Research Article Soil-Landscape Modeling and Remote ...downloads.hindawi.com/journals/aess/2013/798094.pdf · of a soil-landscape model using DEM and remote sensing to predict the

Hindawi Publishing CorporationApplied and Environmental Soil ScienceVolume 2013 Article ID 798094 11 pageshttpdxdoiorg1011552013798094

Research ArticleSoil-Landscape Modeling and Remote Sensing to Provide SpatialRepresentation of Soil Attributes for an Ethiopian Watershed

Nurhussen Mehammednur Seid1 Birru Yitaferu2 Kibebew Kibret3 and Feras Ziadat4

1 Burie Agricultural College PO Box Rissneleden 19 Sundbyberg 17453 Stockholm Sweden2 Amhara Regional Agricultural Research Institute (ARARI) PO Box 527 Bahir Dar Ethiopia3 School of Natural Resources Management and Environmental Science Haramaya University PO Box 138 Dire Dawa Ethiopia4 International Center for Agricultural Research in the Dry Areas (ICARDA) PO Box 950764 Amman 11195 Jordan

Correspondence should be addressed to Feras Ziadat fziadatcgiarorg

Received 5 June 2013 Revised 29 August 2013 Accepted 3 September 2013

Academic Editor Davey Jones

Copyright copy 2013 Nurhussen Mehammednur Seid et al This is an open access article distributed under the Creative CommonsAttribution License which permits unrestricted use distribution and reproduction in any medium provided the original work isproperly cited

Information about the spatial distribution of soil properties is necessary for natural resources modeling however the cost of soilsurveys limits the development of high-resolution soil maps The objective of this study was to provide an approach for predictingsoil attributes Topographic attributes and the normalized difference vegetation index (NDVI) were used to provide informationabout the spatial distribution of soil properties using clustering and statistical techniques for the 56 km2 Gumara-Maksegnitwatershed in Ethiopia Multiple linear regression models implemented within classified subwatersheds explained 6ndash85 of thevariations in soil depth texture organic matter bulk density pH total nitrogen available phosphorous and stone content Thepredictionmodel was favorably comparable with the interpolation using the inverse distanceweighted algorithmTheuse of satelliteimages improved the predictionThe soil depth prediction accuracy dropped gradually from 98 when 180 field observations wereused to 65 using only 25 field observations Soil attributes were predicted with acceptable accuracy even with a low density ofobservations (1-2 observations2 km2) This is because the model utilizes topographic and satellite data to support the statisticalprediction of soil properties between two observations Hence the use of DEM and remote sensing with minimum field dataprovides an alternative source of spatially continuous soil attributes

1 Introduction

Quantitative information and spatial distribution of soilproperties are among the main prerequisites for achievingsustainable land management The accuracy of soil infor-mation determines to a large extent the reliability of landresources management decisions [1ndash3] Conventional soilsurveys are usually used to derive information about soilsand their distribution [4] however limited areas are coveredby detailed soil information mainly due to the high costsof surveys [5] Furthermore the spatial distribution of soilcharacteristics as represented by a conventional soil map doesnot reflect the distribution in the field because of the polygon-based mapping employed [6] In polygon-based mappingsoils are assumed to be homogeneouswithin the polygon andabrupt changes take place at the boundaries between poly-gons [7 8] Soils usually show a diffuse spatial distribution

that is hard to address in polygon-based soil maps [9] Manyresearchers have suggested continuous raster maps as a betteralternative to mapping soils and their properties [6 7 9]

In soil science the implementation of geomaticsmdashGISGPS remote sensing and digital elevationmodels (DEMs)mdashis suggesting new alternatives [3 10] Global coverage of high-resolution imagery and terrain data is increasing and thisenhances the popularity of digital soil mapping to generateaccurate soil maps that capture many properties with reason-able effort [11] The benefit of using GIS is the production ofdigital soil attributes and digital soil maps to optimize theneed for costly field work and laboratory analysis [10 12]

