Phosphorus dynamics in agricultural surface runoff at the edge of the field and in ditches during overbank flooding conditions in the Red River Valley

Abstract Agricultural fields in the Red River Valley of the Northern Great Plains are located on flat clay soils, often drained by shallow, roadside ditches that are not graded and lacking relief. These conditions can result in flow reversals and subsequent flooding of adjacent fields during large runoff events, which can mobilize phosphorus (P). Surface runoff from two agricultural fields and their adjacent ditches was monitored from 2015 to 2017 in southern Manitoba, Canada. Overbank flooding of fields adjacent to ditches was observed in 5 of 21 hydrologic events, and such events dominated annual runoff and P budgets (>83% of losses over the 3-year study period). Flooding events were often dominated by soluble P fractions (57–67%) relative to events where flooding was not observed (39–63%). Concentrations of soluble reactive P in water standing on fields increased with time during flooding events, suggesting that P was mobilized during such events; however, the source of the soluble reactive P is not clear. This study has highlighted temporal differences in hydrologic and biogeochemical interactions between fields and ditches and demonstrated the need for an improved understanding of mechanisms of P mobilization in the landscape, which has direct implications for predicting P mobility in agricultural watersheds.


Introduction
Freshwater bodies in North America such as Lake Winnipeg in Manitoba, Canada, have been impacted by harmful and nuisance algal blooms. Agricultural runoff is considered the primary contributor to the current deterioration of the health of these freshwater bodies (Schindler, Hecky, and McCullough 2012). Elevated edge-of-field phosphorus (P) and nitrogen (N) losses typically occur upstream of Lake Winnipeg in the Red River Valley during major hydrologic events, which are most often associated with spring snowmelt (Kokulan et al. 2019a;Tiessen et al. 2010), but can also occur during large convective storms throughout the growing season (Kokulan et al. 2019a). The occurrence of large runoff events is expected to increase given that an increase in the frequency of rain-on-snow events and multi-day spring and summer storms has been observed in this region (Il Jeong and Sushama 2018;Shook and Pomeroy 2012). Given the disproportionate contribution of large events to annual nutrient losses from agricultural fields (Kokulan et al. 2019a), the more frequent occurrence of large events may be accompanied by increased runoff-related nutrient losses to water bodies. To better understand how agricultural systems will respond to an intensification of the hydrological cycle, field studies that explore sources and processes that drive P mobilization during large events are needed. This knowledge can provide insight into the efficacy of current management practices in mitigating P losses and can assist land managers in the selection of appropriate beneficial management practices under both contemporary and future climates.
Artificial drainage and drainage water management may play a central role in P losses from agricultural systems. Roadside ditches are an integral part of farm drainage systems that connect farms with regional streams (Buchanan et al. 2013) when runoff from agricultural fields flows into ditches through naturally existing or constructed surface swales (or surface drains). Historically, ditches have been examined from a water conveyance standpoint, but their role from a water quality standpoint is less understood. In addition to receiving runoff, ditches also receive sediments, dust and salt mixtures from nearby roads via erosion processes (Cade-Menun et al. 2017). The roles of roadside ditches and their surrounding riparian areas in P dynamics are unclear and appear to differ regionally (Moloney, Fenton, and Daly 2020;Smith et al. 2005;Van Nguyen and Maeda 2016). For example, studies have shown that P export in runoff is reduced through retention by ditch sediments and vegetation buffers in warmer temperate regions (Haggard, Smith, and Brye 2007), whereas recent work has shown that vegetated buffers are less effective in retaining P in cold regions (Kieta et al. 2018;Vanrobaeys et al. 2019), and may even act as P sources. Thus, an improved understanding of the role of ditch sediments and riparian areas in P mobilization or immobilization in the Red River Valley is needed.
