Geoderma 195–196 (2013) 192–200

Contents lists available at SciVerse ScienceDirect

Geoderma

j ourna l homepage: www.e lsev ie r .com/ locate /geoderma

Evaluating indicators of land degradation in smallholder farming systems ofwestern Kenya

Boaz S. Waswa a,⁎, Paul L.G. Vlek a, Lulseged D. Tamene b, Peter Okoth c, David Mbakaya d, Shamie Zingore e

a Center for Development Research (ZEF), University of Bonn, Germanyb International Center for Tropical Agriculture (CIAT), Lilongwe, Malawic International Center for Tropical Agriculture (CIAT), Nairobi, Kenyad Kenya Agricultural Research Institute (KARI), Kakamega, Kenyae International Plant Nutrition Institute (IPNI), Nairobi, Kenya

⁎ Corresponding author at: Centre for DevelopmentBonn, Walter-Flex-Str. 3 53113, Bonn, Germany. Fax: +

E-mail addresses: bswaswa@yahoo.com, bwaswa@u

0016-7061/$ – see front matter © 2012 Elsevier B.V. Allhttp://dx.doi.org/10.1016/j.geoderma.2012.11.007

a b s t r a c t

a r t i c l e i n f o

Article history:Received 9 May 2012Received in revised form 14 November 2012Accepted 21 November 2012Available online 7 January 2013

Keywords:Integrated soil fertility management (ISFM)Land degradationPrincipal component analysis (PCA)Soil fertility indicatorsSoil variabilitySustainable land management (SLM)

Understanding the patterns of land degradation indicators can help to identify areas under threat as basis fordesigning and implementing site-specific management options. This study sort to identify and assess thepatterns of land degradation indicators in selected districts of western Kenya. The study employed the useof Land Degradation Sampling Framework (LDSF) to characterize the sites. LDSF a spatially stratified,random sampling design framework consisting of 10 km×10 km blocks and clusters of plots. The studybroadly identified and classified the indicators and attributes of land degradation into soil and site stability,hydrologic function and biotic integrity. Assessment of general vegetation structure showed that over 70%of the land was under cropland with forests accounting for 8% of the area. Sheet erosion was the majorform of soil loss. High variability was observed for the soil properties and this can be due to both inherentsoil characteristics as well as land management practices. There was distinct variation in the soil propertiesbetween the topsoil (0–20 cm) and the subsoil (20–30 cm) with the topsoil having higher values for mostof the parameters compared to the subsoil. Using coefficient of variation (CV) as criteria for expressing var-iability, Ca, TON, Mg, SOC and silt were most variable soil properties for the 0–20 cm depth. Moderate vari-ability (CV 0.15–0.35) was observed for CEC, P, K and clay while Na, Sand and pH had the least variability(CVb0.15). For the subsoil (20–30 cm), Ca, Mg and silt were the most variable. About 94% of the farms sam-pled were recorded to have very strongly acidic soil levels (pH 4.5–5.5) while 6% of the farms had moderatelyacidic soil levels (pH 5.6–6.0). Over 55% of the farms had low (b2%) total organic carbon levels and this variedwith land use. Soils with SOM below this ‘critical level’ are at a threat of degradation if not well managed. Theprincipal component analysis (PCA) identified three main explanatory factors for soil variability: ‘soil fertilitypotential’, ‘soil physical properties’ and ‘available P’. Improving productivity of land therefore calls for theadoption of integrated soil fertility management (ISFM) options as a strategy to ensuring nutrient availabilitywhile at the same time building the natural nutrient reserve through soil organic matter build up.

© 2012 Elsevier B.V. All rights reserved.

1. Introduction

Information on variability of soil properties can aid to map outpatterns of land degradation and to guide decisions on selection ofappropriate restoration options. However, ecological processes aredifficult to observe in the field due to the complexity of ecosystems(Pellant et al., 2005). To determine whether the land is degraded ornot, direct assessment can be done based on different criteria suchas soil stability, vegetation, nutrient cycling and many other aspects(NRC, 1994). This therefore calls for the defining and use of attributesand indicators that may take a qualitative or quantitative form.

Research (ZEF), University of49 228/73 18 89.

ni-bonn.de (B.S. Waswa).

rights reserved.

Classifying or rating the attribute indicators along ordinal orcategorical scales is often used to capture the degree of departurefrom expected levels for each indicator and hence level of degrada-tion. Spatial variability in landscapes arises from a combination ofintrinsic and extrinsic factors while temporal variability is causedmainly by changes in soil characteristics and rainfall patternsover time (Rao and Wagenet, 1985). Intrinsic spatial variabilityrefers to natural variations in soil characteristics, often a result ofsoil formation processes, such as weathering, erosion, or deposi-tion processes, and variability in organic matter content due tothe architecture of native plant communities (Zacharias, 1998).On the other hand, extrinsic spatial variability refers to the varia-tions caused by lack of uniformity in management practices suchas chemical application, tillage, and irrigation (Vieira et al., 2002;Zacharias, 1998).

Fig. 1. Study site and the sampling blocks.

193B.S. Waswa et al. / Geoderma 195–196 (2013) 192–200

Awareness and interest by scientists in the spatial variation of fieldsoils date back to the early 1990s (Pendleton, 1919; Smith, 1938) butit is only until the 1960s and 1970s that field scientists began to studysoil variability in a systematic way (Beckett and Webster, 1971;Webster, 1994). With growing interest in soil variability, the term“Pedometrics” was coined at an international workshop of theInternational Soil Science Society in Wageningen, The Netherlandto describe the quantitative study of the variation of field soils(Webster, 1994). Over the years, various authors have used attri-butes and indicators to assess land degradation. For example,Pyke et al. (2002) identified 17 indicator attributes, and amongthem are the types of erosion, plant community composition, litteramount, annual production and invasive plants to assess rangelanddegradation patterns in the US. Other studies such as the LandscapeFunction Analysis (LFA) based the assessments on processes in-volved in surface hydrology, i.e. rainfall, infiltration, runoff, ero-sion, plant growth and nutrient cycling (Tongway, 2010).

Soil quality indicators can be identified using a wide range ofstatistical techniques including rating factors based on soil-relatedconstraints to crop productions (Lal, 1994), standardized scoring func-tions based on threshold functions (Karlen and Stott, 1994), linear ormultiple regression analysis (Doran and Parkin, 1994; Li and Lindstrom,2001) and principal component analysis (PCA) (Adhikari et al., 2011;Kosaki and Juo, 1989; Oluwole, 1985; Salami et al., 2011). According tovan Es et al. (1999), soil variability can be accessed as the relative mag-nitude of the sources of variability on a soil property as well as thecombined effect of variability of some of these properties. Coefficient ofvariation (CV) is the most commonly used measure of soil variability(Oluwole, 1985; Wilding, 1985). Using this measure, soil properties canbe expressed by dividing the coefficient of variation (CV) into differentranges, for example, least (b15%), moderate (15%-35%), and most(>35%) (Wilding, 1985). The use of PCA in particular has increasedover time due to its ability to reduce the dimensions of data with-out significant loss of key information about the initial parametersmeasured (Garten et al., 2007). Using this technique, it is possibleto group the often numerous soil physical and chemical parametersinto functional groups thus easing interpretation and targeting ofland management interventions.

