APP下载

Soil particle size range correction for improved calibration relationship between the laser-diffraction method and sieve-pipette method

2021-03-17WeiwenQIUWeiHUDenisCURTINandLidiaMOTOI

Pedosphere 2021年1期

Weiwen QIU,Wei HU,Denis CURTIN and Lidia MOTOI

New Zealand Institute for Plant & Food Research Limited,Private Bag 4704,Christchurch 8140(New Zealand)

(Received September 27,2019;revised November 25,2019)

ABSTRACT Particle size fraction(clay,silt,and sand)is an important characteristic that influences several soil functions.The laser-diffraction method(LDM)provides a fast and cost-effective measurement of particle size distribution,but the results usually differ from those obtained by the traditional sieve-pipette method(SPM).This difference can persist even when calibration is applied between the two methods.This partly relates to the different size ranges of particles measured by the two methods as a result of different operational principles,i.e.,particle sedimentation according to Stokes’Law vs.Mie theory for laser beam scattering.The objective of this study was to identify particle size ranges of LDM equivalent to those measured by SPM and evaluate whether new calibration models based on size range correction can be used to improve LDM-estimated particle size fractions,using 51 soil samples with various texture collected from five soil orders in New Zealand.Particle size distribution was determined using both LDM and SPM.Compared with SPM,original data from LDM underestimated the clay fraction(<2µm),overestimated the silt fraction(2-53µm),but provided a good estimation of the sand fraction(53-2 000µm).Results from three statistical indices,including Pearson’s correlation coefficient,slope,and Lin’s concordance correlation coefficient,showed that the size ranges of<2 and 2-53µm defined by SPM corresponded with the<5 and 5-53µm size ranges by LDM,respectively.Compared with the traditional calibration(based on the same particle size ranges),new calibration models(based on the corrected size ranges of these two methods)improved the estimation of clay and silt contents by LDM.Compared with soil-specific models(i.e.,different models were developed for different soils),a universal model may be more parsimonious for estimating particle size fractions if the samples to be assessed represent multiple soil orders.

Key Words:laser diffraction,Lin’s concordance correlation coefficient,particle size distribution,Pearson’s correlation coefficient,sedimentation method,soil separate,soil texture

INTRODUCTION

Particle size distribution(PSD)is a fundamental characteristic of soil that has a strong influence on soil physical and biological properties(Arriagaet al.,2006;Rawlinset al.,2013).The sieve-pipette method(SPM)has been the standard method for measuring PSD for many decades(Grangeonet al.,2012;Alaryet al.,2013;Wanget al.,2013;Weiet al.,2013).However,the sedimentation-based SPM is time-consuming and usually overestimates clay content,especially in soils with platy clay particles(Vitton and Sadler,1997;Di Stefanoet al.,2010).This has been verified by scanning electron microscopy data,which showed that silt-sized particles were wrongly included in the clay fraction determined by SPM(Yanget al.,2019).For these reasons,SPM is considered to be incapable of accurately and rapidly analyzing a large number of samples,and a more efficient method is required for PSD determination(Zobeck,2004).

The laser-diffraction method(LDM),among several new techniques,has been widely used to estimate PSD in recent years(Virtoet al.,2011).This method enables a fast analysis,with high repeatability for soils with a wide particle size range.Moreover,LDM allows the creation of a continuous size distribution curve,thus providing more detailed information than SPM(Eshelet al.,2004;Schulteet al.,2016).However,LDM may underestimate the clay fraction and overestimate the silt fraction,compared with SPM,especially for soils with flat disc-shaped/non-spherical particles(Konert and Vandenberghe,1997;Duret al.,2004;Eshelet al.,2004).Thus,PSD measured by LDM is not directly applicable for the soil textural classification.Several studies have been conducted to develop calibration models to determine SPM-soil separates(clay,silt,and sand)from LDM data(McCaveet al.,1986;Jonasz,1991;Loizeauet al.,1994;Beuselincket al.,1998;Eshelet al.,2004).However,these calibration models provide a weaker correlation between LDM and SPM for finer fractions(Di Stefanoet al.,2010;Yanget al.,2015).

