Agronomic soil tests can be used to estimate dissolved reactive phosphorus loss

Introduction

Phosphorus (P) is a primary plant nutrient that contributes to eutrophication in waterways through algal blooms, deoxygenation, and fish kills (Hodgkin and Hamilton 1993; Smil 2000). While a range of land use sectors contribute, diffuse agricultural sources often contribute the largest quantities of P because of its extensiveness (Kronvang et al. 2009). Reducing P loss is a priority to maintain beneficial uses of waterways, and to improve the efficiency and economics of limited farm inputs (Smil 2000; Ashley et al. 2011).

A range of factors contribute to the risk of P loss from agricultural soils, the forms of P that are lost, P mobilisation, and the pathways of P delivery to stream networks (Haygarth et al. 2005). These factors include but are not limited to landscape slope and shape, proximity to watercourses, depth to groundwater, the degree of waterlogging, soil P buffering, management factors (such as drainage, pasture type and cover, fertiliser rate and timing, soil test P and the degree of P saturation, and stocking rate), and factors beyond management control such as rainfall quantity and intensity.

Phosphorus loss risk indexes that account for one or more source, transport, and management factors have been developed to assist with nutrient management planning of agricultural land (Gburek et al. 2000; Birr and Mulla 2001; Eghball and Gilley 2001; Mallarino et al. 2002; DeLaune et al. 2004; Djodjic and Bergström 2005; Gourley et al. 2007; Heckrath et al. 2008). These indexes have been correlated with measurements of P loss (Sharpley et al. 2008), and most include agronomic soil test values (Olsen et al. 1954; Colwell 1965; Mehlich 1984) or measures of the degree of P saturation (DPS) (Nair et al. 2004) to contribute to the source component of the index. The DPS measures are most often based around molar ratios of P extracted in a soil test to the sum of iron (Fe) and aluminium (Al) extracted in the soil test, although ratios of other agronomic tests that mimic DPS have also been used. For example, in Sweden Blombäck et al. (2021) explored ratios of either Olsen P (Olsen et al. 1954) or ammonium lactate extractable P to the single point P sorption index of Bache and Williams (1971). They concluded that each of the DPS measures explored were equally capable of predicting dissolved P, and that well established agronomic tests could be used to determine soil fertility as well as the risk of P leaching. In Australia, Moody (2011) proposed a ratio of widely used agronomic tests such as Colwell P (Colwell 1965) and P buffering index (PBI) (Burkitt et al. 2002) as an environmental risk measure rather than introducing additional tests. This ratio of Colwell P to PBI is sometimes called the P environmental risk index (PERI). Moody (2011) demonstrated that PERI was strongly correlated with dilute CaCl₂-extractable P (CaCl₂-P), the benchmark method for the estimation of dissolved reactive P (DRP) leaving agricultural soils, and also identified a negative correlation between CaCl₂-P and PBI. Similarly, strong positive relationships have been identified between DPS and P concentration in run-off (Nair et al. 2004; Tarkalson and Mikkelsen 2004; Withers et al. 2017; Dari et al. 2018), as have relationships between soil test P and surface runoff P concentrations in rainfall simulation experiments (Burkitt et al. 2010). McDowell and Condron (2004) identified strong correlations between CaCl₂-P measured in soil extracts (0–7.5 cm) and DRP in subsurface flow, and demonstrated positive correlations between the ratio of Olsen P to P retention measures and CaCl₂-P or H₂O-extractable P, suggesting such ratios as a potential method to predict CaCl₂-P or H₂O-extractable P. Burkitt et al. (2010) developed empirical relationships between CaCl₂-P and Olsen P for a soil with high PBI, while Moody (2011) found strong relationships between CaCl₂-P and the ratio of Colwell P to PBI (R = 0.925); however, no empirical relationship between CaCl₂-P and Colwell P or PERI was provided. Furthermore, Bloesch and Rayment (2006) explored the relationship between agronomic soil tests and the potential to release soluble P from soils in the Great Barrier Reef catchments. They identified that CaCl₂-P declined with increasing soil PBI, and was correlated with the molar ratio of Mehlich-3 extractable P to Fe plus Al (R² = 0.67). Bloesch and Rayment (2006) found poor correlations between CaCl₂-P and Colwell P and suggested that separate relationships may be necessary for different land uses or different soil types but did not differentiate the data according to PBI or soil type. The authors concluded that CaCl₂-P could be a useful tool to identify soluble P loss risk and suggested that values >2 mg kg⁻¹ warranted further scrutiny from a nutrient management perspective, while Tyson et al. (2020) proposed an environmental threshold of 0.25 mg kg⁻¹ based on the trigger values for water quality in ANZECC and ARMCANZ (2000).

The previously identified positive correlations between CaCl₂-P and measures of DPS (McDowell and Condron 2004; Moody 2011) suggest that DRP loss risk will increase as legacy P stocks in soils increase (Burkitt et al. 2010; McCrackin et al. 2018). In addition, the nature of a quotient such as Colwell P to PBI ratio is that the risk of DRP loss increases as either Colwell P increases, or PBI decreases. It has been well documented for regions such as south-west Western Australia (SWWA), for example, that the soils have both low P buffering, and a high percentage (~75%) have Colwell P levels that exceed the requirement to achieve 90% or 95% of relative pasture yield (Weaver and Wong 2011; Rogers et al. 2021; Weaver and Summers 2021). This region therefore has a high risk of DRP loss, and DRP has been identified as the main P form lost from agricultural landscapes in this region (Summers et al. 2014; Weaver and Summers 2014). There is therefore significant potential through soil testing to inform the land manager to apply an evidence-based approach to fertiliser management, reduce legacy P stocks, and subsequently reduce DRP loss risk. This contrasts with a set-and-forget approach to fertiliser management in which the same nutrient application is applied year-in-year-out, and legacy P stocks and DRP loss risk have the potential to increase (Weaver and Summers 2021).

