Journal list menu

Volume 48, Issue 4 p. 841-849
Special Section: Agricultural Water Quality in Cold Environment
Open Access

Landscape Controls on Nutrient Export during Snowmelt and an Extreme Rainfall Runoff Event in Northern Agricultural Watersheds

Henry F. Wilson

Corresponding Author

Henry F. Wilson

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

Corresponding author ([email protected]).Search for more papers by this author
Nora J. Casson

Nora J. Casson

Dep. of Geography, Univ. of Winnipeg, Winnipeg, MB, R3B 2E9 Canada

Search for more papers by this author
Aaron J. Glenn

Aaron J. Glenn

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

Search for more papers by this author
Pascal Badiou

Pascal Badiou

Ducks Unlimited Canada, Institute for Wetland and Waterfowl Research, Stonewall, MB, R0C 2Z0 Canada

Search for more papers by this author
Lyle Boychuk

Lyle Boychuk

Ducks Unlimited Canada, Institute for Wetland and Waterfowl Research, Stonewall, MB, R0C 2Z0 Canada

Search for more papers by this author
First published: 01 July 2019
Citations: 20

Assigned to Associate Editor Jon Duncan.

Supplemental material is available online for this article.


In the northern Great Plains, most runoff transport of N, and P to surface waters has historically occurred with snowmelt. In recent years, significant rainfall runoff events have become more frequent and intense in the region. Here, we examine the influence of landscape characteristics on hydrology and nutrient export in nine tributary watersheds of the Assiniboine River in Manitoba, Canada, during snowmelt runoff and with an early summer extreme rainfall runoff event (ERRE). All watersheds included in the study have land use that is primarily agricultural, but with differing proportions of land remaining as wetlands, grassland, and that has been artificially drained. Those watersheds with greater capacity for storage of water in surface depressions (noneffective contributing areas) exhibited lower rates of runoff and nutrient export with snowmelt. During the ERRE, higher export of total P (TP), but not total N, was observed from those watersheds with larger amounts of contributing area that had been added through artificial surface drainage, and this was associated primarily with higher TP concentrations. Increasing or restoring the storage of water on the landscape is likely to reduce nutrient export; however, the importance of antecedent conditions was evident during the ERRE, when small surface depressions were at or near capacity from snowmelt. Total P concentrations observed during the summer ERRE were as high as those observed with snowmelt, and N/P ratios were significantly lower. If the frequency of summer ERREs increases with climate change, this is likely to result in negative water quality outcomes.

Core Ideas

  • Rainfall runoff P concentrations and export were as high as with snowmelt.
  • Watersheds with drainage ditches had lower N/P export ratios in summer.
  • Drained watersheds show greater export of P with extreme summer rainfall.
  • Restoring water storage potential on the landscape may reduce P export.


  • ARW
  • Assiniboine River watershed
  • AWC
  • available water capacity
  • ECA
  • effective contributing area
  • ERRE
  • extreme rainfall runoff event
  • FWMC
  • flow-weighted mean concentration
  • PLS
  • partial least squares
  • SOC
  • soil organic carbon
  • TN
  • total nitrogen
  • TP
  • total phosphorus
  • VIP
  • variable importance factor
  • High rates of loading of P from nonpoint sources cause eutrophication and associated harmful algal blooms in inland waters. Eutrophication associated with loading of N is also a global concern for ocean ecosystems (Schindler et al., 2016). An extensive body of research has been focused on mitigation of this loading through the management of agricultural inputs (Kleinman, 2005). Interactions at a watershed scale between agricultural management, ongoing alteration to surface drainage network structure, and changes in the frequency and timing of extreme hydrological events have received less attention, but in the face of ongoing climatic changes, these considerations will be particularly important for the development of management strategies (McCullough et al., 2012; Schindler et al., 2012; Baulch et al., 2019).

    The eutrophication of aquatic ecosystems is a significant environmental problem within the northern Great Plains (Baulch et al., 2019). Eutrophication in the region is characterized by harmful algal blooms and driven by high nutrient concentrations, particularly of P. Soils and aquatic systems in the region are naturally fertile with relatively high concentrations of P (Barica and Allan, 1988); however, since the 1970s, eutrophication issues have continued to increase. A trend of increasing P loading from the watershed and an increased frequency of harmful algal blooms has been observed in regional water bodies including Lake Winnipeg, the Red River (McCullough et al., 2012; Bunting et al., 2016), Lake Manitoba, and the Assiniboine River (Jones and Armstrong, 2001; Page, 2011).

