Conversion of soil particle size distribution and texture classification from ISSS system to FAO/USDA system in Japanese paddy soils

ABSTRACT

The conversion between the two different systems, ISSS and FAO/USDA, of particle size distribution and soil texture classification is useful to characterize soil physical properties and usage of each published. The objective of this study is to test some functions that have been published for conversion from ISSS to FAO/USDA system for Japanese paddy soils and to select the best method. We tested the topsoils of 267 Japanese paddy fields using the log-linear method, log-normal method, multiple regression method, Skaggs et al.’s method, and Minasny and McBratney’s method. The least AIC was obtained using multiple regression method, and the equation derived was given as follows:

si_FAO/USDA = 1.305si_ISSS + 0.396fs_ISSS−0.100cs_ISSS − 12.323

where si, fs, and cs are the percentage of silt, fine sand, and coarse sand respectively; ISSS and FAO/USDA is the fractionation system; and its RMSE was 3.1%. For the case that only the total sand content (s) is available instead of fine sand and coarse sand, the following equation was obtained:

si_FAO/USDA = 1.533si_ISSS + 0.314s_ISSS − 20.903 (RMSE = 3.7%)

Among the non-empirical methods, the best estimation method was Skaggs et al.’s method, and its RMSE was 3.3%. The soil texture classification by FAO/USDA system using estimated particle size fractions by the above equation can be classified to correct categories. The accuracy ratio of the classification was 93-97%.

KEY WORDS:

• Soil texture
• particle size distribution
• Japanese paddy soil
• FAO/USDA system
• ISSS system

View correction statement:

Correction

1. Introduction

There are two major soil particle size fraction systems in the world, the system of International Soil Science Society (ISSS) and the system of FAO/USDA (Minasny and McBratney 2001). The particle size limit for fractionation of each system is as follows:

      ISSS       FAO/USDA

clay     <2 μm      <2 μm

silt     2–20 μm     2–50 μm

sand     20–2000 μm    50–2000 μm

(fine sand     20–200 μm)

(coarse sand     200–2000 μm)

In addition, there are various soil texture classifications in the world (Figure 1). While the common classification system used in Japan is that of ISSS, the difference in the classification of particle size fractions and texture from that of FAO/USDA forced inconvenience for the immediate adoption of past outcomes based on FAO/USDA system. For instance, Allen et al. (1998) show the relationship between soil texture and field capacity and wilting point (Table 19 in Allen et al.) using the FAO/USDA soil texture classification. It is hard to use these data in Japan because of the two different classification systems.

Figure 1. Ternary diagram of soil texture classification by ISSS and FAO/USDA system (n = 267). Every used soil is plotted by each particle size fraction and texture classification

To overcome this problem, several studies are available which support the conversion from one system to the other system (Buchan 1989; Shirazi, Boersma, and Johnson 2001; Shirazi, Boersma, and Hart 1988). The conversion functions from ISSS system to FAO/USDA system are published for the soils in Australia (Minasny and McBratney 2001), Bulgaria (Rousseva 1997), China (Wei et al. 2014), Netherland, and Germany (Nemes et al. 1999). However, no study has been reported for conversion functions using Japanese paddy soils. Recently, Global Soil Partnership (GSP) proposes harmonizing soil analytical data for comparative analysis to manage soil resources. According to Pillar five working group of GSP (2017), ‘harmonization could be seen as the next step to standardization, provides the ability to describe, sample, classify, and analyze the soil in a way that allows the use of the results for later scientific use.’ Comparison and evaluation of past attempts to conversion of two texture systems could be valuable as a case study of harmonization of soil data in Japan.

The objective of this study is to test whether the functions that have been published for conversion from ISSS to FAO/USDA system can estimate correct value in the Japanese soil and to select the best method for Japanese paddy soils. Furthermore, we also summarize the relationship between soil textures in FAO/USDA system and some corresponding soil moisture properties (i.e., field capacity and wilting point) in Japanese paddy soils.

2. Materials and Methods

2.1. Materials, soil particle size distribution, and classification of soil texture

For the study, we collected 267 topsoil samples from the eastern parts of many Japanese paddy fields (Figure 2). The soil type of each soil sample was determined using e-SoilMap (https://soil-inventory.dc.affrc.go.jp/eSoilMap.html), which indicates the soil type using the comprehensive soil classification system of Japan First Approximation (Obara et al. 2011; Table 1). The majority of our soil samples belong to Lowland soil type. Of the 267 soil samples, Lowland soil comprised 69% (183), and Organic soils comprised 13% (34). This is a typical distribution of soil type of paddy soils in Japan.

