An assessment of the variation of soil properties with landscape attributes in the highlands of Cameroon

2.1 Study site and experimental design

The study site is located in the western highlands of Cameroon (Figure 1). The area is characterized by diverse and mosaic landscape with complex topography, varying vegetation patterns, heterogeneous attributes and is dominated by intense agricultural and livestock activities. The elevation of the site ranged from 1,300 to 1,700 m, and mean annual precipitation ranged from 1,500 to 1,800 mm. The vegetation is savannah type with montane forests and patches of montane grassland, and subalpine grasslands and shrublands located in waterways, watersheds, and protected areas. The cropping systems can best be described as multiple cropping with mixed inter-cropping of crops and trees, and crop yields vary greatly with management practices and the location of the plot along the toposequence. Data were collected using the Land Degradation Surveillance Framework, which is designed to provide a biophysical baseline at the landscape level, and a monitoring and evaluation framework for assessing land degradation and recovery over time (Vågen, Winowiecki, Tamene, & Tondoh, 2010). The experimental design was based on a hierarchical survey approach in a site of 10 × 10 km (100 km²). The site was subdivided into 16 clusters, and 10 sampling plots (1,000 m² each) were randomly distributed in each cluster, giving a total of 160 sampling plots. Each sampling plot contained four subplots (100 m² each; Takoutsing et al., 2016; Vågen et al., 2010).

Four landscape attributes were selected for this study: soil depth, land use type, slope gradient, and topographic position. These attributes were selected because previous research has demonstrated that they can significantly affect soil properties (Fernández-Romero et al., 2014; Mora et al., 2014; Ritchie et al., 2007; Sun et al., 2014; Takoutsing et al., 2015; Takoutsing et al., 2017). Data were collected at the levels of plots and subplots. The 160 sampling plots were characterized in terms of geographical location (longitude, latitude, and elevation using a GPS), land use type (classes = cropland, fallow, forest, grassland, or pasture), slope gradient (classes = level [<10%], sloping [10–20%], or steep [>20%]), and topographic position (classes = bottomland, footslope, midslope, ridge/crest, or upland). Data for trees (woody plants with height > 3 m) and shrubs (woody plants with height 1.5–3 m) were recorded at the subplot level. Tree and shrub densities were estimated by counting all stems in the subplot. Height and diameter at breast height were measured on trees, and height, crown length, and width were measured on shrubs. Percent cover of woody and herbaceous vegetation was rated on a 6-point scale: absent, <4%, 4–15%, 15–40%, 40–65%, and > 65%. Soil samples were collected at two depths (topsoil [0–20 cm] and subsoil [20–50 cm]) in the subplots using an auger. For each depth, the samples from the four subplots were pooled together and thoroughly mixed to obtain the homogenous composite sample for each of the 160 sampling plots. Three subsoil samples were not obtained because of depth restriction, so the total number of soil samples was 317 (160 for topsoil and 157 for subsoil).

2.2 Soil analytical procedures

Conventional laboratory analyses methods were used to analyze 10% of the samples for SOC, texture, ExBas, electric conductivity (EC), boron (B), manganese (Mn), phosphorus (P), and soil reaction (pH). These were previously selected as important site-specific soil properties for the study site, and they can be used to rapidly assess the quality of soils to guide land management decisions and indirect measures of soil functions (Takoutsing et al., 2016). The pH was analyzed in a 1:1 H₂O solution. Exchangeable bases, phosphorous, manganese, and boron were analyzed using a Melich-3 extraction procedures, whereas EC was determined using a 1:2 soil : water extract at the Crop Nutrition Laboratory (www.cropnuts.com) in Nairobi, Kenya. Soil organic carbon and total carbon (%) were determined by chromic acid digestion and spectrophotometric analysis. Soil texture was measured with a laser diffraction particle size analyzer at the World Agroforestry Centre Plant and Soil Spectroscopy Laboratory in Nairobi, Kenya. Then all samples were analyzed using mid-infrared reflectance based on methods described in Terhoeven-Urselmans, Vagen, Spaargaren, and Shepherd (2010) and Hengl et al. (2015). Infrared spectroscopy has been demonstrated to be a rapid, affordable, and non-destructive method for analyzing a range of materials (Reeves & Smith, 2009; Shepherd & Walsh, 2002; Viscarra Rossel, DJJ, AB, Janik, & Skjemstad, 2006).

2.3 Statistical analyses

Descriptive statistics (mean, median, standard deviation, etc.) were calculated for the soil properties. The general linear model was used to test the effects of landscape attributes on soil properties. The general linear model included the following sources of variation: soil depths, land use types, slope gradients, and topographic positions as main effects (treated as fixed factors), and two-way and three-way interactions between the main effects. Significant differences in means of soil properties between the classes of each landscape attribute were assessed by Tukey's honestly significant difference test. These analyses were done using R software version 2.15.0 (R Development Core Team, 2013).

