assessment of soil physical degradation in eastern kenya by use of a sequential soil testing...

13
Assessment of soil physical degradation in Eastern Kenya by use of a sequential soil testing protocol C.T. Omuto * Department of Environmental and Biosystems Engineering, University of Nairobi, P.O. Box 30197-0100, Nairobi, Kenya 1. Introduction Soil physical degradation entails the destruction of number and arrangement of soil pores and peds. It is a gradual process through a number of steps: beginning at first with structural deterioration and ending in differential loss of finer particles through erosion. It can therefore take some time before manifesting visual degrada- tion symptoms in the field. During its progress, physical degradation destroys the soil matrix which carries air, moisture, and nutrients for supporting plant biomass production. It also affects soil surface characteristics with negative consequences in water infiltration, capacity of land to buffer atmospheric heat, and the general beauty of the landscape (Feddema, 1998). In spite of the recognized negative impacts and advancing rate, there is still a lack of adequate protocols to assess soil physical degradation (Lal, 2000; Paglia and Jones, 2002). Knowledge of these characteristics is vital for effective control and monitoring of the degradation. Capturing soil physical degradation characteristics requires a clear protocol for identifying degraded from non-degraded soil. Once developed, the protocol can then be used to separate different stages of degradation development and their occurrence in the landscape. There are varied attempts in literature to develop such a protocol. According to Lal (2000), it may entail series of tests on a suite of soil physical properties over time. The tests are then used to allocate degradation to cases where negative changes have occurred in soil physical properties. Some researchers have successfully used this approach. Chong and Green (1983) used sorptivity while Ball et al. (1997) used bulk density to define relative compaction as index of soil physical degradation. Recently, Dexter (2004) developed a slope index from water retention characteristics for determining soil physical quality. Apart from soil physical properties, other researchers have also developed a protocol with emphasis on observable degradation features in the field. They argue that degraded plots have observable symptoms separating them from non-degraded plots. Agriculture, Ecosystems and Environment 128 (2008) 199–211 ARTICLE INFO Article history: Received 17 March 2008 Received in revised form 27 May 2008 Accepted 2 June 2008 Available online 26 July 2008 Keywords: Physical degradation Physical properties Degradation symptoms RUSLE Spectral reflectance Sequential testing ABSTRACT Soil physical degradation is a gradual process of many steps beginning from structural deterioration and ending in differential loss of finer particles through erosion. Control of the degradation remains a challenge to many scientists due to lack of proper assessment protocols. This study developed a sequential protocol with emphasis on definition of physical degradation and successive soil testing to determine the stages of degradation development. The protocol was tested in Cambisols, Arenosols, and Ferralsols in Eastern Kenya. Soil physical degradation due to 10 years land use change was defined as more than 25% drop in infiltration and water retention characteristics and aggregate stability and more than 30% increase in bulk density and silt content. Then a soil testing model was sequentially applied to identify physical degradation phases. Visual assessment of degradation symptoms, RUSLE model, and diffuse infrared spectral reflectance were used in the soil testing model as predictors of physical degradation. Visual assessment was found to be cheap and fast method for identifying final stages of physical degradation with 60% accuracy. Visual assessment combined with RUSLE model improved the assessment accuracy to 80%. Infrared spectral reflectance, which is sensitive to subtle changes in soil physical conditions, was also found as a potential surrogate predictor of early-warning signs of soil physical degradation. Inclusion of spectra into the assessment model improved the accuracy to 95%. This protocol is effective in identifying phases of soil physical degradation, which are useful for planning degradation control and monitoring schemes. Its further testing and worldwide application is recommended. ß 2008 Elsevier B.V. All rights reserved. * Tel.: +254 736 107383. E-mail address: [email protected]. Contents lists available at ScienceDirect Agriculture, Ecosystems and Environment journal homepage: www.elsevier.com/locate/agee 0167-8809/$ – see front matter ß 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.agee.2008.06.006

Upload: ct-omuto

Post on 26-Aug-2016

217 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Assessment of soil physical degradation in Eastern Kenya by use of a sequential soil testing protocol

Agriculture, Ecosystems and Environment 128 (2008) 199–211

Assessment of soil physical degradation in Eastern Kenya by useof a sequential soil testing protocol

C.T. Omuto *

Department of Environmental and Biosystems Engineering, University of Nairobi, P.O. Box 30197-0100, Nairobi, Kenya

A R T I C L E I N F O

Article history:

Received 17 March 2008

Received in revised form 27 May 2008

Accepted 2 June 2008

Available online 26 July 2008

Keywords:

Physical degradation

Physical properties

Degradation symptoms

RUSLE

Spectral reflectance

Sequential testing

A B S T R A C T

Soil physical degradation is a gradual process of many steps beginning from structural deterioration and

ending in differential loss of finer particles through erosion. Control of the degradation remains a

challenge to many scientists due to lack of proper assessment protocols. This study developed a

sequential protocol with emphasis on definition of physical degradation and successive soil testing to

determine the stages of degradation development. The protocol was tested in Cambisols, Arenosols, and

Ferralsols in Eastern Kenya. Soil physical degradation due to 10 years land use change was defined as

more than 25% drop in infiltration and water retention characteristics and aggregate stability and more

than 30% increase in bulk density and silt content. Then a soil testing model was sequentially applied to

identify physical degradation phases. Visual assessment of degradation symptoms, RUSLE model, and

diffuse infrared spectral reflectance were used in the soil testing model as predictors of physical

degradation. Visual assessment was found to be cheap and fast method for identifying final stages of

physical degradation with 60% accuracy. Visual assessment combined with RUSLE model improved the

assessment accuracy to 80%. Infrared spectral reflectance, which is sensitive to subtle changes in soil

physical conditions, was also found as a potential surrogate predictor of early-warning signs of soil

physical degradation. Inclusion of spectra into the assessment model improved the accuracy to 95%. This

protocol is effective in identifying phases of soil physical degradation, which are useful for planning

degradation control and monitoring schemes. Its further testing and worldwide application is

recommended.

� 2008 Elsevier B.V. All rights reserved.

Contents lists available at ScienceDirect

Agriculture, Ecosystems and Environment

journal homepage: www.e lsev ier .com/ locate /agee

* Tel.: +254 736 107383.

E-mail address: [email protected].

0167-8809/$ – see front matter � 2008 Elsevier B.V. All rights reserved.

doi:10.1016/j.agee.2008.06.006

2000; Paglia and Jones, 2002). Knowledge of these characteristics isvital for effective control and monitoring of the degradation.

Capturing soil physical degradation characteristics requires aclear protocol for identifying degraded from non-degraded soil.Once developed, the protocol can then be used to separate differentstages of degradation development and their occurrence in thelandscape. There are varied attempts in literature to develop such aprotocol. According to Lal (2000), it may entail series of tests on asuite of soil physical properties over time. The tests are then usedto allocate degradation to cases where negative changes haveoccurred in soil physical properties. Some researchers havesuccessfully used this approach. Chong and Green (1983) usedsorptivity while Ball et al. (1997) used bulk density to definerelative compaction as index of soil physical degradation. Recently,Dexter (2004) developed a slope index from water retentioncharacteristics for determining soil physical quality.

Apart from soil physical properties, other researchers have alsodeveloped a protocol with emphasis on observable degradationfeatures in the field. They argue that degraded plots haveobservable symptoms separating them from non-degraded plots.

1. Introduction

Soil physical degradation entails the destruction of number andarrangement of soil pores and peds. It is a gradual process througha number of steps: beginning at first with structural deteriorationand ending in differential loss of finer particles through erosion. Itcan therefore take some time before manifesting visual degrada-tion symptoms in the field. During its progress, physicaldegradation destroys the soil matrix which carries air, moisture,and nutrients for supporting plant biomass production. It alsoaffects soil surface characteristics with negative consequences inwater infiltration, capacity of land to buffer atmospheric heat, andthe general beauty of the landscape (Feddema, 1998). In spite ofthe recognized negative impacts and advancing rate, there is still alack of adequate protocols to assess soil physical degradation (Lal,

Page 2: Assessment of soil physical degradation in Eastern Kenya by use of a sequential soil testing protocol

C.T. Omuto / Agriculture, Ecosystems and Environment 128 (2008) 199–211200

The use of expert opinion (Oldeman et al., 1991) and fieldassessment procedures by Stocking and Murnaghan (2001) areapplication examples of this protocol. The protocol has alsoreceived a technological boost from the recent advancements inspace-borne remote sensing (Vrieling, 2006). Degradation symp-toms can now be determined with increased spatial and temporalcoverage using remote sensing. Furthermore, improvements inremote sensing such as hyperspectral infrared spectroscopy arecurrently under tests for their suitability in detecting subtlechanges in soil physical conditions (Ben-Dor et al., 2003; Eshelet al., 2004).