Table 1. Distribution of soil groups of used soils by comprehensive soil classification system of Japan First Approximation (Obara et al. 2011)

Great soil group	Sub-total	Soil group	n
Man-made soils	0	Artifactual soils	0
		Reformed soils	0
Organic soils	34	Peat soils	34
Podzols		Podzols	0
Andosols	31	Regosolic Andosols	0
		Gleyed Andosols	8
		Wet Andosols	16
		Fulvic Andosols	0
		Non-allophanic Andosols	2
		Allophanic Andosols	5
Dark Red soils		Calcaric Dark Red soils	0
		Dystic Dark Red soils	0
		Eutric Dark Red soils	0
Lowland soils	183	Lowland Paddy soils	22
		Gley Lowland soils	83
		Gray Lowland soils	56
		Brown Lowland soils	15
		Regsolic Lowland soils	7
Red–Yellow soils	0	Argic Red–Yellow soils	0
		Cambic Red–Yellow soils	0
Stagnic soils	2	Stagnogley soils	0
		Pseudogley soild	2
Brown Forest soils	0	Brown Forest soils	0
Regosols	0	Volcanogenous Regosols	0
		Sandy Regosols	0
		Lithosols	0
		Terrestrial Regosols	0
Unknown			17

Figure 2. Sampling locations of used soils

For pretreatment, the soil organic matter in used soils was removed using H₂O₂. Dispersion technique was performed using sodium hexametaphosphate for non-volcanic ash soils, and ultrasonic treatment was performed for volcanic ash soils (Gee and Or 2002). Then, soil particle size distributions were obtained both as ISSS fractionation system and as FAO/USDA fractionation system. In addition, only for fractionation by ISSS system, fine sand, and coarse sand were separated. The particle size was fractionated using wet sieving for particles of size limit larger than 50 µm and using pipette method for particles of lower size limit (Gee and Or 2002).

2.2. Evaluation of different procedures to interpolate particle size distribution

To interpolate particle size distribution, we evaluated five representative procedures, namely log-linear conversion (Nemes et al. 1999), log-normal conversion (Shirazi and Boersma 1984; Buchan 1989), multiple regression, Skaggs et al.’s (2001) method, and Minasny and McBratney’s (2001) method.

Log-linear model is a method used to predict unknown particle fraction from the relationship as Equation (1)(1) $\mathrm{C}\mathrm{P}=\mathrm{a}\mathrm{L}\mathrm{n}\left(i\right)+b$(1) :

(1) $\mathrm{C}\mathrm{P}=\mathrm{a}\mathrm{L}\mathrm{n}\left(i\right)+b$(1)

where CP is the cumulative amount of particle ratio, i is the upper limit of each fraction (mm), and a and b are the parameters. We calculated a and b by regression for each soil and estimated si_FAO/USDA.

Log-normal regression is a model based on the assumption that particle size distribution can be expressed as log-normal equation (Buchan 1989).

(2) $\mathrm{f}\left(i\right)=\frac{exp\left[-{\left(x-a\right)}^2/2b^2\right]}{b{\left(2\pi \right)}^{1/2}}$(2)

where i is the upper limit of each fraction, f(i) is the log-normal density, x = log(i), and a and b are the logarithm of geometric mean particle size diameter and geometric particle size standard deviation. In our study, si_FAO/USDA was calculated by Equation (2)(2) $\mathrm{f}\left(i\right)=\frac{exp\left[-{\left(x-a\right)}^2/2b^2\right]}{b{\left(2\pi \right)}^{1/2}}$(2) and normal distribution function in a spreadsheet application of computers.

The multiple linear regression was also conducted. Each parameter of the explanatory variables was determined using the cross-validation method. First, the total data were split into four groups randomly. And three of the four groups were used for regression analysis and the rest one group was used for validation of the regression equation. This process was conducted four times and the average of the parameters was obtained.

The selection of explanatory variables is important for the above three estimation methods. Logically, we can use a freely selected combination of particle size from four limits (cs, fs, si, and cl. Hereafter we call it ‘case of fs–cs’). However, we are often forced to use total sand content only for the upper limit because fine sand and coarse sand values are not shown as separated fractions (hereafter we call it ‘case of s’). In this study, we examine every combination for the case of fs–cs and s. Because the total of each fraction is 100%, used number of explanatory variables for multiple regression must be available fraction minus 1. For multiple regression, we examined every combination satisfying the above conditions, including single regression.

