Nitrate leaching losses from wheat-based cropping systems are projected to increase significantly by 2050 in the Northern Great Plains, a region where nitrogen fertilizer is generally the largest input cost for producers. Despite this, a system for determining areas of fields that are more or less susceptible to nitrogen loss has not been established. Maximum likelihood classification and dimension reductionality techniques are being applied to satellite and drone imagery to investigate spatial variability in soil nitrate at spring planting, with the goal of predicting optimal nitrogen fertilizer rates for spring and winter wheat at the subfield scale. Factors known to affect overwinter soil nitrate change, including soil depth, bulk density, and overwinter CO2 flux, as well as aboveground biomass, grain yield, and harvest index of the previous crop, have been mapped at 5-cm resolution using imagery from a single drone flight in 2017 at one location in central Montana. Mapping efforts based on satellite imagery of coarser spatial (1 m) and finer temporal (biennial) resolutions are ongoing at five commercial farms in the region. A 2019 survey suggested many large operators in this area were unfamiliar with (32%) or directly challenged (24%) current climate projections. More than 40% of operators reported soil sampling every other year or less often, and 22% reported not sampling at all. Together, these results suggest central Montana farmers may be more vulnerable to economic loss from over- or under-fertilization, especially if precipitation, temperature, and nitrate leaching patterns change as predicted. This research has potential to 1) increase nitrogen use efficiency, 2) reduce suboptimal yield/protein associated with under-fertilization, 3) increase return on investment of precision agriculture technologies (i.e., variable rate fertilizer application), and 4) mitigate impacts of changing weather patterns for farming systems in the Northern Great Plains.
It is known that freeze-thaw events can lead to biophysical changes in soil that affect processes like nitrogen (N) mineralization (Roth & Fox, 1990; Heaney et al., 1992; McCracken et al., 1994; Ryan et al., 2000), denitrification (Christensen & Tiedje, 1990; Heaney et al., 1992; Burton & Beauchamp, 1994; Rover et al., 1998) and leaching (Campbell et al., 1984; Liang et al., 1991), which in turn affect the magnitude and direction of overwinter soil nitrate changes (ONCs). Most studies have documented net positive ONCs (McCracken et al., 1994; Mitchell et al., 1996; Jacinthe et al., 2002), with some notable exceptions. Jokela et al. (1992) reported net negative ONCs, though spring sampling occurred after plant uptake was underway, which likely skewed the results. Losses >10 kg ha -1 were reported after two particularly wet winters on the Moccasin Terrace in central Montana (Chen, unpub., Miller unpub., Jones et al., 2011). ONCs ranged from -26 to 68 kg N ha -1 and averaged 18 kg N ha -1 based on this three-year study (Jones et al., 2011). An average ONC of 18 kg N ha -1 suggests that the 65% of farmers who soil sample less than once per year in this area may be more likely to over-fertilize their crops. This fact may help explain the rising groundwater nitrate concentrations in the area (Schmidt & Mulder, 2010; Miller, 2013; John et al., 2017, Sigler et al., 2018). If ONCs are not accounted for, over- or under-fertilization can occur when growing wheat, barley, and other crops. This is true whether fertilizing to past years’ soil test results (e.g., based on long-term fertilizer-protein or fertilizer-yield curves) or to fall test results. Here, we define over-fertilization as fertilizer applied in excess of the recommended rate after accounting for ONC, where ONC is simply Spring Test N (kg N ha-1) minus Fall Test N (kg N ha-1):
Eq. 1 ONC = Spring Test N – Fall Test N
Over- and under-fertilization occur at positive and negative values of Δ Fertilizer:
Eq. 2 Δ Fertilizer = Applied – Recommended
where Applied is applied fertilizer (kg N ha-1), given by cooperating farmer records, and Recommended is the recommended fertilizer rate (kg N ha-1), given by:
Eq. 3 Recommended = State Guideline * Yield Target – ONC – Fall Test N
where State Guideline is the yield-normalized available N recommendation for the state of Montana (0.055 and 0.045 kg N kg-1 for winter wheat and spring wheat, respectively) and Yield Target is the grower’s target yield (kg ha-1). Eq. 3 simplifies to:
Eq. 4 Recommended = State Guideline * Yield Target – Spring Test N
Eq. 5 Δ Fertilizer = Applied – State Guideline * Yield Target + Spring Test N
Given Applied, State Guideline, and Yield Target obtained from producer records, subfield scale estimates of Δ Fertilizer will be obtained. Note that by setting Δ Fertilizer to zero and solving for Applied, recommended N fertilizer rate can be estimated on a pixel-to-pixel basis prior to spring planting.
