Ammonia volatilisation losses from urea applied to acidic cropping soils is regulated by pH buffering capacity

Study site and sample collection

Soil samples (0–50 mm) were collected between December 2018 and April 2019 from four long-term cropping research sites in southwestern Australia (Table 1), established to investigate the effects of different land management practices on soil properties and crop growth (Fisk et al. 2015; Betti et al. 2017; Sarker et al. 2018; Azam and Gazey 2021). Land management practices included liming, stubble management (crop residue retained or burnt), organic matter addition and clay incorporation (Table 1), with three to four replicate field plots per treatment. A randomised block design was used at each study location. Soil sampling for this study was restricted to the surface from each plot, as the NH₃–NH₄⁺ equilibrium reaction following hydrolysis of surface applied urea is a localised reaction occurring within a 30–45 mm radius from the urea granule (Black et al. 1987; Janke et al. 2021).

Experimental design and approach

We compared the different soils to assess how different soil parameters [clay content, organic carbon, cation exchange capacity (CEC), initial soil pH and pHBC] influence NH₃ volatilisation loss potential. The laboratory experiment included soils collected from the four long-term field experiments described above. Each field site included two to four treatments depending on location (Table 1), resulting in 12 field treatments with three to four replicates per treatment, totalling 43 soil samples. Urea prills (1–2 mm diameter) were surface applied to each soil sample at two rates: 0 (control) or 50 kg N ha⁻¹. In southwestern Australia, 50 kg N ha⁻¹ is a typical N fertiliser rate for cereal and oilseed crops (Harries et al. 2021).

Measuring NH₃ volatilisation

To measure NH₃ volatilisation, 150 g of soil (<2 mm; equivalent to ~20 mm depth) was weighed into a 1-L plastic container (90 mm diameter, 100 mm height) before adding deionised water to wet the soil to 60% of maximum water holding capacity (WHC) (Table 2). This soil water content was expected to be sufficient for the urea to undergo hydrolysis without forming a film of water on the soil surface that would impede gas exchange between the soil surface and container headspace. The soil was lightly compacted to form a flat soil surface after being evenly wetted and left to equilibrate for 8 h at 18°C. Urea prills [50 kg N ha⁻¹, ~3 prills (1–2 mm diameter) per treatment] were then evenly applied to the soil surface before placing a phosphoric acid trap (10 mL of 9.8% v/v H₃PO₄ in a 30-mL beaker) within the plastic container and sealing with an airtight lid (Fig. 1). The phosphoric acid trap was secured 60 mm above the soil surface using a wire scaffold to make available the entire soil surface area of the incubation container. The incubation was conducted in a controlled temperature room at 24°C for 3 weeks, with the phosphoric acid trap changed every 7 days. No further water was applied to the soil surface, simulating the surface broadcast application of urea to moist soils in the absence of rainfall events.

Fig. 1.

Incubation apparatus (sealed 1-L container) for capturing volatilised NH₃ using a suspended 10-mL phosphoric acid trap (9.8 v/v %). Incubations were conducted in a controlled temperature room at 24°C.

Volatilisation of NH₃ was measured in the laboratory using phosphoric acid to ‘trap’ the NH₃ volatilised from the urea applied to the soil surface. The apparatus used was based on Sunderlage and Cook (2018) (Fig. 1), but with the soil surface area increased from 1810 mm² to 6362 mm² and an acid solution (10 mL of 9.8% v/v phosphoric acid) instead of an acid-soaked cotton pad. The NH₃ in the container headspace has a high affinity for the phosphoric acid trap, where it is reduced to NH₄⁺ in solution. The stoichiometric capacity of the phosphoric acid solution was sufficient to ensure that if all the applied urea-N volatilised within the closure period the acid trap would not become saturated from captured volatilised NH₃ in the headspace (Ndegwa et al. 2009). Colourimetric analysis using a flow injection analyser (Lachat Instruments, Loveday, CO, USA; Method 10-107-06-2-O: Ammonia in surface water and wastewater) determined the amount of NH₃ captured by the phosphoric acid solution. The NH₃ volatilisation loss was calculated as the amount of N trapped in the phosphoric acid solution, expressed as a percentage of the applied N.