Repeatability when using or interpreting the indicators for landdegradation can be a challenge if there is no standardized protocolto guide assessors. Selection of which indicator to use must takeinto cognizance the objective of assessment, time, expertise andresources available. Several protocols for land degradation assess-ment have been developed to educate assessors on using obser-vable indicators in order to interpret and assess ecosystemhealth. Examples of such protocols or frameworks include theLandscape Function Analysis (LFA) (Tongway, 2010; Tongwayand Hindley, 2004), Visual Soil–Field Assessment Tool (VS–Fast)designed to support and enhance the Land Degradation Assess-ment in Drylands (LADA) project of the Food and AgricultureOrganization (FAO) (McGarry, 2004) and the Land DegradationSampling Framework (LDSF) Vågen et al. (2010) designed tosupport the African Soil Information Service (AfSIS) Project. Theabove frameworks are based on standard scientific principles andattempts have been made to make these frameworks farmer-usable and based on visual assessment of soil condition and health,with particular emphasis on simple, repeatable methods using everydayand low cost apparatus (Kapalanga, 2008). Where applied someof these methods have proven to be simple yet robust, ensuringimmediate data availability, farmer acceptance and rapid updateof the descriptive and measurement tools, leading to rapid assess-ment of the current condition with a potential for longer-termmonitoring (McGarry, 2004).

The above opportunities for discerning patterns in land degra-dation indicators were the motivation to conduct this study. Thisstudy targeted the intensively fragmented smallholder systems in

western Kenya and sort to evaluate the spatial and temporalpatterns of indicators of land degradation as basis for targetingsustainable land management practices.

2. Method

2.1. Study area

This study was conducted in western Kenya covering areas ofKakamega, Butere-Mumias and Siaya Districts (Fig. 1). The studyarea lies at 34° 2′ 48″–34° 58′ 45″ E and 0° 4′ 26″ S–0° 36′ 15″ Nand covers an area of about 3800 km2. The region is dominated bysubsistence agricultural systems where farmers grow food cropssuch as maize, beans, millet, among others. The region is character-ized by a high population density (over 500 persons per sq. km)implying higher pressure on existing land.

2.1.1. GeomorphologyThe districts have varied land forms ranging from undulating hills

and broad valleys, moderate lowlands and swamps (Yala). The regionis drained by several rivers, and among them are the Nzoia River andthe Yala River that drain into Lake Victoria.

The study districts traverse across four agroecological zones (AEZ):humid (Zone I), sub-humid (Zone II), semi-humid (Zone III) andsemi-humid to semi arid (Zone IV). The area is classified as moistmid-altitude zone (MM) (Lynam and Hassan, 1998). The MM zoneforms a belt around Lake Victoria, from its shores at an altitude of1110 m, up to an altitude of about 1500 m above sea level. The dis-tricts experience bimodal rainfall and the distribution and amountsare greatly influenced by the relief and altitude as well as the

194 B.S. Waswa et al. / Geoderma 195–196 (2013) 192–200

presence of Lake Victoria. The rain falls in two peak seasons — longrains (March to May) and short rains (October to December).

2.1.2. SoilsThe study area is characterized by a wide range of soil types but

the dominant ones are the ferralsol (well drained soil found mostlyon level to undulating land), acrisol (clay-rich soils, associated withhumid tropical climates and supports forestry) and nitisol (deep,well-drained, red, tropical soils found mostly in the highlands). Thesoils are generally deficient in N, P and K (Lijzenga, 1998).

2.2. Sampling framework

The Land Degradation Sampling Framework (LDSF) was adoptedto characterize the benchmark sites (Vågen et al., 2010). LDSFis a spatially stratified, random sampling design frameworkbuilt around a hierarchical field survey and sampling protocolcomprising of “Blocks” and “Clusters”. In the wider study, three (3) sam-pling “Blocks” measuring 100 km2 (10 km×10 km) were demarcatedand sampled in the lower (Sidindi, Siaya), middle (Bunyala, Kakamega)and upper (Malava, Kakamega) catchment of the study area (Fig. 1) tocapture the differences in AEZ, and land uses in the region. Each Blockwas further subdivided into 16 tiles (2.5 km×2.5 km in size) in which a“Cluster” of 10 plots were randomly allocated in each tile for detailedsampling (Fig. 2). All coordinates for the randomly allocated plots wereentered into a Global Positioning System (GPS). This paper reports resultsmainly from one of the blocks (Malava) located in Kakamega North andonly draws general trends and observations on the land uses andsocio-economic data from the rest of the blocks that represent thewider western Kenya region.

2.2.1. Plot level measurementsNavigation to the randomly allocated plots was done using a GPS,

field maps, Google Earth maps and with guidance from the local fieldassistants. At the plot level, basic site characteristics were describedand recorded in standardized data entry sheets. All plots were givenan identification number (ID) and the center point was georeferencedusing a GPS. The geographic positions — easting, northing and eleva-tion were taken and recorded. Other data recorded at plot level

Fig. 2. The 10 km×10 km sampling block with the sampling clusters.

included slope and the presence or absence of soil and water conser-vation structures. The position of the plot along the topographicsequence was also described as either being upland, ridge/crest,midslope or footslope.

Land cover of the plot was recorded using a simplified version of theFAO Land Cover Classification System (LCCS) (http://www.africover.org). Using the “binary phase” of this classification, the following broadclasses were identified: i) cultivated or managed terrestrial areas,ii) natural or semi-natural vegetation, iii) cultivated aquatic orregularly flooded areas, iv) natural or semi-natural aquatic orregularly flooded vegetation, and, v) bare areas.

2.2.2. Sub-plot level measurementsEach plot consisted of 4 sub-plots which were sampled and

characterized for land degradation indicators as per the layoutbelow (Fig. 3).

At sub-plot level, several attributes and indicators of land degra-dation were identified, assessed visually and coded on either categor-ical or ordinal rating scales. Selection of the indicators was made inline with the three main attributes— soil and site stability, hydrologicfunction and biotic integrity (Pyke et al., 2002). Soil or site stabilityattributes refer to indicators showing the capacity of the site to limitredistribution and loss of soil resources by wind or water. Suchinclude indicators such as the proportion of bare ground, presenceof pedestals, presence of rills and gullies, soil surface loss such assheet erosion, among others. Attributes of the hydrologic functionrefer to indicators of the site to capture, store and safely releasewater from rainfall such as ground cover and many of the soilstability indicators listed. The attributes for integrity of bioticcommunity refer to indicators of the site to support characteristicfunctional and structural communities in the context of normalvariability and to resist loss of this function and structure causedby disturbance, and to recover following each disturbance. Among theindicators assessed included the presence or absence of invasive species,proportion of the functional vegetation groups (herbaceous vs woodcover), indicator weed species, crop and pasture health and performance,among others. The occurrence and/or severity of the indicators werequalitatively and quantitatively measured and scored using ordinal orcategorical scales.

Fig. 3. Plot and sub-plot sampling layout.

195B.S. Waswa et al. / Geoderma 195–196 (2013) 192–200

2.3. Soil sampling and analysis

The soil samples were collected using a 5 cm diameter and 40 cmlong Eijkelkamp (model 04.17) undisturbed split-tube-soil sampler.The soil core was then sectioned at two depths (0–20 cm and20–30 cm) and composite samples prepared for each depth. Thesamples were then submitted to the laboratory for processing andanalysis of soil chemical properties of interest. The LDSF frameworkdescribed above facilitated sampling of 160 (16 clusters×10 points)randomly allocated points per block.