In some other applications, researchers are using erosionmodeling to identify risk of degradation. Many models have beenapplied with varied focus on the risk factors for soil physicaldegradation (Paglia and Jones, 2002; Pieri et al., 2007). Somemodels such as the Revised Universal Soil Loss Equation (RUSLE)have been integrated with GIS to improve modeling and area-wideapplications (Renard et al., 1997).

Although these attempts seem successful, they tend toconcentrate only on one or two stages of soil physical degradation.Hence, they have not been quite able to make a huge impact incontrolling worldwide degradation progress rate. There is poten-tial for integrating all degradation aspects to develop an exhaustiveprotocol for assessing the stages of soil physical degradation. Thisstudy used long-term changes in a suite of soil physical propertiesto define physical degradation and develop a protocol for assessingits stages of development.

2. Methods and measurements

2.1. Study area

The study was carried out in the Upper Athi river watershed inEastern Kenya. The watershed stretches from the latitude 18090 to18590 South and from the longitude 368560 to 378450 East andcovers 4513 km2 (Fig. 1). It is gently sloping to almost flat aroundthe centre and south-eastern parts with altitudes of less than1500 m a.s.l. Steep slopes (>20%) occur in the south and north-western parts where the altitude is above 1500 m a.s.l. More than50 years ago, most parts of the watershed were under either intactsavannah shrublands or thickets (Tiffen et al., 1994). However, a

Fig. 1. Stud

large proportion of these original land use types have beenconverted to croplands or developed for human settlement. Thesoil is predominantly loam to sandy Cambisols, Ferralsols, andArenosols with low nutrient content (Sombroek et al., 1982).Cambisols are common in high altitudes while Arenosols dominatethe footslopes and lower parts of the watershed. The remainingparts of the watershed have Ferralsols.

The watershed receives rainfall in March through June (asperiod of long rains of about 1200 mm) and in September andOctober (as period of short rains of about 600 mm) (Tiffen et al.,1994). Mean annual temperature is 25 8C. During dry spell thetemperatures may go above 30 8C thus creating high evaporativedemand on the soil surface.

2.2. Data collection

2.2.1. Secondary data

Secondary data for the sequential soil testing protocol wereLandsat TM images for February 11, 2005, 30 m Digital ElevationModel (DEM), Land use maps for 1995 and 2005, and monthlyrainfall amounts (from 1995 till 2005) for 15 stations in and aroundthe study area.

Landsat images were downloaded from http://glcf.umiacs.um-d.edu/index.shtml on 20 June 2005. These images were correctedas indicated in Omuto and Shrestha (2007). Image correctioninvolved conversion of image digital numbers to ground reflec-tance (i.e. radiometric correction) and coordinate transformationof image projection to correspond to coordinates on the ground(i.e. geometric correction). The land use maps and DEM wereobtained from the Department of Environmental and BiosystemsEngineering of the University of Nairobi while rainfall data wereobtained from the Kenya Meteorology Department.

2.2.2. Ground sampling framework for primary data collection

Ground sampling involved field measurements and observa-tions of the indicators of soil physical degradation. The measure-ments were done on soil physical properties (infiltration and waterretention characteristics, soil texture, bulk density, aggregatestability, and infrared spectral reflectance) while observationswere made on signs of physical degradation and for the presence ofany soil conservation practices.

y area.

Page 3: Assessment of soil physical degradation in Eastern Kenya by use of a sequential soil testing protocol

C.T. Omuto / Agriculture, Ecosystems and Environment 128 (2008) 199–211 201

A hierarchical sampling framework was used to spatiallysample the study area and exhaustively characterize potential soilvariability at every sampling location. The framework consisted ofsampling points placed within square plots and the plots groupedinto clusters (Fig. 2). The plots in each cluster were arranged in a Y-frame; as a convenient method to represent all parts of a cluster.Altogether, there were 45 clusters randomly placed to cover themajor land use types, soil types, and climatic regimes of the studyarea. Each Y-cluster had four plots: one plot at the centre and theother three plots on the arms of Y at 480 m from the centre plot(Fig. 2). The plots were georeferenced at their centre with aGARMIN1 GPS (GARMIN International, 2002). Each plot had threesampling points: P1, P2, and P3, such that P2 was at the centre andthe other two points (P1 and P3) placed 10 m away from the centre(P2) as shown in Fig. 2.

2.2.3. Collection of primary data

Sampling and/or measurement of degradation indicators weremade at each point in the plots. Measurement involved infiltrationtests on the soil surface while soil samples were collected fortopsoil only (0–20 cm from soil surface). This depth was chosen forconvenience to test the sequential protocol.

Infiltration tests were done using three single-ring infiltrom-eters (with an internal diameter of 30 and 25 cm deep) on pre-wetted soil surface. There was an infiltrometer for every samplingpoint in a plot; so that infiltration tests for sampling points P1, P2,and P3 were simultaneously done for the plot. Pre-wetting wasdone 24 h in advance by sprinkling about 5 l of water on a gunnysack placed on the soil surface. Pre-wetting prevents soil surfacedisturbance during installation of infiltrometers and also standar-dizes antecedent moisture content for early infiltration character-istics. Infiltrometers were carefully inserted into the pre-wetted

Fig. 2. Ground sam

soil surfaces up to 5–10 cm deep and infiltration rates determinedas the rate of drop in water level inside the infiltrometers (Elrickand Reynolds, 2002; Diamond and Shanley, 2003). Fig. 3a shows aplot of measured infiltration characteristics from the study area.

Undisturbed soil samples for water retention and bulk densitywere collected using 100 cm3 stainless steel core-rings. Thesesamples were collected about 1 m away from the centre of theinfiltrometers. The core-rings were placed inside a ring-holder andthen inserted into the soil surface by hammering the ring-holderwith an impact absorbing hammer (Dirksen, 1999). The sampleswere then carefully removed and transported to a laboratory fordetermination of water retention characteristics and bulk density.Water retention characteristics were determined by removingmoisture from the soil samples at increasing values of suctionpressure heads. A sandbox apparatus was used for pressure heads��1.0 m and a pressure chamber for pressure heads <�1.0 m(Dirksen, 1999). After moisture removal, the samples were oven-dried for 48 h at 105 8C. Altogether, 13 levels of tension heads wereused (�150, �100, �50, �25, �5, �3, �2, �1, �0.79, �0.63, �0.32,�0.1, 0.001 m). Fig. 3b shows a plot of measured water retentioncharacteristics for the study area. Bulk density was determinedfrom the ratio of volume and weight of dry soil after oven drying.

Soil samples for aggregate stability, texture, and spectralreflectance were collected using Edelman soil auger. They werecarefully packed in polythene bags and transported to thelaboratory for further testing. In the laboratory, they were air-dried and sieved to pass through 2 mm sieve. Samples foraggregate stability were then analyzed using a wet-sieveapparatus while samples for texture were analyzed using thehydrometer method after pre-treating them with hydrogenperoxide to remove organic matter (Gee and Bauder, 1986;Eijkelkamp Agrisearch Equipment, 2006).

pling protocol.

Page 4: Assessment of soil physical degradation in Eastern Kenya by use of a sequential soil testing protocol

Fig. 3. Measured infiltration and water retention characteristics.

C.T. Omuto / Agriculture, Ecosystems and Environment 128 (2008) 199–211202

Soil samples for spectral reflectance were scanned with aFieldSpec FR spectroradiometer (Analytical Spectral Devices Inc.,1997). Scanning was performed in the soil spectral laboratory atWorld Agroforestry Centre (ICRAF). This spectrometer givesdiffuse infrared spectral reflectance with wavelengths between0.35 and 2.5 mm and spectral resolution between 0.003 and0.01 mm (Analytical Spectral Devices Inc., 1997). The scanningprocess, spectral averaging, and resampling methods used werethose reported in Shepherd et al. (2003). Fig. 4 shows examplesof infrared spectral reflectance for soil samples from the studyarea.

2.3. Definition of physically degraded soil

2.3.1. Estimation of soil physical properties

Soil physical properties for definition of physical degradationwere soil properties related to infiltration and water retentioncharacteristics, bulk density, aggregate stability, and texture. Thesoil properties pertaining to infiltration characteristics are steadyinfiltration rate and sorptivity. Steady infiltration rate is thecapacity of soil to transmit water in saturated conditions whilesorptivity is a measure of the uptake of water by soil in unsaturatedconditions. Both sorptivity and steady infiltration rate areinfluenced by the arrangement and number of soil pores; thus,they are a reflection of soil structural conditions (Kutilek and

Fig. 4. Examples of infrared spectral reflectance for soil samples from the study area.

