Estimation of soil properties with mid-infrared soil spectroscopy across yam production landscapes in West Africa

Low soil fertility is challenging the sustainable production of yam and other staple crops in the yam belt of West Africa. Quantitative soil measures are needed to assess soil fertility decline and to improve crop nutrient supply in the region. We developed and tested a mid-infrared (mid-IR) soil spectral library to enable timely and cost-efficient assessments of soil properties. Our collection included 80 soil samples from four landscapes (10 km × 10 km) and 20 fields per landscape across a gradient from humid forest to savannah and 14 additional samples from one landscape that had been sampled within the Land Health Degradation Framework. We derived partial least squares regression models to spectrally estimate soil properties. The models produced accurate cross-validated estimates of total carbon, total nitrogen, total sulfur, total iron, total aluminum, total potassium, total calcium, exchangeable calcium, effective cation exchange capacity, and diethylenetriaminepentaacetic acid (DTPA)-extractable iron and clay content (R²>0.75). The estimates of total zinc, pH, exchangeable magnesium, bioavailable copper, and manganese were less predictable (R²>0.50). Our results confirm that mid-IR spectroscopy is a reliable and quick method to assess the regional-level variation of most soil properties, especially the ones closely associated with soil organic matter. Although the relatively small mid-IR library shows satisfactory performance, we expect that frequent but small model updates will be needed to adapt the library to the variation of soil quality within individual fields in the regions and their temporal fluctuations.

1 Introduction

Yam (Dioscorea spp.) is an important food and cash crop in West Africa. The yam belt of West Africa spans across the central zone of coastal countries in West Africa, located across the humid forest zone and northern Guinean savanna. It contributes to about 92 % of total world yam production, e.g., a total yield of 73×10⁶ t in 2017 (Food and Agriculture Organization of the United Nations, 2019). The cropping area in the West African yam belt has been expanded with accelerated population growth. The deforestation and expansion of agricultural land has in many places caused soil degradation. Furthermore, there has been a trend of shortened fallow periods in the cropping areas of West Africa over the last decades, which has further exacerbated the decline in soil fertility across the yam belt. Traditionally, yam is grown without external input in these areas. Therefore, the production of yam and other crops grown in the region depends on soil organic matter (SOM) status (Padwick, 1983), which serves as a main pool of plant-available nutrients and provides cation exchange surfaces for soil nutrients (Syers et al., 1970; Soares and Alleoni, 2008). A particularly strong positive relationship between high organic matter stocks and yam productivity is reported after fallow and when no fertilizer is added (Diby et al., 2009; Kassi et al., 2017). Thus, maintaining or increasing SOM and available nutrient levels is of utmost importance for sustainable production of yam and other crops in West Africa (Carsky et al., 2010). Furthermore, linking soil properties and yam yields (Frossard et al., 2017) and accounting for soil macro- and micronutrient status (O'Sullivan and Jenner, 2006) are fundamental to improving crop yields and soil management strategies.

Soil fertility is an integrative measure of soil attributes and their interactions that support the long-term agricultural production potential. Soil fertility is commonly decomposed into physical, chemical, and biological major components (Abbott and Murphy, 2007). Here, it is important to interpret soil fertility in the form of soil conditions and functions at an adequate resolution over time and space and in relation to the crop of interest. For yam, low tuber yields are often attributed to an unbalanced ratio of essential nutrients (i.e., N, P, K) available in the soil (Enyi, 1972) and a fast mineralization and hence depletion of organic matter (Carsky et al., 2010; Hgaza et al., 2011). Yet, the relationship between soil properties and tuber yield is not fully understood (Frossard et al., 2017). The reason is that the response of yam to mineral fertilization is highly variable because of confounding environmental and management variables, such as climate, soil type, inherent soil fertility, micronutrient deficiencies, tillage, seed tuber quality, planting date and density, staking, and disease pressure across the yam belt (Kang and Wilson, 1981; O'Sullivan and Jenner, 2006; Cornet et al., 2016; Enesi et al., 2018). Further, there are no soil fertility recommendations specific for yam under West African conditions. For this reason, establishing yam field trials designed with different organic and mineral fertilization strategies within different yam-growing regions is required to optimize yam nutrient supply targeting regional soil and environmental conditions (Frossard et al., 2017). Despite the importance of soil fertility, it is challenging to quantify soil measures at sufficient temporal and spatial resolution to relate them to yam productivity together with other management effects.