    High rates of N and P export have been linked with climatic changes that have increased the frequency of high-flow and flooding events (McCullough et al., 2012). Most streamflow in the region is generated in spring with snowmelt over frozen soils. However, increases in the proportion of precipitation as rainfall and the number of multiday rain events have been observed at multiple sites throughout the Canadian prairie portion of the northern Great Plains between 1951 and 2000 (Shook and Pomeroy, 2012). Multiple studies have also suggested that flood volume and peak discharge rates in the region have been amplified by an expanded network of agricultural drainage ditches and loss of surface water storage capacity in depressional or wetland features (Dumanski et al., 2015; Szeto et al., 2015; Blais et al., 2016). Drainage of surface depressions has occurred extensively within the northern Great Plains, and estimates of overall loss of wetlands for agricultural use within the Canadian prairie pothole portion of the region range between 40 and 71% (Federal Provincial and Territorial Governments of Canada, 2010).

    Surface depressions and wetlands in the region have historically retained large amounts of N and P transported from upland areas through runoff and erosion. This has created a pattern of elevated P concentrations in the soil and sediment of surface depressions (Wilson et al., 2016; Brown et al., 2017; Badiou et al., 2018). The fertile nature of agricultural soils in the region and the presence of many nutrient-rich depressions has created conditions where limitation of export tends to occur by transport rather than source availability (Ali et al., 2017). Together, the potential for increased volume of runoff and greater connectivity of nutrient-rich landscape features after drainage of surface depressions has created concern that N and P export with runoff events will be elevated. Understanding of interactions between landscape modification and climatic changes in the northern Great Plains has been limited by the availability of data to characterize artificial drainage and hydrology at a regional scale (Baulch et al., 2019). How watershed landscape characteristics and modification of that landscape influence biogeochemical and hydrological response to extreme climatic events at broader spatial scales has not yet been characterized in the region.

    The research presented here focuses on interactions between land use, artificial surface drainage, and hydrological extremes in tributaries of the Assiniboine River watershed (ARW), which drains 182,000 km2 of Saskatchewan, Manitoba, and North Dakota (Supplemental Fig. S1). The ARW is located within the broader Lake Winnipeg watershed. Land use within the ARW is predominantly agricultural and is typical of the northern Great Plains. Climate changes observed in the ARW are also representative of changes that are occurring more broadly throughout the region. For the May–June period in the ARW the average rainfall, number of consecutive days with rain >1 mm, and proportion of rain occurring on days with rain rated in the 95th percentile have increased significantly since 1960, and this pattern may be linked to anthropogenic climate forcing (Szeto et al., 2015). In the ARW, peak flow rates observed in 2011 and in 2014 were the highest recorded since hydrometric measurement was initiated in 1913. Flooding in both 2011 and 2014 were driven by high amounts of spring rainfall in combination with already high snowmelt flow (as in 2011) and wet antecedent moisture conditions (in both 2011 and 2014) (Ahmari et al., 2016; Blais et al., 2016).

    The objective of this study was to evaluate patterns of water, N, and P export across nine tributary watersheds within the ARW in Manitoba, Canada, with varying degrees of watershed landscape modification. Responses of these tributaries during both the snowmelt and the rainfall-driven flood events of 2014 are presented. An extreme rainfall runoff event (ERRE) in 2014 was unique in that the event was entirely due to rainfall, with streamflow from a large snowmelt event having already receded (Ahmari et al., 2016). Also, rainfall received over the lower ARW was relatively consistent duration, amount, and intensity. For this reason, the 2014 yr was well-suited for a multiwatershed analysis to address two major research questions:
    1. Does the influence of watershed landscape characteristics on stream hydrology and biogeochemistry differ between large snowmelt and rainfall driven runoff events?
    2. Do the watershed landscape characteristics influencing export differ for water N and P?

    Materials and Methods

    Study Area

    The watersheds of focus for this study are located within the ARW in Manitoba (Fig. 1). Prior to the addition of surface drainage ditches to agricultural systems in the ARW, large areas of the watershed were hydrologically disconnected and would retain rather than contribute water to the downstream drainage network under normal conditions (Godwin and Martin, 1975). Study watersheds were selected to include landscapes that are naturally well drained with fewer wetlands, those that include many wetlands and have capacity to store larger volumes of water, and those that have been extensively altered through the addition of surface drainage ditches (0–44% of watershed area as effectively contributing area added by drainage, Table 1). The history of agricultural development in these watersheds, delineation of watershed boundaries, and calculation of slope is described in greater detail in the Supplemental Methods section.

    Details are in the caption following the image

    Panel a shows the Assiniboine River watershed and the location of tributary watersheds and Environment and Climate Change Canada meteorological stations included the study. Political boundaries and major water bodies are also indicated. Panel b shows tributary watershed boundaries and location of meteorological stations at a finer scale. A gray, dashed circle is shown in both panels for spatial reference.