Nielsen, 1994). Philip (1957) proposed a function combining thesetwo soil properties as shown in Eq. (1).

iðtÞ ¼ f c þ 0:5S=ffiffitp

(1)

where i(t) is the infiltration rate, fc is the steady infiltration rate, S isthe sorptivity, and t is the time.

Soil physical properties pertaining to water retention char-acteristics include saturated moisture content (us) and air-entrypotential (ha) (Brooks and Corey, 1964). van Genuchten (1980)proposed Eq. (2) for combining these soil physical properties withsoil moisture levels.

uðhÞ ¼ ur þ ðus � urÞ 1þ h

ha

� �n� ��ð1�n�1Þ

(2)

where u(h) is the soil moisture content at h suction potential and n

and ur are the shape parameters for the water retention curve (vanGenuchten, 1980).

Soil physical properties in Eqs. (1) and (2) were determinedfrom the measured infiltration and water retention character-istics using nonlinear mixed effects (NLME) approach (Omutoet al., 2006). This approach considers sampling design, globaland individual variability of infiltration and water retentioncharacteristics, and any possible parameter interdependence todetermine the physical properties (Omuto et al., 2006).Consequently, it gives fairly more accurate and reliableresults compared to other parameter estimation methods(Omuto et al., 2006). In this study, NLME approach wasimplemented using a HydroMe package (Omuto, 2007). Thepackage is freely downloadable at http://cran.r-project.org/web/packages/HydroMe/index.html and is executable in R (R Devel-opment Core Team, 2008, www.cran.r-project.org). A compar-ison of NLME fitted and measured data showed that Eqs. (1) and(2) were adequate in estimating the above soil physicalproperties (Fig. 5).

Although the shape parameters (n and ur) in Eq. (2) are oftenreferred to as non-physical parameters of water release char-acteristics (Hodnett and Tomasella, 2002), they have been widelyused to derive other important indices related to soil physicalconditions. Minasny and McBratney (2003) used them to obtainingintegral energy as a measure of soil-water availability while Dexter(2004) used them to develop an index of soil physical quality.Hence, they may still be useful in predicting soil physicaldegradation. In this study, the Dexter (2004) index of soil physicalquality was used to combine n, us, and ur for application alongside

Page 5: Assessment of soil physical degradation in Eastern Kenya by use of a sequential soil testing protocol

Fig. 5. NLME predicted infiltration and water retention characteristics. RSE is the standard error of estimate while r2 is the coefficient of determination.

C.T. Omuto / Agriculture, Ecosystems and Environment 128 (2008) 199–211 203

other soil physical properties in defining physical degradation.Dexter (2004) index is given in Eq. (3).

SI ¼ nðus � urÞ2n� 1

n� 1

� �ð1=nÞ�2

(3)

Bulk density represents relative compaction of soil and is aknown index of compaction or crusting (Ball et al., 1997; Lal, 2000;Paglia and Jones, 2002). It was computed from the ratio of mass ofdry soil to volume of the dry soil as shown in Eq. (4).

bulk density ðrbÞ ¼Mass of dry soil

Volume of dry soil(4)

Soil aggregate stability is the resistance of soil structure againstmechanical, physical or chemical destructive forces (Marshallet al., 1996). Wet sieving apparatus is one of the commonly usedmethods for determining aggregate stability (Diaz-Zorita et al.,2002; Eijkelkamp Agrisearch Equipment, 2006). It uses theprinciple of aggregate breakdown on impact with low energystress such as wetting by water to separate unstable from stableaggregates (Diaz-Zorita et al., 2002). Unstable aggregates easilybreakdown leaving stable aggregates intact if they are both slowlyimmersed in water. The ratio of the amount of stable aggregates tototal aggregates for a given soil sample is the index of aggregatestability (Eq. (5)).

Stable aggregates ðAsÞ ¼Amount of stable aggregates

Total amount of aggregates(5)

Eq. (5) was used in this study to determine the index of aggregatestability. The amount of stable aggregates and total aggregateswere obtained by slow immersion of soil samples in water usingthe wet sieve apparatus (Eijkelkamp Agrisearch Equipment,2006).

Altogether, the soil physical properties used in defining physicaldegradation were: steady infiltration rate (fc), sorptivity (S),porosity (us), air-entry potential (ha), Dexter’s index of soil physicalquality (SI), index of aggregate stability (As), sand, silt, and clayfractions, and bulk density (rb). There were three replicates ofthese soil properties for every plot in the Y-clusters.

2.3.2. Definition of degraded soils

Soil physical degradation was defined using negative changes inthe above soil physical properties. The negative changes werethose associated with historic land use changes. Five main land usetypes considered were: thickets, open savannah shrublands,

croplands, built-up areas, and grasslands (Tiffen et al., 1994). Soilphysical properties for plots where land use had remained intactbetween 1995 and 2005 were compared to where land use hadchanged by 2005. The comparison was paired for plots with similarsoil types (Wessels et al., 2004). Plots where soil physicalproperties had changed for the worst during this period (1995–2005) were considered degraded. Otherwise, they were considerednon-degraded if the soil physical properties had remained within95% confidence limit of the average in 1995 or had positivelyimproved.

Percent average differences for soil physical properties indegraded and non-degraded plots were explored with a hierarch-ical tree model to understand variable importance in the definitionof soil physical degradation. Hierarchical tree models are suitablein understanding variables importance (and their interactions) inpartitioning different classes in a dataset (Breiman et al., 1984).The most important variables appear at the top of the model whileleast important variables appear at the bottom (Breiman et al.,1984).

2.4. Data preparation for sequential testing of soil

After defining soil physical degradation, some commonly usedmethods for assessing physical degradation were sequentiallyapplied to identify soil physical degradation phases. The methodstested were visual assessment of degradation symptoms in thefield, risk of soil loss using RUSLE model, and infrared spectralreflectance (Millward and Mersey, 1999; Omuto and Shrestha,2007).

2.4.1. Observable symptoms of degradation

Symptoms of soil physical degradation for visual assessmentwere obtained within the plots during the field survey. Theyincluded evidence of sheet, rill or gully erosion and signs of in situstructural deterioration such as crusting, hardsetting, and/orcompaction. They were obtained using the guidelines in Table 1(Omuto and Shrestha, 2007).

2.4.2. Risk of soil loss using RUSLE

Risk of soil loss was estimated using the Revised Universal SoilLoss Equation (RUSLE) proposed by Renard et al. (1997). Thismodel combines soil loss risk factors such as climate, soil type, landcover, topography, and land use practices to predict the risk of soilloss (Morgan, 1995). The model, which is shown in Eq. (6), has beenwidely used due to its low data demand (Lufafa et al., 2003; Lewis

Page 6: Assessment of soil physical degradation in Eastern Kenya by use of a sequential soil testing protocol

Table 1Criteria for visual assessment of soil physical degradation in the field

Form of degradation Type Degradation symptoms

In situ physical degradation Crusting and sealing Hard layers on soil surface

Algae-strengthen pedestals, biological crusts in sheet eroded fields

Hard and difficult-to-auger surfaces

Compaction Compaction Signs of water logging

Uneven plant/grass growth

Mottling of subsoil

Hard and difficult-to-auger subsoil

Soil loss Sheet Small heaps of washed sand

Fine soil particle in small depressions

Soil deposits in high sides of small obstructions such as wood splinters or fences

Rill Small depressions (<30 cm)

Exposure of plant roots

Gully Deep depression (>30 cm)

Exposure of lower soil depths

C.T. Omuto / Agriculture, Ecosystems and Environment 128 (2008) 199–211204

et al., 2005).

E ¼ R� K � LSt� C � P (6)

where E is the risk of soil loss in tonnes ha�1 yr�1, R is the climatefactor known as rainfall erosivity (in MJ mm ha�1 h�1 yr�1), K is thesoil factor known as erodibility (in tonnes ha h ha�1 MJ�1 mm�1), C

is the land cover factor, and P is the support practice factor(Morgan, 1995). Input data for this model include soil physicalproperties, rainfall characteristics, DEM, Landsat image, and landmanagement factors (van der Kniff et al., 1999; Lee and Lee, 2006).

Within the RUSLE model, rainfall erosivity (R) is estimated asthe average of EI30 (the energy of 30 min rainfall intensity)measurements over a period of 1 year (Morgan, 1995). Severalmodifications have been undertaken which include the changefrom RUSLE annual to seasonal format (Cook et al., 1985) and use ofrainfall amounts in place of EI30 (Renard and Freidmund, 1994; Yinet al., 2007; Shamshad et al., 2008). A document by Moore (1979)provides a means of predicting R based on rainfall amounts inEastern and Southern Africa. The models in this document havebeen successfully used to predict R in the greater Horn of Africa(Lufafa et al., 2003; Hammad et al., 2004). This study applied Eq. (7)from Moore (1979) to estimate erosivity.