Agronomic soil samples for high rainfall pastures are typically collected from a single depth (0–7.5, 0–10, 0–15, or 0–20 cm) designed to represent the zone where plant roots will access most nutrients, as soil tests from these depths have been correlated with grass–legume pasture responses to nutrients (Gourley et al. 2019). Sampling depths designed for agronomic purposes, however, do not consider the dominant hydrological pathways of P transport, and hence use of agronomic sampling depths may limit the utility of these tests to identify locations and pathways of P loss risk. For example, studies show that P can be highly stratified in the soil (Ryan et al. 2017; Weaver and Summers 2021) with as much as 50% of the P stored to 1 m depth of soil found in the top 5 cm, and that stratified soil test P measured in shallow depths (0–5 cm) was better correlated with edge of field losses of dissolved and total P than sampling from 0–20 cm depth (Osterholz et al. 2020). When there is limited interaction of rainfall and runoff with the soil profile in inundated areas, or where there is strong stratification of P in the soil, soil tests derived from agronomic sampling depths may not provide an appropriate indication of DRP loss risk. Soils which have a highly-stratified P concentration in the top few centimetres of soil may desorb P into runoff water (Dougherty et al. 2006; Rivers 2012), particularly when there is a shallow water table and saturation excess, or during heavy rainfall and infiltration excess or in combination with steeper slopes (Zhang et al. 2019). A risk assessment based on an agronomic sampling depth of 0–10 cm may be misleading since the top few centimetres of soil with high DPS would be diluted by lower DPS values deeper in the collected sample. For example, a P-retentive clay soil which is strongly stratified with P in the surface when combined with a low infiltration rate may result in high soluble P in runoff due to P desorbing from P rich surface layers. A 0–10 cm soil sample would not directly identify this surface runoff risk. In contrast, a sandy, low P-retentive, P saturated soil would be identified as a P leaching risk with measures such as PERI derived from a 0–10 cm soil sample whereas a P stratified, poorly infiltrating, high P-retentive soil with potential for P runoff would fail to be identified.

It is possible, however, that P stratification in the soil profile may vary systematically, allowing analytes and derived variables at a specified depth to be estimated from a 0–10 cm agronomic sample. This is not an unreasonable expectation in undisturbed agricultural soils where P fertiliser is surface applied, and where ingress of P deeper into the soil profile is subject to the kinetics of P adsorption and desorption (Novak et al. 1975; Shah et al. 1975). Identification of systematic P stratification would increase the utility of agronomic samples, and when combined with knowledge of dominant hydrological pathways would provide a richer understanding of DRP loss risk from catchments.

This paper explores relationships between CaCl₂-P, measures of DPS derived from agronomic soil tests (e.g. PERI, PBI, and soil P fertility), and stratification of P in the soil. Much of the data used here are from land that has been farmed for more than 50 years, allowing study of the impacts of long-term P fertilisation at low P rates (10–20 kg P ha⁻¹ year⁻¹) on soil P fertility and the resulting reductions in soil P sorption capacity (Weaver and Wong 2011) as measured by unadjusted PBI (Burkitt et al. 2008). The findings are applied to the catchments of the Peel Harvey, Leschenault, Vasse Geographe, Scott River and Lower Blackwood, Wilson Inlet, Torbay Inlet, and Oyster Harbour, which are areas at risk of losing DRP in SWWA. The following hypotheses are tested.

Materials and methods

Data curation and analysis

The available data were presented as violin plots (Hintze and Nelson 1998) and scatterplots, stratified and symbolised by sampling depth, PBI, and CaCl₂-P. In addition, linear, non-linear regression, analysis of covariance (ANCOVA), and Spearman rank correlation were undertaken using the exploratory data analysis and presentation tools RStudio (rstudio.com), DataDesk 8.3 (datadesk.com), and Igor Pro v9 (wavemetrics.com). The midpoint of depth ranges was used where depth was included as a variable. The ANCOVA was used to model linear relationships between CaCl₂-P and Colwell P for overlapping PBI ranges, and to determine if systematic changes in the slopes of these relationships occurred with PBI. Overlapping PBI ranges were 0–5, 0–2.5, 2.5–7.5, 5–10, 7.5–15, 10–20, 15–30, 20–40, 30–80, 60–90, 80–120, 90–130, 100–150, 125–200, 175–250, 225–350, 300–400, 350–500, 450–700, 550–800, and 600–1200. Mean PBI within the overlapping PBI ranges was used as the dependent variable.

Data from the 88 soil profiles were presented by depth and symbolised by PFI and PBI. The paired 0–10 cm sample for each profile was classified into three PFI ranges (<0.5, 0.5–1.0, and >1.0) and three PBI ranges (<35, 35–280, and >280). Mean values of the corresponding paired data at each depth for the three PFI ranges and PBI ranges were overlain on the symbolised data to explore whether there were systematic changes in profile values based on a 0–10 cm soil sample. In addition, systematic changes in relationships for CaCl₂-P, PERI, PFI, and Colwell P at different sampling depths as the dependent variable, and the paired 0–10 cm sample as the independent variable, were explored using ANCOVA. The CaCl₂-P, PERI, PFI, and Colwell P were log transformed for this analysis.

Tradeoffs between pasture productivity and DRP risk (measured as CaCl₂-P) were modelled using a dataset of 30 981 geolocated 0–10 cm soil samples collected in SWWA during 2009–2022. This study established relationships between CaCl₂-P and Colwell P for different PBI ranges, allowing estimation of CaCl₂-P from Colwell P. The current DRP risk was determined from measurements of Colwell P and PBI in the dataset. Future DRP risk was estimated by predicting CaCl₂-P levels in the same samples at critical Colwell P values to achieve 80–95% relative pasture yield in southern Australia (Gourley et al. 2019), and also for 97% and 98% RY (data not shown). Eqns 4 and 8 from Gourley et al. (2019) were used to predict critical Colwell P values (Fig. 1), with Eqn 8 being a modification of Eqn 4 to use lower critical Colwell P values when soil PBI < 15 (Yeates 1993; Moody 2007). These equations will be referred to as Eqns 3 and 4 in future references.

