A standardized soil quality index for diverse field conditions

Science of the Total Environment 541 (2016) 424–434

Contents lists available at ScienceDirect

Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv

A standardized soil quality index for diverse field conditions Vincent de Paul Obade ⁎, Rattan Lal The Ohio State University, Carbon Management and Sequestration Center, School of Environment and Natural Resources, 2021 Coffey Road, Columbus, OH, USA

H I G H L I G H T S

G R A P H I C A L

A B S T R A C T

• Tested 4 methods for constructing a Soil Quality Index (SQI) • Partial Least Squares Regression (PLSR) method optimal for constructing SQI • Identified significant soil attributes influencing site specific soil quality • New SQI highly correlates with crop yields.

a r t i c l e

i n f o

Article history: Received 10 August 2015 Received in revised form 18 September 2015 Accepted 18 September 2015 Available online xxxx Editor: D. Barcelo Keywords: Land management Minimum dataset Soil properties Soil quality index

a b s t r a c t Understanding the nexus between soil quality and productivity is constrained by data artifacts, compounded by limitations of the existing models. Here, we explore the potential of 4 regression methods (i.e., Reduced Regression (RR), SIMPLS, Principal Component Regression (PCR), and Partial Least Squares Regression (PLSR)), to synthesize 10 soil physical and chemical properties acquired from 3 major management practices and different soil layers, into an unbiased soil quality index (SQI) capable of evaluating soil functions (e.g., biomass production). The data was acquired from privately owned fields within the state of Ohio, USA, at the following land use and management sites: natural vegetation (NV) or woodlands, conventional till (CT), and no-till (NT). The soils were sampled at similar landscape positions (i.e., summit) at depth intervals of 0–10, 10–20, 20–40 and 40–60 cm, and analyzed for bulk density (ρb), carbon/nitrogen (C/N) ratio, soil organic C (SOC), total N (TN), available water capacity (AWC), pH and electrical conductivity (EC). Preliminary analyses revealed the PLSR method as the most robust. The PLSR Variable Importance of Projection (VIP) was calculated, transformed into the SQI score and compared with yield data. SOC, ρb, C/N and EC were identified as the major variables influencing soil quality status. The data shows that the quality of Pewamo silty clay loam (Pw) soil was higher than Crosby Celina loams (CtA), Kibbie fine sandy loam (kbA), Glynwood silt loam (GWA) and Crosby silt loam (CrA), respectively. In 2012, the mean SQI was 42.9%, with corn and soybean yields of 7 and 2 Mg/ha. The R2 of SQI versus yield was 0.74 for corn (Zea mays L.), and 0.89 for soybean (Glycine max (L.) Merr.). Future studies will investigate techniques for mapping this SQI. © 2015 Elsevier B.V. All rights reserved.

⁎ Corresponding author. E-mail addresses: [email protected] (V. de Paul Obade), [email protected] (R. Lal).

http://dx.doi.org/10.1016/j.scitotenv.2015.09.096 0048-9697/© 2015 Elsevier B.V. All rights reserved.

V. de Paul Obade, R. Lal / Science of the Total Environment 541 (2016) 424–434

1. Introduction Sustainable agro-ecosystem management requires tools that provide discernible soil quality status information. The variability in soil quality influences biogeochemical cycling, weather patterns, biodiversity, agricultural productivity, and thus food, energy and water security; socio-economic viability and human-wellbeing (de la Paz Jimenez et al., 2002; Doran et al., 1996; Lal, 2009; Ohlson, 2014). Unlike air and water quality, legislation and policy on soil quality are scantily defined. This deficiency in legislation may be attributed to the fuzzy definition of soil quality, accentuated by the inherent difficulty in quantifying and mapping its spatial variability. Notwithstanding the importance of soil quality information, many models depicting global issues related to climate change or to food, water and energy security embody limited soil expertise (McBratney et al., 2014). Quantifying soil quality changes visà-vis soil functions are construed by data artifacts and lack of data, absence of clear reference points or baseline values, model omissions and validation challenges (Andrews and Carroll, 2001; Andrews et al., 2003; McBratney et al., 2014). Moreover, because soil has many functions (e.g., pollution control, biomass production, climate regulation etc.), simply measuring a single or specific soil property to infer soil quality is insufficient. Incoherence in soil quality information can result in disasters such as landslides, disease outbreaks from contaminated water etc., not being detected on time. Accurate, repeatable, systematic, and transparent soil quality measurements can enhance interpretation and comparability between sites (Andrews and Carroll, 2001; de Paul Obade and Lal, 2014). Soil quality entails the capacity of a specific kind of soil to function and sustain plant and animal survival within natural or managed boundaries, without jeopardizing the environmental quality (Andrews et al., 2004; Doran and Zeiss, 2000; NRCS, 2012). Soil quality cannot be directly determined, but can be inferred by measuring soil physical, chemical and biological properties. Practically, soil quality information is gauged by either: (i) the soil test kit and laboratory-based analyses (Wienhold et al., 2004), or (ii) the Munsell soil color chart (Gobin et al., 2000; Staff, 1951), or (iii) remote sensing (Cohen et al., 2007; Minasny and Hartemink, 2011). However, the soil test kit, is not only cumbersome, but also relies on extracting powders which may dissolve poorly with the soil leading to inaccurate results. Furthermore, the soil test kit does not determine the conjoined assessment of soil properties, but measures only specific soil constituents, such as nitrate-N, P2O5, K2O and pH etc. Utilizing specific soil properties to gauge soil quality vis-àvis biomass production, is non-comprehensive and subjective because of the concept that plants intake different nutrients simultaneously at varying environmental gradients (Ohlson, 2014). Besides, laboratorybased methods disturb the soil; can be expensive and time-consuming especially for analyses done over large spatial extents, which require substantial data inputs. Antithetically, the Munsell color chart is subjective, and depends on the human visual perception. With the Munsell color chart, darker soils are assumed to have a higher soil organic matter, therefore perceived to be of a higher quality (McBratney et al., 2002; Shepherd and Walsh, 2002). Although remote sensing is a nondestructive technique that continuously acquires field data even at inaccessible locations, its flaws include: (a) spectral ambiguity which lower the signal-to-noise ratio (SNR) attributed to signal attenuation, adjacency effects, and atmospheric scattering, (b) rigorous data processing and calibration, (c) mismatches between spatial, spectral and temporal resolution, (d) absence of long-term data (e.g., Landsat data archived from 1972) (de Paul Obade et al., 2013). Soil properties (e.g., SOC, moisture) have been predicted by field remote sensors scanning within the visible and Near Infra-Red (NIR) spectrum (Gogé et al., 2014; Kinoshita et al., 2012; Marín-González et al., 2013). Soil organic Carbon (SOC) concentration is considered a proxy of soil quality because it optimally typifies soil biota dynamics and plays a key role in fertility, soil water availability and aggregate stability in croplands (de Moraes Sá et al., 2013; McBratney et al., 2014; Stockmann

