Phosphorus sorption on tropical soils with relevance to Earth system model needs

Additional keywords: adsorption isotherm, clay, dissolved organic carbon, iron oxide, phosphorus adsorption.

Tropical rainforests play a key role in the uptake of carbon dioxide (CO₂) due to their substantial productivity (Pan et al. 2011; Yang et al. 2014). Many tropical soils are highly weathered, and tropical forests are generally considered P limited, such that the extent to which tropical forests can continue to take up CO₂ in the future is critically influenced by P availability (Walker and Syers 1976; Elser et al. 2007; Pan et al. 2011; Yang et al. 2014; Wieder et al. 2015; Sun et al. 2017). However, the P cycle has only recently been incorporated into Earth system models (ESMs) (Wang et al. 2010b; Goll et al. 2012; Yang et al. 2014). Reed et al. (2015) identified critical P dynamic processes controlling soil P availability that could be better represented in ESMs, with particular emphasis on P interactions with mineral soils.

Strong adsorption of inorganic phosphate anions (PO₄³⁻) to mineral soils exerts a key control on the availability of P to plants and microbes (Lloyd et al. 2001). Phosphate anions initially attach to the highest energy binding sites and then fill lower energy sites (McGechan and Lewis 2002). As P anions accumulate at binding sites, the soil particle surfaces become increasingly negative, decreasing the affinity for P anions as P concentrations increase (Barrow 1978). The prioritisation of higher energy sites before lower energy sites, and a decreasing binding affinity with increasing concentrations results in a nonlinear isotherm where P removal is very strong at low concentrations (Barrow 1978; McGechan and Lewis 2002). Traditionally, the sorption properties of soils have been estimated through two parameters in the Langmuir equation, the maximum sorption capacity (Qmax) and the binding energy coefficient (k) (Barrow 1978; Anghinoni et al. 1996). At least three current ESMs include a P cycle, and all assume sorption capacity is higher in weathered and reactive soils common to the tropics, e.g. Oxisols and Andisols (Wang et al. 2007; Goll et al. 2012; Yang et al. 2014; Yang et al. 2016). However, there is a lack of P sorption studies related to tropical soils (Fontes and Weed 1996; Sato and Comerford 2005; Munhoz et al. 2011) compared with other environments (Anghinoni et al. 1996; Gichangi et al. 2008; Cannon 2010; Bastounopoulou et al. 2011).

A wide spectrum of Qmax and k values in P isotherm studies have been reported for both temperate and tropical soils. For temperate soils, reported Qmax values for North American Ultisols ranged from a mean value of 243 mg P kg⁻¹ (Cannon 2010) to a mean value of 2222 mg P kg⁻¹ (Anghinoni et al. 1996) (Table S1, available as Supplementary Material to this paper). The same variability among and within soil orders exists for tropical soils. Phosphorus Qmax values ranged from 305 to 4487 mg P kg⁻¹ in Brazilian Oxisols (de Campos et al. 2016), and from 3237 mg P kg⁻¹ in an Ultisol to 78616 mg P kg⁻¹ in an Inceptisol (Poudel and West 1999). The specific causes of wide variability have yet to be pin-pointed, but there are three sparsely discussed but critical aspects of P sorption. First, Qmax can be underestimated due to low initial P concentrations added to the soils (Harter 1984). Tropical soils are particularly prone to this potential source of error due to their high P sorption capacities (Walker and Syers 1976; Fontes and Weed 1996; Poudel and West 1999; Elser et al. 2007; Auxtero et al. 2008; Yang et al. 2014; de Campos et al. 2016). Further, there is an absence of reported parameter uncertainties in the current literature, such as standard error or confidence intervals (Ryden et al. 1977; Fox 1982; Anghinoni et al. 1996; Fontes and Weed 1996; Gichangi et al. 2008; Bastounopoulou et al. 2011; Munhoz et al. 2011). Without these uncertainty values, it is difficult to compare sorption parameters from different studies in the same context. Finally, sorption isotherms are often fitted to the Hanes–Woolf linearised form of the Langmuir equation (Ryden et al. 1977; Fox 1982; Anghinoni et al. 1996; Fontes and Weed 1996; Gichangi et al. 2008; Bastounopoulou et al. 2011; Munhoz et al. 2011). The axes of a linearised isotherm are not truly independent of one another because the sorbed concentration (y-axis) is calculated from the equilibrium solute concentration (x-axis). A nonlinear fit is now recommended to avoid an overestimation of correlation between axes (Bolster and Hornberger 2007).