(3)Critical Colwell P=19.6+1.1 × PBI0.55

(4)Critical Colwell P=ln( RY−100 −100 )(−0.196+ ( 0.045−0.227×exp(−0.201×PBI) )× PBI 0.179 )

Fig. 1.

Relationship between soil PBI and critical Colwell P based on Eqns 3 (dashed lines) and 4 (solid lines). Line thickness shows critical values for different relative yield targets.

Mean CaCl₂-P of samples that were geolocated within the soil groups of WA (Schoknecht and Pathan 2013) were determined to estimate both current and future DRP risk if Colwell P values were set to achieve 90% RY. The fractional change in CaCl₂-P from current levels to those in catchments managed at 90% RY were estimated and mapped. The DRP leaching and runoff potential was distinguished by modelled soil measures at sampling depths of 0–10 and 0–1 cm respectively.

Results

Mean pH (CaCl₂) was 4.9, with range 3.2–7.1 (Fig. 2a), and decreased with depth (Table 1). Mean PBI was 97.3, with range 1–908 (Fig. 2b), and had no correlation with depth (Table 1). Soil OC had a mean of 3.9%, range 0.2–8.2%, and decreased with depth (Table 1). Samples collected from <5 cm tended to have OC > 5%, while those >10 cm tended to be <2%. The 0–10 cm OC values mostly decreased between the shallow and deeper samples (Fig. 2c). Of the 88 profiles, 81 had either uniform or two adjacent textures (sand, sandy loam, sandy clay loam, or clay loam). The remaining seven profiles had a mixture of three or four textures. No heavier textures were recorded.

Fig. 2.

Split violin plots showing the distribution of (a) pH, (b) PBI, (c) OC, (d) Colwell P, (e) P₉₅ fertility index, (f) PERI, (g) CaCl₂-P, (h) Colwell K, and (i) KCl40S. Different sample depths shown by symbol size and colour. Arrows identify critical values or ranges where appropriate. Black triangle shows mean value.

Mean Colwell P was 49, with range 2–421, and decreased with sampling depth (Table 1). For samples collected from paired profiles, mean Colwell P in the 0–1 cm layer was 127 mg P kg⁻¹, compared to 15 and 48 mg P kg⁻¹ in the 20–30 and 0–10 cm layers, respectively (Fig. 2d). The SR of mean Colwell P_{20–30 cm: 0–10 cm} was 0.32 and Colwell P_{0–1 cm: 0–10 cm} was 2.7. The PFI₉₅ of the samples ranged within 0.05–32.4, with mean of 1.7, and decreased with depth (Table 1). Mean PFI₉₅ was 4.80, 1.76, and 0.64 in the 0–1, 0–10, and 20–30 cm layers, respectively (Fig. 2e), resulting in a SR of 0.36 and 2.7 for PFI_{20–30 cm: 0–10 cm} and PFI_{0–1 cm: 0–10 cm}, respectively.

Mean PERI in the 0–1 cm layer was 6.6, decreasing with depth (Fig. 2f, Table 1), consistent with a SR for mean PERI_{20–30 cm: 0–10 cm} of 0.42 and 2.8 for PERI_{0–1 cm: 0–10 cm}. Mean CaCl₂-P in the 0–1 cm layer was 26.5 mg P kg⁻¹, decreasing with depth (Table 1) and resulting in a SR of 4.9 for mean CaCl₂-P_{0–1 cm: 0–10 cm} and 0.18 for CaCl₂-P_{20–30 cm: 0–10 cm} (Fig. 2g).

Mean Colwell K in the 0–1 cm layer was 578 mg K kg⁻¹, decreasing with depth (Table 1); 32% of 0–10 cm samples exceeded a critical Colwell K value of 161 mg P kg⁻¹ for pastures grown on clay loam soils (Gourley et al. 2019). The KCl40-S decreased with depth (Table 1). Mean KCl40-S was 25.2 mg kg⁻¹, well above the pasture critical value of 8 mg S kg⁻¹ for 0–10 cm (Gourley et al. 2019).

All analytes were negatively correlated with depth (Table 1), supporting hypothesis 1, while there were also strong negative correlations of PBI with CaCl₂-P and PERI. Strong positive correlations were identified of PBI with Colwell P, Colwell K, and KCl40-S; of CaCl₂-P with PERI and PFI; of Colwell P with PFI and Colwell K; and of PFI with PERI and Colwell K. The OC was also positively correlated with PBI, CaCl₂-P, Colwell P, PFI Colwell K, and KCl40-S.

In support of hypothesis 2, CaCl₂-P decreased with increasing PBI (Fig. 3). Mean values of CaCl₂-P were 12.4, 9.4, 7.9, 11.3, 5.8, 0.64, 0.59, and 0.10 mg P kg⁻¹ for the 0–5, 5–10, 10–15, 15–35, 35–70, 70–140, 140–280, and 280–840 PBI ranges, respectively. Within a PBI range, CaCl₂-P tended to be lower for samples collected deeper in the profile. A distinct separation in CaCl₂-P for specific depths was seen in the PBI range of 35–70, with samples from 0–5, 0–10, and 10–30 cm having a mean CaCl₂-P of 22.1, 1.3, and 0.34 mg P kg⁻¹, respectively.

Fig. 3.

Violin plots of CaCl₂-P for different PBI classes. Size and colour of points represent sampling depth. Median values shown for each PBI group (●).

In support of hypothesis 3, log₁₀CaCl₂-P was linearly related with log₁₀PERI (Fig. 4) with an r² of 0.67. Soils with lower PBI tended to have higher PERI and CaCl₂-P than soils with high PBI. Despite these tendencies, high CaCl₂-P was not always restricted to shallow depths or when PBI was low.

Fig. 4.

