Subsoil testing required to detect the rundown of soil potassium to deficient levels for wheat production on loam-textured soils

Introduction

Long term rundown of soil potassium (K) reserves in agricultural soils is common globally (Brownlie et al. 2024; Wang et al. 2024). Negative K balances (removal in harvested products exceeds inputs) over decades have been reported for Australia, India, sub-Saharan Africa and the southern cone of Latin America (Majumdar et al. 2021; Das et al. 2022; Ma et al. 2022). Where plant-available K is above the level where deficiency occurs, negative K balances are an economically rational approach because there is no immediate economic benefit from applying fertiliser K. However, this approach relies on accurate predictions by empirical models to identify when plant-available K has been depleted to deficient levels and a profitable response to fertiliser K is likely.

Soil testing for plant-available K is an important tool for monitoring soil K rundown and predicting when K deficiency is limiting the growth of crops or pastures. Soil test calibration curves are used to define the relationship between a soil test level and relative yield, and to quantify critical levels (and critical ranges), typically the soil test level required for 90% of maximum yield (Brennan and Bell 2013). Soil test calibration curves can differ between crop species, soil test method and soil type (Siatwiinda et al. 2024). The development of these calibration curves needs to include regionally relevant combinations of the factors above and be reassessed over time as soil conditions and production systems change.

Plant testing is an also a valuable tool for monitoring the nutrient supply from soils used for crop production. Like soil test calibration curves, plant test calibration curves relate relative yield to shoot nutrient concentration in whole shoots or specific shoot parts (Murrell and Pitchay 2021; Bell 2023), and the critical level is usually defined as the shoot nutrient concentration that relates to a relative yield of 90%. Plant test calibrations differ from soil testing in that the critical level is dependent on growth stage because the concentration required for maximum yield decreases as the growth stage increases (Bell 2023). Current critical levels for K in use for the major grain crops in Australia are derived from glasshouse (Brennan and Bolland 2004; Brennan and Bolland 2007) and field experiments (Brennan and Bolland 2006). Plant testing complements soil testing because it is possible to respond tactically with fertiliser application in the growing season when deficiencies are detected.

The occurrence of K deficiency in soils used for crop production in Western Australia (WA) has correlated with the native K status of the major soil groups used for agricultural production. Potassium deficiency symptoms were first detected in pastures grown on deep sands in the 1950s (Fitzpatrick and Dunne 1956; Drover 1961; Toms and Fitzpatrick 1961) and in crops grown on duplex soils approximately four decades later (Wong and Witwer 1997; Wong et al. 2000). Based on these findings, research on K nutrition of grain crops has focussed on soils with a sand texture at the surface; 77% of trials in a national database were done on sandy soils in WA (Brennan and Bell 2013). Recently, our monitoring of shoot nutrient concentrations has revealed shoot K concentrations that are low, or marginal in comparison to critical shoot concentrations (Reuter and Robinson 1997) in crops grown on loam-textured soils where Colwell K for 0–10 cm depth was above the current critical soil test levels, for example 40 mg kg⁻¹ in Chromosols (Brennan and Bell 2013). The aims of this study were to examine whether (1) K deficiency is emerging on loam-textured soils via grain yield or quality responses to fertiliser and (2) if a soil test calibration curve can be identified for these soils.

Methods

Results

Relationships between soil chemical properties

Measurements of soil K were highly correlated among methods of extraction and between soil layers. The Pearson correlation coefficient for Colwell K and Exch-K or AA-K was greater than or equal to 0.95 at all soil depths, and for Exch-K or AA-K was 0.99 or 1 at all soil depths (Data for 0–10 cm shown in Fig. 1). Deming regression showed that Colwell K was about 60% of TPB-K at 0–10 cm depth, and about 30% of TPB-K at 20–30 cm, 30–40 cm and 40–50 cm (Fig. S1). Colwell K was 20% to 30% higher than Exch-K or AA-K at 0–10 cm and 10–20 cm, 10% higher at 20–30 cm and equivalent at 30–40 cm and 40–50 cm (Fig. S2, S3). Colwell K for each depth was highly correlated (r > 0.8) with depth increments that were adjacent (above or below) but less so for the 0–10 cm vs 30–40 or 40–50 cm depths (Table S1).