ESMs could readily exploit quantitative relationships between more widely measured soil physicochemical properties and phosphate sorption, if such relationships were available. Sorption of P in tropical soils has been shown to have a strong association with iron (Fe) and aluminium (Al) oxides (Fontes and Weed 1996), which is expected given that the content of Fe- and Al-oxide minerals and the buffering capacity of acidic soils are often directly proportional (McGechan and Lewis 2002). The small particle sizes and high specific surface areas of clays can enhance binding of P (McGechan and Lewis 2002). Soil pH may also impact soil P sorption parameters, as a lower pH results in a more positive charge on hydrous oxides, thus attracting P anions (Sparks 1995). The role of physical characteristics in P sorption is complex, but quantitative relationships between soil properties and sorption can be obtained (Kothawala et al. 2009; Mayes et al. 2012).

The goal of this work is to provide improved quantitative information on P sorption in tropical soils that can be used to improve parameterisation in ESMs. Our specific objectives were to generate P sorption isotherms and Langmuir parameters Qmax and k of A and B horizon soils from locations across the tropical climate zone, and to determine whether any soil characteristics significantly contributed to the variation of observed soil P sorption parameters.

Twenty-three tropical soils and one sub-tropical soil from both A and B horizons were selected for analysis from an archive of air-dried and sieved (2 mm) soils at Oak Ridge National Laboratory (ORNL) to provide a broad representation of tropical regions (Table 1). Five soil orders (Ultisols, Aridisols, Oxisols, Andisols, and Inceptisols) were obtained from the following sources (Ryan et al. 2004; Bruno et al. 2006; Porder et al. 2006; Russell et al. 2007; Wang et al. 2010a; Vieira et al. 2011; Yavitt et al. 2011; Mage and Porder 2013; Jagadamma et al. 2014; Mayor et al. 2014). Further information on soil characteristics (e.g. soil origin, climate, and parent material) and reference publications can be found in Table S2 (Supplementary Material).

Table 1.  Subsoils used in this study
Additional information on soil origin can be found in Table S2 (Supplementary Material)

Samples were air-dried at room temperature for 48 h, ground with mortar and pestle, and passed through a 2-mm sieve to remove roots and small rocks. Soil pH was measured in a 0.005 M CaCl₂ solution at a 2 : 1 solution to solid ratio. Total Fe and Al oxides were extracted and determined by the dithionate-citrate-bicarbonate (DCB) method (Loeppert and Inskeep 1996) and analysed by inductively coupled plasma mass spectrometry (ELAN-1600, Perkin-Elmer Corporation, Waltham, MA, USA). Particle size was ascertained using the Buoycous hydrometer method (Gee and Or 2002). For carbon (C), nitrogen (N), and total organic carbon (TOC) content, soil samples were oven-dried overnight at 70°C, ground to a fine powder, and analysed on a LECO TruSpec elemental analyser (LECO Corporation, St. Joseph, MI, USA). The TOC was determined by removing inorganic C by treating the soil with 3 M HCl for 1 h and rinsing 3 times. Total P was prepared using a Lachat digestion block and measured by a Lachat in-line sample preparation and determination instrument (Hach, Loveland, CO, USA) following the Kjeldahl method 13–115–01–1-B (Lachat Instruments, Milwaukee, WI, USA). All concentrations are expressed as a function of soil dry weight.