N surpluses associated with over-fertilization can contribute to groundwater nitrate pollution, while both surpluses and deficits represent economic loss to the farmer, either through unutilized inputs or sub-optimal yield and protein. Because of the large ONCs observed in previous work (Jones et al., 2011), it is recommended that farmers and crop advisors conduct soil sampling in spring versus fall (Jones & Olson-Rutz, 2019). However, springtime soil sampling is often impractical due to a combination of factors, including field inaccessibility and heavy springtime workloads for farmers and ag professionals. Even if spring soil sampling is successful, test results can lead to added stress, as fertilizer may not be available or a farmer may feel pressured into making a last-minute financial decision (i.e., a large fertilizer purchase) for which he or she may not have budgeted. Furthermore, ONC variability could increase if Montana winters/springs become wetter and warmer, as predicted by studies highlighted in Whitlock et al. (2017). These changes could lead to greater nitrate exports caused by higher leaching rates in the Northern Great Plains (He et al., 2018). Increased mineralization rates from warmer conditions could also lead to greater N increases from fall to spring.
A 15 km transect of the Moccasin Terrace defines the region of interest in the current study. This landform is a 260-km2 strath terrace with shallow water tables (1-10 m) and soils with high infiltration and low storage capacity overlying a conductive gravel layer (Sigler et al., 2018) which in some locations is only 10 cm below the soil surface. Because of higher observed leaching rates where gravel contacts on this landform are shallow (i.e., areas of shallow depth to gravel; DTG), John et al. (2017) suggested that these areas be targeted for fallow replacement and other management practices that might reduce the potential for deep percolation. In fact, the Moccasin Terrace has been the focus of extensive environmental, hydrogeologic, and agronomic research (Schmidt & Mulder, 2010; Jones et al., 2011; Miller, 2013; John et al., 2017; Sigler et al., 2018) which has laid the groundwork for well-informed predictions about the magnitude and direction of ONCs on specific areas of the landform, including those of varying soil water and groundwater storage. For example, Sigler et al. (2018) hypothesized that 1) thicker soils (i.e., soils in areas of deep DTG with high soil water storage capacity) may allow more infiltrated precipitation to remain in the root zone where it is subject to evapotranspiration rather than deep percolation/leaching, and 2) areas of shallow depth to shale (DTS) have low groundwater storage capacity and therefore soils may be subject to saturation excess overland flow rather than deep percolation/leaching.Therefore, low leaching rates and high spring N concentrations are expected in areas of shallow DTS and deep DTG, due to the combined effects of increased overland flow (low groundwater storage) and longer soil water residence times (high soil water storage). Conversely, increased deep-percolation (high groundwater storage) and shorter soil water residence times (lower soil water storage) are expected to lead to high leaching rates and low spring nitrate concentrations in areas of deep DTS and shallow DTG. Losses to denitrification in saturated, anoxic soils of low-DTS areas and high-DTG areas are likely greater than in high-DTS, low-DTG areas. This would lead to inflated leaching estimates if whole-profile nitrogen cycling dynamics were investigated. This experiment relies on the assumption that most denitrification occurs in subsoils. Therefore, soil nitrate investigations were limited to the uppermost 20 cm of the soil profile.
In this report, the terms ‘soil depth’ and ‘depth to gravel’ are used interchangeably. Generally, these variables are positively correlated with soil moisture storage capacity and negatively correlated with leaching (John et al., 2017). Thicker soils (i.e., soils in areas of deep DTG with high soil moisture storage capacity) may allow more infiltrated precipitation to remain in the root zone where it is subject to evapotranspiration rather than deep percolation/leaching (Sigler et al., 2018). Soil depth is a predictor (P<0.01) of whole-profile nitrate inventory (Jones et al., 2011), which is not especially surprising since deeper soils are more voluminous and are less prone to leaching losses. Perhaps more surprising is the significant positive relationship observed between soil depth and 0-20 cm soil nitrate concentrations in spring. This relationship speaks to the degree of connectivity between surface and groundwater and highlights the importance of nitrate leaching losses on this landform. A 60-cm increase in DTG was associated with a predicted 17.4 ± 8.7 kg/ha decline in nitrate leaching rate (P < 0.05) across years and cropping systems on the Moccasin Terrace (John et al., 2017). Sigler et al. (in submission) showed that Normalized Difference Vegetation Index of spring pea based on satellite imagery acquired at 50% crop senescence is linearly related to DTG at soil depths less than 50 cm. However, timing of 50% crop senescence can be difficult to assess and image acquisition at this exact growth stage and crop rotational phase unlikely.