The weaker correlation between LDM and SPM for finer particle fractions may indicate that the two methods differ in their threshold values of clay-silt and silt-sand boundaries.To this end,studies have been conducted to identify the threshold values using different statistical indices,to enable more accurate prediction of soil separates from LDM-derived PSD.For example,Konert and Vandenberghe(1997)found that<8µm fractions from LDM matched<2µm fractions from SPM for fluvial and Aeolian soils using an index of slope of 1.However,the greatest Pearson’s correlation coefficient(R)was observed at a threshold value of 2µm rather than 8µm.This indicated that statistical tests with different indices(Rand a slope of 1)might result in different threshold values for soil separates.Recently,Fisheret al.(2017)adapted the Lin’s concordance correlation coefficient(CCC)to explore the sand,silt,and clay thresholds for LDM.Not only does the CCC assess how close(R)data fall to the best-fit line,but it also shows how far that line is from the 45°line(slope of 1)through the origin.They found that an LDM threshold of<9µm corresponded to the<2µm fractions derived from the sedimentation method,which was close to that(<8µm)found by Konert and Vandenberghe(1997).The 8µm threshold has been widely used to estimate clay and silt contents by LDM(Kerryet al.,2009).However,Kerryet al.(2009)found that it was more appropriate to use a threshold of 4µm than 8µm for chalk soil.These results suggest that the threshold values of clay and silt between these two methods may not be universal and need to be further tested.Disagreement on the threshold values has restrained the applicability of LDM-derived PSD data for rapid soil texture determination(Fedotovet al.,2007;Kowalenko and Babuin,2013).Moreover,previous studies mainly focused on the threshold boundary between clay and silt,but the threshold boundary between silt and sand received little attention.Although the threshold boundary between clay and silt has previously been identified for LDM data,few studies have examined whether the size range correction can actually be used to improve the prediction of particle fractions from LDM.

The objectives of this study were to:i)identify the corresponding size fractions measured from both LDM and SPM using three statistical indices(i.e.,R,slope,and CCC),ii)evaluate whether new calibration models based on size range corrections can be used to improve soil separate estimation,and iii)determine whether specific soil orders need to be considered to improve the particle fraction estimation.

MATERIALS AND METHODS

Soil sampling

A total of 51 soil samples from five soil orders(Allophanic,Brown,Gley,Pallic,and Recent Soils)according to the New Zealand soil classification system(Hewitt,2010)with different soil textures were collected from different regions in New Zealand.The samples were taken from the top 15 cm using a sampling auger(5.0 cm in diameter).All samples were air-dried,gently crushed,and then sieved using a 2-mm mesh sieve to remove the coarse fragments.Organic carbon content was measured by dry combustion using a LECO TruMac analyzer(LECO Corporation,USA)and ranged from 24 to 144 g kg−1.

Soil separate measurements

Sieve-pipette method.The SPM is based on the Stokes’Law(Gee and Or,2002).Soil was dispersed using an ultrasonic vibrator(ca.20 g soil andca.60 mL water in a beaker;sonicated for 2 min).The sand fraction was collected by passing the soil-water suspension through a 53-µm sieve.A 50 mL sample of suspension(<53µm)was collected to measure silt plus clay.The suspension was allowed to settle for 8 h in a temperature-controlled room(20°C),and a sample was collected with a 50-mL pipette from a depth of 10 cm for determination of clay(Gee and Or,2002).Clay(<2µm),silt(2-53µm),and sand(53-2 000µm)contents were calculated as the percentage(%)of recovered sample mass based on the USDA soil texture classification system.

Laser-diffraction method.The LDM involves the assumption that the spatial distribution(i.e.,angle)of scattered light is a function of sample particle size.An optical model is used to evaluate the scattering data and calculate the PSD.The angle of diffraction is inversely proportional to particle size,and the intensity of the diffracted beam at any angle is a measure of the number of particles with a specific cross-sectional area in the optical path.The scattering model,interpreted by the Mie theory,accurately predicts the light scattering behavior of all materials under all conditions(Malvern Instruments Ltd.,2007)and is accepted by the international standard for laser diffraction ISO 13320-1.Therefore,the Mie theory was used in this study.It assumes that particles are spherical.

