Comparison of soil analytical methods for estimating plant-available potassium in highly weathered soils

Soil collection

Twenty-one soil types with a wide range of properties (pH, organic carbon, clay content and mineralogy) were collected at 0–10, 10–20 and 20–30 cm depths from soils used for grain production in WA. Many of the sites were adjacent to Reference soil sites (McArthur 2004) studied by Pal et al. (1999), but on soil that has been used for grain or pasture production, rather than native vegetation. After being air-dried, the soils were sieved through a 2-mm sieve for soil analyses and the glasshouse experiment. Basic soil properties were measured according to the methods of Rayment and Lyons (2011): pH was measured using 0.01 M CaCl₂ solution with a soil:solution of 1:5 (4B4); electrical conductivity (EC) was measured using 1:5 soil:water extract (3A1); organic carbon (OC) was measured using the Walkley–Black method (6A1); exchangeable cations (Ca, Mg, Na and K) were extracted using a mixture of 0.1 M ammonium chloride and barium chloride at a ratio of 1:10 (15E1). Soil particle size was measured using mid infra-red method at the CSBP Soil and Plant Analysis Laboratory. Soil classification and some basic soil properties are presented in Table 1.

Soil K analysis

Pot experiment

From the 0–10 and 20–30 cm depths of each soil type, 2 kg of soil was weighed into plastic pots (height 17 cm, diameter 14 cm) with three replicates. Basal nutrients minus K were applied at the following rates (mg/kg): 237 Ca(NO₃)₂·4H₂O, 150 NH₄H₂PO₄, 80 MgSO₄·7H₂O, 18 FeSO₄·7H₂O, 14 MnSO₄·H₂O, 8.3 CuSO₄·5H₂O, 9 ZnSO₄·7H₂O, 0.37 H₃BO₃, 0.3 CoSO₄·7H₂O and 0.37 NaMoO₄·2H₂O (Ma and Bell 2016). Twelve pre-germinated seeds of wheat (Triticum aestivum) were sown into each pot and thinned to six plants after 5 days. From the third week, 80 mg of NH₄NO₃ per kg soil was added to all pots every 2 weeks. All pots were watered to 100% field capacity by weighing every day, and the lowest water content between watering events was 75–80% field capacity. The experiment was conducted in a naturally-lit glasshouse with minimum and maximum temperatures of 18 and 25°C, respectively. A completely randomised design was employed and pot positions were changed randomly twice a week to minimise any effects due to spatial variations in growing conditions within the glasshouse. Plants were harvested at anthesis, 56 days after sowing. Plant shoots were cut at the soil surface and rinsed in distilled water, then oven-dried at 70°C for 72 h for dry weight. About 1 g of ground material of shoots was digested in 0.5 M HNO₃ in a Milestone Mars-5 microwave (CEM Corp., Matthews, NC, USA) after the method of Huang et al. (2004). Plant samples were digested by hydrogen peroxide and HNO₃ mixture and then the concentrations of K, Na and other nutrients were determined by inductively coupled plasma atomic emission spectroscopy (ICP-AES) (Vista Simultaneous ICP-AES; Varian, Palo Alto, CA, USA). Soil samples from each pot were collected after harvesting plants and Colwell K was measured as mentioned above.

Statistical analyses

Statistical analysis was performed mainly using R, with SPSS ver. 24.0 software package (IBM Corp 2016) used for the Pearson correlation calculations and their significance. The weighted Deming regression was performed in R using the package SimplyAgree and using the weights described by Linnet (1990). Points were removed as needed according to the optimised residuals and the Shapiro–Wilk and Breusch–Pagan tests. The log fit was determined to be the best relationship for the different K test methods and the plant properties, based on comparison with the linear and quadratic models given the overall fit for all values and assumptions of the model (diminishing returns as the K level increases in the soil). The final linear regression was performed between NH₄OAC K and the total K uptake to determine if the linear approximation would be a better model for the linear component of the log model. This required the removal of four points that were deemed to be high leverage points using Cook’s distance (Di > 1). The stepwise regression was performed using the package leaps with forward selection, using the model selection criteria of the Bayesian Information Criterion (BIC). The linear model assumptions were checked for each model and the model comparisons were performed using ΔBIC = BICi – BICj, with ΔBIC < 2 indicating weak evidence for selecting the lower BIC model and ΔBIC > 10 indicating strong support (Burnham and Anderson 2004).