CaCl₂-P as a function of PERI (Colwell P to PBI ratio) for soils of different PBI and sampling depth. log₁₀CaCl₂P = 1.44 × log₁₀PERI + 0.069; r² = 0.67.

Colwell P of high PBI soils spanned a higher range than for low PBI soils (Fig. 5). Low PBI soils had high CaCl₂-P, even when soil P fertility was low in subsoil layers. Low PBI soils had higher CaCl₂-P than high PBI soils, although some high PBI soils sampled from shallow depths and with high Colwell P values also had high CaCl₂-P. More of the high PBI soils had Colwell P values that exceeded a PFI₉₅ of 1, where RY is nominally 95%. The CaCl₂-P tended to be low for subsoil layers, particularly when soil PBI was high.

Fig. 5.

Colwell P as a function of PBI. Point colour represents CaCl₂-P and size represents sampling depth. Line overlays show P₉₅ fertility index and associated relative yields.

Compared to the raw data (Fig. 6), the SR (Fig. 7) showed less variation at the same depth. Based on the mean values, CaCl₂-P was more strongly stratified with depth than PERI, PFI, or Colwell P. Mean CaCl₂-P SR was 10.5, 4.2, 2.0, 0.71, 0.41, and 0.27; mean PERI SR was 2.9, 1.9, 1.2, 0.78, 0.67, and 0.56; mean PFI SR was 2.9, 1.9, 1.2, 0.71, 0.51, and 0.44; and mean Colwell P SR was 3.3, 2.1, 1.3, 0.71, 0.49, and 0.46 for the 0–1, 1–2, 2–5, 5–10, 10–20, and 20–30 cm depths, respectively.

Fig. 6.

Depth-wise changes in (a) CaCl₂-P, (b) PERI, (c) P₉₅ fertility index, and (d) Colwell P. Symbol colour shows the P₉₅ fertility index ( <0.2, 0.2–0.4, 0.4–0.6, 0.6–0.8, 0.8–1.0, 1.0–2.0, 2.0–3.0, 3.0–5.0,  >5.0), and symbol size shows the PBI range (small <5, 5–10, 10–15, 15–35, 35–70, 70–140, 140–280, and large 280–840). Plots overlain show mean values at each depth when categorised by specified PBI_0–10 and PFI_0–10 ranges. Gaussian noise has been added to depth to improve the visibility of overlapping data points.

Fig. 7.

Depth-wise changes in SR for (a) CaCl₂-P, (b) PERI, (c) P₉₅ fertility index, and (d) Colwell P. Symbol colour shows the P₉₅ fertility index ( <0.2, 0.2–0.4, 0.4–0.6, 0.6–0.8, 0.8–1.0, 1.0–2.0, 2.0–3.0, 3.0–5.0,  >5.0), and symbol size shows the PBI range (small <5, 5–10, 10–15, 15–35, 35–70, 70–140, 140–280, and large 280–840). Plots overlain show mean values at each depth. Gaussian noise has been added to depth to improve the visibility of overlapping data points.

The CaCl₂-P and CaCl₂-P SR decreased with depth (Figs 6 and 7), and were generally associated with decreasing PFI₉₅ down the profile. There were systematic changes in depth-wise mean CaCl₂-P values when classified by the three PBI and PFI₉₅ groups defined by the paired 0–10 cm sample, hereon described as PBI_0–10 and PFI_0–10. The CaCl₂-P decreased as PFI_0–10 decreased at every depth when PBI_0–10 < 35, decreased further for PBI_0–10 35–280 when PFI_0–10 was 0.5–1 and >1, and further again for remaining classifications of PBI_0–10 and PFI_0–10.

The PERI and PERI SR (Figs 6 and 7) tended to decrease with increasing depth, had lower values associated with lower PFI_0–10, and mean PERI values showed systematic changes in depth-wise values when classified by the three PBI_0–10 and PFI_0–10 groups. For PBI_0–10 < 35 PERI decreased as PFI_0–10 decreased to a depth of 7.5 cm and then remained relatively constant deeper in the profile. Mean PERI decreased further for PBI_0–10 35–280 when PFI_0–10 > 1, and further again for PFI_0–10 0.5–1.0 and for remaining classifications of PBI_0–10 and PFI_0–10.

There were systematic changes in depth-wise and classified mean PFI₉₅ values (Fig. 6). Depth-wise trends of PFI₉₅ within a PBI_0–10 group were similar and showed increasing mean PFI₉₅ as PFI_0–10 increased. When PBI_0–10 > 280, PFI decreased more rapidly than for PBI_0–10 < 35 or 35–280.

Colwell P and Colwell P SR decreased with depth and increased as PFI₉₅ increased (Figs 6 and 7). There were systematic changes in depth-wise mean Colwell P values when classified by the three PBI_0–10 and PFI_0–10 groups. For PBI_0–10 < 35, mean Colwell P at each depth decreased as PFI_0–10 decreased. Below 10 cm depth, Colwell P was less variable for all PFI_0–10 groups when PBI_0–10 < 35. When PBI_0–10 was >35, Colwell P continued to decrease with increasing depth.

The ANCOVA results exploring relationships between analytes at different depths are shown in Table 2. In support of hypothesis 4, linear relationships could estimate log₁₀ of CaCl₂-P, PERI, PFI, and Colwell P for a specified sampling depth based on the log₁₀ of the same variable derived from a 0–10 cm sample. A common slope for each of the relationships for CaCl₂-P, PERI, PFI, and Colwell P was identified; however, the intercepts varied systematically with midpoint sample depth as described by the equations and coefficients in Table 2.