Soil physical and chemical analysis

Soil properties were determined using air-dried soil samples (<2 mm) stored at room temperature. The gravimetric water content of air-dried soils was calculated as the mass difference between air-dried soil samples before and after oven drying at 105°C for 24 h. Soil pH(CaCl₂) was measured in a 1:5 soil and 0.01 M CaCl₂ suspension using a glass electrode (Rayment and Lyons 2010). Soil WHC (g H₂O g⁻¹ soil) was calculated as the difference in mass of soils at saturation [determined using a porous ceramic plate technique (Cresswell and Hamilton 2002)] and after oven drying at 105°C for 24 h.

Clay percentage (particle size < 2 μm) was measured by a direct sampling pipette method using Stokes Law that determines clay fraction settling times at a given depth and temperature (Cresswell and Hamilton 2002). Pre-treatments included (1) digestion with hydrogen peroxide to remove organic carbon, (2) chemical dispersion using sodium hexametaphosphate and (3) 10 min of vigorous agitation using a milkshake blender (Klute 1986).

The CEC was measured using a 1:10 extraction solution of 0.1 M ammonium chloride and 0.1 M barium chloride to displace exchangeable cations (Ca²⁺, Mg²⁺, Na⁺, K⁺ and Al³⁺). The cations were measured using inductively coupled plasma spectroscopy (Rayment and Lyons 2010). The low electrical conductivity readings for the samples meant that a pre-wash for soluble salts was not needed. The sum of exchangeable cations [cmol(+) kg⁻¹ soil] (Ca²⁺, Mg²⁺, Na⁺, K⁺ and Al³⁺) was the effective soil CEC (Rayment and Lyons 2010). Total organic carbon was measured using the wet oxidation dichromate method (Rayment and Lyons 2010).

Soil pHBC was calculated using a pedotransfer function (Eqn 2) (Moore 2001) with coefficients calibrated for soils of the southwestern Australian agricultural region (R² = 0.86) as follows:

where OC% is the Walkley–Black percentage organic carbon content, clay% is the soil clay fraction (particle size < 2 μm) and Al is exchangeable aluminium (meq 100 g⁻¹).

Statistical analysis

All data were statistically analysed using the R software program (ver. 4.1.0 GUI). Simple linear regressions of NH₃ volatilisation losses versus soil properties were formed to determine suitable function relationships. Pearson’s correlation was used to analyse the relationship between soil properties and NH₃ volatilisation potential. ANOVA and Tukey’s post hoc test compared NH₃ volatilisation losses between each long-term research cropping site. Principal components analysis biplot based on Euclidian distance dissimilarity measures determined the presence of multicollinearity between the soil properties and identified suitable explanatory variables for the multiple regression model to predict NH₃-N volatilisation loss potential. A multiple log-linear regression was used to model the relationship between NH₃ volatilisation and measured soil properties, with the response variable (NH₃-N loss as % of N applied) transformed using a natural logarithm. A Breusch–Pagan test (P > 0.05) confirmed the assumption of equal variance of residuals for all transformed relations, and a Ramsey regression equation specification error test confirmed the correctness of functional form for the multiple log-linear regression (P < 0.05). A root mean square error (RMSE) analysis compared the predicted NH₃ volatilisation potentials using the Fillery and Khimashia (2016) model with the measured NH₃ volatilisation losses from the incubation experiment.

Soil physical and chemical properties

The four long-term research cropping sites provided a wide distribution of soil properties influencing soil NH₃ volatilisation potentials (Table 2). Range of organic carbon content was 0.58–1.31%, CEC was 2.38–10.15 cmol(+) kg⁻¹ soil, clay content was 1.99–19.48% and pHBC was 0.70–1.11 cmol H⁺ kg⁻¹ soil per pH unit.