    Table 1. Characteristics of nine tributaries to the Assiniboine–Souris River system in Manitoba that were included in this study.
    Characteristic Abbreviation Min. Max. Mean Median CV
    Watershed area (km2) Area 64.80 1165.00 406.31 333.00 83.18
    Effective contributing area added by drainage ditches (%) ECA DD 0.04 44.90 16.48 13.84 102.46
    Naturally effective contributing area (%) ECA Nat 8.50 98.33 43.33 40.70 62.07
    Total modern effective contributing area (%) ECA Mod 12.42 98.87 60.26 65.77 41.97
    Available water capacity (mm, 0–15 cm) AWC 24.31 32.88 27.82 27.76 9.20
    Clay mass fraction (g kg−1, 0–15 cm) Clay 226.8 342.8 290.8 306.3 157.1
    Organic carbon (g kg−1, 0–15 cm) SOC 31.33 50.25 40.36 39.98 16.81
    Sand (g kg−1, 0–15 cm) Sand 303.7 502.5 357.3 322.3 211.8
    Saturated hydraulic conductivity (cm h−1, 0–15 cm) Ksat 2.97 4.06 3.23 3.06 11.62
    Cropland (%) Cropland 48.90 82.90 64.37 62.00 17.47
    Deciduous forest (%) Treed 0.12 6.89 3.98 4.07 62.18
    Grassland and pasture (%) Grass 11.52 35.80 21.19 21.00 38.06
    Mean slope (%) Slope 0.75 1.50 1.13 1.19 21.58
    Roads and built areas (%) Roads 0.78 3.58 2.71 2.86 29.77
    Open water (%) Water 1.90 5.00 3.36 3.48 32.74
    Wetland (%) Wetland 2.72 17.00 8.22 5.81 63.69

    Characterization of Watershed Drainage

    The Prairie Farm Rehabilitation Administration delineated and mapped contributing areas representative of average runoff conditions (2-yr return period) for the Canadian prairie provinces beginning in the 1970s based on surface topography, density of the natural stream network, number and size of wetlands, consultation with local residents, and to reflect conditions prior to installation of drainage ditches (Martin, 2001). However, the density of surface drainage ditches and stream networks have increased significantly in many portions of the ARW since the 1970s, and the portion of these watersheds that might be considered noncontributing under normal conditions (for an event with a return period of 2 yr) has been reduced. Updates of contributing area delineations to reflect modern conditions have not been completed for most of the prairies, but in support of the Canadian Wetland Inventory and related wetland monitoring programs, Ducks Unlimited Canada has delineated surface drainage ditches throughout a large portion of the ARW, and these data were used to calculate modern effective contributing area (ECA) as outlined in the Supplemental Methods. Throughout the study, ECAs are expressed as a percentage of total watershed area (Table 1).

    Characterization of Watershed Soils and Land Cover

    Soils in the study region were formed on gently undulating or kettled calcareous glacial till. Soil characteristics in each watershed were quantified using the 90-m-resolution Gridded Soil Landscapes of Canada data product from the Canadian Soil Information Service (2016), generated in support of the GlobalSoilMap initiative. This dataset provides a variety of soil attributes separated over multiple depths. In the current study, watershed soil characteristics were quantified for the 0- to 15-cm and 15- to 30-cm depths for soil organic C (SOC), pH, clay mass fraction (clay), silt mass fraction, sand mass fraction (sand), bulk density, available water capacity (AWC), electrical conductivity, and saturated hydraulic conductivity (Ksat). Some of these variables were found to differ very little between watersheds, particularly for the 15- to 30-cm depth, so only data for SOC (g kg−1), clay (g kg−1), sand (g kg−1), AWC (mm over depth range), and Ksat (cm h−1) for the 0- to 15-cm depth are presented (Table 1).

    Land cover within each watershed was classified based on the 2006 edition of Land Use/Land Cover Landsat TM Maps from the province of Manitoba (Manitoba Remote Sensing Centre, 2008). Land covers are reported percentages of watershed area under annual cropland, deciduous forest, grassland and pasture, roads and built areas, open water, and wetlands (Table 1).

    Streamflow Measurement and Water Sampling

    Water sample collection and manual measurements of streamflow were completed on 12 dates over the ice-free season and across all sites in 2014. Timing of collection was bimonthly between hydrologic events with more frequent sampling occurring during events to ensure collection of samples on the rising limb, near peak flow, and on the receding limb of event hydrographs (Supplemental Fig. S2). Using this sampling strategy, four water samples were collected at each site during snowmelt in 2014, and four samples were collected at each site over the course of the ERRE that occurred in late June 2014. This approach was used to allow for sampling of a greater number of watersheds on a consistent time step because distances between sampling locations were significant (Fig. 1), requiring multiple days to complete sampling and measurement at each of the nine locations. Also, this strategy was selected because concentrations of P (a primary driver of eutrophication regionally) measured in 2012 and 2013 were observed to follow a consistent seasonal pattern at most sites (Ali et al., 2017; Casson et al., 2019) with a weaker response to changes in flow (noticeable in Supplemental Fig. S2).