R ¼ 0:029 3:96ðX12

i¼1

PiÞ þ 3122

!� 26 (7)

where Pi is the mean monthly rainfall amounts (in mm) for month i

and R is the erosivity in MJ mm ha�1 h�1 yr�1. The input Pi in Eq. (7)was obtained by spatial interpolation of mean monthly rainfallamounts from 15 weather stations in/and around the study area.The spatial interpolation of rainfall amounts was done usingkriging method (Hengl et al., 2007).

Soil factor (K) in the RUSLE model quantifies the cohesivecharacter of a given soil type and its resistance to physicaldegradation by raindrop impact and runoff shear forces. It is afunction of soil structure, texture, organic matter content, andpermeability. It can also be obtained from textural compositions inthe absence of all these data (Wischmeier and Smith, 1978). Eq. (8)was proposed by Morgan (1995) for deriving K from soil texturalcharacteristics.

Erodibility ¼ 0:0035þ 0:00388 exp �0:5ðlog Dg þ 1:519Þ2

0:57517

" #(8)

where Dg is the mean soil particle diameter and which is estimatedby Eq. (9) (Torri et al., 1997).

Dg ¼ expX

i

0:01 f i lnffiffiffiffiffiffiffiffiffiffiffiffidi1di2

p" #(9)

where fi is the particle fraction in percent, d1 is the maximumdiameter (mm) of the soil fraction, and d2 is the minimumdiameter (mm) of the soil fraction. Eqs. (8) and (9) were used toestimate K for each plot. The inputs for this estimation were fi frommeasured soil textural fractions, d1 taken as 2 mm for sand,0.05 mm for silt, and 0.002 mm for clay, and d2 taken as 0.05 mmfor sand, 0.002 mm for silt, and 0.0005 mm for clay (Torri et al.,1997).

Slope-length factor (LSt) in the RUSLE model defines whetherthe soil loss model is in two dimensions or three dimensions(Desmet and Govers, 1996). It is a dimensionless quantity from theproduct of length of slope L and slope angle St. The length of slope,L, is the horizontal distance from the origin of overland flow toeither where the slope decreases to a point at the onset ofdeposition or where runoff becomes focused into a defined channel(Renard et al., 1997). It a sensitive factor in the RUSLE model and itscalculation can lead to flow convergence and deposition especiallyin undulating terrain (Lee and Lee, 2006). In this study, it wasderived from the length of upslope drainage contributing areawithin a three-dimensional space (Desmet and Govers, 1996). Thecalculation of L from upslope drainage area has the potential ofaccounting for the flow divergence and convergence patterns (vanRemortel et al., 2004). Foster and Wischmeier (1974) haveproposed Eq. (10) for deriving L from upslope drainage contribut-ing area using DEM grids as input.

L ¼ðAin þ pixel2Þ

mþ1� Amþ1

in

pixelmþ2ð sin aj j þ cos aj jÞmð22:13Þm(10)

where Ain is the drainage contributing area at the inlet of a grid forwhich L is being estimated, pixel is the DEM grid resolution (whichis 30 m for the DEM used in this study), a is the flow directionwithin the grid, and m is the exponent that addresses the ratio ofrill-to-interrill soil loss. The value of m is taken as 0.4 for slopeangle St > 38, 0.3 for 28 < St � 38, 0.2 for 18 < St � 28, and 0.1 forSt � 18 (Morgan, 1995). Since long lengths of slope can concentrateoverland flows into channels, care must be taken in calculating L

for use in the RUSLE model (Lewis et al., 2005). Renard et al. (1997)recommended a maximum of 200 m for L for modest accuracy. Inthis study, Eq. (10) was solved using computer modules in SAGA

Page 7: Assessment of soil physical degradation in Eastern Kenya by use of a sequential soil testing protocol

C.T. Omuto / Agriculture, Ecosystems and Environment 128 (2008) 199–211 205

GIS (http://saga-gis.org) by imposing an upper sealing of 120 m(four pixels of the DEM map) for the slope length. Slope St in theslope-length factor was derived from the DEM using maximumdownhill slope algorithm by Hickey et al. (1994). This algorithmwas implemented in ArcGIS software (ESRI, 2004). The DEM is theinput for calculating both L and St. The slope St and L werecombined to produce LSt factor using Eq. (11) (Wischmeier andSmith, 1978).

LSt ¼ L

22:13

� �m

0:065þ 4:56 sin Stþ 65:41 sin2 St� �

(11)

where St is the slope in degrees and m is obtained as in Eq. (10).Eq. (10) was determined for the whole study area as LSt map. TheLSt-factor for use in Eq. (6) for each surveyed plot was extractedfrom this map.

Land cover factor (C) in the RUSLE model was determined usingvan der Kniff et al. (1999) model in Eq. (12).

C ¼ exp �aNDVI

b� NDVI

� �(12)

where a and b are the constants. van der Kniff et al. (1999)suggested values of b as 1 and a as 2. The input NDVI for Eq. (12)was obtained from the ratio of the difference and sum of the thirdand fourth bands of the corrected Landsat TM image for 2005(Omuto and Shrestha, 2007). Eq. (12) was also determined for thewhole study area as a map of C factor. C factor for use in Eq. (6) foreach plot was extracted from this map.

Land management practice factor (P) in the RUSLE modelrepresents the effect of soil and water conservation practices forcontrolling soil loss. Wischmeier and Smith (1978) defined P as anindex for comparing the amount of soil lost with a specificconservation practice to the corresponding soil lost with down-slope cultivation. They developed a monograph for allocating P todifferent land use practices. In this study, soil conservationpractices observed during field survey were assigned P indicesusing this monograph. The main soil conservation practicesobserved in the study area were contour ploughing and cut-offdrains. Deep ripping and strips of close-growing vegetation werealso observed in a few places (Tiffen et al., 1994). P-indices wereassigned to plots where the conservation practices were found (orfor plots adjacent to an immediate upslope area with suchconservation practices). A P value of 1 was assumed for plotswithout conservation practises.

Although RUSLE is a widely used model, it is important to notethat it has uncertainties in predicting the risk of soil loss. There aretwo main uncertainties in RUSLE estimates: uncertainties in theinput data and those due to model inadequacies. The uncertaintiesin input data can be estimated as linear propagation of standarderrors of the input data while uncertainties due to modelinadequacies should be assessed by comparing its output withactual measurements. In this study the model uncertainties wereestimated using Eq. (13).

s2 ¼X

s2i

@E

@wiu

� �(13)

where s2 is the standard error of the predicted risk of soil loss, wi isthe input data, and u is the variance–covariance matrix for theinput data. The model was, however, not compared with actualfield measurements due to lack of reliable data.

2.4.3. Infrared spectral reflectance

Soil spectral reflectance consisted of reflectance for wave-lengths between 0.35 and 2.5 mm in 216 contiguous groups of0.01 mm wavebands (Fig. 4). The replicate measurements of

spectra for each plot were calibrated to corresponding measuredsoil physical properties. The calibration tested the relevance ofspectra in relation to soil physical conditions. Partial LeastSquares (PLS) regression method was used for the calibrationprocess (Bro, 1996). After the calibration, an optimum number ofprincipal components (PCs) accounting for over 90% of spectralvariations were selected for use in the sequential soil testingprotocol.

2.5. Sequential testing of soil for physical degradation

The above methods for assessing soil physical degradationwere sequentially applied as predictors of physical degradation totest their ability in identifying different stages of degradationdevelopment. The testing involved recursive partitioning ofsampled plots into their correct degradation statuses usingindices for assessment for the above methods as splittingvariables. Recursive partitioning is the process of successivesplitting of data into homogeneous groups. It begins with aheterogeneous data (known as parent) which is split into two lessheterogeneous groups (known as children). The splitting con-tinues until the children are too homogeneous to be split. The finalgroup of the homogeneous data are then assigned group namecorresponding to their common characteristics (e.g. degraded ornon-degraded soil) (Breiman et al., 1984). The splitting rules andthe class assignment to terminal children have been discussed inBreiman et al. (1984). The ratio of correct classification to the totalnumber of partitioned classes in a terminal group determines theaccuracy of the splitting variables in identifying the homogeneousgroups.

In this study, degradation indices for recursive partitioningwere field-observable symptoms of degradation, risk of soil loss,and infrared spectral reflectance. A hierarchical tree model forthe sequential testing with these indices is shown in Fig. 6. Themodel was implemented using CART1 V 6.0 (Salford SystemsInc., 2008). It was developed on a random selection of two-thirdsof the data and validated on the remaining one-third of the data.The accuracy on validation dataset was used to assess theperformance of each assessment method in identify stages ofdegradation development.

