial soybean and soybean-maize farm with remnant intact tropical forest near the border of the Cerrado savanna and Amazon forest biomes in Eastern Mato Grosso, Brazil. In the early 1980s, approximately half of the ranch's 80,000 ha were deforested for cattle pasture. In the early 2000s, the cattle pastures were converted to soybean fields, and since 2008 parts of the ranch have been converted to soybean and maize double cropping where a shorter-season variety of soybean is grown in the first part of the wet season, harvested, and then maize is planted and harvested in the second half (Spera and others 2016). Each year, maize plants are fertilized with a total of 80 kg N ha -1 , 5 kg N ha -1 drilled at seeding and 75 kg N ha -1 broadcast at the seedling stage. Lime is applied every 1-2 years to maintain surface soil pH above 5.5. Phosphorus at about 38 kg P ha -1 y -1 is applied as rock PO 4 3-, and potassium (K) at 30 kg K ha -1 y -1 is applied as potassium chloride (KCl).
The climate, topography, soils, and cropping system at Tanguro Ranch are representative of similar farms in the region where soybean and soybean-maize double cropping have expanded rapidly. The ranch lies 320-390 m above sea level on the Brazilian Shield on Precambrian gneisses of the Xingu Complex with deep (exceeding 10 m), acidic, Oxisol (Haplustox) soils with a sandy clay texture ( Figure 2). The mean soil textures are 38.3% clay, 11.3% silt, and 50.4% sand from 0-8 m across forests and cropping sites (Table S1), as measured using the hydrometer method for particle size analysis (Bouyoucos 1962). The mean an-nual temperature is 27°C, and mean annual rainfall is 1,700 mm (Neill and others 2013) with a strong dry season from May to September in which total precipitation for the five-month period averages 80 mm (Figueira and others 2016). The native, primary forest is evergreen tropical forest typical of the Southern Amazon (Ivanauskas and others 2004).

Experimental Design
We sampled 12 sites; four replicates within three land uses: forest, soy single-cropping, and soymaize double cropping. At each site, we sampled soils from 0-8 m with soil augers (10 cm diameter) and homogenized and sub-sampled in one-meter increments (although we were not able to auger past 6 m at one forest site after trying in two different locations). We did not collect bulk density from soil samples collected by auger, but matched these samples with bulk density measurements from density collected from nearby soil pits in forest and croplands at Tanguro. We used the nearest pit to each of our sites to calculate bulk density for that profile. Bulk density from the pits was measured using the volumetric core method (Grossman and Reinsch 2018), where samples were collected by placing the rings against soil pit walls, measured in triplicate, and averaged. To obtain concentrations of extractable soil NH 4 + and NO 3 -, we extracted 5 g of field-moist soil in 30 ml 2 M potassium chloride (KCl) for approximately 24 h, filtered them through pre-rinsed Whatman #1 filter paper, and froze them for later analysis on a SmartChem 170 discrete analyzer (Unity Scientific, Milford, MA). We dried a subsample of each soil at 105°C to a constant weight to determine the gravimetric water content. We then air-dried the remainder of the soil samples for further analyses.
We crushed the air-dried soils with a mortar and pestle and sieved to < 2 mm to measure pH and anion exchange capacity. To measure pH, we used a 5 g sample of each depth using a 1:1 soil/solution ratio in solutions of deionized (DI) water and 1 M KCl. Samples were shaken for 1 h and allowed to equilibrate for 30 min before pH measurement.
To measure anion exchange capacity, we used the compulsive exchange method, which controls for pH and ionic strength and is suitable for clay samples with variable-charge minerals (Laird and Fleming 2008). First, soils were saturated with barium chloride to saturate soil anion exchange sites with Clions. Then, we added a low ionic strength magnesium sulfate to exchange the Clions with SO 4 2-. The amount of Clreleased was measured to determine anion exchange capacity. Duplicate extractions on subsamples from each site and depth were conducted at pH 5.4 ± 0.1 and conductivity of 355 lS ± 30 lS. We filtered samples with Whatman #1 filter paper and analyzed Clsamples on a Dionex Integrion HPIC ion chromatograph (San Jose, CA). Because the pH must be held constant to accurately measure anion exchange capacity in variable-charge soils (Gillman 1979;Laird and Fleming 2008), small differences driven by variation in soil pH may be masked in the anion exchange measurements, so we also calculated delta pH values (the difference between soil pH measured in 1 M KCl minus pH in deionized water) as a separate indication of soil charge. We measured bulk soil C and N by combustion on a Vario MAX Cube organic elemental analyzer (Elementar, Ronkonkoma, NY).