    To evaluate the consistency of results generated using differing sampling frequencies, three of the nine study watersheds were sampled at much higher frequency (25–50 samples during snowmelt, 29–32 samples during the ERRE). These three watersheds were selected to ensure contrasting levels of contributing area.

    Streamflow Data

    Streamflow was measured by the Water Survey of Canada near the location of water sampling in five of the nine study watersheds (Supplemental Table S1). For these five sites, daily streamflow data for 2014 were obtained from the online hydrometric database of the Water Survey of Canada ( At the four remaining sampling locations flow was measured by Agriculture and Agri-food Canada and a stage-rating curve was generated using manual streamflow measurements in combination with depth measurements from pressure transducers (Onset HOBO U20-001-04) placed in submerged permeable casings that were anchored in place upstream of a bridge or large culvert (Supplemental Methods).

    Water Sampling and Analysis

    Water samples were collected from a well-mixed location with high velocity. Samples were collected into prewashed polycarbonate bottles and were kept on ice and filtered within 24 h of collection using a precombusted glass fiber filter (nominal pore size = 0.7 μm). Samples were stored and analyzed using established methods (Supplemental Methods) for the determination of total P (TP), total dissolved P, total N (TN), and total dissolved N, as well as nitrate + nitrite (NOx).


    Winter precipitation data were obtained from Environment and Climate Change Canada online records ( for the six meteorological stations with measurement of winter precipitation and located <100 km from the centroid of any study watersheds (Fig. 1). Rainfall estimates were obtained from integrated multisatellite retrievals for Global Precipitation Measurement (IMERG [Integrated Multisatellite Retrieval for Global Precipitation Measurement]) data available through NASA on a 0.1° grid ( that combine data from all passive-microwave instruments in the constellation and precipitation gauge analyses. Additional detail of methods used for precipitation data retrieval and comparison between datasets is provided in the Supplemental Methods.

    Precipitation occurring between 1 Nov. 2013 and full recession of the snowmelt hydrograph for each watershed (ranging between 5 and 31 May) was summed as an indicator of precipitation input likely to contribute to snowmelt runoff. Increases in flow associated with the ERRE began with rainfall in June, so for each watershed, total precipitation received from just prior to the onset of increasing flow in June up until the recession flow at the end of July was summed to calculate total rainfall with potential to contribute to the ERRE that began in late June.

    Calculation of Water Yield, Runoff Ratio, and Chemical Export

    The total volume of water exported per unit watershed area (water yield, expressed as a depth in mm) was calculated using the sum of streamflow over the course of the snowmelt or of the ERRE. Runoff ratio was calculated as the ratio of water yield to precipitation contributing to each event (Supplemental Methods). Because clear concentration–discharge relationships were not observed for each stream and sample timing was based on flow, the export and flow-weighted mean concentration (FWMC) of all elements were calculated by linear interpolation of concentrations within each event (Supplemental Methods). The molar ratio of TN/TP exported during each event was calculated as an indicator of whether potential availability of each element differed between seasons.

    Statistical Analysis

    Relationships between response variables (export, runoff ratio, and stream chemistry) and predictor variables (watershed characteristics and weather) were evaluated using partial least squares (PLS) analysis. Predictor variables included contributing area, land cover, topography, soil, and weather factors and are listed in Table 2. Where transformation of variables was required to meet the assumption of normality, it is noted in the text. Centering and scaling were utilized and validation was completed using the leave-one-out method in the PLS platform of JMP 14 statistical software (SAS Institute, 2018). A large number of predictor variables exhibited variable importance factors (VIPs) >0.8 and could be considered to have a statistically significant influence on the response, but coefficients for most of these factors were low, indicating that the magnitude of influence was low. For this reason, we describe variables with VIP >1.2 as significant for the response of interest (Liu et al., 2013). To avoid overfitting, the best PLS model for each response variable was selected where the number of factors included in the model minimized root mean predicted sum of squares (PRESS) as determined by cross-validation. For each PLS model, VIP and regression coefficients are presented to indicate variable influence. The number of PLS factors and percentage variation explained for the best PLS models are also noted.

