- Email: [email protected]
Statistical relationships between soil colour and soil attributes in semiarid areas
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
Available online at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/issn/15375110
Research Paper
Statistical relationships between soil colour and soil attributes in semiarid areas S. Iba´n˜ez-Asensio a,*, A. Marque´s-Mateu b, H. Moreno-Ramo´n a, S. Balasch c a
Department of Plant Production, Universitat Polite`cnica de Vale`ncia, Camino de Vera s/n, Valencia 46022, Spain Department of Cartographic Engineering, Universitat Polite`cnica de Vale`ncia, Camino de Vera s/n, Valencia 46022, Spain c Department of Statistics and Operations Research, Universitat Polite`cnica de Vale`ncia, Camino de Vera s/n, Valencia 46022, Spain b
article info
Soil colour has become one of the most innovative indicators used to adjust amendment
Article history:
and fertilizer rates in precision agriculture. This paper uses a combination of colour vari-
Received 6 March 2013
ables together with geographical, management and pedologic variables in order to find
Received in revised form
relationships between the three colour components (lightness, hue, and chroma) and
27 May 2013
several soil characteristics, in a semiarid environment. In these areas soils are weakly
Accepted 26 July 2013
developed, and organic matter, nitrogen, phosphorous and iron soil contents are usually
Published online 23 August 2013
low and undergo high spatial variability. Multivariate analysis was used to find statistical relationships that: determine soil colour in those environmental areas; reveal the most appropriate chromatic variables for each case; and determine the interactions between variables that can mask the effects of individual variables. Colour measurements were collected with a trichromatic colorimeter. Eighteen soil variables were used, of which eight resulted in statistically significant correlations with colour components. Those variables were sand (%), clay (%), parent material (marls), soil organic carbon (SOC), carbonate content, total nitrogen (TN), iron, and 1:5 soil:water extract electrical conductivity (EC1:5). Only sand was significant for all three colour components. The content of organic carbon was not significant in multiple regression analysis with respect to soil lightness in this study of semiarid soils. However it was significant in bivariate regression, in the same way as found in other studies. ª 2013 IAgrE. Published by Elsevier Ltd. All rights reserved.
1.
Introduction
Nowadays, it is becoming more important for soil studies to access detailed datasets of soil properties. Site-specific crop management (Bogrekci & Lee, 2005; Eshani, Upadhyaya, Slaugther, Shafii, & Pelletier, 1999), soil erosion and soil degradation control or digital soil mapping (Viscarra,
McKenzie, & Grundy, 2010) require the collection of high resolution soil spatial data. However, soils analyses are expensive and time consuming, both at the regional or the global scale. Soil is an anisotropic natural body and colour is one of the characteristics most used in its classification. Colour provides valuable information on the formation process as well as on constituent elements and other properties.
* Corresponding author. Tel.: þ34 963877333; fax: þ34 963879749. E-mail address: [email protected] (S. Iba´n˜ez-Asensio). 1537-5110/$ e see front matter ª 2013 IAgrE. Published by Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.biosystemseng.2013.07.013
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
Soil colour is the variable of interest in this study, in which we seek significant statistical relationships between soil colour and other pedologic and agricultural variables which are related to both soil formation factors and soil quality. It is commonly accepted that soil colour reflects the proportions of three principal constituents: humus (black), hydroxides of iron (red) and silicic acid, kaolinite and calcium carbonate (white), but there are other soil components, such as oxides of manganese, nitrogen or phosphorus for example, which are important nutrients that can be identified by colour variables such as lightness (Christensen, Bennedsen, Jørgensen, & Nielsen, 2004; Schwertmann, 1993; Simonson, 1993; Torrent & Barro´n, 2003). Other soil characteristics, such as texture, different organic matter, soil moisture, soil erosion, and so on, are also major variables that influence soil colour (Brady & Weil, 2006; Konen, Burras, & Sandor, 2003; Schulze, Nagel, Van Scoyoc, Henderson, & Baumgardner, 1993). When compared to other soil variables, colour is straightforward to measure and does not require complicated sample preparation, so different attempts to find relationships between colour and other soil constituents and properties are abundant in the literature (Doi,Wachrinrat, Teejuntuk, Sakurai, Sahunalu, 2010; He, Vepraskas, Lindbo, & Skaggs, 2003; Mouazen, Karoui, Deckers, De Baerdemaeker, & Ramo´n, 2007; Sa´nchez-Maran˜o´n, Martı´n-Garcı´a, & Delgado, 2011; Viscarra, Fouad, & Walter, 2008). These studies have drawbacks in relation to the precise designation of the influence of each variable that affects soil formation, and indirectly soil colour. In order to find confident correlations it is necessary to take samples from similar pedologic environments. However, it is not easy to find areas with such uniform environmental conditions. Even though the most suitable approach is to sample slopes or transects with similar pedologic environments, it is very difficult to find areas where soil colour differences are exclusively due to the attribute under study, without influence from other forming factors. Some authors have done experiments in wider areas which vary in their soil genesis processes (Ibarra-F et al., 1995; Schulze et al., 1993). Furthermore, most of those studies have been conducted in temperate climate areas with decarbonated soils and high contents of organic matter (Konen et al., 2003; Wills, Burras, Sandor, 2007), nitrogen (Eshani et al., 1999) or phosphorus (Bogrekci & Lee, 2005). In this study, efforts are focused on finding areas homogeneous only with respect to those fundamental variables, such as stoniness, crustability and high carbonate content, leaving out other variables. The few studies in semiarid areas (Sa´nchez-Maran˜o´n, Delgado, Melgosa, Hita, & Delgado, 1997; Sa´nchez-Maran˜o´n et al., 2011; Sa´nchez-Maran˜o´n, Ortega, Miralles, & Soriano, 2007; Sa´nchez-Maran˜o´n, Soriano, Megolsa, Delgado, & Delgado, 2004) support this approach. It is known from previous research conducted in humid areas, that soil colour is strongly conditioned by the presence of chromogen substances, which in turn are indicators of fertility as well as soil condition. However, two main questions remain unanswered. The first question refers to the approach used to study colour-soil relationships. The common procedure consists of
121
studying the influence of each individual chromogen component on soil colour, and developing colour-component regression equations where other variables are assumed to be constant (Viscarra et al., 2008). Furthermore, it should be remembered that the influence of each variable is not independent of the others, and that the soil does not remain unaltered over time, which limits this approach. The second question refers to specific characteristics of arid and semiarid areas. Unlike humid areas, soils in arid and semiarid areas: lack chromogen substances common in humid areas, are rich in substances not present in humid soils, i.e. carbonates and salts, and have very limited range of variability of some components such as soil organic carbon (SOC), N, or P. With regard to colour determination techniques, it is worth noting that the classical approach to communicating soil colour is based on the use of specific Munsell charts under natural illumination conditions (Munsell Color Co., 1994). Colour spaces developed in the early 1930s by the Commission Internationale de l’E´clairage (CIE) are suitable alternatives to visual techniques, i.e. observation of the Munsell charts, to avoid subjectivity-related issues. The use of those spaces in soil science is well documented in the literature (Torrent & Barro´n, 1993; Viscarra, Minasny, Roudier, & McBratney, 2006). Thus, the goal of this paper is to improve the knowledge about the topographic, management and pedologic variables that contribute to soil colour in a semiarid environment, where the presence of chemical chromogen agents is very limited due to weak soil development.
2.
Materials and methods
2.1.
Site description, sampling and processing of data
Soil samples were collected near the river Vinalopo´ in the province of Alicante (southeast Spain). The major soil orders in the area are Aridisols and Entisols (Soil Survey Staff, 2010). A total of 110 topsoil samples (0e15 cm) were collected over an area of 60 km2 with similar climate characteristics, but great diversity in parent material, aspect, elevation, proximity to the river, vegetation, and land management (Fig. 1). The climate is continental Mediterranean with cold winters and hot and dry summers, an annual rainfall of 350.5 mm and an annual average of temperature of 15.2 C. This demonstrates the semiarid character of area, more pronounced during the summer months. The principal types of land use are: crops (60%) with vineyards and fruit trees (olive and almond trees), abandoned fields (19%) and forest (14%) with pines and scrub. Geomorphology is defined by the valley of river Vinalopo´ and the terraces and slopes of two mountain ranges with northeastesouthwest orientation. The mean terrain height is 540 m above sea level, and slopes are less than 15%. The predominant geologic material is quaternary located in the bottom of the valley, whereas limestone, marls, clays and gypsum are present in the slopes (Fig. 2). Samples were first air-dried in the laboratory, crushed and sieved using a brass sieve with 2 mm openings (Soil Survey Staff, 2004). This preprocessing was carried out in order to
122
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
Fig. 1 e Study area with sample points overlaid.
remove uncertainty effects in colour measurements due to water content (Chang, Laird, & Hurburgh, 2005) and particle size of the samples (Barrett, 2002; Sa´nchez-Maran˜o´n et al., 1997; Torrent & Barro´n, 1993). The processed samples were transferred into measurement plates taking up all the available space before colour measurements were made. Each sample was measured three times, shaking the container between measurements. The value that was used in later processing stages was the mean value of the three measurements. The samples were processed and analysed following standard laboratory procedures (Soil Survey Staff, 2004) in order to collect physical, chemical properties and colour coordinates. The laboratory analyses carried out were: carbonate content by Bernard’s calcimeter (CaCO3), electrical conductivity of 1:5 soil:water extract (EC1:5), pH ratio soil:water 1:1, organic carbon by WalkleyeBlack method (SOC), free iron by citrate-dithionite extraction (Fe2O3), nitrogen by Kjeldahl method (TN), phosphorus by Olsen method (TP) and texture (USDA fractions) by Bouyoucos Hydrometer technique. All parameters were analysed according to the existing official methodology. Other variables were obtained by means of geographical information system (GIS) analyses: aspect, elevation and slope; and by observations in field visits: stoniness, land use
(agricultural or forest) and management (irrigation system and conservation techniques). The geographical coordinates collected during field visits were used to create a map of sample points, which allowed georeferencing of variables such as topographic or soil management attributes. The numerical values of these variables were arranged in tables, which were joined to the point map by means of a key identifier. The resulting data set can be divided into four groups of variables (Table 1). In order to determine the relationships between the three colour components and the remaining soil variables, we used a multiple regression model. Qualitative characteristics, i.e. irrigation system, parent material, and land use, were incorporated as explaining variables by means of their corresponding dummy variables (Draper & Smith, 1981). The inevitable existence of collinearity between the explaining variables in the study led us to use the all possible subset regression technique (Jobson, 1991). This computational technique provides a convenient strategy to select subsets of variables which can be further studied in the regression models. The criterion we used to select those subsets was Mallows’ Cp, whose objective is to find models which minimise mean square error in predicted colour observations. This prediction error takes into account both prediction variance and bias. We eventually chose those models
123
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
Fig. 2 e Principal geological materials in area by occupation area: quaternary materials, limestone, clays and gypsum, and marls.