Various statistical models are used to investigate the rela-tionships between the spatial distribution of soil attributesand that of landscape attributes Predictive mapping tech-niques such as geostatistics (ie kriging and cokriging)

2 Applied and Environmental Soil Science

fuzzy logic linear and multiple regression regression treesand neural networks have been used to develop soilmaps [1013] Ongoing research in digital soil mapping indicates thatquantitative prediction models are promising tools to pro-duce soilmapswith acceptable accuracy [14 15] Research hasprovided optimistic results and some researchers obtainedbetter results than traditional soil surveys [16ndash18] The useof satellite data to complement topographic informationimproves the mapping of natural resources [10 13 19]These investigations are based on the relationship betweenlandscape pattern and the differentiation of soil types andattributes [20] The spatial distribution of specific soil prop-erties has been predicted using soil-landscapemodelingThisincluded sand silt and organic matter contents A-horizonthickness solum depth extractable phosphorus and pH [2122] Using different terrain attributes the empirical modelsdeveloped in these studies explained 41ndash68 of the variationin soil properties [21 22]

GIS is used to quantify the relationships between soilsand topographic and remote sensing variables and to aid thedigital soil mapping efforts [2 3 10 23 24] Topographicparameters are derived from DEMs and used to predictthe distribution of soil characteristics [5 22 25 26] Theincreasing availability of high-resolution remote sensing dataprovides a new opportunity for predicting soil character-istics with acceptable accuracy [11] Many researchers havereported on the contribution of remote sensing data inproviding acceptable prediction of soil characteristics [27ndash29] Nevertheless there are many issues that warrant furtherresearch the conjunctive use of satellite data and topographicdata to improve the prediction of soil characteristics and thebest time of year to acquire satellite data The use of variousnumbers (densities) of observations and how that affectsaccuracy will help researchers decide on a reasonable numberof observations to maintain acceptable accuracy with anappropriate amount of field work The use of soil-landscapemodels to predict soil attributes is theoretically better thaninterpolation techniques because the former uses topogra-phy and satellite data as a background to guide predictionwhile interpolation relies solely on the spatial relationshipsbetween adjacent observations without considering the vari-ations in factors that drive soil differentiation Howeverthe merits of using soil-landscape modeling compared withinterpolation techniques need verification and quantificationespecially when a limited number of observations are avail-able The objective of this study is to investigate the utilityof a soil-landscape model using DEM and remote sensing topredict the spatial distribution of soil characteristics with anoptimum number of field observations

2 Materials and Methods

The study area is located in the Lake Tana basin in the Gum-ara-Maksegnit watershed of the Amhara region in Ethiopiawithin 12∘241015840ndash12∘311015840N and 37∘341015840ndash37∘371015840E It is centered45 km southwest of Gondar town and covers an area of56 km2 Altitude within the watershed ranges from 1923 to2851m above sea level with topography ranging from gentle

Observations for predictionObservations for validationStream network

SubwatershedsMain watersheds

N

(km)0 05 1 2

343500 345500 347500 349500

343500 345500 347500 349500

1383

500

1381

500

1379

500

1377

500

1375

500

1373

500

1371

500

1383

500

1381

500

1379

500

1377

500

1375

500

1373

500

1371

500

Figure 1 Distribution of field observations watershed boundarystream network and subwatershed divisions for Gumara-Maksegnitwatershed

to sharp steep slopes The mean annual rainfall is 1052mmThe mean minimum and maximum temperatures are 133and 285∘C respectivelyThe study area is characterized by anustic moisture regime [30] The natural resource base hasbeen depleted to a great extent due to soil erosion andimproper land use which resulted in reduced tree coverbiodiversity agriculture range and pasture productivity

21 Soil Survey and Field Observations The area was dividedinto square grids (500m times 500m) The total area of thewatershed and the heterogeneity of topography and vegeta-tion cover and consequently the soils were used to guidethe selection of this grid size Larger grid size (1 km) wouldresult in a limited number of observations and would limitthe representation of soil variability Soil samples were takenwithin each grid in the watershed (Figure 1) This was a com-promise between grid and free sampling techniques Whilethe surveyor was free to take the sample within the definedgrid to represent the area the distribution of sampling sites