Fig. 1.

Correlation plot for selected soil chemical properties (0–10 cm depth) at the trial sites. Values shown are the Pearson correlation coefficient. Variables are grouped based on their correlations using hierarchical clustering. The symbol ‘X’ shows correlations for which P > 0.05. Correlation plot was produced using the ggcorrplot package in R (Kassambara 2019).

Comparison of soil test calibration curves for different sampling depths and soil properties

Overall, AA-K, Exch-K and Colwell K provided the best calibration curves (Fig. 2). Based on the values within, and width of the 90% CI for r-value and RMSE, the calibration curves improved as sampling depth increased for AA-K, Exch-K and Colwell K, worsened as sampling depth increased for PCT-K, and did not change with sampling depth for the other soil variables.

Fig. 2.

90% confidence intervals obtained from the bootstrap samples for r-value and root mean square error (RMSE) for the arcsine-log calibration curves that relate soil properties for increasing depth intervals to crop relative grain yield. r-value = Pearson correlation coefficient of the transformed soil test and RY data. AA-K = ammonium acetate exchangeable K, Exch-K = ammonium chloride and barium chloride exchangeable K, TPB-K = sodium tetraphenylboron K, PCT-K was calculated from CEC and Exch-K, pH = soil pH (CaCl2), CEC = sum of ammonium chloride and barium chloride exchangeable calcium, magnesium, K and sodium, OC = organic carbon. Soil properties are ordered top to bottom using median r-values for soil properties.

For AA-K, Exch-K and Colwell K, there was a marginal improvement in the 90% CIs for r-value and RMSE as sampling depth increased from 0–10 cm to 0–20 cm, and 0–20 to 0–30 cm, a larger improvement from 0–30 to 0–40 cm, then a marginal improvement from 0–40 to 0–50 cm. For example, the 90% CIs for r-value for AA-K 0–10, 0–20, 0–30, 0–40 and 0–50 cm were 0.32–0.83, 0.37–0.84, 0.47–0.85, 0.56–0.87 and 0.59–0.86, respectively. The 90% CIs for RMSE for AA-K 0–10, 0–20, 0–30, 0–40 and 0–50 cm were 5.1–9.7, 4.8–9.2, 4.3–7.6, 3.7–6.0 and 3.7–6.1, respectively. Our analysis suggests that to gain a meaningful improvement in the soil test calibration curve for AA-K for 0–10 cm, sampling 0–40 cm is required. The calibration curves for AA-K, Exch-K and Colwell for 0–40 cm depth were similar. The 90% CIs for r-value for Exch-K and Colwell K 0–40 cm were 0.59–0.88 and 0.51–0.87, respectively, and the 90% CIs for RMSE for Exch-K and Colwell K were 3.8–6.0 and 4.1–6.9, respectively. Our analysis below focusses on Colwell K because it provides a similar level of accuracy to AA-K and Exch-K and is an industry standard for soil K analysis in Australia.

The soil test calibration curves for Colwell K for the different sampling depths are shown in Fig. 3. This figure illustrates the improvement in the goodness of fit of the calibration curves as sampling depth increases. The CSTV for the 0–20 and 0–30 cm sampling depth was not different to that for 0–10 cm, however, CSTV for 0–40 cm sampling depth is significantly lower than 0–10 cm.

Fig. 3.