The rapid addition of water from snowmelt or convective rain events to soils with low and/or impeded infiltration capacities, combined with the naturally flat topography of the Red River Valley region, leads to frequent flooding events (Schindler, Hecky, and McCullough 2012). Such occurrences are exacerbated by ice jams (in early spring) and the blockage of drainage ditches and culverts (yearround) by regional municipalities to manage the risk of flooding in the downstream areas of the Red River (Wazney and Clark 2016). Impeded drainage prompts roadside ditches to fill and back up into adjacent fields, causing periods of inundation or overbank flooding which last from several days to weeks. These overbank flooding events have implications for P exchange processes between the soil, plants, and the water column, spanning from inundated areas of fields through riparian zones and into ditch soil, sediments and vegetation (Schindler, Hecky, and McCullough 2012). Although it is known that much of the P loss into ditches and tributaries in the Red River Valley is supplied by overland flow (Kokulan et al. 2019a;Tiessen et al. 2010), it is not clear if this is overland flow directly contributed by fields, or, if it occurs as a result of the prolonged inundation of the ditches and the vegetation associated with the ditch, which increases the opportunity for P release to floodwaters. Indeed, the blockage of ditches to control flooding in downstream regions may, in fact, exacerbate P loss from agricultural fields in the Red River Valley by causing overbank flooding, accelerating the transfer of field soil P into ditches.
Biogeochemical processes that operate at the soilsolution interface determine the mobility of P (Devau et al. 2011); however, these processes may be impacted by flooding conditions (Amarawansha et al. 2015), as flooding increases the contact time between P-rich soil pools and more dilute snowmelt or rainwater, potentially leading to the mobilization of P to runoff via diffusion. Prolonged inundation of soils also affects the equilibrium phosphate concentration (EPC) by decreasing the redox potential and subsequently changing the pH of the environment. In alkaline environments such as the Red River Valley, a decrease of redox potential favours the dissolution of CO 2 , thereby shifting the pH towards neutrality (Amarawansha et al. 2015). This, in turn, affects the stability of P complexes and elevates P release from soils to the water column. Laboratory incubation studies indicate that the majority of soils in Manitoba are prone to increased P loss under periods of prolonged inundation when undisturbed (Amarawansha et al. 2015); however, this has not been explored in a field setting under different types of flooding events spanning a range of seasons and energy-regimes (e.g. ponding versus erosive runoff conditions). Higher energy-regimes related to variable flow conditions can modify biogeochemical processes at the surface-water interface and in the overlying water column (Plach et al. 2014). As such, dynamic cycling of P may occur as a result of changing floodwater chemistry and energydriven soil-water mixing (e.g. rapid release of floodwaters from fields when flow from blocked drainage ditches is resumed leading to the erosion of surface soils.) To address abovementioned knowledge gaps, surface runoff and ditch flow in adjacent roadside ditches were monitored from two agricultural fields during 3 open-water seasons, (i.e. between March 1st and September 30th) over the 2015-2017 period in the Red River Valley of Manitoba, Canada. Outside of these months, conditions are typically at or below freezing and runoff does not occur. Objectives of this fieldbased study were to (1) observe and characterize the occurrence and frequency of periods in which ditches flood over their banks, (2) to compare and contrast flooding events and non-flooding events in terms of runoff volumes, runoff phosphorus concentrations, loads and speciation; and (3) use soil and water quality data from fields and ditches to infer mechanisms that may govern P dynamics during flooded periods.

Materials and methods
Two adjacent farm fields (25 ha each) and their roadside ditches in Elm Creek, Manitoba, were instrumented (Supplemental Figure S1). Soils of both fields are classified as Gleyed Humic Vertisols of the Red River Series (U.S taxonomy: Gleyic humicryerts), and topography is characterized by nearly flat terrain (0-2% slope). Both fields drain into adjacent roadside ditches through in-field surface swales. Field A was tiledrained, whereas Field B was surface drained. Tile drain laterals (10 cm diameter) in Field A were systematically located at $1 m depth, with 13 m spacing. Laterals drained into a large 37.5 cm diameter main that discharged to a collection pond and adjacent retention pond (described in Kokulan et al. 2019b). However, tile drainage has a negligible effect on overbank flooding and surface runoff in this landscape, largely due to either frozen soils or convective events, both of which lead to limited infiltration (Kokulan et al. 2019a(Kokulan et al. , 2021. In both fields, canola (Brassica napus L.) was grown in 2015, which was followed by spring wheat (Triticum aestivum L.) and soybeans (Glycine max L. Merr.) in 2016 and 2017, respectively. The fields were annually tilled to 15 cm depth in fall. Phosphorus is subsurface seed placed in spring as mono ammonium phosphate (40 kg ha À1 in 2015, 45 kg ha À1 in 2016, and 20 kg ha À1 in 2017), whereas the N is surface broadcast as urea (127 kg ha À1 in 2015 and 123 kg ha À1 in 2016) and ammonium sulphate (22 kg ha À1 in 2015 and 2016). Nitrogen fertilizers were seed placed in 2017 (7 kg ha À1 urea and 9 kg ha À1 ammonium phosphate).