Modelled depth-wise changes in CaCl₂-P, PERI, PFI₉₅, Colwell P, and SR for notional values of each analyte measured in a 0–10 cm sample are shown in Fig. 8. Similar to Fig. 7, the SR for CaCl₂-P, PERI, PFI₉₅, and Colwell P were less variable than the raw data. Higher values of each analyte measured in a 0–10 cm sample resulted in higher modelled values down the profile. The CaCl₂-P was more strongly stratified than the other analytes. Even though the depth-wise SR for each analyte was similar for each of the modelled 0–10 cm values, the SR values were inversely related to the 0–10 cm values. That is, lower 0–10 cm values led to higher values of SR even though the SR range was narrow compared to the raw data. For example, modelled values of CaCl₂-P SR ranged within 3–5 at a depth of 0.5 cm, while modelled CaCl₂-P ranged within 5.2–64 when CaCl₂-P in a 0–10 cm sample increased from 1 to 20 mg P kg⁻¹. At 30 cm depth, CaCl₂-P SR was in narrow range of 0.1–0.17 when CaCl₂-P in a 0–10 cm sample increased from 1 to 20 mg P kg⁻¹.

Fig. 8.

Modelled depth-wise changes in (a) CaCl₂-P, (c) PERI, (e) P₉₅ fertility index, (g) Colwell P, and (b, d, f, h) respective SR, using equations in Table 2 for example values of each variable measured in a 0–10 cm sample.

In further support of hypothesis 2, within a PBI range, CaCl₂-P was linearly related with Colwell P, resulting in a systematic decrease in slope of these relationships as soil PBI increased (Fig. 9). The ANCOVA identified that the slope of regressions for overlapping PBI ranges significantly differed (P < 0.05) from 0 and decreased with increasing mean PBI within each PBI range (Fig. 9a). Slopes decreased rapidly from near 1 for PBI < 5 to around 0.2 at PBI 30, and 0.03 at PBI > 100. Small significant differences in intercept were found for PBI range of 10–80, but for other PBI ranges the intercept was close to 0. Inclusion of all interactions in this model accounted for 71% of the variation, while removing terms for varying intercepts reduced the variation accounted for to 69%. Subsequent modelled relationships between CaCl₂-P and Colwell P assumed a zero intercept with a PBI dependent slope based on Fig. 9a.

Fig. 9.

(a) Curvilinear relationship between the slope of linear regressions between CaCl₂-P and Colwell P for overlapping PBI ranges. Slope = 0.048 + 0.911 × 0.961^PBI; r² = 0.99. (b) Modelled relationships between CaCl₂-P and Colwell P for soils of specific PBI from analysis of covariance of overlapping PBI ranges compared to 1:1 line. Coloured line overlays show different levels of soil P fertility index where a value of 1 is at the critical value for 95% relative yield. Lower (0.5) or higher fertility index (2, 3, and 4) values are multiples of the critical value of 1. Solid and dashed colour line overlays show P fertility index values based on Eqns 4 and 3, respectively.

Modelled relationships between CaCl₂-P and Colwell P were close to the 1:1 line when soil PBI = 1. The slope decreased as PBI increased (Fig. 9b), with little change in slope for PBI > 100. As PFI₉₅ increased, CaCl₂-P also increased. However, Eqn 3 resulted in CaCl₂-P values for soils with low PBI that were double those of Eqn 4. As PBI approached 30, the difference between the two equations decreased, after which modelled CaCl₂-P was similar.

The proportional change in current mean CaCl₂-P within each PBI group from that based on RY targets determined from Eqns 3 to 4 is shown in Fig. 10a. For low PBI soils (<35), Eqn 3 led to an increase in mean CaCl₂-P compared to Eqn 4 which showed a CaCl₂-P decrease, but this depended on the target RY. Increases in CaCl₂-P occurred when PBI < 35 at RY of 95%, PBI < 15 at RY of 90%, and PBI < 5 at RY of 85% for Eqn 3. For PBI > 35, both equations yielded similar results at all RY targets. That is, decreases in CaCl₂-P for PBI range of 35–280, and some increases in CaCl₂-P for PBI > 280, particularly when RY targeted ≥90%.

Fig. 10.

(a) Modelled proportional changes in mean CaCl₂-P for PBI ranges when the target Colwell P is based on values to achieve 80%, 85%, 90%, and 95% pasture relative yield (RY). (b) Proportion of 30 981 samples collected during 2009–2022 in specified PBI ranges (bars, right hand y-axis) overlain by modelled contribution to change in mean CaCl₂-P (lines, left hand y-axis) for PBI ranges when the target Colwell P is based on values to achieve 80%, 85%, 90%, and 95% pasture RY. Solid and dashed colour line overlays show P fertility index values based on Eqns 4 and 3, respectively. Total reduction in CaCl₂-P annotated for each pasture RY.

Of the 30 981 0–10-cm samples collected during 2009–2022, 80% fell into the PBI range 15–280, 8% into PBI < 15, and 12% into PBI > 280 (Fig. 10b). The greatest contribution to changes in mean CaCl₂-P were for the PBI range 15–280. Eqn 4 resulted in greater changes in CaCl₂-P than Eqn 3, which for PBI < 15 showed an increase in CaCl₂-P. The choice of RY target also influenced the change in mean CaCl₂-P within each PBI group, and the overall modelled change in CaCl₂-P. The overall reduction in CaCl₂-P at 95% RY was 24% and 16% using Eqns 4 and 3, respectively; these reductions increased to 59% and 54% at 80% RY (Fig. 10b). At 97% and 98% RY, the overall reduction in CaCl₂-P was 11% and 0%, respectively, using Eqn 4 (data not shown). At 97% and 98% RY, the CaCl₂-P reduced by 2% and increased by 9%, respectively, using Eqn 3 (data not shown).