Here, we explored whether principle component analysis and un/supervised classification of multi-year, 1-m satellite imagery and single-year, 5-cm drone imagery can accurately predict DTG when poor mapping conditions (i.e., fallow phases of the crop rotation and post-harvest conditions) exist at image acquisition for one or more years of the image collection. The overall goal of this research is to develop a high resolution geospatial model for predicting Spring Test N as a decision support tool for nitrogen fertilizer rate adjustments at farm, field, and subfield scales in the context of changing precipitation and temperature patterns. This research has the potential to increase nitrogen use efficiency, reduce suboptimal yield/protein associated with under-fertilization, increase return on investment of precision agriculture technologies (i.e., variable rate fertilizer application), and mitigate impacts of changing weather patterns for farming systems in the Northern Great Plains.
1. Quantify subfield-scale variability in ONCs toward improved precision ag methods such as
variable rate fertilizer application.
2. Develop a model to predict ONCs and to notify farmers when large ONCs are likely to have
occurred in order to minimize economic and environmental injuries resulting from uninformed
fertilizer management decisions.
3. Demonstrate to farmers the economic and environmental benefits of soil testing and the
implications of changing precipitation and temperature patterns for N management practices.
Observations from the current study can be grouped into two types: time series (TS) and remote sensing (RS) data. Generally, TS data are those expected to vary by meaningful amounts over the course of the two-year study. TS data were collected at shorter sampling intervals and are thus characterized by greater temporal resolution. TS data typically were collected only at the subfield at the MSU Central Agricultural Research Center (CARC). CARC is sometimes referenced as the ‘on-station’ (versus off-station) site in this report. Volumetric water content, whole-profile CO2 efflux, and soil nitrate are examples of TS data collected at 15-min, 1-hr, and 1-mo intervals, respectively, throughout the winter months of 2019-2020 at CARC. RS data collection occurred at most twice per year and at a minimum once prior to study initiation. Rather than high temporal resolutions, these data are characterized by fine spatial resolutions. Gravel depth observations are examples of RS data collected at spatial resolutions from as coarse as 30 measurements ha-1 to as fine 105 measurements ha-1. Shale depth observations are examples of RS data collected prior to the current study at a 0.4 measurements km-2 resolution. Both TS and RS data types will be used as explanatory variables in generalized linear models (Model 1-3) for predicting soil N in spring (Spring Test N). At all off-station sites, Spring Test N and other soils data can be generally classed as RS rather than TS, observed at high spatial resolutions and low temporal ones.
On- and off-station subfields were chosen to fall out along a DTS (i.e., groundwater storage) gradient. Three off-station sites were seeded to winter wheat and two to spring wheat in the 2019 crop year. The on-station site was also cultivated to spring wheat in 2019. A subset of soil samples from this site will be included in off-station modelling efforts. DTS and DTG will be treated either as continuous or categorical variables in three linear mixed models (Models 1-3). A categorical approach will allow factor level replication within Previous Crop (2 levels: winter wheat and spring wheat), DTS (3 levels: Shallow, Moderate, and Deep), and DTG (2 levels: Shallow and Deep). These factors will be included as explanatory variables in a generalized linear model (Model 1) for predicting Spring Test N at the subfield scale in one year of the two-year study. A second linear regression model (Model 2) will include samples from both years with DTG, DTS, and Year as explanatory variables. In Model 2, effects of Previous Crop on Spring Test N will be ignored. Previous Crop effects on Spring Test N are known to be small relative to those of DTG (Jones et al., 2011). However, it is not known whether effects of Previous Crop are small relative to those of DTS. Results of Model 1 will provide insight into whether the exclusion of Previous Crop is justified. Model 3 will incorporate Spring Test N data from Jones et al. (2011) and John et al. (2017) to investigate across years. In addition to previous crop and DTG, Model 3 will include VWC and air temperature as explanatory variables, similar to the net nitrate production model developed by John et al. (2017), which we will attempt to validate in the current study.
Multiple ancillary and reference datasets were used in off-station site selection and in classification of on- and off-station subfields, including CropScape and verbal communication with cooperating farmers (i.e., for information regarding crop rotations), a shale surface digital elevation model of the landform (Sigler et al., 2018), meter and sub-meter imagery from multiple acquisition years (National Agricultural Imagery Program), as well as 5-cm multi-band imagery obtained via an unmanned aerial vehicle in fall 2017. Supervised classification and principal component eigenvector mapping were used to predict DTG at all subfields. A total of 12 fully randomized DTG sampling points per subfield served as reference data in training and test area selection for subsequent cross-validation accuracy assessments.