Applied and Environmental Soil Science 3

still followed an unbiased grid In cases where many gridsshared common features the surveyor reduced the numberof samplings to represent these grids with one sampling siteEach site was excavated to a depth of 100 cm or an impedinglayer using an auger FAO terminology [30] was used todescribe the sampling sites The soil attributes and site char-acteristics recorded were GPS coordinates (easting northingand elevation) soil depth using an auger slope steepness(percent) estimated using clinometers surface stone cover(percent) and stone content of the top soil layer (0ndash25 cm) Asoil sample for each soil observation was taken for laboratoryanalysisThe following soil attributes were analyzed clay siltsand organic matter total nitrogen and available phosphoruscontents pH and bulk density

The hydrometer method outlined by the simplified pro-cedure of Day was used to determine soil particle size distri-bution [31] Hydrogen peroxide (H

2O2) was used to destroy

the organicmatter and sodiumhexametaphosphate (NaPO3)

was used as a dispersing agent The bulk density (Bd) of thesoil was estimated from undisturbed (core) soil samples col-lected using a core sampler weighed at fieldmoisture contentand then determined following the procedures of Blake [32]

Soil pH was measured using a digital pH meter in thesupernatant suspension of 1 25 soil liquid ratio where theliquids were water and 1M KCl solution and pH was calcu-lated by subtracting soil pH (KCl) from soil pH (H

2O) The

organic carbon content of the soil was analyzed following thewet digestionmethod described byWalkley and Black whichinvolves digestion of organic carbon in the soil samples withpotassium dichromate (K

2Cr2O7) in sulfuric acid solution

[33]The Kjeldahl procedure was followed for the determina-tion of total nitrogen as described by Sahlemeden andTaye byoxidizing the organic matter with concentrated sulfuric acidand converting the nitrogen in the organic compounds intoammonium sulfate during the oxidation [34]

Available phosphorus was determined by Olsen et almethod [35]The soil sampleswere shakenwith 05Msodiumbicarbonate at a nearly constant pH 85 in 1 2 soil solventratio for 30min and the extract was obtained by filteringthe suspension as indicated by Olsen et al The availablephosphorus extracted was measured by spectrophotometerfollowing the procedure outlined by Murphy and Riley [36]

22 DEM and Terrain Analysis Shuttle Radar TopographyMission (SRTM) 90m resolution DEM was used for thestudy The free availability and accuracy of this productchecked against field observations were the main reasons forits selection Previous research recommended some terrainattributes as best predictors of soil characteristics [37 38]The following terrain parameters were derived using standardcommands in ArcGIS aspect profile plan and mean curva-tures flow accumulation area and slopeThe average upslopecontributing area for each pixel (upslope flow accumulation)was calculated by multiplying the average flow accumulationgrid by the area of the pixel (flow accumulation times 8100m2)The compound topographic index (CTI) for each pixel wascalculated using the formula [21] CTI = ln (AstanD) whereAs is the average upslope contributing area and D is the

(km)0 05 1 2

Class1234567

Area (ha)3ndash2545ndash703ndash2570ndash10080ndash9020ndash13018ndash45

Slope ()20ndash600ndash400ndash200ndash2030ndash500ndash2020ndash60

N

W E

S

Figure 2 Distribution and characteristics of the seven subwater-sheds classes

average slope degree ARCSWAT was used to derive thestream network and the watershed was also automaticallydivided into a number of subwatersheds (Figure 1)