Soils (3 g) were added to 50 mL glass vials along with a 30 mL aliquot of different KH₂PO₄ concentrations, ranging from 10 to 500 mg P L⁻¹, in duplicate and in a 0.01 M CaCl₂ matrix (Graetz and Nair 2000). Blank soil samples, conducted in triplicate, were exposed only to 0.01 M CaCl₂. Separate glass vials without soil received 30 mL of each concentration of KH₂PO₄ in 0.01 M CaCl₂ to determine the concentration of added P. Two drops of toluene were added to each vial to inhibit microbial uptake (Gichangi et al. 2008). Vials were capped and placed on a shaker at room temperature for 24 h at 85 oscillations per minute, centrifuged at 453×g for 10 min, filtered using a Millex 0.45-µm filter, and acidified before storage and analysis (Gichangi et al. 2008; Bastounopoulou et al. 2011). Phosphorus remaining in the supernatant was measured by the molybdenum blue method (Murphy and Riley 1962; Drummond and Maher 1995), utilising the Lachat as described above. Supernatant P, and resulting P sorption maximum values, were reported as mg of orthophosphate as phosphorus, or mg PO₄-P, per kg dry weight of soil. Dissolved P in the samples was confirmed to be in the form of PO₄⁻³ via ion chromatography (Dionex ICS-2100, Sunnyvale, CA, USA).

Dissolved organic carbon (DOC) isotherms were determined using a C stock acquired from a stream in Georgetown, SC (Gu et al. 1994; Chen et al. 2002). A 30-mL aliquot of DOC concentrations, ranging from 2.5 to 100 mg C L⁻¹ in 0.01 M NaCl, was added to glass vials containing 0.5 g of the <2 mm fraction soil, in duplicate (Mayes et al. 2012). Blank soil samples, conducted in triplicate, were exposed only to 0.01 M NaCl. Vials without soil also received 30 mL of each DOC concentration to determine initial DOC added. Vials were shaken and centrifuged as described above. The supernatant was analysed using a Shimadzu TOC-L CSH/CSN analyser (Shimadzu Corp., Kyoto, Japan).

Phosphorus and DOC sorbed concentrations were calculated using the Fox and Kamprath (1970) procedure, where the amount sorbed was calculated as the difference between the concentration added and the concentration remaining in the supernatant. Sorption data were fitted to the non-linear form of the Langmuir equation (Eqn 1) from Mayes et al. (2012) to provide a curve and sorption parameters, utilising a method developed by Bolster and Hornberger (2007), which also calculated the standard error (s.e.) for each parameter, and the goodness-of-fit for the entire equation. In this case, goodness-of-fit measured model efficiency (E), where E = 1 (or 100%) indicated that the model perfectly fit the data. Equations used to calculate s.e. and E can be found in Bolster and Hornberger (2007).

where RE is the amount of P or DOC adsorbed or desorbed (mg kg⁻¹ soil), X_f is the final measured equilibrium solution concentration (mg L⁻¹ solution), Qmax is the maximum sorption capacity (mg kg⁻¹ soil), and k is the binding affinity coefficient (L solution mg⁻¹).

Data reported as percentage values were square-root transformed to account for constrained data and conformed to parametric data assumptions for analysis. Data were analysed in JMP 13.2.1 (SAS Institute Inc., Cary, NC, USA). For normally distributed data, analysis of variance (ANOVA) was applied to assess variance by soil order, where P < 0.05 indicated significant differences. A post-hoc Tukey’s test was used to determine where those differences occurred. Variance in non-normal data was determined by a Wilcoxon rank test, where P < 0.05 indicated significant difference. A post-hoc Dunn’s multiple comparison test for pairs was performed to determine where those differences occurred. A Spearman rank test was performed to assess the presence of any statistically significant correlations between Langmuir model sorption parameters (Qmax and k) and soil characteristics. Correlations were considered statistically significant for correlation coefficients that had a P < 0.05, and where significant, the strength of correlation was shown by the Spearman correlation coefficient (ρ).

