*2.5. Statistical Analysis*

Paired watershed analysis involves edge-of-field runoff water quality monitoring of two or more fields similar in soil characteristics with hydrologic isolation to enable fields to function as individual small watersheds. Watersheds are treated identically during the calibration/baseline period to account for runoff variation among watersheds, and event-based runoff data is collected. During the treatment phase, the control watershed continues with same treatment while treatment designated watersheds receive different management treatments. Treatment and calibration regression equations are then tested statistically for differences to determine treatment effects. The direction and magnitude of the effect are determined by comparing values predicted by calibration regression to those observed during the treatment period.

In the present study, the treatment period runoff water quality data are presented in relation to previously reported calibration period results [39]. The number of paired runoff events for statistical analysis ranged from *n* = 34 to 39 during the calibration period and *n* = 55 to 61 events during the treatment period. All dependent runoff water quality variables measured (and model residuals) were non-normally distributed. While log transformation is commonly done for non-normality and heteroscedasticity (e.g., [10,19]), log transformations rarely normalized variables and did not adequately resolve variance heterogeneity for this study. A generalized linear mixed modeling approach (log-link function) was also conducted [47] with several distribution types, but heteroscedasticity remained an issue. Given these limitations, permutation tests [48,49] were conducted (rather than analysis of covariance/regression) to test two hypotheses for each dependent variable: (1) calibration period mean equals treatment period mean and (2) calibration period slope equals treatment period slope. A total of 1000 permutations were generated by randomly reassigning calibration and treatment data to the two time periods followed by one-way analysis of variance (Hypothesis 1) or randomly sorting x-axis values followed by estimation of permuted slopes (Hypothesis 2). Hypothesis tests for means and slopes were conducted as two-tailed tests with *p*-values set at 0.25, 0.10, 0.05, and 0.01 to define significance levels. Predicted constituent concentrations/loads were based on

calibration regression equations of Jokela and Casler [39]. The relative impact of BMPs on water quality was expressed as the percent difference between predicted (i.e., from calibration phase equations) and measured values for the treatment. Corn silage yield and soil nutrient data were analyzed by mixed models analysis of variance [50].

#### **3. Results and Discussion**

#### *3.1. Weather and Corn Silage Yield*

Precipitation during the treatment period (2008–2011) was slightly greater (10%) compared to the calibration period (2006–2008) and more pronounced in the growing season (36% greater than the calibration period; Table 2). Treatment phase growing season rainfall was similar to the 30-year average, whereas the calibration period was 32% drier. During the calibration period, measured runoff averaged 14% of total precipitation. For the treatment phase, the mean runoff fraction was considerably higher (26% of precipitation), most likely due to the overall wetter conditions increasing runoff potential in these somewhat poorly drained soils. Average temperatures for calibration and treatment periods were similar to the 30-year mean.

Corn yield averaged across fields was 16.2 Mg ha−<sup>1</sup> for the treatment period and in the range of expected yields for central Wisconsin. There were some yield differences among fields with notably lower yield for FMST, which could have been related to greater NH3 volatilization due to the lack of fall incorporation. However, the lower third of field M3 becomes more imperfectly drained as the main field slope levels out and runoff water collects. It is therefore likely that greater N loss from not incorporating manure and the more imperfectly drained section of M3 contributed to the lower average yield. Averaged across the treatment period, yields ranged from 15 to 17 Mg ha−1; N and P uptake ranged from 177 to 192 and 21.9 to 25.4 kg ha−1, respectively. Yield, N, and P removal differences among fields were also noted for the calibration phase [39]. For the calibration period average yield (17.3 Mg ha−1), N and P removal (197 and 34.7 kg ha−1, respectively) were highest for field M2 (RSMT for the present study), while average yields for the other fields were similar (16 to 16.6 Mg ha<sup>−</sup>1).


