Journal list menu

Volume 48, Issue 4 p. 803-812
Special Section: Agricultural Water Quality in Cold Environment
Open Access

Impacts of Soil Phosphorus Drawdown on Snowmelt and Rainfall Runoff Water Quality

Jian Liu

Corresponding Author

Jian Liu

School of Environment and Sustainability and Global Institute for Water Security, Univ. of Saskatchewan, Saskatoon, SK, S7N 3H5 Canada

Corresponding author ([email protected]).Search for more papers by this author
Jane A. Elliott

Jane A. Elliott

National Hydrology Research Center, Environment and Climate Change Canada, Saskatoon, SK, S7N 3H5 Canada

Search for more papers by this author
Henry F. Wilson

Henry F. Wilson

Brandon Research and Development Centre, Agriculture and Agri-Food Canada, Science and Technology Branch, Brandon, MB, R7A 5Y3 Canada

Search for more papers by this author
Helen M. Baulch

Helen M. Baulch

School of Environment and Sustainability and Global Institute for Water Security, Univ. of Saskatchewan, Saskatoon, SK, S7N 3H5 Canada

Search for more papers by this author
First published: 01 July 2019
Citations: 28

Supplemental material is available online for this article.

Assigned to Associate Editor Merrin Macrae.

Abstract

Managing P export from agricultural land is critical to address freshwater eutrophication. However, soil P management, and options to draw down soil P have received little attention in snowmelt-dominated regions because of limited interaction between soil and snowmelt. Here, we assessed the impacts of soil P drawdown (reducing fertilizer P inputs combined with harvest removal) on soil Olsen P dynamics, runoff P concentrations, and crop yields from 1997 to 2014 in paired fields in Manitoba, Canada. We observed that Olsen P concentrations in the 0- to 5-cm soil layer were negatively correlated with the cumulative P depletion and declined rapidly at the onset of the drawdown practice (3.1 to 5.4 mg kg−1 yr−1 during 2007–2010). In both snowmelt runoff and rainfall runoff, concentrations of total dissolved P (TDP) were positively correlated with the concentrations of soil Olsen P. Soil P drawdown to low to moderate fertility levels significantly decreased mean annual flow-weighted TDP concentrations in snowmelt runoff from 0.60 to 0.30 mg L−1 in the field with high initial soil P and from 1.17 to 0.42 mg L−1 in the field with very high initial soil P. Declines in TDP concentration in rainfall runoff were greater. Critically, yields of wheat (Triticum spp.) and canola (Brassica napus L.) were not affected by soil P depletion. In conclusion, we demonstrate that relatively rapid reductions in P loads are achievable at the field scale via managing P inputs and soil P pools, highlighting a management opportunity that can maintain food security while improving water security in cold regions.

Core Ideas

  • Reducing or ceasing soil P input decreased soil Olsen P rapidly.
  • Soil Olsen P concentrations were negatively correlated with cumulative P depletion.
  • Declining soil P led to decreases in flow-weighted mean dissolved P concentrations.
  • Improved water quality can be achieved with management of soil P.
  • Agricultural productivity can also be maintained with lower soil P.

Abbreviations

  • SWE
  • snow water equivalent
  • TDP
  • total dissolved phosphorus
  • Phosphorus runoff from arable land contributes to the eutrophication of downstream freshwater ecosystems (Schindler et al., 2012; Li et al., 2015; Sharpley et al., 2015). Among the many challenges to reducing P loads from agricultural sources to surface waters is the reality that even where significant reductions in P input into agricultural systems are undertaken to reduce P loading to surface waters, contributions to runoff from legacy P that has accumulated in soil as a result of past land management practices can prevent or significantly delay the achievement of P loading reductions (Kleinman et al., 2011; Sharpley et al., 2013; Schoumans et al., 2014).

    Soil P constitutes a large P pool, and a long-term source of P to surface waters. A number of studies reported that P loss in surface runoff that was induced by rainfall is often strongly related to P concentrations in soils (Pote et al., 1999; Andraski and Bundy, 2003; Vadas et al., 2005), suggesting the potential to reduce runoff P losses through lowering soil test P levels. Reducing or ceasing P applications to crops and mining soil P by crop harvests has been proposed as a measure for soil P drawdown (Koopmans et al., 2004; van der Salm et al., 2009; Kleinman et al., 2011; Svanbäck et al., 2015; Fiorellino et al., 2017; Vadas et al., 2018). However, the efficacy of this strategy within a reasonable timeframe is highly variable. For instance, van der Salm et al. (2009) reported a significant drawdown of P in four grassland soils by mining the P with zero P application and removing grasses for a period of 5 yr. The strategy reduced ammonium lactate-extractable P (Egnér et al., 1960) by 10 to 60% and solution P concentrations by 30 to 90% in the upper 5-cm soil layer. The finding was supported by Svanbäck et al. (2015), who observed an 11 to 37% decline in ammonium lactate-extractable P in the top 20 cm of four arable soils after 7 to 16 yr of zero P application. In a P-enriched Othello silt loam (fine-silty, mixed, active, mesic Typic Endoaquults), nonetheless, no significant differences in Mehlich-III-extractable P (Mehlich, 1984) were detected among different rates of poultry litter applications (including zero P application) for 10 yr (Sharpley et al., 2013). In the same study, Sharpley et al. (2013) reviewed soil P response to P fertilization in 10 field experiments (4–27 yr in duration). The results indicated that ceasing P applications would result in a large range of variation of decline in soil test P (i.e., 0.4–30 mg kg−1 yr−1).