2.3.1. Soil spectral reflectance analysisAll the soil samples were analyzed by diffuse reflectance spectros-

copy using a FieldSpec FR spectroradiometer (Analytical SpectralDevices Inc., Boulder, Colorado) at wavelengths from 0.35 to 2.5 μmwith a spectral sampling interval of 1 nm using the optical setupdescribed in Shepherd et al. (2003). The soil spectral data were an-alyzed by conducting a principal component analysis of the firstderivative spectra and computing the Euclidean distance basedon the scores of the significant principal components. Randomsamples were then selected from each quartile of the ranked Eu-clidean distances to constitute the 20% that would undergo con-ventional wet chemistry analysis (Shepherd and Walsh, 2002,2007).

2.3.2. Conventional wet chemistry analysisThe selected reference soil samples were then analyzed for select-

ed properties following conventional wet chemistry methods fortropical soils (Anderson and Ingram, 1993; ICRAF, 1995; Okalebo etal., 2002). These analyses were conducted at the Crop Nutrition(Cropnut) Laboratory in Nairobi. Soil pH was determined in 1:2.5(w/v) suspensions, exchangeable acidity by NaOH titration using a1:10 soil/solution ratio. Samples with pH>5.5 were assumed tohave zero exchangeable acidity and samples with pHb7.5, zeroexchangeable Na. Exchangeable Ca and Mg was determined by1 M KCl extraction, and exchangeable K and available P by 0.5 MNaHCO3 and 0.01 M EDTA (pH=8.5) using 1:10 soil/solutionratio extraction method. Soil texture was determined using aBouyoucos hydrometer after pre-treatment with H2O2 to removeorganic matter (Gee and Bauder, 1986). Total carbon and nitrogenwere analyzed at ICRAF laboratory by dry combustion using a C/Nanalyzer.

2.3.3. Spectral prediction of soil propertiesMeasured values of the soil samples selected for conventional

laboratory analysis were calibrated to the first derivative of thereflectance spectra using partial least squares regression (PLSR)implemented using the Unscrambler software (CAMO Inc., Corvallis, OR,USA). The regression models were then used to predict soil values forthe rest of the samples under investigation. Table 1 shows the correlationcoefficient and the root mean standard errors of calibration (RMSEC)

Table 1Calibration results for selected soil properties for Malava Block.

Property RMSEC R-squared

Total carbon (%) 0.26 0.85Total nitrogen (%) 0.27 0.81pH (H2O) 0.09 0.33Av. P (mg/kg) 0.67 0.15K (mg/kg) 0.52 0.28Ca (mg/kg) 0.41 0.67Mg (mg/kg) 0.29 0.76C.E.C. (C mol/100 g) 0.25 0.75Silt (%) 0.39 0.62Sand (%) 0.09 0.73Clay (%) 0.13 0.7

between the wet chemistry and the NIRS results for Malava Block. Thelarge correlation between the two results (R2>0.5) for C, N, Ca, Mg,CEC, silt, sand, suggests quite a strong relationship between the resultsof the wet chemistry and the NIRS analysis procedures. Soil pH showedmedium correlation (R2=0.3–4.9) while small correlation (R2b0.3)was observed for K and available P. These results especially for the largecorrelation are to a large extent similar to those reported by (Shepherdand Walsh, 2002; Terhoeven-Urselmans et al., 2010; Vågen et al., 2006).

2.4. Data analysis

Descriptive statistics (mean, standard deviation — SD andminimum–maximum values and coefficient of variation — CV, andskewness) were calculated for measured soil properties. Pearsoncorrelation and regression analyses were performed to understandthe relationships among the soil properties. PCA was computed toidentify the important factors that explain the variability in soilproperties observed across the Malava Block. Prior to performingPCA, the suitability of data for factor analysis was assessed bycomputing the correlation matrix. A correlation matrix was usedbecause it standardizes data with zero mean and unit variancethus making it possible to compare soil properties that have diffe-rent dimensions (are measured and presented in different units).Other test statistics computed during the PCA included theKaiser–Meyer–Oklin (Kaiser, 1970, 1974) and the Barlett's test ofsphericity (Bartlett, 1954), which support the factorability of thecorrelation matrix. The eigenvalues – the amount of varianceexplained by each factor – were then computed. The principalcomponents with eigenvalues greater than 1 were retained whilethose with eigenvalues less than 1 were not considered furtherbecause these explained less variance than that for a measuredattribute (Shukla et al., 2006). This was further proved by theinspection of the scree plot at each soil depth (Catell, 1965, 1966)and further supported by the results of the Monte Carlo PCA forparallel analysis (Watkins, 2000). The retained PCs were subjectedto varimax rotation to maximize the correlations between PC andthe measured attributes by distributing the variance of each factor.Statistical procedures were conducted using SPSS. All the abovestatistical procedures were conducted using SPSS and STATAsoftware.

3. Results

3.1. Site characterization

Assessment of general vegetation structure showed that over 70%of the land was under cropland (Table 2). Forests accounted for 8% ofthe farms sampled. Forest areas are contributed by the Malava andChesero forest fragments.

Vegetation cover assessment in Malava Block showed that thearea has sparse wood cover but high herbaceous cover. The highherbaceous cover rating is contributed by the expansive sugarcaneproduction fields that dominate the farming systems. Over 55% ofthe farms sampled lacked any form of soil and water conservation(SWC) technologies. Where present, either structural or vegetativeSWC techniques or a combination of the two was used by the

Table 2Vegetation structure in Malava Block.

Vegetation structure Frequency Percent (%)

Cropland 119 73.9Wooded grassland 25 15.5Woodland 4 2.5Forest 13 8.1Total 161 100.0

196 B.S. Waswa et al. / Geoderma 195–196 (2013) 192–200

farmers. The most common structures present were cut off drainsand drainage trenches, terraces planted with fodder species suchas napier grass and use of stone bounds and trash lines. Sheeterosion was the most dominant form of soil loss in Malava Blockhaving been observed in over 70% of the farms sampled. Sheeterosion has a significant impact on the general soil fertility sinceit involves removal and loss of the topsoil layer that holds most ofthe nutrients needed by the crops. Agriculture (crop cultivation)was identified as the main activity with highest impact on thehabitat. As indicated earlier, the wood cover rating in the studyarea was low and this translated to the low impact of tree cuttingand fuelwood collection at farm level. The low grazing impact canalso be explained by the increased land fragmentation and conver-sion to crop production thereby leaving little area for livestockrearing.

3.2. Variation in soil properties across Malava Block

There was distinct variation in the soil properties between thetopsoil (0–20 cm) and the subsoil (20–30 cm) (Tables 3 and 4). Thetopsoil had higher values for soil carbon, nitrogen and most cationscompared to the subsoil. Using CV as criteria for expressing variabili-ty, Ca, TON, Mg, TOC and silt were most variable soil properties for the0–20 cm depth. Moderate variability (CV 0.15–0.35) was observed forCEC, P, K and clay while Na, Sand and pH had the least variability(CVb0.15). For the subsoil (20–30 cm), Ca, Mg and silt were themost variable; TON, TOC, P, CEC, K, clay were moderately variablewhile Na and Sand were least variable (Table 3).