Off-station remote sensing
In the process of off-station site selection, priority was given to fields within the boundary of the Moccasin Terrace (Sigler et al., 2018) for which winter wheat or spring wheat was planned during the 2019 crop year. Priority was given to those fields in which high spectral variability in addition to uniform, cereal crop stands were observed in 2017 and/or 2015 NAIP imagery in order to ensure that a viable NDVI map could be generated. Furthermore, selection occurred such that distance between fields was minimized (ca. 15 km) while the range of predicted DTS values was maximized (ca. 1-20 m). This approach served the dual purpose of controlling for potential effects of weather differences among sites while increasing the likelihood of capturing true differences in ONC across DTS environments. Within each 20-80 ha field, a 0.4-ha subfield was targeted based on crop stand uniformity, spectral variability, and degree of subfield-to-field spectral representation at time of image acquisition, as determined by single-band histogram comparisons. DTG in each subfield was randomly sampled at 12 locations. A Giddings hydraulic soil probe retrofitted with a pressure gauge allowed for soil depth observation accuracy assessments using different instruments and different pressure thresholds. Depths at which a 3-cm soil corer and a modified Paul Brown probe achieve downward pressures of 500 and 150 psi, respectively, were averaged to obtain the DTG approximation. The 12 observation points served as reference data for training and test area selection prior to supervised maximum likelihood classification of NAIP imagery. Two DTG classes (i.e., shallow and deep) were assigned three test areas and three training areas, no less than 36 pixels each. This ensured both training and test areas encompassed a cumulative 108 pixels, satisfying the 100 pixel guideline put forth by Campbell and Wynne (2011). Because of the high variability in 1-m resolution imagery as well restrictions on reference data collection and analytical resources, we ignored the requirement for spectral uniformity of training and test areas prior to supervised maximum likelihood classification. One goal of this research is to automate to the highest possible degree NAIP imagery classification schemes for DTG prediction, and, as such, the requirement for spectral uniformity among training and test areas was deemed impractical.
On-station remote sensing
In summer 2017, 5-cm resolution, 4-band imagery (i.e., red, red edge, green, and near-infrared) of a 10-ha field of spring wheat at the Central Ag Research Center was obtained via an unmanned aerial vehicle. The field was planted back to spring wheat in 2018. We intended to re-image the spring wheat crop to obtain duplicate imagery in relatively wet conditions compared to the very dry conditions of 2017. These datasets would have allowed for non-destructive characterization of soils using binary image-change detection techniques outlined in Zipper et al. (2015). However, a drone malfunction in 2018 led to the exploration of alternative classification methods. Past work has shown that Normalized Difference Vegetation Index (NDVI) of spring wheat is a strong predictor of DTG on the Moccasin Terrace when image acquisition occurs at 50% crop senescence (Sigler et al., 2018). However, timing of 50% crop senescence can be difficult to assess and image acquisition at this exact growth stage is challenging. Thus, the viability of alternative classification techniques for predicting DTG on this landform was assessed using imagery acquired post-50% crop senescence, as well as post-harvest, by the National Agricultural Imagery Program (NAIP). DTG was randomly sampled at 42 points on a 0.4-ha subfield, selected for its high spectral variability, uniform crop stand at the time of image acquisition, distance from historical management boundaries, and accessibility. A random subset of 12 (of the original 42) DTG sampling points were used when comparing on- and off-station classification results. In order to find the minimum number of observations required to accurately capture DTG variability (i.e., support), we used cross-validation techniques with observed DTG values and predictions from classifiers discussed above at increasingly coarse image resolutions. These results informed whether DTG prediction was feasible at resolutions >3 m, offering insight into compatibility with precision ag technologies such as variable rate fertilizer application, the effective resolution of which is currently determined by equipment specifications. Moreover, this assessment heightened confidence in the methods of image classification implemented off-station.
Six CFLUX-1 soil respiration chambers were procured as part of a companion study investigating effects of crop rotation on nitrogen mineralization. To test whether soil depth impacted respiration rates, the chambers were deployed along a DTG gradient at the on-station site. Hourly CO2 efflux measurements have been made beginning in October 2019 and continuing to present day (February 2020). Instruments were typically disconnected from power during snowstorms and reconnected following the storm’s end, allowing for in-situ soil respiration measurements during multiple snowmelt events throughout the 2019-2020 winter.