    Table 2. Flow-weighted mean concentrations, total P (TP), total dissolved P (TDP), total N (TN), total dissolved N (TDN), nitrate + nitrite (NOx) N/P ratio of export, water yield, and runoff ratio as measured during snowmelt and for an extreme summer rainfall runoff event in 2014 in nine tributaries of the Assiniboine River in Manitoba, Canada.
    Location n TP TDP TN TDN NOx Molar N/P Water yield Runoff ratio
    mg L−1 mm %
    2014 snowmelt
    Bailey's Creek 4 0.53 0.48 1.77 1.69 0.29 7.7 22 14
    Elgin Creek 4 0.70 0.62 2.95 2.52 0.92 9.6 27 19
    Gopher Creek 4 0.35 0.31 1.87 1.67 0.52 12.2 32 20
    Little Souris River 4 0.52 0.45 3.12 2.57 0.74 13.7 34 24
    Minnewashta Creek 4 0.18 0.14 1.22 1.13 0.28 15.6 29 9
    Oakner Creek 4 0.45 0.41 1.52 1.31 0.36 7.8 56 22
    Oak River 4 0.41 0.34 1.80 1.50 0.28 10.2 40 15
    Willow Creek (east) 4 0.58 0.45 4.32 3.84 2.43 16.9 64 44
    Willow Creek (west) 4 0.63 0.46 6.20 5.75 4.19 22.5 42 22
    All site mean 0.48 0.41 2.75 2.44 1.11 12.9 39 21
    Oak River 25 0.34 0.29 1.66 1.37 0.36 11.1 40 15
    Willow Creek (east) 50 0.57 0.48 4.18 3.82 2.22 16.7 64 44
    Willow Creek (west) 50 0.51 0.44 4.41 4.06 2.62 19.6 42 22
    2014 extreme rainfall event
    Bailey's Creek 4 0.56 0.52 1.48 1.45 0.10 7.0 56 25
    Elgin Creek 4 0.81 0.76 1.68 1.41 0.18 5.5 52 21
    Gopher Creek 4 0.45 0.43 1.51 1.24 0.24 8.8 66 32
    Little Souris River 4 0.33 0.22 1.63 1.38 0.20 13.2 50 22
    Minnewashta Creek 4 0.31 0.27 1.44 1.38 0.06 12.5 37 18
    Oakner Creek 4 0.88 0.70 1.43 1.33 0.12 4.3 67 31
    Oak River 4 0.51 0.44 1.34 1.27 0.07 7.1 47 23
    Willow Creek (east) 4 0.49 0.41 1.40 1.37 0.07 7.6 36 18
    Willow Creek (west) 4 0.54 0.43 1.34 1.27 0.06 6.6 49 24
    All site mean 0.54 0.47 1.47 1.34 0.12 8.1 51 24
    Oak River 29 0.59 0.55 1.80 1.56 0.22 8.1 47 23
    Willow Creek (east) 30 0.44 0.32 1.65 1.49 0.10 10.0 36 18
    Willow Creek (west) 32 0.51 0.49 1.57 1.46 0.12 8.2 49 24
    • Values calculated using higher frequency sampling at three study watersheds.

    Predictor variables influencing response variables are also described using ordinary least squares linear regression on univariate relationships between response variables and predictors with the highest VIPs. Regression analyses were performed in JMP 14 using the fit y by x platform.


    Precipitation and Runoff

    Precipitation prior to the end of snowmelt (1 November to 30 April) ranged between 139 and 260 mm (Supplemental Table S1). Winter precipitation at each Environment and Climate Change Canada weather station was near the normal observed for the 1981 to 2010 period; however, antecedent soil moisture as measured in fall 2013 for the top 0 to 30 cm was 80 to 100% of available water-holding capacity throughout the study area (Riekman, 2013). When combined with a late spring melt, these conditions resulted in relatively high peak flows at snowmelt (Fig. 2), but duration and volume totals were low enough that only minor snowmelt flooding occurred in the watersheds studied.

    Details are in the caption following the image

    Discharge rate measured by the Water Survey of Canada for the Oak River, a tributary to the Assiniboine River located in Manitoba. Rates are shown for the year 2014, the maximum recorded on any given date over the period of record, and the mean recorded for each date over the period of record.

    Precipitation volumes in May and June were relatively high in all watersheds. Meteorological systems bringing this rainfall covered the entire study area. In June and into July of 2014, a large low-pressure system with heavy rain occurred (Ahmari et al., 2016). This weather system in combination with wet antecedent conditions resulted in the ERRE of 2014, with 240 to 360 mm of rain occurring in each watershed between 1 May and 30 June (Supplemental Table S1). The total rainfall accumulation observed in widespread rainfall events in June was very similar between watersheds and ranged between 172 and 204 mm.