Table 1 e Analytical methods, codes and units. General variables Geomorphologic
Pedologic
Management
Variable
Units
Elevation above sea Slope Aspect Geological material
m % Cardinal points
Calcium carbonate pH Electrical conductivity (1:5) Texture
% pH units dS m1 %Clay (2e0.05 mm) %Sand (0.05e0.002 mm) %Silt (>0.002 mm) % % mg kg1 % %
Stoniness Total Nitrogen Total Phosphorous Organic Carbon Iron Crop
Irrigation Colour
Lightness Chroma Hue
Methodology GIS GIS GIS Geological Map
1e 2e 3e 4e
Quaternary material Clay and gypsum Limestone Marl
1e 2e 3e 4e 1e 2e
Fruit trees Vineyard Forestry Abandoned field No Yes
Bernard Calcimeter pH-meter Conductimeter Bouyoucus
Gravimetric Kjeldhal Olsen WalkleyeBlack Dithionite citrate Observation in field
Observation in field CIELAB
Codification
Trichromatic Colorimeter
124
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
according to pedologic criteria that provided better interpretation of the numerical results.
2.2.
3.
Results and discussion
These procedures lead to a number of relationships which are summarised in this section. We provide statistical relationships between colour and soil variables, together with possible explanations and results from other research. It is necessary to note that these results are not based on experimental designs. This statistical approach has known practical limitations regarding interpretation of the results owing to the collinearity that exists between the explaining variables. Specifically, it is not possible to establish true causeeeffect relationships and the results should be considered as a first approach to the problem. The approach in this study is slightly different from previous research. On the one hand, we base relationships on multiple regression models, contrary to most studies that are based on bivariate regression models. On the other hand we analysed all three colour attributes lightness, hue and chroma, whereas soil colour studies focus mainly on lightness.
Elements of colorimetry
In the context of soil science, colour has always been one of the basic properties reported in soil profile descriptions (Simonson, 1993). The common approach is based on the use of specific Munsell charts under natural illumination. According to the Munsell Soil Color Chart, colour is reported in terms of the so-called hue, value, and chroma. However, we used a different laboratory procedure based on CIE standards and colorimeter measurements (Torrent & Barro´n, 1993; Viscarra et al., 2006). Colour characteristics were determined using a Konica Minolta CS-100A colorimeter which outputs chromaticity coordinates xy and luminance Y in the well-known CIE xyY space (CIE, 2004). Laboratory measurements provided xyY coordinates of both soil samples and the white reference target. Chromaticity coordinates were then transformed into tristimulus values XYZ using well-known formulas (CIE, 2004; Westland & Ripamonti, 2004).
3.1.
X ¼ Y$ðx=yÞ Y¼Y Z ¼ Y$ð1 x yÞ=y
Analytical characteristics
The main analytical characteristics of soils are given in Table 2. In summary, the soils were stony, with coarse texture, nonsaline and moderately alkaline. They were light, with predominant yellow and brown hues, and the carbonate content was high, whereas they were low in organic matter and several essential nutrient elements for plant growth, namely Fe, N, and P.
Next, tristimulus values were converted into CIELAB coordinates: 1=3 16 L ¼ 116$ðY=Y nÞ i h a ¼ 500$ ðX=Xn Þ1=3 ðY=Yn Þ1=3 h i b ¼ 500$ ðY=Yn Þ1=3 ðZ=Zn Þ1=3
3.2.
where Xn, Yn, Zn are the tristimulus values of the white reference. It should be noted that there are alternative formulas for CIELAB coordinates computation which should be used when either (X/Xn)1/3, (Y/Yn)1/3, or (Z/Zn)1/3 is less than (6/29)3 (CIE, 2004). Finally, CIELAB coordinates were transformed into the CIE L*C*h representation:
Model selection
A number of regression models were selected on the basis of Mallows’ Cp, following the same procedure for each colour component. With regard to lightness, six models with low values of Mallow’s Cp (values between 3.4 and 3.9) were selected which involved up to 10 variables: stoniness, carbonate content, SOC, TN, iron oxides, sand, aspect, EC, clay and vineyard (Table 3). Seven models were selected for hue with Mallow’s Cp values between 1.0 and 1.9 which include 9 variables as possible indicators of soil hue: stoniness, TN, clay, aspect, sand, marls, irrigation, SOC and phosphorus are the selected variables (Table 3).
h i1=2 2 Cab ¼ ðaÞ2 ðbÞ hab ¼ arctanðb=aÞ The three colour attributes are lightness (L*), chroma (C*), and hue angle (h*), which represent psychophysical correlates of human perception mechanisms.
Table 2 e Statistical summary of chemical, physical and geomorphologic results (n [ 110). CE
pH
1
% Mean Median Std. dev. Variance Maximum Minimum
39 30 31.4 983.9 100 0
EC1:5 dS m
8.3 8.3 0.3 0.1 8.9 7.4
0.6 0.2 0.8 0.6 2.7 0.2
CaCO3 TN Fe2O3 SOC %
%
%
%
30.7 29 14.1 198.9 92.9 6.9
0.1 0.1 0 0 0.2 0
0.5 0.3 0.6 0.3 3 0
1.01 0.99 0.35 1.2 1.58 0
TP mg kg 0.5 0.3 0.7 0.5 4.9 0
Clay 1
Silt
Sand
El
Sl
As
%
%
%
m
%
16.1 14 9.2 84.2 43 3
25.3 23 14.7 216.7 73 2
58.5 59 17.2 295.6 88 19
540.6 533.3 39.1 1525.4 678 479.5
3.5 2.9 2.5 6.2 13.5 0.6
227.5 223.3 46.2 2133.8 356.6 96.7
L*
h*ab
C*ab
Cielab coord. 56.9 56.8 7 48.5 71.3 41.6
68.5 67.9 5.5 29.9 88.6 37.5
17.3 17.4 3.6 12.7 27.8 7.3
CE, stoniness; EC1:5, electrical conductivity 1:5; CaCO3, carbonate content; TN, total nitrogen; Fe2O3, free iron; SOC, soil organic carbon; TP, total phosphorus; El, elevation; Sl, slope; As, aspect; L*, lightness; h*ab, hue; C*ab, chroma.
125
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
Table 3 e Mallows’ Cp values for lightness, hue and chroma regression models. Variable Mallows Cp L*
h*ab
C*ab
7 8 8 8 8 9 6 6 7 7 8 8 9 8 8 9 9 9 9 10
3.9 3.5 3.6 3.7 3.8 3.4 1.6 1.7 1 1.9 1.5 1.6 1.8 3.6 3.7 3.7 3.7 4.2 4.2 4.2
Standard CE EC1:5 CaCO3 TN SOC TP Fe2O3 Clay El Sl As Sand Crop Irr Marls deviation 5.26 5.22 5.22 5.22 5.23 5.19 4.5 4.5 4.5 4.48 4.44 4.45 4.43 2.79 2.79 2.77 2.77 2.78 2.78 2.76
x x x x x x x x x x x x x
x x x x x x
x
x x
x x x x
x
x x x x x x x x x x x x x x x x x x x x
x x x x x x x x x x x x x x x x x x x x
x x x x x x
x x x x x x x x x
x x x
x x x x x x x x x x x x x x x
x x x x x x x x x x x x x x x x x x
x x x x x x x
x x x x x x x x x x x x x x x x x x x x
x
x x x x x
x
x x x x x x x x x x x x x
Irr, Irrigation; other abbreviations in Table 2.
Regarding chroma, seven models with the lower values of Mallow’s Cp (values between 3.6 and 4.2) were tested. The 12 variables related to soil chroma are TN, SOC, phosphorus, clay, elevation, slope, sand, marls, stoniness, EC and irrigation (Table 3).
3.3. Relationship between lightness (L*) and soil characteristics There are five significant variables related to L*, two of them were positively correlated, and three were negatively correlated (Table 4). In this context, positively correlated means that L* has greater values as the corresponding explanatory variable increases, i.e. soil colour is lighter. The value of the determination coefficient (R2) was 49.4%.
Some of the relationships between variables were expected on the basis of research in other environments. Soil lightness is affected by texture because silt and clay particles reflect more visible light than sand particles, resulting in lighter soil (Spielvogel, Knicker, & Ko¨gel-Knabner, 2004). Moreover, sand particles have less specific surface area than those of either silt or clay, which in turn have a colloidal nature. Therefore, sand particles can be easily coated with organic matter, resulting in darker soils. In connection with particle size issues, it should be noted that sample processing itself has known effects on lightness. In this study different sieve sizes were tested (Fig. 3). A 2 mm sieve size was chosen since it allows preparation of more homogeneous samples while preserving the effects of chromogen elements. It is important to use the same size consistently across the experiment.