For the topsoil, the skewness was positive (ranged from 0.20 to2.47) for all soil properties except clay which had negative skewnessof −0.30 (Table 3). A similar pattern was observed for the sub-soilwhere only sand had a negative skewness (Table 4). Both negativekurtosis and positive kurtosis were observed for selected soil proper-ties at the different soil sampling depths. Despite these variations, thedata were not transformed during determination of variability.

3.2.1. Correlation of all the soil propertiesTables 5 and 6 show the Pearson product moment correlation

coefficients of soil properties for the top- and subsoil for the MalavaBlock. Significant correlations (pb0.05) were observed among six ofthe soil attribute pairs at the 0–20 cm depth. Weak but significantpositive correlation was observed between clay and CEC (r=0.182,n=160, pb0.05). Negative significant correlations were observed be-tween Ca and P (r=−0.197, n=160, pb0.05), Mg and P (r=−0.192,n=160, pb0.05), and sand and P (r=−0.162, n=160, pb0.05), clayand pH (r=−0.192, n=160, pb0.05) and sand and pH (r=−0.176,n=160, pb0.05). For the subsoil, significant positive correlation wasobserved between TOC and pH (r=0.196, n=160, pb0.05) and negativecorrelation between clay and Ca (r=−0.172, n=160, pb0.05).

Table 3Descriptive statistics for soil properties for Malava Block at 0–20 cm depth.

Soil property Minimum Maximum Mean

TOC 1.00 5.96 2.10±0.06TON 0.07 0.51 0.16±0.01pH 4.80 5.85 5.18±0.02CEC 5.14 26.59 10.46±0.28P 3.31 13.73 6.82±0.14K 67.75 232.45 116.95±2.15Ca 256.74 3550.42 859.80±37.30Mg 49.26 403.22 129.26±4.33Na 31.62 51.96 40.28±0.30Sand 47.25 73.52 61.25±0.43Clay 17.84 44.59 29.71±0.43Silt 2.48 25.94 7.66±0.25

TOC— total organic carbon (%); TON— total organic nitrogen (%); pH; CEC— cation exchange capa(mg kg−1); Na— sodium (mg kg−1); sand, clay and silt (%).

3.2.2. Soil textureThe soils in the study area generally well drained. Fort the topsoil,

textural analysis indicated that sand was the most dominant propor-tion in Malava Block with a mean of 61% followed by clay (31%) thensilt 8%. The soils in Malava can therefore be described as sandy clayloam.

3.2.3. Soil pHSoil analysis in the top layer (0–20 cm) revealed that Malava

Block was dominated by acidic soils with pH ranging between 4.8and 5.85 and this decreased to between 4.77 and 5.67 for thesubsoil (20–30 cm). About 94% of the farms sampled in Malavadistrict recorded very strongly acidic pH (pH 4.5–5.5) while 6% ofthe farms had moderately acidic (pH 5.6–6.0) soils. For MalavaBlock, forest areas had the highest pH averaging 5.5 but thisdeclined for grasslands and croplands. Extrapolating the pointmeasurements, showed a clear pattern of soil pH and hence soilacidity for Malava Block with areas North East of Malava townhaving more acidic soil characteristics.

3.2.4. Soil organic carbon (C) and nitrogen (N)High variation in soil organic carbon (SOC) was observed across

the study site (Fig. 4). The values ranged from 1 to 5.9% for the topsoil.A critical look at the carbon variation showed that 55% of the farmssampled in Malava Block had low (b2%) SOC. Soil organic carbon(C) also varied with land use. In Malava Block for example, forest orareas adjacent to forests (north — Malava Forest and south east —

Chesero Forest) recorded the highest SOC averaging about 4%compared to 1.9% SOC for most cultivated areas. The pattern ofSOC varied depending on other factors such as soil type, pH, CECamong others. A pattern similar to that observed for SOC was alsonoted for total organic nitrogen (TON) hence this data is notshown.

3.2.5. Cation exchange capacityCEC tended to vary with land use. These were in the order

forest>grassland>cropland>woodland. There was a strong positivecorrelation between CEC and SOC (Fig. 5). This can be due to theobservation that CEC is highly determined by the level of soil or-ganic matter in the soil.

3.3. Factor analysis of the soil properties in Malava Block

3.3.1. Results of the PCA for the 0–20 cm depthPrincipal components analysis (PCA) for the 0–20 cm soil depth

revealed the presence of three components with eigenvaluesexceeding 1, explaining 63.9%, 19.5% and 14.7% of the variance,respectively. Following the varimax rotation, the three-componentsolution explained 98% of the variance, with Component 1 contributing

Std. deviation Skewness Kurtosis CV

0.82 2.19 6.08 0.390.07 2.47 7.59 0.430.20 0.77 0.97 0.043.60 2.32 6.81 0.341.82 0.95 1.85 0.27

27.17 1.71 4.39 0.23471.79 2.47 8.94 0.5554.81 2.10 6.35 0.423.79 0.20 0.24 0.095.38 −0.30 −0.34 0.095.42 0.38 −0.70 0.183.12 1.64 6.48 0.41

city (mEq/100 g); P— phosphorus (mg kg−1); Ca— calcium ((mg kg−1);Mg—magnesium

Table 4Descriptive statistics for soil properties for Malava Block at 20–30 cm depth.

Soil property Minimum Maximum Mean Std. deviation Skewness Kurtosis CV

TOC 0.85 3.71 1.74±0.04 0.56 1.04 1.11 0.32TON 0.06 0.31 0.13±0.00 0.04 1.16 1.67 0.34pH 4.77 5.67 5.15±0.01 0.18 0.35 −0.29 0.04CEC 5.24 19.70 8.75±0.19 2.42 1.42 3.05 0.28P 3.06 12.00 5.98±0.14 1.74 1.05 1.33 0.29K 62.27 191.32 101.43±1.66 20.98 1.00 1.70 0.21Ca 201.35 2505.52 635.90±24.36 308.11 2.07 8.23 0.48Mg 43.44 305.25 101.72±2.98 37.72 1.60 4.93 0.37Na 32.55 61.50 43.65±0.37 4.74 0.16 0.41 0.11Sand 47.80 80.27 63.45±0.47 5.94 −0.08 −0.32 0.09Clay 15.69 46.95 30.31±0.48 6.07 0.50 −0.20 0.20Silt 1.66 18.62 5.89±0.19 2.44 1.35 3.95 0.41

TOC— total organic carbon (%); TON— total organic nitrogen (%); pH; CEC— cation exchange capacity (mEq/100 g); P— phosphorus (mg kg−1); Ca— calcium (mg kg−1); Mg—magnesium(mg kg−1); Na— sodium (mg kg−1); sand, clay and silt (%).

197B.S. Waswa et al. / Geoderma 195–196 (2013) 192–200

57.1%, Component 2 contributing 23.9% and Component 3 contributing17.1% (Table 7). The rotated solution revealed the presence of complexstructures, with the components showing a number of strong loadingsbut also some variables loading on more than one component.

