*3.1. Changes in Vertical Distribution of Soil P*

In this study, M1P increased significantly from 2015 to 2018 at 0–5 and 5–10 cm in both treatments (Table 3). These results indicate an accumulation of P in the top 10 cm layer of soil for both treatments, with 6.1 and 4.9 times increase in median M1P for the 0–5 cm layer and 2 and 1.6 times increase in median M1P for the 5–10 cm soil layer, CHD and STR pastures, respectively. Optimum M1P in Georgia Piedmont pasture soil is 30 mg P kg−<sup>1</sup> [34]. In only a few years, both treatments were able to improve soil P concentrations of the 0–5 cm soil layer from low to moderate (half of the optimum M1P). The increase in M1P at 0–5 cm soil layer in both the treatments was due in part to the manure deposited by the grazing animals. P is relatively immobile in soil [9] and deposition of cattle dung accumulates inorganic P in the top 5 cm soil layer for a wide range of soils [35] while surface applied superphosphate accumulates inorganic P in the top 7.5 cm layer of a silt loam soil [5]. The unexpected increase in M1P at 5–10 cm depth could be explained by the decrease in bulk density [36]. Reduced bulk density results in greater porosity and movement (eluviation and illuviation) of clays and nutrients down the soil profile.

**Table 3.** Median Mehlich-1 phosphorus (M1P) in 2015 (Baseline) and 2018 (Post-Treatment) sampling dates in continuous grazing with hay distribution (CHD) and strategic grazing (STR) pastures.


Medians separated by different upper-case letters denote significant difference between sampling dates within treatments (i.e., CHD, 2015, compared to CHD, 2018). Medians separated by different lower-case letters denote significant difference between treatments on the same sampling date (i.e., CHD, 2015, compared to STR, 2015). Difference is at 0.1 level of significance.

We speculate that soil biology and plant roots were responsible for transporting the available P from the top 5 cm to the 5–10 cm interval. However, this does not fully explain the increase in the 0–5 cm soil layer.

In CHD pastures, the increase at 0–5 cm may be partially explained by hay bales added during the drought in 2016. CHD pastures required 102 hay bales while STR pastures required only 34 hay bales due to sustained vegetation in exclusions late into a 2016 drought. Hay bales distributed at the Eatonton pastures had an average dry weight of 389 kg, 0.28% of which is P [37]. Assuming 1.1 kg P in each hay bale, the amount of P added would be 112 kg P in CHD and 37 kg P in STR pastures. As hay bales were distributed throughout the pasture during treatments and not in areas vulnerable to loss and because P increased in the top two layers, these results indicate that hay distribution can help retain P in pastures. Most P eaten by cattle returns to the soil in the form of manure adding the P eaten in hay and forage. In STR pastures, the increase in the 0–5 cm layer was likely due more to the retention of P deposited by grazing cattle than hay. STR pastures were effective in accumulating P due to redistribution, recycling, and retention of P.

#### *3.2. Spatial Distribution of Phosphorus*

With baseline hotspot analysis, all pastures and depths showed clusters of high M1P concentrations at low-lying areas that had high P transport potential (Figures 2a–c and 3a–c). Such hotspots must have been present due to continuous addition of manure and hay at congregation sites such as feeding areas and trees (natural shades). During Post-Treatment sampling, hotspots were still prevalent at similar locations in CHD pastures, but STR pastures showed such hotspots were more prevalent at higher elevations (Figures 2d–f and 3d–f). This has implications as to the reasons for reduced P losses in runoff from the STR pastures (see below).

**Figure 2.** Hotspots of Mehlich-1 phosphorus (M1P) concentrations at 0–5, 5–10, and 10–20 cm soil depths during Baseline sampling (2015; (**a**–**c**)) and Post-Treatment (2018; (**d**–**f**)) at the Eatonton pastures.