Soil sampling and processing
Throughout the field data collection campaign, it became obvious that the soil sampling scheme employed in this experiment was problematic. Namely, equipment affectations caused by the frequent return to the same soil sampling locations in the field raised concerns about compaction and microtopographic disturbances related to stubble position and snow capture. To address these concerns, the decision was made to increase the number of RS samples collected by simply bracketing the study period with the most intensive sampling efforts. Therefore, 12 samples per location were collected in October 2018 and 36 samples per location are planned for study areas of a larger extent in April 2020. Two samples per site were collected in April of 2019 and two per site were planned for October 2019. However, frequent and unseasonably early snowstorms prevented sampling at off-station sites in fall of 2019. This missed event did not entirely compromise the experiment, given that soil N at spring planting is the variable of greatest interest in this study. It does however limit what can be said about ONCs. Furthermore, in order to avoid multiple, independent soil sample handling and subsampling events, the decision was made to delay chemical and biological assays until the end of the project, so that all samples from the experiment might be processed and analyzed at one time. Therefore, nitrate, ammonium, mineralization, and other soils data will not be available until spring 2020.
On-station soil sampling occurred in three field visits from October, 2018 to April 2019. A total of 5 visits from October 2019 to April 2020 were planned, with three of those having already occurred. Only the first and last visits of the on-station winter sampling effort included off-station sampling in 2018. A Giddings hydraulic soil probe was/will be used to collect 20-cm deep cores on- and off-station from randomly assigned locations within each subfield. A weighed amount of subsample will be removed from all bulk soil samples. Subsamples will be analyzed with cell viability, urease activity, and acid phosphatase assay kits at the Central Ag Research Center (CARC) Soil Microbiology Lab in Moccasin, MT. Aerobic incubations will be performed for additional estimates of mineralizable nitrogen (Vigil et al., 2002; McDonald et al., 2013). Bulk soil samples will be shipped to Ward Laboratories, Inc. for texturing and basic chemical analyses. Samples from the October 2018 trip will be analyzed for total nitrogen, total carbon, pH, buffer pH, soluble salts, organic matter, Olsen phosphorus, total phosphorus, cation exchange capacity, potassium, calcium, magnesium, sodium, and sulfate-sulfur. Samples from all subsequent will be analyzed only for nitrate-nitrogen and ammonium-nitrogen.
Principal Component Images
As noted, a modified Paul Brown probe was mounted to a Giddings hydraulic soil probe and the depth of the probe at 150 psi was recorded as a proxy for DTG. Simple linear regression analyses were performed with DTG and NDVI of spring wheat acquired at 50% (NDVI50) and 100% (NDVI100) crop senescence, as well as PC eigenvector image number 6 (PC6). Of all PC images, PC6 was the strongest predictor of DTG (RMSE = 10 cm; P < 0.01) and was visually similar to NDVI50, despite accounting for only 7% of the total variation in the 24-band image. Conversely, PC image number one was a relatively poor predictor of DTG (RMSE = 15 cm; P = 0.192), but explained nearly one-third of the variation in the data. Of all the linear regressions performed, NDVI50 was the strongest predictor of DTG (RMSE = 8 cm; P < 0.001) while NDVI100 exhibited the third strongest relationship (RMSE = 12 cm; P < 0.05), behind only PC6. These results suggest potential for PC analysis in the prediction of DTG on this landform. However, the high ‘potency’ or maximal information conveyance of the first PC images prevented accurate characterization of DTG relative to that of the ‘intermediately potent’ PC6.
Maximum likelihood classification
Results of maximum likelihood classification (MLC) revealed high accuracy within years where image acquisition occurred on cereal crops pre-harvest, and forage crops pre-and post-harvest. The largest value of kappa (K) resulting from supervised MLC of 5-yr composite NAIP imagery was 63%, observed at the site of deepest DTS, while a K value of -17.6% was observed at the shallowest DTS. No significant relationship was found between DTS and the measures of accuracy summarized in Table 1.
Table 1. Soil depth prediction accuracy (%) of supervised maximum likelihood classification with 5-yr composite NAIP imagery by site, with depth to shale (DTS) in meters.
Figure 1. Shallow (blue) and deep (yellow) soils as predicted by maximum likelihood classification of 5-yr composite imagery at six, 0.4-ha subfields on the Moccasin Terrace, central Montana.However, classified images for sites 1 (Fig 1d) and 2 (Fig 1c) the shallowest DTS sites, are vertically linear in character, despite the relative prominence of diagonally linear features observed in the raw imagery.