### *3.2. Runoff Water Quantity*

Cumulative runoff over the calibration and treatment period ranged from 660 to 980 mm among the four fields (Figure 1). In the two treatments with spring chisel tillage (RSMT and FMST), runoff increased by 131 and 34%, respectively, compared to the falltilled control (FMT), whereas runoff for BFMT was similar to the control (FMT), both of which were fall chisel plowed (Table 3). These results are consistent with those of Stock et al. [14], who reported that chisel tillage significantly reduced runoff and nutrient loading in surface runoff from silt loam soils in southern WI. They attributed the lower runoff to greater surface roughness and increased water storage/infiltration compared to no-tillage. The lower increase in runoff for FMST may be due to physical protection of soil structure, essentially a mulching effect aiding infiltration [51,52]. While several studies show cover crops can contribute to lower surface runoff from greater evapotranspiration and soil structural effects [53–55], delaying tillage for RSMT resulted in more runoff in our study. As stated, the decreased surface roughness from delayed plowing likely contributed to increased runoff. However, while delaying tillage until spring could help reduce erosion potential during the non-growing season when runoff risk is elevated, any such effect would probably be minimal with the limited crop residue remaining after silage harvest. Furthermore, any potential beneficial effect of the rye cover crop on runoff in our study was minimized because of its limited fall growth (seeded in October). Percentage cover of rye and other live plant material was 68, 29, and 32 in November 2008, 2010, and 2011, respectively, (2009 data missing), and rye biomass was 500, 338, and 399 kg ha−<sup>1</sup> in late Apr or early May 2010, 2011, and 2012, respectively, before manure application, tillage, and corn planting.

**Figure 1.** (**a**) Cumulative runoff (mm), (**b**) total P load (kg ha<sup>−</sup>1), (**c**) suspended sediment load (kg ha<sup>−</sup>1), and (**d**) dissolved P load (mg kg<sup>−</sup>1) over the calibration (2006–2008) and treatment periods (2008–2012) for each watershed; FMT = fall applied manure with chisel tillage, RSMT = fall rye (cover crop) with spring-applied manure and chisel tillage; FMST = fall applied manure with spring tillage; BFMT = fall applied manure/chisel tillage with grass buffer.


calibrationandtreatmentperiods,predictedtreatmentperiodsmeans,andhypothesistestresultsforrunoff



treatment period: \*\* *p* ≤ 0.01; \* *p* ≤ 0.05; + *p* ≤ 0.10; # *p* ≤ 0.25, NS ≥ 0.25.

### *3.3. Runoff Water Quality*

Similar to other studies [22,36,56,57], we found significant relationships between runoff volume and SS, TP, and TN loads (calibration and treatment periods combined). A stronger relationship between TN load and runoff (R2 = 0.75) may reflect greater mobility compared to SS and TP (R2 = 0.26; 0.50, respectively); a large portion (20 to >50%) of TN load was NO3-N further supporting this hypothesis (Table 4). Event runoff TP and TN loads were closely associated with SS loads (R2 = 0.83; 0.74, respectively), indicating sediment was an important transporter of P and N in runoff. Jokela and Casler [39] also reported a strong relationship between runoff SS and TP loads (R2 = 0.82), with the exception of two high flow events in early spring 2007 with very high SS.

The influence of management systems on P transport and overall runoff water quality during the treatment period differed. Establishment of a vegetative buffer along with fall manure and tillage (BFMT) significantly decreased SS, TP, TN, and NO3-N concentrations (Table 3). Loads of SS, TP, TN, NO3-N and NH4 +-N were reduced by 55, 28, 34, 41, and 22%, respectively (Table 4), however, mean DRP concentration and load increased by 88 and 15%, respectively, compared to predicted means (Tables 3 and 4). In addition to using means to compare treatment effects, regression slopes (treatments vs. control watersheds) for calibration and treatment periods were also compared for constituent loads and concentrations (Tables 3 and 4). As an example, effects of BFMT on runoff and TP load were based on slope comparisons, as illustrated in Figure 2. Plots of the deviation between predicted and observed values for SS and TP load show minimal effects until well into the treatment period (September 2010; Figure 3), perhaps related to time for development of buffer vegetation. While vegetated buffers are generally effective at mitigating surface runoff sediment and particulate P, DRP removal efficiency is generally lower and in some cases positive [58–60]. Reports of episodic DRP release from biomass after freeze-thaw events, presumably due to cell lysis and runoff mobilization [20,21,61–63], may have also contributed to elevated DRP loads for BFMT.

The combination of a post-harvest rye cover crop and delay of manure application/chisel tillage until spring (RSMT) significantly decreased SS, TP, TN, NO3-N, and NH4 +-N concentrations by 24 to 54% (Table 3), suggesting potential benefit. However, because of the increased runoff volume, discussed above, there were no decreases in runoff load of any of these parameters. The limited rye growth because of late planting, noted earlier, likely contributed to this. The lack of load reduction of DRP, NO3-N, and NH4 +-N is not unexpected because a rye cover crop is considered more effective at reducing erosion and sediment-bound nutrient loss as compared to dissolved nutrients. Unfortunately, some confounding results may have occurred because manure applied in the spring (RSMT) had a 10 to 22% greater nutrient content than that applied in fall treatments, most pronounced in fall 2010/spring 2011 when differences in N, P, and NH4 +-N applied were two-fold or more (Table 1).