By fitting the isotherms to the Langmuir equation, the resultant Qmax values ranged from 733.8 mg PO₄-P kg⁻¹ (Kenya Aridisol, KA-A) to 3775 mg PO₄-P kg⁻¹ (Costa Rica Oxisol, CRO-A) (Table 2), and the mean Qmax value for all soils was 2060 mg PO₄-P kg⁻¹ (standard deviation (s.d.) = 767.7). Hawaii Andisols did not reach a sorption maximum over the range of initial P concentrations used in this study, so their isotherm was not represented using the Langmuir equation. To assess correlation and variance among sorptive and physical soil characteristics, soils were pooled together for correlation analysis, and by soil order for variance. Due to low sample size, only Inceptisol, Oxisol, and Ultisol soils were included in the variance analysis. Despite the broad range of values, P Qmax values did not display any significant difference among soil orders (ANOVA, P = 0.07) (Fig. 1).

Table 2.  Phosphorus sorption isotherm parameters
Phosphorus sorption isotherm maximum (Qmax), binding affinity coefficient (k), respective standard errors (s.e.), and goodness-of-fit (E) values estimated by fitting the isotherms to the Langmuir equation

Fig. 1.  Scatter plots depicting the shape of phosphorus (P) sorption isotherms for (a) Inceptisols, (b) Oxisols, and (c) Ultisols. Legends indicate soil identifiers listed in Table 1.

KA-A and CRO-A also represented the lowest and highest Qmax values for DOC sorption, establishing a DOC Qmax range of 310.0–6212 mg C kg⁻¹ and a mean of 1876 mg C kg⁻¹ (Table 3). Although an initial Wilcoxon test indicated variance (Wilcoxon, P = 0.02), a post-hoc Dunn’s comparison test exhibited no significant differences among DOC Qmax values by soil order (Dunn’s, P > 0.05).

Table 3.  Dissolved organic carbon (DOC) sorption isotherm parameters
DOC sorption isotherm maximum (Qmax), binding affinity coefficient (k), respective standard errors (s.e.), and goodness-of-fit (E) values estimated by fitting the isotherms to the Langmuir equation

The sorption maxima for P and DOC were significantly positively correlated with each other (Fig. 2). There were no significant relationships among the binding coefficients (Table 4).

Fig. 2.  Linear regression displaying the relationship between Langmuir-derived sorption maxima (Qmax) of dissolved organic carbon (DOC) and phosphorus (P).

Table 4.  Spearman correlation coefficients (ρ) among Langmuir-derived sorption parameters
Phosphorus (P) and dissolved organic carbon (DOC) sorption isotherm maxima (P and DOC Qmax respectively), and P and DOC binding affinity coefficient (P and DOC k respectively) estimated by the Langmuir equation. * P < 0.05

Clay content ranged from 2.25% (Hawaii Andisol, HA-FB) to 67.7% (Costa Rica Oxisol, CRO-B) (Table 5). An initial test of variance indicated that clay content varied by soil order (Wilcoxon, P = 0.02), though a post-hoc Dunn’s comparison did not illustrate significant difference (Dunn’s, P > 0.05). Andisols were primarily composed of sand-sized particles (mean = 65.1%, s.d. = 5.60%), and Oxisols had significantly less sand than both Inceptisols and Ultisols (Tukey, P < 0.05).

Table 5.  Selected properties of analysed soils
Carbon and nitrogen content (C, N), total organic carbon (TOC), total phosphorus (P), dithionate-citrate-bicarbonate extracted aluminium- and iron-oxides (Al-oxides and Fe-oxides respectively), pH in CaCl₂. Dashes indicate no data

To determine correlative relationships between Qmax and reactive soil properties, the pooled soils were regressed against particle size, pH, Al- and Fe-oxide content, P, C, and TOC (Kothawala et al. 2009; Mayes et al. 2012). Phosphorus sorption maxima exhibited a significant positive correlation with clay content, and a negative correlation with sand content (Table 6, Fig. 3), but no correlations with particle size were found for DOC sorption maxima (Table 6).