3. Results and discussions

3.1. The variation of soil physical properties with land use

types and soil types

Soil properties were grouped into the main soil types and landuse types. Good soil physical qualities were found in land use typeswith vegetation irrespective of the soil types (Fig. 7). However,differences in soil physical conditions between soil types emergedwhen plots were compared on the basis of the presence ofvegetation cover. Cambisols had the largest differences betweenplots with vegetation (thickets or grasslands) and withoutvegetation (tilled croplands or built-up areas). The largestdifference was among soil properties related to structure suchas porosity and aggregate stability. The average porosity oraggregate stability in croplands and built-up areas was around 50%less than in thickets or grasslands (Fig. 7). These results suggestthat Cambisols are more sensitive to changes in land use/covercharacteristics compared to Arenosols or Ferralsols. Ferralsols,which are deeply weathered soils, remained rather stable betweendifferent land use types (WRB, 2006). However, their porosity canreduce by 40% or more if soil is continuously modified (e.g. manyyears of tillage in croplands) (Fig. 7). Arenosols appeared to beunstable with poor soil physical qualities (Fig. 7). All plots in

Page 8: Assessment of soil physical degradation in Eastern Kenya by use of a sequential soil testing protocol

Fig. 7. Variation of soil physical properties

Fig. 6. Sequential soil testing model.

C.T. Omuto / Agriculture, Ecosystems and Environment 128 (2008) 199–211206

Arenosols had index of aggregate stability less than 0.50 andDexter’s index of physical quality less than 0.03 (except for theplots in shrublands). 50% index of aggregate stability and 0.03 forDexter’s index are the suggested lower limits for stable soils ofgood physical qualities (Marshall et al., 1996; Dexter, 2004). Giventhat most plots in Arenosols were below these limits, they may bereferred to as unstable soils. WRB (2006) also classifies Arenosolsas unstable soils.

High variability and poor physical characteristics wereobserved in croplands, built-up areas, and grasslands than inthickets or shrublands (Table 2). These soil characteristics givethe impression that land use changes from thickets/shrublandsto croplands/grasslands/built-up areas have the potential ofdegrading the soil physical conditions. The poorest physicalcharacteristics and highest variability in soil texture were incroplands and built-up areas (Table 2). Although texture is astable soil property, high variation in texture can occur due toadvanced physical degradation (Lal, 2000; Paglia and Jones,2002). High variation in texture and poor soil physicalcharacteristics in croplands and built-up areas is therefore anindication that these land use types are risk factors for soilphysical degradation.

3.2. Definition of soil physical degradation in the study area

GPS locations for the surveyed plots were superimposed onthe land use maps for 1995 and 2005. The major land usechanges during this period were mainly from thickets orshrublands to grasslands, built-up, or croplands. Out of 180surveyed plots, 113 had land use changes between 1995 and2005. 54 plots in the thickets in 1995 were converted to built-uparea (10 plots), grasslands (20 plots), and croplands (24 plots) by2005. Similarly, 32 plots in the shrublands in 1995 were

between soil types and land use types.

Page 9: Assessment of soil physical degradation in Eastern Kenya by use of a sequential soil testing protocol

Table 3Mean of soil physical properties and corresponding land use changes between 1995 and 2005

Soil property Intact land use types between 1995 and 2005 Land use changes by 2005

Symbol Thickets Croplands Shrublands Grasslands Croplands Built-up

Steady infiltration rate fc (cm h�1) 30.9 (1.05a) 13.8 (2.08) 26.5 (1.45) 26.3 13.4 10.9

Sorptivity S (cm h�0.5) 25 (2.28) 18.1 (3.76) 11.2 (2.21) 16.1 3.51 10.1

Saturated moisture content us (cm3 cm�3) 0.45 (0.01) 0.36 (0.11) 0.33 (0.02) 0.34 0.24 0.14

Air-entry potential ha (m) 0.23 (0.02) 0.37 (0.15) 0.29 (0.03) 0.33 0.50 0.36

Dexter’s index SI 0.06 (0.002) 0.02 (0.001) 0.04 (0.002) 0.03 0.02 0.02

Aggregate stability As 0.61 (0.008) 0.32 (0.01) 0.45 (0.006) 0.81 0.34 0.23

Bulk density rb (g cm�3) 1.19 (0.017) 1.51 (0.07) 1.35 (0.023) 1.33 1.48 1.53

Sand Sand (%) 40.2 (1.2) 36 (0.15) 39 (1.6) 43 30.4 39

Silt Silt (%) 28.2 (0.8) 37 (0.1) 30.4 (0.7) 31.2 36.6 31.6

Clay Clay (%) 31.6 (1.2) 27 (0.22) 29.6 (3.9) 25.8 33 28.4

Number of plots – 15 25 27 31 38 44

a Values in the brackets are the 95% confidence intervals of the means.

Table 2Summary of the soil physical properties according to land use types

Soil property Land use types

Symbol Thickets Shrublands Croplands Grasslands Built-up

Steady infiltration rate fc (cm h�1) 30.9 (1.05a) 26.5 (1.45) 13.6 (1.93) 26.3 (2.37) 10.9 (2.6)

Sorptivity S (cm h�0.5) 25 (2.28) 11.2 (2.21) 11 (3.76) 16.1 (3.44) 10.1 (1.56)

Saturated moisture content us (cm3 cm�3) 0.45 (0.01) 0.33 (0.02) 0.29 (0.04) 0.34 (0.02) 0.14 (0.03)

Air-entry potential ha (m) 0.23 (0.02) 0.29 (0.03) 0.41 (0.12) 0.33 (0.04) 0.36 (0.15)

Dexter’s index SI 0.06 (0.002) 0.04 (0.002) 0.02 (0.009) 0.03 (0.005) 0.02 (0.01)

Aggregate stability As 0.61 (0.008) 0.45 (0.006) 0.33 (0.069) 0.47 (0.045) 0.23 (0.018)

Bulk density rb (g cm�3) 1.19 (0.017) 1.35 (0.023) 1.49 (0.072) 1.42 (0.029) 1.53 (0.09)

Sand Sand (%) 40.2 (1.2) 39 (1.6) 33.3 (4.6) 43 (3.1) 39 (0.8)

Silt Silt (%) 28.2 (0.8) 30.4 (0.7) 35.7 (4.3) 31.2 (1.7) 31.6 (4.1)

Clay Clay (%) 31.6 (1.2) 29.6 (3.9) 31 (2.02) 25.8 (3.1) 28.4 (3.7)

a Values in the brackets are 95% confidence intervals of the means.

C.T. Omuto / Agriculture, Ecosystems and Environment 128 (2008) 199–211 207

converted to croplands (10 plots), grasslands (16 plots), andbuilt-up (6 plots) by 2005 and 16 plots in the croplandsconverted to grassland (4 plots) and built-up areas (12 plots) inthe same period (Table 3).

Almost all plots with land use changes by 2005 had morethan one standard deviation drop (or increase in bulk density)from their average soil properties in 1995. The plots where thechanges were outside 95% confidence limit of their averages in1995 were considered degraded (Table 3). Percent averagedifferences between soil properties in the degraded and non-degraded plots were explored with a tree model to enableunderstanding of variable importance in the definition ofphysical degradation (Breiman et al., 1984). A drop of morethan 25% in Dexter’s index of soil physical quality (SI) was themost important in separating degraded from non-degraded soil(Fig. 8). Thus, it is a possible important indicator for assessingphysical degradation.

The next important variables included changes in aggregatestability (As) and air-entry potential (ha) (Fig. 8). As isdirectly related to soil structure and inversely related to ha

(Reynolds and Elrick, 1990; Marshall et al., 1996). An increase inha and a decrease in As correspond to the deterioration of soilstructure. The relative importance of changes in As and ha inFig. 8 is therefore an illustration of the high priority ofdeterioration of soil structure in the definition of soil physicaldegradation.

The second last important group of variables involvedchanges in steady infiltration rate (fc), bulk density (rb), andsaturated moisture content (us) (Fig. 8). These soil propertiesrepresent relative soil compaction (Valentine and Bresson, 1992;Ball et al., 1997; Jones et al., 2003). Their occurrence below soilproperties related to structure in Fig. 8 suggest that soil

compaction follows deterioration of soil structure in thedefinition of physical degradation. Percent change in silt contentwas the least important variable for defining soil physicaldegradation (Fig. 8). A change in silt content, which represents achange in soil particle distribution (such as due to soil loss),implies that any alteration in soil texture comes after thedeterioration of structure and compaction in the definition ofsoil physical degradation.

The above sequence of variable importance in the definition ofphysical degradation also suggests a possible similar sequence ofsensitivity of the processes in the degradation. Soil physicaldegradation may therefore be regarded as a process which beginswith deterioration of soil structure and ends in deferential loss ofparticles through soil erosion.