Skaggs et al. (2001) proposed a simple empirical method to describe particle size distribution.

(3) $C{P}_i=\frac{100}{1+\left(1/C{P}_{i0}-1\right)exp\left(-a{R}^b\right)}$(3)

(4) $\mathrm{R}=\frac{i-i0}{i0},i\ge i0>0$(4)

We used cl, si_ISSS, and fs_ISSS to obtain parameters a and b and si_FAO/USDA in each soil.

We also tried the method of Minasny and McBratney (2001) that was proposed as the conversion equation from ISSS system to FAO/USDA system using 1210 samples including Australian soils, UNSODA, GRIZZLY, and the USDA–NRCS Soil Survey Laboratory data. The model is shown in Equation (5)(5) $\begin{array}{rl}& s{i}_{FAO/USDA}=-18.3914+2.0971s{i}_{ISSS}+0.6726s_{ISSS}\\ & -0.0142s{i}_{ISSS}^2-0.0049s_{ISSS}^2\\ & ifs{i}_{FAO/USDA}<0thens{i}_{FAO/USDA}=0.8289s{i}_{ISSS}+0.0198s_{ISSS}\end{array}$(5) :

(5) $\begin{array}{rl}& s{i}_{FAO/USDA}=-18.3914+2.0971s{i}_{ISSS}+0.6726s_{ISSS}\\ & -0.0142s{i}_{ISSS}^2-0.0049s_{ISSS}^2\\ & ifs{i}_{FAO/USDA}<0thens{i}_{FAO/USDA}=0.8289s{i}_{ISSS}+0.0198s_{ISSS}\end{array}$(5)

where s represents the total sand contents.

Each model was evaluated based on the accuracy of estimation of silt fraction (si_FAO/USDA), and we mainly compared the Root Mean Squared Error (RMSE) value of each method to choose the best estimator and also calculated Akaike’s Information Criterion (AIC) to avoid overfitting.

2.3. Estimation of FAO/USDA type soil texture classification

We classified used samples by FAO/USDA system with both measured and estimated particle size distribution. The results were shown as an accuracy ratio that was the ratio that soil texture classified with estimated particle size distribution correspond to soil texture classified with measured particle size distribution correctly.

3. Results

3.1. Distribution of soil particle size and classification of soil texture

The distribution of soil particle size of used soil spreads in a wide range (Figure 1). When measured using the system of ISSS, the sand, silt, and clay fractions ranged from 4.3% to 88.4%, from 5.4% to 55.5%, and from 6.2% to 58.1%, respectively. When measured using the system of FAO/USDA, the sand and silt fractions ranged from 2.3% to 85.6% and from 8.2% to 67.4%, respectively. Because the fraction of silt is higher in the FAO/USDA system, the distribution of plots shifted to silt fraction slightly (Figure 1). There are no samples classified by sandy clay, sand, or silt loam with ISSS system, and there are no samples classified by sandy clay, sand, or silt with FAO/USDA system.

3.2. Comparison of conversion methods

Figure 3 shows the comparison of silt contents by measured and estimated values. The selected explanatory variables and the AIC values of each estimation method are indicated in Table 2. In log-linear, log-normal, and multiple regression, the least AIC combination among all combinations is indicated for both ‘the case of fs–cs’ and ‘the case of s.’ The least AIC is obtained by multiple regression method for both ‘the case of fs–cs and s.’ In conclusion, the two regression equations were obtained as follows:

(8) $s{i}_{USDA/FAO}=1.305s{i}_{ISSS}+0.396f{s}_{ISSS}+0.100c{s}_{ISSS}-12.323$(8)

(9) $s{i}_{USDA/FAO}=1.533s{i}_{ISSS}+0.314s_{ISSS}-20.903$(9)

Table 2. Comparison of various methods to estimate silt contents by FAO/USDA system from ISSS soil particle size distribution (n = 267)

Method	Symbol	Variables	RMSE %	Likelihood	AIC
Log-linear	(a)	cl, si, fs, cs	7.8	−478	961
	(b)	cl + si, fs, cs	3.6	−390	784
	(c)	cl, si, s	5.1	−428	860
Log-normal	(d)	cl, si, fs, cs	9.3	−498	1001
	(e)	cl, si, s	8.6	−489	982
Multiple regression	(f)	si, fs, cs	3.1	−370	748
	(g)	si, s	3.7	−393	792
Skaggs et al. (2001)	(h)	cl, si, fs	3.3	−379	764
Minasny and McBratney (2001)	(i)	si, s	10.9	−517	1060

cl, clay; si, silt; fs, fine sand; s, sand; cs, coarse sand.