    The rate of decline in soil test P can be influenced by several natural and management factors including climates, soil properties, cropping systems, and available soil P pools (Vadas et al., 2018). In cold climates, notably, crop growing seasons are shorter than in warm climates, and the turnover of P is different due to long, cold winter seasons. Still, it is not clear to what extent soil P can be drawn down by harvesting crops in combination with ceasing or reducing P applications in cold climates. Furthermore, the potential impacts of soil P drawdown on crop production are also unknown. Ideally, soil P drawdown should be aimed at reducing P loss from soils while maintaining soil P fertility at an appropriate level so that it does not harm crop yields.

    Although it is expected that lower soil test P will result in reduced runoff P concentrations, there has been a general paucity of scientific validation due to the rarity of long-term field research (van der Salm et al., 2009; Sharpley et al., 2013). In particular, little is known about the impacts of soil P drawdown on P in snowmelt runoff. It has been hypothesized that soil nutrient management may have only a minor benefit to water quality in snowmelt-dominated cold regions such as the Canadian prairies. This is because snowmelt water is generally assumed to interact less with frozen soils than during rainfall runoff events (Granger et al., 1984). Moreover, the contribution of soil P may be masked by fall or winter P applications and by the P in crops or crop residues, two potentially important sources of P for losses in spring snowmelt runoff (Elliott, 2013; Liu et al., 2013a; Liu et al., 2018).

    On the Canadian prairies, snowmelt runoff often accounts for 80 to 90% of the annual surface runoff (Nicholaichuk, 1967; Glozier et al., 2006). However, in the future, less snowfall and more spring rainfall is anticipated than in the past (Szeto et al., 2015), and in some areas of the region, a growing importance of rainfall runoff has been noted (Dumanski et al., 2015). Thus, solutions to reduce P in both snowmelt and rainfall runoff are relevant in this region. The objectives of this study were (i) to examine the potential to lower soil P levels by reducing P fertilizer input without reducing crop yield in cold climates, and (ii) to assess the impact of soil P drawdown on water quality in snowmelt runoff and rainfall runoff, and on agronomic productivity.

    Materials and Methods

    Study Site

    The study was conducted on paired fields in the South Tobacco Creek watershed (49°20′ N, 98°22′ W), located in the Canadian prairie province of Manitoba (Supplemental Fig. S1). The South Tobacco Creek watershed is a long-term experimental watershed operated by the Agriculture and Agri-Food Canada Brandon Research and Development Centre in collaboration with the Environment and Climate Change Canada National Hydrology Research Centre. The 76-km2 watershed is situated on the edge of the Manitoba escarpment and is a headwater stream of the Red River that drains to the eutrophic, 23,750-km2–sized Lake Winnipeg. The watershed has been intensively monitored at field, subwatershed, and watershed scales since 1993 to study temporal and spatial drivers controlling nutrient losses and to assess the impacts of various nutrient and water management practices on water quality (Tiessen et al., 2010; Liu et al., 2013b, 2014; Chen et al., 2017).

    The study region is typical of the Pembina Hills upland, as a transition area between the lower Manitoba plain and the higher Saskatchewan plain (Michalyna et al., 1988). It has a subhumid continental climate with long, cold winters and short, warm summers (Environment Canada, 2009). The mean annual temperature is ∼3°C, and the mean annual precipitation is ∼550 mm, of which 25 to 30% occurs as snowfall. Soils are primarily clay-loam (27% clay) formed on moderately to strongly calcareous glacial till that overlays shale bedrock, with dominant soil series as Dark Gray Chernozems (Mollisols).

    Field Experiments

    The paired experimental fields were established in 1993 and designated as F10 (4.2 ha) and F11 (5.1 ha), respectively (Supplemental Fig. S1). Both fields are north facing, with undulating landscapes and slope gradients of ∼5%. The assessment of P management impacts on water quality in this study is closely related to the previous tillage research in the fields. The field experiments were designed to assess the impacts of contrasting tillage methods on water quality from 1997 to 2007 (details are provided in Tiessen et al., 2010). During this period, the field F10 was treated with conventional tillage typically implemented as combined light-duty cultivation, heavy-duty cultivation, and harrow and packers, and F11 with conservation tillage as harrow, packers, and no-till. The conventional tillage was defined as leaving <30% of the residue from the previous crop on the soil surface after seeding, while conservation tillage left at least 30% residue (Lal, 2003). Both fields had otherwise identical management. They were planted in a cereal–oilseed rotation that included 3 yr of spring wheat (Triticum spp.), 3 yr of canola (Brassica napus L.), 3 yr of flax (Linum usitatissimum L.), 1 yr of barley (Hordeum vulgare L.), and 1 yr of oats (Avena sativa L.). From 2008 to 2014, both fields were tilled conventionally, almost identically in the same year, and the crops used were limited to a rotation of wheat and canola (Table 1). As measured in 2004 (the earliest year, when the measurement started), F10 had an organic matter content of 3.4% and F11 had 4.5% in the top 15 cm soil.