425

et al., 2013). SOC can be predicted and mapped using regression models, geostatistics, or by pedotransfer functions (PTFs). PTFs translate measured soil attributes into estimates of unmeasured variables, whereas geostatistics predicts unsampled points based on the distance and degree of variation between sample pairs of adjacent measured points using a variogram. Accuracy in geostatistics depends on sampling density of field data. Researchers postulate that soil property information can be gleaned from environmental covariates, abbreviated as scorpan factors, comprising (1) s: soil, other or previously measured attributes of the soil at a point location; (2) c: climate, climatic properties of the environment at the location; (3) o: organisms, including land cover and natural vegetation; (4) r: topography, including terrain classes; (5) p: parent material, including lithology; (6) a: age, the time factor; (7) n: geographic position (Grinand et al., 2008; Lacoste et al., 2014; McBratney et al., 2002). Thus, the challenge remains how to construct indicators of soil quality that incorporates both qualitative and quantitative information? Soil Quality Indices (SQIs) synthesize soil attributes into a format that enhances the understanding of soil processes to inform on appropriate management or policy interventions (Boote et al., 1996; Wienhold et al., 2004; Wienhold et al., 2009). Examples of soil attributes include the organic matter (OM) content and stock, bulk density (ρ), respiration rate, soil depth, electrical conductivity (EC), pH etc. Fig. 1 depicts a typical SQI paradigm, encompassing emerging issues on soil quality assessment, tenets for a robust SQI and inherent limitations (Andrews and Carroll, 2001; Karlen et al., 1997; Nortcliff, 2002). The “scoring function” concept is applied in SQIs to decipher the interconnection between soil properties, soil processes (e.g., mineralization), management systems and social perspectives (Andrews et al., 2002a; b; Andrews et al., 2002b; Karlen et al., 1994a; Karlen et al., 1994b; Wienhold et al., 2004). The Soil Management Assessment Framework (SMAF) is an example of a score-based indicator that operates in two synergistic steps: (i) indicator selection and interpretation, and (ii) aggregation (Andrews et al., 2004). The indicator selection and interpretation process entails the transformation of measured, or observed data such as soil nutrients or contaminant concentration, into unitless indicator scores; whereas “aggregation” step combines the individual indicator scores into a single value (Karlen et al., 2008). Despite its prominence as an emerging research domain, quantifying soil biota is not a practical undertaking due to challenges such as: (i) inaccuracies in earthworm counts (i.e., by hand), (ii) difficulty in accounting microbial species diversity, and (iii) difficulty in interpreting the soil respiration tests (Arshad and Martin, 2002; de Paul Obade and Lal, 2014; McBratney et al., 2014). The question then is how to synthesize and transform soil property information sampled from diverse landscapes into a versatile SQI? To enhance the fidelity of SQIs requires credible information acquired through baseline data and accurate models. Besides, science based techniques are required for establishing a minimum data set (MDS) consisting of critical soil variables (Andrews and Carroll, 2001; Yemefack et al., 2006). Although all models can be considered deficient, some can be useful (Box George and Draper, 1987). The critical question in constructing models is whether: (i) new knowledge can be garnered, or (ii) this knowledge can improve human wellbeing and the overall environmental quality. In essence, statistical models for evaluating trends in complex data are either parametric or non-parametric. Unlike parametric, non- parametric statistical methods being parsimonious, do not require the following assumptions to hold true: (i) independence of observation, (ii) homoscedasticity, and (iii) normally distributed errors (Chong and Jun, 2005; Mehmood et al., 2012). Thus, this study explores the potential of 4 non-parametric methods to develop a new SQI computed by aggregating soil attributes under different management and soil layers that: (i) objectively identifies the minimum data set (MDS) consisting of key soil variables, (ii) investigates the contribution of land use/management on soil quality, and (iii) rates soil quality vis-àvis crop yields.

426

V. de Paul Obade, R. Lal / Science of the Total Environment 541 (2016) 424–434

Fig. 1. Soil quality index (SQI) components and attributes.

2. Materials and methods A total of 204 soil samples were obtained between April to early May, 2012 from the following privately owned field sites located in Ohio, USA: Miami (40° 10′ 12″ N, 84°07′ 41.7″ W), Seneca site 1 (41° 00′ 25″ N, 85°16′ 21″ W), Seneca site 2 (41° 12′ 43″ N, 82°54′ 39″ W), Preble (39° 46′ 09″ N, 84°36′ 52″ W and & 39° 41′ 45″ N, 84°40′ 36″ W), and Auglaize (40° 27′ 34.5″ N, 84°26′ 14.8″ W). The soil types were CrA (Crosby silt loam; taxonomic class: fine, mixed, active, mesic Aeric Epiaqualfs), kbA (Kibbie fine sandy loam; taxonomic class: fineloamy, mixed, active, mesic Aquollic Hapludalfs), GWA (Glynwood silt loam; taxonomic class: fine, illitic, mesic Aquic Hapludalfs), CtA (Crosby Celina loams; taxonomic class; fine, mixed, active, mesic Aquic Hapludalfs), and Pw (Pewamo silty clay loam: taxonomic class; fine, mixed, active, mesic Typic Argiaquolls). A brief description of these soil types is available at the web soil survey (http://websoilsurvey. nrcs.usda.gov/app/WebSoilSurvey.aspx). The total annual precipitation in Ohio averages between 100 to 120 cm, whereas the temperature varies from 8.1 to 10.7 °C (DeForest et al., 2012). In the year 2012, drought was experienced in the Midwestern regions of the United States (Lal et al., 2012), resulting to a reduction in total rainfall by between 15 to 35 cm (http://www.ncdc.noaa.gov/sotc/drought/2012/ 8#MRCC). The management and land use practices were no-till (NT) with or without manure (M) and cover crops (cc), natural vegetation (NV i.e., forest), and conventional tillage (CT) (Table 1). The CT fields at Miami, Seneca, and Preble were chisel plowed to 20–25 cm depth, except for the Auglaize site which was disked. The surface residue cover was b 10% in CT managed fields compared with NT which was N20%. The NT management was practiced for over 5 years in sections of the 4 sites, except Preble site which had over 25 years NT field partitions. In each of the 5 sites, three replicates of undisturbed core (54 mm diameter and 60 mm length) and bulk soil samples were sampled from similar landscape positions (i.e., summit), at 0–10, 10–20, 20–40, 40–60 cm depth increments. The USDA-NIFA project guidelines (project web site: sustainablecorn.org) was the protocol used in the laboratory analyses (Kladivko et al., 2014). Soil ρb was assayed using core method, and without stones. Soil moisture content was determined gravimetrically by oven drying a fraction of the soil at 105 °C (Topp and Ferre, 2002), and