3.3. Characteristics of the evidence of physical degradation

Soil properties were compared for plots with observablesymptoms of degradation and without any degradation symptoms.Majority of plots with symptoms of degradation had lower averagesoil properties (and high bulk density) than plots withoutdegradation symptoms (Table 4). This correlation points to thepotential of observable degradation symptoms in identifyingdegraded from non-degraded soil.

The average risk of soil loss from all the plots was 22.3tonnes ha�1 yr�1 with standard deviation of 2.5 tonnes ha�1 yr�1.There was low standard deviation (1.1 tonnes ha�1 yr�1) for plotswith low risk of soil loss (<15 tonnes ha�1 yr�1) and high standarddeviation (5.5 tonnes ha�1 yr�1) for plots with high risk of soil loss(>22 tonnes ha�1 yr�1). The increase in uncertainty with predictedrisk of soil loss suggests that the input data need to be furtherassessed to improve accuracy of the RUSLE model. Alternatively, it

Page 10: Assessment of soil physical degradation in Eastern Kenya by use of a sequential soil testing protocol

Fig. 8. Hierarchical tree model for the definition of soil physical degradation in Cambisols, Arenosols, and Ferralsols of the Upper Athi River Basin in Eastern Kenya.

C.T. Omuto / Agriculture, Ecosystems and Environment 128 (2008) 199–211208

could also suggest that the model was only fairly accurate inestimating low risk of soil loss. Other studies have also shown thatthe model is unreliable in predicting high risk of soil loss (Lewis et al.,2005; Lee and Lee, 2006). In spite of RUSLE’s unreliability inpredicting high risk of soil loss, its comparative accuracy inpredicting low risk of soil loss could still support the objective ofidentifying non-degraded from degraded soil. Hence, its applicationwas still valid in the sequential testing of soil for physicaldegradation.

The average potential risk of soil loss from degraded siteswas 27 tonnes ha�1 yr�1 with standard deviation of 2.8tonnes ha�1 yr�1 while the average potential risk of soil lossfrom non-degraded sites was 17 tonnes ha�1 yr�1 with standarddeviation of 1.5 tonnes ha�1 yr�1. These averages were signifi-cantly different at 95% confidence interval. Hence, there was apromise in using RUSLE outputs to separate degraded from non-degraded soil.

Table 4Characteristics of soil physical properties and degradation symptoms

Soil property Symbol Observable signs of degradation

Without signs

(mean)

Without s

(standard

Steady infiltration rate fc (cm h�1) 50.8 3.05

Sorptivity S (cm h�0.5) 31.9 1.51

Saturated moisture content us (cm3 cm�3) 0.43 0.05

Air-entry potential ha (m) 0.19 0.05

Dexter’s index SI 0.03 0.002

Aggregate stability As 0.68 0.05

Bulk density rb (g cm�3) 1.36 0.09

Sand Sand (%) 41 0.35

Silt Silt (%) 24 1.37

Clay Clay (%) 35 1.28

Number of plots – 71

Infrared spectral reflectance was also found to have somepotential in predicting soil physical conditions. This is illu-strated in the summary of calibrations between infrared spectralreflectance and soil physical properties in Table 5. Relativelygood validation statistics were found between spectra andbulk density, steady infiltration rate, index of aggregate stability,and clay content (r2 > 0.6). Some of these soil propertieswere among the important variables defining physical degrada-tion (Fig. 8). Their relationship with spectra implies that spectralreflectance is a possible surrogate indicator of soil physicaldegradation.

Apart from the calibration with soil physical properties, thespectra also differentiated between degraded and non-degradedsoil. A plot of the scores of first principal component (explaining57% of the spectral variation) and second principal component(explaining 19% of the spectral variation) attempted to separatedegraded from non-degraded soil (Fig. 9).

Difference in mean

igns

deviation)

With signs

(mean)

With signs

(standard deviation)

p-Value (5%)

18.1 3.95 0.0001

13.0 2.87 0.0001

0.27 0.07 0.0001

0.43 0.05 0.0001

0.02 0.003 0.001

0.37 0.05 0.0001

1.46 0.05 0.0001

40 0.34 0.15

35 4.33 0.0001

25 2.24 0.0001

109

Page 11: Assessment of soil physical degradation in Eastern Kenya by use of a sequential soil testing protocol

Table 5Summary statistics of calibration between spectra and soil physical properties

Soil property Symbol Calibration set

(120 plots)

Validation set

(60 plots)

r2a RMSEb r2 RMSE

Steady infiltration rate fc (cm h�1) 0.72 0.08 0.68 0.10

Sorptivity S (cm h�0.5) 0.57 0.70 0.54 0.80

Saturated moisture content us (cm3 cm�3) 0.72 0.02 0.67 0.03

Air-entry potential ha (m) 0.64 0.36 0.58 0.39

Dexter’s index SI 0.63 0.17 0.61 0.21

Aggregate stability As 0.7 0.01 0.66 0.01

Bulk density rb (g cm�3) 0.81 0.05 0.76 0.02

Sand Sand (%) 0.77 0.11 0.71 0.23

Silt Silt (%) 0.61 0.27 0.57 0.30

Clay Clay (%) 0.71 0.05 0.67 0.12

a r2 is the coefficient of determination.b RMSE is the root mean square error.

C.T. Omuto / Agriculture, Ecosystems and Environment 128 (2008) 199–211 209

3.4. Sequential testing of soil for physical degradation

3.4.1. Using observable symptoms of the degradation in the field

Soil testing model using observable symptoms of degradationseparated degraded from non-degraded soil with a validationaccuracy of 64% and a confidence level of 57% (Table 6). Therewere eleven degraded and seven non-degraded plots misclassi-fied by the model. Misclassification of degraded plots waslargely due to lack of prominent symptoms of degradation. Nineof these plots belonged to cropland while the remaining twowere in grassland. Surface modification by tillage operations incroplands and the presence of grass cover in grasslands couldhave masked the symptoms of degradation and thereforeleading to their misclassification. Non-degraded plots weremisclassified due to the presence of desert-like features, whichcould be easily mistaken for degradation symptoms. Four ofthese plots belonged to savannah shrublands while theremaining three were on rocky slopes.

Majority of correctly identified degraded plots had high bulkdensity, high proportion of silt, and very low infiltrationcharacteristics. These soil characteristics are associated with thelate stages of progressing soil physical degradation (Fig. 8). Theircorrelation with observable symptoms of degradation means thatthe symptoms are also associated with final stages of physicaldegradation. It is important to note that although the model isquite uncertain in its prediction (low confidence interval of 57%), it

Fig. 9. Scores of the first two principal components for infrared spectral reflectance.

is more accurate in identifying degraded than non-degraded plots(Table 6). Having association with late stages of degradationdevelopment and relatively high model sensitivity imply that theobservable symptoms of degradation are only fairly accurate indetecting advanced physical degradation.

3.4.2. Use of the risk of soil loss from RUSLE model

When the risk of soil loss was added to the soil testing modelwith observable symptoms of degradation, the validation accuracyimproved to 82% (Table 7). This model managed to detect the threedegraded lots from croplands and three non-degraded plots fromrocky slopes previously misclassified by observable degradationsymptoms (Table 7). The plots from croplands had high silt contentand were positively picked by RUSLE model as degraded. Thepresence of high silt fractions could not be easily detected by visualobservation; thus, the plots were misclassified by observablesymptoms of degradation. The plots from rocky slopes misclassi-fied by observable symptoms had low risk of soil loss due to low LS-factors arising from short lengths of slope.

The soil testing model with RUSLE outputs still did notcorrectly classify eight degraded and four non-degraded plots inthe validation set (Table 7). The misclassified degraded plotswere from a sedimentation plane with a mix of tall grass andsparse vegetation. Nevertheless, the model was able to correctlyidentify misjudgements by visual assessment and improved thesoil testing accuracy by over 10%. It also improved the confidenceinterval to 80%; thus, indicating improved repeatability ofcombined application of visual assessment and soil lossmodelling. These improvements reinforce the argument thatsoil degradation assessment by expert opinion should beaugmented with erosion modelling to improve the assessmentaccuracy (FAO, 2003).

3.4.3. Use of infrared spectral reflectance

Further improvement on the soil testing protocol was achievedby using infrared spectral reflectance. Seven principal compo-nents, which accounted for 97% of the spectral variation, wereadded to the soil testing model. The validation accuracy for themodel improved to 95% (Table 8). Infrared spectral reflectancemanaged to resolve soil physical conditions of some plotspreviously misclassified by the combined application of visualassessment and risk of soil loss. The spectra correctly identifiedthe three non-degraded plots in the shrublands which werepreviously misclassified by RUSLE and visual assessment.Although these plots had desert-like features, they were foundto be spectrally similar to other non-degraded plots. Similarly, sixplots from the sedimentation plane were also found to bespectrally similar to degraded plots. They had no degradationsymptoms and were therefore misclassified as non-degraded byvisual assessment. They were also misclassified by the RUSLEmodel due to the model’s poor accountability of sand deposits(Renard et al., 1997).