The practice of fall-applied manure with spring chisel tillage (FMST) is not considered a BMP, however, it is a fairly common practice for central WI dairy farms and, therefore, the evaluation was deemed important. Not unexpectedly, there were large concentration increases (46, 312, and 184% increases for TP, DRP, and NH4 +-N, respectively) and large loading increases for DRP and NH4 +-N (376 and 197%); TP load also showed a numerical increase of 39%, but was NS (Tables 3 and 4; Figure 3). Previous studies have also demonstrated the importance of manure incorporation or injection to mitigate runoff N and P in corn and hay forage systems [51,52,64,65]. There was a significant decrease in SS, TN, and NO3-N concentration, presumably due to the protective soil mulching effect described above regarding runoff effects, but there were no load decreases, except for NO3-N, because of the increase in runoff.


#### *Soil Syst.* **2021** , *5*, 1

164

**Figure 2.** Relationships between each watershed (BFMT = fall applied manure/chisel tillage with grass buffer (**top**); RSMT = fall rye (cover crop) with spring-applied manure and chisel tillage (**center**); or FMST = fall applied manure with spring tillage (**bottom**) and FMT (fall applied manure with chisel tillage) during the calibration and treatment period for runoff quantity (mm) (**left**) and total P (TP) load (kg ha−1) (**right**). The solid and dashed lines are linear regressions for the calibration and treatment periods, respectively.

**Figure 3.** Observed minus predicted values for runoff events throughout the treatment period for each system BFMT = fall applied manure/chisel tillage with grass buffer (**top**); RSMT = fall rye (cover crop) with spring applied manure and chisel tillage (**center**); or FMST = fall applied manure with spring tillage (**bottom**) for suspended solid (SS) load (mg kg<sup>−</sup>1) (**left**); total P (TP) load (mg kg<sup>−</sup>1) (**middle**); and dissolved reactive P (DRP) load (mg kg<sup>−</sup>1) (**right**).

The distribution of runoff events in a given hydrologic year is often skewed, with a few large events contributing a majority of the annual runoff, sediment, and/or nutrient loss [10,66,67]. We observed a similar phenomenon, with much of the cumulative SS and TP load over the course of the trial coming from a small number of events in early fall after silage harvest when soil is exposed (Figure 1). As a result, the effectiveness of conservation practices may be greatly reduced if they are overwhelmed by one or a few large events.

Much of the annual runoff in cold climate regions is from snowmelt [11,13,34,35,39,68,69]. An average of 53% of annual runoff (averaged across watersheds and control/treatment periods) was derived from snowmelt. We also found greater surface runoff yields during the non-growing season when soils are wetter due to lower evapotranspiration rates. In our study, snowmelt runoff had a much greater proportion of nutrients in dissolved form, for example, 70% of the TP load for FMST was DRP for snowmelt events as compared to 17% for rain events. This is consistent with other studies reporting greater DRP/TP ratios for snowmelt compared to rainfall [13,14,17,39,70,71], potentially exacerbating water quality risk due to coincident elevated DRP mobility and high runoff volumes. Runoff DRP and NH4 +-N loads were generally larger for snowmelt (Table S2) and early spring rainfall events, a loss pattern noted elsewhere [13,19,69,72,73].

Separate analyses of snowmelt and rainfall events showed some differences in relative treatment effects, as indicated by the change between observed and predicted treatment period means (Tables S2 and S3). For rain events, BFMT showed significant decreases in mean runoff concentrations of SS, TP, and all N species (40 to 68%) and in TP, DRP, and NH4 +-N loads (34 to 84%; Table S3). In contrast, snowmelt events had no significant decreases (some concentrations increased) with significant load decreases only for SS and NO3-N (Table S2). While BFMT was much less effective for snowmelt than rain, the 80% reduction in SS load indicates the potential of a vegetative buffer to mitigate particulate P and SS for snowmelt runoff events [11,13,14,17,19,26,35,67,69]. However, while a vegetated buffer may contribute to a reduction in SS and particulate P transport year-round, buffers are less effective at DRP removal, particularly under conditions of frozen or partially frozen soils [29]. In addition, studies show that snowmelt runoff tends to have a greater proportion of DRP compared to rainfall-induced surface runoff [26,67]. In our study, approximately 43% of TP was in DRP form averaged across watersheds versus 3% for rain (Tables S2 and S3).