the water retention determined using a combination of a tension table (Blanco-Canqui and Lal, 2007; Clement, 1996), and the pressure plate extractors (Klute, 1986; Klute and Dirksen, 1986). The available water capacity (AWC) was computed from the difference in volumetric water content at field capacity (FC) (− 33 kPa), and at permanent wilting point (PWP) (− 1500 kPa) (Dane and Hopmans, 2002; Jemai et al., 2013). This study assumed soil texture to be a fixed soil property that is not influenced by management or even climate (Askari and Holden, 2014; Bonfante and Bouma, 2015). The chemical analysis was done on the bulk disturbed soil samples that were air dried, pulverized, and passed through a 2-mm sieve for determination of pH, EC, and through a 250 μm sieve before analyzing for Carbon/Nitrogen (C/N) ratio. The total C and total N concentrations were determined by the dry combustion method at 900 °C with a Vario Max C:N analyzer (Nelson and Sommers, 1996). Soil pH and EC were measured in a 1: 1 soil: water suspension using a hand-held portable probe1 (Lal, 1996; Peech, 1965). Because, the soil pH was approximately 7, no soil samples had significant inorganic carbonates, therefore, SOC concentration was considered equivalent to the total C (Brown et al., 2006; De Vos et al., 2005). The SOC stocks were computed by multiplying the SOC concentration by specific gravity (ρb/ρw, where ρb is the bulk density of soil and ρw the density of H2O) for respective soil layers, and expressed in Mg ha−1. The Nitrate and Nitrite concentration was determined from fresh soil samples stored in cool conditions up to analyses by Ion Chromatograph (Zhang et al., 2013). The corn (Zea mays L.) and soybean (Glycine max (L.) Merr.) were grown according to the specific agronomic practices for Ohio county; and hand harvested from August to September, 2012 at the same geographic location as the soil sampling. The yield was determined as follows: (i) the ears of the corn (i.e., hand harvested from a dimension of 2 m long and 2 rows wide); and the soybeans (i.e., harvest area 1 m2) were weighed in the field, (ii) the harvested corn and soybean were air dried, shelled, and the dry weight of the cobs, kennels, beans, and remaining above ground vegetative biomass (e.g., stalks) measured. The moisture content was determined by oven drying the samples at 60 °C for 96 h, and the yields computed with reference to 15.5% moisture for

1

Thermo Scientific, Orion Star Series, Made in Singapore

V. de Paul Obade, R. Lal / Science of the Total Environment 541 (2016) 424–434

corn, and 14% for soybean. The Harvest Index (HI) was determined as the ratio of the harvest weights to the total above ground vegetative biomass (Johnson et al., 2006). Although a minimum of 3 seasons yield data is the preferential standard, only one season data for the year 2012 is used to visualize the general trend.

2.1. Statistical analyses and model fitting

Table 2 Model Comparisons between SIMPLS, PCR, PLSR and the Reduced Rank (RR). Model

Predictor (explanatory) %

Response %

SIMPLS PCR PLSR Reduced Rank (RR)

11.1 17.7 11.6 10.7

13.1 0.6 39.6 (~66× N PCR) 41.2

PCR: Principal Component Regression. PLSR: Partial Least Squares Regression.

Prior to model selection in this study, the accuracy and robustness of the following non parametric methods: (i) Reduced Regression (RR), (ii) SIMPLS (de Jong, 1993), (iii) the Principal Component Regression (PCR) and (iv) Partial Least Squares Regression (PLSR) were evaluated using the PROC PLS in SAS 9.2 and pls code in R version 3.0.2 statistical software, at 5% significance level. The detailed description of the mathematics on PLSR, and other sundry methods such as the Reduced Regression (RR), SIMPLS, Principal Component Regression (PCR) are beyond the scope of this work, but can be accessed on the following references (Chong and Jun, 2005; de Jong, 1993; Mehmood et al., 2011, 2012). Based on the information adduced from the model comparisons (Table 2) supported by literature review (Chong and Jun, 2005; Mehmood et al., 2011, 2012), PLSR was preferentially selected for building the SQI; under the hypothesis that the soil types and management practices determine the location-specific soil physical and chemical properties. PLSR analyzes both qualitative and quantitative data, and projects predictors (X variables) and responses (Y variables) into a set of orthogonal variables called latent variables, thereby maximizing the covariance between the response and predictors (Chong and Jun, 2005). A cross-validation after 500 iterations was applied to select the optimal number of PLSR components to avoid over-fitting. The regression coefficients in the PLSR signify the weights of the independent variables or predictors; whereas the coefficient of determination reflects the degree to which the modeled dependent (i.e., response) variables explain the independent variables. In this study, 12 site characteristic predictor variables (i.e., soil type, management, soil depth increments) and 10 soil physical and chemical properties response variables were modeled. The covariance was determined between the response variables, namely; the available water capacity (AWC), field capacity (FC), permanent wilting point (PWP), soil ρb, electrical conductivity (EC), pH, Total Nitrogen (TN), Nitrate, Nitrite and SOC concentration, C/N ratio; and the predictors comprising the CrA, kbA, GWA, CtA, and Pw soils under NT, CT, NV management, for different soil layers (i.e., 0–10, 10–20, 20–40, and 40–60 cm). The Variable Importance of Projection (VIP) initialed as vj score (1) was the data reduction technique used for selecting the Table 1 Field sites, crop sequence, management practices in Ohio. Soil type description follows the USDA soil classification system. Site

Soil type

Crop sequence

Management

Miami Seneca (1) Seneca (2) Preble Auglaize

CrA kbA GWA CtA Pw

csc ccs csc ch c

NV, NT, NTcc, CT NV, NTccm, NTcc, CT NV, NTcc, CT NV, NT, CT NV, NT, CT

CrA (Crosby silt loam). kbA(Kibbie fine sandy loam). GWA (Glynwood silt loam). CtA (Crosby Celina silt loams). Pw (Pewamo silty clay loam). cc: cover crop. m: manure. CT: Conventional Tillage. NT: No Till. NV: Natural Vegetation (e.g., forest). c: corn. s: soybean. h: hay.