Comparison of soil test calibration curves for Colwell K at different soil sampling depths. Data shown are the relative yield (RY) and Colwell K for replicates within the field trials. Soil test calibration curves were modelled with the arcsine-log calibration curve (ALCC) method in the soiltestcorr package (Correndo et al. 2023) in R (R Core Team 2024). r-value = Pearson correlation coefficient of the transformed soil test and RY data, RMSE = root mean square error for RY, CSTV = critical soil test value for 90% RY. Values shown are the 90% confidence interval (CI) from the bootstrap samples. Dashed vertical lines are the 90% CI for CSTV. Dashed horizontal line is 90% RY.

Comparison of yield responses with shoot K concentration

Relative yield showed a closer relationship with shoot K concentration at stem elongation than at flowering (Fig. 4). The 90% CI for r-value was higher for stem elongation sampling than flowering, and the 90% CI for RMSE was lower for the stem elongation sampling than flowering.

Fig. 4.

Comparison of plant test calibration curves for shoot K concentration measured at stem elongation (a) and flowering (b). Data shown are the relative yield (RY) and shoot K concentration for each treatment and replicate within the trials. Plant test calibration curves were modelled with the arcsine-log calibration curve (ALCC) method in the soiltestcorr package (Correndo et al. 2023) in R (R Core Team 2024). r-value = Pearson correlation coefficient of the transformed plant test and RY data, RMSE = root mean square error for RY, CPTV = critical plant test value for 90% RY. Values shown are the 90% confidence interval (CI) from the bootstrap samples. Dashed vertical lines are the 90% CI for CPTV. Dashed horizontal line is 90% RY.

Discussion

Significant grain yield responses of 0.69 to 1.37 t ha⁻¹ to fertiliser K in 4 of 8 field experiments provided confirmation that on some loam-textured soils, grain yield loss is occurring to K deficiency if no K fertiliser is applied. This is the first report of K deficiency for wheat on these soils in WA.

The relationship between yield response to fertiliser K and methods that measure exchangeable K identified in our analysis is in agreement with a glasshouse study using a wider range of soils (Cheng et al. 2023). In our study, the 90% CI for r-value and RMSE were similar for AA-K, Exch-K and Colwell at each depth, meaning that any of these methods could be used to develop reliable soil test calibration curves for the soil types included in this study. In the glasshouse study, wheat K uptake showed a better relationship with AA-K than Colwell K, but the difference in goodness of fit was small (ΔBIC, the change in the Bayesian information criterion, = 8). Combined, these studies show that Colwell K is adequate for measuring plant-available K over a range of soils used for cropping in WA. Also, both studies showed closer relationships between crop growth (K uptake in the glasshouse study, RY in the current study) and Colwell K or AA-K than with measurements of non-exchangeable K. The poorer fit of calibration curves for TPB-K compared to AA-K, Exch-K or Colwell K in this study indicate that non-exchangeable K is currently not a major component of K supply to crops on these soils.

Increasing sampling depth improved the goodness of fit of soil test calibration curve. The 90% CI for r-value for the calibration curve for Colwell K at 0–10 cm depth in the current study was 0.32–0.82, overlapping the value for Chromosols (r = 0.46, n = 46) and Kandodols (r = 0.78, n = 37) in a national study (Brennan and Bell 2013). Our results showed a marginal improvement in the soil test calibration curve by sampling 0–20 cm, or 0–30 cm compared to 0–10 cm, but a more meaningful improvement when sampling 0–40 cm, compared to 0–10 cm. The improvement in the goodness of fit as sampling depth increases is attributed to crop uptake from both surface and subsoil layers. The contribution of K from the subsoil to total uptake can be significant; on average 34% of K uptake by wheat was derived from the subsoil (Kuhlmann 1990), and it can change the grain yield response to K fertiliser on soils with low exchangeable K in the surface layer (Scanlan et al. 2015).