Roadside ditches adjacent to both fields received their runoff water predominantly from the agricultural fields, with smaller contributions from the adjacent road and from direct precipitation (snow or rain) (Kokulan et al. 2019a). The two ditches were separated by a farmstead, and flow in the two ditches occurred in opposite directions (the ditch associated with Field A flowed in a north-to-south direction while Field B ditch flowed in a south-to-north direction). Both ditches subsequently drained into provincial drainage channels through culverts. Provincial drains are equipped with passive one-way stop closures. These gates can also be physically blocked (notpassive) to keep water in the fields and lessen the risk of downstream flooding. Drainage ditches adjacent to the fields were parabolic, approximately 2 m wide and are sloped 0.3% toward culverts into which they drain. Ditches and the adjacent fields were separated by a thin (1-2 m width) grassed riparian buffer, which hosted the same vegetation as ditches. Ditch and riparian vegetation were generally comprised of bromegrass (Bromus riparius), quack grass (Elymus repens) and alfalfa (Medicago sativa L.) and was not managed during the study periods.

Instrumentation and data collection
In the current paper, the stage of water in ditches and fields is the main focus rather than flow, as this permits the determination of periods in which surface runoff from fields converged with ditch water, creating 'flooding' conditions (Supplemental Figure S2). Surface runoff was monitored in both fields and adjacent ditches from 2015 to 2016, covering annual spring snowmelt and several spring and summer storms. In 2017, surface runoff was only monitored in Field A and the ditch adjacent to it. The flow from each field was monitored at two surface swales. A Vnotch weir was established in each surface swale to monitor overland flow (Kokulan et al. 2019b). Each weir was equipped with an SR-50A ultrasonic sensor (Campbell Scientific Ltd) and a capacitance sensor (Odyssey, Dataflow Systems Ltd.) to measure water levels. Capacitance sensors (Odyssey, Dataflow Systems Ltd.) were installed in the ditches, both at the upstream and downstream ends, to measure ditch water levels. Water levels in weirs and ditches were recorded at 15 minute intervals, and frequent manual water level measurements were taken to validate sensor readings during runoff periods. A meteorological station (CR10x, Campbell Scientific Ltd) was installed at the site to take hourly measures of rainfall (TE525M, Texas Electronics) and air temperature (HMC45C, Campbell Scientific Ltd.).
In 2016, a Flo-tote 3 and FL 900 series logger (Hach Ltd.) were installed in the downstream culvert of Field A to create a rating curve for estimating flow in ditches. Two rating curves were used for spring (r 2 ¼ 0.96) and summer (r 2 ¼ 0.78) to estimate surface runoff during flooded events. Kindsvater-Shen equation was used to estimate surface runoff from Vnotch weirs when flooding was not observed. This method may have slightly overestimated runoff volumes by accounting already available ditch water as overland flow. Error estimates, considering the water available prior to flow resumption, were 2.6, 3.5 and Twenty-one runoff events were monitored during the 3-year study period (Table 1). These included 3 early spring snowmelt events, 6 late spring rainstorms, and 12 summer thunderstorms. Spring storms were differentiated from thunderstorms that occurred later in the season because soils were near-saturated and typically partially frozen during spring storms, impeding soil water drainage (Kokulan et al. 2019b), whereas soils were fully thawed throughout the remainder of the season. One spring storm (event #2) generated surface runoff 6 days after fertilizer was applied to both fields. Surface runoff samples from the swales were collected at each weir with programmable water samplers (AS950, Hach Ltd.) and acid-washed 1 L polyethylene bottles. Samples were collected at 2 to 4-hour sampling intervals during the rising limbs of the overland flow hydrographs and 6 to 12-hour sampling intervals during the peak and falling limbs of the hydrographs. Additional overland flow grab samples were collected at a daily time step during small flow periods (<1 L s À1 ). Ditch water samples were collected manually in 500 mL polyethylene bottles both upstream and downstream of the roadside ditches at 24-48-hour intervals during runoff events. Runoff samples for events #14 and #17 were not collected in both fields due to autosampler failures.