Table 6.  Spearman correlation coefficients (ρ) for relations among soil properties and phosphorus (P) and dissolved organic carbon (DOC) sorption maxima (Qmax)
* P < 0.05

Fig. 3.  Linear regression displaying the relationship between Langmuir-derived phosphorus sorption maxima (P Qmax) and (a) clay content, and (b) sand content.

Although Al- and Fe-oxide concentrations did not exhibit significant differences by soil order (ANOVA, P = 0.19 and 0.33 respectively), both covered a wide range of values, with Andisols representing the highest observed Al- and Fe-oxide concentrations. The greatest Al-oxide concentration (9.93 mg g⁻¹, Hawaii Andisol, HA-FA) was greater than the lowest observed concentration (0.35 mg g⁻¹, Puerto Rico Ultisol, IC-V) by a factor of 28, and the greatest Fe-oxide concentration (90.5 mg g⁻¹, Hawaii Andisol, HA-FB) was greater than the lowest observed concentration (1.55 mg g⁻¹, Brazil Ultisol, BU-BA) by a factor of 58 (Table 5). Total P, Al- and Fe-oxides were all positively correlated with P Qmax values (Table 6, Fig. 4), whereas Fe-oxides were the only physical characteristic correlated with DOC Qmax values (Table 6).

Fig. 4.  Linear regression displaying the relationship between Langmuir-derived phosphorus sorption maxima (P Qmax) and (a) aluminium (Al) oxide content, and (b) iron (Fe) oxide content.

The C content ranged from 0.65% (Brazil Ultisol, BU-BB) to 10.5% (Hawaii Andisol, HA-FB) (Table 5). The TOC was lowest in the Kenya Aridisol (KA-A, 50.8 mg kg⁻¹), and greatest in the Panama Oxisol P-23 (465 mg kg⁻¹) (Table 5). Neither C content nor TOC displayed variance by soil order (ANOVA, P = 0.75; Wilcoxon, P = 0.45 respectively). Total P, however, was sensitive to soil order (Tukey, P = 0.02); Oxisols in our database (mean = 612 mg P kg⁻¹, s.d. = 219, n = 10) had significantly more total P than Ultisols (mean = 295 mg P kg⁻¹, s.d. = 193, n = 5) (Table 5). No correlations were observed between sorption maxima and C, TOC, or P in the pooled soils (Table 7).

Table 7.  Spearman correlation coefficients (ρ) for carbon content (C), total organic C (TOC), total phosphorus (P), and aluminium (Al) and iron (Fe) oxides
* P < 0.05

The soils were acidic, with a mean pH of 4.70 (s.d. = 0.73) (Table 5), which was not sensitive to soil order, though Hawaii Andisols exhibited the highest observed pH values (5.90 and 5.84, HA-FA and HA-FB respectively), and Costa Rica and China Ultisols the lowest (3.58 and 3.66, CRU-V and CU-HSA respectively). No correlations were observed between sorption maxima and pH in the pooled soils (Table 6).

Some correlations between physical properties were observed. The C content was positively correlated with TOC, Al-oxides, and Fe-oxides, but not with total P, although total P was correlated with TOC (Table 7). Total P, Al- and Fe-oxides were also all positively correlated with one another (Table 7).