427

important variables. The selected variables had vj N 0.83 (Chong and Jun, 2005; Mehmood et al., 2012). To determine the overall SQI, the vj for each soil type, management, and soil layer, were rescaled by squaring, summing and converting to %. The rescaled % vj were subsequently aggregated according to the site specific definition of the soil type, management and respective depth increments. For example for CrA soil type, under NT management, for 0– 20 cm soil layers; the SQI = (soil CrA (% vj) + NT (% vj) + soil depth (0-10 cm) (% vj) + soil depth (10–20%) (% vj)).

vj ¼

ffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi h  i A A ðpÞ∑a¼1 SSa waj jkwa k2 =∑a¼1 ðSSa Þ

ð1Þ

• SSa = sum of squares by ath component; • p = predictor variables • vj = importance measure/weights based on covariance of each PLS component • waj = loading weight for jth variable using ath component. • (waj/ wa )2 = importance of the jth variable. Alternately, the SQI was computed directly from the PLSR regression model based on the aggregated soil attributes for each soil type, management, or soil layer. For instance, in the regression model (2) represented as: yi ¼ m1 x þ m2 x…… þ c

ð2Þ

yi = Σ Soil Attributes Index (SAI) for any soil type (e.g., kbA soil), is the response variable computed from the regression coefficients multiplied by the soil physico-chemical properties, the “x” being the laboratory measured values for each sample, and the corresponding regression coefficients are m1 = regression coefficient for nitrate, and m3 = regression coefficient for AWC etc., “c” is the error term which was centered and automatically rescaled to zero. Next, the site specific Σ SAI inclusive of selected soil layer or depth was standardized and transformed into % SQI as follows: SQI ð%Þ ¼ ðyi =ytotal Þ  100

ð3Þ

where yi = Σ SAI computed from the mean soil attribute values from the regression for selected soil type (e.g., GWA) or management (e.g., NT) or soil layer (e.g., 0–10 cm) whereas the ytotal constitutes the combined Σ SAI for each soil type, inclusive of management practices (i.e., NT + CT + NV) from 0 to 60 cm soil depth. Thus, the regression models relate aggregated soil physical and chemical properties vis-àvis specific management practice, soil layers, or soil type. For monitoring purposes, which is the jist for SQIs, this logic should hold so long as “oranges are compared with oranges”, that is, the same soil physical and chemical properties are compared, which also imply that during the relative comparison the units cancel out. Finally, correlation analyses between the SQI and the yields were computed.

428

V. de Paul Obade, R. Lal / Science of the Total Environment 541 (2016) 424–434

V. de Paul Obade, R. Lal / Science of the Total Environment 541 (2016) 424–434

Fig. 3. Partial Least Squares Regression (PLSR) correlation loading plot. CrA (Crosby silt loam) kbA(Kibbie fine sandy loam) GWA (Glynwood silt loam) CtA (Crosby Celina silt loams) mngt: management Pw (Pewamo silty clay loam) CT: Conventional Tillage NT: No Till NV: Natural Vegetation (e.g., forest) Depth 10, 20, 40, and 60 represent soil depths 0–10, 10–20, 20–40 and 40–60 respectively.

3. Results and discussion 3.1. Assessing the model efficacy Fig. 2 a, b and c documents the rationale for selecting the best model to build the SQI, from the PLSR, RR, SIMPLS, and PCR multivariate techniques used to relate 10 soil physical and chemical properties, with 12 determinant variables (i.e., (i) the 5 soil types (i.e., CrA, kbA, GWA, CtA, Pw), (ii) 3 management practices (i.e., NT, CT, NV) and (iii) 4 soil layers (i.e., 0–10, 10–20, 20–40 and 40–60 cm). Based on the “absolute minimum Predicted Residual Sum of Squares (PRESS)”, the PLS (Fig. 2a), RR (Fig. 2b), and SIMPLS (not shown) models performed optimally with only 3 out of a total 10 factors required, compared with 9 for PCR. The explanatory power of the PLSR, RR, SIMPLS, and PCR models (Table 2) were as follows: (i) the RR had 41.2% of the response variation, compared with only 10.7% for the predictor; (ii) the PCR accounted for 17.7% of the predictor, but only 0.6% of the response variation; (iii) the PLS had slightly less predictor variation (11.6%) compared with the PCR, but 39.6% response variation, approximately 66 times that of PCR, and (iv) the SIMPLS had 11.1% predictor and 13.1% response variation. These concur with previously reported findings of Chong and Jun (2005) and Mehmood et al. (2011, 2012), that PLSR is more informative. Fig. 3 shows the relative variable weights generated from the first 2 PLSR factors. The weights or loadings are notably low, hypothetically due to the complex interactions and spatial heterogeneity of soil properties. For example, the 1st factor had only 11.1% the predictors accurately explained by the model, and 13.1% for the response variables; whereas the 2nd factor had 12.7% predictor fit and 10.1% response fit, respectively. The information provided by the circles embedded in %