For the topsoil, the first component (PC1) was represented by highloadings on SOC, TON, pH, CEC, K, Ca, Mg, and silt. These variableshave to do with the organic matter component and exchangeablebase status and are very relevant in explaining the potential fertilitylevels of the soil. The second component (PC2) gave high loadingson sand and clay. The SOC, TON and CEC, which are important compo-nents of SOM, partly belonged to PC2. Sand and clay play a major rolein influencing soil physical properties such as texture, bulk density,available pore space and the capacity to store and release plant nutri-ents. Soil physical properties influence the amount of SOM and henceCEC in any soils. The negative loading of sand in particular impliesthat soils with high sand content will generally tend to have lowSOM and CEC. The third component (PC3) gave high negative loadingfor P and high positive loading for Na. Soil pH partly belonged to PC3.Since the soils in this region do not have major problems of sodium,this component can be taken to represent the phosphorus availabilityin the soil. The co-loading of pH, P and Na on this PC shows theinfluence of soil acidity on P availability.

3.3.2. Results of the PCA for 20–30 cm depthAt the 20–30 cm depth, the initial principal components analysis

revealed the presence of three components with eigenvalues exceed-ing 1, explaining 61.3%, 24.6% and 12.2% of the variance, respectively.However upon rotation, the three components explained 98.2% of thetotal variance with PC1 explaining 50.9%, PC3 explaining 28.9% andPC3 explaining 18.4% of the total variance (Table 8). PC1 had highloadings on Ca, Mg, Silt, CEC, pH and K. This component also had

Table 5Pearson product moment correlations between soil properties in Malava Block for the 0–20

Soil property TOC TON pH CEC P K

TOC 1TON .996⁎⁎ 1pH .563⁎⁎ .623⁎⁎ 1CEC .918⁎⁎ .944⁎⁎ .802⁎⁎ 1P −0.074 −0.135 − .656⁎⁎ − .253⁎⁎ 1K .967⁎⁎ .969⁎⁎ .631⁎⁎ .954⁎⁎ −0.024 1Ca .817⁎⁎ .852⁎⁎ .810⁎⁎ .961⁎⁎ − .197⁎ .9Mg .846⁎⁎ .878⁎⁎ .817⁎⁎ .975⁎⁎ − .192⁎ .9Na − .366⁎⁎ − .334⁎⁎ 0.02 − .344⁎⁎ − .736⁎⁎ −Sand − .846⁎⁎ − .802⁎⁎ − .176⁎ − .612⁎⁎ − .162⁎ −Clay .532⁎⁎ .474⁎⁎ − .192⁎ .182⁎ 0.132 .3Silt .674⁎⁎ .701⁎⁎ .702⁎⁎ .838⁎⁎ 0.023 .8

Depth=0–20 cm.⁎ Correlation is significant at the 0.05 level (2-tailed).

⁎⁎ Correlation is significant at the 0.01 level (2-tailed).

co-loading for SOC and TON. These variables best explain the fertilitycomponent of the soil. The PC2 had high negative loading for sandand positive loading for clay, SOC and TON. The soil properties inthis component best explain the relationship between soil physicalproperties and SOM. Soils with low clay content will tend to havelower SOM contents. The third component (PC3), as in the topsoil,had high positive loading for P and negative loading for Na. Thiscomponent also had negative loading for pH. This componentrepresents P availability.

4. Discussion

The study identified some key indicators to monitor extent ofdegradation of the farms sampled. Rating the indicators on a categor-ical scale facilitated assigning of a degradation score on the plot. Asimilar approach has been used by Manske (2001) to illustrate thehealth condition of a rangeland. This process enables identificationof patches within the landscape that are either degrading, stable orimproving and it echoes the arguments under the Ecosystem DistressSyndrome (Rapport et al., 1985) theoretical framework.

The most variability for the topsoil was observed among Ca, TON,Mg, SOC and silt. This variability could be due to spatial differences insoil forming processes as well as land management practices. Differ-ent soils have different behavior in as far as their physical and chem-ical properties are concerned (Momtaz et al., 2009). Similarly thevariability could be due to the sampling design since the samplinglocations were randomly allocated and the unknown locations onlyaccessed using the GPS. This type of sampling enabled the capturingof diverse locations in the landscape such as managed and naturalecosystems, major land forms (upland, crest, and bottomlands) anddifferent gradients (level, sloping, and steep). Momtaz et al. (2009)

cm depth.

Ca Mg Na Sand Clay Silt

03⁎⁎ 125⁎⁎ .996⁎⁎ 1.507⁎⁎ − .421⁎⁎ − .433⁎⁎ 1.774⁎⁎ − .458⁎⁎ − .504⁎⁎ .378⁎⁎ 164⁎⁎ −0.01 0.033 0.001 − .839⁎⁎ 123⁎⁎ .940⁎⁎ .933⁎⁎ − .632⁎⁎ − .357⁎⁎ −0.153 1

Table 6Pearson product moment correlations between soil properties in Malava Block for the 20–30 cm depth.

Soil property TOC TON pH CEC P K Ca Mg Na Sand Clay Silt

TOC 1TON .994⁎⁎ 1pH .196⁎ .292⁎⁎ 1CEC .829⁎⁎ .879⁎⁎ .635⁎⁎ 1P .356⁎⁎ .287⁎⁎ − .605⁎⁎ 0.054 1K .959⁎⁎ .972⁎⁎ .329⁎⁎ .915⁎⁎ .364⁎⁎ 1Ca .675⁎⁎ .741⁎⁎ .693⁎⁎ .944⁎⁎ 0.055 .825⁎⁎ 1Mg .722⁎⁎ .783⁎⁎ .686⁎⁎ .962⁎⁎ 0.081 .859⁎⁎ .995⁎⁎ 1Na − .609⁎⁎ − .590⁎⁎ 0.06 − .548⁎⁎ − .799⁎⁎ − .717⁎⁎ − .570⁎⁎ − .595⁎⁎ 1Sand − .889⁎⁎ − .847⁎⁎ 0.125 − .543⁎⁎ − .409⁎⁎ − .770⁎⁎ − .321⁎⁎ − .379⁎⁎ .481⁎⁎ 1Clay .571⁎⁎ .500⁎⁎ − .437⁎⁎ 0.062 .297⁎⁎ .353⁎⁎ − .172⁎ −0.118 −0.097 − .843⁎⁎ 1Silt .608⁎⁎ .663⁎⁎ .591⁎⁎ .855⁎⁎ .238⁎⁎ .786⁎⁎ .959⁎⁎ .956⁎⁎ − .711⁎⁎ − .259⁎⁎ − .234⁎⁎ 1

Depth=20–30 cm.⁎ Correlation is significant at the 0.05 level (2-tailed).

⁎⁎ Correlation is significant at the 0.01 level (2-tailed).

198 B.S. Waswa et al. / Geoderma 195–196 (2013) 192–200

and Seyfried and Wilcox (1995) observed that the distribution andvariability of soil properties are scale dependent — the bigger thescale the higher the variability and vice versa. Despite this pattern,the variability provided a better view of the status of the soil acrossthe study area. Such information on the soil variability can also beused to guide researchers designing future sampling exerciseswhere the soil properties with the highest CV will require the mostnumber of samples to be taken to estimate their mean value withina certain percentage at any significance level.