Current modelled CaCl₂-P exceeded a threshold of 2 mg kg⁻¹ for the study catchments based on 0–10 cm (Fig. 11a) and 0–1 cm (Fig. 11d) sampling depths, with much greater exceedance for the 0–1 cm depth. Modelled CaCl₂-P reduced when Colwell P levels were reduced to achieve 90% RY for the 0–10 cm (Fig. 11b) and 0–1 cm (Fig. 11e) layers. Extensive areas in the study catchments had CaCl₂-P below a threshold of 2 mg kg⁻¹ for the 0–10 cm layer once the critical Colwell P levels for 90% RY were achieved (Fig. 11b), although extensive areas above the CaCl₂-P threshold of 2 mg kg⁻¹ also remained for the 0–10 and 0–1 cm layers (Fig. 11b, e). The mean fractional changes in CaCl₂-P at both depths were extensive once the critical Colwell P levels for 90% RY were achieved (Fig. 11c, f), and were typically of the magnitude of 25–50%. The greatest fractional change from reducing the CaCl₂-P to target 90% RY was in the areas near to the receiving estuarine waterbodies, especially the Peel Harvey, Leschenault, and Vasse Geographe catchments. These catchments also have high P loads in the runoff associated with high Colwell P in the soil (Fig. 11c, inset).

Fig. 11.

Spatial distribution of modelled (a, d) mean CaCl₂-P within soil groups (Schoknecht and Pathan 2013) of 30 981 soil samples collected during 2009–2022 for the 0–10 and 0–1 cm layers, respectively; (b, e) mean CaCl₂-P if soil Colwell P achieves RY of 90% (using Eqn 4) for the 0–10 and 0–1 cm layers, respectively; and (c, f) fractional change (b−aa and e−de respectively) in mean CaCl₂-P within soil groups for the 0–10 and 0–1 cm layers, respectively. Inset bar chart in (c) shows current annual P loads (Ruprecht et al. 2013) in selected catchments and estimated annual P loads once critical Colwell P levels in soil for 90% RY have been achieved.

Discussion

Coarse-textured soils with low PBI are unlikely to be the sole source of water quality problems associated with DRP because DRP loss risk also increases (Fig. 3, Table 1) with shallow sampling depths for any PBI (Fig. 2). Soils with high PBI, particularly when sampled from shallow depths, also contain elevated Colwell P levels sufficient to support CaCl₂-P at or above those of low PBI soils (Fig. 5). The implication is that soils of any PBI and texture subject to saturation excess or infiltration excess flow can contribute DRP, while coarse-textured soils with low PBI can also contribute DRP via leaching pathways. The DRP contribution of fine-textured high PBI soils should not be overlooked since they can sustain Colwell P levels of an order of magnitude higher than low PBI soils (Fig. 5). In addition, surface soil also likely contains elevated levels of Total P (Weaver and Summers 2021), which when eroded via surface runoff can contribute to particulate P losses.

While analytes such as Colwell P and PBI can be utilised to estimate the risk of DRP loss (Fig. 4), a standardised agronomic sampling depth of 10 cm may limit their utility. Knowledge of DRP loss risk using a measure like PERI from an agronomic soil test alone is insufficient. Prior to sampling for the purpose of estimating DRP loss risk, consideration should be given to the dominant flow pathways for DRP transport since sampling at agronomic depths will not necessarily align with these flow pathways. Since Colwell P and CaCl₂-P were highly stratified (Table 1), sampling to 0–10 cm will dilute the estimate of DRP loss risk compared to sampling to shallower depths. This is consistent with the findings of Osterholz et al. (2020) who identified a stronger relationship between shallow sampling depths and P concentrations in surface runoff and tile discharge than deeper sampling. For surface runoff, the exposure pathway aligns with the highest concentrations of soil P (0–1 cm) for soils of any PBI. For leaching situations, even though the zone of interaction between water and the volume of soil is expanded (0–10 cm), leaching usually occurs for coarse-textured soils with low PBI where opportunities for P sorption are limited. The result will be significant loss of DRP.

The risk of DRP loss, as determined by CaCl₂-P, can be estimated by PERI, making use of existing agronomic analytes (Fig. 4), or from Colwell P and PBI (Fig. 9). However, to be useful as an estimate of DRP loss risk, sampling depth should be adjusted, or the soil test values estimated for a depth that matches the dominant flow pathway, for example 0–1 cm when surface runoff is dominant. This may only be possible for permanent pasture-based systems where the nutrient stratification profile remains undisturbed. Such adjustments can be made based on the equations and coefficients in Table 2, where CaCl₂-P, PERI, PFI, and Colwell P at a specified soil depth can be estimated from the same analyte determined from a 0–10 cm sample (Fig. 8). For example, assume a 0–10 cm soil sample has been collected and analysed with a reported PBI of 10 and Colwell P of 25 mg kg⁻¹, and hence a PERI_0–10 of 2.5. Surface runoff has been identified as the dominant transport pathway and estimates of various soil parameters at a midpoint depth of 0.5 cm (0–1 cm sample) are required. Four approaches to estimate DRP loss risk using equations provided in this paper can be undertaken. Firstly, CaCl₂-P_0–10 can be estimated from PERI_0–10 (Fig. 4), with adjustments of CaCl₂-P to a midpoint depth using Table 2. Secondly, PERI_0–10 can be adjusted to a midpoint depth of 0.5 cm using Table 2, with subsequent estimation of CaCl₂-P from the depth-adjusted PERI value (Fig. 4). Thirdly, Colwell P_0–10 can be adjusted to a midpoint depth of 0.5 cm using Table 2 with subsequent estimation of CaCl₂-P from the depth-adjusted PERI value (Fig. 4). Fourthly, CaCl₂-P_0–10 can be estimated from the ANCOVA results (Fig. 9) based on the change in the slope of the relationship between CaCl₂-P_0–10 and Colwell P_0–10 for different PBI. The resulting CaCl₂-P_0–10 can be adjusted to a midpoint depth of 0.5 cm using equations in Table 2.