Soil P concentrations and solid-phase inorganic P partitioning were examined within each field to determine what form the P was held in to infer mobility. Soil samples were collected in Fall 2017 following the harvest (before Fall till) along 225 m-long transects (Supplemental Figure S1). In Field A, surface soil samples (0-15 cm) were collected along 3 transects with a gouge auger (3 samples per transect, one each from the downfield, midfield and up field). In Field B, surface soil samples (0-15 cm) were collected along 2 transects a with a 1 00 soil probe. Each sample was a composite of 5 subsamples taken within a 1 m radius. Collected soil samples were separated by depth (0-6 cm and 6-15 cm) in the field immediately following sample collection. Samples were also collected from adjacent ditches. Collected samples were air-dried, ground to remove any stones or other debris and passed through a 2-mm sieve prior to chemical analysis.

Laboratory analyses
Upon return to the laboratory, water samples were filtered through a Whatman glass microfiber filter paper (0.45 mm pore size, GE Life Sciences) and refrigerated at 4 C. Samples were analyzed for soluble reactive phosphorus (SRP) (QuikChem Method 10-115-01-1-A) with a QuikChem 8500 series 2 system (Lachat instruments) through flow injection analysis colorimetry within 48 hours of collection. Once filtered, filter papers with sediments were dried at 105 C for 24 hours and weighed for total suspended solids (TSS). A non-filtered sub-sample (200 mL) was stored at À18 C and subsequently analyzed for total phosphorus (TP) with a QuikChem 8500 series 2 system (QuikChem Method 10-115-01-4-C, Lachat Applications Group) within one to three weeks from the collection date. Standard checks (0.1 mg P/L) were used per 20 samples for quality control. Five percent of samples were analyzed in duplicate for both SRP and TP. Samples that exceeded the standard spectrum (0 À 0.2 mg P/L for SRP; 0-1 mg P/L for TP) were diluted and re-analyzed.
The detection limits for SRP and TP were 0.001 and 0.005 mg P/L, respectively.
The P content of collected soil samples was determined by the Olsen-P colorimetry method (Sims 2009a). One gram of soil was shaken with 20 ml pH adjusted (8.5) 0.5 M NaHCO 3 extracting solution in a mechanical shaker for 30 minutes at room temperature. The solution was then filtered through a Whatman no-42 filter paper and analyzed for P with a spectrophotometer. For the subsequent analysis, samples that were taken from a particular landscape position (i.e. ditch, downfield, midfield and up field) were bulked to form a composite sample considering the time and resource limitation. The phosphorus sorption index (PSI) was determined from a modified single point sorption isotherm method, as described by Sims (2009b). For PSI analysis, 1 g of soil sample was shaken with 20 ml P sorption solution (75 mg P L À1 ) in a mechanical shaker for 18 hours. The solution was then filtered with 0.45 mm pore size syringe filters and analyzed for SRP with by colorimetry (ammonium molybdate ascorbic-acid, Bran Luebbe AA3, Seal Analytical Ltd.; Method No. G-175-96 Rev.13 for SRP). Loosely bound and moderately soluble (Sol-SRP), reducible bound (Red-SRP), and acid-soluble (Ca-SRP) SRP fractions of soil samples were estimated according to the procedure described by Zhang and Kovar (2009) for calcareous soils. For Sol-SRP, 0.5 g of soil was shaken with 25 ml of 0.1 M NaOH þ 1 M NaCl solution for 17 hours in a mechanical shaker. For Red-SRP, soil residue was then heated in a water bath with 20 ml 0.3 M Na 3 C 6 H 5 O 7 .2H 2 O, 2.5 ml of 1 M NaHCO 3 and 0.5 g Na 2 S 2 O 7 in a water bath at 85 C. Soil residue was then shaken with a 0.5 M HCl solution for 1 hour to extract the Ca-SRP. At the end of every extraction step, soil was centrifuged at 10 000 rpm for 5 minutes (11,180 x g) and washed with saturated NaCl. The wash was added to each extract prior to analysis. Reducible and Ca-SRP P extracts were also neutralized with either 2 M NaOH or 2 M HCl with p-nitrophenol indicator prior to P determination by an autoanalyzer. The sum of these 3 fractions represents the soil total extractable-SRP concentration. Extract from Red-SRP was also analyzed for soil reducible-Fe (FeOx) with colorimetry (510 nm) with a UV spectrometer (FerroVerV R Hach Ltd.). Soil organic matter (SOM) and calcium carbonate equivalent (CCE) of soils were determined using the loss-on-ignition method (Dean 1974). For soil analyses, 20% of samples were analyzed in duplicate.