    Water Yields and Runoff Ratio

    Water yields and runoff ratio during snowmelt of 2014 were similar to those for the early summer ERRE (Table 2). Results of PLS analysis (Supplemental Table S2) indicate that factors influencing storage potential in surface depressions and connectivity (contributing area) or infiltration and direction of water to storage in soils (clay, SOC, and sand) were important in influencing water yield during snowmelt. Only those factors reflecting overall potential for storage in surface depressions (natural and modern ECAs) or land covers of greater or lesser abundance in nondrained landscapes (cropland and wetland) were identified as critical factors in the prediction of runoff ratio. Although natural and modern ECAs are correlated, the univariate relationship between the natural log of snowmelt runoff ratio and modern ECA was stronger (r2 = 0.74, p < 0.01 vs. r2 = 0.69, p < 0.01) because the two watersheds with the most extensive artificial drainage exhibited much higher runoff ratios than would be predicted based on natural ECA (Fig. 3).

    Details are in the caption following the image

    Univariate relationships are shown between best landscape or weather predictor variables and runoff ratio, total P load, or flow-weighted mean total P concentration as measured in nine tributaries of the Assiniboine River in 2014 during snowmelt and an extreme rainfall runoff event (ERRE). Where significant correlation was observed between the best univariate predictor variable and a secondary predictor identified as important with partial least squares analysis, the value for the second predictor is indicated using color coding (blue to red) as noted by the scale in each panel. Triangles indicate values recalculated for three sites using higher frequency chemistry and are plotted for comparison, but high-frequency measurements were not used to calculate the trendlines displayed in each panel. All regression coefficients are for simple linear regression, with the exception of that for snowmelt runoff ratio, where natural log transformation was applied for normality.

    Multiple factors influenced water yield and runoff ratio during the 2014 ERRE with fewer potential links identified to landscape drivers than for snowmelt (Supplemental Table S2). Runoff ratio and water yield were lower where more road or open water area was present in the watershed. Runoff ratio and water yield were higher where watersheds had a greater ECA added by artificial drainage ditches and in watersheds with less AWC in the soil (Fig. 3).

    Water Chemistry

    In 2014, FWMCs of both total and dissolved forms P were slightly higher for the ERRE than for snowmelt across study watersheds in the ARW, and FWMCs for all forms of N were lower during the ERRE (Table 2). High concentrations of P observed for both snowmelt and the ERRE, along with lower concentrations of N in summer, resulted in lower N/P of export from all watersheds during the ERRE (Table 2).

    The FWMCs of total dissolved P and TP that were calculated with lower frequency sampling (four hydrograph-targeted samples per site per event) were within 20% of those calculated with a higher frequency sampling strategy (25–50 samples per event). Calculated FWMCs for N forms varied more significantly with 40 to 60% higher values being calculated with the lower frequency sampling strategy at the high N concentration Willow Creek (west) site for snowmelt and an underestimation of NOx of 68% at the Oak River during the ERRE (Table 2). This pattern indicates that higher frequency sampling is required to fully capture N concentration changes over the course of hydrological events in the region. The lower frequency sampling strategy added uncertainty to FWMC and export values for all elements sampled, but for N species in particular, this caveat should be kept in mind in the interpretation of PLS analyses to identify influence of watershed characteristics. The N/P ratios calculated using the two sampling strategies were within 20% and driven largely by concentrations of P. For all evaluation of univariate relationships between water chemistry and environmental predictors, those values calculated using higher frequency sampling are plotted to ensure consistency and to illustrate the potential impact of added uncertainty (Fig. 3).

    The Influence of Watershed Landscape Characteristics on Nitrogen and Phosphorus Export

    Nitrogen and P were predominantly present in dissolved forms for both snowmelt and the ERRE, so results of PLS analysis to identify important factors influencing export and concentration were nearly identical for dissolved and particulate forms of the same element (Supplemental Table S2). The export of N and P with snowmelt was primarily driven by the volume of runoff, with the most important PLS factors being those associated with storage of water in surface depressions (modern and natural ECAs).

    Both the concentration and export of P observed during the ERRE related strongly to ECA added by drainage ditches (Fig. 3). A weaker influence on TP load and FWMC was also observed for grassland and pasture area. Those factors most strongly influencing TN loads during the ERRE were the same as those influencing water yields. The FWMCs of TN, total dissolved N, and NOx observed with the ERRE were lower and showed less variability between sites than those observed with snowmelt. For this reason, the volume of runoff appears to have been the primary factor in defining rates of export during the ERRE.

    No significant univariate predictor of snowmelt N/P was identified (p > 0.05); however, for the ERRE, the best univariate predictor of N/P was drainage ditch ECA (r2 = 0.52, p < 0.03) and the same as that observed for FWMC TP. Together these results indicate that in watersheds where P concentrations were elevated with export, proportional increases in the export of N were not observed.