Soil properties

The 63 soil samples from 21 sites represented a wide range of soil types with different properties (Table 1). Of all soils, 43% were Kandosols (nine sites), 24% were Sodosols (five sites), 19% were Dermosols (four sites), 10% were Arenosols (two sites) and 5% were Calcarosols (one site). Soil texture ranged from sand to clay, with 70% of soils being sandy loam and sandy clay loam because clay-rich soils were targeted in this study. The samples had clay contents of range 19–450 g/kg (median 160 g/kg) and silt content range 10–210 g/kg (median 69 g/kg). The OC range was 1.4–37.4 g/kg (median 6.0 g/kg), pH (CaCl₂) was 4.1–7.9 (median 5.4), cation exchange capacity (CEC) was 0.8–42.9 cmol_c/kg (median 5.4 cmolc/kg) and EC (1:5) was 0.02–0.39 dS/m (median 0.05 dS/m).

Clay minerals

Random powder XRD and subsequent FTIR analysis revealed that quartz was the dominant mineral in all soils, comprising 21–99% (median 72%) (Supplementary Table S1, Fig. 1). The exception was the subsoil sample of soil 15, where no quartz was identified. Out of 42 samples, 35 samples contained K-feldspars (1–54%, median 7%), and plagioclase was identified in 32 samples (1–34%, median 4%). About 95% of samples contained kaolinite and the proportion range was 0.3–25% with median 4%. Montmorillonite was identified in 50% of samples and range was 0.5–35% (median 4.7%). Nontronite and white mica were identified in less than 10% samples and contributed only minor amounts (up to 2%).

Fig. 1.

Semi-quantitative mineralogy of whole soil samples according to random powder X-ray diffraction and Fourier-transform infra-red spectroscopy analysis.

Soil solution K

Water-soluble K ranged widely within 1–110 mg/kg and median of 8 mg/kg (Table S2, Fig. 2a). The CaCl₂-extractable K range was 3–431 mg/kg, with median of 28 mg/kg. Although water-soluble K concentration was highly correlated with CaCl₂-extractable K (r = 0.885, P < 0.01), CaCl₂-extractable K was consistently more highly correlated with other soil K measurements (Table S3). The amount of CaCl₂-extractable K was about 3–4 times higher than water-soluble K, and the weighted Deming regression also showed strong proportional bias, as CaCl₂-extractable K = 3.758 water-soluble K – 0.215 (r = 0.89) (Fig. 3). Overall, soil solution K had a weak but significant relationship with soil EC (r = 0.268 and 0.322, P < 0.05), and consistently decreased with soil depth with the exception of two duplex soils (soils 4 and 15), where solution K increased with soil depth (Tables 3 and S2).

Fig. 2.

Amounts of K extracted by different extractants for soil samples: (a) soil solution K, including water-extractable K and CaCl₂-extractable K, (b) soil exchangeable K, including AgTU⁺ K, Colwell K and NH₄OAc K and (c) soil non-exchangeable K, including TBK_1h, TBK_4h, Nitric K and aqua regia K.

Fig. 3.

Weighted Deming regression of water-extractable K with CaCl₂-extractable K. Soil samples in triangles were removed based on being extreme outliers from observing the optimised residuals.