Data analyses and interpretation
A surface runoff event was determined to have commenced when water levels > 1 cm were detected in V-notch weirs (Kokulan et al. 2019b) and was deemed to have ended when all flow stopped and no water flowed from fields. Event hydrographs for both overland flow and ditches were split into three phases to assess the evolution of P concentrations based on flow and stage characteristics (Supplemental Figure S2). The hydrograph phase classified as 'rising' refers to the rising limb of the hydrograph, where surface runoff exits fields and enters the ditch. During the second phase, referred to as 'merged', the ditch stage was at its peak, but the flow was impeded in both fields and ditches, and ditch water and overland flow were connected. Such merged conditions were only apparent during events classified as 'flooding' events. If such merged conditions did not occur at any point in the event, the event was classified as 'non-flood'. The third and final phase of the hydrograph was the 'falling' phase, defined as when flow resumed in the swales following either the rising or the 'merged' phases, coinciding with the falling limb of a hydrograph. As flow-weighted mean concentrations could not be precisely determined for periods of 'merged' with impeded flow, daily P concentrations in water were used to compare P concentrations between flooded and non-flooded events. Daily P concentrations were estimated from the average of individual water samples collected in a day. This was done to reduce the sampling bias that could arise from uneven water samples between sampling days. Data could not be transformed to meet the assumption of normality, and consequently, Mann-Whitney U test was used to compare daily P concentrations between flooded (Field A: 50, Field B: 34) and non-flooded events (Field A: 29, Field B: 21) and between overland flow and ditch water (Field A: 36, Field B: 28). Differences in Olsen-P concentrations in soils from fields and adjacent ditches were also examined using Mann-Whitney U test. Statistical tests were not performed for other soil properties due to lack of replication (1 composite from the ditch and 3 composites from the field). Differences deemed significant at p < 0.05.
Partitioning coefficients (K d ), the ratio of solid and aqueous phases of an element, were calculated as an indicator of potential shifts in P soil-water interactions related to water-column chemistry and energyregime (Gormley-Gallagher, Douglas, and Rippey 2015; Plach et al. 2014). Higher K d values indicated a more sorptive environment within the floodwater and vice versa. Partitioning coefficients for P (L kg À1 ) were calculated for Field A flooding events using the equation [1]. We were unable to estimate K d s for Field B as we could not quantify the flow and nutrient loads. (1)

Results and discussion
Occurrence of overbank flooding of ditches As roadside ditches are ephemeral and ditch flow is largely generated by surface runoff (Kokulan et al. 2019a), all events that produced a hydrologic response resulted in both surface runoff and ditch flow (Figure 1 (Field A), Supplemental Figure S3 (Field B)). Of these event responses, only 5 resulted in 'flooding' conditions in Field A during which water on fields and in ditches merged (Table 1, Supplemental Figure S4). Four events with flooding conditions were observed in Field B in 2015 and 2016 (Supplemental Figure S3). Such conditions occurred following multi-day events during which ditches filled and merged with surface runoff in swales: these were observed during spring snowmelt as well as during late spring rainstorms and one fall thunderstorm event (Figure 1). During these periods of flooding ('merged' condition), substantial portions of the fields were submerged, lengthening the contact time between surface soil and runoff (Supplemental Figure S4). In general, periods of merged conditions during the spring followed rapid snowmelt associated with rain-on-snow events, creating a rapid addition of large volumes of water (Table 1). This occurred in 2 out of the 3 study years. Flooding events that occurred in late spring followed multi-day spring storms on seasonally wet (postsnowmelt) soils. Only one flooding event occurred that was not associated with either snowmelt or rain on seasonally wet soils: it was also observed in September 2016, following a high-intensity, short-duration thunderstorm that contributed a large volume of rainfall (85 mm rain in 2 hours). Although there were many other events that occurred over the study period with varying rainfall volumes and antecedent moisture conditions, most of these events failed to produce flooding conditions. It is important to note that