The watershed subdivisions were considered as a baseunit in the prediction model analyses This is because thecorrelation between terrain attributes and soil characteristicsfor the whole watershed was very low and therefore unsuit-able to predict the latter from the former The subdivisionof the whole watershed into smaller subwatersheds groupedthe soil observations into homogeneous units and so enabledthe establishment of better statistical relationships [37] Eachsmall subwatershed was divided by the passing streamlineinto two facets (subdivisions) (Figure 1)The generated facetswere grouped into classes (Figure 2) based on their charac-teristics of area and slope If individual subwatersheds wereconsidered directly the number of observations within eachsubwatershed was not enough to establish a rigorous statisti-cal relationship The grouping of subwatersheds into classesbased on characteristics such as area and slope increasedthe number of observations within each class and enabledgood prediction The selection of an appropriate number


of classes to cluster the subwatershed facets is important ingenerating good results [37] The classification (grouping) ofsubwatershed facets was repeated to produce a classificationthat led to an approximately equal number of observations ineach class (at least 15 observations)This allowed a reasonableregression model to be established for each classThe clusterswere used in the prediction of soil attributes Seven classeswere found to be optimum for predicting soil attributes in thestudy area

23 Generating the Normalized Difference Vegetation Index(NDVI) Map ASTER satellite images taken on January 30and March 19 2007 and SPOT satellite images taken onOctober 8 and November 23 2007 were analyzed in orderto select the best image(s) for the study The images werecorrected radiometrically and geometrically using ENVIsoftware Radiometric correction was done separately forASTER and SPOT imagery on an individual band basisGeometric correction for the images within the study areawas done by taking the October SPOT image as a base imageThe corrected images were resampled to the same spatialresolution (15mtimes 15m)TheNDVI values for each pixel werecalculated using the following formula [39]

NDVI = NIR minus 119877NIR + 119877

(1)

where NIR is reflection in near infrared and 119877 is reflectionin red wavelength NDVI values of the four images werecompared based on their vegetation biomass distributionTwo images with maximum and minimum vegetation coverwere selected an ASTER image taken on March 19 whenmost of the watershed area was bare and a SPOT imagetaken on October 8 when the most was vegetated Thisenabled assessment of the best time for satellite images to betaken for improving soil prediction The other two imagesprovided less extra information and were not considered infurther analysis Theoretically the image in March reflectedvariations in soil appearance or color because the soil wasbare at that time in Ethiopia (direct effect) while the image inOctober reflected the indirect effect of soil characteristics onvegetation cover and greenness because the soil was coveredwith vegetation at that time

Terrain attributes and satellite data for each soil observa-tion were extracted Statistical analyses were implementedwithin each class between the derived terrain attributessatellite data and collected soil attributes of field observationsusing SPSS Multiple linear regression models are usuallyemployed to predict dependent (soil attributes) from inde-pendent variables (satellite data and terrain attributes) Froma total of 220 observations 180 were randomly selected foranalysis and the remaining 40 observations were used toassess the accuracy of the prediction model A regressionmodel was established for each class to predict soil attributesfrom terrain attributes and satellite data Map algebra ofArcGIS was used to obtain predicted soil attribute gridsusing the regression models and the raster grids for eachclass (slope percent contributing area CTI aspect classes

curvature classes and NDVI values of ASTER and SPOTimages) An example of these equations follows

Clay content (class 1)

= 1769 + ((grid slope) lowast 007)

+ ((grid flowaccumulation) lowast 000002)

+ ((grid CTI) lowast 012)

+ ((grid aspect) lowast 025)

+ ((grid meancurvature) lowast 054)

+ ((grid NDVI-March) lowast minus0004)

+ ((grid NDVI-October) lowast minus0006) (2)

Predictions were first executed for individual classes andwere then merged together to generate predicted values forthe whole watershedThe prediction accuracy was verified intwo ways first by comparing the predicted and observed val-ues using 40 randomly selected field observations (Figure 1)and second by comparing the predicted soil attributes withthose derived using the inverse distance weighted (IDW)technique