Figure 3. Comparison of various methods to estimate silt contents of FAO/USDA (si_FAO/USDA) from ISSS soil particle size distribution. The alphabet in the parenthesis in the first line of each graph correspond to the symbols in Table 2

The best equation was Equation (8)(8) $s{i}_{USDA/FAO}=1.305s{i}_{ISSS}+0.396f{s}_{ISSS}+0.100c{s}_{ISSS}-12.323$(8) , and the RMSE calculated by this equation was 3.1%. The standard deviations of each coefficient of Equations (8)(8) $s{i}_{USDA/FAO}=1.305s{i}_{ISSS}+0.396f{s}_{ISSS}+0.100c{s}_{ISSS}-12.323$(8) and (9(9) $s{i}_{USDA/FAO}=1.533s{i}_{ISSS}+0.314s_{ISSS}-20.903$(9) ) obtained by cross-validating four times are shown in Table 3.

Table 3. The standard deviations (sd) and coefficient of variations (cv) of four cross-validations by using multiple regression Equations (8)(8) $s{i}_{USDA/FAO}=1.305s{i}_{ISSS}+0.396f{s}_{ISSS}+0.100c{s}_{ISSS}-12.323$(8) and (9(9) $s{i}_{USDA/FAO}=1.533s{i}_{ISSS}+0.314s_{ISSS}-20.903$(9) )

	Equation (8)(8) $s{i}_{USDA/FAO}=1.305s{i}_{ISSS}+0.396f{s}_{ISSS}+0.100c{s}_{ISSS}-12.323$(8)				Equation (9)(9) $s{i}_{USDA/FAO}=1.533s{i}_{ISSS}+0.314s_{ISSS}-20.903$(9)
	Coefficient	sd	cv		Coefficient	sd	cv
Silt (si)	1.305	0.086	6.6		1.533	0.111	7.2
Fine sand (fs)	0.396	0.044	11.1		−
Coarse sand (cs)	0.100	0.057	56.7		−
Sand (s)	−				0.314	0.043	13.8
Intercept	−12.323	4.514	36.6		−20.903	5.523	26.4

The estimated silt contents using log-linear method spread in a narrow range (Figure 3). The best combination of variables for this method is cl + si, fs, and cs (e.g., using a size limit of 20 µm, 200 µm, and 2000 µm, respectively). The log-normal method can estimate the measured values poorly. Skaggs et al.’s method (2001) worked well but the AIC is slightly less than that of Equation (8)(8) $s{i}_{USDA/FAO}=1.305s{i}_{ISSS}+0.396f{s}_{ISSS}+0.100c{s}_{ISSS}-12.323$(8) . The estimation of values using Minasny and McBratney’s (2001) study overestimated the measured values.

3.3. Estimation of FAO/USDA type soil texture classification

As used samples were classified by FAO/USDA classification system, they were grouped into nine categories (Table 4). The classification using ISSS data with estimation by methods of Equations (8)(8) $s{i}_{USDA/FAO}=1.305s{i}_{ISSS}+0.396f{s}_{ISSS}+0.100c{s}_{ISSS}-12.323$(8) and (9(9) $s{i}_{USDA/FAO}=1.533s{i}_{ISSS}+0.314s_{ISSS}-20.903$(9) ), which are the best estimation methods by case of fs–cs and s, respectively, was conducted. Both equations can estimate soil texture classification with high accuracy. Estimation by Equation (8)(8) $s{i}_{USDA/FAO}=1.305s{i}_{ISSS}+0.396f{s}_{ISSS}+0.100c{s}_{ISSS}-12.323$(8) corresponded to correct classification by 97% of accuracy ratio. And it is 93% using Equation (9)(9) $s{i}_{USDA/FAO}=1.533s{i}_{ISSS}+0.314s_{ISSS}-20.903$(9) . The accuracy of estimation about silty clay loam was the least in both equations.