In contrast to BFMT, RSMT had significantly lower mean concentrations of SS and most P and N species than predicted in both rain and snowmelt runoff, with somewhat greater reductions in snowmelt. However, because of increased runoff, these were translated into significant decreases in loads only for SS, TN, and NO3-N in snowmelt. Overall, there were only modest treatment effects on P loads and the other measures for either rainfall or snowmelt events, (although some slopes indicated modest changes), indicating the greatest impact of a rye cover crop occurred outside the growing season, even with minimal rye growth. Spring manure application and tillage minimally impacted nutrient loss likely due to timing, occurring immediately prior to planting, typically during a drier window of time. Only DRP had a slightly (7%) but significantly greater mean load in the treatment period, most of the higher DRP concentrations found during events occurring post snowmelt but prior to planting (Figure 4).

**Figure 4.** Concentrations (mg L<sup>−</sup>1) of dissolved reactive P (DRP) over the study period.

Broadcast application of manure in the fall with tillage delayed until spring (FMST) leaves manure on the soil surface through the non-growing season, resulting in marked water quality contrasts between rain and snowmelt runoff. In rain events, the small (11%) reduction in SS concentration led to 17% and 2% decreases in TN concentration and load, respectively, but no other decreases. There were increases in DRP and NH4 +-N concentration of about 70% and loads of 190 and 16%, respectively. In snowmelt runoff, there were significant decreases in SS, TN, and NO3-N concentrations and SS and NO3-N load. Perhaps the most striking effects were the very large increases in TP, DRP, and NH4 +-N (579, 647, and 232% concentrations, respectively, and 605, 784, and 411% loads, respectively), emphasizing the runoff water quality risk posed by unincorporated manure left on the soil surface over winter.

#### *3.4. Research Implications*

Both BFMT and RSMT significantly reduced runoff SS, TP, TN, and NO3-N concentrations while BFMT also reduced SS, TP, TN, and NO3-N loads; however, neither significantly reduced DRP concentration or load. Similar tradeoffs with respect to practices designed to control erosion have been identified in the Lake Erie and other watersheds, where practices

such as reduced tillage and tile drains can mitigate erosion and TP loss in surface runoff but have the unintended consequence of contributing to elevated runoff DRP risk [74,75]. Encouragingly, RSMT significantly reduced concentrations of SS and all P and N species and loads of SS, TN, and NO3-N for snowmelt events (Table S2), suggesting a rye cover crop in the fall (with spring tillage) may indeed have potential in corn silage systems to mitigate sediment and P loss in surface runoff. Our results indicate that FMST is not an advisable practice, given its much higher risk of surface runoff DRP loss, as evidenced by the large total and dissolved P load increases (Table 4; Figure 3), especially in snowmelt runoff (on average 605% and 784%, respectively).

An important result from our study was the relatively large runoff DRP concentrations, especially during the treatment period. Average event runoff DRP concentrations (for all events and/or snowmelt events) were at or above the reported eutrophication threshold of 0.05 mg L−<sup>1</sup> (Table 3, Figure 4), indicating runoff would present a freshwater eutrophication risk if discharged directly to streams [76,77]. Overall, DRP increased from about 2.5% of runoff TP in the calibration period to 12.2% during the treatment phase. Rye cover crop biomass and/or grass buffer strips may have contributed to increased runoff DRP from those treatments as a result of DRP release from biomass after freeze-thaw events, as previously noted [21,29,61–63]; grass buffer and rye biomass likely altered SS and particulate P transport. However, we hypothesize P from manure and labile soil P forms were more important P sources to runoff water. Manure can be a direct source of nutrients to runoff, especially when left on the surface (i.e., FMST) where labile N and P forms are vulnerable to mobilization in surface runoff. The soluble NH4 +-N form was about half of total N in manure (Table 1) and may partially explain the large increase, especially for FMST, whereas more variable effects were noted for TN (which includes low soluble organic forms associated with sediment). It is also well established that dairy manure applications increase labile soil P concentrations as measured by agronomic soil tests [78,79]. Soil samples were taken in 2009 and 2012; mean B1P in fall 2009 was 22 ± 2 mg kg−<sup>1</sup> and increased to 29.5 ± 1.9 mg kg−<sup>1</sup> (37% increase) by spring 2012. This P increase represents a shift from "optimum" to "high" with respect to the University of Wisconsin soil test P fertility categories [41]. Several studies show that P release potential to surface runoff increases with greater labile soil P concentrations [80–82]. In WI, B1P is used as an agronomic indicator of plant-available P and is the basis for determining needed crop P inputs; it is also an important component in the WI P index, which is used to determine P loss risk and develop manure management plans [17,26,80,83]. For a range of soil series in WI, Laboski and Lamb [78] showed that both water-soluble P concentrations and the degree of P saturation increased linearly with B1P, while P sorption strength decreased. It is likely that annual manure applications contributed to the measured B1P increases and runoff DRP concentrations observed in our study. We consider manure P and soil-bound P to be the two main sources of runoff P in our study, further highlighting the importance of conservation practices to mitigate both particulate and dissolved P forms.