The role of satellite data in improving the prediction accu-racy of soil attributes using multiple linear regression modelswas assessed by repeating the previous analysis without usingthe satellite images as an independent variable This wasdone for one of the soil attributes (clay content) Regressioncoefficients (1198772) of the models without including remotesensing data were compared with those that included satel-lite data The quality of the predictions with and withoutsatellite data was also tested by examining the relationshipbetween observed and predicted values when some differ-ences between them were allowed This comparison wasimplemented by allowing a difference between predictedand observed clay content of 3 5 or 7mdashbecause it wasnot expected that predicted values would exactly equal theobserved values (where the difference allowed was zero) Infact within a distance of 1m in the field there are somedifferences in some soil attributes within these ranges

24 Selection of the Optimum Number of Observations for SoilPrediction The effect of the number of field observationsused to build the regression models on the accuracy of thepredictionmodel (sensitivity analysis) was investigated usingsoil depth as an exampleThiswas to determine theminimumnumber of observations to maintain acceptable predictionaccuracy of soil attributes Various numbers of observationswere selected starting from 180 and gradually decreasing150 120 90 60 40 30 25 and finally 20 observations Theprediction model was implemented until a threshold wasreached where the prediction accuracy declined sharply toa low level The classes used in the earlier modeling wereamalgamated to get an optimum number of observations perclass The accuracies of predicting soil depth as a result of


Table 1 Descriptive statistics of the soil attributes and site characteristics for soil observations

Variable Minimum Maximum Mean Mode Std deviationSoil depth (cm) 0 101 435 101 330Elevation (m) 1954 2852 2252 2020 220Slope () 2 90 348 30 233Stone cover at the surface () 0 65 154 10 142Stone in the soil () 0 738 174 0 132Clay content () 116 678 286 203 126Silt content () 40 498 340 398 77Sand content () 146 726 373 426 93Organic matter () 021 997 282 112 19Total nitrogen () 003 906 026 022 06Available phosphorus (mg kgminus1) 05 972 141 45 153Bulk density (g cmminus3) 081 170 125 121 02pH 534 798 672 662 041Erosion status Low Severe mdash Moderate mdashErosion type Sheet Gully mdash Rill mdash

using different numbers of soil observations (density) wereestimated using different validation indices [27 40] such asthe root mean square error (RMSE) and the mean absoluteestimation error (MAEE) As the difference between pre-dicted values andobserved values increases so do bothRMSEand MAEE Consider the following

RMSE = 1119899

119899

sum

119894=1

radic(119894minus 119910119894)

2

MAEE = 1119899

119899

sum

119894=1

10038161003816100381610038161003816119894minus 119910119894

10038161003816100381610038161003816

(3)

where n is the number of verification points 119894is predicted

soil depth and 119910119894is observed soil depth The quality of

the prediction using different densities of observations wasalso tested by examining the relationship between observedand predicted values by allowing some differences betweenthemThis enabled easy observation of the gradual change inpercentage of observations predicted correctly when differentobservation densities were used

3 Results and Discussion

Descriptive statistics for soil attributes and site characteristicswere recorded for each site (Table 1)The correlation betweensoil attributes and terrain attributes and satellite data whenconsidering the whole set of soil observations was very low(00ndash050) thus building a regression model to predict soilattributes would not produce acceptable results Howeverthe range of 1198772 between soil attributes and terrain attributesand satellite data when the whole watershed was divided intosmaller subwatersheds was 006ndash085 (Table 2) 1198772 dependedon the type of soil attribute and the subwatershed facet classfor which the relationship was being established In generalthe1198772 values were acceptable comparedwith previous studies

[40ndash42] and were used to generate predictions of the var-ious soil attributes within the seven classes However thefinal judgment of the prediction accuracy is made throughthe comparison between predicted and measured values ofsoil characteristics Significant correlations between terrainattributes and satellite images with soil attributes varied forthe same soil attribute for different classes (Table 3) Forexample in class 2 clay content was significantly correlatedwith slope percent upslope contributing area and theASTERimage taken inMarch whereas in classes 3 and 7 clay contentwas highly correlated with slope percent aspect curvaturethe ASTER image taken in March and the SPOT imagetaken in OctoberThese relationships indicate that classifyingthe watershed into classes improved the prediction of soilattributes The results were supported by findings of otherresearchers who reported low 1198772 (00ndash019) between varioussoil attributes and terrain attributes for the whole watershedwhich was also improved significantly for the classifiedsubwatersheds and consequently improved the predictionaccuracy of soil attributes [37]