The vertical linearity of these classified images suggests equipment affectations may be more severe in areas of shallow DTS. This makes sense intuitively, since the water table is likely nearer to the surface in these areas and farmers are probably more often forced to conduct field work in wetter-than-ideal conditions. Further, MLC results indicating relatively high accuracy at site 2 suggest highly compacted areas may have prevented accurate soil depth readings, or the hydraulic probing method used in this and other studies (Jones et al., 2011; John et al., 2017) is a combined measure of DTG and compaction. Broadly, MLC shows moderate potential for predicting DTG, unless years in which image acquisition occurs on cereal crops pre-harvest or forage crops pre- and post-harvest are singled out. However, in these cases, NDVI is likely a stronger predictor of DTG.
Some climate change impact models forecast declining N exports based on the combined effects of reduced mineralization and reduced streamflow in hotter, drier conditions (Ye & Grimm, 2013), while others forecast up to 317% increases in nitrate leaching, depending on location and climate scenario (He et al., 2018). The current effort builds upon a three-year study from Jones et al. (2011) which monitored overwinter soil nitrate changes in shallow (0-20 cm) and deep (> 20 cm) soils (Fig 2a,d) following several different crops (Fig 2b,e) on the Moccasin Terrace from 2007-2010 (Fig 2c,f). Soil samples were taken from the same locations (within 30 cm) in fall, mid-winter, and spring and then analyzed for nitrate. Broadly, results indicate that year and gravel depth (given by the maximum depth of the hydraulic soil probe) explained most of the variability in ONC.
Figure 2. Soil nitrate changes from fall to mid-winter (a-c) and from mid-winter to spring (d-f) at the Central Agricultural Research Center in Moccasin, MT. Color and shape indicate soil depth (shallow = 0-8 inches; deep > 8 inches) previous crop (A.le = annual legume; fall = fallow; Oils = oilseed; Sm G = small grain), or year. Solid line indicates 1:1 (Jones et al., 2011).
Large increases in soil nitrate were observed from fall to mid-winter in 2008-09 and 2009-10, while nitrate was virtually unchanged over this period in 2007-08 (Fig 2c), suggesting conditions were favorable to N mineralization and nitrification in fall-winter of 2008-09 and 2009-10 relative to 2007-08. It is possible that conditions were favorable to nitrification in fall-winter 2007-08 as well, but that leaching/denitrification losses were relatively higher. However, virtually unchanged 20-cm soil moisture during this period suggests the opposite is true. Furthermore, average daily temperatures in 2007 dropped steadily from about 24 °C to 4 °C from September through November, then dipped below freezing and more or less stayed there until April of 2008. A preliminary assessment of 2019 soil respiration data suggests most microbial activity in these soils occurs above air temperature of 0 °C. Soil depth and cumulative growing degree units (base = 0) from August through April had additive effects on nitrate inventory in spring (P < 0.001; R2 = 0.8). It follows that conditions in early fall 2007 would have been conducive to nitrification, but by mid-fall nitrification rates would have declined drastically. There were more days with air temperatures above 0 °C and temperatures were generally more variable in the falls of 2008 and 2009, which could explain the soil nitrate increases from fall to mid-winter in these years.
From mid-winter to spring, large nitrate losses were observed in 2009-10, but virtually no changes were observed over this period in 2007-08 or 2008-09 (Fig 2f). Losses to leaching/denitrification may have been higher or inputs from nitrification may have been lower in 2009-10. However, Fig 2d shows that most soils losing nitrate from mid-winter to spring 2009-2010 were less than 20 cm deep, suggesting that leaching was the main contributor to the 2009-10 losses rather than suppressed nitrification. In early March of 2010, air temperatures rose to well above freezing and stayed there through April. These thawing conditions may have been conducive to leaching, which would help explain the losses from mid-winter to spring in 2009-10.
Preliminary results from the current study indicate that areal images subjected to principal component analysis and maximum likelihood classification can accurately predict DTG when image acquisition occurs at various crop stages and various phases of the crop rotation. This provides a convenient method for determining DTG within a field when NDVI utility is limited due to circumstantially poor mapping conditions at image acquisition. Furthermore, DTG maps can be generated using public data (i.e., USDA National Agricultural Imagery Program) conveniently accessed by programs such as Google Earth Engine. These data layers may broaden the currently narrow profit margins associated with variable rate fertilizer technology, especially if implemented in combination with split-application and/or slow-release technologies. At a given location on the Moccasin Terrace, DTS determines groundwater storage capacity, which likely influences leaching rates (Sigler et al., 2018) and possibly affects denitrification rates (Sigler et al., in review) . Soils in areas of shallow DTS may be subject to saturation excess overland flow and therefore reduced deep percolation/leaching rates relative to those in areas of deep DTS (Sigler et al., 2018). Saturated soils in these areas may also create anoxic conditions necessary for denitrification (Seiler et al., 2005; Otero et al., 2009). As noted, gravel depth measurements were made along a predicted DTS gradient from shallow (<5 m) to deep (>10 m) shale depth in order to assess whether DTS and DTG interact to affect soil nitrate concentration in spring. This assessment is ongoing. However, a preliminary assessment suggests the effects of shale depth on wet bulk density are gravel depth dependent (P < 0.01). This is significant because BDWET has been reported to correlate with nitrous oxide efflux in fertilized soils (Ball et al., 2000). Nitrous oxide flux was not measured in the current study. However, Sigler et al. (2018) found that apparent leaching losses were greater in areas of shallow shale. Indeed, higher losses to denitrification in areas of shallow DTS would lead to lower actual and apparent leaching losses.