To quickly assess key soil properties, such as soil organic carbon (SOC) and cation exchange capacity (CEC), we need more cost- and time-efficient methods in addition to the traditional wet chemistry laboratory analyses that are often cost-intensive and time-consuming. Proximal sensing is a method that can provide reliable, rapid, and inexpensive soil measurements (UNEP, 2012). Soil visible and near-infrared (vis–NIR) and mid-infrared (mid-IR) diffuse reflectance spectroscopy has gained popularity over the past 30 years to assess soil properties in a complementary manner to conventional laboratory analytical methods (Nocita et al., 2015). For model development and calibration but, importantly, also for validation purposes, soil IR spectroscopy requires laboratory reference analysis data. Previous studies have shown successful spectroscopic predictions of soil properties, such as organic C, texture, cation exchange capacity (CEC), and exchangeable K (Viscarra Rossel et al., 2006; Cécillon et al., 2009; Nocita et al., 2015; Sila et al., 2016). Many soil chemical and physical properties, such as soil mineralogy and the concentration, forms, and distribution of SOM, are closely associated with IR spectral diversity. However, for determining a range of extraction-based soil properties, the predictive capability seems variable. This can be caused by complex surface chemical processes that are not directly related to soil organic matter and/or insufficient densities available at local scale to represent such locally complex relationships (Viscarra Rossel et al., 2006; Abdi et al., 2012; Sanderman et al., 2020). Further, a library that includes a broad range of soil biophysical conditions found in the region in which it is used needs to be established. Depending on the geographical extent of the study – field (e.g., Cambou et al., 2016), region, country (e.g., Clairotte et al., 2016), continent (e.g., Sila et al., 2016), world (e.g., Viscarra Rossel et al., 2016) – various statistical predictive modeling strategies are typically employed to account for geographically regional variability in soil properties and determine empirical relationships between spectra and soil attributes. Particular subsets of and features in spectra are characteristic of functional groups of soil components, and thus, elucidating spectral features that are important for the prediction of a particular soil attribute helps to understand and validate the mechanisms based on which the empirically models predict the soil properties.

In this work, we aim to develop mid-IR spectroscopy as a diagnostic tool for key analytical soil variables within four climatically, ecologically, and agriculturally distinct landscapes in Burkina Faso and Côte d’Ivoire. For yam and other cash crops, there is a lack of soil diagnostic tools to identify factors limiting yields and to derive site-specific fertilizer recommendations within and across landscapes. In these regions, yam has substantial economic importance for small-holder farmers. As land management and soil status is a key factor not only for yam but also for other high-value crops in the region, quick and cost-effective soil status assessments should be transferable to other crops with similar nutrient demands. Thus, the main objectives of this study were to (1) develop and evaluate openly accessible and reusable mid-IR spectroscopic models to estimate soil properties for selected landscapes representing major soil and climatic conditions in the West African yam belt, (2) to determine important spectral features for specific soil properties, and (3) to build a new soil spectral library in four landscapes of the West African yam belt for soil prediction and assessment. Finally, we make specific recommendations on whether and how specific mid-infrared diagnostic measures are applicable for different soil management and screening purposes. We also discuss the spectroscopic evaluation of the soil's capacity to retain and release nutrients for sustained and improved cropping in the region.

2 Materials and methods

2.1 Landscapes and soil sampling

Our study area covered the climatic and soil biophysical conditions representative of the West African yam belt. We selected four landscapes, two in Côte d’Ivoire and two in Burkina Faso. Each landscape (approximately 10 km × 10 km) represents a diverse geographic ecoregion. The landscapes cover a gradient between humid forest and the northern Guinean savannah. Specifically, the landscape Liliyo in Côte d’Ivoire is at 5.88^∘ N and in the humid forest zone. The predominant soil type is Ferralsol (IUSS Working Group WRB, 2015). The landscape Tiéningboué in Côte d’Ivoire is at 8.14^∘ N and belongs to the forest savannah transitional zone. The soils are dominated by Nitisols and Lixisols (IUSS Working Group WRB, 2015). The landscape Midebdo is at 9.97^∘ N and in the sub-humid savannah of Burkina Faso. Its dominant soil types include Lixisols, Gleysols, and Leptosols (IUSS Working Group WRB, 2015). The landscape Léo is at 11.07^∘ N and in the northern Guinean savannah of Burkina Faso and has Lixisols and Vertisols as the dominant soil types (IUSS Working Group WRB, 2015). The mean annual rainfall was approximately 1300 mm in Liliyo, and 900 mm in Tiéningboué, Midebdo, and Léo.