    The Importance of Antecedent Conditions and Landscape Characteristics in Defining Water Yields and Runoff Ratio

    The lower water yield and runoff ratio with snowmelt may partially be a function of sublimation losses over winter (Ehsanzadeh et al., 2012), but storage of snowmelt in surface depressions is also likely to have contributed to reductions. The increased runoff ratio in watersheds with intensive artificial drainage (Fig. 3) suggests that ditching to drain surface depressions has increased the volume of runoff occurring with snowmelt in the ARW. Unexplained variation in runoff ratio prediction may result from the modern ECA metric being a static estimate. In reality, ECA is of dynamic nature, changes with antecedent conditions, and is influenced by the size and distribution of depressions (Spence, 2007; Ehsanzadeh et al., 2012; Ali et al., 2018). Detailed studies occurring over time in other small watersheds in the ARW (Dumanski et al., 2015) rather than across space (as in the current study) have also shown increases in runoff per unit precipitation along with the loss of surface depression storage. Together, these results indicate that efforts to reduce wetland drainage and to restore or develop new surface water storage on the landscape are likely to reduce volumes of runoff with snowmelt.

    Late snowmelt combined with precipitation in May and early June resulted in nearly full surface depression storage in much of the watershed ahead of the late June ERRE (Ahmari et al., 2016). These conditions may account for the lack of relationship between runoff ratio and overall ECA for the ERRE. Runoff ratio and water yield were lower where greater areas of road and open water are present in the watershed. This may relate to forms of surface water storage not well accounted for in the ECA characterizations:
    1. Larger lakes and wetlands (more open water) have been identified as having the potential to store larger volumes of water than smaller surface depressions and for “gate-keeping” type effects on water volume moving downstream (Shook and Pomeroy, 2011).
    2. In the landscape of the ARW, most water passes through the road network in culverts, many of which are small enough in diameter that flow occurring during the ERRE was restricted, resulting in extensive back-flooding, into farm fields, filling of ditches, and increasing the potential for infiltration of ponded water. Examples such as that shown in Supplemental Fig. S3 were extremely common throughout those watersheds studied and led to numerous road closures with hundreds of communities declaring states of emergency throughout western Manitoba during the ERRE (Healy, 2014).
    3. The negative association between AWC of watershed soils and runoff ratio suggests that even under high-moisture conditions, water to storage in soils remains an important component of the regional water balance (Ehsanzadeh et al., 2012). In contrast, greater area of ECA added by artificial drainage ditches is likely to increase drainage density and the rate at which water is conveyed downstream, as suggested by the positive association of drainage ditch ECA with both water yield and runoff ratio (Fig. 3).

    Seasonal Differences in Water Chemistry

    Given the changes in chemistry that have been observed for smaller rainfall runoff events in the northern Great Plains region, it has been suggested that in the absence of point source inputs, concentrations of N and P are highest with snowmelt after seasonal release from frozen vegetation and soils (Li et al., 2011; Rattan et al., 2017). However, the higher concentrations of P observed during the ERRE indicate that for events of similar magnitude, the potential for elevated concentrations of P with large rainfall-driven runoff events is equal to that of snowmelt in the region. Soil P is relatively abundant in agricultural soils of the ARW, and P concentrations in surface runoff have been observed to scale with the amount of soluble P at the soil surface for both snowmelt (Jensen et al., 2011) and rainfall-driven surface runoff (Duncan et al., 2017). The presence of a relatively large source pool of these elements in each watershed and consistently high concentrations with both snowmelt and the ERRE suggest the importance of transport, solubility, and mobilization-related processes (Ali et al., 2017). Those factors controlling N transport with snowmelt are unique from rainfall, and lower FWMCs with the ERRE may be the result of soluble N loss over the course of snowmelt and a continued decline under wetter and warmer spring conditions that are conducive to denitrification and assimilation (Wagner-Riddle et al., 1997; Costa et al., 2017; Satchithanantham et al., 2019).

    Consistently high concentrations of P were observed for both snowmelt and the ERRE, but lower N concentrations resulted in a reduced ratio of N/P with the ERRE (Table 2). It is important to note that N-fixing cyanobacteria species tend to be favored under low-N/P conditions (Schindler et al., 2016) such as those observed with the ERRE, and that these algae produce harmful toxins such as microcystin (Orihel et al., 2012). However, evaluation of N/P of export over a wider range of hydrologic events is required before broader generalizations can be made about implications for N/P of seasonal timing of extreme runoff events.