429

reflect the relative weights for each of the soil attributes, under different management and depth increments; and increase from the origin or center, for both factors; with the 100% circle representing maximum weight. The positively correlated response and predictor model variables, as illustrated by the 1st factor were ρb, NT management and Pw soil; whereas the negatively correlated were NV managed soils, C, N, SOC, EC, FC, CtA soil at all depths. The CrA, kbA, GWA, and Pw negatively varied with the soil physical and chemical characteristics only at the surface (b 10 cm). The C, N and SOC were the most sensitive soil indicators to management perturbations; which corroborates the fact that N cycle being inextricably linked to the C cycle exhibit similar soil dynamics (Flessa et al., 2002; Govaerts et al., 2006). On the other hand, positively correlated model variables as shown by the 2nd factor were the soils under NV management, kbA and Pw soils, C/N; whereas the negatively correlated were SOC concentration, EC, pH, FC, PWP, CT managed soils, CtA and GWA soil, but surface soils had negligible correlation with the remaining model variables. The significant predictor variables (i.e., N 25%) in the 1st factor were SOC concentration, ρb, C, N and EC (numerical details provided in Table 3 for the first 3 factors). In comparison, the significant determinant model variables in decreasing order were the NV (75%) and NT (50%) managed soils; Pw, CtA and kbA (25%) soils, respectively. Similarly, Fig. 4a computed by the Wold's criterion, shows that the Pw, CtA, kbA, soil's, NV and NT managed soils, and the surface (b10 cm) soils were significant model variables. Based on the information adduced from the regression coefficients (Table 4), the CrA soil had less (i.e., −0.02) Nitrate concentration than other soils, lower ρb (−0.02), lower AWC (− 0.01); but had relatively higher SOC (+ 0.03) and Nitrite (+ 0.01) concentration, higher pH (+0.07), EC (+0.05), C/N (+0.02), FC (+0.04) and PWP (+0.05). Similar trends were observed for the CtA and GWA soils. However, the Pw and kbA soil, and CT managed soils had relatively lower C, N, FC, EC, PWP, pH, Nitrite and SOC concentration as reflected by their respective negative coefficients; but higher Nitrate concentration, AWC, and ρb. Interestingly, the NT managed soils, had relatively lower (− ve correlation) C/N and SOC concentration, AWC; but higher ρb, EC, FC, PWP, pH, and Nitrite concentration, than the NV and CT managed soils. Yet, the NV managed soil had less ρb, pH, EC, FC, Nitrite and Nitrate concentration, and relatively higher C/N ratio, AWC, PWP, and SOC concentration. As expected, the surface soil layer (b 10 cm depth) had relatively less ρb, pH, PWP, Nitrite concentration, FC, but positive for EC, SOC, EC, Nitrate concentration, and the AWC. Comparatively, the deeper soil layers (i.e., N 40 to 60 cm) had less Nitrate and SOC concentration, EC, AWC, pH, and higher with FC, ρb, PWP, Nitrite concentration, and C/N. This variability in soil attributes with depth signify complex soil processes that are hypothetically driven by respiration of soil microbes, bioturbation, erosion or leaching. Fig. 5a to e, reveal that the NV managed kbA soils had the highest SQI (i.e., 32%) followed by the Pw, GWA, CtA and CrA soils, respectively. Similarly kbA soils had the highest SQI under NT management, but the SQI in decreasing order for NT soils was: CrA ≈ GWA, CtA ≈ Pw. Alternately, Pw surface soil layers (i.e., b20 cm) under CT management which occupied a relatively larger land area had the highest SQI, followed by the CrA, CtA, GWA, kbA soils. 3.2. Interconnection between SQI, land management practices and yield Agro-ecosystems are complex entities encompassing interacting biotic and abiotic components (Blecker et al., 2012). This study opted to build the SQI using PLSR; because unlike PCA which is based on variance of predictor variables and therefore extracts information from only the

Fig. 2. a. Partial Least Squares Regression (PLSR) validation. Top graph: A prediction model with 3 factors provides similar accuracy as one with all the 10 PLS factors, as interpreted by the minimum Root Mean Predicted Residual Sum of Squares (PRESS). Bottom graph: The regression model attained full potential (i.e., the accuracy of the dependent variable remained constant) with only 3 PLS components or factors as shown by the maximum R2 = 0.2 (20%). b. Reduced Rank (RR) regression validation. Top graph: The prediction model required only 3 out of the 10 computed RR factors to have the best accuracy as shown by the minimum Predicted Residual Sum of Squares (PRESS) (i.e., similar to PLSR). Bottom graph: The regression model attained full potential (i.e., the accuracy of the dependent variable remained constant) with only 3 RR components or factors as shown by the maximum R2 = 0.2 (20%). c. Principal Component Regression (PCR) validation. Top graph: The prediction model requires 9 out of 10 PCR factors to have minimum Predicted Residual Sum of Squares (PRESS). Bottom graph: The regression model attained full potential (i.e., the accuracy of the dependent variable remained constant) with 6 PCR components or factors as shown by the maximum R2 = 0.2 (20%).

430

V. de Paul Obade, R. Lal / Science of the Total Environment 541 (2016) 424–434

Table 3 The weights of soil variables for each of the first 3 factors computed by Partial Least Squares Regression (PLSR). No. of extracted factors

Nitrate

Nitrite

AWC

FC

PWP

ρb

C/N

SOC

pH

EC

1 2 3

0.03 0.03 0.55

0.02 −0.10 −0.09

−0.07 0.09 0.28

−0.10 −0.35 −0.25

−0.03 −0.38 −0.49

0.57 −0.02 −0.08

−0.45 0.14 −0.42

−0.58 −0.05 −0.01

0.05 −0.70 0.07

−0.33 −0.45 0.35

AWC: Available Water Capacity (cm3 of water cm−3 of soil). SOC: Soil Organic Carbon (Mg ha−1). EC: Electrical Conductivity (μS cm−1). ρb: Soil Bulk Density (Mg m−3). FC: Field Capacity (cm3 of water cm−3 of soil). PWP: Permanent Wilting Point (cm3 of water cm−3 of soil). C/N: carbon/nitrogen ratio.

Fig. 4. a. Variable Importance Plot (VIP) from Wold's Criterion and computed by Partial Least Squares Regression (PLSR). The selected variables had vj N 0.83 b. Soil Quality Index (SQI) computed from the Wold's Criterion for each soil type, management, and soil depth increment, and rescaled in %. CrA (Crosby silt loam) kbA(Kibbie fine sandy loam) GWA (Glynwood silt loam) CtA (Crosby Celina silt loams) Pw (Pewamo silty clay loam) mngt: management CT: Conventional Tillage NT: No Till NV: Natural Vegetation (e.g., forest) Depth 10, 20, 40, and 60 represent soil depths 0–10, 10–20, 20–40 and 40–60 respectively.

V. de Paul Obade, R. Lal / Science of the Total Environment 541 (2016) 424–434

431

Table 4 The centered and scaled regression coefficients computed from the Partial Least Squares Regression (PLSR).