the passive one-way valves in the provincial drains located downstream of the ditches may have contributed to flooding conditions by preventing drainage. However, such closures only occur under very wet conditions, as they are passively closed by elevated flow in streams, reopening once water levels drop in downstream channels. Although flood conditions occurred in only a small proportion of events, such events can be expected in most seasons, particularly associated with snowmelt. However, the frequency of such flooding events was lower after spring. Soil, weather and anthropogenic factors influence the progression of spring runoff and its tendency of turning into multi-day inundation events in the Red River Valley. For example, a wetter fall season followed by a severe winter substantially decreases the infiltrability of meltwater through wet, frozen soils, thus increasing the chance for spring snowmelt flooding (Wazney and Clark 2016). This flooding risk is further exacerbated by rain-on-snow events, which not only add additional water for runoff but also rapidly melt/ripen the snowpacks (Il Jeong and Sushama 2018). In general, meltwater exited the monitored fields once the adjacent ditches began to thaw. However, during rapid melts, such as those accompanied by rain on snow events, ditch water often flows back into fields, potentially due to frozen culverts and ice jams in the nearby provincial channel where the ditches drain. During the study period, such conditions eventually led to prolonged flooding periods during snowmelt (1-7 days flooding out of a 7-14-day event).
There is also significant potential for the regular occurrence of flooding following late spring storms. Indeed, this study observed two major late-spring flooding events in 2015 (event #2) and 2016 (event #8). Both events had substantial rainfall volumes (64 mm in event #2; 38 mm in event #8); however, events of similar magnitude that occurred later in each season did not lead to flooding conditions (Table 1). It is likely that the flooding conditions were created by a combination of wet antecedent conditions and decreased storage potential but also due to the presence of partially frozen ground (Kokulan et al. 2021). Previous studies discussing the occurrence of flooding events in the Red River Valley have related them only to early spring snowmelt runoff (e.g. Cordeiro et al. 2017;Schindler, Hecky, and McCullough 2012), and less is known about flooding during late spring or summer events, in this region where flat topography makes the relevance of these event to water and nutrient budgets particularly important. Our study shows that sizeable late spring multi-day rain events and summer thunderstorms also have the potential to flood agricultural fields in this region, which may have implications in the future as more frequent multi-day spring and summer storms are forecasted for the Prairies (Shook and Pomeroy 2012).
Major summer thunderstorms occurred in all 3 study years (Table 1); however, only one (event #17) produced flooding conditions. Although thunderstorms, which are associated with high-intensity rainfall, frequently trigger surface runoff in this landscape (Kokulan et al. 2019b), they seldom lead to field scale flooding, likely due to higher evapotranspiration and crop water demand, which increase soil water storage potential.

Contribution of flooding events to edge of field runoff and P loss
Although flooding events represented only a small fraction of the observed events (5 out of 21), such events contributed to the majority of annual runoff and P losses (Table 1). On Field A, these 5 events, during which prolonged periods of flooding were observed, accounted for 70% of the edge of field runoff losses (overland flow þ tile flow), 83% of the SRP losses, and 84% of TP losses over the entire study period. Overland flow was particularly important during these events, accounting for 87% of the total overland flow hydrologic losses, 86% of SRP losses, and 88% of the TP losses over the 3-year period. Although flooding was observed within these larger events, the duration was relatively short. For example, events typically occurred over 10-14 days, and flooding conditions occurred throughout approximately 10-50% (1-7 days) of each event. Given that flooding generally occurred when ditches were blocked by passive drains to prevent flooding downstream, flow was generally minimal during periods of 'flooding' (i.e. stagnant water, standing on fields). The majority of flow (75-99%) exiting fields via ditches during these large events occurred after the merging of the swales and ditches had occurred (falling limb of hydrograph), presumably once the passive drains had reopened.