A biplot cluster analysis represents differences in soil physical and chemical properties at each field site, where Euclidian distances between sample site points indicate sample dissimilarity – the further the distance between points, the more dissimilar (Fig. 2). Badgingarra and Merredin B sites had exclusive clustering in the biplot distribution, with dissimilar soil parameters to the other two sites. The Badgingarra site had the lowest soil pHBC, clay content, CEC and organic carbon content of all sampling sites, while Merredin B had the highest soil clay content, pHBC and CEC. The Buntine site had the widest ellipse, indicating the greatest variance in soil properties between samples from one site, reflecting the persistence of the historic organic matter addition treatment.

Fig. 2.

Principal components analysis based on Euclidian distance dissimilarity ordination plot showing the interaction between NH₃-N loss (NH₃.loss) from southwestern Australia cropping soils and various soil properties [clay content (clay), cation exchange capacity (CEC), pH buffering capacity (pHBC), organic carbon (OC) and initial pH(CaCl₂) (ipH)]. PC1 and PC2 represent 85.89% of data variance, with the percent variance explained by each coordinate axis. Bold ellipses encapsulate the variance within each experimental cropping site (see Table 1 for details) in relation to PC1 and PC2 at a 95% confidence interval.

NH₃ volatilisation

Total NH₃ loss after 3 weeks ranged from 0.9% (Merredin B, unburnt treatment) to 25% (Badgingarra, 0 t ha⁻¹ clay treatment) of N applied (Table 3). Cumulative NH₃ volatilised from surface broadcast urea (50 kg N ha⁻¹) varied significantly between all field sites (P < 0.001) (Table 4). No NH₃ was detected from the N fertiliser treatments after 2 weeks of incubation or from the nil N fertilised treatments during the 3 weeks of incubation.

Relationship between NH₃ volatilisation and soil properties

All measured soil properties were significantly correlated with total NH₃ volatilisation losses (P < 0.05) (Table 5). Soil pHBC most strongly related to NH₃ volatilisation losses (R² = 0.77). Negative relationships between NH₃ volatilisation losses and organic carbon, CEC, clay content and pHBC demonstrated that these soil properties helped prevent NH₃ volatilisation (Table 5, Fig. 3). In contrast, initial pH had a weak positive relationship with NH₃ volatilisation.

Fig. 3.

Exponential regressions between NH₃ volatilisation losses (% N applied) and soil physiochemical properties: (a) cation exchange capacity, (b) pH (CaCl₂), (c) clay content, (d) organic carbon content and (e) pH buffering capacity (pHBC). Cumulative NH₃ volatilisation of southwestern Australia cropping soils was measured for 3 weeks after applying 50 kg N ha⁻¹ as urea prills to the surface of soils wetted to 60% water holding capacity.

Analysing the soil properties as multiple log-linear regression to predict NH₃ volatilisation rates identified organic carbon, CEC and soil pH(CaCl₂) with the highest prediction capacities (adjusted R² = 0.92, P < 0.001; Table 6). The independence of these explanatory variables is described by the perpendicular vector angles between soil properties organic carbon, CEC and soil pH in the principal components biplot (Fig. 2). Soil pHBC and clay percentage were omitted from the equation as they demonstrated multicollinearity with organic carbon and CEC properties, respectively. Each soil property included in the multiple linear-log regression had significant regression coefficients (Table 5). The multi log-linear regression equation follows:

where NH₃-N% loss is the cumulative percentage NH₃-N loss of the applied N over 3 weeks, OC is soil organic carbon percentage of the soil sample and pH is pH(CaCl₂) of the soil sample.

Predictive capacity of the NH₃ volatilisation prediction model

The measured cumulative NH₃ volatilisation losses were poorly correlated with the predicted NH₃ volatilisation loss potential using the predictive model of Fillery and Khimashia (2016). The RMSE of 36% indicated that the model had poor predictive power in estimating NH₃ volatilisation losses from some of these soils, with predicted NH₃ volatilisation rates with range of 14.4–23.5%. In contrast, the range of experimental volatilisation losses was 0.9–25.2% (Table 3). The comparison analysis demonstrated that the model overestimated the NH₃ loss potentials for all field site soils except for Badgingarra. A positive exponential relationship occurred between organic carbon content and the difference between experimental and model-predicted NH₃ volatilisation values (R² = 0.78, P < 0.05; Fig. 4).

Fig. 4.