Table 4. Accuracy ratio of classification of soil texture to FAO/USDA system from estimation by multiple regression of Equation (8)(8) $s{i}_{USDA/FAO}=1.305s{i}_{ISSS}+0.396f{s}_{ISSS}+0.100c{s}_{ISSS}-12.323$(8) and Equation (9)(9) $s{i}_{USDA/FAO}=1.533s{i}_{ISSS}+0.314s_{ISSS}-20.903$(9) with ISSS soil particle size distribution

		Accuracy ratio (%)
Soil texture	n	Equation (8)(8) $s{i}_{USDA/FAO}=1.305s{i}_{ISSS}+0.396f{s}_{ISSS}+0.100c{s}_{ISSS}-12.323$(8)	Equation (9)(9) $s{i}_{USDA/FAO}=1.533s{i}_{ISSS}+0.314s_{ISSS}-20.903$(9)
Sand	―	―	―
Loamy sand	1	100	100
Sandy loam	27	100	96
Loam	85	98	95
Silt loam	26	100	88
Silt	―	―	―
Sandy clay loam	2	100	100
Clay loam	65	98	95
Silty clay loam	36	86	81
Sandy clay	―	―	―
Silty clay	13	100	100
Clay	12	100	100
Total	267	97	93

4. Discussions

4.1. Recommended method for conversion from ISSS to FAO/USDA system

According to the results of RMSE and AIC values, the method that can convert from ISSS to FAO/USDA system the most precisely is the multiple regression method. The order of RMSE in each method is completely consisting of that of AIC, it means that overfitting does not affect adversely the values of RMSE. Minasny and McBratney (2001) also tested multiple regression and log-linear conversion for 1610 samples including Australian databases, UNSODA, GRIZZLY, and USDA–NRCS, and concluded that empirical regression is the best method to estimate. However, the proposed equation by Minasny and McBratney (2001) did not show good RMSE and AIC values in our samples, and almost all estimated values are over the measured values (Figure 3 [i]). As the reason that the equation proposed by Minasny and McBratney (2001) did not give a good result for Japanese paddy soils is unclear, we recommend using our empirical regression equations given as Equations (8)(8) $s{i}_{USDA/FAO}=1.305s{i}_{ISSS}+0.396f{s}_{ISSS}+0.100c{s}_{ISSS}-12.323$(8) or (9(9) $s{i}_{USDA/FAO}=1.533s{i}_{ISSS}+0.314s_{ISSS}-20.903$(9) ) for Japanese paddy soils.

Empirical equations have the nature that obtained coefficients cannot be generalized unquestionably. The standard deviations of each coefficient obtained by cross-validating four times are shown in Table 3. The results indicate that the standard deviation was relatively high in the coefficient of coarse sand and intercept. Adaptability of the equations may have an unknown limit. The alternative to avoid this problem, which is essential in the empirical equation, is the use of Skaggs et al.’s (2001) method. The RMSE obtained by this method (3.3%) is very close to that of the multiple regression (3.1%). If users avoid using an empirical function, this method is used as a probable alternative. Four fractions of particle size are needed for this method, namely only the case of fs–cs is available to estimate. This could be one of the disadvantages.

Classifications by FAO/USDA soil texture using estimated particle size fractions can mostly correspond to these using measured particle size fractions. Table 4 indicates the required accuracy for the estimated classification. As the accuracy ratio in silty clay loam and silt loam is lower (Table 4), there is a tendency that accuracy ratio decreases in the soils in which silt contents are over about 55% (Figure 1). One advantage of using FAO/USDA classification would be that the soil texture categorization can characterize the soil particle size distribution well because the particle size distributions in the ternary diagram of FAO/USDA system spread in a wider range than that of ISSS (Figure 1); nearly all soils are classified as light clay or clay loam by the ISSS system.

4.2. Corresponding of some soil moisture properties

Some specific values about water retention properties are often summarized using FAO/USDA soil texture system. For instance, Allen et al. (1998) show the relationship between soil texture and field capacity and wilting point (Table 19 in Allen et al.). It would be convenient to show these relationships for Japanese paddy soils, particularly for field capacity, because field capacity of Japan is determined as water contents at −3.1 kPa or −6.3 kPa water potential (Hasegawa 1997), which is quite different from international common value, −33 kPa (Minasny, McBratney, and Bristow 1999). In our dataset, 199 data are available to show the relationship between soil texture and water contents at −3.1 kPa, and 244 data can be available to show the relationship between soil texture and water contents at the wilting point, −1500 kPa. In addition, Nakano et al. (2018) show further 151 and 173 data that can relate ISSS particle size fraction with water content at −3.1 kPa and −1500 kPa, respectively. We converted the data of Nakano et al. (2018) to FAO/USDA classification system by Equation (8)(8) $s{i}_{USDA/FAO}=1.305s{i}_{ISSS}+0.396f{s}_{ISSS}+0.100c{s}_{ISSS}-12.323$(8) , merged to our dataset, and then summarized the relationship between the water contents at these water potential and soil texture (Figure 4). Table 5 is rearranged from Figure 4 for reading each value easily.