Ball BC, Horgan GW, Parker JP (2000). Short‐range spatial variation of nitrous oxide fluxes in relation to compaction and straw residues. European Journal of Soil Science, 51(4), 607-616.
Burton DC, Beauchamp EG (1994) Profile of nitrous oxide and carbon dioxide concentrations in a soil subject to freezing. Soil Sci Soc Am J 58:115-122
Campbell CA, deJong R, Zentner RP (1984) Effect of cropping, summer fallow and fertilizer nitrogen on nitrate-nitrogen lost by leaching on a brown chernozemic loam. Can J Soil Sci 64:61-74
Campbell JB, Wynne RH (2011) Introduction to remote sensing. Guilford Press.
Christensen S, Tiedje JM (1990) Brief and vigorous N2O production by soil at spring thaw. J Soil Sci 41:1-4
He W, Yang JY, Qian B, Drury CF, Hoogenboom G, He P, Lapen, D, Zhou W (2018) Climate change impacts on crop yield, soil water balance and nitrate leaching in the semiarid and humid regions of Canada. PLoS ONE. 13(11) e0207370
Heaney DJ, Nyborg M, Solberg ED, Malhi SS, Ashworth J (1992) Overwinter nitrate loss and denitrification potential of cultivated soil in Alberta. Soil Biol Biochem 24:877-884
Jacinthe PA, Dick WA, Owens LB (2002) Overwinter soil denitrification activity and mineral nitrogen pools as affected by management practices. Biol Fertil Soils 36:1-9
Jackson-Smith D, Armstrong A (2012). Summary Report. 2012 Judith River Watershed Farmer Survey. Online. http://waterquality.montana.edu
John AA, Jones CA, Ewing SA, Sigler WA, Bekkerman A, Miller PR (2017) Fallow replacement and alternative nitrogen management for reducing nitrate leaching in a semiarid region. Nutr Cycl Agroecosyst. Volume pages. doi: 10.1007/s10705-017-9855-9
Jones CA, Chen C, McVay K, Stougaard B, Westcott M, Eckhoff J, Lenssen A, Weeding J, Greenwood M (2011) Measured and predicted temporal changes in soil nitrate-N levels from late summer to early spring in Montana. Western Nutrient Mgmt Conference. Vol 9
Jones C, Oson-Rutz K (2019) Soil Testing: Getting a good sample. Montana State University. http://landresources.montana.edu/soilfertility/documents/PDF/sscoop/SoilTestGetGoodSampleSS.pdf
Liang BV, Remillard M, MacKenzie AF (1991) Influence of fertilizer, irrigation, and non-growing season precipitation on soil nitrate-nitrogen under corn. J Environ Qual 20:123-128
McCracken DV, Smith MS, Grove JH, MacKown CT, Blevins RL (1994) Nitrate leaching as influenced by cover cropping and nitrogen source. Soil Sci Soc Am J 58:1476-1483
McDonald NT, Watson CJ, Lalor STJ, Laughlin RJ, Wall DP (2013) Evaluation of soil tests for predicting nitrogen mineralization in temperate grassland soils. Soil Sci Soc Am J 78: 1051-1064
Mitchell MJ, Driscoll CT, Kahl JS, Likens GE, Murdoch PS, Pardo LH (1996) Climatic control of nitrate loss from forested watersheds in the northeastern United States. Environ Sci Technol 30:2609-2612
Miller CR (2013) Groundwater nitrate transport and residence time in a vulnerable aquifer under dryland cereal production. Thesis, Montana State University, Bozeman, MT
Miller P. Montana soil test results. Unpublished raw data
Chen C. Montana soil test results. Unpublished raw data
Otero N, Torrentó C, Soler A, Menció A, Mas-Pla J (2009) Monitoring groundwater nitrate attenuation in a regional system coupling hydrogeology with multi-isotopic methods: the case of Plana de Vic (Osona, Spain). Agric. Ecosyst. Environ. 133, 103-113
Roth GW, Fox RH (1990) Soil nitrate accumulations following nitrogen-fertilized corn in Pennsylvania. J Environ Qual 19:243-248
Rover M, Heinemeyer O, Kaiser EA (1998) Microbial induced nitrous oxide emissions from an arable soil during winter. Soil Biol Biochem 14:1859-1865
Ryan MC, Kachanoski RG, Gillham RW (2000) Overwinter soil nitrogen dynamics in seasonally frozen soils. Can J Soil Sci 80:541-550
Schmidt C, Mulder R (2010) Groundwater and surface water monitoring for pesticides and nitrate in the Judith River Basin, central Montana. Montana Department of Agriculture. http://agr.mt.gov/agr/Programs/NaturalResources/Groundwater/Reports/. Accessed 9 Sept 2018