Intercept CrA CtA GWA Pw kbA CT NT NV 0–10 10–20 20–40 40–60

Nitrate

Nitrite

AWC

FC

PWP

ρb

C/N

SOC

pH

EC

0 −0.02 −0.07 −0.08 0.11 0.07 0.10 0 −0.10 0.17 0.03 −0.05 −0.16

0 0.01 0.04 0.02 −0.04 −0.04 0 0.04 −0.04 −0.03 −0.01 0.01 0.03

0 −0.01 −0.04 −0.04 0.05 0.05 0.04 −0.05 0.02 0.12 0.01 −0.04 −0.09

0 0.04 0.17 0.11 −0.20 −0.17 −0.01 0.06 −0.05 −0.01 −0.03 −0.01 0.05

0 0.05 0.19 0.14 −0.22 −0.20 −0.05 0.09 −0.06 −0.12 −0.04 0.03 0.14

0 −0.02 −0.17 −0.09 0.22 0.09 0.04 0.22 −0.27 −0.27 0.03 0.13 0.12

0 0.02 0.13 0.11 −0.21 −0.07 −0.14 −0.21 0.35 0.04 −0.04 −0.04 0.04

0 0.03 0.20 0.12 −0.26 −0.13 −0.05 −0.21 0.27 0.25 −0.04 −0.12 −0.10

0 0.07 0.22 0.10 −0.20 −0.25 0.11 0.21 −0.32 0.06 −0.02 −0.02 −0.02

0 0.05 0.23 0.09 −0.23 −0.20 0.09 0 −0.08 0.30 −0.02 −0.13 −0.16

CrA (Crosby silt loam). kbA(Kibbie fine sandy loam). GWA (Glynwood silt loam). CtA (Crosby Celina silt loams). Pw (Pewamo silty clay loam). CT: Conventional Tillage. NT: No Till. NV: Natural Vegetation (e.g., forest). AWC: Available Water Capacity (cm3 of water cm−3 of soil). SOC: Soil Organic Carbon (Mg ha−1). EC: Electrical Conductivity (μS cm−1). ρb: Soil Bulk Density (Mg m−3). FC: Field Capacity (cm3 of water cm−3 of soil). PWP: Permanent Wilting Point (cm3 of water cm−3 of soil). C/N: carbon/nitrogen ratio.

first few components (i.e., b3) thus losing some information; the PLSR which evaluates the covariance between predictor and response variables utilizes all model information. Moreover, PLSR synthesizes both qualitatitive and quantitative data. Therefore, information useful for elucidating the complex interrelationship between the soil properties and management practices can be missed using the PCA. Thus, soil functionality was evaluated by comparing the yields at each site against the PLSR derived SQI (Fig. 6): (i) inclusive of the aggregated weight of soil attributes up to 60 cm soil depth (Fig. 4b), and (ii) assuming that soil depths had zero weights, or no influence on soil physical and chemical property variability, synonymous with the “zero-option” based on the arguments of Bouma and McBratney (2013). The yield comparison against SQI inclusive of soil depth weights (Fig. 6) revealed that the NT managed Pw soil from Auglaize site had the highest SQI (47%), but yielded only 2.5 Mg/ha of corn in 2012, but the NT managed CrA soil from Miami site had the least SQI (37.2%) with a high 9.24 Mg/ha corn yield. In contrast, the CT managed CrA soil with the least SQI (29%) had a high corn yield (9.9 Mg/ha), whereas the Pw soil under similar CT management produced only 5.1 Mg/ha corn yield despite having a high SQI (39.6%). The highest soybean yields (2.3 Mg/ha) were in the CT managed kbA soil from Seneca site 1 (SQI = 36%), compared with the GWA soil under NT from Seneca site 2 that yielded only 1.2 Mg/ha at SQI of 37.9%. Based on the combined data from all the sites, the computed mean SQI, corn and soybean yields for the year 2012, was 42.9%, 7 and 2 Mg/ha, with the PLSR model fit of SQI versus corn and soybean having an R2 of 0.74, 0.89, respectively. This SQI technique opens up new prospects for unbiased assessment of soil quality from one location relative to that of another. This SQI concept can be innovated by replacing the soil type descriptions with arbitrary ranges of soil texture descriptions to enable regional SQI comparisons. Besides management choices (e.g., NT, CT), other factors that reportedly influence agricultural production, and were not modeled in this study include: crop hybrids, microbial activity, pests occurrence, solar insolation, planting dates, fertilizer application and intake rates, and the weather patterns (Bouma, 2002; Bouma and

McBratney, 2013; Manandhar and Odeh, 2014). Because, the year 2012 was a drought year, this SQI approach will be tested with data from subsequent years to determine its efficacy. In essence, long-term data-sets integrated into dynamic process-oriented simulation models are required to critically appraise SQIs (Bonfante and Bouma, 2015; Bouma, 2015). Finally, this SQI technique is a precursor towards the development of a versatile soil quality monitoring tool useful for: (i) sequencing soil quality vis-à-vis soil functionality (e.g., agricultural production), and (ii) providing information for guiding sustainable agro-ecosystem management.

4. Conclusion This study exemplifies a simple yet comprehensive PLSR method that synthesizes soil attributes into a SQI useful for explaining the interconnection between management practices, soil quality and crop yields. Unlike other models that relate single soil properties (e.g., AWC, ρb,) to crop yields; this SQI designed from conjoined soil attributes, provides comprehensive evaluation of SQI vis-à-vis crop yields, because plants intake different nutrients simultaneously at varied environmental gradients. Besides, Pw soil were generally of a higher quality than CtA, kbA, GWA and CrA, respectively. SOC, ρb, C/N ratio and EC were the important variables influencing site specific soil quality. Future research should investigate: (i) techniques for mapping or spatially interpolating this SQI, because one field can have several soil types, and (ii) the utility of this SQI to monitor biogeochemical cycling.

Acknowledgments This research is part of a regional collaborative project supported by the USDA-NIFA, Award No. 2011-68002-30190 “Cropping Systems Coordinated Agricultural Project (CAP): Climate Change, Mitigation, and Adaptation in Corn-based Cropping Systems.”

432

V. de Paul Obade, R. Lal / Science of the Total Environment 541 (2016) 424–434

Fig. 5. a to e. Soil Quality Index (SQI) computed directly from the aggregated soil attributes in the regression model for different management practices and soil layers. Error bars represent standard deviation. . CrA (Crosby silt loam) kbA(Kibbie fine sandy loam) GWA (Glynwood silt loam) CtA (Crosby Celina silt loams) mngt: management Pw (Pewamo silty clay loam) CT: Conventional Tillage NT: No Till NV: Natural Vegetation (e.g., forest) Depth 10, 20, 40, and 60 represent soil depths 0–10, 10–20, 20–40 and 40–60 respectively.

References Andrews, S.S., Carroll, C.R., 2001. Designing a soil quality assessment tool for sustainable agroecosystem management. Ecol. Appl. 11, 1573–1585. Andrews, S.S., Flora, C.B., Mitchell, J.P., Karlen, D.L., 2003. Growers' perceptions and acceptance of soil quality indices. Geoderma 114, 187–213.

Andrews, S.S., Karlen, D.L., Cambardella, C.A., 2004. The soil management assessment framework: a quantitative soil quality evaluation method. Soil Sci. Soc. Am. J. 68, 1945–1962. Andrews, S.S., Karlen, D.L., Mitchell, J.P., 2002a. A comparison of soil quality indexing methods for vegetable production systems in Northern California. Agric. Ecosyst. Environ. 90, 25–45.