The Langmuir isotherm fits provided adequate representations of the P and DOC isotherms, with acceptable s.e. and E values. Both DOC and P isotherms were also fit to the Freundlich equation because it has been shown to provide a better fit than Langmuir for P (Bolster and Hornberger 2007); however, our Langmuir fits were superior with lower s.e. and higher E values (Brenner et al. 2018). Consequently, only Langmuir fits are discussed here. Despite the wide range of P Qmax values in this study, there was no significant difference among soil orders. Studies have shown that highly weathered soils (i.e. Oxisols and Ultisols) have greater P sorption capacities than less weathered soils (i.e. Inceptisols) (Oades et al. 1989; Yang and Post 2011), and soils with higher amounts of variable charge minerals (i.e. Oxisols and Andisols) also typically have greater P sorption capacities (Oades et al. 1989). Although the absence of significantly different P sorption maxima observed here was unexpected based on the existing literature, it is likely due to the small sample size of the individual soil orders studied, and the relatively high variation in sorption values, particularly the Ultisols (Fig. 1c).

Phosphorus sorption maxima were significantly correlated with both Al- and Fe-oxides, which is in agreement with others that have assessed sorption parameter relationships (Lopez-Hernandez and Burnham 1974; Sakadevan and Bavor 1998; Börling et al. 2001; McGechan and Lewis 2002). However, when graphing the relationship between P sorption maxima and either Al- or Fe-oxides, only 12% and 7.6% of the variation in P sorption maxima could be explained by the resulting linear regression equation respectively (Fig. 4). Gichangi et al. (2008) described the relationship between P Qmax values and DCB-extracted Al- and Fe-oxides with a quadratic function, explaining 89% and 70% of variation in P Qmax values respectively, indicating that these parameters provide a reliable estimation for P sorption. A quadratic function minimally improved the R² values for describing the relationship between P sorption maxima and either Al- or Fe-oxides (23% and 11% respectively), but since a quadratic fit is unsupported by a mechanistic rationale, the fits are not shown. The low R² in our linear regressions are likely a result of the small overall sample size and global distribution of soil samples; a larger sample size could potentially provide a statistically reliable equation for estimating P Qmax from DCB-extracted Al- and Fe-oxide concentrations.

Although some studies have found that Al plays a more significant role than Fe in P sorption mechanics for both temperate soils (Lopez-Hernandez and Burnham 1974; Sakadevan and Bavor 1998; Börling et al. 2001) and tropical soils (Fontes and Weed 1996; Poudel and West 1999; Hartono et al. 2005; Gichangi et al. 2008; Alleoni et al. 2012; Chimdi et al. 2013; de Campos et al. 2016), our study found that DCB-extracted Al- and Fe-oxides were similarly correlated with the P Qmax values (ρ = 0.470 and 0.461 respectively). Considering the low R² values for these parameters in describing P Qmax, this lack of discrepancy in parameter influence is not surprising. Nonetheless, our data did indicate that particle size had a stronger influence on P sorption maxima than Al- or Fe-oxides. Clay was strongly correlated with P Qmax values, where 56% of the variation could be explained by the corresponding linear regression equation (Fig. 3). The high weathering rate of tropical soils leads to an abundance of clay minerals, providing a high surface area and numerous binding sites (McGechan and Lewis 2002). For ESMs, this provides evidence that information about clay content could significantly aid in the estimation of P sorption parameters for tropical soils (Lopez-Hernandez and Burnham 1974; McGroddy et al. 2008; Bruland and DeMent 2009; Moazed et al. 2010; Alleoni et al. 2012). Although P sorption isotherms are not often performed in routine soil analyses, particle size is a frequently measured parameter. Therefore, ESMs could derive sorption parameters via measured clay content, or even estimated clay content based on soil weathering rates, if a solid correlative relationship was available. Other researchers who have assessed relationships with physicochemical parameters disagree about which parameter has the greatest influence. Lopez-Hernandez and Burnham (1974) found that Al- or Fe-oxides had significantly more influence on P sorption than clay content in tropical Oxisols and Ultisols. Hartono et al. (2005) found that Al- and Fe-oxides and clay content were similarly influential in tropical Ultisols, as did de Campos et al. (2016) in a study that analysed tropical Alfisols, Entisols, Inceptisols, Histosols, Mollisols, Oxisols, and Ultisols.