During July and August 2016, we sampled the soil from a total of 80 fields under yam cultivation across the four landscapes, i.e., 20 yam fields in each landscape. The fields were selected in advance by taking into account visual variation in soil color and texture across the landscape. The yam fields selected contained the maximum soil variability based on soil color and cropping history, taking into account both local farmers' knowledge on soil fertility and agronomic extension expertise. Yam is typically planted on soil mounds, ranging from 5000 to 10 000 mounds ha⁻¹ with a single yam plant per mound. Within each field, we sampled the soil at four adjacent mounds in square arrangement, which were spaced between 0.5 and 2 m. At each mound, six to eight auger cores (25 mm in diameter) to the 0.3 m depth were taken at a radius between 0.15 and 0.3 m away from the center of a mound, depending on the size of the mounds. Then the soils from the four mounds were combined into one composite sample per field (around 500 to 1000 g of soil).

An additional set of 14 composite soil samples was collected by the International Center for Research in Agroforestry (ICRAF) at Liliyo from one sentinel site called “Petit-Bouaké” (UNEP, 2012). Sampling took place between 25 and 29 August 2015 at positions that were previously selected for the Land Degradation Surveillance Framework (LDSF) in a spatially stratified manner (Vagen et al., 2010). The soil samples received from ICRAF were within the same landscape as the sampled soils in Liliyo within YAMSYS but sampled from different positions. All soil samples were air-dried and stored in plastic bags until further analysis.

2.2 Soil reference analyses

The air-dried soil samples were crushed and sieved at 2 mm. About 60 to 70 g of the sieved soil was oven-dried at 60^∘C for 24 h, of which 20 g was ball-milled. All chemical analyses except soil pH were conducted both on the soils sampled in yam fields (n=80) and the LDSF soils obtained from ICRAF (n=14).

The milled soils were analyzed for total C and macronutrient (N and S) concentrations using an elemental analyzer (vario PYRO cube, Elementar Analysensysteme GmbH, Germany). For each of the four landscapes, two soils were selected and analyzed based on three analytical replicates for quantifying within-sample variance of the elemental analysis. For the remaining samples, the analysis was not repeated. Sulfanilamide was used as a calibration standard for the dry combustion. For pH determination, 10 g of air-dried soil per sample was placed in a 50 mL Falcon tube, and 20 mL of de-ionized water was added. The samples were shaken in a horizontal shaker for 1.5 h and measured for pH using a pH electrode (Benchtop pH/ISE meter model 720A, Orion Research Inc., USA).

Bioavailable micronutrient (Fe, Mn, Zn, and Cu) concentrations in soils were determined with the diethylenetriaminepentaacetic acid (DTPA) extraction method, as described in Lindsay and Norvell (1978). The extracting solution consisted of 0.0005 M DTPA, 0.01 M CaCl₂, and 0.1 M triethanolamine. Briefly, 10 g of the sieved (< 2 mm) soils was extracted with 20 mL of DTPA solution. Micronutrient concentrations in the filtrates were measured by inductively coupled plasma optical emission spectroscopy (ICP-OES; using a Shimadzu ICPE-9820 plasma atomic emission spectrometer). Final DTPA-extractable concentrations of Fe, Mn, Zn, and Cu were calculated back to per kilogram of dry soil. For each landscape, two soils were selected and analyzed in triplicate to assess analytical errors. For the remaining soils the analysis was not repeated.

