Soil test phosphorus and cumulative phosphorus budgets in fertilized grassland

Soil samples were collected at Ercé and Gramond from 2001 to 2006; at Les Verrières in 1993, 1995, 2001, 2005, and 2008; at Lévis from 2001 to 2006; and at Maaninka and Siikajoki from 2003 to 2010. All soils were collected prior to fertilizer application, air-dried, passed through a 2-mm sieve, and stored at room temperature before analysis. Sampling depth and STP determination were based on existing recommendations in each country: 0–5 cm and Olsen extractable P (P _Ol) for Ercé and Gramond (Olsen et al. 1954); 0–10 cm and a mixture of 0.5 M ammonium acetate, 0.5 M acetic acid, and 0.02 M EDTA extractable P (P _AAE) for Les Verrières (Demaria et al. 2005); 0–15 cm and Mehlich-3 extractable P (P _M3) for Lévis (Mehlich 1984); and 0–20 cm and acidic ammonium acetate extractable P (P _AAC) for Maaninka and Siikajoki (Vuorinen and Mäkitie 1955) (Table 2). All sites were ungrazed, so there was no P return as livestock manure. At each site, the forage crop was harvested twice a year and dry matter (DM) yield was recorded. At Maaninka and Siikajoki, winter barley was harvested in August between 2003 and 2007. Forage P concentrations were determined on samples from Maaninka and Siikajoki following thermal oxidation at 550 °C and dissolution with HCl, and at the other sites using a mixture of sulfuric and selenious acids, as described by Isaac and Johnson (1976). The P removal for an individual cut was calculated by multiplying the forage DM yield by its P concentration. For a given year, annual P removal was obtained as the sum of P removal with the two cuts. The annual P budget was calculated as the difference between P applied as fertilizer and annual P removal, and B _cumP was calculated by summing up the annual P budgets (Messiga et al. 2012a, b, 2014a).

Data normality was verified using the Shapiro–Wilk statistic, and homogeneity was verified visually with graphics of the residuals (SAS Institute Inc. 2001). Analysis of variance was conducted for each site using the MIXED procedure of SAS. For B _cumP, P _M3, P _Ol, P _AAC, and P _AAE, the factors considered in the analysis of variance were replicate as random effect, year as repeated measurement, and P application as fixed effect. Differences between least square means (LSMEANS) for all treatment pairs were tested at a significance level of P = 0.05. In addition, linear and quadratic effects of P fertilizers on STP were determined for the Les Verrières and Lévis sites using polynomial contrasts. A linear-plateau model (SAS Institute Inc. 2001) was used to fit the STP data to B _cumP. When the linear-plateau model was significant, a deflection point (the point where two straight-line regressions intersect) with (x, y) coordinates in the range of data of STP plotted against B _cumP was determined. When the linear-plateau model was not significant, the parameters of the linear regressions between STP and B _cumP were estimated using the REG procedure of SAS.

The value of B _cumP varied greatly across sites and P application rates, from a minimum of −287 kg P ha⁻¹ at Gramond (Fig. 1b) after nine years without any P application to a maximum of 209 kg P ha⁻¹ at Lévis (Fig. 1d) after eight years of P application. At all sites and for the years following the start of the experiments, B _cumP significantly differed between P application rates (Fig. 1). Negative and decreasing B _cumP values were associated with annual P applications lower than 15 kg P ha⁻¹ (Fig. 1a,b), while positive and increasing B _cumP values were associated with annual P applications greater than or equal to 26 kg P ha⁻¹.

In the two experiments in France, the average annual P budget for the P₅₀ treatment was 8 kg P ha⁻¹ year⁻¹ at Ercé (Fig. 1a) and 11 kg P ha⁻¹ year⁻¹ at Gramond (Fig. 1b). These values are within the range of annual P budgets observed in 21 French agricultural regions (Senthilkumar et al. 2012b) and comparable to the 8 kg P ha⁻¹ year⁻¹ average P surplus reported for the European Union (Richards and Dawson 2008; Smit et al. 2009). In France, the national annual P budget in 2006 was 4 kg P ha⁻¹ as a result of a national strategy to decrease the total P inflow to soils (Senthilkumar et al. 2012a). Our results indicate that annual applications of less than 50 kg P ha⁻¹ are needed to reach this national target on French semi-permanent grasslands.