The dominance of a small number of large magnitude events in annual edge-of-field losses has been described in other North American agricultural regions such as southern Ontario, Canada (Macrae et al. 2007;Van Esbroeck et al. 2017) and the Midwestern US (Carpenter, Booth, and Kucharik 2018). However, this is particularly relevant in the Red River Valley region where the snowmelt period (spring freshet) has a predominant role in the annual hydrologic cycle (Dumanski, Pomeroy, and Westbrook 2015;Kokulan et al. 2019a) due to snowmelt on frozen ground and dry summers due to the dry Prairie climate.
Although flooding events contributed the majority of annual edge-of-field P losses, this was driven by runoff volumes, as there was no significant difference (p > 0.05) between daily SRP concentration for flooded (Field A 0.20 mg P/L; Field B 0.08 mg P/L) and non-flooded (Field A 0.24 mg P/L; Field B 0.06 mg P/L) conditions (Table 1). The lack of difference in SRP concentration between events with and without flooding was consistent both on the rising and falling limb of the event hydrographs ( Figure 2).
Although there was not a difference in daily SRP concentrations, TP concentrations were elevated in non-flooded events relative to flooded events for the field with high P (Field A 0.26 mg P/L for flooded events and 0.39 mg P/L for non-flooded events, p < 0.05; Field B 0.09 mg P/L for flooded events and 0.14 mg P/L for non-flooded events, p ¼ 0.13). This difference in TP concentration was revealed through SRP:TP and partitioning coefficients (K d ) that demonstrated differences in energy dynamics between the flooded and non-flooded conditions in Field A ( Figure 3). Overall, ratios of SRP:TP were greater (mean SRP:TP ¼ 0.74 for Field A and 0.61 for Field B) and K d were lower (Figure 3) for flooded events relative to non-flooded events (mean SRP:TP ¼ 0.64 for Field A (p < 0.05) and 0.47 for Field B (p ¼ 0.4). Non-flood events had higher K d in Field A (p < 0.05) but not in Field B (p ¼ 0.4). The lack of significant results from Field B could be due to smaller SRP and TP concentrations relative to Field A. Reduced particulate material during flooded events likely indicates potential impeded flow and the settling of particulates under quiescent conditions. In contrast, greater K d values (i.e. greater P partitioning to particulates) during non-flooded events likely reflects turbulence due to rapidly draining water and the re-suspension of eroded surface soils.
Larger SRP:TP and smaller K d values during flood events may also indicate a potential mobilization of SRP from soils during the merged phases of flooding events (i.e. desorption, and diffusion into the overlying water column). Although water was largely stagnant in ditches and on fields during the 'merged' phase increases in P concentration were observed in standing water over time for snowmelt flooded events (Figure 4). An increase in P concentration was often observed 50 hours after the onset of flooding. However, this change in P concentration was not . SRP and TP concentrations following flooded periods during the study (i.e. hours since the flood water from both field and ditch initially merged). Stage, SRP, and TP concentrations have been normalized by field and event for comparison due to differences in initial conditions and event magnitude, where stage is a proportion of event-maximum stage, and P concentrations are a departure from field and event means, normalized by field and event standard deviations (z-scores).
observed for flooded periods that occurred following spring storms. This may partially reflect runoff from enriched SRP ditchwater but may also indicate SRP mobilization from field soils into floodwaters. For example, during flooding, soil SRP could have been mobilized through desorption, pH-mediated dissolution or redox-mediated dissolution reactions (Amarawansha et al. 2015;Chow and Eanes 2001;Liu et al. 2013). These results are in agreement with other Manitoban studies that reported a gradual increase in floodwater SRP concentration in the early stages of flooding (0-21 days), potentially due to diffusion of P from porewater to floodwater (Vitharana et al. 2021).