Calculations and Statistical Analyses
We used our measurements of anion exchange capacity and extractable soil NO 3 to estimate an upper bound of soil NO 3 --N sorption capacity. We converted the anion exchange capacity and bulk density measurements to predict maximum NO 3 sorption in the soil in kg N ha -1 m soil -1 , assuming that all anion exchange sites were filled with NO 3 ions. Then, we compared our actual measurements of NO 3 concentrations through the soil profile with the predicted maximum.
where N measured is measured soil NO 3 --N (kg N ha -1 m soil -1 ), and N max is the maximum soil NO 3 --N adsorption estimate (kg N ha -1 m soil -1 ) based on the soil's anion exchange capacity. We multiplied the soil N saturation (%) by a factor of 2 under the assumption that roughly half of anion exchange sites are occupied by other anions and so are not available for nitrate ions. Given that this is a simple multiplier, it is straightforward to assess how different factors would influence the results.
All analyses and figures were done in R v4.0.3 (R Core Team 2020). We used ANOVA and least square means post-hoc comparisons (with Tukey Pvalue adjustments) to test for land use differences. We log-transformed total soil NO 3 concentrations to normalize the data for an ANOVA test (Figure 3c, Table S2). We used a linear regression on total NO 3 accumulated versus time since each site was deforested (Table S3).

Soil exchangeable NO 3
concentrations to 8 m ( Figure 3a) were much larger than soil exchangeable NH 4 + concentrations (Figure 3b) in forest and croplands. Total soil exchangeable NO 3 to 8 m was nearly an order of magnitude higher in the cropping systems (396-2,830 kg N ha -1 , means of 1,052-1,161 kg N ha -1 ) compared with forest (36-314 kg N ha -1 , mean of 143 kg N ha -1 ) (Figure 2c, Table S4).
Anion exchange capacity varied from 0.03 to 4.09 cmol c kg soil -1 , with a mean of 1.26 cmol c kg soil -1 (Table S6). Anion exchange capacity increased with depth between 0 and 3 m but remained relatively constant between 3-8 m (Figure 4a). Anion exchange capacity was similar among land uses for the most part, although it was lower in forest than in soybean-maize double cropped sites at two depth intervals (2-3 m and 7-8 m; Figure 4a, Table S5). Soil texture fractions were also similar among land use types, with the clay fraction ranging from 21.9%-53.2% (Table S1). The clay fraction is likely underestimated due to incomplete disaggregation of larger particles during the particle size analysis, with some clay particles still included in larger sized aggregates (Donagemma and others 2003).
Anion exchange capacity was more than sufficient to explain the pools of exchangeable NO 3 -. We multiplied anion exchange values with the total mass of soil to estimate the potential NO 3 sorption capacity based on the soil charge. The NO 3 --N sorption capacity summed from 0-8 m was 9,234-44,860 with a median of 32,420 kg N ha -1 (Figure 4b).
To estimate the proportion of potential NO 3 exchange sites that are currently occupied by Anion Exchange Capacity Explains Deep Soil Nitrate Accumulation NO 3 -, we divided our NO 3 pool size measurements by this NO 3 sorption capacity estimate. If all anion exchange sites could be occupied by NO 3 -, then these soils were 0-41.9% saturated with NO 3 -. Some of the exchange sites must be occupied with other anions, so overall NO 3 saturation is somewhat higher. If we assume that NO 3 could fill half of available anion exchange sites, soil NO 3 saturation ranged from 0-83.7% for individual replicates, with a median of 2.5% (Figure 4c, Table S6), indicating that the soil has many more exchange sites available for NO 3 sorption. There were significant differences in the soil N saturation estimate among land uses between 2-4 m depth: the forest was lower than soybean-maize between 2-3 m and lower than both cropping systems between 3-4 m (Table S9). Even in the cropping systems with large existing NO 3 pools, deep soils still had a large capacity to accumulate additional nitrate.
Soil pH in 1 M KCl varied from 3.9-6.6 with a mean of 5.2 (Figure 5a). Soil pH in deionized water varied from 3.9-7.2 with a mean of 5.3 (Figure 5b, Table S6). Soil pH in KCl and water was higher in the top 1 m in the maize double cropped sites compared with the forest (Table S7, S8), but this difference generally diminished below 2 m (Table S7). More than 75% of DpH (pH KCl -pH H2O ) values ( Figure 5c) were above -0.5, which indicates the presence of variable-charged clays that produce net anion exchange capacity (Uehara and Gillman 1981). DpH was highest between 2-5 m in the cropping system, which was generally the depth at which the largest NO 3 accumulations occurred (Figure 5c).