    Table 1. Summary of long-term management practices implemented in the paired F10 and F11 fields located in the South Tobacco Creek watershed in Manitoba, Canada. All years referred to crop years, roughly from October of the previous year to September of the crop year. A negative value for soil P depletion reflects a net increase in P for that year, whereas a positive value reflects depletion, or drawdown.
    Crop year Crop P fertilizer rates Residue from previous year Tillage methods Grain yield Grain P removal Straw P removal Soil P depletion
    F10 & F11 F10 & F11 F10 F11 F10 & F11 F10 & F11 F10 & F11 F10 & F11
    kg ha−1 Mg ha−1 kg ha−1
    Before P drawdown
    1997 Flax 0 Wheat Heavy-duty cultivator in fall; light-duty cultivator and harrow + packers in spring No till 1.07 5.3 1.0 6.3
    1998 Flax 0 Flax Heavy-duty cultivator and harrow + packers in fall; light-duty cultivator and harrow + packers in spring Harrow + packers in fall 2.20 10.9 2.0 12.9
    1999 Wheat 14.8 Flax Heavy-duty cultivator and anhydrous rig in fall; light-duty cultivator and harrow + packers in spring No till 2.02 8.7 0 −6.1
    2000 Canola 17.2 Wheat Heavy-duty cultivator in fall; light-duty cultivator and harrow + packers in spring Harrow + packers in spring 2.97 27.0 0 9.8
    2001 Oats 14.8 Canola Heavy-duty cultivator in fall; light-duty cultivator and harrow + packers in spring Harrow + packers in fall 0.76 2.6 0.9 −11.3
    2002 Flax 0 Oats Heavy-duty cultivator and harrow + packers in fall; light-duty cultivator and harrow + packers in spring Harrow + packers in fall 3.14 15.5 2.8 18.3
    2003 Wheat 12.3 Flax Heavy-duty cultivator in fall; light-duty cultivator in spring No till 2.02 8.7 2.1 1.5
    2004 Canola 17.2 Wheat Heavy-duty cultivator in fall; light-duty cultivator in spring No till 2.80 25.5 0 8.3
    2005 Barley 0 Canola Heavy-duty cultivator and light harrow in fall; light-duty cultivator in spring Light harrow in fall 1.08 4.2 1.1 5.3
    P drawdown period
    2006 Canola 4.9 Barley Heavy-duty cultivator in fall; light-duty cultivator in spring No till 2.52 23.0 0 18.1
    2007 Wheat 4.9 Canola Heavy-duty cultivator in fall; light-duty cultivator in spring No till 3.03 13.1 3.2 11.4
    2008 Canola 4.9 Wheat Heavy-duty cultivator in fall; light-duty cultivator in spring Heavy-duty cultivator in fall 2.80 25.5 0 20.6
    2009 Wheat 4.9 Canola Heavy-duty cultivator in fall; heavy-duty and light-duty cultivators in spring Harrow + packers in spring 4.03 17.4 4.2 16.7
    2010 Canola 9.9 Wheat Heavy-duty cultivator in fall; light-duty cultivator in spring Heavy-duty cultivator in fall; light-duty cultivator in spring 2.80 25.5 0 15.6
    2011 Wheat 7.4 Canola Heavy-duty cultivator in fall; anhydrous rig in spring in spring Heavy-duty cultivator in fall; anhydrous rig in spring in spring 2.69 11.6 2.8 7.0
    2012 Canola 7.4 Wheat Heavy-duty cultivator in fall; light-duty cultivator in spring Heavy-duty cultivator in fall; light-duty cultivator in spring 3.08 28.1 0 20.7
    2013 Wheat 4.9 Canola Light-duty cultivator and harrow + packers in spring Light-duty cultivator and harrow + packers in spring 3.03 13.6 3.2 11.9
    2014 Canola 7.4 Wheat Deep tiller in fall; light-duty cultivator and harrow + packers in spring Deep tiller in fall; light-duty cultivator and harrow + packers in spring 3.36 30.6 0 23.2

    Throughout the field experiments from 1997 to 2014, two P fertilization regimes were implemented (Table 1). During 1997 to 2005, the crops received chemical P fertilizers at rates in approximate balance with the agronomic needs of wheat (13.6 kg P ha−1 yr−1), canola (17.2 kg P ha−1 yr−1), and oats (14.8 kg P ha−1 yr−1), and no P fertilizers for flax and barley. During 2006 to 2014, in contrast, wheat and canola were supplied with much lower rates of P fertilizers (wheat: 5.5 kg P ha−1 yr−1; canola: 6.9 kg P ha−1 yr−1). These operations resulted in substantial differences in soil P balances between the two periods (i.e., a much larger soil P depletion in the later period than in the former). In this study, we defined the two periods as before P drawdown and P drawdown period, respectively, and evaluated the patterns of P in snowmelt runoff and rainfall runoff in the two periods. All P fertilizers were applied in spring, with seed, broadcast, or banded. Throughout 1997 to 2014, N fertilizer rates varied between crops but were the same on both fields. About 90% of the N was applied with seed, broadcast, or banded in spring, whereas the rest was applied in the fall. On average, barley, canola, flax, oats, and wheat received 67, 101, 60, 56, and 90 kg N ha−1 yr−1, respectively. Details of crop sequences, tillage methods, and P fertilizer inputs are presented in Table 1.

    Runoff Monitoring and Water Sampling

    In the fields, monitoring of snowmelt runoff and rainfall runoff was initiated in 1993, but only the data for the period 1997 to 2014 were used for this study. The runoff monitoring used a compound angle V-notched weir, which was installed at the lower edge of each field (Supplemental Fig. S1). The water level at the weir was measured at 5-min intervals by an ultrasonic senor with a data logger, and it was converted to flow rate using a standard V-notched weir flow equation (Smith, 1985). For quality control purposes, the water level measurements were adjusted according to ambient air temperature that was monitored on site. During each runoff event, water was sampled using a Sigma autosampler (800SL) installed adjacent to the weir, and additional grab samples were collected during low-flow events. Depending on flow volume and duration, the number of samples ranged from 3 to >20 per event. Water samples (2 L) were retrieved each day from the sampler, packed on ice, and transported to the laboratory for analysis of P. More details on the runoff monitoring and water sampling methods can be found in Tiessen et al. (2010). Total P, from unfiltered water samples, and total dissolved P (TDP), after 0.45-μm filtration, were determined as soluble reactive P by reduction using stannous chloride, after digesting the water samples with a sulfuric acid and persulfate mixture (Environment Canada, 1979a, 1979b). Particulate P was calculated as the difference between total P and TDP.