For each sample, the concentrations of total elements (Fe, Si, Al, K, Ca, P, Zn, Cu, and Mn) in the soil were assessed by energy dispersive X-ray fluorescence spectrometry (ED-XRF) measurements on 4 g of the milled soil with a SPECTRO XEPOS instrument (SPECTRO Analytical Instruments GmbH, Germany). The soil was mixed with an equal amount of wax using a ball mill and pressed into pellets. Exchangeable cations (Ca²⁺, Mg²⁺, K⁺, Na⁺, and Al³⁺) were determined with the BaCl₂ method (Hendershot and Duquette, 1986). About 2 g of the air-dried soil (< 2 mm) was extracted by shaking for 2 h with 30 mL of 0.1 M BaCl₂ on a horizontal shaker (120 cycles min⁻¹). The suspension was filtered through no. 40 filter paper (Whatman, Brentford, UK). For each landscape, two soils were analyzed in analytical triplicate. The concentrations of exchangeable cations in the BaCl₂ extract were measured by ICP-OES (using a Shimadzu ICPE-9820 plasma atomic emission spectrometer). Different BaCl₂ extract dilutions were used in order to obtain an optimal signal intensity for the quantification of specific elements across all samples. Concentration of H⁺ per kilogram of dry soil was calculated based on the pH measured in the BaCl₂ extractant. The BaCl₂ extraction does only slightly modify pH and is therefore an appropriate method to calculate effective CEC (CEC_eff) at native soil pH. Using the concentrations of the BaCl₂-extractable cations (i.e., Ca²⁺, Mg²⁺, K⁺, Na⁺, Al³⁺, and H⁺), CEC_eff was calculated as the sum of exchangeable cations in centimoles (cmol) of cation charge per kilogram of dry soil. Exchangeable acidity was defined by the sum of exchangeable Al³⁺ and H⁺. Base saturation in percent was calculated as a ratio of the sum of basic cations (Ca²⁺, Mg²⁺, K⁺, and Na⁺) in cmol(+) per kilogram of soil to the CEC_eff multiplied by 100.

Particle size analysis was conducted by the International Institute of Tropical Agriculture (IITA) in Cameroon, as described in Bouyoucos (1951). Briefly, 50 g of dried 2 mm sieved soil was stirred with 50 mL 4 % sodium hexametaphosphate and 100 mL of deionized water in a mixer, to break down the aggregates into individual particles. Readings with a hydrometer (ASTM 152 H, Thermco, New Jersey, USA) were taken after letting it stand in the suspension for 30 min. The silt content was calculated by subtracting the measured proportion of sand and clay from 100 %.

Spectroscopic measurements

The milled soils (n=94) were measured on a Bruker ALPHA DRIFT spectrometer (Bruker Optics GmbH, Ettingen, Germany), which was equipped with a ZnSe optics device, a KBr beamsplitter, and a DTGS (deuterated triglycine sulfate) detector. Mid-IR spectra were recorded between 4000 and 500 cm⁻¹ with a spectral resolution of 4 cm⁻¹ and a sampling resolution of 2 cm⁻¹. Reflectance (R) spectra were transformed to apparent absorbance (A) using $A={\log }_{\mathrm{10}}(\mathrm{1}/R)$ and corrected for atmospheric CO₂ using macros within the OPUS spectrometer software (Bruker Corporation, US). The spectra were referenced to a IR-grade fine ground potassium bromide (KBr) powder spectrum, which was measured prior to the first soil sample and measured every hour again. All spectra were recorded by averaging 128 scans (internal measurements) to improve the signal-to-noise ratio for each of the three independent replicate samples of each soil.

2.3 Spectroscopic modeling

2.3.1 Processing of soil spectra

Three replicates of spectra were averaged for each sample. The spectra were transformed by using a Savitzky–Golay-smoothed first derivative using a third-order polynomial and a window size of 21 points (42 cm⁻¹ at a spectrum interval of 2 cm⁻¹) (Savitzky and Golay, 1964). Prior to spectral modeling, Savitzky–Golay-preprocessed spectra were further mean-centered and scaled (divided by standard deviation) at each wavenumber.

2.3.2 Model development and validation

The measured soil properties were modeled by applying partial least squares regression (PLSR) (Wold et al., 1983) with the preprocessed spectra as predictors. The models were fitted using the orthogonal scores' PLSR algorithm. A 10-fold cross-validation, repeated five times, was performed to provide unbiased and precise assessment of PLSR model performance (Molinaro et al., 2005; Kim, 2009). For each individual soil property, the number of factors for the most accurate PLSR model was tuned separately. For each soil property model, the sample set was repeatedly randomly split into k=10 (approximately) equally sized subsets without replacement for all repeats $r=\mathrm{1},\mathrm{2},\mathrm{\dots },\mathrm{5}$ and all candidate values in the tuning grid with the number of PLSR factors (ncomp) $=\mathrm{1},\mathrm{2},\mathrm{\dots },\mathrm{10}$. Within each of the $r\times \mathrm{ncomp}=\mathrm{5}\times \mathrm{10}=\mathrm{50}$ resampling data set splits, each of the 10 possible held-out and model fitting set combinations (folds) was subjected to candidate model building at the respective ncomp, using $k-\mathrm{1}=\mathrm{9}$ out of 10 subsets, and remaining held-out samples were predicted based on the fitted models. The root mean square error (RMSE; Eq. 1) of the held-out samples was calculated by aggregating all repeated K-fold cross-validation predictions (${\widehat{y}}_i$) and corresponding observed values (y_i) grouped by ncomp, which resulted in a cross-validated performance profile RMSE vs. ncomp.