Each soil sample(2 g)was pretreated by gradually adding 20 mL of 30% H2O2for the removal of organic matter.The samples were cooled to room temperature(20°C)and then shaken overnight at 180 r min−1using an endover-end shaker before the PSD measurement.A preliminary experiment,using soils with contrasting texture,indicated that sample pre-treatment using 30% H2O2and sonication produced comparable PSD data with the SPM method.To minimize the possible influence of soil color on the refractive index(RI)and LDM-derived PSD(Mady and Shein,2017),peroxidation,rather than sonication,was used for sample pre-treatment before LDM measurement.Particle size distribution was measured on a volume basis with the wet unit(Hydro2000S)of the Mastersizer 2000(Malvern Instruments Ltd.,UK).The PSD values were measured within a size range of 0.02-2 000µm.A total of 1 000 equally spaced data points were obtained on a logarithmic scale of particle size.Each soil sample was repeated three times,and the averaged values were used for data analyses.The parameters were selected to ensure the optimum distribution of the sample into the dispersant(water).It was necessary to ensure that the samples were measured in a stable,reproducible state of dispersion.The RI values for particles and dispersant(water)were set as 1.544 and 1.330,respectively.The absorption value was set as 0.008(Eshelet al.,2004;Fisheret al.,2017).The stirrer speed was 2 800 r min−1,and the obscuration range was between 5% and 30%.To ensure the uniform dispersion of particles,ultrasonic sound was applied before the measurement of each sample for a duration of 60 s at a level of 30%.

Statistical methods

Size range corrections.Three statistical indices were used to identify the threshold values:R,slope,and CCC(Fisheret al.,2017).These three statistical indices relating SPM clay(<2µm)and LDM-derived cumulative content of each particle size range,starting with the finest size(0.02µm),were calculated using Matlab R2018a(MathWorks,Inc.,USA).This resulted in 1 000R,slope,and CCC values.Similar calculations were done for silt(2-53µm)and sand(53-2 000µm).The LDM size range corresponding to the various SPM fractions(clay,silt,and sand)was identified as that giving the greatestR,slope closest to 1,or the greatest CCC.This method was based on the assumption that the greatestR,slope closest to 1,and the greatest CCC are obtained when the two methods measure exactly the same particles.Note that LDM produced highly dense(1 000 points)data,and all the combinations of size ranges were considered in our statistical analyses,increasing confidence in the results.

Calibration model development.Two types of linear regression calibration models were compared in this study(Table I):the traditional model(MT)and new calibrationmodel(MN).For MT,the particle size ranges were kept as<2µm for clay,2-53µm for silt,and 53-2 000µm for sand for both SPM and LDM.For MN,although size ranges were maintained as<2,2-53,and 53-2 000µm for clay,silt,and sand,respectively,for SPM,the optimal size ranges for LDM were identified universally(MNu)using the combined data or soil-specifically(MNs)using data for each soil order.For each model(MT,MNu,and MNs),both universal(MT-U,MNu-U,and MNs-U)and soil-specific(MT-S,MNu-S,and MNs-S)calibration equations were developed to evaluate whether soil-specific equations are needed to predict soil separates.Therefore,this study involved two(MTand MN),three(MT,MNu,and MNs),and six models(MT-U,MT-S,MNu-U,MNu-S,MNs-U,and MNs-S),at three levels(Table I).

TABLE ICalibration models with different levels of information(model designation in brackets)

The whole dataset was first used to develop different models to test whether the new model(MN)outperformed the traditional one(MT).For the comparison of the prediction capability of these two types of models,we randomly allocated about half(26)of the samples to a calibration dataset and the remaining(25)samples to a validation dataset.For each calibration dataset,only universal equations(MT-U,MNu-U,and MNs-U)were developed for predicting the validation dataset.The same procedure was repeated 200 times for each model.