The study using documented ‘critical’ values of some soil proper-ties was able to gauge the quality of the soil (Okalebo et al., 2002).Soil acidity was noted to be a major problem across the study sites.Studies indicate that acid soils cover over half a million ha of maizegrowing areas in Kenya (Kanyanjua et al., 2002). In western Kenyaalone, about 57,670 ha of the soils are acidic. Soil acidity can reduceyields through reduced phosphorus (P) availability and increasedaluminium (Al) and manganese (Mn) toxicity (Hue and Licudine,1999; O'Hallorans et al., 1997). Crop tolerance to soil acidity varies.For example, maize the staple crop for the farmers in the region liesin the medium tolerance range and would do well in soils ofpH 5.5–6.0 but as indicated most soils have the pH below thiscritical level. These results justify the investments by variousorganizations in the region for the use of agricultural lime bysmallholder farmers. Positive responses to use of lime have beenrecorded among farmers in Malava region.

The results indicated that majority of the farms had very low soilorganic carbon contents. Researchers generally agree that despitevariation in behavior of different types of soils, 2% soil organic carbon

0

1

2

3

4

5

6

7

0 20 40 60 80 100 120 140 160 180

Per

cen

t C

arb

on

(%

)

Farms

'Critical' SOC level

Fig. 4. Soil organic carbon (C) variation across the farms sampled in Malava Block.

(SOC) (ca. 3.4% SOM) is a critical threshold below which potentiallyserious decline in soil quality will occur (Greenland et al., 1975;Kemper and Koch, 1966; Loveland and Webb, 2003; Pretty, 1998).Considering the above threshold, about 55% of farms in Malava areconsidered having SOM below the critical level, hence are at a threatof degradation if not well managed. SOM is important as a “revolvingnutrient fund”; and as an agent to improve soil structure, maintaintilth and minimize erosion (FAO, 2005). SOM contributes to thecation exchange capacity of a soil. Cation exchange capacity deter-mines a soil's ability to retain positively charged plant nutrients,such as NH4+, K+, Ca2+, Mg2+, and Na+. Soils with low SOMhave low nutrient availability and poor water holding capacityand also exhibit poor responses to fertilizer application (Tittonellet al., 2005).

This first principal component of the soil chemical propertiesencompassed high loading for TOC, TON, pH, CEC, K, Ca, Mg, and silt.These variables have to do with organic matter content of the soiland exchangeable base status, and are thus very relevant inexplaining the ‘potential fertility’ levels of the soil. A study by(Salami et al., 2011) separated the measured soil variables intofive components of which the first component containing Mg, K,Ca, ECEC, ECEC and clay and was referred to as the ‘potentialfertility’ component while the third component with high loadingsof C and N and was referred to as the ‘organic matter’ component. Inanother study, Kosaki and Juo (1989) identified PC1 as ‘inherentfertility’ when it was dominated by Mg, sand, silt, Ca, K and clay.In the present study, however, the variables explaining ‘inherentsoil fertility’ and ‘organic matter’ were grouped together. This

y = 0.2086x - 0.0833 R² = 0.8428

0.0

1.0

2.0

3.0

4.0

5.0

6.0

7.0

0.0 5.0 10.0 15.0 20.0 25.0 30.0

So

il O

rgan

ic C

arb

on

(%

)

Cation Exchange Capacity (C mol/kg)

Forest points

Fig. 5. Correlation between cation exchange capacity and soil organic carbon forMalava Block.

Table 7Rotated factor loadings for the three principal components (PC) for Malava Block forthe 0–20 cm depth.

Soil property PC1 PC2 PC3 CE

Mg .987 .142 − .038 0.997Ca .987 .096 − .037 0.985CEC .945 .302 .054 0.988Silt .942 − .070 − .286 0.973pH .867 − .106 .440 0.957K .866 .469 − .153 0.993TON .804 .581 .005 0.983TOC .763 .634 − .044 0.986Clay − .110 .986 − .004 0.984Sand − .380 − .889 .188 0.971P − .252 .106 − .952 0.982Na − .406 − .040 .896 0.969% of variance explained 57.05 23.92 17.09

Extraction method: principal component analysis; rotation method: Varimax withKaiser Normalization; rotation converged in 4 iterations.

199B.S. Waswa et al. / Geoderma 195–196 (2013) 192–200

explains the intricate relationship between improving soil fertilityand SOM in these smallholder cropping systems. Soil organicmatter plays a key role in improving the exchangeable bases espe-cially in highly weathered soils such as those found in the studyarea. Efforts to improve plant nutrition should aim at achievingbalanced nutrition by providing all the key elements and at thesame time improving the capacity of the soil to store the same byincreasing the SOM content. This observation supports the currentinitiatives on integrated soil fertility management (ISFM) beingpromoted among the farmers in the region (Bationo and Waswa,2011; Bationo et al., 2007).

This study identified PC2 as characterized by high loadings forsand and clay and described this PC as the ‘soil physical property’component. The fifth component in the study by Salami et al.(2011) had high positive loadings for silt and sand, and was termedas the ‘sand–silt’ component. Sand and silt explain soil physical prop-erties such as texture, bulk density, available pore space and thecapacity to store and release plant nutrients (Troeh and Thompson,2005).

The third component (PC3) in this study gave a high negativeloading for P and high positive loading for Na. Since the soils in theregion do not have major problems regarding Na, this componentcan be taken to represent the ‘phosphorus availability’ pool of thesoil. Kosaki and Juo (1989) described their second component (PC2)as ‘available P’ since it was dominated by P and Ca. Similarly,Salami et al. (2011) defined PC2 as the ‘available phosphorus’component due to the high loadings for P, K, exchangeable acidity

Table 8Rotated factor loadings for the three principal components (PC) for Malava Block forthe 20–30 cm depth.

Soil Property PC1 PC2 PC3 CE

Ca .986 .091 .068 0.985Mg .984 .148 .079 0.997Silt .948 − .021 .278 0.997CEC .926 .350 .009 0.980pH .779 − .197 − .560 0.960K .767 .574 .272 0.991Clay − .272 .946 .089 0.978Sand − .228 − .936 − .228 0.979TOC .602 .764 .212 0.990TON .675 .715 .154 0.991P − .030 .214 .970 0.988Na − .520 − .171 − .815 0.965% of variance explained 50.89 28.85 18.42

CE — communality estimates; extraction method: principal component analysis; rotationmethod: Varimax with Kaiser normalization; rotation converged in 5 iterations.Numbers in bold indicate soil properties with high loadings on the respective PCs.

and Mn. Studies in western Kenya show that P is one of the mainlimiting nutrients for crop production (Bunemann, 2003; Sanchez,2002). Phosphorus deficiency in many of the soils is largely due to theinherent low P concentration in the parent material (Bunemann, 2003)as well as P-fixation (Van der Eijk, 1997). The PCA results confirm theimportance of P as a key parameter for understanding the soil fertilityproblem in the study area.

There is a wide variation in the results observed from the use ofPCA to understand soil variability. For example, a study by Oluwole(1985) identified three PCs' for soils in Nigeria. PC1 representedtexture, PC2 soil acidity and PC3 the SOM pool. The author concludedthat the main variation in the soils of the study area could be due tothe differences in clay content, and the associated differences in soilnutrient levels and pH values, as well as differences in the organicmatter status brought about by differences in land use. In anotherstudy, Adhikari et al. (2011) was able to group soil properties into 4PCs explaining soil sodicity (PC1), water transport through the soil(PC2), soil texture (PC3) and organic matter (PC4). The variationsobserved across studies could be attributed to the specific soilforming processes, management-related factors and soil conditions,which are site specific (Jiang and Thelen, 2004). Despite the differ-ences, PCA is able to help assign soils to functional groups therebyfacilitating targeting of management options.