Soil N ( Figure 6a) and C (Figure 6b) concentrations decreased sharply between 0-1 and 1-2 m but varied much less below 2 m. There were no significant differences in soil total N concentrations among land uses or within depth increments (Table S12). Although soil C concentrations were significantly different among land uses across all depths, they were not significantly different within any depth specific increments (Table S13). We found no statistically significant differences in total soil N (Table S14) or total soil C (Table S15), though median bulk soil N increased from 16 Mg N ha -1 in forests to 18 Mg N ha -1 in soybean-maize double cropped sites ( Figure 6).  a, b, and c), which are jittered vertically in (a,b) and horizontally in (c). Significant differences in total soil NO 3 from ANOVA and a Tukey's honestly significant differences test on log 10 -transformed total 0-8 m soil NO 3 data (Table S2)

are indicated by different letters (c).
Extractable soil NO 3 was $ 0.9% of total N in the forest and $ 6.5% of total N in the croplands.
We examined the relationships between soil NO 3 pools and time since land was converted from forest to cropland ( Figure S2) using a statistical fit (Table S3). We estimated the rate of past NO 3 accumulation by regressing total soil NO 3 against the time since conversion from forest to pasture ( Figure S2, Table S3). Based on these measurements, soil NO 3 accumulated at 27 kg N ha -1 yr -1 (R 2 = 0.37, P = 0.02, Figure S2, Table S3) in the croplands.   2009), and in sugarcane croplands on deep, highlyweathered Oxisols in tropical Australia (Rasiah and others 2003). Furthermore, greater accumulation of NO 3 in croplands compared with forest is not unique to our study; it is consistent with a previous report from one forest location and one crop field at Tanguro (Jankowski and others 2018). In addition to showing up in direct measurements, large NO 3 retention in this landscape is consistent with previous observations of other N cycling metrics. Other studies at this site have observed rapid and nearly complete nitrification of ammonium or urea fertilizers applied to surface soils (Jankowski and others 2018), low NO 3 concentrations in soil solution (Jankowski and others 2018), low NO 3 exports from forests, and no increase in NO 3 exports from croplands (Riskin and others 2017).

DISCUSSION
After confirming these trends in deep NO 3 storage, we tested our hypothesis about the proximate cause: that anion exchange capacity and the large volume of soil explain the pattern of soil NO 3 storage. Our measurements of anion exchange capacity and DpH (pH KCl -pH H2O ) data (Figures 4, 5) both suggest that net anion exchange capacity is the mechanism for nitrate retention. DpH values above -0.5 indicate the presence of variable-charge clays, which can have net anion exchange capacity (Uehara and Gillman 1981).
The positive DpH values in Oxisols and Ultisols may also indicate low organic matter (Mekaru and Uehara 1972) or KCl-extractable aluminum in the soil (Oades and others 1989). Higher pH at the surface of cropping sites is to be expected from liming (Figure 5a, b). Conversion from forest with deep roots to croplands with shallower roots may Figure 6. Soil total N (a) and total C (b) concentrations (mg N g soil -1 ) by soil depth, and summed soil total N from 0-8 m (Mg N ha -1 ) (c), and summed soil total C from 0-8 m (Mg C ha -1 ) (d). The different colors correspond to different land uses. The lines connect the medians (triangles) in (a, b) within each land use and depth. There were four replicate sites per land use (circles) (a, b, c, and d), which are jittered vertically in (a,b). There were significant differences among depths but not land uses in a two-factor ANOVA on total soil N and C (a, b, Tables S12, S13), but no significant differences in an ANOVA on total soil N or C summed from 0-8 m by land use (c, d).
have resulted in losses of soil organic matter because of the loss of root carbon inputs in the subsoil (Trumbore and others 1995;Davidson and others 2011), reducing N mineralization of soil organic matter. And increased soil organic matter has been shown to lower the pH at which a soil has no net charge, effectively lowering anion exchange capacity (Sanchez 2019). Thus, the loss of soil organic matter may effectively increase anion exchange capacity and the DpH.