Relationships between soil properties and landscape attributes were further explored using redundancy analysis (RDA). Developed by Hotelling (1936), RDA has been considerably used in ecology (Bowles, Acosta-Martínez, Calderón, & Jackson, 2014; Campos-Herrera, El-Borai, Rodríguez Martín, & Duncan, 2016; Gutiérrez et al., 2016; Rodríguez Martín et al., 2014; Rousseau, Rioux, & Dostaler, 2006) to assess relationships between a set of dependent variables and a set of explanatory variables. One of the advantages of RDA is that it integrates the techniques of ordination and multiple regression (Borcard, 2006; Legendre & Legendre, 1998; van den Wollenberg, 1977). In this study, RDA was used to identify the combination of landscape attributes that best explains the variation in soil properties. Canonical ordinations are usually carried out with standardized explanatory variables, which are not always dimensionally homogeneous (Borcard, 2006). In this context, the soil properties were transformed to standard normal variables with mean = 0 and variance = 1 (transformed variable = [untransformed variable-mean]/standard deviation). The results of RDA were graphically presented with a bi-plot of the dependent variables (soil properties) and the explanatory variables (landscape attributes). RDA was done using XLSTAT software (Addinsoft Version 2012.2.02).

3.1 Ecological characterization

The study site is a mosaic landscape with diverse physical features and ecological conditions that influence soil properties. Land suitable and favorable for agricultural production based on slope classification (slope < 10%) accounts for 44% of the site. The distribution of land use types based on the 160 observations plots illustrates that 88% of the area is either cultivated or under fallow, suggesting that farmers exploit nearly every hectare of land, including those with steep slopes (slope > 20%). Uncultivated plots representing 12% of the area are either unsuitable for agriculture, protected areas, have very steep slopes, or are not easily accessible to farmers. The annual herbaceous cover including crops varied between 15% and 40% for cultivated plots and between 40% and 65% for uncultivated plots. Mean tree and shrub densities were estimated at 140 trees ha⁻¹ and 190 shrubs ha⁻¹ whereas the median values were lower (134 trees ha⁻¹ and 159 shrubs ha⁻¹), indicating variation among land use types. Higher tree and shrub densities were found in uncultivated plots. The high woody cover observed across the area is due to the fact that trees are maintained or planted for a range of products and services including food, farm demarcation, land ownership, and as a legacy for future generations.

3.2 Analysis of variance of soil properties

Analysis of variance (Table 1) indicated that the majority of soil properties were highly affected by all the landscape attributes, but less affected by their interactions. The effects of landscape attributes on soil properties were not significant in only four cases: soil depths on pH, slope gradients on EC and pH, and topographic positions on B. Significant interactions existed between land use types and topographic positions for SOC, EC, ExBas, and pH, between slope gradients and topographic positions for pH, and between land use types and slope gradients for Mn. These interactions mean that the effects of some landscape attributes on these soil properties differed depending on the other landscape attributes. For example, pastures at midslope positions had lower values of SOC compared with upland plots probably due to the effects of erosion, and croplands in the bottomlands and footslopes had higher values of SOC compared with croplands located in the ridge/crest. Fallows had higher values of SOC at bottomland and footslope positions and lower values at upland positions (Figure 2).

In general, a consistent trend was observed for soil depths (Table 2), as all indicators with the exception of clay content were significantly higher in the topsoil. This is probably the result of higher organic matter content (not measured in the study) or the fine texture of the clay-rich soils that slows percolation and reduces leaching (Yimer, Ledin, & Abdelkadir, 2007). In addition, the influence of biological activities (fauna and flora) and management (additional inputs such as organic and inorganic fertilizers) are more pronounced on the upper soil layers, resulting in higher values for most nutrients.

Mean values of most soil properties with the exception of SOC tended to be higher in forests and grasslands compared with other land use types and in plots located in bottomlands and footslopes compared with other topographic positions (Table 2). The lower values observed in croplands and pasture are attributed to the influence of anthropogenic activities such as farming and intensive grazing that reduce organic matter inputs into the soils and reduce vegetation cover (Takoutsing et al., 2013). Higher clay content observed in croplands and pastures is due to reduced vegetation cover caused by anthropogenic activities including agriculture and livestock rearing (Liu, Zhang, Chang, Kan, & Lin, 2012). Clay content did not differ significantly between bottomland and midslope positions, but was higher in ridge/crest and upland positions, probably as the result of the effects of water erosion.

Differences observed in soil properties with topographic positions may be attributed to the combined effects of topography on soil moisture content, depth of soil formation, erosion, and deposition processes. In the study area, different topographic positions have similar vegetation cover, so variation in the values of soil properties at different positions is attributed to differences in soil erosion rates (not quantified in this study), or soil development and deposition processes. Nutrients in the ridge/crest plots are eroded by runoff and deposited in the footslopes and bottomlands, therefore influencing the nutrient concentrations (Figure 3).