$$\begin{array}{}\text{(1)}& \mathrm{RMSE}=\sqrt{{\displaystyle \frac{{\sum }_{i=\mathrm{1}}^n{\left({\widehat{y}}_i-y_i\right)}^{\mathrm{2}}}{n}}}\end{array}$$

Based on this performance profile, the minimal ncomp among the models, whose performance was within a single standard error (“one standard error rule”; Breiman et al., 1984) of the lowest numerical value of RMSE, was selected.

Model assessment was done with the best factors for each property using cross-validation hold outs. We reported the cross-validated measures' RMSE, R² (coefficient of determination) obtained via linear least squares regression, and ratio of performance to deviation (RPD), after averaging predictions across repeats. The RPD index is the ratio of the chemical reference data standard deviation (s_y) to the RMSE of prediction.

$$\begin{array}{}\text{(2)}& \mathrm{RPD}={\displaystyle \frac{s_y}{\mathrm{RMSE}}}\end{array}$$

Besides calculating the above listed performance measures, the uncertainty of spectral estimates was graphically reported for each soil sample, using prediction means and 95 % confidence intervals derived from cross-validation repeats ($n=r=\mathrm{5}$; Eqs. 3 and 4).

$$\begin{array}{}\text{(3)}& {\displaystyle }{S}_n^{\mathrm{2}}={\displaystyle \frac{\mathrm{1}}{n-\mathrm{1}}}\sum _{i=\mathrm{1}}^n{\left(y_i-\stackrel{\mathrm{‾}}{\widehat{y_i}}\right)}^{\mathrm{2}}\text{(4)}& {\displaystyle }\stackrel{\mathrm{‾}}{\widehat{y_i}}\pm t\left(n-\mathrm{1},\mathrm{1}-\mathit{\alpha }/\mathrm{2}\right){\displaystyle \frac{S_n}{\sqrt{n}}};\mathit{\alpha }=\mathrm{0.05}\end{array}$$

To cover the full training data space in the models for future sample predictions, the final PLSR models were rebuilt using the entire training set and the respective values of the optimal final number of PLSR components determined by the procedure described above.

2.3.3 Model interpretation

The mid-IR spectra contain complex information about soil composition and properties. To establish a predictive relationship, statistical models need to find relevant spectral features for each soil property. Model interpretation requires a variable importance assessment to decide on the contribution of spectral variables to prediction and to explain spectral mechanisms. Therefore, we conducted model interpretation based on the variable importance in projection (VIP) method (Wold et al., 1993; Chong and Jun, 2005), using the model at the respective best number of factors (ncomp). The VIP measure v_j was calculated for each wavenumber variable j as

$$\begin{array}{}\text{(5)}& v_j=\sqrt{p\sum _{a=\mathrm{1}}^A\left[{\mathrm{SS}}_a\right(w_{aj}/\Vert w_{aj}\Vert {)}^{\mathrm{2}}]/\sum _{a=\mathrm{1}}^A({\mathrm{SS}}_a)},\end{array}$$

where w_aj are the PLSR weights for the ath component for each of the wavenumber variables, and SS_a is the sum of squares explained by the ath component:

$$\begin{array}{}\text{(6)}& {\mathrm{SS}}_a=q_a^{\mathrm{2}}{t}_a^T{t}_a,\end{array}$$

where q_a are the scores of the predicted variable y, and t_a are the scores of the predictors X. These VIP scores account for multicollinearity found in spectra and are considered to be a robust measure to identify relevant predictors. Important wavenumbers were classified with a VIP score above 1. A variable with VIP above 1 contributes more than the average to the model prediction. For model interpretation, we only computed VIP at the respective finally chosen number of PLS (partial least squares) components a_final for each considered model. We focused on a selection of three well-performing models with R² ≥ 0.8 (RPD ≥ 2.3) to illustrate model interpretation. These were total C, total N, and clay content.

2.4 Statistical software

The entire analysis was performed using the R statistical computing language and environment (version 3.6.0) (R Core Team, 2017). We used the pls (Mevik et al., 2019) package for PLSR, as described by Martens and Naes (1989). Cross-validation resampling, model tuning, and assessment was done using the caret package (Kuhn et al., 2019). Custom functions from the simplerspec package were used for spectroscopic modeling (Baumann, 2019). All data and code to reproduce the results of this study are available online via Zenodo (Baumann, 2020).