The lack of effective control of DRP in runoff was not totally unexpected because the practices evaluated are known to be more effective for erosion and particulate P reduction than dissolved P, suggesting a need for alternative approaches to control dissolved P. One option is the use of injection or other low-disturbance manure application methods, which have been shown to reduce DRP losses substantially compared to surface manure or incorporation by tillage [23,51]. Another is to limit supplementary P addition to livestock feed, which has resulted in significant reductions in runoff DRP from the application of dairy manure [18,84,85]. While our study highlights the importance of controlling erosion and SS transport in order to mitigate P loss, it also demonstrates the inadequacy of the evaluated practices to effectively reduce DRP accumulation in surface soils and the associated greater risk of P transport in surface runoff. Depending on the proximity to open water or other surface water conveyances, our results indicate that additional practices may be needed to mitigate both particulate and dissolved P transport simultaneously and to a degree necessary to protect water quality.

#### **4. Conclusions**

Fall manure application and incorporation via chisel plow or other tillage is a common practice on many dairy farms in central WI and other parts of the US. Results from our paired watershed study in central WI indicate that each of the three manure-tillage management systems tested had different potential impacts on surface runoff water quality compared to the control of fall-applied manure with chisel tillage. Both BMPs (BFMT and RSMT) reduced runoff SS, TN, and TP concentrations, but effects on loads were variable, and DRP loads increased slightly. Addition of a vegetative buffer with fall manure application/chisel tillage reduced SS, TP, and total and dissolved N losses, which made it the most effective BMP. A fall rye cover crop with spring manure application and chisel tillage also shows potential for mitigating sediment, N, and P transport, but it would be more effective with earlier seeding to achieve more fall growth and ground cover. Ineffective control of DRP losses suggests the need for additional management practices, such as low-disturbance manure application and limiting livestock dietary P. Fall-applied manure with spring tillage incorporation is not recommended due to high runoff NH4 +-N and DRP losses.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/2571 -8789/5/1/1/s1, Table S1: Summary of Abbreviations used, Table S2: Event-based means and regression slopes for calibration and treatment periods, predicted treatment periods means, and hypothesis test results for runoff volume, suspended sediment, and nutrient concentrations and loads for snowmelt events only, Table S3: Event-based means and regression slopes for calibration and treatment periods, predicted treatment periods means, and hypothesis test results for runoff volume, suspended sediment, and nutrient concentrations and loads for rain events only.

**Author Contributions:** This research project and article were contributed to by the authors in the following way: conceptualization, W.E.J.; methodology, W.E.J.; formal analysis, M.D.C.; investigation, W.E.J.; resources, J.C. and W.K.C.; data curation, J.F.S.; writing—original draft preparation, J.F.S.; writing—review and editing, E.O.Y., W.E.J., and M.D.C.; supervision, W.K.C. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was supported in part by the U.S. Department of Agriculture, Agricultural Research Service.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data is contained within the article or supplementary material. The data presented in this study are available [insert article and supplementary material here].

**Acknowledgments:** The authors would like to thank Craig Simson, Matt Volenec, and Ashley Braun for excellent technical support in the field and lab and Mike Bertram and the UW MARS staff for equipment operation and assistance with field maintenance. The findings and conclusions in this publication are those of the authors and should not be construed to represent any official USDA or U.S. Government determination or policy.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**