Table 4 e Regression analysis for Lightness (L*) Chroma (C*ab) and Hue (h*ab).
2
R Constant CaCO3 Sand Fe2O3 TN EC1:5 SOC Clay Marls
Lightness
Hue
Chroma
Estimate
Estimate
Estimate
49.4 65.28 0.094* 0.11* 2.08** 46.83** 0.0021** e e e
32.4 84.94 e 0.22*** e 34.45* e 0.36* 0.24*** e
36.7 13.76 e 0.13*** e e e 0.28*** 0.09* 2.5*
[*P < 0.05; **P < 0.01; ***P < 0.001]; Estimate ¼ Estimated values of regression model coefficients; Constant ¼ independent term of regression equation.
Fig. 3 e Effect of particle size on luminance.
126
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
Samples with higher carbonate contents had lighter colours, as in the research of Sa´nchez-Maran˜o´n et al. (1997) who studied crushed soil samples from climatic environments similar to ours, with low carbonate and organic matter contents. Spielvogel et al. (2004) studied fine textured Mollisols with higher organic matter and low carbonate contents, which are different from the soils used here, and obtained the same behaviour. It is well known that CaCO3, has considerable influence on soil lightness (Simonson, 1993). Limestone and marls contain large amounts of carbonates and are the parent materials of many of the soils studied for this paper. Therefore, a positive relationship between CaCO3 and lightness was expected. Spielvogel et al. (2004) found higher values of L* in calcareous soils (L* > 56.1) than in non-calcareous soils (L* < 55.3). It should be noted that their calcareous soils were mainly clayey and silty, whereas those in this study are sandy (Table 2). Iron oxide content also influences soil colour, mainly in the hue component (Schwertmann, 1993). Sa´nchez-Maran˜o´n et al. (1997) did not find significant influence of Fe2O3 on lightness either in crushed samples or in aggregates of red Mediterranean soils. However, we found a significant relationship between lightness and hue angle (correlation 0.58, P < 0.01), which may be indirectly associated with the influence of iron oxides on lightness. It can be observed from the data that redder soils with lower hue angles are darker than white or yellower soils with larger hue angles, that is, reddish soils are darker than yellowish soils in the present study area (note that a hue angle of 0 represents red and 90 represents yellow in CIELAB space). With regard to total nitrogen (TN), the results show a decrease in lightness as nitrogen content increases. These results agree with those of Qian, Klinka, and Lavkulich (1993) in forest mineral soils from Canada, despite the higher contents of TN in their soils. There are also obvious climate differences between the study areas. Unlike other soil components, the identification of nitrogen content from soil colour has not been greatly studied in the past. This trend has changed with the advent of site-specific management which requires precise amounts of fertilizer to be applied to crops in different parts of fields. The study of this component as a determinant of topsoil colour could be particularly important in arid and semiarid areas since it is common to find abandoned agricultural fields which have been colonised by leguminous plants. In the present study, nitrogen contents were higher in abandoned fields than in cultivated lands (fruit trees and vineyards). This may well be a consequence of nitrogen fixation carried out by leguminous symbiotic microorganisms. In order to gain more insight into this subject, we specifically studied TN contents under four different land use classes (fruit trees, vineyards, forests, and abandoned fields) by means of analysis of variance. The analysis of variance shows significant differences with land use (P ¼ 0.002) as depicted by Tukey’s HSD intervals (Fig. 4). In the present study, higher electrical conductivities are associated with lighter soils, which can be used to determine the presence of salts. In areas where evaporation largely exceeds precipitation there is no leaching, and sometimes white crusts are visible on the soil surface. White saline crusts
Fig. 4 e Tukey’s HSD intervals from the analysis of TN and vegetation and land use (1: Fruit trees [n [ 54]; 2: vineyard [n [ 13]; 3: forestry [n [ 19]; 4: abandoned field [n [ 24]).
mainly consist of sodium and magnesium chlorates and sulphates (Valentin, 1993), and soils with such crusts are usually affected by wind erosion as they dry (Aubert, 1976). From this viewpoint, soil colour could be considered as a good indicator of desertification in areas with an abundance of salts, either of geologic origin or derived from inadequate water management in agriculture (Shepherd & Walsh, 2002). Clay content, SOC, stoniness and aspect are soil variables which appeared in the initial models given by the all possible subsets regression analysis. However, they did not exhibit significant relationships to lightness and were not included in the final models. The existing literature on soil colour always reports significant correlations between clay content and lightness (positive sign), as well as between SOC and lightness (negative sign). Most of this research studied the influence of these variables on soil colour individually, without taking into account interactions with other soil properties. Bivariate linear regressions were computed to determine whether or not our data exhibited such behaviour. Due to the nature of our data (limited range of values for some soil attributes, and the large extent and heterogeneity of the study area) correlation coefficients can reflect weak relationships (Schulze et al., 1993). Table 5 shows the trends of lightness with respect to SOC and clay, and also includes TN due to its great effect on crop production, which makes it a variable of interest in the development of predictive equations on which to base precision agriculture approaches. SOC showed a negative relationship (P < 0.0001), however multivariate analysis showed a positive relationship of lightness with organic carbon (P ¼ 0.06), which may appear
Table 5 e Linear regression analysis for Lightness (L). Linear regression SOC Clay TN
Estimate
R2
4.06*** 0.03 1.63
0.38 0.04 0.14
*P < 0.05; **P < 0.01; ***P < 0.001; Estimate, estimated values of regression model coefficients.
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
surprising at first glance. The origin and texture of the samples can be of critical importance in this regard. Some authors suggest that the relationships between lightness and SOC depend to a large extent on soil texture and its homogeneity (Spielvogel et al., 2004). Red clay materials were found in this study, but the number of clayey samples was low in relation to the total number of analysed samples. Moreover, carbonate content and texture affect the relationship between soil organic carbon content and soil lightness. Spielvogel et al. (2004) found that in silty and clayey soils with similar organic carbon contents, carbonate-free samples were darker than calcareous samples. The soils in the present study had high contents of carbonates which can mask the influence of either organic matter or clay on lightness. Accordingly, the fact that TN is not significant in the bivariate analysis suggests interactions between this variable and land use. Stoniness and aspect were the other two soil variables which appeared in the initial models (Table 3) and were not included in the final models. Stoniness produces light colours, and a positive relationship was found between lightness and aspect in south-oriented samples as well as in north-oriented samples.
3.4.
Hue (hab)
There are four significant variables related to hue, one positively correlated, and three negatively correlated (Table 4). When dealing with hue in CIELAB space, positively correlated means that soil colour becomes yellowish, and negatively correlated means that soil colour becomes reddish as the explanatory variable increases. The value of the determination coefficient (R2) was 32.4%. Much of the scatter is probably associated with differences in soil texture and in material origin, orientation and other pedologic variables in the samples (Schulze et al., 1993). Sand and clay are negatively correlated, which suggests that high contents of sand and clay produce lower values of hue (in this context this means that soil colour approaches to the red colour.) This agrees with hue angle measurements of Sa´nchez-Maran˜o´n et al. (2011) on individual particles of Regosols and Umbrisols from southeast Spain. The negative relationship between sandy samples and hue may occur because the particles in the sand fraction have been weathered. Williams and Yaalon (1977) obtained reddish sandy particles in the laboratory by the alteration of iron-rich heavy minerals and the precipitation of iron on surrounding quartz grains. In the case of clayey soils, more clayey samples have lower hues. In the present study area the detrital fine particle rocks are red clays (Fig. 2). SOC and TN also have an effect on soil hue. The chromogen nature of SOC component causes different colorations in topsoil depending on its content (Schulze et al., 1993). These authors highlighted the influence exerted by organic matter on hue, value, and chroma, being specifically relevant the composition of the organic matter. In the soils for this study, higher contents of SOC give reddish hues, whilst greater TN gives yellowish hues. Marls and stoniness are two soil variables which appeared in the initial models given by the all possible subsets regression analysis. However, they did not exhibit significant
127
relationships to hue (P ¼ 0.097 and P ¼ 0.07 respectively) and were not included in the final models. In general terms, topsoil developed on limestone or marls have yellower colorations than those developed on quaternary, clay, or gypseous materials. Similarly, less stony topsoils are redder than stony topsoils. In both cases, soil colours are determined by the presence or absence of carbonates in parent materials (Simonson, 1993). The relationships found by Sa´nchez-Maran˜o´n et al. (1997) in red soils of Spain and Italy do not fully agree with the results presented here. They found relationships of hue and iron in processed samples, and horizon type (A or B), elevation and clay in aggregates.
3.5.
Chroma (C*ab)
There are four significant variables related to chroma. Marls and SOC are negatively correlated, whereas sand and clay content are positively correlated (Table 4). The value of R2 is 36.7%. It seems that the colour of samples located on white marls is less saturated than for other soils, which is in agreement with the expected result. Organic matter content has the same effect. Results obtained by Konen et al. (2003) in mollic epipedions are similar to the results presented here. On the other hand, finer textured soils have more saturated colours. This may be due to the red colours of fine grained sedimentary rocks as a consequence of high contents of iron in the oxidised state. Likewise, sandy samples have larger chromas. Sa´nchezMaran˜o´n et al. (2011) found similar results in sandy particles obtained by dispersion from non-calcareous materials. Chromas were higher for particle sizes under 0.05 mm than those for the 1e2 mm fraction. This relationship is possible when the particles in the sand fraction have been weathered, much like in the case of hue (Williams & Yaalon, 1977). We found similarities in hue and chroma components which may suggest some sort of interaction between these two variables. Such interaction seems to be supported by the fact that some definitions of chroma rely on hue, and that neutral colours, i.e. colours without hue, must have a chroma of zero. The issue of the hue chroma interaction has been reported previously in the colour difference literature (CIE, 2004) as well as in more specific soil science literature (Buntley & Westin, 1965). In the latter case, numerical functions of Munsell hue and chroma are used as indices of soil development. The bivariate regression analysis of hue and chroma seems to support this (R2 ¼ 65%).