V. de Paul Obade, R. Lal / Science of the Total Environment 541 (2016) 424–434

433

Fig. 6. Soil Quality Index (SQI) versus yield. The yield data for NT/kbA is not available. CrA (Crosby silt loam) kbA(Kibbie fine sandy loam) GWA (Glynwood silt loam) CtA (Crosby Celina silt loams) Pw (Pewamo silty clay loam) CT: Conventional Tillage NT: No Till NV: Natural Vegetation (e.g., forest).

Andrews, S.S., Mitchell, J.P., Mancinelli, R., Karlen, D.L., Hartz, T.K., Horwath, W.R., et al., 2002b. On-farm assessment of soil quality in California's central valley. Agron. J. 94, 12–23. Arshad, M.A., Martin, S., 2002. Identifying critical limits for soil quality indicators in agroecosystems. Agric. Ecosyst. Environ. 88, 153–160. Askari, M.S., Holden, N.M., 2014. Indices for quantitative evaluation of soil quality under grassland management. Geoderma 230–231, 131–142. Blanco-Canqui, H., Lal, R., 2007. Impacts of long-term wheat straw management on soil hydraulic properties under no-tillage. Soil Sci. Soc. Am. J. 71, 1166–1173. Blecker, S.W., Stillings, L.L., Amacher, M.C., Ippolito, J.A., DeCrappeo, N.M., 2012. Development of vegetation based soil quality indices for mineralized terrane in arid and semi-arid regions. Ecol. Indic. 20, 65–74. Bonfante, A., Bouma, J., 2015. The role of soil series in quantitative land evaluation when expressing effects of climate change and crop breeding on future land use. Geoderma 259–260, 187–195. Boote, K.J., Jones, J.W., Pickering, N.B., 1996. Potential uses and limitations of crop models. Agron. J. 88, 704–716. Bouma, J., 2002. Land quality indicators of sustainable land management across scales. Agric. Ecosyst. Environ. 88, 129–136. Bouma, J., 2015. Reaching out from the soil-box in pursuit of soil security. Soil Sci. Plant Nutr. 1-10. Bouma, J., McBratney, A., 2013. Framing soils as an actor when dealing with wicked environmental problems. Geoderma 200–201, 130–139. Box George, E.P., Draper, N.R., 1987. Empirical Model-Building and Response Surfaces. Wiley. Brown, D.J., Shepherd, K.D., Walsh, M.G., Mays, M.D., Reinsch, T.G., 2006. Global soil characterization with VNIR diffuse reflectance spectroscopy. Geoderma 132, 273–290. Chong, I.G., Jun, C.H., 2005. Performance of some variable selection methods when multicollinearity is present. Chemom. Intell. Lab. Syst. 78, 103–112. Clement, C.R., 1996. A simple and liable tension table. J. Soil Sci. 17. Cohen, M., Mylavarapu, R.S., Bogrekci, I., Lee, W.S., Clark, M.W., 2007. Reflectance spectroscopy for routine agronomic soil analyses. Soil Sci. 172, 469–485. Dane, J.H., Hopmans, J.W., 2002. Water retention and storage. In: Dane, J.H., Topp, G.C. (Eds.), Methods of Soil Analysis: Part 4 Physical Methods. 5. Agronomy Monograph, Madison, WI, pp. 675–719. de Jong, S., 1993. SIMPLS: an alternative approach to partial least squares regression. Chemom. Intell. Lab. Syst. 18, 251–263. de la Paz Jimenez, M., de la Horra, A., Pruzzo, L., Palma, M., 2002. Soil quality: a new index based on microbiological and biochemical parameters. Biol. Fertil. Soils 35, 302–306. de Moraes Sá, J.C., Bürkner dos Santos, J., Lal, R., de Moraes, A., Tivet, F., Machado Sá, M.F., et al., 2013. Soil-specific inventories of landscape carbon and nitrogen stocks under no-till and native vegetation to estimate carbon offset in a subtropical ecosystem. Soil Sci. Soc. Am. J. 77, 2094–2110. de Paul Obade, V., Lal, R., 2014. Using meta-analyses to assess pedo-variability under different land uses and soil management in central Ohio, USA. Geoderma 232–234, 56–68. de Paul Obade, V., Lal, R., Chen, J., 2013. Remote sensing of soil and water quality in agroecosystems. Water Air Soil Pollut. 224, 1–27. De Vos, B., Van Meirvenne, M., Quataert, P., Deckers, J., Muys, B., 2005. Predictive quality of pedotransfer functions for estimating bulk density of forest soils. Soil Sci. Soc. Am. J. 69, 500–510. DeForest, J.L., Smemo, K.A., Burke, D.J., Elliott, H.L., Becker, J.C., 2012. Soil microbial responses to elevated phosphorus and pH in acidic temperate deciduous forests. Biogeochemistry 109, 189–202. Doran, J.W., Sarrantonio, M., Liebig, M.A., 1996. Soil health and sustainability. In: Sparks, D.L. (Ed.), Advances in Agronomy. Vol. 56, pp. 1–54.