Accumulation and Patterns
Several potential mechanisms apart from anion exchange capacity may contribute to NO 3 accumulations in deep soils and the differences in NO 3 patterns that we observed across forest and croplands. These are not exclusive of each other, and all could potentially contribute to the observed patterns. The first mechanism is a within-soil redistribution of NO 3 in the cropping systems following the initial clearing of the forest from decay of the cleared forest biomass at the surface. Based on bulk density (Riskin and others 2013) and total soil N stocks (Figueira and others 2016; Wong and others 2019) from intact forest and soybean single cropped sites measured in 2009, we estimate that the total soil N stock in the top 10 cm of 2,362 kg N ha -1 in the forest declined to 1,271 kg N ha -1 in soybean fields, demonstrating that about 1,000 kg N ha -1 may have been mineralized from the soil surface and redistributed deeper during the land use transitions from forest to pasture and cropping. Removal of deep NO 3 by plant uptake likely occurs in the forest because tree roots extend 8 m deep in the seasonally-dry but evergreen Amazon forest (Nepstad and others 1994). This recycling of deep soil N would occur to a much less extent in croplands because soybean and maize roots only extend to about 2 m (Feldman 1994;Battisti and Sentelhas 2017).
Subsequent turnover of N in dead root biomass or loss of new organic inputs from roots to the deep soil could also affect deep soil NO 3 pools. Decay of fine roots from the original forest based on fine root biomass of 1 mg cm -3 in deep soil (Nepstad and others 1994) and a fine root C:N ratio of 30 would likely only contribute another < 20 kg N ha -1 ). We found a slight trend of increasing median bulk soil N from forests to soybean-maize double cropped sites (Figure 6c), but we do not have enough statistical power to demonstrate that land use change resulted in significant changes to soil N or C stocks-and therefore potential N mineralization fluxes. However, small changes in total soil N stocks could have important implications for NO 3 redistribution at these sites because the total soil N pool is roughly one to two orders of magnitude larger than the exchangeable NO 3 pool. Changes in gaseous losses of N such as N 2, NO, and N 2 O after land use conversion or fertilization could also have important implications for the NO 3 budget of tropical soils (Davidson and others 2007), especially for fertilized tropical cropland soils (Huddell and others 2020). There are some known changes to gaseous N losses with land use change at this site, but these differences are not large enough to account for the large differences in NO 3 accumulation. N 2 losses of NO 3 via denitrification may be substantial: unpublished estimates of N 2 losses ranged from 10 to 20 kg N ha -1 y -1 for croplands and ranged from 2 to 12 kg N ha -1 y -1 for forests at this site (Fox and others 2017). However, a recent analysis of NO and N 2 O emissions found small losses of only $ 1-2 kg N ha -1 y -1 at this site, with no significant differences in annual emissions across land uses under current fertilization and management practices (Huddell and others 2021).
Some NO 3 that accumulates at depth likely comes from N inputs to croplands. We estimated soil NO 3 accumulation rates of 27 kg N ha -1 y -1 (R 2 = 0.37, P = 0.02, Figure S2, Table S3), which is similar to what a partial N budget based on empirical measurements at Tanguro would predict: N fixation in the soybean crop of 227 kg N ha -1 y -1 (Figueira and others 2016), maize fertilization rates of 80 kg N ha -1 y -1 , harvest removal of soybean N (184 kg N ha -1 y -1 ) and maize N (104 kg N ha -1 y -1 ) (Jankowski and others 2018), annual biological N fixation rates in the forest of < 1.2 kg N ha -1 y -1 (Wong and others 2019), and N surpluses of 43 kg N ha -1 y -1 in soybean single cropped sites and 19 kg N ha -1 y -1 in soybean-maize double cropped sites. Overall, this indicated that NO 3 has accumulated in deep soils of the cropping systems at a rate that aligned closely with expectations based on crop N budgets.
Finally, more water moving through soil profiles in cropland compared with forest may influence NO 3 distribution by moving NO 3 down to the deep soil, where it can adsorb to exchange sites. Forest clearing reduces evapotranspiration from approximately 1,095 mm y -1 in forest watersheds to 625 mm y -1 in soybean watersheds and results in much larger annual stream discharges in cropped watersheds, about 614 mm y -1 , compared with 141 mm y -1 in forest watersheds (Riskin and others 2017). These differences in water balance indi-Anion Exchange Capacity Explains Deep Soil Nitrate Accumulation cate that about four times more water flows through the surface and subsurface soils of croplands compared with forest. Rates of soil infiltration and soil hydraulic conductivity in both forest and croplands exceed maximum rainfall intensities and generally preclude production of horizontal water flowpaths that would bypass water movement through deep soils (Scheffler and others 2011). However, hydraulic conductivity in the top meter is lower in croplands than in forest (Scheffler and others 2011) and hydraulic conductivity declines below 2 m in both forest and croplands (K. Jahn, personal communication). Slower water movement in croplands may allow for more exchange of NO 3 on soil exchange sites and aligns with our observation of increased exchangeable NO 3 below 2 m (Figure 3a).