4.
Conclusions
All of the pedologic variables, except stoniness, pH and phosphorus, exhibited significant relationships with some of the three colour attributes. None of topographic and management variables had significant influence on soil colour. It seems that there exists considerable interaction between hue and chroma (R2 ¼ 65%) and it should be noted that determination coefficients (R2) between soil variables and soil colour are always less than 50%, which is usual in studies
128
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
where observational data are not subjected to experimental design. There are six variables that have an effect on the three soil colour components. Sand content is the only significant variable for all three colour attributes. Nitrogen explains lightness and hue, and clay and SOC explain chroma and hue. Carbonate, iron and EC are significant in the lightness models, whereas marls are significant in the chroma models. Soil texture, nitrogen and soil organic matter content show a remarkable relationship with lightness. Experience shows important interactions between SOC, clay, iron and carbonate content, as well as between TN and land use. The significant correlations found between soil colour and several soil variables demonstrate the validity of this approach, although further studies are needed. This would help in obtaining predictive equations which may be useful in precision agriculture and other engineering applications. From the results of this study, colour attributes could be suitable to predict other soil characteristics (for instance SOC content) provided that land managers are careful when using predictive equations derived from data analyses. In arid and semiarid environments with irregular relief, it is quite common that close locations, even at the individual plot level, exhibit high spatial variability in topsoil texture and chemical composition. In these cases, it may be necessary to have a specific equation for each different homogeneous soil unit.
references
Aubert, G. (1976). La morphologie des sols affecte´s par le sel, reconnaissance et pre´vision, surveillance continue. In FAO soils bulletin: Vol. 31. Prognosis of salinity and alkalinity (pp. 187e194). Roma. Barrett, L. R. (2002). Spectrophotometric color measurement in situ in well drained sandy soils. Geoderma, 108, 49e77. Bogrekci, I., & Lee, W. S. (2005). Spectral phosphorus mapping using diffuse reflectance of soils and grass. Biosystems Engineering, 91, 305e312. Brady, N., & Weil, R. R. (2006). The nature and properties of soils. New Jersey: Prentice Hall. Buntley, G. J., & Westin, F. C. (1965). A comparative study of developmental color in a ChestnuteChernozemeBrunizem soil climosequence. Soil Science Society Proceedings, 29, 579e582. Chang, C.-W., Laird, D. A., & Hurburgh, G. R. (2005). Influence of soil moisture on near infrared reflectance spectroscopic measurement of soil properties. Soil Science, 170, 244e255. CIE. (2004). Colorimetry e Technical report. Vienna: Commission Internationale de l’E´clairage. Christensen, L. K., Bennedsen, B. S., Jørgensen, R. N., & Nielsen, H. (2004). Modelling nitrogen and phosphorus content at early growth Barley using Hyperspectral Line Scanning. Biosystems Engineering, 88, 19e24. Doi, R., Wachrinrat, C., Teejuntuk, S., Sakurai, K., & Sahunalu, P. (2010). Semiquantitative color profiling of soils over a land degradation gradient in Sakaerat, Thailand. Environmental Monitoring Assessment, 170, 301e309. Draper, N., & Smith, H. (1981). Applied regression analysis. New York: Wiley. Eshani, M. R., Upadhyaya, S. K., Slaugther, D., Shafii, S., & Pelletier, M. (1999). A NIR technique for rapid determination of soil mineral nitrogen. Precision Agriculture, 1, 217e234.
He, X., Vepraskas, M. J., Lindbo, D. L., & Skaggs, R. W. (2003). A method to predict soil saturation frequency and duration from soil color. Soil Science Society of America Journal, 67, 961e969. Ibarra-F, F. A., Cox, J. R., Martin-R, M. H., Crowl, T. A., Post, D. F., Miller, R. W., et al. (1995). Relationships between buffelgrass survival, organic carbon and soil color in Mexico. Soil Science Society of America Journal, 59, 1120e1125. Jobson, J. D. (1991). Applied multivariate data analysis. Volume I: Regression and experimental design. New York: Springer-Verlag. Konen, M. E., Burras, C. L., & Sandor, J. A. (2003). Organic carbon, texture and quantitative color measurement relationships for cultivated soils in North Central Iowa. Soil Science Society of America Journal, 67, 1823e1830. Mouazen, A. M., Karoui, R., Deckers, J., De Baerdemaeker, J., & Ramon, H. (2007). Potential of visible and near-infrared spectroscopy to derive colour groups utilising the Munsell soil colour charts. Biosystems Engineering, 97, 131e143. Munsell Color Co.. (1994). Munsell Soil Color Charts. New York: Munsell Color Company. Qian, H., Klinka, K., & Lavkulich, L. M. (1993). Relationships between color value and nitrogen in forest mineral soils. Canadian Journal of Soil Science, 73, 61e72. Sa´nchez-Maran˜o´n, M., Delgado, G., Melgosa, M., Hita, E., & Delgado, R. (1997). CIELAB color parameters and their relationships to soil characteristics in Mediterranean red soils. Soil Science, 162(11), 833e842. Sa´nchez-Maran˜o´n, M., Martı´n-Garcı´a, J. M., & Delgado, R. (2011). Effects of the fabric on the relationship between aggregate stability and color in a RegosoleUmbrisol soilscape. Geoderma, 162, 86e95. Sa´nchez-Maran˜o´n, M., Ortega, R. B., Miralles, I., & Soriano, M. (2007). Estimating the mass wetness of Spanish arid soils from lightness measurements. Geoderma, 141, 397e406. Sa´nchez-Maran˜o´n, M., Soriano, M., Megolsa, M., Delgado, G., & Delgado, R. (2004). Quantifying the effects of aggregation, particle size and components on the colour of Mediterranean soils. European Journal of Soil Science, 55, 551e565. Schulze, D. G., Nagel, J. L., Van Scoyoc, G. E., Henderson, T. L., & Baumgardner, M. F. (1993). Significance of organic matter in determining soil colors. In J. M. Bigham, & E. J. Ciolkosz (Eds.), Soil color, SSSA Special Publication No 31 (pp. 71e90). Madison: Soil Society of America. Schwertmann, U. (1993). Relations between iron oxides, soil color and soil formation. In J. M. Bigham, & E. J. Ciolkosz (Eds.), Soil color, SSSA Special Publication No 31 (pp. 51e70). Madison: Soil Society of America. Shepherd, K. D., & Walsh, M. G. (2002). Development of reflectance spectral libraries for characterization of soil properties. Soil Science Society of America Journal, 66, 988e998. Simonson, R. W. (1993). Soil color standards and terms for field-use. History of their development. In J. M. Bigham, & E. J. Ciolkosz (Eds.), Soil color, SSSA Special Publication No 31 (pp. 1e20). Madison: Soil Society of America. Soil Survey Staff. (2004). Soil survey laboratory methods manual e Soil survey investigations report no. 42. Burt. R. U.S. Department of Agriculture, Natural Resources Conservation Service. Soil Survey Staff. (2010). Keys to soil taxonomy (11th ed.). Washington, DC: USDA-Natural Resources Conservation Service. Spielvogel, S., Knicker, H., & Ko¨gel-Knabner, I. (2004). Soil organic matter composition and soil lightness. Journal of Plant Nutrition and Soil Science, 167, 545e555. Torrent, J., & Barro´n, V. (1993). Laboratory measurement of soil color: theory and practice. In J. M. Bigham, & E. J. Ciolkosz (Eds.), Soil color, SSSA Special Publication No 31 (pp. 21e34). Madison: Soil Society of America.
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
Torrent, J., & Barro´n, V. (2003). The visible diffuse reflectance spectrum in relation to the color and crystal properties of hematite. Clays and Clay Minerals, 51, 309e317. Valentin, C. (1993). Soil crusting and sealing in West Africa and possible approaches to improved management. In FAO soils bulletin: Vol. 69. Soil tillage in Africa: Needs and challenges. Viscarra, R. A., Fouad, Y., & Walter, C. (2008). Using a digital camera to measure soil organic carbon and iron contents. Biosystems Engineering, 100, 149e159. Viscarra, R., McKenzie, N., & Grundy, M. (2010). Using proximal soil sensors for digital soil mapping. In J. L. Boettinger, D. W. Howell, A. C. Moore, A. E. Hartemink, & S. Kienast-
129
Brown (Eds.), Digital soil mapping: Bridging research, production and environmental application and operation (pp. 79e92). London e New York: Springer. Viscarra, R. A., Minasny, B., Roudier, P., & McBratney, A. B. (2006). Color spaces for soil science. Geoderma, 133, 320e377. Westland, S., & Ripamonti, C. (2004). Computational color science using MATLAB. Chichester: John Wiley & Sons. Williams, C., & Yaalon, D. H. (1977). An experimental investigation of reddening in dune sand. Geoderma, 17, 181e191. Wills, S. A., Burras, C. L., & Sandor, J. A. (2007). Prediction of soil organic carbon content using field and laboratory measurements of soil color. Soil Science Society of America Journal, 71, 380e388.