Doran, J.W., Zeiss, M.R., 2000. Soil health and sustainability: managing the biotic component of soil quality. Appl. Soil Ecol. 15, 3–11. Flessa, H., Ruser, R., Dörsch, P., Kamp, T., Jimenez, M.A., et al., 2002. Integrated evaluation of greenhouse gas emissions (CO2, CH4, N2O) from two farming systems in southern Germany. Agric. Ecosyst. Environ. 91, 175–189. Gobin, A., Campling, P., Deckers, J., Feyen, J., 2000. Quantifying soil morphology in tropical environments methods and application in soil classification. Soil Sci. Soc. Am. J. 64, 1423–1433. Gogé, F., Gomez, C., Jolivet, C., Joffre, R., 2014. Which strategy is best to predict soil properties of a local site from a national Vis–NIR database? Geoderma 213, 1–9. Govaerts, B., Sayre, K.D., Deckers, J., 2006. A minimum data set for soil quality assessment of wheat and maize cropping in the highlands of Mexico. Soil Tillage Res. 87, 163–174. Grinand, C., Arrouays, D., Laroche, B., Martin, M.P., 2008. Extrapolating regional soil landscapes from an existing soil map: sampling intensity, validation procedures, and integration of spatial context. Geoderma 143, 180–190. Jemai, I., Ben Aissa, N., Ben Guirat, S., Ben-Hammouda, M., Gallali, T., 2013. Impact of three and seven years of no-tillage on the soil water storage, in the plant root zone, under a dry subhumid Tunisian climate. Soil Tillage Res. 126, 26–33. Johnson, J.M.-F., Allmaras, R.R., Reicosky, D.C., 2006. Estimating source carbon from crop residues, roots and rhizodeposits using the national grain-yield database. Agron. J. 98, 622–636. Karlen, D.L., Mausbach, M.J., Doran, J.W., Cline, R.G., Harris, R.F., Schuman, G.E., 1997. Soil quality: a concept, definition, and framework for evaluation. Soil Sci. Soc. Am. J. 61, 4–10. Karlen, D.L., Tomer, M.D., Neppel, J., Cambardella, C.A., 2008. A preliminary watershed scale soil quality assessment in north central Iowa, USA. Soil Tillage Res. 99, 291–299. Karlen, D.L., Wollenhaupt, N.C., Erbach, D.C., Berry, E.C., Swan, J.B., Eash, N.S., et al., 1994a. Long-term tillage effects on soil quality. Soil Tillage Res. 32, 313–327. Karlen, D.L., Wollenhaupt, N.C., Erbach, D.C., Berry, E.C., Swan, J.B., Eash, N.S., et al., 1994b. Crop residue effects on soil quality following 10-years of no-till corn. Soil Tillage Res. 31, 149–167. Kinoshita, R., Moebius-Clune, B.N., van Es, H.M., Hively, W.D., Bilgilis, A.V., 2012. Strategies for soil quality assessment using visible and near-infrared reflectance spectroscopy in a western Kenya chronosequence. Soil Sci. Soc. Am. J. 76, 1776–1788. Kladivko, E.J., Helmers, M.J., Abendroth, L.J., Herzmann, D., Lal, R., Castellano, M.J., et al., 2014. Standardized research protocols enable transdisciplinary research of climate variation impacts in corn production systems. J. Soil Water Conserv. 69, 532–542. Klute, A., 1986. Water Retention: Laboratory Methods. Methods of Soil Analysis: Part 1 — Physical and Mineralogical Methods pp. 635–662. Klute A, Dirksen C. Hydraulic Conductivity and Diffusivity: Laboratory Methods. Methods of Soil Analysis: Part 1 — Physical and Mineralogical Methods 1986; sssabookseries: 687–734. Lacoste, M., Minasny, B., McBratney, A., Michot, D., Viaud, V., Walter, C., 2014. High resolution 3D mapping of soil organic carbon in a heterogeneous agricultural landscape. Geoderma 213, 296–311. Lal R. Deforestation and land-use effects on soil degradation and rehabilitation in western Nigeria. Soil chemical properties. Land Degrad. Dev. 1996; 7: 87–98. Lal, R., 2009. Ten tenets of sustainable soil management. J. Soil Water Conserv. 64, 20A–21A. Lal, R., Delgado, J.A., Gulliford, J., Nielsen, D., Rice, C.W., Van Pelt, R.S., 2012. Adapting agriculture to drought and extreme events. J. Soil Water Conserv. 67, 162A–166A. Manandhar, R., Odeh, I., 2014. Interrelationships of land use/cover change and topography with soil acidity and salinity as indicators of land degradation. Land 3, 282–299.

434

V. de Paul Obade, R. Lal / Science of the Total Environment 541 (2016) 424–434

Marín-González, O., Kuang, B., Quraishi, M.Z., Munóz-García, M.Á., Mouazen, A.M., 2013. On-line measurement of soil properties without direct spectral response in near infrared spectral range. Soil Tillage Res. 132, 21–29. McBratney, A., Field, D.J., Koch, A., 2014. The dimensions of soil security. Geoderma 213, 203–213. McBratney, A.B., Minasny, B., Cattle, S.R., Vervoort, R.W., 2002. From pedotransfer functions to soil inference systems. Geoderma 109, 41–73. Mehmood, T., Liland, K.H., Snipen, L., Saebo, S., 2012. A review of variable selection methods in Partial Least Squares Regression. Chemom. Intell. Lab. Syst. 118, 62–69. Mehmood, T., Martens, H., Saebo, S., Warringer, J., Snipen, L., 2011. A Partial Least Squares based algorithm for parsimonious variable selection. Algorithms Mol. Biol. 6. Minasny, B., Hartemink, A.E., 2011. Predicting soil properties in the tropics. Earth Sci. Rev. 106, 52–62. Nelson DW, Sommers LE. Total carbon, organic carbon, and organic matter. In: Sparks DL, et al., editors. Am. Soc. of Agron. 9. SSA, Madison, WI., 1996, pp. 961–1010. Nortcliff, S., 2002. Standardisation of soil quality attributes. Agric. Ecosyst. Environ. 88, 161–168. NRCS, 2012. Soil Quality Concept. Natural Resources Conservation Service (NRCS). Ohlson, K., 2014. The soil will save us: how scientists, farmers, and foodies are healing the soil to save the planet: Rodale.

Peech, M., 1965. Hydrogen ion activity. ASA Monograph. 9. ASA, Madison, WI, pp. 914–925. Shepherd, K.D., Walsh, M.G., 2002. Development of reflectance spectral libraries for characterization of soil properties. Soil Sci. Soc. Am. J. 66, 988–998. Staff, S.S., 1951. Soil Survey Manual. USDA, Washington DC, p. 503. Stockmann, U., Adams, M.A., Crawford, J.W., Field, D.J., Henakaarchchi, N., Jenkins, M., et al., 2013. The knowns, known unknowns and unknowns of sequestration of soil organic carbon. Agric. Ecosyst. Environ. 164, 80–99. Topp, G.C., Ferre, P.A., 2002. Water content. In: Dane, J.H., Topp, G.C. (Eds.), Methods of Soil Analysis: Part 4 Physical Methods. 5. SSSA, Agronomy Monograph, Madison, WI, pp. 417–422. Wienhold, B.J., Andrews, S.S., Karlen, D.L., 2004. Soil quality: a review of the science and experiences in the USA. Environ. Geochem. Health 26, 89–95. Wienhold, B.J., Karlen, D.L., Andrews, S.S., Stott, D.E., 2009. Protocol for indicator scoring in the soil management assessment framework (SMAF). Renew. Agr. Food Syst. 24, 260–266. Yemefack, M., Jetten, V.G., Rossiter, D.G., 2006. Developing a minimum data set for characterizing soil dynamics in shifting cultivation systems. Soil Tillage Res. 86, 84–98. Zhang, S., Zhao, T.B., Wang, J., Qu, X.L., Chen, W., Han, Y., 2013. Determination of fluorine, chlorine and bromine in household products by means of oxygen bomb combustion and ion chromatography. J. Chromatogr. Sci. 51, 65–69.