Soil exchangeable K

Soil exchangeable K for AgTU⁺ K, Colwell K and NH₄OAc K had ranges of 13–909, 20–1167 and 19–1047 mg/kg, with median values of 67, 100 and 89 mg/kg, respectively (Fig. 2b). Overall, soil exchangeable K was significantly correlated with soil EC, pH, CEC and clay content (P < 0.01); weakly correlated with KBC and montmorillonite content (P < 0.05); and negatively correlated with sand content (P < 0.01) (Table 3). There were strong positive correlations (r > 0.96) among AgTU⁺ K, Colwell K and NH₄OAc K (Table S3, Fig. 4). However, as the magnitude of soil exchangeable K increased, the disparity between the amount of K extracted by AgTU⁺ compared to Colwell K and NH₄OAc K increased due to the proportional bias evident in their respective weighted Deming regressions, where AgTU⁺ K = 0.76Colwell K + 0.09 (r = 0.97), and AgTU⁺ K = 0.73NH₄OAc K + 2.98 (r = 0.96) (Table 4, Fig. 4). In contrast, there was no proportional or systematic bias between NH₄OAc K and Colwell K where NH₄OAc K = 0.98Colwell K – 2.22 (r = 0.98), except several extreme outliers that showed 25–35% higher K extracted by NH₄OAc than Colwell K in soils 5, 6 (only 20–30 cm) and 15.

Fig. 4.

Weighted Deming regressions for methods of extracting soil exchangeable K, (a) AgTU⁺ K and NH₄OAc K, (b) Colwell K and NH₄OAc K, (c) AgTU⁺ K and Colwell K. Soil samples in triangles were removed based on being extreme outliers from observing the optimised residuals.

Soil non-exchangeable K

The ranges for amounts of TBK_1h, TBK_4h and Nitric K were 22–1606, 25–1582 and 23–1860 mg/kg, with median values of 219, 268 and 216 mg/kg, respectively (Fig. 2c). Aqua regia K had significant correlations with TBK_1h, TBK_4h and Nitric K, but aqua regia K values were 1–6 times higher than TBK_1h, TBK_4h and Nitric K values (Table S3, Fig. 2c). Similarly to soil exchangeable K, non-exchangeable K was significantly correlated with soil EC, pH, CEC, clay content, KBC and montmorillonite content (P < 0.01), weakly correlated with the amount of quartz (P < 0.05) and negatively correlated with sand content (P < 0.01) (Table 3). Similarly to exchangeable K, there were strong positive correlations (r > 0.94) among TBK_1h, TBK_4h and Nitric K (Table S3). The amount of TBK_4h was on average ~20% higher than TBK_1h and Nitric K, and showed proportional bias based on the weighted Deming regression where TBK_4h = 1.24TBK_1h – 1.5 (r = 0.99) and TBK_4h = 1.22 Nitric K – 4.69 (r = 0.94) (Table 4, Fig. 5). There was no systematic difference or proportional bias between Nitric K and TBK_1h where Nitric K = 1.0TBK_1h + 2.25 (r = 0.94), except the extreme outliers (soils 5 and 20) in which Nitric K extracted 23–67% more K than TBK_1h.

Fig. 5.

Weighted Deming regressions for methods of extracting soil non-exchangeable K, (a) TBK_4h and Nitric K, (b) TBK_4h and TBK_1h, (c) Nitric K and TBK_1h. Soil samples in triangles were removed based on being extreme outliers from observing the optimised residuals.

Relationship between plant and soil parameters

Although adequate levels of all basal nutrients minus K were added to all pots, and soil water supply was maintained at 75–100% of field capacity, wheat shoot dry weight showed large variations among soil types, with range 2.2–13.3 g per pot (Table S4). Shoot weight had significant positive correlations with EC, CEC, pH and OC, clay and silt contents, but negative correlation with sand content (Table 5). Regarding clay minerals, shoot weight was weakly positively correlated with contents of montmorillonite, nontronite and white mica (r = 0.24–0.29, P < 0.05), but negatively correlated with kaolinite content (r = −0.19, P < 0.05) (Table 5). Shoot weight was significantly related to all measured soil K pools (Table 6). The change in exchangeable K (initial level of Colwell K minus level of Colwell K after harvest) showed the best relationship (R² = 0.74) with shoot weight, followed by CaCl₂-extractable K, NH₄OAc K, AgTU⁺ K and Colwell K (R² = 0.60–0.65).