Exponential regression between organic carbon (%) and the error variance between predicted and observed experimental NH₃ volatilisation losses from southwestern Australian cropping soils. Predicted NH₃ volatilisation losses were calculated using the model developed by Fillery and Khimashia (2016).

NH₃ volatilisation from urea applied to sandy/sandy loam acidic soils

Volatilisation of NH₃ is not an agronomic concern exclusive to alkaline soils. Our study indicated that acidic soils with low pH buffering capacities could be subject to substantial NH₃ volatilisation losses. Importantly, the extent of NH₃ volatilisation in this laboratory study indicated that the susceptibility of NH₃-N losses from acidic to neutral soils is highly variable, ranging from negligible, <1–25% (Table 3). The greatest loss potential of 25% of applied NH₃-N occurred for the Badgingarra sandy soil [pH(CaCl₂) 6.5], comparable to the 27% measured loss in a field experiment following surface application of urea to an acidic sandy loam [pH(CaCl₂) 5.5] in the central grain belt of southwestern Australia (Fillery and Khimashia 2016). These values are some of the highest NH₃ volatilisation losses recorded from Australian cropping soils where urea has been applied (Barton et al. 2022).

Investigating the NH₃ volatilisation loss potentials of soils across a wide pH range, including acidic soils in this study and studies by Fillery and Khimashia (2016) and Ohnemus et al. (2021), has indicated that acidic soils are not only at risk to NH₃ volatilisation but could have the greatest NH₃ loss potential. Collectively these findings challenge current views that the magnitude of NH₃ volatilisation loss is greater when applying urea to alkaline soils (pH > 7) than acidic soils (Ernst and Massey 1960; Zhenghu and Honglang 2000; Hutchings et al. 2019). Past evaluation of NH₃ loss potential assumed all soils have a constant pH buffering capacity with a uniform pH shift following urea hydrolysis. Our study highlights that poorly buffered acidic soils can be subject to substantial NH₃ losses, likely driven by a severe pH shift as a consequence of poor pH buffering. Recently, Ohnemus et al. (2021) proposed that the greatest NH₃ loss potentials from soils with an initial pH in the range of 5.0–6.5 occur due to the lower CEC typical of many soils within this acidic pH range. Our results support this finding in some instances; however, there is large variability in NH₃ loss potentials from soils across this acidic to neutral pH range (Fig. 3), suggesting that initial pH is likely not the main regulatory factor determining NH₃ loss potential. Thus, relying on pH as an indicator for evaluating NH₃ loss potential could be misleading.

Three main reasons restrict the predictive capacity of initial pH: (1) initial soil pH is a temporal indication of risk in a system that is potentially subject to a significant pH change following urea hydrolysis (Black et al. 1985), (2) initial soil pH typically does not represent the peak pH evolved to or duration of the peak following urea hydrolysis (Ohnemus et al. 2021) and (3) pH evolution following urea hydrolysis is highly variable between soil types as it relates to soil pHBC (Ferguson et al. 1984; Izaurralde et al. 1987). These factors significantly impact the amount of NH₃ available in the soil system and are, thus, likely to have caused the soil property of initial pH to be the weakest regulatory property of NH₃ loss assessed in our study (R² = 0.40), demonstrated by the observed variance in volatilisation responses at the same initial pH. For example, at an initial pH of 6.2, Merredin L soil had 6% NH₃ losses compared to 14.5% from Buntine soil. As soil pH does not remain constant during urea hydrolysis, we suggest basing a greater predictive capacity for NH₃ volatilisation loss on soil properties that represent the soil’s pH response to the urea hydrolysis reaction.