The soil testing model with spectral reflectance was able tocorrectly identify degraded soil misclassified by the combinedapplication of soil loss modelling and visual assessment.The improved accuracy with spectra shows that infraredspectral reflectance can detect subtle changes in soil physicalconditions which are not yet apparent to visual assessmentand soil loss modelling. In this regard, infrared spectralreflectance behaves as a sensitive indicator of soil physicaldegradation. Due to improved accuracy and repeatability(confidence interval of 95%), the spectral can be regarded as areliable indicator for incipient degradation. This is particularlyimportant for the assessment of early-warning signs of soilphysical degradation.

Page 12: Assessment of soil physical degradation in Eastern Kenya by use of a sequential soil testing protocol

Table 7Confusion matrix for the soil testing model using observed symptoms of degradation and risk of soil loss

Calibration set Validation set

Model predictions Model predictions

Degradation classes Non-degraded Degraded Total Degradation classes Non-degraded Degraded Total

Non-degraded 40 8 48 Non-degraded 13 4 17

Degraded 9 77 86 Degraded 8 21 29

aSensitivity = 90% and specificity = 83% Sensitivity = 76% and specificity = 72%

Overall predicted accuracy is 80% on the validation set and confidence interval is 80%

a Sensitivity = correct identification of degraded soil and specificity = correct identification of non-degraded soil.

Table 8Confusion matrix for the soil testing model using observed symptoms of degradation, risk of soil loss, and infrared spectral reflectance

Calibration set Validation set

Model predictions Model predictions

Degradation classes Non-degraded Degraded Total Degradation classes Non-degraded Degraded Total

Non-degraded 46 2 48 Non-degraded 16 1 17

Degraded 4 84 88 Degraded 2 27 29

aSensitivity = 96% and specificity = 95% Sensitivity = 94% and specificity = 93%

Overall predicted accuracy is 95% on the validation set and confidence interval is 95%

a Sensitivity = correct identification of degraded soil and specificity = correct identification of non-degraded soil.

Table 6Confusion matrix for the soil testing model using observed symptoms of degradation

Calibration set Validation set

Model predictions Model predictions

Degradation classes Non-degraded Degraded Total Degradation classes Non-degraded Degraded Total

Non-degraded 29 19 48 Non-degraded 10 7 17

Degraded 21 65 86 Degraded 11 18 29

aSensitivity = 76% and specificity = 60% Sensitivity = 62% and specificity = 59%

Overall predicted accuracy is 64% on the validation set and confidence interval is 57%

a Sensitivity = correct identification of degraded soil and specificity = correct identification of non-degraded soil.

C.T. Omuto / Agriculture, Ecosystems and Environment 128 (2008) 199–211210

4. Conclusion

This study shows a consistent evidence of the decline in soilstructure which gives way to compaction and finally soil loss. Thisevidence supports the suggestion that soil physical degradation is agradual process which begins with structural deterioration andends into the differential loss of soil particles through erosion. Asequential soil testing protocol presented in the study has thepotential of screening soil samples for evidence of any stage of soilphysical degradation. The protocol takes into account changes insoil physical properties as the basic foundation for defining soilphysical degradation. Although soil physical properties are difficultand expensive to measure, limited sampling over time of selectedimportant soil properties can still facilitate the definition ofphysical degradation. For assessing soil physical degradation inCambisols, Arenosols, and Ferralsols in the Upper Athi Riverwatershed in Eastern Kenya, the selected soil properties includeDexter’s index of soil quality, index of aggregate stability, air-entrypotential, steady infiltration rate, bulk density, porosity, and siltcontent.

The sequential soil testing protocol showed that visualassessment of observable degradation symptoms is a lessexpensive and rapid method for assessing advancing physicaldegradation. However, the repeatability of the method is quite low

(with about 60% confidence interval). Improvements in thismethod can be achieved through augmented use with erosionmodeling. About 80% repeatability and accuracy can be achievedwith the combined application of erosion modeling and visualassessment of soil physical degradation. This study has also shownthat diffuse infrared spectral reflectance is a potential surrogatepredictor of soil physical conditions. Spectral reflectance isespecially sensitive to subtle changes in soil physical conditionsnot readily amenable to visual assessment and erosion modeling.Hence, it can be used to test for early-warning signs of soil physicaldegradation. Inclusion of spectra in the sequential soil testingimproved the accuracy and repeatability of testing to 95%.

The sequential soil testing protocol presented in this study maybe useful in designing a soil degradation monitoring framework fortimely advice to control soil physical degradation. Further testingand worldwide application with different models in different soiltypes and landscapes is recommended.

Acknowledgements

This study was funded by International Foundation for Sciences(IFS, www.ifs.se) through the project number C/9353-1. Soilspectral reflectance was determined at ICRAF, through thegenerous support from Dr. Keith Shepherd and the entire ICRAF

Page 13: Assessment of soil physical degradation in Eastern Kenya by use of a sequential soil testing protocol

C.T. Omuto / Agriculture, Ecosystems and Environment 128 (2008) 199–211 211

staff. Prof. Alex McBratney and Dr. Budiman Minasny of SydneyUniversity are also acknowledged for their valuable input duringthe development of this manuscript.

References

Analytical Spectral Devices Inc., 1997. FieldSpecTM User’s Guide. Analytical SpectralDevices Inc., Boulder, CO.

Ball, B.C., Campbell, D.J., Douglas, J.T., Henshall, J.K., O’ullivan, M.F., 1997. Soilstructural quality, compaction and land management. Eur. J. Soil Sci. 48,593–601.

Ben-Dor, E., Goldlshleger, N., Benyamini, Y., Agassi, M., Blumberg, D.G., 2003. Thespectral reflectance properties of soil structural crusts in the 1.2–2.5 mmspectral region. Soil Sci. Soc. Am. J. 67, 289–299.

Breiman, L.J., Friedman, J., Stone, C.J., Olshen, R.A., 1984. Classification and Regres-sion Trees. CRC Press, Boca Raton, FL, 358 pp.

Bro, R., 1996. Multiway calibration: multilinear PLS. J. Chemometrics 10, 47–61.Brooks, R.H., Corey, A.T., 1964. Hydraulic Properties of Porous Medium. Hydrology

Paper, No. 3. Colorado State University (Fort Collins).Chong, S.K., Green, R.E., 1983. Sorptivity measurement and its application. In: ASAE

(Eds.), Advances in Infiltration: Proceedings of the National Conference onAdvances in Infiltration. Chicago, December 12–13. America Society of Agri-cultural Engineers, Illinois.

Cook, D.J., Dickinson, W.T., Rudra, R.P., 1985. GAMES-Guelph Model for Evaluatingthe Effects of Agricultural Management Systems on Erosion and Sedimentation.University of Guelph. School of Engineering, Guelph.

Desmet, P., Govers, G., 1996. GIS-based procedure for automatically calculating theUSLE LS factor on topographically complex landscape units. J. Soil WaterConserv. 51, 427–433.

Dexter, A.R., 2004. Soil physical quality. Part I. Theory, effects of soil texture, density,and organic matter, and effects on root growth. Geoderma 120, 201–214.

Diamond, J., Shanley, T., 2003. Infiltration rate assessment of some major soils. IrishGeog. 36, 32–46.

Diaz-Zorita, M., Grove, J.H., Perfect, E., 2002. Aggregation, fragmentation, andstructural stability measurement. Marcel Dekker Encyclo. Soil Sci. 37–40.

Dirksen, C., 1999. Soil Physics Measurements. Catena-Verlag, Reiskirchen, 154 pp.Eijkelkamp Agrisearch Equipment, 2006. Wet-sieve Apparatus: User’s Manual.

Eijkelkamp Agrisearch Equipment, Giesbeek, Netherlands.Elrick, D.E., Reynolds, W.D., 2002. Measuring water transmission parameters in

vadose zone using ponded infiltration techniques. Agric. Sci. 7, 17–22.Eshel, G., Levy, G.J., Singer, M.J., 2004. Spectral reflectance properties of crusted soils

under solar illumination. Soil Sci. Soc. Am. J. 68, 1982–1991.ESRI, 2004. ArcGIS1 8.3. Redlands, CA, www.esri.com.FAO, 2003. Data Sets, Indicators and Methods to Assess Land Degradation in

Drylands. FAO, Rome, Italy.Feddema, J.J., 1998. Estimated impacts of soil degradation on the African water

balance and climate. Clim. Res. 10, 127–141.Foster, G.R., Wischmeier, W.H., 1974. Evaluating irregular slopes for soil loss

prediction. Trans. Am. Soc. Agric. Eng. 17, 305–309.GARMIN International, 2002. GARMIN1 12 XL Handheld GPS. Olathe, Kansas.Gee, G.W., Bauder, J.W., 1986. Particle-size analysis. In: Page, A.L. (Ed.), Methods of

Soil Analysis. Part I. Physical and Mineralogical Methods. 2nd ed. Agronomy, pp.383–411.