3 Results

Measured properties and mid-IR estimates of yam soils

The distribution of soil properties of the yam fields showed a wide variation across the landscapes (Fig. 1). Total C concentrations across all fields ranged from 2.4 to 24.7 g C kg⁻¹. Total C values at the landscape scale were the lowest (median) in Léo and the highest in Tiéningboué. Soils from yam fields in the two landscapes from Côte d’Ivoire (13.0 ± 5.4 g C kg⁻¹; mean ± standard deviation) had relatively higher total C compared with the fields in the landscapes in Burkina Faso (6.1 ± 3.6 g C kg⁻¹). The median value and variation of CEC_eff exhibited similar patterns across the landscapes to total C. Total N concentrations across all fields ranged from 0.18 to 2.48 g N kg⁻¹. Total N within and across the four landscapes exhibited a similar pattern to total C. Generally, the landscapes in Burkina Faso were low in total N compared to those from Côte d’Ivoire (0.44 ± 0.24 g N kg⁻¹ vs. 1.09 ± 0.46 g N kg⁻¹). Median total N concentrations were almost identical for Liliyo and Tiéningboué, with 1.1 g N kg⁻¹. Total S concentrations varied between 41 and 242 mg S kg⁻¹ across all fields and showed a similar pattern to total C and N. The yam fields in the landscapes of Burkina Faso had on average more than 2 times higher total S than the other landscapes. Total P concentrations were in a similar range for the landscapes Léo, Midebdo, and Liliyo. In Tiéningboué, total P values were almost 2 times higher than the other fields (817 mg S kg⁻¹ vs. 453 mg P kg⁻¹), with more within-landscape variation.

Figure 1 Reference measurements of soil chemical properties. Léo and Midebdo are two yam-growing landscapes in Burkina Faso, and Liliyo and Tiéningboué are in Côte d'Ivoire. The chemically analyzed soils (n=94) originated from 20 yam fields per landscape and 14 additional soils from the Liliyo region were provided by the World Agroforestry Center (ICRAF). C is carbon, N is nitrogen, P is phosphorus, Fe is iron, Al is aluminum, Si is silicon, Ca is calcium, Zn is zinc, Cu is copper, K is potassium, and Mn is manganese. Bioavailable micronutrients were measured by the diethylenetriaminepentaacetic acid (DTPA) extraction method. Ca(exch.), Mg(exch.), K(exch.), and Al(exch.) signify exchangeable elements determined with BaCl₂ extraction. CEC_eff is the effective cation exchange capacity, and BS_eff is the effective base saturation. The number of soils analyzed for each individual property is indicated above the 75 % percentile.

The concentrations of total Fe, total Al, total Ca, total Zn, and total Cu in the soil tended to be higher for the landscapes in Côte d’Ivoire than in Burkina Faso (Fig. 1). To give an example, median concentrations of total Ca were 2.16 g Ca kg⁻¹ in fields sampled from the Tiéningboué region and similar in Liliyo (i.e., 1.90 g Ca kg⁻¹), while they were markedly lower in Léo and Midebdo (i.e., 0.90 vs. 1.26 g Ca kg⁻¹). In general, the ranges for total micronutrient contents were more variable in the landscapes of Côte d’Ivoire (e.g., range = 14.0–57.0 mg Zn kg⁻¹ in Liliyo; lowest range in Léo = 12.2–19.7 mg Zn kg⁻¹). Total K concentration was highly variable within and across the landscapes (overall range = 0.5–34.1 g K kg⁻¹) and lowest in Midebdo (range = 0.9–8.9 g K kg⁻¹), while the highest total K median was measured in yam fields of Léo (range = 4.1–25.0 g K kg⁻¹).

Median extractable Fe and its interquartile ranges were comparable across the landscapes (see Fig. 1). However, there were some fields where extractable Fe reached values higher than 100 mg Fe kg⁻¹. Median extractable Zn values showed a similar pattern to total C, with the highest median values and interquartile range in Tiéningboué, and had the lowest in Léo. In comparison, the highest median values and interquartile range of extractable Cu and Mn were found in Liliyo. For extractable Zn, Cu, and Mn, median values and interquartile range were higher in the two landscapes in Côte d’Ivoire than the two landscapes in Burkina Faso.