The rundown of subsoil K in long-term experiments provides further evidence of the contribution of K from subsoils to crop uptake. The depletion of AA-K and TPB-K stocks was greater in subsoils (20–100 cm) than topsoils (0–20 cm) over 9 years of continuous cropping (Correndo et al. 2021). In a 16-year study, the annual rate of decline (mg kg⁻¹ year⁻¹) for AA-K and nitric acid-K at 20–40 cm was the same as for 0–20 cm. These results, in combination with the improvement in calibration curve with increased sampling depth, suggest that sampling to 40 cm on loam-textured soils, at intervals several years apart e.g. 5 years, is required to monitor the rundown of soil K reserves.

The impact of subsoil K on soil test calibrations in previous studies in WA is mixed. Brennan and Bolland (2006) found that the threshold of Colwell K for a response to fertiliser K was similar for 0–10, 10–20 and 20–30 cm. However, we reanalysed their data using the same approach as in this study (ALCC method for mean soil test values for sampling depth, confidence intervals calculated from bootstrap samples) and found that the calibration curve improved as sampling depth increased (Fig. S4). The improvement in goodness of fit as sampling depth increases supports sampling 0–20 cm for canola to provide a more reliable prediction of relative yield than 0–10 cm. However, in a previous study with wheat conducted on Chromosols, Wong et al. (2000) found that a depth-weighted function for the soil K profile K did not provide a better prediction of relative yield than Colwell K at 0–10 cm depth alone. In the study by Wong et al. (2000), Colwell K at 0–10 cm depth was less than 50 mg kg⁻¹ at all sites and was less than 50 mg kg⁻¹ below 10 cm depth for 8 of the 10 sites. It is likely that the greater range of Colwell K concentrations at all depths in our current study and the study by Brennan and Bolland (2006) has allowed the impact of subsoil K on grain yield response to fertiliser K to be detected.

The 90% CI for Colwell K at 0–10 cm depth in our current study (60–90 mg kg⁻¹) is higher than reported for wheat in a national analysis based on Colwell K (Brennan and Bell 2013). The CSTV (and range) was 45 mg kg⁻¹ (41–49) for all soils in the national study (n = 201) and ranged from 41 mg kg⁻¹ (32–52) for Tenosols (n = 70) to 64 mg kg⁻¹ (57–70) for Brown Ferrosols (n = 18). The 90% CI for Colwell K (0–10 cm) in this study only overlapped with Brown Ferrosols in the national study. The reason for the higher critical level in this study is not clear. It may be due to higher yields in this study than in previous research, but the impact of grain yield potential was not consistent in the national study; critical level was influenced by yield level in Tenosols but not Kandosols (Brennan and Bell 2013). In previous work on N × K interactions for wheat, RY was lower when 138 kg N ha⁻¹ was applied [RY = 0.65] compared to the nil N control [RY = 0.81] (Fig. 1, Brennan and Bolland 2009). The higher critical level may be due to higher levels of CEC in this study, but it was not reported in the national study, so the relationship between CEC and critical level over the historical and current data cannot be assessed. The influence of sampling depth on CSTV was not reported for the national study, so it’s not possible to compare this aspect of sampling between the two studies either.

The long-term rundown in Colwell K and negative K balances in all soils of the cropping regions in WA (Harries et al. 2021; Ma et al. 2022) suggests that the incidence of K deficiency on coarse (sand, loamy sand, sandy loam) and medium (loam, silt loam, silt, sandy clay loam) textured soils will increase in the future. As a result, there will be increasing demand for empirical models to include a wider range of soils than in the current study. Future research designed to calibrate these models should include measurements of soil properties that are likely to influence K buffering; the proportions of minerals present in the clay fraction (Breker et al. 2019), clay content and CEC (Gourley et al. 2019). Although RY was not closely related to TPB-K in this study, we recommend this measurement be included in future work as the influence of non-exchangeable K on crop response to fertiliser K may increase as exchangeable K is further depleted and as a wider diversity of loamy and clay-textured soils are assessed. Also, the influence of fertiliser K placement should be examined; it is likely that there will be a benefit from banding compared to surface application on loam and clay-textured soils where CEC is greater than 15 mol_c⁺ kg⁻¹ (Bell et al. 2021).