Regulation of NH₃ volatilisation by soil properties

Volatilisation loss of NH₃ is regulated by a complex collection of physical and chemical soil interactions (O’Toole et al. 1985; Ohnemus et al. 2021). Controlled laboratory conditions can exclude the influences of environmental and agronomic regulatory parameters, isolating the regulatory influence of physical and chemical soil properties on NH₃ volatilisation losses. Our study found the strongest predictive capacity when NH₃ loss potential was considered a function of multiple regulatory parameters, using a multiple log-linear regression model (Eqn 3) (R² = 0.92). The regression model combined the influences of organic carbon content, CEC and initial soil pH on NH₃ loss potential. The function comprises three properties that represent three main regulatory mechanisms of NH₃ volatilisation potential: (1) soil capacity to buffer the alkalising urea hydrolysis reaction (Fillery and Khimashia 2016), i.e. soil pHBC, (2) soil’s affinity for complexing NH₄⁺ (Baldock et al. 2009), i.e. soil CEC and (3) pH dissociation preference for NH₃ or NH₄⁺ availability (Freney et al. 1983), i.e. soil pH. While this laboratory study investigated the NH₃ volatilisation potential under controlled environmental conditions, it offers insights into the soil properties that regulate NH₃ losses from acidic soils, a rarely considered factor compared to alkaline cropping soils (Barton et al. 2022).

Soil pHBC had the strongest capacity for predicting NH₃ volatilisation potential in this study (R² = 0.77) (Fig. 2). Soil pHBC was determined using a pedotransfer function (Eqn 2) that considered the proton adsorbing and releasing capacities of aluminium oxides, organic carbon content and clay content (Moore 2001). Other studies have recognised pHBC as an explanatory variable for NH₃ volatilisation risk (Freney et al. 1983; Kissel et al. 2008; Fillery and Khimashia 2016); however, using a pedotransfer function to define soil pHBC is a novel method for assessing NH₃ volatilisation loss potential. While the pedotransfer function method improves access by removing the lengthy analytical titration method of Aitken and Moody (1994), it is limited by the lack of universal application that the titration method provides, requiring region-specific functions that recognise differences in clay mineralogy and organic carbon functional groups (Wong and Wittwer 2009). For example, Coleman and Thomas (1964) demonstrated that a pHBC pedotransfer function developed for northeastern Australian soils overestimated the pHBC of southwestern Australia soils by two- to three-fold. This overestimation was attributed to differences in silicate clay structure and organic matter functional groups specific to geomorphic and climate regions (Wong and Wittwer 2009). Further research into methods as universal as Aitken and Moody’s (1994) titration method but as time efficient as Moore’s (2001) pedotransfer function would help improve access to soil pHBC data, providing a more accurate assessment in determining associated NH₃ loss potential than the weaker regulatory property of initial pH.

Predictive capacity of an NH₃ volatilisation model

The predictive model of Fillery and Khimashia (2016) tended to overestimate NH₃ losses from the urea applied to acid and neutral soils in the present study. The model best predicted NH₃ losses from the Badgingarra soils (<5.5% overestimation) as the site with the lowest organic carbon content. For the remaining sites, NH₃ loss predictions were overestimated by up to 18% (Fig. 3). The overestimation of NH₃ losses appears to be a consequence of excluding organic carbon content as a regulatory parameter in the NH₃ loss prediction model by Fillery and Khimashia (2016) (Fig. 4).

Variance in the model’s predictive capacity across different soil types likely reflects the assignment of clay content as a surrogate for soil pHBC, which assumes that all clays provide uniform pHBCs. However, clay types differ in their layer silicate structure and abundance of sesquioxide complexes, both of which affect the pHBC (Weaver et al. 2004; Wong et al. 2013). Therefore, the model’s pHBC factor does not always accurately represent actual soil pHBC and misrepresents the NH₃ volatilisation inhibitory capacity of the soil. For example, soils dominant in 2:1 clay types (e.g. smectite and vermiculite) have pHBC strongly related to clay content (Moore 2001), whereas 1:1 clay soils have pHBC strongly related to organic carbon content. Poor estimation of NH₃ loss potential in kaolinitic soils with medium to high organic carbon contents (i.e. >1.0%) could be a trade-off that comes with using accessible input variables to assess NH₃ volatilisation loss potential in the Fillery and Khimashia (2016) predictive model. Future work should incorporate an improved method for estimating or directly measuring pHBC. Better estimation could be gained by adopting pedotransfer functions to estimate pHBC or including more soil properties that account for pHBC, such as organic carbon content or recognise the different buffering capacities of clay mineralogy. Model updates should then be validated in the field, including environmental parameters that regulate volatilisation losses.