Shoot K concentrations had range of 5–35 g/kg, while the total K uptake range was 10–168 mg/kg soil (Table S4). All plants were healthy and showed no nutrient deficiency symptoms during the experiment period, with the exception of wheat grown in soils with low initial NH₄OAc K (<70 mg/kg), which showed symptoms of K deficiency and contained shoot K concentration below 10 g/kg. Shoot K concentration and total K uptake had stronger correlations with EC, CEC and clay and sand contents (r > 0.56), and weaker correlations with OC and silt contents (Table 5). In addition, both were weakly positively correlated with montmorillonite and white mica contents (r = 0.19–0.32, P < 0.05), but negatively correlated with quartz content (r ~−0.21, P < 0.05) (Table 5). As with shoot weight, there were significant correlations between all measured soil K pools and shoot K concentration and total K uptake, but a logarithmic equation fitted the data better than linear and quadratic models (Table 6). Among all methods, NH₄OAc K was the best to predict shoot K concentration (R² = 0.815) and total K uptake (R² = 0.895) (Fig. 6), with Colwell K the next best predictor for both responses and showing similar model quality (5 < ΔBIC < 10). Stepwise multiple regression was conducted after log transformation, and the equations and R² for the most predictive combinations of independent variables are shown in Table S5. The results showed that the addition of KBC improved the prediction of wheat biomass using log CaCl₂ K (ΔBIC > 2), increasing R² from 0.647 to 0.670. For the prediction of shoot K concentration, the added terms of pH, sand content and EC produced a better model, increasing adjusted R² from 0.815 to 0.891, with the most important property added being soil pH. Finally, the prediction of total K uptake improved by including OC and EC, increasing adjusted R² from 0.895 to 0.924.

Fig. 6.

Best relationship between measured K in initial soils and (a) wheat shoot biomass, (b) shoot K concentration and (c) total shoot K uptake.

Notably, when NH₄OAc K exceeded 400 mg/kg, both shoot K concentration and total K uptake plateaued and did not change much with further increases in soil NH₄OAc K (Fig. 6). This then drove the selection of the linear model. After removing four points deemed to be high leverage points using Cook’s distance (Di > 1), it had a significant positive linear relationship with total K uptake (R² = 0.93) and a much smaller BIC, indicating that the linear model was better for prediction for NH₄OAc K below approximately 400 mg/kg (Fig. 7a). The slope of the regression line between NH₄OAc K and total K uptake was only 0.50, indicating that on average only 50% of NH₄OAc K was taken up from soil by wheat shoots. The amount of shoot K uptake was close to the change in exchangeable K (initial level of Colwell K minus level of Colwell K after harvest) in all soils (Fig. 7b). Since the linear relationship between soil K and plant parameters only appeared when the four high leverage points (>400 mg/kg exchangeable K) were removed, stepwise multiple regression was conducted for soils with <400 mg/kg exchangeable K where no log transformation was needed (Table 7). Similar results showed that the combination of CaCl₂ K and pH predicted 63% of the variation in shoot biomass. However, NH₄OAc K was the best indicator of shoot K concentration and the addition of clay content only increased R² from 0.84 to 0.90. Furthermore, NH₄OAc K explained about 93% of the variation of total K uptake, while addition of EC and KBC increased the R² to 0.95.

Fig. 7.

Linear relationship of total shoot K uptake with NH₄OAc K in (a) initial soils and (b) with the change in exchangeable K. Triangles in (a) show the points deemed to be high leverage points that were removed to fit a linear model for soils with <400 mg/kg initial NH₄OAc K.

Soil tests for assessing different soil K pools

Overall, the concentrations of soil solution K and exchangeable K were reasonably high, especially in surface soil (11–431 and 38–1047 mg/kg, respectively). This may be attributed to the fertiliser application or cycling of K to the surface via plant residues since the soils have been used for crop and pasture production for several decades. Normally, soil solution K concentration ranges within 1–10 mg/kg soil (Power and Prasad 1997). However, in this study 80% of surface soil and 30% of subsoil had solution K concentration above 10 mg/kg. The median and maximum values of Colwell K were 5- and 2-fold higher than for the soils collected in south-western Australia under native vegetation by Pal et al. (1999). For the sub-surface soil, the median values of Colwell K were 2–3 times higher than those in native vegetation soils, although the maximum values were similar. If we consider 45 mg/kg Colwell K as the critical value for wheat growth as suggested by Brennan and Bell (2013), only 2 of 21 surface soils were K deficient.