Statistical parameters for optimizing calibration models.Widely used parameters such as root mean square error(RMSE)and adjusted coefficient of determination(R2)were used to evaluate the goodness-of-fit of models.These parameters are expressed as:

whereR2adjis the adjustedR2,XLDMandXSPMrefer to the clay,silt,or sand content measured by LDM and SPM,respectively,jis the index of summation with starting point of 1 and stopping point of sample sizem,andkis the number of parameters.A smaller RMSE or larger adjustedR2indicates a better performance of a model to estimate SPM-soil separates from LDM data.

When soil-specific models(MT-S,MNu-S,and MNs-S)are developed,more parameters are introduced for evaluating soil separates compared with the universal models(MT-U,MNu-U,and MNs-U).Increasing the number of fitting parameters can promote the fitting performance.However,it may also increase the possibility of“excessive parameterization”.It is,therefore,crucial to select the most parsimonious model to minimize the“over fitting”risk.In order to evaluate models for parsimony,corrected Akaike’s information criterion(AICc)values were calculated(Burnham and Anderson,2002):

where RSS is the residual sum of squares.A lower AICc value implies that a model is more parsimonious.

RESULTS AND DISCUSSION

Particle size distribution measured by LDM without calibration

Figure 1 illustrates the particle size distribution and cumulative particle size distribution measured by LDM for three soils with contrasting textures.The particle size at the peak volume for the clay,silt loam,and loam soils was approximately 4,40,and 90µm,respectively,showing an increasing trend as the soil became coarser(Fig.1a).This trend was in agreement with the cumulative particle size distribution,i.e.,the cumulative particle volume was greater for finer soils at a wide range of particle sizes(Fig.1b).This confirmed the feasibility of using LDM to differentiate the PSD for various textures.

Size fraction data for the 51 investigated soils,measured by SPM and LDM without calibration,are presented in Fig.2.The LDM underestimated the clay fraction and overestimated the silt fraction when compared with SPM,but the two methods resulted in comparable sand fraction for most soils.On average,the clay content was underestimated by 62%,and the silt content was overestimated by 23%.This result is consistent with previous studies(Konert and Vandenberghe,1997;Pieriet al.,2006).For example,Yanget al.(2015)found that LDM underestimated the clay content by 45% and overestimated the silt content by 18% when compared with SPM for 235 samples.They also observed that differences between the two methods were positively correlated with the clay content.Interestingly,larger deviation of the clay fraction between the two methods were also observed for finer soils in the present study(Fig.2).This confirms that LDM-particle size data require calibration,especially for fine-textured soils(Yanget al.,2015;Fisheret al.,2017).

Soil separate measurements with MT

Fig.1 Particle size distribution(a)and cumulative distribution(b)for three types of soil texture,clay,silt loam,and loam,of three soil orders(Pallic,Gley,and Recent Soils).

Fig.2 Relationships between clay(<2µm),silt(2-53µm),and sand(53-2 000µm)fractions from five soil orders(Allophanic,Brown,Gley,Pallic,and Recent Soils)obtained by laser-diffraction method(LDM)and sieve-pipette method(SPM).Fitting equations with adjusted coefficient of determination(R2adj)are shown without fitting lines.

TABLE IIModel parameters and goodness-of-fit indices of traditional calibration models(MT)a)for estimating sieve-pipette method size fractions from five soil orders(Allophanic,Brown,Gley,Pallic,and Recent Soils)from laser-diffraction method data

Table II lists the parameters of MT.When all data were included,soil separates were estimated well by LDM,with adjustedR2of 0.35,0.44,and 0.60 for clay,silt,and sand contents,respectively(Cohen,1988).The adjustedR2values imply that a better prediction of the coarser fractions would be expected from LDM.Soil order affected the performance of the calibration model.While the clay fraction was well estimated for Allophanic,Brown,and Gley Soils(adjustedR2=0.39-0.79),the silt fraction was well estimated for Pallic and Recent Soils(adjustedR2=0.31 and 0.81,respectively).The better prediction of the clay content for Allophanic,Brown,and Gley Soils may be related to the incomplete removal of organic matter(Allophanics)and soil minerals being coated by Fe/Mn oxides(Brown and Gley soils),which would result in higher RI values(>1.6).According to Eshelet al.(2004),the overestimation of clay by reducing RI(from>1.6 to 1.544)may balance clay underestimation by LDM,resulting in a better prediction of the clay content.The sand content was,in general,accurately predicted for all soils except for Allophanics.The reason for the poor prediction of the sand fraction in the Allophanic soils is not known.However,the Allophanic soils,which had higher sand content compared with other soils(29.4%vs.17.3%-21.1%),exhibited the best clay content estimation by LDM.