Available online at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/issn/15375110
Research Paper
Statistical relationships between soil colour and soil attributes in semiarid areas S. Iba´n˜ez-Asensio a,*, A. Marque´s-Mateu b, H. Moreno-Ramo´n a, S. Balasch c a
Department of Plant Production, Universitat Polite`cnica de Vale`ncia, Camino de Vera s/n, Valencia 46022, Spain Department of Cartographic Engineering, Universitat Polite`cnica de Vale`ncia, Camino de Vera s/n, Valencia 46022, Spain c Department of Statistics and Operations Research, Universitat Polite`cnica de Vale`ncia, Camino de Vera s/n, Valencia 46022, Spain b
article info
Soil colour has become one of the most innovative indicators used to adjust amendment
Article history:
and fertilizer rates in precision agriculture. This paper uses a combination of colour vari-
Received 6 March 2013
ables together with geographical, management and pedologic variables in order to find
Received in revised form
relationships between the three colour components (lightness, hue, and chroma) and
27 May 2013
several soil characteristics, in a semiarid environment. In these areas soils are weakly
Accepted 26 July 2013
developed, and organic matter, nitrogen, phosphorous and iron soil contents are usually
Published online 23 August 2013
low and undergo high spatial variability. Multivariate analysis was used to find statistical relationships that: determine soil colour in those environmental areas; reveal the most appropriate chromatic variables for each case; and determine the interactions between variables that can mask the effects of individual variables. Colour measurements were collected with a trichromatic colorimeter. Eighteen soil variables were used, of which eight resulted in statistically significant correlations with colour components. Those variables were sand (%), clay (%), parent material (marls), soil organic carbon (SOC), carbonate content, total nitrogen (TN), iron, and 1:5 soil:water extract electrical conductivity (EC1:5). Only sand was significant for all three colour components. The content of organic carbon was not significant in multiple regression analysis with respect to soil lightness in this study of semiarid soils. However it was significant in bivariate regression, in the same way as found in other studies. ª 2013 IAgrE. Published by Elsevier Ltd. All rights reserved.
1.
Introduction
Nowadays, it is becoming more important for soil studies to access detailed datasets of soil properties. Site-specific crop management (Bogrekci & Lee, 2005; Eshani, Upadhyaya, Slaugther, Shafii, & Pelletier, 1999), soil erosion and soil degradation control or digital soil mapping (Viscarra,
McKenzie, & Grundy, 2010) require the collection of high resolution soil spatial data. However, soils analyses are expensive and time consuming, both at the regional or the global scale. Soil is an anisotropic natural body and colour is one of the characteristics most used in its classification. Colour provides valuable information on the formation process as well as on constituent elements and other properties.
* Corresponding author. Tel.: þ34 963877333; fax: þ34 963879749. E-mail address: [email protected] (S. Iba´n˜ez-Asensio). 1537-5110/$ e see front matter ª 2013 IAgrE. Published by Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.biosystemseng.2013.07.013
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
Soil colour is the variable of interest in this study, in which we seek significant statistical relationships between soil colour and other pedologic and agricultural variables which are related to both soil formation factors and soil quality. It is commonly accepted that soil colour reflects the proportions of three principal constituents: humus (black), hydroxides of iron (red) and silicic acid, kaolinite and calcium carbonate (white), but there are other soil components, such as oxides of manganese, nitrogen or phosphorus for example, which are important nutrients that can be identified by colour variables such as lightness (Christensen, Bennedsen, Jørgensen, & Nielsen, 2004; Schwertmann, 1993; Simonson, 1993; Torrent & Barro´n, 2003). Other soil characteristics, such as texture, different organic matter, soil moisture, soil erosion, and so on, are also major variables that influence soil colour (Brady & Weil, 2006; Konen, Burras, & Sandor, 2003; Schulze, Nagel, Van Scoyoc, Henderson, & Baumgardner, 1993). When compared to other soil variables, colour is straightforward to measure and does not require complicated sample preparation, so different attempts to find relationships between colour and other soil constituents and properties are abundant in the literature (Doi,Wachrinrat, Teejuntuk, Sakurai, Sahunalu, 2010; He, Vepraskas, Lindbo, & Skaggs, 2003; Mouazen, Karoui, Deckers, De Baerdemaeker, & Ramo´n, 2007; Sa´nchez-Maran˜o´n, Martı´n-Garcı´a, & Delgado, 2011; Viscarra, Fouad, & Walter, 2008). These studies have drawbacks in relation to the precise designation of the influence of each variable that affects soil formation, and indirectly soil colour. In order to find confident correlations it is necessary to take samples from similar pedologic environments. However, it is not easy to find areas with such uniform environmental conditions. Even though the most suitable approach is to sample slopes or transects with similar pedologic environments, it is very difficult to find areas where soil colour differences are exclusively due to the attribute under study, without influence from other forming factors. Some authors have done experiments in wider areas which vary in their soil genesis processes (Ibarra-F et al., 1995; Schulze et al., 1993). Furthermore, most of those studies have been conducted in temperate climate areas with decarbonated soils and high contents of organic matter (Konen et al., 2003; Wills, Burras, Sandor, 2007), nitrogen (Eshani et al., 1999) or phosphorus (Bogrekci & Lee, 2005). In this study, efforts are focused on finding areas homogeneous only with respect to those fundamental variables, such as stoniness, crustability and high carbonate content, leaving out other variables. The few studies in semiarid areas (Sa´nchez-Maran˜o´n, Delgado, Melgosa, Hita, & Delgado, 1997; Sa´nchez-Maran˜o´n et al., 2011; Sa´nchez-Maran˜o´n, Ortega, Miralles, & Soriano, 2007; Sa´nchez-Maran˜o´n, Soriano, Megolsa, Delgado, & Delgado, 2004) support this approach. It is known from previous research conducted in humid areas, that soil colour is strongly conditioned by the presence of chromogen substances, which in turn are indicators of fertility as well as soil condition. However, two main questions remain unanswered. The first question refers to the approach used to study colour-soil relationships. The common procedure consists of
121
studying the influence of each individual chromogen component on soil colour, and developing colour-component regression equations where other variables are assumed to be constant (Viscarra et al., 2008). Furthermore, it should be remembered that the influence of each variable is not independent of the others, and that the soil does not remain unaltered over time, which limits this approach. The second question refers to specific characteristics of arid and semiarid areas. Unlike humid areas, soils in arid and semiarid areas: lack chromogen substances common in humid areas, are rich in substances not present in humid soils, i.e. carbonates and salts, and have very limited range of variability of some components such as soil organic carbon (SOC), N, or P. With regard to colour determination techniques, it is worth noting that the classical approach to communicating soil colour is based on the use of specific Munsell charts under natural illumination conditions (Munsell Color Co., 1994). Colour spaces developed in the early 1930s by the Commission Internationale de l’E´clairage (CIE) are suitable alternatives to visual techniques, i.e. observation of the Munsell charts, to avoid subjectivity-related issues. The use of those spaces in soil science is well documented in the literature (Torrent & Barro´n, 1993; Viscarra, Minasny, Roudier, & McBratney, 2006). Thus, the goal of this paper is to improve the knowledge about the topographic, management and pedologic variables that contribute to soil colour in a semiarid environment, where the presence of chemical chromogen agents is very limited due to weak soil development.
2.
Materials and methods
2.1.
Site description, sampling and processing of data
Soil samples were collected near the river Vinalopo´ in the province of Alicante (southeast Spain). The major soil orders in the area are Aridisols and Entisols (Soil Survey Staff, 2010). A total of 110 topsoil samples (0e15 cm) were collected over an area of 60 km2 with similar climate characteristics, but great diversity in parent material, aspect, elevation, proximity to the river, vegetation, and land management (Fig. 1). The climate is continental Mediterranean with cold winters and hot and dry summers, an annual rainfall of 350.5 mm and an annual average of temperature of 15.2 C. This demonstrates the semiarid character of area, more pronounced during the summer months. The principal types of land use are: crops (60%) with vineyards and fruit trees (olive and almond trees), abandoned fields (19%) and forest (14%) with pines and scrub. Geomorphology is defined by the valley of river Vinalopo´ and the terraces and slopes of two mountain ranges with northeastesouthwest orientation. The mean terrain height is 540 m above sea level, and slopes are less than 15%. The predominant geologic material is quaternary located in the bottom of the valley, whereas limestone, marls, clays and gypsum are present in the slopes (Fig. 2). Samples were first air-dried in the laboratory, crushed and sieved using a brass sieve with 2 mm openings (Soil Survey Staff, 2004). This preprocessing was carried out in order to
122
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
Fig. 1 e Study area with sample points overlaid.
remove uncertainty effects in colour measurements due to water content (Chang, Laird, & Hurburgh, 2005) and particle size of the samples (Barrett, 2002; Sa´nchez-Maran˜o´n et al., 1997; Torrent & Barro´n, 1993). The processed samples were transferred into measurement plates taking up all the available space before colour measurements were made. Each sample was measured three times, shaking the container between measurements. The value that was used in later processing stages was the mean value of the three measurements. The samples were processed and analysed following standard laboratory procedures (Soil Survey Staff, 2004) in order to collect physical, chemical properties and colour coordinates. The laboratory analyses carried out were: carbonate content by Bernard’s calcimeter (CaCO3), electrical conductivity of 1:5 soil:water extract (EC1:5), pH ratio soil:water 1:1, organic carbon by WalkleyeBlack method (SOC), free iron by citrate-dithionite extraction (Fe2O3), nitrogen by Kjeldahl method (TN), phosphorus by Olsen method (TP) and texture (USDA fractions) by Bouyoucos Hydrometer technique. All parameters were analysed according to the existing official methodology. Other variables were obtained by means of geographical information system (GIS) analyses: aspect, elevation and slope; and by observations in field visits: stoniness, land use
(agricultural or forest) and management (irrigation system and conservation techniques). The geographical coordinates collected during field visits were used to create a map of sample points, which allowed georeferencing of variables such as topographic or soil management attributes. The numerical values of these variables were arranged in tables, which were joined to the point map by means of a key identifier. The resulting data set can be divided into four groups of variables (Table 1). In order to determine the relationships between the three colour components and the remaining soil variables, we used a multiple regression model. Qualitative characteristics, i.e. irrigation system, parent material, and land use, were incorporated as explaining variables by means of their corresponding dummy variables (Draper & Smith, 1981). The inevitable existence of collinearity between the explaining variables in the study led us to use the all possible subset regression technique (Jobson, 1991). This computational technique provides a convenient strategy to select subsets of variables which can be further studied in the regression models. The criterion we used to select those subsets was Mallows’ Cp, whose objective is to find models which minimise mean square error in predicted colour observations. This prediction error takes into account both prediction variance and bias. We eventually chose those models
123
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
Fig. 2 e Principal geological materials in area by occupation area: quaternary materials, limestone, clays and gypsum, and marls.