In the Swiss experiment, the average annual P budget was 10, 2, and −2 kg P ha⁻¹ year⁻¹ for the P ₂₆, P ₁₇, and P ₉ treatments, respectively (Fig. 1c). The current national mean P budget in Switzerland is 5.5 kg P ha⁻¹ as a result of direct payments to farmers for environmental programs aimed at decreasing nutrient surpluses (Spiess 2011). Our results indicate that to meet this target, the annual P application to grasslands should be less than 26 kg P ha⁻¹.

In the experiment conducted in eastern Canada, the average annual P budgets were 26, 10, and −4.5 for the P ₄₅, P ₃₀, and P ₁₅ treatments, respectively (Fig. 1d). Local recommendations in eastern Canada are based on P _M3 and Mehlich-3 extractable Al (Al_M3) content. For soils with P _M3 (61–90 kg P _M3 ha⁻¹) and Al_M3 (1100–1600 mg Al_M3 ha⁻¹), i.e., values similar to those of the Lévis site, the recommended P application rate ranges from 24 to 37 kg P ha⁻¹ (CRAAQ 2010). Our results suggest that for this fertility class, fertilizer P recommendations will mostly lead to positive B _cumP values in grassland and, consequently, to the build-up of STP and greater associated risks of P losses. In addition, continuous build-up of STP will decrease the probability of a positive yield response to fertilizer P application.

In the Finnish experiments, the average annual P budget for the fertilized plots was −8 kg P ha⁻¹ year⁻¹ at Maaninka and −9 kg P ha⁻¹ year⁻¹ at Siikajoki. This negative B _cumP with the P _x treatments at both sites (Fig. 1e, f) confirms that P applications corresponding to local recommendations (Valkama et al. 2009) were lower than P removal. The national average application of fertilizer P in Finland was about 30 kg P ha⁻¹ in the 1970s and 1980s, but dropped to 12 kg P ha⁻¹ during the 1990s in order to reduce the P status of cultivated soils (Saarela et al. 2004). As a consequence of the P fertilization regulations of the national agri-environmental support system since 1995, the national mean P budget decreased from 20 to 8 kg P ha⁻¹ year⁻¹ during the period 1990–2004 (Organization for Economic Cooperation and Development 2008). The consequence of current Finnish fertilizer P recommendations for grassland, as observed at Maaninka and Siikajoki, is a negative P budget. This is in line with the Helsinki Convention (Helsinki Commission 2005), which has adopted targets to reduce P discharge to waters by 50 % and to improve agricultural practices to maintain or even increase yield levels while lowering nutrient inputs.

The concentration of P _Ol at Ercé and Gramond was significantly (P < 0.001) increased by P application, but the extent of the increase varied with year (P application × year, P = 0.002 at Ercé and P < 0.001 at Gramond). This positive effect of P application occurred in all years, but the extent of the effect varied slightly from year to year and with no clear trend. The concentration of P _Ol at Ercé varied from 10.3 mg kg⁻¹ in 2003 for the P ₀ treatment to 45.3 mg kg⁻¹ in 2007 for the P ₅₀ treatment (Fig. 2a), while at Gramond, it varied from 20.6 mg kg⁻¹ in 2001 for the P ₀ treatment to 87.9 mg kg⁻¹ in 2004 for the P ₅₀ treatment (Fig. 2b).

The concentration of P _AAE at Les Verrières was significantly affected by P application, and the extent of change was affected by year (P application × year, P < 0.001). The P _AAE concentration varied from a minimum of 6.7 mg kg⁻¹ in 2008 for the P ₀ treatment to a maximum of 68.4 mg kg⁻¹ in 2008 for the P ₂₆ treatment (Fig. 2c). It significantly increased with time for the P ₁₇ and P ₂₆ treatments, and decreased slightly with time for the P ₉ and P ₀ treatments. Overall, P _AAE values tended to have a quadratic relationship with P application rate.