Compared with the universal calibration model(MT-U),soil-specific calibration models(MT-S)resulted in an improved agreement between SPM and LDM for each size fraction,as indicated by the decreased RMSE(Table III).However,soil-specific calibration only increased the adjustedR2for clay fractions and did not decrease AICc values compared with the universal model,suggesting that the universal model was more parsimonious.This implies that it may be worthwhile to develop a soil-specific model only if one soil order is being investigated.However,from the statistical point of view,this model will not result in a major improvement in the estimates of soil separates using LDM when multiple soil orders are involved.

Identification of the corresponding size ranges between SPM and LDM

Figure 3 shows the values of three statistical indices for particles with different size ranges measured by LDM and soil clay,silt,and sand contents measured by SPM whendata from all 51 samples were included.All statistical index values changed with the size range.Using clay as an example,when coarser particles were included,bothRand CCC first increased to reach a peak and then decreased,while slope decreased consistently(Fig.3a).It is interesting that the CCC peak(0.71)aligned with a slope of 1 at the particle size range of<5µm,where a high value ofR(0.71)was also observed.Therefore,all three statistical indices suggested that 5µm was the boundary between clay and silt for LDM.From Fig.3b,the 5-53µm fraction from LDM had the highestR(0.74)and CCC(0.73),and unit slope with the silt content(2-53µm)of SPM.This further affirmed that 53µm was the boundary between silt and sand for LDM.Therefore,as coarser sizes were included in the sand fractions,bothRand CCC gradually increased and slope decreased to a stable value(Fig.3c).Overall,<5 and 5-53µm from LDM best matched with the<2 and 2-53µm fractions from SPM,respectively,while the sand fractions(53-2 000µm)measured by the two methods matched well.Therefore,the clay and silt particles were identified as different particle size ranges by the two methods.

TABLE IIIGoodness-of-fit indicesa)of traditional(MT)and new(MN)calibration modelsb)for estimating sieve-pipette method soil separates from laser-diffraction method(LDM)data

Fig.3 Statistical indices(Pearson’s correlation coefficient(R),slope,and the Lin’s concordance correlation coefficient(CCC))between fractions of different size ranges,clay(<2µm,a),silt(2-53µm,b),and sand(53-2 000µm,c)measured by sieve-pipette method and laser-diffraction method.The x-axis is shown on a logarithmic scale and represents the upper limit of particle size ranges with lower limits of 0(a),5(b),and 53(c)µm.

The optimized size threshold between clay,silt,and sand differed within different soil orders(Table IV).However,the size deviations depended on the statistical index used.WhileRdetected a large variation of threshold values for clay-silt(2-14µm)and silt-sand(21-447µm),slope and CCC showed small differences in threshold values for clay-silt(4-6µm)and silt-sand(45-64µm)among soils.Furthermore,a slope of 1 and CCC basically identified the same threshold values for different soils,with average values being close to those identified using the combined data.It seems thatRmay not be a suitable statistical index to identify the threshold values,given that an unacceptable threshold value was found for Allophanic soils.Part of the reason may be related to the fact thatRis sensitive to outliers.Therefore,only the threshold values determined by CCC would be used for developing new calibration models(MNuand MNs).