The effects of slope gradients on soil properties were inconsistent and did not agree with our hypothesis. Some soil properties were higher on steep plots (e.g., B and Mn) whereas others were higher on level plots (e.g., SOC and ExBas). From observations, slope gradient classes have similar vegetation cover, and land use types are distributed irrespective of the slope. It was generally observed that the intensity and type of management applied to a plot depended on its accessibility by farmers or animals and its proximity to the households. Perhaps the accessibility and proximity to households account for the inconsistent results, but we did not quantify or test these factors in this study.

3.3 Relationships between soil properties and landscape attributes

The RDA revealed several sets of significant relationships between soil properties and landscape attributes (Figure 4). The first two ordination axes (F1 and F2) explained about 92% of the effects of landscape attributes on soil properties (66% from the first axis and 26% from the second axis). The black arrows represent soil properties, and the different color points represent the landscape attributes. The arrows indicate the direction of the change in soil properties, and the lengths of the arrows indicate the extent to which the soil parameters are affected by the landscape attributes. In RDA, most of the soil properties were related to the first axis in quadrants Q3 and Q4, whereas clay content was associated with quadrant Q2 and negatively related to the rest of the indicators (Figure 4). This can be interpreted by the fact that clay content is influenced by soil depth with high values in the subsoil and by the local-scale influences associated with land use (cropland). According to the groupings of RDA, soil properties were divided to assess the effects of landscape attributes: These groups are discussed below.

3.5.Implication for ecological intensification in complex topographic landscapes

A worldwide concern is how to meet the increased demand for ecosystem services caused by the rapidly increasing human population. Because there is little if any scope for agricultural land expansion while also preserving natural forests, meeting the demand can only be achieved through the optimal utilization of available ecosystem services during the agricultural production processes, referred to as ecological intensification (Bommarco, Kleijn, & Potts, 2012). Ecological intensification of agricultural practices enhances productivity through sustainable management practices including intercropping, crop rotations, system diversification, and low chemical inputs (Kovács-Hostyánszki et al., 2017). Quantifying the effects of natural processes on the implementation of ecological intensification presents challenges that are more pronounced in areas with complex topography such as the highlands of Cameroon, where agricultural practices and land management interventions are influenced by climate, soil types, and topography. A holistic intensification cannot be achieved without considering the potential of a given soil to fulfill its functions.

The results of our study revealed that some of the soil properties responded significantly to the types of field management practices, location of the plot along the toposequence and vegetation cover, whereas others such as clay content and pH were less affected by these landscape attributes. Previous studies have reported similar and mixed results and attributed the results to the physical features of the landscapes (Gao, He, Yu, Chen, & Wang, 2014; Sun et al., 2015; Takoutsing et al., 2015). Another study in the same site demonstrated significant spatial variation and well-defined patterns of higher values for key soil functional properties in lowlands and valleys, as well as in zones with high and permanent vegetation cover (Takoutsing et al., 2017). In order to understand this variation and address the challenges faced by ecological intensification, further studies are needed at various scales to develop and test site-specific options for sustainable intensification (Takoutsing et al., 2017; Winowiecki et al., 2016). In addition, there is also a need for systematic assessment of the spatial variability of soil properties within landscapes to generate information relevant for agricultural policy and evidence-informed land management at both plot and landscape levels (Shepherd, Shepherd, & Walsh, 2015; Winowiecki et al., 2016). Soil scientists need to take into account local knowledge in formulating hypotheses on soil functioning that determine soil health, and then set up an approach that is integrative, systems-oriented, and multidisciplinary to ensure sustainability of soil health. In this context, it is fundamental for all stakeholders promoting agricultural intensification to apply low-cost and environmental friendly methods that mitigate nutrient loss through leaching and runoff, soil and water contamination, water erosion, and minimize the production of greenhouse gases. Researchers should investigate how to enhance terrestrial ecosystem services through ecological processes. In addition, efforts are also required to set up effective capacity building programs at various levels to ensure accumulation of necessary baseline data and information on land health necessary to design appropriate interventions as soil health is the first requirement for agricultural development and environmental sustainability (Allen, Singh, & Dalal, 2011). Because the definition of soil health cannot be generalized due to site specific contexts, key soil functional properties to assess the quality for a given soil should be selected based on both anthropogenic and environmental factors. The selection based on one-time field measurement does not suffice, and multiyear sampling is required. If possible, biological properties related to the microbial and faunal activities in soils should be incorporated in the set of properties. Another important challenge for researchers is how to make the research findings relevant to small-scale farmers. Emphasis has been given to influencing policy and decision makers with the hope that there would be positive impacts on smallholder farming systems (Ramisch, 2005; Scoones & Toulmin, 1998). None of the previous approaches has targeted farmers directly and is readily applicable in the farmers' context. This is a limitation for most of the results generated by this and other studies.