The concentration of P _M3 at Lévis was significantly influenced by P application (P < 0.001), and the extent of change was affected by year (P application × year, P < 0.001). The P _M3 concentration varied from a minimum of 23.8 mg kg⁻¹ in 2006 for P ₀ to a maximum of 83.9 mg kg⁻¹ in 2006 for P ₄₅ (Fig. 2d). It significantly increased with time for the P₃₀ and P₄₅ treatments, and decreased with time for the P ₀ and P ₁₅ treatments. Overall, P _M3 values tended to have a linear relationship with P application rate.

The concentration of P _AAC at Maaninka was significantly influenced by P application (P = 0.011) and year (P < 0.001). It varied from 15.5 mg kg⁻¹ in 2003 to 11.0 mg kg⁻¹ in 2010 for the P ₀ treatment and from 16.0 mg kg⁻¹ in 2003 to 12.8 mg kg⁻¹ in 2010 for the P _x treatment (Fig. 2e). The concentration of P _AAC at Siikajoki was not significantly influenced by P application, but significantly decreased with time (P < 0.001) from 14.9 mg kg⁻¹ in 2003 to 10.9 mg kg⁻¹ in 2010 (Fig. 2f).

Similarly, STP concentrations have been reported in other grassland studies with varying P application rates in France (Jouany et al. 2004), Switzerland (Gallet et al. 2003), Canada (Malhi et al. 2003, 2009), and Finland (Lkhagvasuren et al. 2011), indicating that soil type, soil P content, and management practices encountered in our long-term experiments were representative of grassland management systems in the respective countries.

Differences between the P ₀ treatment and P-fertilized plots (i.e., P ₅₀ for Ercé and Gramond, P ₁₇ and P ₂₆ for Les Verrières, and P ₃₀ and P ₄₅ for Lévis) in terms of STP concentrations could be explained by high residual P from fertilizer applied in the P-fertilized plots. The P fertilizer applied to grassland is not mixed with the soil and crop residues, but broadcast on the surface (Whitehead 2000). The transformation of residual P derived from fertilizer to less labile forms might be limited in grassland as this broadcast fertilizer P interacts with plant residues, blocking the sorption sites and, therefore, their adsorption or fixation on the solid phase (Messiga et al. 2012b). Grass stands and residues may alter P availability in fertilized grassland soils by reducing the transformation of soluble P derived from fertilizers to less labile forms by (i) organic anion replacement by H₂PO₄ on adsorption sites, (ii) coating of Fe and Al oxides by humus to form a protective cover and reduce P adsorption, and (iii) formation of stable organic complexes with Fe and Al, preventing their reaction with H₂PO₄ (Havlin et al. 2014). In addition, the limited transfer of residual P from the enriched upper layer to depleted lower layers due to the low mobility of P also contributes to differences in STP between unfertilized and P-fertilized plots (Messiga et al. 2013).

The STP generally increased with increasing B _cumP at all sites. At five of the six sites, however, a deflection point was observed with (x, y) coordinates of −133 kg P ha⁻¹ and 10.1 mg P _Ol kg⁻¹ at Ercé (Fig. 3a); −150 kg P ha⁻¹ and 22.0 mg P _Ol kg⁻¹ at Gramond (Fig. 3b); −61 kg P ha⁻¹ and 9.3 mg P _AAE kg⁻¹ at Les Verrières (Fig. 3c); −79 kg P ha⁻¹ and 26.4 mg P_M3 kg⁻¹ at Lévis (Fig. 3d); and −116 kg P ha⁻¹ and 11.3 mg P _AAC kg⁻¹ at Siikajoki (Fig. 3f). At Maaninka, however, no deflection point was observed in the relationship between STP and B _cumP (Fig. 3e).