    Seasonal Differences in Watershed Landscape Characteristics Influencing Water Chemistry and Nutrient Loads

    Loads of TN and TP exported during snowmelt were primarily controlled by the volume of water leaving each watershed and those landscape factors controlling it; however, variation in TN and TP concentration was observed to occur with additional landscape factors. The positive association of FWMC TP and TN with cropland, but the negative association with wetlands (Supplemental Table S2), may reflect the greater likelihood for accumulation of soluble N and P in soils of land receiving annual fertilizer application (Carpenter et al., 1998), the reduced potential for retention in wetlands and riparian buffers where high rates of historical loading have led to saturation (Satchithanantham et al., 2019), and the tendency for higher concentration in runoff water from watersheds in the region with greater soil P (Salvano et al., 2009). However, the positive univariate relationship between cropland and snowmelt FWMC TP was relatively weak (r2 = 0.5, p = 0.03) compared with an observed negative association with winter precipitation (r2 = 0.75, p < 0.01; Fig. 3). This negative relationship between FWMC TP in snowmelt and winter precipitation suggests that dilution of concentration may occur where release of water from a large and relatively low-P-content snowpack may exceed rates of release from soil or vegetation. Both snow retention and soil P are factors that can be influenced by agricultural management, so a finer scale study of the mechanistic links between these factors and snowmelt export within these agricultural watersheds could help to inform future management decisions.

    The PLS results for FWMC of TN during the ERRE indicate a potential connection to factors that may influence denitrification rates, including soil properties that can influence the development of anoxia under wet conditions (clay, AWC, and sand) and the availability of C (SOC) (Bouwman et al., 2002; Rochette et al., 2008). However, the higher level of uncertainty for FWMC of TN associated with sampling frequency makes further interpretation challenging.

    The most important landscape predictor of TP load and FWMC with the ERRE was drainage ditch ECA (Fig. 3). The results presented here also suggest that disproportionate loss of P versus N with runoff might be anticipated from drained landscapes with ERREs. This pattern is consistent with observed decreases in capacity for P sorption, and increased solubility of P in the soils of drained prairie wetlands (Brown et al., 2017; Badiou et al., 2018) without significant changes in TN or NO3 (Brown et al., 2017). Data are not available to identify the timing of wetland drainage within the watersheds studied here in the ARW, but it is likely that the time since drainage may be an important factor in defining variation not explained in relationships with metrics of ECA change (Brown et al., 2017). In addition, the extent to which agricultural management of drained lands is likely to influence N and P export remains to be defined. Rates and timing of fertilizer input are likely to influence extractability. Drained wetlands tend to remain low-lying areas and are susceptible to differing forms of flooding. The effectiveness of artificial drainage and the potential for release from fertilized but frequently flooded land are topics requiring further research to develop effective mitigation options (Baulch et al., 2019). The secondary influence of grassland and pasture area on P concentration and export may also reflect the nature of regional drainage management. Those portions of the landscape where seasonal excess moisture may limit utilization as cropland are often managed for grazing by cattle. However, where use of cropland has been prioritized over grazing, surface drainage to increase arable land area is a likely outcome. Landscapes with less intensive surface drainage (drainage ditch ECA) also tended to have larger areas maintained as grassland (r2 = 0.80, p = 0.04), so identification of a separate mechanism is not possible with the dataset presented here.


    This study highlights the importance of antecedent conditions in defining solute source and hydrologic connectivity and the need for continued measurement over a range of conditions to fully define those mechanisms that will control watershed responses to future events (McMillan et al., 2018). The development of management strategies to reduce water yield increases associated with artificial surface drainage and restoration or the development of new surface water storage capacity on the landscape may be required to reduce downstream excess water issues and nutrient transport with snowmelt in the ARW. Under the extremely wet antecedent conditions that preceded the ERRE of 2014, the surface water storage capacity was essentially full at the time of runoff generation, regardless of watershed ECA. Under these conditions, higher TP concentrations and associated loads occurred in watersheds with more extensive artificial drainage, but this was not observed for TN. If the frequency of summer ERREs increases in the region with climate change, this is likely to result in negative water quality outcomes. High concentrations and loading of P from the watershed were observed for both snowmelt and rainfall runoff, but proportionate increases in TN were not observed, resulting in lower N/P ratios.

    Supplemental Material

    In the supplemental material, further detail is provided on study location, watershed characteristics, sampling and analysis methods, water chemistry data, and precipitation data. Images are also provided of observed flooding conditions in upland and near-stream locations.

    Conflict of Interest

    The authors declare no conflict of interest.


    This research was supported by funding under Agriculture and Agri-Food Canada's Growing Forward 2 Program. Marilyn Makortoff provided technical assistance in the field and laboratory. Norma Sweetland provided technical assistance in the laboratory.