The corresponding particle sizes of clay between SPM and LDM have been discussed in previous studies.Both theoretical approaches and experimental data generally indicate that 8-9µm is the boundary between clay and silt based on LDM data(Konert and Vandenberghe,1997;Luet al.,2000;Pabstet al.,2000;Fisheret al.,2017).However,Kerryet al.(2009)confirmed that the use of 8µm as a cut-offcould cause over-estimation of the clay size range,and they recommended 4µm as an appropriate clay-silt threshold for chalk soils when comparing SPM and LDM data.Our study showed that the boundary(4-6µm with an average of 5µm)between clay and silt differed slightly between soil orders in New Zealand.This threshold was close to that suggested by Kerryet al.(2009),indicating again that 8µm is not a universal boundary between clay and silt.Note that different pre-treatment methods were used for LDM and SPM,although the two pre-treatment methods resulted in comparable SPM results in a preliminary study(data not shown).The possible influence of sample pre-treatment on the threshold values merits further investigation.

Soil separate measurements with MN

Soil separates were,in general,distributed along a 1:1 line after calibration using both MNu-U(universal thresholdbased)and MNs-U(soil-specific threshold-based)(Fig.4;data for MNu-Sand MNs-Sare not shown).Tables V and VI listmodel parameters and goodness-of-fit indices for estimating clay and silt contents using MNuand MNs,respectively.When a universal model(MNu-Uor MNs-U)was used,the RMSE decreased and the adjustedR2increased for both clay and silt contents as compared with the traditional model MT-U.For example,the adjustedR2increased from 0.35(MT-U)to 0.48(MNu-U)or 0.54(MNs-U)for the clay fraction,and from 0.44(MT-U)to 0.53(MNu-U)or 0.58(MNs-U)for the silt fraction(Table III).This indicates that the new models,by considering the size range corrections,can improve the capability to estimate clay and silt contents using LDM.In addition,MNs-Uproduced a slightly higher adjustedR2than MNu-Ufor all soil separates,indicating that the use of soil-specific threshold values improved the estimation of soil separates(compared with the use of universal threshold values in universal calibration equations).

TABLE IVOptimized size threshold(OST)for the boundary between clay,silt,and sand identified by soil orders(Allophanic,Brown,Gley,Pallic,and Recent Soils)using different statistical indices

Fig.4 Soil clay,silt,and sand contents from five soil orders(Allophanic,Brown,Gley,Pallic,and Recent Soils)measured by sieve-pipette method(SPM)vs.those measured by laser-diffraction method(LDM)using new calibration models with universal calibration equations and universal(MNu-U,a-c)or soil-specific(MNs-U,d-f)threshold value,for clay(a and d),silt(b and e),and sand(c and f).Fitting equations with adjusted coefficient of determination(R2adj)are shown without fitting lines.For MNu-U,the corresponding size ranges are<5,5-53,and 53-2 000µm for clay,silt,and sand,respectively;for MNs-U,the soil-specific threshold values identified by the Lin’s concordance correlation coefficient were used.

TABLE VModel parameters and goodness-of-fit indices of new calibration models with universal threshold values(MNu)a)for estimating soil separates(clay,silt,and sand)from five soil orders(Allophanic,Brown,Gley,Pallic,and Recent Soils)of sieve-pipette method(SPM)from laser-diffraction method(LDM)data

TABLE VIModel parameters and goodness-of-fit indices of new calibration models with soil-specific threshold values(MNs)a)for estimating soil separates(clay,silt,and sand)from five soil orders(Allophanic,Brown,Gley,Pallic,and Recent Soils)of sieve-pipette method(SPM)from laser-diffraction method(LDM)data

Similarly,the new calibration models based on soilspecific regressions(MNu-Sand MNs-S)improved the estimation of clay and silt contents by LDM in all cases,except for the clay fraction of Allophanic soil and the silt fractions of Brown and Gley soils.This was because the clay-silt threshold modification in the new method can correct the underestimation of the clay fraction by LDM.The clay fractions of Pallic and Recent soils consist of large quantities of platy shaped illite minerals,which affect the settling velocity of clay particles.This explains why the estimation of clay and silt contents of these two soils was greatly improved using the size range correction.In these situations,however,it seems that soil-specific threshold values(MNs-S)do not produce significant improvement in soil separate prediction compared with universal threshold values(MNu-S)(Tables V and VI).