    Rainfall was monitored onsite using a tipping bucket rain gauge. Snow depth and snow density were measured in late winter (just before the spring snowmelt) along two transects in each field, and they were used to calculate snow water equivalent (SWE), an equivalent depth of water stored before snowmelt (Tiessen et al., 2010).

    Soil Sampling and Phosphorus Measurement

    To relate tillage and fertilization practices to soil P dynamics and eventually to runoff P patterns, surface soils in both fields were sampled from 1997 to 2014. However, the design of soil sampling changed over time, toward a greater intensity. Prior to 2007, soil samples were collected from few locations in the fields and the samples were taken at a depth of 0 to 15 cm in most of the years (two to three samples per field in a year). Since 2007, soil samples were collected both from a shallower depth of 0 to 5 cm and from 5 to 15 cm, and from a greater number of locations in each field. Each year, 12 soil samples were collected from each field, with four locations representing each of the three landscape positions (i.e., upper, mid-, and lower slope). Notably, each soil sample was composited from four to seven soil cores collected from a location, and the sampling locations were georeferenced and resampled each year. All soil samples were collected after harvest in the fall, but before the fields were tilled. The samples were sent to Agvise Laboratories (Northwood, ND) for standard Olsen P analyses (Olsen et al., 1954). In this study, only the Olsen results for the period 2007 to 2014 were presented, due to (i) the differences in soil sampling intensity and depth in comparison with the earlier sampling dates, and (ii) the fact that spatial variation in soil Olsen P was large, making it difficult to compare between pre-2007 and post-2007, when soil samples were collected from different sets of locations.

    Data Analyses

    Soil P balance (surplus or depletion) indicates the size of P source that has the potential to be lost in runoff water (Liu et al., 2012). In this study, we estimated soil P depletion as the difference between soil P output with crop removal and soil P input with mineral fertilizers. A full P budget would also include P input as atmospheric deposition and P output in runoff, but these quantities are very small in comparison with crop removal P and fertilizer P (Roste, 2015). The estimation of soil P depletion (PDepletion) is expressed as
    urn:x-wiley:00472425:equation:jeq2jeq2018120437-math-2433(1)
    Phosphorus output by grain harvest (PGrain harvest) was calculated as
    urn:x-wiley:00472425:equation:jeq2jeq2018120437-math-0002(2)
    where YGrain is grain yield (Mg ha−1), which was recorded annually by the farmer. The RP removal (i.e., P removal rate [kg P ha−1/Mg grain ha−1]) was not measured but estimated from literature values (Canadian Fertilizer Institute, 2001), which were 4.31 for wheat, 3.84 for barley, 3.51 for oats, 9.18 for canola, and 5.12 for flax, respectively.
    Phosphorus removal associated with straw removal (PStraw removal) was estimated as
    urn:x-wiley:00472425:equation:jeq2jeq2018120437-math-0003(3)
    where RStraw:Grain is the ratio of total straw biomass to grain yield. Here, we used the values reported for the study region: 1.34 for wheat (Tiessen et al., 2005), 2.33 for canola (Nuttall et al., 1992), 1.44 for oats (Carefoot and Janzen, 1997), 1.02 for barley (McAndrew et al., 1994), and 1.94 for flax (Campbell et al., 2005). The straw removal fraction (FStraw removal) was identified based on actual straw bailing operations and it equaled to 0.6 for a baling operation but 0 for no baling. The RP removal ratio (kg P ha−1/Mg straw ha−1) was again adapted from Canadian Fertilizer Institute (2001), which was 1.30 for wheat, 1.24 for barley, 1.43 for oats, 1.62 for canola, and 0.77 for flax, respectively.

    All statistical analyses were conducted using SAS 9.4 (SAS Institute, 2012). A mixed model was used to assess the effects of P drawdown on annual flow-weighted mean P concentrations in snowmelt and rainfall runoff, event-based mean runoff P concentrations, annual soil Olsen P, and annual crop yield. For assessing P drawdown effects on runoff P, the time period (before P drawdown and P drawdown period) was used as a fixed effect, and crop type and precipitation amount (SWE or rainfall) were used as random effects. It should be noted that agronomic data were aligned to the hydrologic year (October–September) or from postharvest of the previous crop to harvest of the crop year. When assessing temporal changes in soil P, years were set as a fixed effect and sampling points as a random effect. For yield assessment, the time period was used as a fixed effect. All the fixed effects were assessed by using the Tukey–Kramer least square means method. Moreover, a Proc Model procedure was used to test the logarithmic correlation between soil Olsen P concentration and cumulative soil P depletion. A Proc Stepwise procedure for multiple regressions was used to describe relationships among annual flow-weighted mean TDP concentrations in runoff, soil Olsen P, and SWE or rainfall for the data from 2007 to 2014. Residuals of all data followed a normal distribution according to normality tests. A significance level of α = 0.1 was used throughout the study, unless it was otherwise specified. This probability level is used to account for the high variability inherent in field and watershed-based experiments as suggested by Hansen et al. (2000) and Tiessen et al. (2010), for example.