As expected, very low or undetectable levels of non-exchangeable K occurred in the light-textured soils. Interestingly, the median values of non-exchangeable K were 2–3 times those in native vegetation soils, while the maximum value was only 20–30% of those in native vegetation soils (Pal et al. 1999). The high non-exchangeable K in the loam- and clay-textured soils could be attributed to the high amount of feldspars, since little mica and/or illite was identified. The mineralogy analysis showed that quartz was the dominant mineral in all the collected soils, followed by K-feldspars, plagioclase and kaolinite, with small amounts of montmorillonite, nontronite and white mica. This is not unexpected as soils in south-west WA are highly weathered. Similar results were reported by Pal et al. (2002) who found that sand and silt fractions were dominated by quartz but contained significant amounts of K-feldspars, while kaolinite was the dominant mineral in the clay fraction. Although illte/mica were detected previously in some WA soils, the peak of illite/mica was small and no percentage data were provided (Pal et al. 2001a, 2001b, 2002). The mineralogical analysis of the clay fraction from Reference soils of south-western WA showed 6–20% of illite in the subsurface (15–30 or 45 cm) of soils TUT 7, GTN 6, GTN 9, BEA 2 and MER 2 (McArthur 2004). However, in this study the absence of illite/mica in the soils paired with those Reference soils (soils 1, 2, 7, 10, 12 and 14) might be for several reasons. Firstly, in this study those soils were not collected from the Reference soil sites under native vegetation, but from nearby farms adjacent to the sites, which had been used for grain or pasture production. The variation in illite content between this study and the Reference soil could potentially be attributed to the subtle differences between the soils under native vegetation and agricultural cultivation. Secondly, in this study, whole soil samples were used for mineralogy determination, rather than the clay fraction only, thus the small proportions (1–5%) of illite/mica in the whole sample might not be detectable. The reason for using bulk soil rather than the clay fraction in this study was that the soil non-exchangeable K could be present in various K-bearing minerals that exist in the whole range of particle sizes. Previous studies showed that feldspar was a significant source of K in some WA soils and extractable K was positively correlated with soil silt content, not clay content (Pal et al. 2001a, 2002). Therefore, the contribution of different minerals to soil K pools and plant uptake was evaluated by the bulk soil samples in this study.

With regard to the soil test methods for soil exchangeable K, the high correlations among AgTU⁺ K, Colwell K and NH₄OAc K confirmed that these three methods extracted K from similar K pools. However, the amount of K extracted by AgTU⁺ was consistently ~25% lower than that extracted by NH₄OAc and Colwell in all soils. Unbuffered AgTU⁺ was commonly used for rapid determination of effective CEC and exchangeable cations in soils dominated by low-activity clays with variable surface charges (Pleysier and Juo 1980). However, Searle (1986) found that the performances of AgTU⁺ and NH₄OAc were not consistent even for soils with similar mineralogy, resulting in poor correlation between AgTU⁺ K and NH₄OAc K. This discrepancy could be attributed to the absence of pH buffering in AgTU⁺, which extracts cations under variable soil solution pH conditions across different soils. In contrast, the pH of NH₄OAc and Colwell solutions were buffered at 7 and 8.5, respectively, potentially enhancing the extractability of cations. Notably, there was no significant difference between NH₄OAc K and Colwell K for most soils, consistent with the report by Zhang et al. (2017).