Table 1 e Analytical methods, codes and units. General variables Geomorphologic
Pedologic
Management
Variable
Units
Elevation above sea Slope Aspect Geological material
m % Cardinal points
Calcium carbonate pH Electrical conductivity (1:5) Texture
% pH units dS m1 %Clay (2e0.05 mm) %Sand (0.05e0.002 mm) %Silt (>0.002 mm) % % mg kg1 % %
Stoniness Total Nitrogen Total Phosphorous Organic Carbon Iron Crop
Irrigation Colour
Lightness Chroma Hue
Methodology GIS GIS GIS Geological Map
1e 2e 3e 4e
Quaternary material Clay and gypsum Limestone Marl
1e 2e 3e 4e 1e 2e
Fruit trees Vineyard Forestry Abandoned field No Yes
Bernard Calcimeter pH-meter Conductimeter Bouyoucus
Gravimetric Kjeldhal Olsen WalkleyeBlack Dithionite citrate Observation in field
Observation in field CIELAB
Codification
Trichromatic Colorimeter
124
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
according to pedologic criteria that provided better interpretation of the numerical results.
2.2.
3.
Results and discussion
These procedures lead to a number of relationships which are summarised in this section. We provide statistical relationships between colour and soil variables, together with possible explanations and results from other research. It is necessary to note that these results are not based on experimental designs. This statistical approach has known practical limitations regarding interpretation of the results owing to the collinearity that exists between the explaining variables. Specifically, it is not possible to establish true causeeeffect relationships and the results should be considered as a first approach to the problem. The approach in this study is slightly different from previous research. On the one hand, we base relationships on multiple regression models, contrary to most studies that are based on bivariate regression models. On the other hand we analysed all three colour attributes lightness, hue and chroma, whereas soil colour studies focus mainly on lightness.
Elements of colorimetry
In the context of soil science, colour has always been one of the basic properties reported in soil profile descriptions (Simonson, 1993). The common approach is based on the use of specific Munsell charts under natural illumination. According to the Munsell Soil Color Chart, colour is reported in terms of the so-called hue, value, and chroma. However, we used a different laboratory procedure based on CIE standards and colorimeter measurements (Torrent & Barro´n, 1993; Viscarra et al., 2006). Colour characteristics were determined using a Konica Minolta CS-100A colorimeter which outputs chromaticity coordinates xy and luminance Y in the well-known CIE xyY space (CIE, 2004). Laboratory measurements provided xyY coordinates of both soil samples and the white reference target. Chromaticity coordinates were then transformed into tristimulus values XYZ using well-known formulas (CIE, 2004; Westland & Ripamonti, 2004).
3.1.
X ¼ Y$ðx=yÞ Y¼Y Z ¼ Y$ð1 x yÞ=y
Analytical characteristics
The main analytical characteristics of soils are given in Table 2. In summary, the soils were stony, with coarse texture, nonsaline and moderately alkaline. They were light, with predominant yellow and brown hues, and the carbonate content was high, whereas they were low in organic matter and several essential nutrient elements for plant growth, namely Fe, N, and P.
Next, tristimulus values were converted into CIELAB coordinates: 1=3 16 L ¼ 116$ðY=Y nÞ i h a ¼ 500$ ðX=Xn Þ1=3 ðY=Yn Þ1=3 h i b ¼ 500$ ðY=Yn Þ1=3 ðZ=Zn Þ1=3
3.2.
where Xn, Yn, Zn are the tristimulus values of the white reference. It should be noted that there are alternative formulas for CIELAB coordinates computation which should be used when either (X/Xn)1/3, (Y/Yn)1/3, or (Z/Zn)1/3 is less than (6/29)3 (CIE, 2004). Finally, CIELAB coordinates were transformed into the CIE L*C*h representation:
Model selection
A number of regression models were selected on the basis of Mallows’ Cp, following the same procedure for each colour component. With regard to lightness, six models with low values of Mallow’s Cp (values between 3.4 and 3.9) were selected which involved up to 10 variables: stoniness, carbonate content, SOC, TN, iron oxides, sand, aspect, EC, clay and vineyard (Table 3). Seven models were selected for hue with Mallow’s Cp values between 1.0 and 1.9 which include 9 variables as possible indicators of soil hue: stoniness, TN, clay, aspect, sand, marls, irrigation, SOC and phosphorus are the selected variables (Table 3).
h i1=2 2 Cab ¼ ðaÞ2 ðbÞ hab ¼ arctanðb=aÞ The three colour attributes are lightness (L*), chroma (C*), and hue angle (h*), which represent psychophysical correlates of human perception mechanisms.
Table 2 e Statistical summary of chemical, physical and geomorphologic results (n [ 110). CE
pH
1
% Mean Median Std. dev. Variance Maximum Minimum
39 30 31.4 983.9 100 0
EC1:5 dS m
8.3 8.3 0.3 0.1 8.9 7.4
0.6 0.2 0.8 0.6 2.7 0.2
CaCO3 TN Fe2O3 SOC %
%
%
%
30.7 29 14.1 198.9 92.9 6.9
0.1 0.1 0 0 0.2 0
0.5 0.3 0.6 0.3 3 0
1.01 0.99 0.35 1.2 1.58 0
TP mg kg 0.5 0.3 0.7 0.5 4.9 0
Clay 1
Silt
Sand
El
Sl
As
%
%
%
m
%
16.1 14 9.2 84.2 43 3
25.3 23 14.7 216.7 73 2
58.5 59 17.2 295.6 88 19
540.6 533.3 39.1 1525.4 678 479.5
3.5 2.9 2.5 6.2 13.5 0.6
227.5 223.3 46.2 2133.8 356.6 96.7
L*
h*ab
C*ab
Cielab coord. 56.9 56.8 7 48.5 71.3 41.6
68.5 67.9 5.5 29.9 88.6 37.5
17.3 17.4 3.6 12.7 27.8 7.3
CE, stoniness; EC1:5, electrical conductivity 1:5; CaCO3, carbonate content; TN, total nitrogen; Fe2O3, free iron; SOC, soil organic carbon; TP, total phosphorus; El, elevation; Sl, slope; As, aspect; L*, lightness; h*ab, hue; C*ab, chroma.
125
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
Table 3 e Mallows’ Cp values for lightness, hue and chroma regression models. Variable Mallows Cp L*
h*ab
C*ab
7 8 8 8 8 9 6 6 7 7 8 8 9 8 8 9 9 9 9 10
3.9 3.5 3.6 3.7 3.8 3.4 1.6 1.7 1 1.9 1.5 1.6 1.8 3.6 3.7 3.7 3.7 4.2 4.2 4.2
Standard CE EC1:5 CaCO3 TN SOC TP Fe2O3 Clay El Sl As Sand Crop Irr Marls deviation 5.26 5.22 5.22 5.22 5.23 5.19 4.5 4.5 4.5 4.48 4.44 4.45 4.43 2.79 2.79 2.77 2.77 2.78 2.78 2.76
x x x x x x x x x x x x x
x x x x x x
x
x x
x x x x
x
x x x x x x x x x x x x x x x x x x x x
x x x x x x x x x x x x x x x x x x x x
x x x x x x
x x x x x x x x x
x x x
x x x x x x x x x x x x x x x
x x x x x x x x x x x x x x x x x x
x x x x x x x
x x x x x x x x x x x x x x x x x x x x
x
x x x x x
x
x x x x x x x x x x x x x
Irr, Irrigation; other abbreviations in Table 2.
Regarding chroma, seven models with the lower values of Mallow’s Cp (values between 3.6 and 4.2) were tested. The 12 variables related to soil chroma are TN, SOC, phosphorus, clay, elevation, slope, sand, marls, stoniness, EC and irrigation (Table 3).
3.3. Relationship between lightness (L*) and soil characteristics There are five significant variables related to L*, two of them were positively correlated, and three were negatively correlated (Table 4). In this context, positively correlated means that L* has greater values as the corresponding explanatory variable increases, i.e. soil colour is lighter. The value of the determination coefficient (R2) was 49.4%.
Some of the relationships between variables were expected on the basis of research in other environments. Soil lightness is affected by texture because silt and clay particles reflect more visible light than sand particles, resulting in lighter soil (Spielvogel, Knicker, & Ko¨gel-Knabner, 2004). Moreover, sand particles have less specific surface area than those of either silt or clay, which in turn have a colloidal nature. Therefore, sand particles can be easily coated with organic matter, resulting in darker soils. In connection with particle size issues, it should be noted that sample processing itself has known effects on lightness. In this study different sieve sizes were tested (Fig. 3). A 2 mm sieve size was chosen since it allows preparation of more homogeneous samples while preserving the effects of chromogen elements. It is important to use the same size consistently across the experiment.