5. Conclusion

The complexity of ecosystems means that land degradationcannot be viewed from a single lens but by examining indicatorsassociated with ecosystem attributes or factors of interest. Thestudy evaluated various indicators of soil and site stability, hydro-logic function and biotic integrity and noted that the presenceand magnitude of these indicators varied across the landscape.The study area had sparse wood cover but high herbaceous cover.Over 55% of the farms sampled completely lacked soil and waterconservation (SWC) technologies. Sheet erosion was observed inover 70% of the farms. Assessment of the general vegetation struc-ture showed that over 70% of the land was under cropland and 8%under forests. There was a high variability in the soil propertiesmeasured. Soil acidity and general low soil organic carbon wereevident across the study area. By grouping soil C and major macronutrients together, the PCA revealed the importance of soil organicmatter and nutrient availability in ensuring sustained food produc-tion in these cropping systems that are already facing problem ofdeclining soil fertility and reduced productivity. The PCA resultsfurther affirm the need for integrated soil fertility management(ISFM) as a strategy to ensure nutrient availability while at thesame time building the natural nutrient reserve through soil organicmatter build up.

Acknowledgements

This research was made possible through the PhD scholarshipfrom German Academic Exchange Program (DAAD) and researchsupport by the Center for Development Research (ZEF), Universityof Bonn, Germany.Wewould also like to thank Tropical Soil Biology andFertility (TSBF) and specifically the African Soil Information Services(AfSIS) Project for additional financial and logistical support towardsthe field work. The authors are grateful to the International PlantNutrition Institute (IPNI) for the award that was used partly topay for soil analysis and travel to present the results. Facilitationto the field sites was made possible though collaboration with theKenya Agriculture Research Institute (KARI), Kakamega RegionalResearch Centre Lime Project Team. Gratitude also goes to theresearch assistants and the farmers in the study sites.

200 B.S. Waswa et al. / Geoderma 195–196 (2013) 192–200

References

Adhikari, P., Shukla, M.K., Mexal, J.G., Sharma, P., 2011. Assessment of the soil physicaland chemical properties of desert soils irrigated with treated wastewater usingprincipal component analysis. Soil Science 176, 356–366.

Anderson, J.M., Ingram, J.S., 1993. Tropical Soil Biology and Fertility: A Handbook ofMethods. CAB International, Nairobi, Kenya.

Bartlett, M.S., 1954. A note on the multiplying factors for various chi squareapproximations. Journal of the Royal Statistical Society 16 (Series B), 296–298.

Bationo, A., Waswa, B.S., 2011. New challenges and opportunities for integrated soilfertility management in Africa. In: Bationo, A., Waswa, B., Okeyo, J.M., Maina, F.,Kihara, J.M. (Eds.), Innovations as Key to the Green Revolution in Africa: Exploringthe Scientific Facts. Springer, Dordrecht, The Netherlands, pp. 3–17.

Bationo, A., Waswa, B.S., Kihara, J., Kimetu, J., 2007. Advances in Integrated Soil FertilityManagement in Sub-Saharan Africa: Challenges and Opportunities. Springer,Dordrecht, Netherlands . (1094 pp.).

Beckett, P.H.T., Webster, R., 1971. Soil variability: a review. Soils and Fertilizers 34 (1),1–15.

Bunemann, E.K., 2003. Phosphorus Dynamics in a Ferralsol Under Maize-fallowRotations: The Role of the Soil Microbial Biomass. Swiss Federal Institute ofTechnology, Zurich . (162 pp.).

Catell, R.B., 1965. Factor analysis: an introduction to essentials. Biometrics 21, 190–215.Catell, R.B., 1966. The scree test for number of factors. Multivariate Behavioral Research

1, 245–276.Doran, J.W., Parkin, T.B., 1994. Defining and assessing soil quality. Defining Soil Quality

for Sustainable Environment: In: Doran, J.W., Coleman, D.C., Bezdicek, D.F.,Stewart, B.A. (Eds.), Soil Sci. Soc. Am. Special publication No. 35 (Madison,Wisconsin, US).

FAO, 2005. The Importance of Soil Organic Matter: Key to Drought-resistant Soil andSustained Food Production. Food and Agriculture Organization of the UnitedNations (FAO), Rome, Italy.

Garten, C.T., Kang, S., Brice, D.J., Schadt, C.W., Zhou, J., 2007. Variability in soil propertiesat different spatial scales (1 m–1 km) in a deciduous forest ecosystem. Soil Biologyand Biochemistry 39, 2621–2627.

Gee, G.W., Bauder, J.W., 1986. Particle-size analysis. In: Klute, A. (Ed.), Methods of SoilAnalysis, Part 1, Physical and Mineralogical Methods. Agronomy 9. 2nd ed. ASA,Madison, Wisconsin, USA, pp. 383–411.

Greenland, D.J., Rimmer, D., Payne, D., 1975. Determination of the structural stabilityclass of English and Welsh soils, using a water coherence test. Journal of SoilScience 26, 294–303.

Hue, N.V., Licudine, D.l, 1999. Amelioration of subsoil acidity through surfaceapplication of organic manures. Journal of Environmental Quality 28, 623–663.

ICRAF, 1995. International Centre for Research in Agroforestry: Laboratory Methods ofSoil and Plant Analysis. ICRAF, Nairobi, Kenya.

Jiang, P., Thelen, K.D., 2004. Effect of soil and topographic properties on crop yield in anorth-Central corn–soybean cropping system. Agronomy Journal 96, 252–258.

Kaiser, H., 1970. A second generation little jiffy. Psychometrika 35, 401–415.Kaiser, H., 1974. An index of factorial simplicity. Psychometrika 39, 31–36.Kanyanjua, S.M., Ireri, L., Wambua, S., Nandwa, S.M., 2002. Acidic Soils in Kenya:

Constraints and Remedial Options. Publications Unit KARI Headquarters, Nairobi,Kenya.

Kapalanga, T.S., 2008. A review of land degradation assessment methods, LandRestoration Training Programme. Final Project 2008, Keldnaholt, 112 Reykjavík, Iceland.

Karlen, D.L., Stott, D.E., 1994. A framework for evaluating physical and chemicalindicators of soil quality. Defining Soil Quality for Sustainable Environment. In:Doran, J.W., Coleman, D.C., Bezdicek, D.F., Stewart, B.A. (Eds.), SSSA PublicationNo. 35. SSSA and ASA, Madison, WI, pp. 54–72.

Kemper, W.D., Koch, E.J., 1966. Aggregate Stability of Soils from Western United Statesand Canada . (Washington, DC, USA).

Kosaki, T., Juo, A.S.R., 1989. Multivariate approach to grouping soils in small fields 1:extraction of factors causing soil variation by PCA. Soil Science and Plant Nutrition35 (3), 469–477.

Lal, R., 1994. Methods and Guidelines for Assessing Sustainable Use to Soil and WaterResources in the Tropics. Soil Management Support Services, USDA-NRCS,Washington, DC.

Li, Y., Lindstrom, M.J., 2001. Evaluating soil quality soil redistribution relationship onterraces and steep hillslope. Soil Science Society of America Journal 65, 1500–1508.

Lijzenga, M., 1998. Maize Response to NPK in Relation to Soil Fertility Indices inWestern Kenya: Plant Production Systems. Wageningen Agricultural University,Wageningen, The Netherlands.