    Results

    Soil Phosphorus Drawdown by Crop Harvests

    During the experiment from 1997 to 2014, both F10 and F11 received relatively low rates of fertilizer P application (Table 1). Crop harvests repeatedly removed even greater amounts of P than fertilizer P inputs and thus resulted in depletion of soil P, particularly since 2006. The mean annual P depletion ranged from 5.0 kg ha−1 before P drawdown (1997–2005) to 16.1 kg ha−1 during the drawdown period (2006–2014). The depletion of P had a significant impact on the Olsen P concentrations in the surface soil. The measurements in 2007 to 2014, when a uniform soil sampling method was used, showed that soil Olsen P concentrations in the upper 5 cm soil depth were negatively correlated with cumulative soil P depletion in both F10 (R2 = 0.95, p < 0.0001) and F11 (R2 = 0.91, p = 0.001), with a logarithmic function (Fig. 1a-1). Similar correlations were also found in the upper 15-cm soil depth (Fig. 1a-2): F10 (R2 = 0.86, p = 0.001) and F11 (R2 = 0.70, p = 0.02).

    Details are in the caption following the image

    Trends of soil Olsen P concentrations. (a) Relationship between soil Olsen P concentrations and cumulative soil P depletion for the paired experimental fields of F10 and F11 at soil depths of (a-1) 0–5 and (a-2) 0–15 cm, respectively. (b) temporal changes of the soil Olsen P concentrations from 2007 to 2014 for F10 and F11: (b-1) 0- to 5-cm soil depth in F10, (b-2) 0- to 5-cm soil depth in F11, (b-3) 0- to 15-cm soil depth in F10, and (b-4) 0- to 15-cm soil depth in F11. Spatial variations of soil Olsen P within individual years are presented as boxplots, and mean values are presented as blue dots. Significant differences between the means are indicated by different letters, at p < 0.1. The soil data in 2014 was excluded for F11 due to manure application to the field in the fall of 2013.

    The fields F10 and F11 had very different soil Olsen P concentrations in the upper 5-cm soil depth in 2007 (22.3 vs. 33.8 mg kg−1), because of the contrasting tillage methods implemented during 1997 to 2006. Although F10 remained in conventional, heavy tillage throughout the study, F11 was converted from conservation tillage during 1997 to 2007 to conventional tillage for 2008 to 2014 (Table 1). The conservation tillage enriched P in the upper 5-cm topsoil of F11 (Tiessen et al., 2010). Despite differences in tillage practices, soil Olsen P concentrations in the upper 5-cm soil depth had very similar temporal trends in F10 and F11, which declined rapidly from 2007 to 2010 by an average of 3.1 mg kg−1 yr−1 in F10 and 5.4 mg kg−1 yr−1 in F11, and then leveled off in both fields (Fig. 1b-1 and 1b-2). In the upper 15-cm soil depth, meanwhile, Olsen P concentrations declined from 16 mg kg−1 in 2007 to 8 mg kg−1 in 2014 in F10 and from 21 mg kg−1 in 2007 to 13 mg kg−1 in 2013 in F11 (Fig. 1b-3 and 1b-4). It should be noted that crop yield did not decrease with soil P drawdown. In fact, canola yield remained stable and wheat yield even increased significantly during 2006 to 2014 as compared with 1997 to 2005 (Table 2).

    Table 2. Summary of fertilizer P inputs, soil Olsen P, crop yield, runoff depth, flow-weighted mean P concentrations, and P loads in snowmelt runoff and rainfall runoff before and during P drawdown in the experimental fields of F10 and F11, respectively. All values are presented as mean annual numbers, for which the statistical differences between the two periods were indicated by different letters following the numbers (p < 0.1).
    Field P drawdown period Overall tillage method P input Soil P depletion Soil Olsen P Wheat yield Canola yield Precipitation§ Runoff TDP Particulate P Total P Total P
    kg ha−1 mg kg−1 Mg ha−1 mm mg L−1 kg ha−1
    Snowmelt runoff
    F10 Before (1997–2005) Conventional 8.5 5.0a 22.3 2.02a 2.89a 49a 42a 0.60b 0.14a 0.74b 0.20
    During (2006–2014) Conventional 6.3 16.1b 15.4 3.20b 2.91a 67a 56a 0.30a 0.12a 0.42a 0.14
    F11 Before (1997–2005) Conservation 8.5 5.0a 33.8 2.02a 2.89a 62a 54a 1.17b 0.15a 1.32b 0.53
    During (2006–2013)# Conventional 6.1 16.1b 23.4 3.20b 2.91a 66a 55a 0.42a 0.20a 0.62a 0.32
    Rainfall runoff
    F10 Before (1997–2005) Conventional 8.5 5.0a 22.3 2.02a 2.89a 387a 16a 0.82b 0.46a 1.28a 0.10
    During (2006–2014) Conventional 6.3 16.1b 15.4 3.20b 2.91a 355a 41a 0.18a 0.92b 1.10a 0.15
    F11 Before (1997–2005) Conservation 8.5 5.0a 33.8 2.02a 2.89a 387a 15a 0.93b 0.25a 1.18a 0.13
    During (2006–2013)# Conventional 6.1 16.1b 23.4 3.20b 2.91a 379a 11a 0.31a 0.39a 0.70a 0.03
    Annual runoff
    F10 Before (1997–2005) Conventional 8.5 5.0a 22.3 2.02a 2.89a 436a 50a 0.54b 0.31a 0.85b 0.30
    During (2006–2014) Conventional 6.3 16.1b 15.4 3.20b 2.91a 422a 72a 0.23a 0.26a 0.49a 0.29
    F11 Before (1997–2005) Conservation 8.5 5.0a 33.8 2.02a 2.89a 449a 66a 1.10b 0.16a 1.26b 0.67
    During (2006–2013)# Conventional 6.1 16.1b 23.4 3.20b 2.91a 445a 62a 0.40a 0.21a 0.61a 0.35
    • Exceptions exist in 2009 for F11 and in 2013 for both F10 and F11 (Table 1).
    • The change of sampling design in 2007 made it difficult to compare soil Olsen P values between the two periods (i.e., 1997–2005 vs. 2006–2014). The soil Olsen P values determined at the beginning of soil P drawdown (i.e., 2007) were used to represent soil P for the period before P drawdown treatment, and the mean values calculated 2007–2014 were used for the P drawdown period.
    • § This column indicates snow water equivalent for snowmelt runoff and rainfall for rainfall runoff.
    • TDP, total dissolved P.
    • # Manure was applied to F11 in fall 2013, and thus the data in the crop year of 2014 were excluded from the analysis.