Table 5. The relationship between soil texture classification by FAO/USDA and some corresponded moisture properties in Japanese paddy soils and comparison to the ‘typical values’ given by Allen et al. (1998)

	Field capacity (−3.1 kPa)				Wilting point (−1500 kPa)
	The 1st quartile	The 3rd quartile	n		The 1st quartile	The 3rd quartile	n	Field capacity†	Wilting point†
Sand								7 − 17	2 − 7
Loamy sand	26	26	1		7	8	3	11 − 19	3 − 10
Sandy loam	33	41	43		11	14	52	18 − 28	6 − 16
Loam	33	44	127		13	17	156	20 − 30	7 − 17
Silt loam	36	45	20		13	16	25	22 − 36	9 − 21
Silt								28 − 36	12 − 22
Sandy clay loam	29	43	6		13	15	7
Clay loam	37	47	97		17	20	108
Silty clay loam	38	46	36		18	21	41	30 − 37	17 − 24
Sandy clay
Silty clay	29	46	10		19	22	11	30 − 42	17 − 29
Clay	38	48	10		17	21	14	32 − 40	20 − 24

†: Data from Figure 19 in Allen et al. (1998)

Figure 4. Boxplots for the relationship between soil texture categorized by FAO/USDA classification system and field capacity (−3.1 kPa; n = 350) or wilting point (−1500 kPa; n = 417) in Japanese paddy soils

It is clearly shown in Figure 4 that wilting point (volumetric water content at −1500 kPa water potential) is influenced by soil texture, and the results are very close to the ‘typical values’ given by Allen et al. (1998). The soil texture of FAO/USDA system can be classified by texture-by-feel analysis easily (Thien 1979) and wilting point could be estimated by the classified soil texture approximately using Table 5 even in Japanese paddy soils. It is one of the advantages to use FAO/USDA system. On the other hand, the distributions of values of field capacity (−3.1 kPa) are quite higher than the results of Allen et al. (1998) and seem to be difficult to categorize by soil texture (Figure 4). The field capacity in Japan (−3.1 kPa) is more independent of soil texture than that in international common value (at −33 kPa). The estimation of field capacity in Japanese paddy field still remains for further researches.

Acknowledgments

The authors thank Mr. Yuji Ohashi (Hokkaido Research Organization), Tadayuki Kudo (Agriculture Research Institute, Aomori Prefectural Industrial Technology Research Center), Norimasa Tanikawa (Agriculture Research Institute, Aomori Prefectural Industrial Technology Research Center), Tomohiro Konno (Miyagi Prefectural Furukawa Agricultural Experiment Station), Yuki Kato (Yamagata Integrated Agricultural Research Center), Tadashi Ando (Yamagata Integrated Agricultural Research Center), Makiko Moriya (Yamagata Integrated Agricultural Research Center), Akira Matsuda (Yamagata Integrated Agricultural Research Center), Yoshihumi Nagumo (Niigata Agricultural Research Institute), Masaharu Aoki (Nagano Agricultural Experiment Station), Tadayoshi Uehara (Nagano Agricultural Experiment Station), Kiyoshi Okamoto (Nagano Vegetable and Ornamental Crops Experiment Station), Hideyuki Umemoto (Ishikawa Agriculture Research Center), Yoshihiro Mukai (Ishikawa Agriculture Research Center), and Kenji Nakamura (Ibaraki-ken Agricultural Experiment Institute) for sending soil samples and field capacity data. They also thank Ms. Yuko Itoh for her help in soil analysis.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This study was supported by the Ministry of Agriculture, Forestry, and Fisheries of Japan under a grant for the project entitled ‘Development of diagnostic methods and countermeasure techniques for overcoming high yield inhibitory factors’. This work was also supported by the Council for Science, Technology and Innovation (CSTI), the Cross-ministerial Strategic Innovation Promotion Program (SIP), the ‘Technologies for creating next-generation agriculture, forestry and fisheries’ (funding agency: Bio-oriented Technology Research Advancement Institution, NARO).