Sigler WA, Ewing SA, Payn RA, Jones CA, Weissmann GS (2013) Linking soil water balance and water age with leaching of nitrate to groundwater in an agricultural setting. AGU, Dec. 9-13 2013, San Francisco, CA; Poster H53F-1482
Sigler WA, Ewing SA, Jones CA, Payn RA, Brookshire ENJ, Klassen JK, Jackson-Smith D, Weissmann GS (2018) Connections among soil, ground, and surface water chemistries characterize nitrogen loss from an agricultural landscape in the upper Missouri River Basin. J Hydrol. 556, 247-261
Sigler WA, Ewing SA, Jones CA, Payn RA, Miller P, Maneta M (in review). Water and nitrate loss from dryland agricultural soils is controlled by management, soils, and weather.
Seiler, RL (2005) Combined us of 15N and 18O of nitrate and 11B to evaluate nitrate contamination in groundwater. Appl. Geochem. 20, 1626-1636
Vigil MF, Eghball B, Cabrera ML, BR Jakubowski, Davis JG (2002) Accounting for seasonal nitrogen mineralization: an overview. J Soil & Water Conserv Soc. 57: 464-469
Whitlock C, Cross W, Maxwell B, Silverman N, Wade AA (2017) 2017 Montana Climate Assessment. Bozeman and Missoula MT: Montana State University and University of Montana, Montana Institute on Ecosystems. 318 p. doi:10.15788/m2ww8w
Zipper SC, Soylu ME, Booth EG, Loheide II SP (2015) Untangling the effects of shallow groundwater and soil texture as drivers of subfield-scale yield variability. Water Resour Res 10.1002/2015WR017522
Educational & Outreach Activities
As noted above, Spring Test N is being monitored at five commercial farms and major findings will be communicated to farmers at the end of the two-year study. Cooperating farmers assisted in site selection by providing historical crop rotation, fertility, and other management records of fields and subfields. Cooperators also revised and edited surveys designed to assess pre-existing knowledge of ONCs and climate predictions prior to distribution. These surveys were ultimately distributed in February 2019. Multiple attempts have been made to convene a group cooperator meeting, but schedule conflicts and high workloads associated with poor weather and delayed field operations (including delayed fall planting in 2018 and delayed harvest in 2019) have thus far prevented such a meeting. However, one-on-one conversations with cooperators in-person and over the phone, as well as via email, were held on multiple occasions throughout the course of the project. The first discussions with potential cooperators occurred in October, 2018 and have continued regularly since.
At the CARC summer field day in June 2019 (attendance ≈ 70), a presentation was made summarizing results of past work (Jones et al., 2011), study methods, and implications of early findings for local farming operations. Additionally, a presentation outlining the purpose and proposed methods of the experiment was given as part of the Department of Land Resources and Environmental Sciences graduate seminar course on the campus of Montana State University, Bozeman, MT in December of 2019 (Fordyce, 2019). A public presentation at CARC’s annual ‘winter research roundup’ (attendance ≈ 70) will be held in December 2020 to communicate major findings. Additionally, we anticipate three scholarly publications from this work: one focused on methodology of imagery classification, one focused on geospatial modelling of Spring Test N, and a third focused on implications of warmer, wetter winters/spring for fertilizer management in the Northern Great Plains. The Journal of Applied Remote Sensing, the Journal of Nutrient Cycling in Agroecosystems, and the Journal of Environmental Quality, respectively, are possible outlets for these papers. In addition to scholarly publications, a raster layer of modelled Spring Test N compatible with Google Earth will be made available for download in spring of 2021 and after, depending on the public response. Furthermore, emails will be sent out via the Montana State University ‘Ag Alert’ email chain to forewarn growers of high or low modelled Spring Test N. Finally, informational handouts and brochures will be made available at CARC field days and other public events.