Table 4 e Regression analysis for Lightness (L*) Chroma (C*ab) and Hue (h*ab).
2
R Constant CaCO3 Sand Fe2O3 TN EC1:5 SOC Clay Marls
Lightness
Hue
Chroma
Estimate
Estimate
Estimate
49.4 65.28 0.094* 0.11* 2.08** 46.83** 0.0021** e e e
32.4 84.94 e 0.22*** e 34.45* e 0.36* 0.24*** e
36.7 13.76 e 0.13*** e e e 0.28*** 0.09* 2.5*
[*P < 0.05; **P < 0.01; ***P < 0.001]; Estimate ¼ Estimated values of regression model coefficients; Constant ¼ independent term of regression equation.
Fig. 3 e Effect of particle size on luminance.
126
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
Samples with higher carbonate contents had lighter colours, as in the research of Sa´nchez-Maran˜o´n et al. (1997) who studied crushed soil samples from climatic environments similar to ours, with low carbonate and organic matter contents. Spielvogel et al. (2004) studied fine textured Mollisols with higher organic matter and low carbonate contents, which are different from the soils used here, and obtained the same behaviour. It is well known that CaCO3, has considerable influence on soil lightness (Simonson, 1993). Limestone and marls contain large amounts of carbonates and are the parent materials of many of the soils studied for this paper. Therefore, a positive relationship between CaCO3 and lightness was expected. Spielvogel et al. (2004) found higher values of L* in calcareous soils (L* > 56.1) than in non-calcareous soils (L* < 55.3). It should be noted that their calcareous soils were mainly clayey and silty, whereas those in this study are sandy (Table 2). Iron oxide content also influences soil colour, mainly in the hue component (Schwertmann, 1993). Sa´nchez-Maran˜o´n et al. (1997) did not find significant influence of Fe2O3 on lightness either in crushed samples or in aggregates of red Mediterranean soils. However, we found a significant relationship between lightness and hue angle (correlation 0.58, P < 0.01), which may be indirectly associated with the influence of iron oxides on lightness. It can be observed from the data that redder soils with lower hue angles are darker than white or yellower soils with larger hue angles, that is, reddish soils are darker than yellowish soils in the present study area (note that a hue angle of 0 represents red and 90 represents yellow in CIELAB space). With regard to total nitrogen (TN), the results show a decrease in lightness as nitrogen content increases. These results agree with those of Qian, Klinka, and Lavkulich (1993) in forest mineral soils from Canada, despite the higher contents of TN in their soils. There are also obvious climate differences between the study areas. Unlike other soil components, the identification of nitrogen content from soil colour has not been greatly studied in the past. This trend has changed with the advent of site-specific management which requires precise amounts of fertilizer to be applied to crops in different parts of fields. The study of this component as a determinant of topsoil colour could be particularly important in arid and semiarid areas since it is common to find abandoned agricultural fields which have been colonised by leguminous plants. In the present study, nitrogen contents were higher in abandoned fields than in cultivated lands (fruit trees and vineyards). This may well be a consequence of nitrogen fixation carried out by leguminous symbiotic microorganisms. In order to gain more insight into this subject, we specifically studied TN contents under four different land use classes (fruit trees, vineyards, forests, and abandoned fields) by means of analysis of variance. The analysis of variance shows significant differences with land use (P ¼ 0.002) as depicted by Tukey’s HSD intervals (Fig. 4). In the present study, higher electrical conductivities are associated with lighter soils, which can be used to determine the presence of salts. In areas where evaporation largely exceeds precipitation there is no leaching, and sometimes white crusts are visible on the soil surface. White saline crusts
Fig. 4 e Tukey’s HSD intervals from the analysis of TN and vegetation and land use (1: Fruit trees [n [ 54]; 2: vineyard [n [ 13]; 3: forestry [n [ 19]; 4: abandoned field [n [ 24]).
mainly consist of sodium and magnesium chlorates and sulphates (Valentin, 1993), and soils with such crusts are usually affected by wind erosion as they dry (Aubert, 1976). From this viewpoint, soil colour could be considered as a good indicator of desertification in areas with an abundance of salts, either of geologic origin or derived from inadequate water management in agriculture (Shepherd & Walsh, 2002). Clay content, SOC, stoniness and aspect are soil variables which appeared in the initial models given by the all possible subsets regression analysis. However, they did not exhibit significant relationships to lightness and were not included in the final models. The existing literature on soil colour always reports significant correlations between clay content and lightness (positive sign), as well as between SOC and lightness (negative sign). Most of this research studied the influence of these variables on soil colour individually, without taking into account interactions with other soil properties. Bivariate linear regressions were computed to determine whether or not our data exhibited such behaviour. Due to the nature of our data (limited range of values for some soil attributes, and the large extent and heterogeneity of the study area) correlation coefficients can reflect weak relationships (Schulze et al., 1993). Table 5 shows the trends of lightness with respect to SOC and clay, and also includes TN due to its great effect on crop production, which makes it a variable of interest in the development of predictive equations on which to base precision agriculture approaches. SOC showed a negative relationship (P < 0.0001), however multivariate analysis showed a positive relationship of lightness with organic carbon (P ¼ 0.06), which may appear
Table 5 e Linear regression analysis for Lightness (L). Linear regression SOC Clay TN
Estimate
R2
4.06*** 0.03 1.63
0.38 0.04 0.14
*P < 0.05; **P < 0.01; ***P < 0.001; Estimate, estimated values of regression model coefficients.
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
surprising at first glance. The origin and texture of the samples can be of critical importance in this regard. Some authors suggest that the relationships between lightness and SOC depend to a large extent on soil texture and its homogeneity (Spielvogel et al., 2004). Red clay materials were found in this study, but the number of clayey samples was low in relation to the total number of analysed samples. Moreover, carbonate content and texture affect the relationship between soil organic carbon content and soil lightness. Spielvogel et al. (2004) found that in silty and clayey soils with similar organic carbon contents, carbonate-free samples were darker than calcareous samples. The soils in the present study had high contents of carbonates which can mask the influence of either organic matter or clay on lightness. Accordingly, the fact that TN is not significant in the bivariate analysis suggests interactions between this variable and land use. Stoniness and aspect were the other two soil variables which appeared in the initial models (Table 3) and were not included in the final models. Stoniness produces light colours, and a positive relationship was found between lightness and aspect in south-oriented samples as well as in north-oriented samples.
3.4.
Hue (hab)
There are four significant variables related to hue, one positively correlated, and three negatively correlated (Table 4). When dealing with hue in CIELAB space, positively correlated means that soil colour becomes yellowish, and negatively correlated means that soil colour becomes reddish as the explanatory variable increases. The value of the determination coefficient (R2) was 32.4%. Much of the scatter is probably associated with differences in soil texture and in material origin, orientation and other pedologic variables in the samples (Schulze et al., 1993). Sand and clay are negatively correlated, which suggests that high contents of sand and clay produce lower values of hue (in this context this means that soil colour approaches to the red colour.) This agrees with hue angle measurements of Sa´nchez-Maran˜o´n et al. (2011) on individual particles of Regosols and Umbrisols from southeast Spain. The negative relationship between sandy samples and hue may occur because the particles in the sand fraction have been weathered. Williams and Yaalon (1977) obtained reddish sandy particles in the laboratory by the alteration of iron-rich heavy minerals and the precipitation of iron on surrounding quartz grains. In the case of clayey soils, more clayey samples have lower hues. In the present study area the detrital fine particle rocks are red clays (Fig. 2). SOC and TN also have an effect on soil hue. The chromogen nature of SOC component causes different colorations in topsoil depending on its content (Schulze et al., 1993). These authors highlighted the influence exerted by organic matter on hue, value, and chroma, being specifically relevant the composition of the organic matter. In the soils for this study, higher contents of SOC give reddish hues, whilst greater TN gives yellowish hues. Marls and stoniness are two soil variables which appeared in the initial models given by the all possible subsets regression analysis. However, they did not exhibit significant
127
relationships to hue (P ¼ 0.097 and P ¼ 0.07 respectively) and were not included in the final models. In general terms, topsoil developed on limestone or marls have yellower colorations than those developed on quaternary, clay, or gypseous materials. Similarly, less stony topsoils are redder than stony topsoils. In both cases, soil colours are determined by the presence or absence of carbonates in parent materials (Simonson, 1993). The relationships found by Sa´nchez-Maran˜o´n et al. (1997) in red soils of Spain and Italy do not fully agree with the results presented here. They found relationships of hue and iron in processed samples, and horizon type (A or B), elevation and clay in aggregates.
3.5.
Chroma (C*ab)
There are four significant variables related to chroma. Marls and SOC are negatively correlated, whereas sand and clay content are positively correlated (Table 4). The value of R2 is 36.7%. It seems that the colour of samples located on white marls is less saturated than for other soils, which is in agreement with the expected result. Organic matter content has the same effect. Results obtained by Konen et al. (2003) in mollic epipedions are similar to the results presented here. On the other hand, finer textured soils have more saturated colours. This may be due to the red colours of fine grained sedimentary rocks as a consequence of high contents of iron in the oxidised state. Likewise, sandy samples have larger chromas. Sa´nchezMaran˜o´n et al. (2011) found similar results in sandy particles obtained by dispersion from non-calcareous materials. Chromas were higher for particle sizes under 0.05 mm than those for the 1e2 mm fraction. This relationship is possible when the particles in the sand fraction have been weathered, much like in the case of hue (Williams & Yaalon, 1977). We found similarities in hue and chroma components which may suggest some sort of interaction between these two variables. Such interaction seems to be supported by the fact that some definitions of chroma rely on hue, and that neutral colours, i.e. colours without hue, must have a chroma of zero. The issue of the hue chroma interaction has been reported previously in the colour difference literature (CIE, 2004) as well as in more specific soil science literature (Buntley & Westin, 1965). In the latter case, numerical functions of Munsell hue and chroma are used as indices of soil development. The bivariate regression analysis of hue and chroma seems to support this (R2 ¼ 65%).