The first approach estimated a CaCl₂-P of 4.4 mg kg⁻¹ for the 0–10 cm sample (Fig. 4), and therefore 18 mg kg⁻¹ at midpoint depth of 0.5 cm (Table 2). A CaCl₂-P SR of 4.1 identifies that the surface loss risk of DRP from the 0–1 cm layer is over four times that from leaching, assuming uniform leaching and no short-circuit loss pathways such as macropores. The second approach estimated a PERI of 4.4, a PERI SR of 1.8 at a midpoint depth of 0.5 cm (Table 2), and a CaCl₂-P of 10 mg kg⁻¹ (Fig. 4). The third approach estimated a Colwell P of 68, Colwell P SR of 2.7, PERI of 6.8 at a midpoint depth of 0.5 cm (Table 2), and a CaCl₂-P of 33 mg kg⁻¹ (Fig. 4). The fourth approach estimated a CaCl₂-P_0–10 of 16.5 mg kg⁻¹ and CaCl₂-P at a midpoint depth of 0.5 cm of 54.4 mg kg⁻¹. While estimates from these approaches varied, they consistently provided higher values of CaCl₂-P at shallow depths of around 2–4 times that of a 0–10 cm sample estimate (Fig. 8b). The approach to estimate DRP risk will depend on what soil test parameters are readily available, and multiple approaches are recommended with ground truthing dependent on application. There are also cases where soil P fertility is determined using other measures such as Olsen P, which for agronomic purposes does not require PBI for its interpretation (Gourley et al. 2019). While Olsen P and an equivalent PERI using Olsen P rather than Colwell P were not explored in this paper, the P loss risks and stratification are likely similar given the strong relationship between the two measures (Moody et al. 2013). It will therefore be as important for soils managed with Olsen P tests to comply with critical soil test values (Gourley et al. 2019) to minimise DRP loss risk. In addition, PBI_u may be a useful adjunct measure where Olsen P is the default measure of soil P fertility to track changes in P sorption, particularly for soils at high risk of P leaching (Summers and Weaver 2022).

The study suggests that compliance with critical Colwell P values for fertiliser management can decrease legacy Colwell P and DRP loss. Hypothesis 5 is supported by the results, and the data in Fig. 10b illustrate the potential reduction under full compliance. The actual reduction in DRP, however, depends on several factors, some within the control of farm managers, agronomists, and fertiliser company decision support tools, and others not. Factors within control include the decision to use soil testing, the choice of equation to determine critical Colwell P values, the choice of RY targets, and the choice to modify P stratification in the soil profile. Factors such as rainfall amount and intensity, soil physical features such as texture and the dominance of surface runoff or leaching as primary transport pathways, as well as soil chemical properties such as PBI are less within the control of farm and catchment managers.

Soil testing, and adherence to critical Colwell P targets reduced DRP by a minimum of 16–24% at 95% RY, and up to 54–59% if lower RY were adopted (Fig. 10), while at 97% and 98% RY small decreases (2–11%) or increases (0–9%) occurred, respectively (data not shown). This is largely because critical Colwell P targets to achieve 90% or 95% RY were exceeded for a high percentage (~75%) of soils, except when PBI > 840 (Fig. 10a), in which case, applications of P are required, and CaCl₂-P loss risk will increase. The very small proportion of these high PBI soils significantly limits the potential for overall DRP loss risk. For low PBI soils (<35), there is potential to reduce or increase DRP loss risk dependent on the equation used to determine critical Colwell P values. Use of Eqn 4, which includes lower critical Colwell P values for PBI < 15, reduced DRP loss risk while Eqn 3 increased DRP loss risk. Overall, the potential for DRP loss risk between the two equations for PBI < 15 did not greatly differ because of the small proportion of soils with PBI < 15. However, soils with PBI < 15 can be located close to estuaries, limiting the potential for assimilation of P as it is transported within the stream network (Weaver and Summers 2021), hence they have a proximity risk which should not be ignored and Eqn 4 is recommended.

The choice of RY target and the critical Colwell P associated with that has a much greater impact on DRP loss risk than the choice of equation used to determine critical Colwell P values (Fig. 10). From an agronomic perspective, there is little or no benefit in operating above 95% RY (Gourley et al. 2019; Rogers et al. 2021). At critical Colwell P values that support 95% RY, there is a minimum reduction of DRP loss of 16–24%. However, it is usually only dairy farms or hay paddocks that need to operate at 95% RY. Much of the pasture grown for livestock in SWWA is for beef and sheep grazing and, for many growers, other nutrient constraints such as K or soil acidity, pasture utilisation, and lifestyle choices reduce the need to operate at 95% RY. Operating at 80–85% RY is a more realistic proposition for many growers, meaning that the reduction in DRP loss may be closer to 50%. This is not an unreasonable estimate of DRP reduction for SWWA given that water-limited pasture yield in this high rainfall zone is around 9–10 tonnes of dry matter ha⁻¹ year⁻¹, based on a notional 15 kg dry matter mm⁻¹ of growing season rainfall (Blumenthal and Ison 1993). Pasture utilisation on dairy farms in SWWA that have Colwell P levels in soil well above that which would support 95% RY has been measured at 4.5–6.5 tonnes ha⁻¹ year⁻¹ (Bolland et al. 2011). After accounting for the requirement to retain some residual pasture, RY targets are more likely in the range of 60–80%, and therefore a reduction in DRP of 50% is not unreasonable. In addition, the critical Colwell P values defined in Gourley et al. (2019) are based on growth response in a grass–legume system where clover is relied upon to fix nitrogen. Because clover has higher P requirement than grasses (Ozanne et al. 1969; Sandral et al. 2019), the potential for DRP reduction is likely to be greater for ryegrass-dominant pastures.

This estimate of DRP reduction was based on expected changes in CaCl₂-P from 30 981 0–10-cm samples. However, CaCl₂-P along with other soil analytes or derivates is highly stratified in the soil profile (Figs 6, 7, and 8), consistent with reports that repeated application of P to the topsoil has resulted in stratification of P (Dougherty et al. 2006; McLaughlin et al. 2011; Weaver and Summers 2021). The dominant transport pathway, leaching or runoff, and its interaction with stratified layers will therefore influence the magnitude of DRP loss and potential reduction. There is potential to interrupt these transport pathways and reduce stratification by burying P-saturated topsoil and bringing P-retentive soil to the surface, which must be considered against potential N or C losses. For example, Nash et al. (2007) reported that soil disturbance caused by laser levelling of irrigation bays resulted in 41% reduction in P loss to runoff and suggested that regular cultivation for pasture renovation decreases P export. On lower PBI soils (<35), the leaching pathway can be interrupted by amending the soil with P-retentive materials such as bauxite residue, which has been shown to reduce P leaching by more than 70% (Vlahos et al. 1989; Summers et al. 1993).