    Overall Patterns of Runoff Phosphorus as Affected by Soil Phosphorus Drawdown

    Annual precipitation, flow, and P loss varied widely from 1997 to 2014. Although SWE ranged from 6 to 99 mm, annual rainfall ranged from 166 to 779 mm. On average, flow occurred in one to eight snowmelt events and zero to nine rainfall events per year in the paired fields F10 and F11, and annual snowmelt runoff and rainfall runoff varied from 2 to 131 mm and from 0 to 79 mm, respectively. Annual total P loss ranged from 0.01 to 1.17 kg ha−1, and 22 to 100% of that occurred as snowmelt runoff (mean = 71%). Total dissolved P constituted 51 to 92% of the total P in snowmelt runoff (mean = 74%) and 32 to 95% of the total P in rainfall runoff (mean = 54%), with particulate P constituting the rest of the total P.

    Although the interannual variations in hydrology, P fertilizer application rate, tillage, and amount of crop residue that is incorporated into soil may have contributed to the variability of P concentrations in runoff, results show that the P concentrations were related to soil P drawdown (Table 2). The flow-weighted mean concentrations of TDP in snowmelt runoff decreased significantly from 0.60 mg L−1 before P drawdown to 0.30 mg L−1 during the period of P drawdown in F10 (p = 0.06), and from 1.17 to 0.42 mg L−1 in F11 (p = 0.002). In rainfall runoff, the flow-weighted mean concentrations of TDP decreased from 0.82 to 0.18 mg L−1 in F10 (p = 0.04) and from 0.93 to 0.31 mg L−1 in F11 (p = 0.02). On an annual basis, the concentrations of TDP during the P drawdown period was 57 and 64% lower than before drawdown in F10 and F11, respectively. In contrast with the patterns of TDP, the flow-weighted mean concentrations of particulate P in snowmelt runoff tended to decrease in F10 but increase in F11 during 2006 to 2014 as compared with 1997 to 2005 (Table 2). In rainfall runoff, the particulate P concentrations appeared to increase in both fields, although the increase was significant in F10 only, in association with soil erosion in 2010. In summary, P drawdown significantly reduced total P concentrations in snowmelt runoff but did not reduce total P in rainfall runoff (Table 2). Furthermore, soil P drawdown appeared to affect the composition of total P. On average, in F10 and F11, annual mean proportion of total P as TDP was reduced from 85 to 70% in snowmelt runoff and from 71 to 30% in rainfall runoff during the P drawdown period.

    Relationship between Runoff Phosphorus, and Soil Olsen Phosphorus and Snow Water Equivalent or Rainfall

    There was a strong, positive correlation between annual flow-weighted mean TDP concentrations and soil Olsen P concentrations in the 0- to 5-cm soil depth for both snowmelt runoff and rainfall runoff (Table 3). In addition, annual flow-weighted mean TDP concentrations in snowmelt runoff were negatively correlated with SWE. Soil Olsen P concentrations and SWE collectively explained 40% of the variability of TDP concentrations in snowmelt runoff. Soil Olsen P alone explained 58% of the variability of TDP in rainfall runoff.

    Table 3. Stepwise regressions between annual flow-weighted mean total dissolved P concentrations in snowmelt runoff or rainfall runoff (mg L−1), and Olsen P concentrations in the upper 5-cm soil depth (mg kg−1) and snow water equivalent or rainfall depth (mm) for the data from 2007 to 2014.
    Runoff n Regression Partial 1 Partial 2
    Equation R2 p Variable R2 p Variable R2 p
    Snowmelt 14 TDP = 0.0144 × Olsen P 0.00395 × SWE + 0.306 0.40 0.06 Soil Olsen P 0.19 0.089 SWE 0.21 0.098
    Rainfall 10 TDP = 0.0121 × Olsen P + 0.0276 0.58 0.01 Soil Olsen P 0.58 0.01 Rainfall >0.15
    • TDP, total dissolved P; SWE, snow water equivalent.

    Event-Level Trends of Total Dissolved Phosphorus as Affected by Soil Phosphorus Drawdown

    The overall patterns of TDP as affected by soil P drawdown were confirmed by event-level TDP trends. When all the snowmelt and rainfall events were grouped according to runoff depths (i.e., 0–5, 5–15, 15–25, and >25 mm), the mean TDP concentrations for the events during the P drawdown period were 37 to 74% lower than those before P drawdown for almost all runoff depth categories and both fields (Fig. 2). The only exception was found for F10 at the rainfall runoff depth of 5 to 15 mm, where the two time periods had similar mean TDP concentrations (0.19 vs. 0.20 mg L−1). In particular, the difference in the TDP concentrations between the two periods was significant in most of the cases when the number of samples for comparison was sufficiently high, at 0- to 5-mm snowmelt runoff in both F10 (p = 0.003) and F11 (p = 0.002), 5- to 15-mm snowmelt runoff in F10 (p = 0.01), and 0- to 5-mm rainfall runoff in F11 (p = 0.03). Although the P drawdown period had, on average, a 57% lower mean TDP concentration at 0- to 5-mm rainfall runoff in F10 than before, the difference was not statistically significant due to high variability (p = 0.2).