Change in M1P distribution was mapped for the Eatonton (Figure S1) and the Watkinsville (Figure S2) study pastures calculated as 2018 raster–2015 raster. Difference maps of Eatonton and Watkinsville showed increase in M1P at locations of greater elevations in the pastures as compared to lower elevations. The high lying areas in pastures, denoted by yellowish to red color in the elevation model, showed greater increase in M1P at all three sampling depths in 2018 as compared to 2015. Increased availability of soil P at higher locations would mean greater time for runoff sediments to settle, possibly leading to lower losses of particulate P in runoff.

STR pastures showed lower availability of M1P at low lying exclusions close to streams. However, CHD pastures showed no change with higher M1P at low-lying, edge-of-stream areas making them more prone to lose P in soil as DRP and TKP. These areas were not excluded and over seeded to provide soil cover. It was interesting to see how M1P values were still high in areas at the previous hay-feeding locations in both CHD and STR pastures. STR pastures due to more uniform distribution of change (no hotspots) presented the potential of better redistribution of P compared to CHD.

**Figure 3.** Hotspots of Mehlich-1 phosphorus (M1P) concentrations at 0–5, 5–10, and 10–20 cm soil depths during Baseline sampling (2015; (**a**–**c**)) and Post-Treatment (2018; (**d**–**f**)) at the Watkinsville pastures.

#### *3.3. E*ff*ect of Exclusions*

Both excluded and non-excluded areas, in STR pastures only, showed an almost five-fold increase from Baseline to Post-Treatment in the 0–5 cm layer. M1P in the 5–10 cm layer was almost twice as high in exclusions than in non-exclusions in 2018 (Table 4). Significant increases in M1P concentration in exclusions of the STR pastures at 0–5 and 5–10 cm layers suggest more P being captured by exclusions. The over-seeded exclusions could have helped slow runoff water and allowed for greater retention of particulate P even at 5–10 cm depth due to decreased bulk density and continued cover on the soil surface.



<sup>1</sup> Medians separated by different upper-case letters denote significant difference between sampling dates in either non-exclusion (Non-Exclusions, 2015 vs. Non-Exclusion, 2018) or exclusion (Exclusions, 2015 vs. Exclusion, 2018). <sup>2</sup> Medians separated by lower-case letters denote significant difference between non-exclusion and exclusion at individual sampling dates (Non-Exclusions, 2015 vs. Exclusions, 2015). Difference is at 0.05 level of significance.

#### *3.4. Changes in Runo*ff *Water Phosphorus*

Median concentrations and corresponding loads of DRP in runoff samples from CHD treatment was 0.14 mg P L−<sup>1</sup> and 0.03 kg P ha−<sup>1</sup> in 2015, 0.20 mg L−<sup>1</sup> and 0.11 kg ha−<sup>1</sup> in 2017, and 0.34 mg P L−<sup>1</sup> and 0.12 kg P ha−<sup>1</sup> in 2018, respectively. Similarly, in STR pastures, the median concentrations and corresponding loads of DRP were 0.38 mg P L−<sup>1</sup> and 0.11 kg P ha−<sup>1</sup> in 2015, 0.41 mg P L−<sup>1</sup> and 0.17 kg P ha−<sup>1</sup> in 2017, and 0.32 mg P L−<sup>1</sup> and 0.12 kg P ha−<sup>1</sup> in 2018, respectively. The DRP concentrations and loads for the CHD treatments were significantly greater in 2018 when compared to 2015 and 2017 concentrations and loads (Figure 4). Stratification of P near the soil surface makes it more prone to loss in runoff water through desorption of P from soil surface and residues (litter and shoot) by water [38]. In the STR pastures no significant differences in either concentrations or loads of DRP were observed between sampling dates. For a given sampling period, the only significant difference between treatments for DRP concentrations and loads was noted in 2015 data, where STR 2015 was significantly greater than the CHD 2015. Distribution of DRP concentration in runoff samples from the two grazing treatments during the three sampling years revealed the absence of <sup>≥</sup>2 mg P L−<sup>1</sup> concentration in Post-Treatment runoff samples in the STR pastures. We make note of this as there were two tropical storms that occurred during the period 2017–2018 suggesting that, even during extreme events, the STR grazing system was able to reduce large pulses of DRP in runoff. For example, in 2015 the maximum rainfall event was 163.3 mm with a rainfall intensity of 2.0 mm hr−<sup>1</sup> and in 2017 the maximum rainfall event (Hurricane Irma) was 103.6 mm with a rainfall intensity of 4.2 mm hr<sup>−</sup>1. Yet the maximum concentration in runoff was <2 mg P L<sup>−</sup>1.