Above the deflection point, the fitted linear regression lines were significant (0.33 < r ² < 0.72) for Ercé, Les Verrières, Lévis, and Siikajoki, but not for Gramond (Fig. 3). Using the slope of the linear regression above the deflection point at four of the sites and for the whole range of data at Maaninka, we calculated that a change in P budget of 100 kg P ha⁻¹ would result in a change of 18.9 mg P _Ol kg⁻¹ at Ercé (Fig. 3a), 25 mg P _AAE kg⁻¹ at Les Verrières (Fig. 3c), 22 mg P _M3 kg⁻¹ at Lévis (Fig. 3d), 2.7 mg P _AAC kg⁻¹ at Maaninka (Fig. 3e), and 3.1 mg P _AAC kg⁻¹ at Siikajoki (Fig. 3f). The STP changes for a unit of P budget estimated in this study tended to be greater than values encountered in other field crops under conventionally tilled soils. For example, in a long-term maize monoculture under conventional tillage, Messiga et al. (2010) reported a change of 3.3 mg P _Ol kg⁻¹ and 14 mg P _M3 kg⁻¹ for a change in P budget of 100 kg P ha⁻¹. Zhang et al. (1995) reported a change of 7 mg P _M3 kg⁻¹ in a maize monoculture under conventional tillage in eastern Canada. We could not find values for P _AAC and P _AAE in the literature due to a lack of similar studies in the corresponding countries. The greater change rates obtained under grassland indicate the maintenance of high residual P derived from fertilizer. Messiga et al. (2012b) obtained a high change rate of STP for a unit of P budget under no-till systems that translated into a polynomial distribution of STP data plotted against P budget. Soil sampling procedures can also introduce some variability in grassland samples, leading to greater soil test P. For example, shallow soil sampling at Ercé (0–5 cm), Les Verrières (0–10 cm), and, to some extent, Lévis (0–15 cm) resulted in greater change rates compared with Maaninka and Siikajoki (0–20 cm). Differences between extraction methods across the locations could also explain the differences in change rates, but it is difficult to draw conclusions on this point because the effects of this parameter were not tested at our study locations.

Below the deflection point, however, the fitted regression lines were not significant at the sites tested, and there was a slow or negligible STP decline in the absence of P fertilization. The differing relationships between STP and B _cumP on either side of the deflection point observed at Ercé, Les Verrières, Lévis, and Siikajoki have also been reported in other field crop studies (Webb et al. 1992; Ciampitti et al. 2011). This slow or negligible STP decline in the absence of P fertilization might be explained by the high buffering capacity of the solid phase (Power et al. 2005; Tunney et al. 2010). Ma et al. (2009) reported that in soils with field crops grown for several years without P fertilization, P _Ol always decreases before stabilizing at a lower limit of 3 mg kg⁻¹ despite further net losses of soil P. In a recent study, Messiga et al. (2014b) assessed the depletion of soil P following sequential extractions at several sites in eastern Canada, including the Lévis site, and found that the concentration of P _Ol by the end of the series of extractions remained constant at 2 mg kg⁻¹. At the Rothamsted experimental site, a negative P budget of −509 kg P ha⁻¹ was obtained for the period 1901–1974, but the decrease in P _Ol in the top 23 cm accounted for only 182 kg P ha⁻¹ or 36 % of the total P removed in the crops grown in this period (Johnston and Poulton 1977). In another long-term experiment at Saxmundham in Suffolk, UK, the decrease in P _Ol relative to a negative P budget in the period 1968–1984 accounted for only 10 % of the total P removed in the crops grown in the same period (Johnston et al. 2001). The authors attributed this to plant roots taking up P retained at sites in the soil from which it was not extracted by the Olsen reagent. Another possible explanation for this slow or negligible STP decline with no or little P fertilization might be that plant roots can take up P below the area used for soil sampling. Further studies with more than one sampling depth and sampling at greater depths at different sites would help confirm this hypothesis of P uptake at greater depths in situations of no or low P applications over several years.

Our results confirm the overall positive relationship between STP and B _cumP, but they also indicate that changes in STP can be less in situations of negative B _cumP and low STP values than in situations of positive B _cumP and high STP values. Consequently, this relationship between STP and the cumulative P budget should be used with caution when attempts are made at predicting changes in STP. This is particularly true in situations of low STP values. The deflection points reported above provide some values of STP with different soil extractants below which the relationship between STP and B _cumP should not be used to predict STP changes in grasslands. Above the deflection point, however, the relationship provides a valid estimate of expected changes in STP over time as a function of B _cumP and could be used as part of a strategy aimed at lowering the P content of P-saturated grassland soils.