    Details are in the caption following the image

    Distribution of total dissolved P concentrations for all runoff events at various snowmelt and rainfall runoff depths: (a) before P drawdown (1997–2005), and (b) during the P drawdown period (2006–2014). Variations in the P concentrations for runoff events within each runoff depth interval are presented as boxplots, and mean concentration values are presented as blue dots. The means were compared for the same category between the two periods, and asterisks indicate significant differences at p < 0.1. The number of samples for each category was indicated by n.

    Discussion

    In cold regions, snowmelt runoff often constitutes a significant proportion of annual runoff and nutrient losses (Liu et al., 2013b; Ulén et al., 2019). Indeed, although snowfall accounted for a relatively small proportion of the annual precipitation over the 18 monitoring years in the paired fields (here SWE/rainfall ≈ 1:5), snowmelt runoff contributed an average of 75% of the annual runoff and 71% of the annual total P loss. Total dissolved P, which is a major form of P extracted from the easily soluble P pools in soils (Pote et al., 1999; Vadas et al., 2005) and frozen-thawed plants (Elliott, 2013; Liu et al., 2013a), accounted for 74% of the total P loss. The small proportion of particulate P in total P loss (26%) is associated with the flat, nonerosive landscape of these fields. These results are consistent with the patterns reported for other watersheds in the Canadian prairies (Nicholaichuk, 1967; Ontkean et al., 2005) and highlight the importance of reducing P loads in snowmelt runoff to minimize their downstream impacts. Nonetheless, one should be aware that rainfall runoff is also relevant in the cold, semiarid prairie climates, and it can dominate P losses in some years.

    Few downstream options are available to mitigate P losses in cold environments (Baulch et al., 2019). For example, plant-related mitigation strategies such as buffer strips frequently fail to work efficiently in winter (Kieta et al., 2018), and plants can even become a source of P losses after plant cells are burst by frost (Bechmann et al., 2005; Elliott, 2013; Liu et al., 2013a). Thus, soil P management is especially critical to minimize the potential of P loss. In this study, we observed a clear declining trend of dissolved P concentrations in both snowmelt runoff and rainfall runoff during the P drawdown period (2006–2014) as compared with before (1997–2005), pointing to the impacts of soil P drawdown. This was supported by the positive correlations between TDP concentrations in runoff and Olsen P concentrations in the 0- to 5-cm soil depth (Table 3). In contrast with the patterns of TDP, the temporal trends of the particulate P concentrations were complicated by tillage practices. Whereas in F10, where conventional tillage was used throughout the study, the particulate P concentrations in snowmelt runoff tended to decrease as soil P drew down, the particulate P concentrations in snowmelt runoff from F11 tended to increase during 2006 to 2014 as compared with 1997 to 2005 after a conversion from conservation tillage to conventional tillage (Table 2). As compared with conservation tillage, conventional tillage increased soil erosion, and the transport of sediment and particulate P in the runoff water (Baulch et al., 2019). A similar pattern was found for rainfall runoff from F11, where the particulate P concentrations appeared to increase after the change in tillage. Although the particulate P concentrations also increased in F10, it was mainly associated with some severe soil erosion events in 2010 rather than tillage.

    In fields, a number of factors can influence runoff P patterns including soil test P (Vadas et al., 2005), P application source, form, rate, and placement (Liu et al., 2016), crops and crop residues (Elliott, 2013), rainfall (Sharpley et al., 2008) and snowmelt regimes (Elliott, 2013), and hydrological processes (Gburek and Sharpley, 1998). Several factors changed over time in our fields. In addition to the changes in the amount of fertilizer P input and soil P, SWE and rainfall depths as well as crop types, tillage, and incorporation of crop residue also varied during the monitoring period and may have contributed to the variability of the TDP concentrations. For example, changes in application rates and P fertilizer placement may have affected TDP concentrations in rainfall runoff after P applications. The potentials of different types and amounts of crop residues to release P after freeze–thaw cycles (Elliott, 2013) may also have contributed to variable TDP concentrations in snowmelt runoff. However, the trends of TDP concentrations in F10, where conventional tillage was used and crop residue was incorporated almost consistently throughout the study, did show significantly lower TDP concentrations during the P drawdown period than before P drawdown (Table 2, Fig. 2). Moreover, notably, mean SWE and rainfall depths observed before and during P drawdown did not differ significantly and cover a similar range of environmental conditions (Table 2).