Across all samples and landscapes, soil pH varied between 4.7 and 8.4. Median pH was comparable in Tiéningboué (i.e., 6.4), Liliyo (i.e., 6.5), and Midebdo (i.e., 6.5). Median pH of yam fields in Léo (i.e., 6.0) was lower than in the other landscapes. Exchangeable K, Ca, and Mg concentrations showed similar patterns across the four landscapes. In Burkina Faso, each of the exchangeable cations showed relatively low median concentrations across the fields and less landscape-level variation than in Côte d’Ivoire. In general, the highest median and variation of exchangeable cations among the landscapes were measured in the yam field soils of Tiéningboué. Median exchangeable Al values were comparable among the landscapes, although there were some outliers with exchangeable Al > 20 mg kg⁻¹ for Midebdo, Liliyo, and Tiéningboué. The CEC_eff ranged from 0.9 to 14.6 cmol(+) kg⁻¹ across all fields and landscapes. Median CEC_eff tended to decrease in the following order across landscapes: Léo > Midebdo > Liliyo > Tiéningboué. The interquartile range of CEC_eff was also the greatest in Tiéningboué and the smallest in Léo.

Reference measurements for total N, S, exchangeable Ca, exchangeable Mg, and CEC_eff. were closely correlated with total C (Fig. 2; $\mathrm{0.71}\le r\le \mathrm{0.92}$ (CEC_eff.)). Also, total Ca, Al, and clay content correlated closely with total C (r>0.70). Clay contents were weakly related to silt (r=0.21), while sand had a markedly negative relationship with silt ($r=-\mathrm{0.89}$). Bioavailable Zn (DTPA) was covarying with both CEC_eff. (r=0.58) and total Zn (r=0.59). Bioavailable Cu (DTPA) had a strongly positive association to total Cu (r=0.90). Exchangeable K (BaCl₂) had the strongest relationship with total C and CEC_eff. (r=0.63, and r=0.64).

Figure 2 Correlation matrix of soil properties measured on each of the 20 soils sampled from individual yam fields per landscape and 14 additional agricultural soils received from the World Agroforestry Center (n=94; see Fig. 1 for further details and abbreviated chemical properties). Pearson correlation coefficients (r) were rounded to one decimal point.

3.1 Soil mid-IR spectroscopic models

Among the measured soil properties, mid-IR PLSR models for total K (R²=0.96) and total Al (R²=0.97) performed the best (Table 1). Out of a total of 27 soil attributes, 11 were well quantified by the models ($R_{\mathrm{cv}}^{\mathrm{2}}\mathit{⩾}\mathrm{0.75}$; Fig. 3). The confidence intervals derived from cross-validation prediction were very narrow, showing that all PLSR models were stable. Within this group of stable models, four soil attributes are directly related to the mineralogy (total Fe, Al, K, and Ca), three are related to soil organic matter (total C, N, and S), one is related to texture (clay fraction), one is related to plant nutrition (exchangeable Fe), and two are related to mineralogy and plant nutrition (exchangeable Ca and CEC_eff). More specifically, total C was accurately predicted, with an R² of 0.92 and a RMSE of 1.6 g C kg⁻¹. The models were also able to predict total N well (R²=0.89; RMSE = 0.16 g N kg⁻¹). Prediction accuracy of total S was slightly lower than for total C, but its goodness-of-fit and RMSE suggest that the model was reliable for prediction. However, exchangeable K (R²=0.28) and BS_eff (R²=0.24) were poorly predicted (Table 1). Predictions for percent clay were reliable (R²=0.81; RMSE=2.1 %), whereas predictions for percent sand (R²=0.45; RMSE=8.1 %) and percent silt (R²=0.41; RMSE=6.5 %) were not accurate. Finally, chosen models of all soil attributes had between one and nine PLSR components.

Table 1 Descriptive summary of measured (meas.) soil reference data (see Fig. 1) and evaluation results of cross-validated PLSR models. All samples across the four landscapes were aggregated into a single model per respective soil property. Model evaluation was done on held-out predictions of 10-fold cross-validation (cv) repeated five times at the finally selected number of PLSR components (ncomp). CV is the coefficient of variation, RMSE is the root mean square error, and RPD is the ratio of performance to deviation. C is carbon, N is nitrogen, P is phosphorus, Fe is iron, Al is aluminum, Si is silicon, Ca is calcium, Zn is zinc, Cu is copper, K is potassium, and Mn is manganese. Bioavailable micronutrients were measured by diethylenetriaminepentaacetic acid (DTPA) extraction. Ca(exch.), Mg(exch.), K(exch.), and Al(exch.) signify exchangeable elements determined with BaCl₂ extraction. CEC_eff is the effective cation exchange capacity, and BS_eff is the effective base saturation.