Loveland, P., Webb, J., 2003. Is there a critical level of organic matter in the agriculturalsoils of temperate regions: a review. Soil and Tillage Research 70, 1–18.

Lynam, J., Hassan, R.M., 1998. A New Approach to Securing Sustained Growth inKenya's Maize Sector. CAB International, Wallingford, UK.

Manske, L.L., 2001. Grassland ecosystem monitoring: a non-quantitative procedure forassessing rangeland performance status. Range Management Report DREC. NorthDakota State University Dickinson Research Extension Center, Dickinson, NorthDakota.

McGarry, D., 2004. Final Consultancy Report for FAO on the Development of Methodologyof a Visual Soil–Field Assessment Tool— To Support and Enhance the LADA Program.FAO, China.

Momtaz, H.R., et al., 2009. An assessment of the variation in soil properties within andbetween landform in the Amol region, Iran. Geoderma 149 (1–2), 10–18.

NRC, 1994. Rangeland Health: New Methods to Classify, Inventory, and MonitorRangelands. National Academy Press, Washington, DC.

O'Hallorans, J., Muñoz, M.A., Marquez, P.E., 1997. Chicken manure as an amendment tocorrect soil acidity and fertility. Journal of Agriculture of the University of PuertoRico 81, 1–8.

Okalebo, R.J., Gathua, W.K., Woomer, P.L., 2002. Laboratory Methods of Soil and PlantAnalysis: A Working Manual. TSBF-CIAT and Sacred Africa, Nairobi, Kenya.

Oluwole, A., 1985. Variability of soils developed on migmatites in a part of the MiddleBelt of Nigeria. Applied Geography 5, 309–323.

Pellant, M., Shaver, P.L., Pyke, D.A., Herrick, J.E., 2005. Interpreting Indicators ofRangeland Health. Version 4. US Department of the Interior, Bureau of LandManagement, Denver, CO, USA.

Pendleton, R.L., 1919. Are soils mapped under a given type name by the Bureau of soilsmethod closely similar to one another? University of California Publications inAgricultural Sciences 3, 369-348.

Pretty, J., 1998. The Living Land. Earthscan, London.Pyke, D.A., Herrick, J.E., Shaver, P., Pellant, M., 2002. Rangeland health attributes and

indicators for qualitative assessment. Journal of Range Management 55 (55),584–597.

Rao, P.S.C., Wagenet, R.J., 1985. Spatial variability of pesticides in field soils: methods ofdata analysis and consequences. Weed Science 33 (2), 18–24.

Rapport, D.J., Regier, H.A., Hutchinson, T.C., 1985. Ecosystem behavior under stress. TheAmerican Naturalist 125, 617–640.

Sanchez, P.A., 2002. Soil fertility and hunger in Africa. Science 295 (5562), 2019–2020.Seyfried, M.S., Wilcox, B.P., 1995. Scale and nature of spatial variability: field examples

having implications for hydrologic modeling. Water Resources Research 31,173–184.

Shepherd, K.D., Walsh, M.G., 2002. Development of reflectance spectral libraries forcharacterization of soil properties. Soil Science Society of America Journal 66,988–998.

Shepherd, K.D., Palm, C.A., Gachengo, C.N., Vanlauwe, B., 2003. Rapid characterizationof organic resource quality for soil and livestock management in tropicalagroecosystems using near-infrared spectroscopy. Agronomy Journal 95,1314–1322.

Shepherd, K.D., Walsh, M.G., 2007. Infrared spectroscopy — enabling an evidence-baseddiagnostic surveillance approach to agricultural and environmental management indeveloping countries. Journal of Near Infrared Spectroscopy 15, 1–20.

Shukla, M.K., Lal, R., Ebinger, M., 2006. Determining soil quality indicators by factoranalysis. Soil and Tillage Research 87, 194–204.

Smith, H.F., 1938. An empirical law describing heterogeneity in the yield of agriculturalcrops. Journal of Agricultural Science 28, 1–23.

Terhoeven-Urselmans, T., Vågen, T.-G., Spaargaren, O., Shepherd, K.D., 2010. Predictionof soil fertility properties from a globally distributed soil mid-infrared spectrallibrary. Soil Science Society of America Journal 74 (5), 1792–1799.

Salami, B.T., Ade, J.B., Sanginga, N., 2011. Delineation of management zones byclassification of soil physico-chemical properties in the Northern Savanna ofNigeria. African Journal of Agricultural Research 6 (6), 1572–1579.

Tittonell, P., Vanlauwe, B., Leffelaar, P.A., Shepherd, K.D., Giller, K.E., 2005. Exploringdiversity in soil fertility management of smallholder farms in western Kenya II:within-farm variability in resource allocation, nutrient flows and soil fertilitystatus. Agriculture, Ecosystems and Environment 110, 166–184.

Tongway, D.J., 2010. Teaching the assessment of landscape function in the field:enabling the design and selection of appropriate restoration techniques. EcologicalRestoration 28 (2), 182–187.

Tongway, D.J., Hindley, N.L., 2004. Landscape Function Analysis: Procedures forMonitoring and Assessing Landscapes. CSIRO Sustainable Ecosystems, Canberra,Australia.

Troeh, F.R., Thompson, L.M., 2005. Soils and Soil Fertility. Blackwell Publishing,New Jersey.

Vågen, T.-G., Shepherd, K.D., Walsh, M.G., 2006. Sensing landscape level change in soilfertility following deforestation and conversion in the highlands of Madagascarusing Vis-NIR spectroscopy. Geoderma 133, 281–294.

Vågen, T.-G., Winowiecki, L., Tamene, T., Tondoh, J.E., 2010. The Land DegradationSurveillance Framework (LDSF): Field Guide. World Agroforestry Centre (ICRAF).

Van der Eijk, D., 1997. Phosphate Fixation and the Response of Maize to Fertilizer Phosphatein Kenyan Soils. Wageningen University, Wageningen, The Netherlands,(162 pp.).

van Es, H.M., Ogden, C.B., Hill, R.L., Schindelbeck, R.R., Tsegaye, T., 1999. Integratedassessment of space, time, and management-related variability of soil hydraulicproperties. Soil Science Society of America Journal 63, 1599–1608.

Vieira, S.R., Millete, J., Topp, G.C., Reynolds, W.D., 2002. Handbook for geostatisticalanalysis of variability in soil and climate data. In: Alvarez, V.V.H., Schaefer,C.E.G.R., Barros, N.F., Mello, J.W.V., Costa, L.M. (Eds.), Tópicos em ciência do solo.Sociedade Brasileira de Ciência do Solo, Viçosa, pp. 1–45.

Watkins, M.W., 2000. MonteCarlo PCA for Parallel Analysis [Computer Software]. Ed &Psych Associates, State College, PA.

Webster, R., 1994. The development of pedometrics. Geoderma 62, 1–15.Wilding, L.P., 1985. Spatial variability: its documentation, accommodation, and

implication to soil surveys. In: Nielsen, D.R., Bouma, J. (Eds.), Soil SpatialVariability (Pudoc, Wageningen, Netherlands).

Zacharias, S., 1998. Modeling Spatial Variability of Field-scale Solute Transport in theVadose Zone. Virginia Polytechnic Institute and State University, Blacksburg,Virginia.