The rate of rundown of soil K levels from adequate to deficient on loam-textured soils will depend on the yields of grain removed but will be accelerated by hay production in the crop sequence. Based on the 5-year average grain yield in this region [DPIRD Agzone H3] (Planfarm 2019, 2020, 2021, 2022, 2023) and typical grain K concentrations in WA (Ma et al. 2022), annual removal for wheat, barley and canola would be 13, 15 and 16 kg K ha⁻¹ per year respectively. In this region, dry matter yield for oaten hay is typically 6 to 8 t ha⁻¹, removing 102 to 136 kg K ha⁻¹ per year (Blackwood 2007). Previous work with WA soils has shown that cumulative K removal over six harvests of ryegrass was closely related to exchangeable K plus water-soluble K with relatively small changes in non-exchangeable K (Pal et al. 2001). The dependence on exchangeable K in WA cropping soils identified in previous work (Pal et al. 2001; Cheng et al. 2023) suggests that plant-available K in loam-textured soils is susceptible to rundown, but the rate of rundown, and equilibrium between K pools has not been characterised for these soils and should be addressed in future research. The equilibrium between these pools is critical, because this study has shown significant reserves of non-exchangeable K in the sub-surface layers in loam-textured soils; TPB-K levels were about three times Colwell K.

Our analysis of plant test calibrations curves showed that shoot K analysis at stem elongation is valuable for monitoring crop K status. Analysis of shoot K concentration at stem elongation provides some flexibility in K management; if a deficiency is identified with plant testing it is possible that a yield response to applying fertiliser K could occur at this stage (about 8 weeks after sowing) (Ma et al. 2015). The 90% CI for critical plant test value for the stem elongation calibration curve is higher than the value reported from a glasshouse study by Brennan and Bolland (2007) (2.25% K) for 90% RY, and for the flowering calibration curve, falls between the critical level in whole shoots at head emergence [Z50] (1.5% K) and milky-dough [Z75) (0.9% K) reported by Reuter et al. (1997). It’s possible that the higher critical level for stem elongation in the current study compared to previous research is due to higher yield levels, but there is not sufficient data to explore this relationship.

The emergence of K deficiency on loam-textured soils poses a new management challenge for grain production on these soil types: management of spatial variation. The loam-textured soils of the Avon sub-region are part of dissected landscapes where soil types vary from shallow sandy, gravelly surfaced soils in the upslope, to duplex soils with sandy-loam to loam-textured A horizons and sandy-clay to clay-textured B horizons in the mid-slope, and sand over clay duplex soils in valley floors (McArthur 2004). Airborne and ground-based radiometric surveys have potential for identifying the presence, or depth of ironstone gravel (Taylor et al. 2002; Pracilio et al. 2006; Wong et al. 2009) in these landscapes, which may have practical value in delineating soil zones that differ in K supply, but these relationships need to be investigated. It’s likely that a combination of visual cues – better growth on harvester windrows from the previous year (Brennan et al. 2004) or visual leaf symptoms of K deficiency (e.g. Murrell and Pitchay 2021) – and plant and soil sampling within soil management zones will be required to optimise the spatial management of K on these soils.

Conclusion

This study showed wheat grain yield response to the application of fertiliser K (P < 0.05) on loam-textured soils used for crop production in WA, confirming the emergence of K deficiency on these soils that were previously considered non-responsive to K. Wheat grain yield response to fertiliser K was related to exchangeable K but the soil test calibration curve for all measurements of exchangeable K improved as the depth of sampling increased. Soil testing to 40 cm depth was required for the most accurate prediction of relative yield. The improved prediction of relative yield with increased sampling depth is attributed to the contribution of subsoil K to total K uptake. Analysis of shoot K concentration at stem elongation provides complementary data to soil testing and can be used to guide in-season fertiliser K applications.

Supplementary material

Supplementary material is available online.