For both traditional(MT)and new(MNuand MNs)models,soil-specific calibration equations improved the correspondence between SPM and LPM values for the clay and silt fractions(i.e.,increased adjustedR2and decreased RMSE values compared with the universal calibration equations)(Table III).However,the AICc values for the soil-specific equations increased slightly,suggesting that the universal model was more parsimonious.Therefore,if different soils are included in the estimation of soil separates,it is not efficient to develop soil-specific regression models.It may be preferable to estimate soil separates by developing a model with a universal equation based on soil-specific threshold values(MNs-U)for New Zealand soils.

Comparison of different methods in predicting soil separates with LDM data

The 51 soil samples were divided into two datasets for model calibration and validation using both traditional and new models built by universal equations.Figure 5 shows the results of one of 200 scenarios.Regardless of calibration or validation,the new model resulted in increased adjustedR2(Fig.6)and decreased RMSE and AICc values(not shown)for both clay and silt contents.For example,in the validation dataset,the mean values of adjustedR2for the traditional models(MT-U)were 0.34,0.43,and 0.59 for clay,silt,and sand contents,respectively,while the corresponding values for the new models were 0.47,0.52,and 0.59(for MNu-U),and 0.52,0.55,and 0.60(for MNs-U).This,again,indicates that the new model is better than the traditional model in terms of soil separate prediction.Compared with MNu-U,MNs-Uslightly improved the prediction of clay,silt,and sand contents.Interestingly,the adjustedR2values from the 200 scenarios were comparable between the calibration and validation datasets.This implies that the model developed during the calibration process provides a robust prediction of soil separates.

Fig.5 Calibration(C)and validation(V)of soil separate estimation by sieve-pipette method(SPM)and laser-diffraction method(LDM)using the traditional model with universal calibration equation(MT-U)and two new calibration models with universal calibration equation and universal(MNu-U)or soil-specific(MNs-U)threshold value.This figure shows the results of one typical scenario.Adjusted coefficient of determination values are included in brackets in the legends.

Fig.6 Adjusted coefficient of determination(R2)for soil separate measurements during the calibration(a)and validation(b)processes by sieve-pipette method and laser-diffraction method using the traditional model with universal calibration equation(MT-U)and two new calibration models with universal calibration equation and universal(MNu-U)or soil-specific(MNs-U)threshold value.

CONCLUSIONS

Laser diffraction data are commonly used to calculate soil separates(clay,silt,and sand)for determining soil texture.To this end,the traditional method has been widely used to estimate soil separates of SPM from the data measured by LDM without size range corrections(i.e.,assuming that the two methods measure the same particle size ranges).This study measured the PSD of 51 soils from different soil orders across New Zealand using both SPM and LDM.Three statistical indices showed that the<2 and 2-53µm SPM fractions corresponded with the<5 and 5-53µm LDM fractions when data for all soils were considered.The threshold values identified by CCC and slope ranged from 4 to 6µm for clay-silt and from 45 to 64µm for silt-sand,whileRwas observed not to be a good statistical index to identify the threshold values.New calibration models based on the size range corrections,using either universal or soilspecific threshold value,improved the capability to estimate soil separates.Although soil order affected the goodnessof-fit for both clay and silt contents,a universal regression model was acceptable in terms of model parsimony.This study emphasizes the importance of size range correction for developing calibration models to predict the traditional SPM size fractions(sand,silt,and clay)from laser diffraction data.

ACKNOWLEDGEMENTS

This study was completed as part of the Manaaki Whenua-Landcare Research-led MBIE Program,Soil Health and Resilience-A Pathway to Prosperity and Wellbeing(No.P/442062/01),Next Generation S-Map-Smarter Decisions(No.P/443063/01),and the Plant & Food Research-led Strategic Science Investment Fund Program,Sustainable Agro-Ecosystems,all funded by the New Zealand Ministry of Business,Innovation and Employment.We are grateful to Craig Tregurtha,Peg Gosden,Sarah Glasson,Kathryn Lehto,and Chris Dunlop from New Zealand Institute for Plant & Food Research Limited for their kind help in sieving,archiving soils,and providing the chemical reagent for this study.


登录APP查看全文