In this study, the amounts of K released after 1 or 4 h of incubation time for the NaTPB method were compared. Previously, the 4-h incubation time was recommended by Rayment and Lyons (2011) and Carey et al. (2011) for the routine use of the NaTPB method since it provided a better fit with operational requirements of commercial laboratories and gave slightly better repeatability and precision and correlation with the NaTPB salt precipitant method developed by Jackson (1985) than the 1-h extraction time. Our results showed that the amount of TBK_4h was ~24% higher than that of TBK_1h in all soils. It has been reported that the short reaction time is correlated with release of K from vermiculite interlayers while the longer reaction time extracted more K from primary minerals (Cox et al. 1996; Bell et al. 2021a). Thus, the amount of K extracted by NaTPB would be expected to increase with incubation time when reserves of K are mostly in fixed and/or structural forms (Moody and Bell 2006).

The strong correlation between Nitric K and TBK_1h or TBK_4h indicated that K was extracted from similar K pools. The amount of Nitric K was similar to that of TBK_1h, similar to findings for all the selected agricultural soils from south-east Queensland, Australia (Moody and Bell 2006). Compared to HNO₃ and NaTPB methods, aqua regia had much stronger extractability, evidenced by the 1–6 times higher concentrations of extractable K than TBK and Nitric K. Andrist-Rangel et al. (2010) found that aqua-regia-extractable K could be best explained by the sum of phyllosilicate K, but not K in feldspars. In this study, aqua regia K was significantly correlated with soil properties, including pH, EC, CEC and clay content. However, previous studies showed an inconsistent effect of soil texture on the extractability, since the extractability of K by aqua regia was highest for clays and lowest for sands (Andrist-Rangel et al. 2006), but no clear relationships were found in the study by Andrist-Rangel et al. (2010).

Performance of soil test to predict plant K uptake

Overall, there were significant relationships between all measured soil K parameters and shoot biomass, K concentration or total K uptake. Since plant growth may be strongly influenced by soil physical and chemical properties in addition to K level, shoot K concentration and total K uptake by plants will be discussed as a more reliable index of the efficacy of soil K tests than shoot biomass.

The correlations between K extracted by different methods and wheat K concentration and the step-wise multiple regression suggested that exchangeable K, particularly NH₄OAc K, was a satisfactory diagnostic indicator of K status for the soils in this study. The present study did not support our hypothesis that loam- and clay-textured soils would require different K test methods for K management on different soils, although the total K uptake was weakly correlated with the contents of clay, montmorillonite and white mica. This could be attributed to the relatively high initial exchangeable K in most soils but limited plant uptake in this short-term study since plants were harvested after only 56 days of growth – more time may be needed for wheat root systems to access non-exchangeable K pools. This was supported by the slope of the linear regression between NH₄OAc K and total K uptake, indicating only 50% of soil NH₄Oac K was taken up by wheat shoots within 56 days (Fig. 7a). For longer-season crops and multiple crop seasons, the release or replenishment of slowly available K from interlayer and structural pools is likely to be more important for plant uptake because of the higher and more prolonged K demand. Previous glasshouse K exhaustion experiments showed that plant-available K was well related to NH₄OAc K in soils with low non-exchangeable K contribution, and well related to TBK in a wide range of Canadian soils with high amount of non-exchangeable K (Cox et al. 1999). However, under exhaustive cut-and-removal pot trials, the slope of the regression line between soil readily available K (water-soluble K + exchangeable K) and cumulative K uptake by ryegrass increased progressively from less than 0.5 (Pal et al. 2001a) or 0.59 (Darunsontaya et al. 2012) after the first harvest to about 1 after the sixth harvest, indicating that soil water-soluble K and exchangeable K were almost the only forms of K taken by grass in the highly weathered soils from both WA and Thailand (Pal et al. 2001a; Darunsontaya et al. 2012). The reliance of plant K uptake on exchangeable K when there is a low contribution of non-exchangeable K suggests that soils formed from weathered parent materials will be vulnerable to K rundown due to negative K balances (Ma et al. 2022). Further experiments should compare the performance of NH₄OAc and TBK on the prediction of plant-available K and assess the contribution of non-exchangeable K to plant K uptake after multiple crop seasons, or exhaustive pot experiments, using soils with variable K reserves and mineralogy.