    Fields F10 and F11 were tilled differently during 1997 to 2005, and F11 had generally greater soil P concentrations than F10 (Fig. 1). Trends of TDP concentrations (Table 2) were consistent with the trends of soil Olsen P in the upper 5-cm soil depth with respect to both that when each field was considered separately and that when the two fields were compared (Fig. 1). With similar mean annual snowmelt runoff amounts in the two periods, the flow-weighted mean concentrations of TDP decreased by 50% in F10 and 64% in F11 during the P drawdown period as compared with the earlier period (Table 2). The results provide scientific support for the proposal that P drawdown can mitigate P losses from soils, a proposal for which long-term field research and research across regions have been lacking (van der Salm et al., 2009; Sharpley et al., 2013). Because of the ongoing variation in tillage, weather, and crop type that occurred between years over the course of our study period, we are unable to fully explore the secondary influence of these factors accounting for interannual variation in P runoff losses. Although SWE was shown to significantly affect annual flow-weighted mean TDP concentrations (Table 3), event-based analyses have demonstrated apparent trends of declining TDP concentrations during the P drawdown period as compared with before P drawdown, when runoff volumes were similar during the two periods (Fig. 2). This further confirms the impacts of soil P drawdown on the temporal trends of the annual TDP concentrations (Table 2). Nonetheless, prediction of the implications of future climate scenarios on P runoff losses will require targeted research to more fully evaluate these factors.

    Our results confirmed that crop harvests in combination with low P inputs significantly depleted plant available P pools in the soils (Fig. 1b), and that these changes can occur quickly. For example, the decline of Olsen P in the upper 5-cm soil depth occurred at 3.1 mg kg−1 yr−1 in F10 and 5.4 mg kg−1 yr−1 in F11 during the first 3 yr of P drawdown. We found that soil Olsen P concentrations were negatively correlated with the cumulative P depletion in both experimental fields, despite their different initial soil P concentrations associated with the historically contrasting tillage methods (Fig. 1a). Across the literature, studies have reported varying rates of decline in soil test P following an increase in P depletion (van der Salm et al., 2009; Sharpley et al., 2013; Svanbäck et al., 2015), which highlight the need of more research on understanding the mechanisms controlling soil P drawdown in different soils. In this context, computational models could be a cost-effective tool for assessing the effects of soil P drawdown across a large range of climates and soils (Vadas et al., 2018).

    At the start of the soil P drawdown practice in 2007, the soil Olsen P concentrations in the 0- to 15-cm soil depth were 16 mg kg−1 in F10 and 21 mg kg−1 in F11 (Fig. 1b-3 and 1b-4), which fell in high and very high soil P fertility categories, respectively, according to the fertilizer recommendations in Manitoba (Manitoba Agriculture, 2007). After 7 yr of P drawdown, the Olsen P concentrations had declined to 8 mg kg−1 in F10 and 13 mg kg−1 in F11 (Fig. 1b-3 and 1b-4), corresponding to low and moderate soil P fertility categories, respectively. According to the recommendations for cereal and canola production, F10 needs to be supplied with 17 kg P ha−1 yr−1 and F11 needs 10 kg P ha−1 yr−1. These application rates will continue to deplete P from the soils if the current crop P removal rates of 17 kg P ha−1 yr−1 by wheat and 27 kg P ha−1 yr−1 by canola continue and may lead to further decline in runoff TDP concentrations over long term. However, it should be noted that although soil P drawdown is aimed at mitigating the impacts of soil P loss on water quality, it should be targeted at maintaining soil P fertility at an appropriate level so that it does not harm crop yields. From this standpoint, lowering soil Olsen P to a low soil P fertility category (i.e., F10) may not be a preferred agronomic and economic option. Low soil P fertility in cold climates is especially concerning because cold conditions after seed germination can impede P acquisition (Sheppard and Racz, 1984; Grant et al., 2001).

    In this study, notably, soil P depletion and drawdown did not decrease wheat or canola yield in either field. In fact, the yield of wheat significantly increased during 2006 to 2014 as compared with that in 1997 to 2005 (Table 2), which may be due to more favorable weather conditions, improved genetics, and patterns of N fertilization. This indicates that P was not a limiting factor for wheat or canola production under the conditions observed at our study sites. In conclusion, crop harvests in combination with low P inputs can be an effective strategy for removing P from high-P soils without trading off crop yields, with water quality benefits apparent over timescales of years.

    Conclusions

    In cold regions such as the Canadian prairies, few options are available to reduce runoff P from agricultural soils. Thus, identifying effective management strategies are especially important. In this study, soil P drawdown by lowering fertilizer P inputs showed a large potential to reduce dissolved P concentrations in both snowmelt runoff and rainfall runoff in two paired Manitoba fields, without affecting crop yields. The finding provides direct evidence that managing soil P can improve the regional water quality in the Canadian prairies where relevant research has been scarce. Decreases in soil P and runoff P concentrations can be observed over relatively short timescales measured in years, suggesting the large potential of adopting the soil P drawdown strategy as an effective strategy for P loss reduction. The effect of soil P on runoff P can interact with other factors, including hydrology and cropping systems, emphasizing the need for further work on this topic across regions. Moreover, soil P content below an agronomic optimum value may harm crop production and increase fertilizer expenses. Thus, further research is needed to identify soil nutrient management strategies that can simultaneously satisfy water quality, food security, and farm economy, although this work suggests that reducing P loads is achievable via P management and drawdown, without affecting food production.

    Supplemental Material

    Supplemental Fig. S1 is provided as supplemental material. The figure shows a map and experimental setup of the paired fields (F10 and F11) in the South Tobacco Creek watershed in Manitoba, Canada.

    Conflict of Interest

    The authors declare no conflict of interest.

    Acknowledgments

    This research was funded by Agriculture and Agri-Food Canada (through A-Base project funding and the Watershed Evaluation of Beneficial Management Practices project), Environment and Climate Change Canada (Lake Winnipeg Basin Initiative), and by the Canada First Excellence Research Fund (Global Water Futures Program–Agricultural Water Futures project). Field support was provided by Deerwood Soil and Water Management Association personnel, B. Turner, D. Cruikshank, and K. Hildebrand. We also thank the cooperating producer on the farm where our research was conducted.