The varying influence of soil physical and chemical parameters on P Qmax values, as well as the variance of P Qmax values themselves, may be further influenced by the soil order or parent material. This was illustrated in particular by the Hawaii Andisols, which never reached a sorption maximum. The Andisols were primarily composed of sand (Table 5), even though sand was negatively correlated with P Qmax values throughout the dataset (Table 6, Fig. 3). The isotherm was approximately linear, indicating strong sorption, which is consistent with documented high sorption capacities of Andisols (Dahlgren et al. 2004; Oades et al. 1989; Fox and Searle 1978; Parfitt 1978). Therefore, it appears that the influence of Al- and Fe-oxides outweighed the influence of particle size among Andisols (Liptzin and Silver 2015; Dahlgren et al. 2004; McGechan and Lewis 2002; Anghinoni et al. 1996; Mehadi and Taylor 1988; Parfitt 1978).

The relationship between P and DOC sorption maxima, though positively correlated, does not tell a straightforward story. Dissolved organic matter and humic acids can compete with P for sorption sites, which can lead to the release of sorbed P, or P can be incorporated into the sorbed complex (Tiessen et al. 1994; Lloyd et al. 2001; Frizano et al. 2003; Guppy et al. 2005). The correlation between DOC and P Qmax values suggests that similar binding sites could be involved; however, the vast difference between DOC and P Qmax values for different soils (ranging from 41.1 mg kg⁻¹ to 2519 mg kg⁻¹) confirms a more complex relationship. This is further complicated by the different correlations between the Qmax parameters and other soil properties. Although P Qmax values were significantly correlated with sand content, clay content, and Al- and Fe-oxides, DOC Qmax values were only significantly correlated with Fe-oxides. However, because our DOC and P sorption maxima were measured separately, the extent of competitive interactions is not known.

Our Langmuir-calculated P sorption maxima exceeded many of the Qmax values used by Yang et al. (2014) in their ESM (Anghinoni et al. 1996; Gichangi et al. 2008; Bastounopoulou et al. 2011; Munhoz et al. 2011; Wang and Liang 2014). These studies also used notably different maximum P concentrations to calculate the P sorption isotherm, which could lead to an underestimation of P Qmax values (Harter 1984). However temperate soils have been shown to have smaller P sorption maxima (Anghinoni et al. 1996; Cannon 2010; Bastounopoulou et al. 2011; Yang and Post 2011), so a smaller range of initial P concentrations might be appropriate for the soils in those studies. For example, an initial maximum P concentration of 50 mg P L⁻¹ was used for temperate Appalachian soils (Anghinoni et al. 1996). However, because tropical soils are high in oxides and clay-sized particles and tend to have higher P sorption capacities (Walker and Syers 1976; Fontes and Weed 1996; Poudel and West 1999; Elser et al. 2007; Auxtero et al. 2008; Yang et al. 2014; de Campos et al. 2016), higher initial P concentrations may be required to reach the plateau of the isotherm and appropriately fit the Langmuir Qmax parameter. Despite the need for high initial P concentrations, many studies with tropical soils used maximum P concentrations as low as 100 mg P L⁻¹ for soils in Indonesia (Hartono et al. 2005). In our study, the maximum initial P concentration, and resulting Qmax values, was five times as high as concentrations in Hartono et al. (2005). A study by Tiessen et al. (1994) that generated P sorption isotherms for volcanic Oxisols observed Qmax values ranging from 10 to 360 mg PO₄-P kg⁻¹, which were notably lower than those observed in our study (Table 2). This discrepancy is most likely due to the low maximum concentration of P added to create their isotherm: 50 mg P L⁻¹, the same maximum concentration used for temperate soils in the study by Anghinoni et al. (1996). In some cases, we found no evidence of P remaining in the supernatant in our study, even with an initial concentration as high as 100 mg P L⁻¹, indicating that the range of initial P concentrations for tropical soils should probably always exceed 100 mg P L⁻¹. An example of our P sorption isotherms can be seen in Fig. 5, where the Qmax and standard error are illustrated for a Puerto Rico Oxisol (EV-V). For this soil, P in solution was below the limit of detection for initial P concentrations ranging from 5 to 50 mg P L⁻¹. Once enough P had been added to the soils to allow P to be detectable in the supernatant, initial P concentrations of 100 and 150 mg P L⁻¹ still only contributed to the linear portion of the isotherm; it was not until an initial P input of 200 mg P L⁻¹ that the isotherm began to curve, therefore confidently establishing a plateau and enabling fitting of Qmax.