Despite the potential for up to 50% DRP reduction through the adoption of soil testing and evidence-based fertiliser management, it cannot be expected that these reductions will occur as soon as adoption of soil testing occurs, or that the magnitude of reduction will be correlated 1:1 with level of adoption. There are lags in the adoption of practices, time lags for practices to produce the required effect (Meals et al. 2010; Chen et al. 2019; McDowell et al. 2020; von Arb et al. 2021), and catchment feedbacks due to equilibrium P chemistry (Taylor and Kunishi 1971; Froelich 1988). These feedbacks mean that stream sediment will release P in response to reduced P concentrations from the edge of fields once practices are effective until a new in-stream equilibrium is established.

For practices to produce the required effect, a simplified paddock-based P balance example provides some insight into potential time lags. For a typical beef farm with annual P inputs from fertiliser of 10 kg P ha⁻¹ and exports in animal products of 2 kg P ha⁻¹ (Weaver and Wong 2011), the annual P balance excluding environmental losses is +8 kg P ha⁻¹. If we assume that soil testing and evidence-based fertiliser management are now adopted, a cessation of P inputs would result in most cases because current soil Colwell P levels exceed critical values and P applications are not required. Annual P inputs from fertiliser are now 0 kg P ha⁻¹ and animal products will contribute to an export of 2 kg P ha⁻¹, resulting in a P balance of −2 kg P ha⁻¹ until critical Colwell P levels are reached. The result is that under a fertilisation scenario, soil P levels will build at a faster rate than they will fall under a soil testing scenario with a complete cessation of P inputs (Schulte et al. 2010; Dodd et al. 2012; Tyson et al. 2020). This will likely lead to time periods for soil testing and evidence-based fertiliser management to have an effect that are greater than the time that was required to build soil P fertility to the current excessive levels. Similarly, typical P exports from sheep grazing of 1 kg P ha⁻¹ will lead to longer lag times than beef grazing, and 7 kg P ha⁻¹ for dairy shorter lag times, assuming a cessation of P inputs. These slow declines in legacy P stocks are supported by Bolland et al. (2011) who identified declines in Colwell P of 4.4–7.1 mg kg⁻¹ year⁻¹ following cessation of P inputs into pastures grazed by dairy cattle. More rapid declines in legacy P stocks may be achieved by excluding stock. Stock recycle 80–90% of the consumed P back to pastures (Hutton et al. 1967), while removing a 10 tonne hay crop from the farm and catchment would reduce legacy P stocks by 25–30 kg P ha⁻¹ year⁻¹ through phytomining (McDowell et al. 2020).

There are trade-offs between agricultural productivity and water quality. Based on the modelled estimates of current levels of CaCl₂-P, these trade-offs do not currently favour water quality because legacy stocks of soil Colwell P exceed critical values, contributing to unnecessarily high losses of DRP. Neither do the trade-offs currently favour agriculture because Colwell P in excess of critical values provides no agronomic benefit. Reducing legacy stocks of soil Colwell P to the critical values results in no disbenefit to agriculture while significantly reducing the loss of DRP to waterways. Once critical Colwell P levels are achieved, a trade-off will likely remain, albeit significantly less than currently exists (Fig. 11c, inset), and which represents the cost of conducting optimal agriculture. The journey to achieve critical Colwell P levels will in most cases require a cessation of P inputs until soil testing suggests otherwise, and a change in fertiliser practice on arrival. Typically the change in fertiliser practice will be away from capital P applications, and towards maintenance P applications based around product P outputs when critical Colwell P levels are achieved (Barrow 2015).

Water quality improvement plans developed for at-risk catchments in SWWA (Environmental Protection Authority 2008; Department of Water 2010; Hugues-dit-Ciles et al. 2012; White 2012) specify current and target P loads. Target P loads are typically 30–60% lower than current annual P loads. Evidence-based fertiliser management based on soil testing and compliance with critical Colwell P levels can make a significant contribution to the required reduction in DRP and P loads (Figs 10 and 11). This could be achieved with minimal cost and likely increase in farm profit by reducing the cost of P fertiliser inputs and increasing productivity by diverting some of these savings to addressing other limiting factors.

Conclusions

Existing agronomic tests such as Colwell P and PBI, and the derived measure of PERI, or the slope of the relationship between CaCl₂-P and Colwell P for a specified PBI are useful to estimate CaCl₂-P, and when coupled with knowledge on the dominant hydrological pathways can provide a richer understanding of DRP loss risk. Data from a 0–10 cm sample collected for agronomic purposes can be used to estimate DRP loss risk from other depths. The risk of DRP loss is greater from near surface depths than deeper in the profile due to stratification. The DRP loss risk can be significantly reduced without loss of utilised pasture in SWWA because of already elevated levels of P fertility in soils. Soil testing and evidence-based fertiliser management such as use of Eqn 4 for P recommendations have a significant role in reducing DRP loss risk. The DRP loss risk could be reduced by as much as 60% through compliance with recommendations from soil testing, use of appropriate critical Colwell P values, and aiming for realistic and economical relative yields.

Data availability

The data that support this study cannot be publicly shared due to ethical or privacy reasons and may be shared upon reasonable request to the corresponding author if appropriate.

Conflicts of interest

The authors declare that they have no conflicts of interest.

Declaration of funding

The soil samples collected and analysed in this project were supported by funding through the WA Government’s Royalties for Regions programmes: Regional Estuaries Initiative and Healthy Estuaries Western Australia, along with support from the Western Australian Department of Primary Industries and Regional Development.