**Figure 4.** Comparison of dissolved reactive phosphorus (DRP). (**a**) DRP concentrations (mg L<sup>−</sup>1), (**b**) DRP loads (kg ha<sup>−</sup>1), (**c**) total Kjeldahl phosphorus (TKP) concentrations (mg L<sup>−</sup>1), and (**d**) TKP loads (kg ha<sup>−</sup>1) from Baseline (2015) to 2017 and 2018 in continuous grazing with hay distribution (CHD) and strategic grazing (STR) pastures. Upper-case letters denote comparison between sampling years for individual treatment (example: CHD, 2015 vs. CHD, 2017 vs. CHD, 2018) and lower-case letters denote comparison between CHD and STR treatments for individual sampling periods (example: CHD, 2015 vs. STR, 2015). Different letters denote significance at 0.05 level.

Median concentrations and corresponding loads of TKP in runoff samples from CHD treatment were 0.94 mg P L−<sup>1</sup> and 0.30 kg P ha−<sup>1</sup> in 2015, 0.85 mg L−<sup>1</sup> and 0.40 kg ha−<sup>1</sup> in 2017, and 0.74 mg P L−<sup>1</sup> and 0.24 kg P ha−<sup>1</sup> in 2018, respectively. Similarly, in STR pastures, the median concentrations and corresponding loads of TKP were 1.26 mg P L−<sup>1</sup> and 0.33 kg P ha−<sup>1</sup> in 2015, 1.03 mg P L−<sup>1</sup> and 0.35 kg P ha−<sup>1</sup> in 2017, and 0.90 mg P L−<sup>1</sup> and 0.30 kg P ha−<sup>1</sup> in 2018, respectively. Comparison of TKP concentrations in runoff water between sampling years revealed significantly lower TKP concentration in 2017 sampling compared to 2015 in both treatments. In 2018, however, only STR showed significant decrease in TKP concentration in comparison to both 2015 and 2017. Over-seeded exclusions in STR treatment could have helped reduce the concentration of TKP in runoff water through retention.

#### *3.5. Relationship between Soil Phosphorus and Phosphorus in Runo*ff

The relationship between M1P, the plant available fraction of soil P and DRP and TKP loads in surface runoff was studied for CHD and STR grazing treatments over a three-year study period. We found that M1P at 0–5 cm depth was significantly correlated with DRP load in runoff water for both grazing treatments during both sampling dates (Figure 5). As for DRP loads, regression slopes derived from the relationship, DRP versus M1P (0–5 cm soil depth), during Baseline and Post-Treatment were compared using a simple regression model. No differences in the slopes were indicated for CHD pastures between Baseline and Post-Treatment, while the slope in STR pasture was significantly lower for Post-Treatment as compared to Baseline.

**Figure 5.** Relationship of (**a**) dissolved reactive phosphorus (DRP) load and (**b**) total Kjeldahl phosphorus (TKP) in runoff water and Mehlich-1 Phosphorus (M1P) in 0–5 cm depth soil compared between Baseline (2015) and Post-Treatment (2016–2018) sampling dates in continuous grazing with hay distribution (CHD) and strategic (STR) grazing treatments. \*, \*\*, and \*\*\* represent statistical significance at ≤0.05, ≤0.01, and ≤0.001, respectively.