Linkages between soil P and water chemistry in fields and ditches
Olsen P concentrations and soluble SRP concentrations (i.e. Sol-SRP; P loosely adsorbed to clays and metal oxides surfaces) were numerically elevated in field soils relative to ditches (p ¼ 0.08 for Field A and p ¼ 0.07 for Field B), suggesting the potential for SRP to be mobilized from field soils during inundation due to increased soil-water contact during flooding events (Table 2). Indeed, soil available P concentrations correlate well with concentrations of P in runoff in the Prairies . Surprisingly, runoff P concentrations were smaller in Field B than Field A, despite the fact that Sol-SRP and Olsen-P concentrations were similar in both fields. Furthermore, Sol-SRP concentrations were smaller in ditches than in fields. Despite differences in soil/sediment composition and chemistry between fields and ditches ( Table 2), concentrations of P in ditches and surface runoff from the field did not significantly differ from one another (Figure 1, p > 0.05).
Loosely-bound SRP (i.e. Sol-SRP) comprised the smallest fraction ($ 10%) of the total extractable SRP stored in surface soils of both fields. However, a substantial amount of soil SRP was held in the reducible phase (i.e. Red-SRP; $37%), particularly in Field B (Table 2). Prolonged soil inundation can mobilize P due to reductive dissolution of P binding Mn 4þ and Fe 3þ oxides. In fact, the majority of Manitoban soils have been found to be prone to redox associated P losses (Amarawansha et al. 2015;Dharmakeerthi et al. 2019). Increases in SRP concentrations in surface water with time during the 'flood' phase of the hydrograph may reflect P release due to redox processes. Unfortunately, dissolved oxygen, redox status or Fe 2þ /Fe 3þ concentrations of the floodwater were not monitored in this study. As such, future research of spatial and temporal drivers (e.g. duration of inundation, soil redox conditions, Fe speciation, soil FeOx reactivity) behind potential redox-mediated soil P release to runoff is needed.
Although there was considerable P stored in the loosely-bound and reducible forms, the acid-soluble SRP pool (i.e. Ca-SRP) represented the dominant fraction for soil SRP retention (45-65%) ( Table 2) in both fields and ditches. However, mean concentrations of Ca-SRP in ditch sediments were up to 35 to 40% higher than in field soils. Elevated Ca-SRP concentrations coincided with greater carbonate content and lower organic matter in ditch sediments relative to field soils (Table 2). They also coincided with generally greater P sorption indices (PSI) within ditch sediments (median: 390, range: 372-408 l kg À1 ) relative to the two fields (median:306, range: 274-381 l 2) a Samples from the same landscape position were composited for all analysis except for the Olsen-P. As such, there were no replicates for ditch samples. kg À1 ) ( Table 2). Thus, rather than acting as a P source to runoff, ditch sediments may have served as an SRP sink (greater Ca-SRP and PSI relative to field soils). Under alkaline conditions, the presence of Ca 2þ in floodwater could reduce floodwater SRP concentration by precipitating as b-tricalcium phosphate and octacalcium phosphate (Attanayake et al., 2022). In addition, adjacent roads are graded with limestone, which may further assist in the immobilization of SRP in runoff.
Although not measured in the present study, the potential of plant material to release SRP when frozen (Cober, Macrae, and Van Eerd 2018;Kieta et al. 2018) or decaying (Cober, Macrae, and Van Eerd 2018), and as a P source for snowmelt runoff, is well documented in cold environments (Liu et al. 2019). Even though crop residue was hayed and removed from the fields, ditch vegetation was not managed. Future research should consider assessing the role of ditch vegetation on P mobilization (early spring snowmelt) and immobilization (intake by ditch vegetation during late spring).

Conclusions
This study documented the occurrence of flooding of roadside ditches into adjacent fields in agricultural lands in the Red River Valley of Manitoba. Although events with flooding periods comprised a small proportion of all total runoff events, such flooding events contributed to the majority of annual runoff and edge of field SRP and TP losses. Further, while concentrations of P did not differ between flood and non-flood events, flooding events appeared to have a greater proportion of dissolved P relative to events without flooding. The results of this field-based study indicate that P concentrations and speciation in surface runoff may be driven by highly dynamic soil-water P cycling in ditches and fields as well as runoff flow regime, which impact floodwater chemistry. An improved understanding of the relative influences of management practices such as ditch closures, road gradation, and vegetation management on these dynamic processes is needed.