Hammad, A., Lundekvam, H., Borresen, T., 2004. Adaptation of RUSLE in the easternpart of the Mediterranean region. Environ. Manage. 34, 829–841.

Hengl, T., Heuvelink, G.B.M., Rossiter, D.G., 2007. Regression kriging: from equationsto case studies. Comp. Geosci. 33, 1301–1315.

Hickey, R., Smith, A., Jankowski, P., 1994. Slope length calculations from a DEMwithin Arc/Info grid. Comp. Environ. Urban Sys. 18, 365–380.

Hodnett, M.G., Tomasella, J., 2002. Marked difference between van Genuchten soilwater-retention parameters for temperate and tropical soils: a new water-retention pedo-transfer function developed for tropical soils. Geoderma 108,155–180.

Jones, J.A.R., Spoor, G., Thomasson, A.J., 2003. Vulnerability of subsoils in Europe tocompaction: a preliminary analysis. Soil Till. Res. 73, 131–143.

Kutilek, M., Nielsen, D., 1994. 364 pp. In: Soil Hydrology, CATENA-VERLAG, Crem-lingen-Destedt.

Lal, R., 2000. Physical management of soils of the tropics: priorities for the 21stcentaury. Soil Sci. 165, 191–207.

Lee, G.S., Lee, K.H., 2006. Scaling effect for estimating soil loss in the RUSLE modelusing remotely sensed geospatial data in Korea. Hydro. Earth Syst. Sci. Discus-sions 3, 135–157.

Lewis, L.A., Verstraeten, G., Zhu, H., 2005. RUSLE applied in a GIS framework:calculating the LS factor and deriving homogeneous patches for estimatingsoil loss. Int. J. Geog. Inf. Sci. 19, 809–831.

Lufafa, A., Tenywa, M.M., Isabirye, M., Majaliwa, M.J.G., Woomer, P.L., 2003. Pre-diction of soil erosion in a Lake Victoria basin catchment using a GIS-baseduniversal soil loss model. Agric. Sys. 76, 883–894.

Marshall, T.J., Holmes, J.W., Rose, W.C., 1996. Soil Physics. Cambridge UniversityPress, London, 346 pp.

Millward, A.A., Mersey, J.E., 1999. Adapting the RUSLE to model soil erosionpotential in a mountainous tropical watershed. CATENA 38, 109–129.

Minasny, B., McBratney, A.B., 2003. Integral energy as a measure of soil-wateravailability. Plant Soil 249, 253–262.

Moore, T.R., 1979. Rainfall erosivity in east Africa: Kenya, Tanzania, Uganda. Geog.Ann. Ser. A Phys. Geog. 61, 147–156.

Morgan, R.P.C., 1995. Soil Erosion and Conservation. Addison-Wesley Longman,Edinburgh, 198 pp.

Oldeman, L.R., Hakkeling, R.T.A., Sombroek, W.G., 1991. World Map of the Status ofHuman-Induced Degradation: An Explanatory Note. ISRIC, Wageningen, TheNetherlands/UNEP, Nairobi, Kenya.

Omuto, C.T., 2007. HydroMe: estimation of soil hydraulic parameters from experi-mental data. http://cran.r-project.org/web/packages/HydroMe/index.html. RProject, USA.

Omuto, C.T., Shrestha, D., 2007. Remote sensing techniques for rapid detection ofsoil physical degradation. Int. J. Rem. Sens. 28, 4785–4805.

Omuto, C.T., Minasny, B., Mcbratney, A.B., Biamah, E.K., 2006. Nonlinear mixedeffects modelling for improved estimation of water retention and infiltrationparameters. J. Hydrol. 330, 748–758.

Paglia, M., Jones, R., 2002. Sustainable Land Management—Environmental Protec-tion: A Soil Physical Approach. Catena-Verlag, Resikisrchen, Germany, 588 pp.

Philip, J.R., 1957. The theory of infiltration. I. The infiltration equation and itssolutions. Soil Sci. 83, 345–357.

Pieri, L., Bittelli, M., Wu, J.Q., Dun, S., Flanagan, D.C., Pisa, P.R., Ventura, F., Salvator-elli, F., 2007. Using the water erosion prediction project (WEPP) model tosimulate field-observed runoff and erosion in the Apennines mountain range,Italy. J. Hydrol. 336, 84–97.

Renard, K.G., Freidmund, J.R., 1994. Using monthly precipitation data to estimatethe R-factor in the RUSLE. J. Hydrol. 157, 287–306.

Renard, K.G., Foster, G.R., Weesies, G.A., McCool, D.K., 1997. Predicting soil erosionby water: a guide to conservation planning with the revised Universal Soil LossEquation (RUSLE). USDA Agricultural Handbook, vol. 703. p. 404.

R Development Core Team, 2008. R: A Language and Environment for StatisticalComputing. R Foundation for Statistical Computing, Vienna, Austria. , In: http://www.r-project.org.

Reynolds, W.D., Elrick, D.E., 1990. Ponded infiltration from a single ring. I. Analysisof steady flow. Soil Sci. Soc. Am. J. 54, 1233–1241.

Salford Systems Inc., 2008. CART1 V 6.0. Salford Systems Inc., San Diego.Shamshad, A., Azhari, M.N., Isa, M.H., Wan Hussin, W.M.A., Parida, B.P., 2008.

Development of an appropriate procedure for estimation of RUSLE EI30 indexand preparation of erosivity maps for Palau Penang in Peninsula Malaysia.CATENA 72, 423–432.

Shepherd, K.D., Palm, C.A., Gachengo, C.N., Vanlauwe, B, 2003. Rapid character-ization of organic resource quality for soil and livestock management intropical agroecosystems using near-infrared spectroscopy. Agron. J. 95,1314–1322.

Sombroek, W.G., Braun, H.M.H., van der Pouw, B.J.A., 1982. Exploratory Soil Map andAgro-climatic Zone Map of Kenya, 1980, Scale 1:1.000.000. Kenya Soil Survey.Ministry of Agriculture, Nairobi.

Stocking, M., Murnaghan, N., 2001. Handbook for the Field Assessment of LandDegradation. Earthscan Publishers Ltd., London, 121 pp.

Tiffen, M., Mortimore, M., Gichuki, F., 1994. More People, Less Erosion: Environ-mental Recovery in Kenya. John Willey and Sons, Overseas DevelopmentInstitute, England.

Torri, D., Poessen, J., Borselli, L., 1997. Predictability and uncertainty of the soilerodibility factor using global dataset. CATENA 31, 1–22.

Valentine, C., Bresson, L.M., 1992. Morphology, genesis and classification of surfacecrusting loamy and sandy soils. Geoderma 55, 225–245.

van der Kniff, J.M., Jones, R.J.A., Montanarella, L., 1999. Soil Erosion Risk Assessmentin Italy. JRC, Ispra, Italy.

van Genuchten, M.Th., 1980. A closed-form equation for predicting the hydraulicconductivity of unsaturated soils. Soil Sci. Soc. Am. J. 44, 892–898.

van Remortel, R.D., Maichle, R.W., Hickey, R.J., 2004. Computing the LS factor forthe revised universal soil loss equation through array-based slope proces-sing of digital elevation data using a C++ executable. Comp. Geosci. 30,1043–1053.

Vrieling, A., 2006. Satellite remote sensing for water erosion assessment: a review.CATENA 65, 2–18.

Wessels, K.J., Prince, S.D., Frost, P.E., van Zyl, D., 2004. Assessing the effects ofhuman-induced land degradation in the former homelands of northern SouthAfrica with a 1 km AVHRR NDVI time-series. Rem. Sens. Env. 94, 47–67.

Wischmeier, W.H., Smith, D.D., 1978. Predicting rainfall erosion losses—a guide toconservation planning. USDA Agriculture Handbook No. 537.

WRB (World Reference Base), 2006. World reference base for soil resources: aframework for international classification, correlation and communication.World Soil Resources Reports No. 103. FAO, Rome. 132 pp.

Yin, S., Xie, Y., Nearing, M.A., Wang, C., 2007. Estimation of rainfall erosivity using 5-to 60-min fixed interval rainfall data from China. CATENA 70, 306–312.