Similar relationships between the M1P in soil and TKP in runoff water were studied for the two grazing systems during the two sampling dates. Results revealed significant positive correlation between the soil P and TKP in runoff water. Comparison of regression slopes for the two sampling periods indicated that the slopes were three-fold and two-fold lower during Post-Treatment as compared to Baseline in STR and CHD, respectively. Load of DRP in runoff water per the amount of M1P found in the soil was near half during Post-Treatment sampling compared to Baseline in STR. The intercept of lines relating soil P with TKP loads was significant for both treatments during the Baseline and Post-Treatment samplings. Positive intercept values suggest that the source of total P, other than M1P, could be organic matter at the soil surface [39], in our case, very likely dung and hay residues.

Transport of different P fractions in runoff and interactions with soil P, in exclusion and non-exclusion areas, provide additional insights into loss and transport mechanisms, especially in areas with high P transport potential. P application in areas with high P transport potential should be avoided [40]. More explicitly, areas of high P transport potential need to be managed to reduce P losses in runoff [15]. These would include: (1) steep areas close to streams and prone to erosion, (2) low-lying areas close to the streams, (3) concentrated flow-paths, and (4) high elevation areas with greater slopes. Exclusions placed in areas with high P transport potential can help reduce the P loss from these areas. In our study, exclusions at low-lying areas aided in the retention of particulate P and its vertical movement deeper into the rhizosphere. Vegetation in the exclusions served as buffers that interrupted the direct interaction of runoff water with the soil and slowed runoff water down, therefore facilitating deposition of particulate P. Furthermore, allowance of flash grazing of these areas resulted in continued use of these areas as grazing lands, while still reducing the amount of time cattle were present in the vulnerable areas causing chronic, direct deposition of manure. Vegetated exclusions acted as buffers and provided opportunity for the P in runoff to settle and infiltrate 10 cm into the soil while also reducing the amount of particulate P lost in runoff.

### **4. Conclusions**

After three years of application of STR grazing system, soil P in the 0–5 and 5–10 cm depths increased significantly compared to historically continuously grazed systems that had hay-feeding and watering at the same locations yearly. The CHD grazing system with hay distribution at different locations also increased soil P. In STR pastures, exclusions increased soil P at 0–5 and 5–10 cm depths in Post-Treatment showing retention of P and reduction in total P losses in runoff water. Exclusions also provided added benefits of forage availability during drought, and reduced interaction of animals with vulnerable low-lying locations in pastures. Combining rotational grazing (every 5 to 10 days) and lure management of cattle with movable equipages aided in recycling of soil P to less vulnerable high-lying portions of the pastures, which was demonstrated by hotspot analysis. Hotspots of soil P prevalent at vulnerable areas during the baseline period moved to higher locations in STR pastures allowing greater retention of P. With the increase in soil P, DRP losses increased in both treatments, however, less occurrence of larger P losses during extreme events in the STR treatment suggests that STR grazing systems could be more resistant to extreme weather events. TKP losses were reduced significantly with CHD and STR grazing management systems as they ensured distribution of P in areas less vulnerable to loss in runoff. Hence, CHD and STR grazing managements present considerable potential to retain P for forage use rather than being exported to aquatic systems. Further research is needed to determine the effectiveness of these management practices on pastures fertilized with broiler litter and in other landscapes beyond the Georgia Piedmont.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2571-8789/4/4/66/s1, Figure S1: Spatial distribution of change in Mehlich-1 phosphorus (change mg kg−<sup>1</sup> = 2018 raster−2015 raster) at Eatonton location. Figure S2: Difference and DEM of Watkinsville pastures: Spatial distribution of change in Mehlich-1 phosphorus (change mg kg−<sup>1</sup> = 2018 raster–2015 raster) at Watkinsville location.

**Author Contributions:** Conceptualization, D.F.; methodology, D.F. and A.S.; formal analysis, A.S.; investigation, A.S., D.F., S.D., and A.M.; writing—original draft preparation, A.S.; writing—review and editing, A.S., D.F., and M.C.; funding acquisition, D.F. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by NRCS-USDA, Conservation Innovation Grant. Grant number 69-3A75-14-251.

**Acknowledgments:** The authors are grateful to NRCS-USDA for their assistance with the first order soil survey, and to the Sustainable Agriculture Laboratory team, and Charles Trumbo at the University of Georgia for their support in lab and in field.

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