Future NO 3 -Storage: Maximum Sorption and Caveats to Estimates
Our storage capacity calculations indicated that the amount of anion exchange capacity was more than sufficient to store the pools of NO 3 we observed. We estimated an upper bound on NO 3 sorption capacity based on the soil charge by combining anion exchange capacity measurements with the total mass of soil. This calculation showed that the maximum NO 3 --N sorption capacity summed from 0-8 m across all land uses ranged from 9,234-44,860, with a median of 32,420 kg N ha -1 (Figure 3b). In reality, actual levels of NO 3 sorption would be limited by competition with other anions such as PO 4 3-, SO 4 2-, and Cl -. Of these anions, PO 4 3binds most strongly, so it may occupy exchange sites near the soil surface and displacing other anions deeper (Donn and Menzies 2005). A study at this site found that PO 4 3from fertilization (applied at 38 kg P ha -1 y -1 ) does not move below 20 cm (Riskin and others 2013). There are no studies of SO 4 2or Clin our sites, but in studies on Oxisols in tropical Australia, large SO 4 2concentrations near the surface from sea salt accession displaced or excluded NO 3 and Clfrom shallow exchange sites and caused NO 3 and Clto leach deeper over time (Donn and others 2004). The maximum concentrations of 2.65 mmol SO 4 2kg -1 soil measured between 80-100 cm from some Brazilian Oxisols (Barreal and others 2003) are similar to our maximum of 2.56 mmol NO 3 kg -1 soil. Using the annual N surplus estimate of 19 kg N ha -1 y -1 from a partial N budget of soybean-maize double cropped sites, we estimated that N can continue to accumulate for roughly 341 years (median; the range across replicates was 80-570 years) before the anion exchange sites are saturated. However, some competing effects may reduce this timeline. Preferential flow paths through soil macropores or increases in overland flow caused by soil compaction that results from continuous cropping could allow NO 3 to bypass much of the bulk soil and effectively reduce the contact of NO 3 with available exchange sites for sorption (Morales and others 2010). Although current cropping practices and the near-exclusive use of minimum tillage in the Amazon-Cerrado cropping region minimize compaction (Neill and Macedo 2016) and maintain for the currently high surface infiltration rates (Scheffler and others 2011), compaction might increase in future. Also, higher soil organic matter content along preferential flow paths (Bundt and others 2001) or an increase in soil pH caused by leaching of lime along those paths could also reduce the effective NO 3 sorption capacity if soil organic matter reduces anion exchange capacity. However, preferential flows generally only occur in the top $ 2 m at Tanguro (personal communication, K. Jahn). Finally, anion exchange capacity was measured on soils that had been ground and shaken in solution. Both of these processes could disrupt the strong soil aggregates that give Oxisols strong structure and drainage capacity (Sanchez 2019) and expose more anion exchange sites in the laboratory measurements than would be exposed to soil solution in situ.

CONCLUSIONS AND IMPLICATIONS
We found that NO 3 naturally accumulates in deep soils of the native forests, that conversion to agriculture increases the deep NO 3 pool significantly, and that the anion exchange capacity of these soils is more than enough to account for the NO 3 storage. The deep soil accumulation of NO 3 in croplands is consistent with low levels of NO 3 in soil solution and NO 3 currently exported from watersheds in streams draining both forest and croplands. We also found that our crop N budget aligns well with the rate of accumulation in cropping systems, and at the current rate of accumulation, NO 3 could continue to be stored in deep soil in this region for about 340 years. Time lags in NO 3 arrival to streams caused by long groundwater residence times (Wang and others 2013) might delay the leaching of NO 3 to surface waters, but cannot completely explain low groundwater and streamwater NO 3 concentrations because croplands lie adjacent to streams in most cropland watersheds. Agriculture has expanded in the state of Mato Grosso, Brazil to 58,000 km 2 since 2001 (Spera and others 2014), catapulting Brazil to become the world's leading exporter of soybeans and secondlargest exporter of maize by 2018 (http://faostat.fa o.org/). Because most of that expansion has occurred on deep, highly-weathered soils our findings suggest that NO 3 storage in deep soils will delay for a number of decades the eutrophication (Carpenter and others 1998) and degradation of water quality (Rabalais and others 2002;Galloway and others 2003) found in many temperate cropping systems. However, identifying factors that will influence the magnitude of deep soil NO 3 retention, such as competition for soil anion exchange sites with other anions, preferential flow paths that bypass exchange sites in the soil matrix, and artifacts of laboratory measurements of anion exchange, will be critical to accurately predicting the timing and magnitude of NO 3 movement from croplands in this expanding cropland region.