4.
Conclusions
All of the pedologic variables, except stoniness, pH and phosphorus, exhibited significant relationships with some of the three colour attributes. None of topographic and management variables had significant influence on soil colour. It seems that there exists considerable interaction between hue and chroma (R2 ¼ 65%) and it should be noted that determination coefficients (R2) between soil variables and soil colour are always less than 50%, which is usual in studies
128
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
where observational data are not subjected to experimental design. There are six variables that have an effect on the three soil colour components. Sand content is the only significant variable for all three colour attributes. Nitrogen explains lightness and hue, and clay and SOC explain chroma and hue. Carbonate, iron and EC are significant in the lightness models, whereas marls are significant in the chroma models. Soil texture, nitrogen and soil organic matter content show a remarkable relationship with lightness. Experience shows important interactions between SOC, clay, iron and carbonate content, as well as between TN and land use. The significant correlations found between soil colour and several soil variables demonstrate the validity of this approach, although further studies are needed. This would help in obtaining predictive equations which may be useful in precision agriculture and other engineering applications. From the results of this study, colour attributes could be suitable to predict other soil characteristics (for instance SOC content) provided that land managers are careful when using predictive equations derived from data analyses. In arid and semiarid environments with irregular relief, it is quite common that close locations, even at the individual plot level, exhibit high spatial variability in topsoil texture and chemical composition. In these cases, it may be necessary to have a specific equation for each different homogeneous soil unit.
references
Aubert, G. (1976). La morphologie des sols affecte´s par le sel, reconnaissance et pre´vision, surveillance continue. In FAO soils bulletin: Vol. 31. Prognosis of salinity and alkalinity (pp. 187e194). Roma. Barrett, L. R. (2002). Spectrophotometric color measurement in situ in well drained sandy soils. Geoderma, 108, 49e77. Bogrekci, I., & Lee, W. S. (2005). Spectral phosphorus mapping using diffuse reflectance of soils and grass. Biosystems Engineering, 91, 305e312. Brady, N., & Weil, R. R. (2006). The nature and properties of soils. New Jersey: Prentice Hall. Buntley, G. J., & Westin, F. C. (1965). A comparative study of developmental color in a ChestnuteChernozemeBrunizem soil climosequence. Soil Science Society Proceedings, 29, 579e582. Chang, C.-W., Laird, D. A., & Hurburgh, G. R. (2005). Influence of soil moisture on near infrared reflectance spectroscopic measurement of soil properties. Soil Science, 170, 244e255. CIE. (2004). Colorimetry e Technical report. Vienna: Commission Internationale de l’E´clairage. Christensen, L. K., Bennedsen, B. S., Jørgensen, R. N., & Nielsen, H. (2004). Modelling nitrogen and phosphorus content at early growth Barley using Hyperspectral Line Scanning. Biosystems Engineering, 88, 19e24. Doi, R., Wachrinrat, C., Teejuntuk, S., Sakurai, K., & Sahunalu, P. (2010). Semiquantitative color profiling of soils over a land degradation gradient in Sakaerat, Thailand. Environmental Monitoring Assessment, 170, 301e309. Draper, N., & Smith, H. (1981). Applied regression analysis. New York: Wiley. Eshani, M. R., Upadhyaya, S. K., Slaugther, D., Shafii, S., & Pelletier, M. (1999). A NIR technique for rapid determination of soil mineral nitrogen. Precision Agriculture, 1, 217e234.
He, X., Vepraskas, M. J., Lindbo, D. L., & Skaggs, R. W. (2003). A method to predict soil saturation frequency and duration from soil color. Soil Science Society of America Journal, 67, 961e969. Ibarra-F, F. A., Cox, J. R., Martin-R, M. H., Crowl, T. A., Post, D. F., Miller, R. W., et al. (1995). Relationships between buffelgrass survival, organic carbon and soil color in Mexico. Soil Science Society of America Journal, 59, 1120e1125. Jobson, J. D. (1991). Applied multivariate data analysis. Volume I: Regression and experimental design. New York: Springer-Verlag. Konen, M. E., Burras, C. L., & Sandor, J. A. (2003). Organic carbon, texture and quantitative color measurement relationships for cultivated soils in North Central Iowa. Soil Science Society of America Journal, 67, 1823e1830. Mouazen, A. M., Karoui, R., Deckers, J., De Baerdemaeker, J., & Ramon, H. (2007). Potential of visible and near-infrared spectroscopy to derive colour groups utilising the Munsell soil colour charts. Biosystems Engineering, 97, 131e143. Munsell Color Co.. (1994). Munsell Soil Color Charts. New York: Munsell Color Company. Qian, H., Klinka, K., & Lavkulich, L. M. (1993). Relationships between color value and nitrogen in forest mineral soils. Canadian Journal of Soil Science, 73, 61e72. Sa´nchez-Maran˜o´n, M., Delgado, G., Melgosa, M., Hita, E., & Delgado, R. (1997). CIELAB color parameters and their relationships to soil characteristics in Mediterranean red soils. Soil Science, 162(11), 833e842. Sa´nchez-Maran˜o´n, M., Martı´n-Garcı´a, J. M., & Delgado, R. (2011). Effects of the fabric on the relationship between aggregate stability and color in a RegosoleUmbrisol soilscape. Geoderma, 162, 86e95. Sa´nchez-Maran˜o´n, M., Ortega, R. B., Miralles, I., & Soriano, M. (2007). Estimating the mass wetness of Spanish arid soils from lightness measurements. Geoderma, 141, 397e406. Sa´nchez-Maran˜o´n, M., Soriano, M., Megolsa, M., Delgado, G., & Delgado, R. (2004). Quantifying the effects of aggregation, particle size and components on the colour of Mediterranean soils. European Journal of Soil Science, 55, 551e565. Schulze, D. G., Nagel, J. L., Van Scoyoc, G. E., Henderson, T. L., & Baumgardner, M. F. (1993). Significance of organic matter in determining soil colors. In J. M. Bigham, & E. J. Ciolkosz (Eds.), Soil color, SSSA Special Publication No 31 (pp. 71e90). Madison: Soil Society of America. Schwertmann, U. (1993). Relations between iron oxides, soil color and soil formation. In J. M. Bigham, & E. J. Ciolkosz (Eds.), Soil color, SSSA Special Publication No 31 (pp. 51e70). Madison: Soil Society of America. Shepherd, K. D., & Walsh, M. G. (2002). Development of reflectance spectral libraries for characterization of soil properties. Soil Science Society of America Journal, 66, 988e998. Simonson, R. W. (1993). Soil color standards and terms for field-use. History of their development. In J. M. Bigham, & E. J. Ciolkosz (Eds.), Soil color, SSSA Special Publication No 31 (pp. 1e20). Madison: Soil Society of America. Soil Survey Staff. (2004). Soil survey laboratory methods manual e Soil survey investigations report no. 42. Burt. R. U.S. Department of Agriculture, Natural Resources Conservation Service. Soil Survey Staff. (2010). Keys to soil taxonomy (11th ed.). Washington, DC: USDA-Natural Resources Conservation Service. Spielvogel, S., Knicker, H., & Ko¨gel-Knabner, I. (2004). Soil organic matter composition and soil lightness. Journal of Plant Nutrition and Soil Science, 167, 545e555. Torrent, J., & Barro´n, V. (1993). Laboratory measurement of soil color: theory and practice. In J. M. Bigham, & E. J. Ciolkosz (Eds.), Soil color, SSSA Special Publication No 31 (pp. 21e34). Madison: Soil Society of America.
b i o s y s t e m s e n g i n e e r i n g 1 1 6 ( 2 0 1 3 ) 1 2 0 e1 2 9
Torrent, J., & Barro´n, V. (2003). The visible diffuse reflectance spectrum in relation to the color and crystal properties of hematite. Clays and Clay Minerals, 51, 309e317. Valentin, C. (1993). Soil crusting and sealing in West Africa and possible approaches to improved management. In FAO soils bulletin: Vol. 69. Soil tillage in Africa: Needs and challenges. Viscarra, R. A., Fouad, Y., & Walter, C. (2008). Using a digital camera to measure soil organic carbon and iron contents. Biosystems Engineering, 100, 149e159. Viscarra, R., McKenzie, N., & Grundy, M. (2010). Using proximal soil sensors for digital soil mapping. In J. L. Boettinger, D. W. Howell, A. C. Moore, A. E. Hartemink, & S. Kienast-
129
Brown (Eds.), Digital soil mapping: Bridging research, production and environmental application and operation (pp. 79e92). London e New York: Springer. Viscarra, R. A., Minasny, B., Roudier, P., & McBratney, A. B. (2006). Color spaces for soil science. Geoderma, 133, 320e377. Westland, S., & Ripamonti, C. (2004). Computational color science using MATLAB. Chichester: John Wiley & Sons. Williams, C., & Yaalon, D. H. (1977). An experimental investigation of reddening in dune sand. Geoderma, 17, 181e191. Wills, S. A., Burras, C. L., & Sandor, J. A. (2007). Prediction of soil organic carbon content using field and laboratory measurements of soil color. Soil Science Society of America Journal, 71, 380e388.