Fig. 5.  Langmuir-derived P sorption isotherm for one of the soils analysed in this study, a Puerto Rico Oxisol (EV-V). P sorption maximum capacity (Qmax) is indicated by a solid grey line, with the standard error confidence interval depicted by dotted grey lines. Initial P concentrations (mg P L⁻¹) for resulting data points are labelled alongside markers.

Studies that used a maximum P concentration comparable to or exceeding our own also found higher P Qmax values. Alleoni et al. (2012) used an initial maximum P concentration of 1200 mg P L⁻¹ for P sorption isotherms in Brazilian Oxisols, and their resulting P Qmax ranged from 2400 to 2600 mg PO₄-P kg⁻¹. Poudel and West (1999) analysed tropical volcanic Oxisols, Ultisols, and Inceptisols from the Philippines using initial maximum P concentration of 1000 mg P L⁻¹ and found a mean P Qmax of 12 856 mg PO₄-P kg⁻¹. A Spearman rank test on the surveyed literature (Table S1) further cemented the influence that initial P concentration has on final P sorption maxima (ρ = 0.687, P < 0.0001), indicating that Qmax values increase when a higher maximum initial P concentration is used. The need for a greater range of initial P concentrations was also illustrated by our inability to produce a P sorption maximum for the Hawaii Andisols. Their high sorption capacity supports the need to carefully estimate sorption maxima in soils of volcanic origin and in soils with notable concentrations of clays and Al- and Fe-oxides.

As ESMs begin to incorporate the P cycle (Wang et al. 2010b; Goll et al. 2012; Yang et al. 2014; Reed et al. 2015), it is important to provide more thorough information on soils in tropical regions, where soil P is low and forest productivity is generally considered high. The studies used by Yang et al. (2014) to estimate P sorption maxima (Anghinoni et al. 1996; Gichangi et al. 2008; Bastounopoulou et al. 2011; Munhoz et al. 2011; Wang and Liang 2014) represent P sorption parameters of non-tropical soils based on isotherms with lower maximum P concentrations, and therefore lower Qmax values may have underrepresented P sorption maxima for tropical ecosystems. Better estimates of P sorption capacity can help to predict plant productivity in highly weathered soils of tropical regions, thus increasing model efficacy. Furthermore, a thorough analysis of the relationships among Qmax, k, particle size, and Al- and Fe-oxides across a wide range of tropical soils could offer stronger linear regressions for estimating sorption parameters, thereby reducing the need for niche P sorption isotherm studies. ESMs could instead obtain sufficient information from more readily available studies that include particle size and Al- and Fe-oxide concentrations.

The Langmuir equation provided a suitable description of P and DOC sorption properties of the tropical soils. Sorption isotherms are infrequently utilized in soil characterization analyses, but this study demonstrated that clay content could be a significant predictor of P Qmax values, offering a strong potential for ESMs to estimate P sorption parameters with more widely available data. To expand the efficacy of this proposed method, it would be useful to analyse these correlations among larger sample sizes from different soil orders, so that the differing influence of soil order and parent material could be better accounted for in ESMs. Our results also suggested that the initial maximum P concentration used in generating isotherms is critical for appropriately estimating P Qmax values. Previous studies may have underestimated P sorption maxima due to insufficient initial P input concentrations. The high sorption capacity of weathered, tropical soils necessitates a maximum concentration that properly approximates the sorption plateau of the isotherm.

The authors declare no conflicts of interest.