Figure 3 Cross-validated predictions of soil properties derived from best mid-infrared (mid-IR) partial least squares regression (PLSR) models vs. laboratory reference measurements (see Fig. 1). Average estimates, their confidence intervals (error bars), and evaluation metrics were derived with 10-fold cross-validation repeated five times. “ncomp” is the number of PLSR components of most accurate final models, RSME is the root mean square error, and RPD is the ratio of performance to deviation. Only soil properties modeled with R²>0.75 are shown. CEC_eff is the effective cation exchange capacity. Exchangeable (exch.) elements were determined with BaCl₂. Bioavailable Fe was determined via diethylenetriaminepentaacetic acid (DTPA) extraction.

3.1.1 Model interpretation

A large proportion of absorptions had VIP>1 for each of the total C, total N, and clay models (Fig. 4). Important wavenumbers (VIP>1) for total C were mostly between 3140 and 1230 cm⁻¹. Besides clear absorption peaks, there were relatively continuous spectral features that were important to the models. For example, the relatively continuous and smooth spectral region between the alkyl C−H vibrations at 2855 and 2362 cm⁻¹ had a comparable contribution to the model as peak regions associated with total C prediction. The VIP patterns across wavenumbers were almost identical for total C and N models, and its reference measurements were strongly correlated (r=0.94; Fig. 2). In contrast, the clay content model deviated from the total C model in particular regions, for example around the kaolinite OH− feature at 3620 cm⁻¹ or at kaolinite $\mathrm{Al}-\mathrm{O}-\mathrm{H}$ vibrations at 934 and 914 cm⁻¹.

Figure 4 Variable importance analysis of partial least squares regression (PLSR) models for the concentrations of total soil C and total N and clay content, including overlaid raw and preprocessed spectra. The top panel shows resampled mean sample absorbance spectra (n=94). Prominent peaks were identified as local maxima with a span of 10 points 20 cm⁻¹ for the selected wavenumbers. Fundamental mid-IR vibrations that are well described in the literature (e.g., Madejová et al., 2002; Rossel and Behrens, 2010; Stevens et al., 2013) were added as labels when identified peaks matched literature assignments. (Q) stands for quartz and (K) for kaolinite. The middle panel depicts preprocessed spectra (Savitzky–Golay first derivative with a window size of 21 points (42 cm⁻¹); third-order polynomial fit). The bottom panel shows variable importance in the projection (VIP) for three selected well-performing PLSR models (total C, total N and clay; R²>0.81). The horizontal black line at VIP=1 indicates the threshold above which absorbance at a given wavenumber contributes more than the average (wavenumber) to the spectral variance explained of a certain soil property. Dashed points closely below the y=0 line of the VIP graph visualize positive (above y=0) and negative (below y=0) PLSR β coefficients.

4 Conclusions

We developed models with mid-IR spectra to estimate soil chemical and physical properties relevant to the production of yam and other staple crops in four landscapes in the yam belt of West Africa. We tested the models for the important soil properties that are applied widely for agronomic performance evaluation. We showed that mid-IR spectroscopy models have the potential for the cost-effective and rapid determination of the distribution and variability of important soil properties across highly variable yam production landscapes in West Africa. Specifically, total C, total N, total S, total Fe, total Al, total K, total Ca, exchangeable Ca, CEC_eff, bioavailable Fe, and clay content can be quantified with RPD>2 and R²>0.75 when aiming to predict in the range of soil property values found in the environmental conditions covered by this study. We achieved spectral estimates with quite small uncertainties that are typically reported for libraries at the geographical extent of a field or farm. The correlation analysis of measured values together with spectral inference helps improve our understanding of how soil properties are interrelated with soil functional composition. This study delivered parsimonious, unbiased, and accurate mid-IR spectroscopy-based models to monitor and predict soil quality and to manage crop nutrition. Hence, we envision this pilot study as being a starting point to continuously update and adapt the mid-IR model library for more efficient site-specific and agronomically relevant soil estimates in the West African yam belt. This can create a better capacity to diagnose and monitor soils in the long term compared with traditional wet chemistry and will hopefully ameliorate the soil conditions for sustainably meeting the demand of yam and other important staple crops in the regions.