The regression coefficients (Table 3) indicated that digitalterrain attributes had a stronger influence on soil propertiesThis was supported by previous studies [43 44] In naturesoil properties are highly spatially variable [45 46] andfor accurate estimation of soil properties this continuousvariability should be considered In addition in nature landis not flat (two dimensional) as it is represented on themap Hence prediction will be more reliable if a three-dimensionalmodel is used inwhich terrain attributes providea quantification of landform shape and connectivity thatdefine geomorphometry and water flow patterns [37 47ndash49]

The RMSE between predicted soil characteristics andthose observed in the field (Table 4) indicated good accuracyof prediction using the regression models for most soilcharacteristics when compared with previous studies [38 4050] Bayer et al found that two regression techniques hadsimilar capabilities to provide significant prediction models


Table 2 Regression coefficients (1198772) of the predicted soil attributes for different classes

Class Soil depth(cm)

Claycontent()

Siltcontent()

Sandcontent()

Organicmatter ()

Bulkdensity

(gm cmminus3)pH

Totalnitrogen()

Availablephosphorus

(ppm)

Stone atsurface()

Stone insoil ()

1 045 012 021 016 025 034 039 019 016 010 0132 045 050 026 053 032 021 018 009 007 037 0583 072 080 073 081 064 054 048 072 068 026 0324 076 054 042 057 085 065 037 078 071 069 0795 053 038 026 006 026 048 079 019 050 021 0546 034 052 024 052 018 026 014 026 018 007 0287 025 033 033 026 025 036 032 019 015 017 018

Table 3 Regression models to predict clay percent of the surface layer using terrain attributes and satellite images

Class no Constant Slope () As CTI Aspect Curvature ASTER March SPOT October 1198772

1 1769 007 000002 012 025 054 minus0004 minus0006 0122 4293 minus033

lowastlowast

00001lowast

minus075 031 108 minus008lowastlowast 004 050

3 3916 minus053lowast 00001 134 393

lowastlowast

minus263lowastlowast 004 minus020 080

4 2212 minus137lowast

minus00001 368 279 076 minus002 minus013 0545 minus282 minus006 minus00001 274 194 015 minus001 minus004 0386 5377 minus025

lowastlowast

00001lowastlowast

minus120 minus172lowast

minus020 minus005lowastlowast 0007 052


minus000003 175 140 162 minus001lowast

minus004lowast 033

lowastSignificant at the 005 probability level lowastlowastsignificant at the 001 probability levelAs upslope contributing area CTI compound topographic index

for soil organic carbon and iron oxides [51] In the presentstudy comparing this RMSE with that derived from spatialinterpolation using the IDW method indicated favorablycomparable accuracy of the prediction model RMSE of theprediction model was lower than that of the IDW in allcases Similar results were also observed when the number ofobservations used was reduced to 60 (Table 4) In additioncomparing the RMSE differences (between the predictionmodel and the IDW) using 180 with that of 60 observationsindicated higher increments in error term differences asthe number of observations used decreased (Table 4) Asa result the prediction model was highly preferred due tohigher accuracy than IDW especially when a limited numberof observations (60) were used This is because the soil-landscape model provides estimates based on the character-istics of the terrain between two observations and thereforeimproves accuracy This is not the case when interpolationis done with the assumption that closest observations arealso closer in their soil characteristics which is not alwaystrue The soil-landscape model uses the background topog-raphy and satellite data to characterize each observation andthen predict their soil attributes with consideration of thecharacteristics of these observations This is consistent withsoil genesis theories that are not considered in interpolationtechniques Previous research indicated that the use of differ-ent regression techniques enables the prediction of differenttopsoil parameters in a rapid and nondestructive mannerwhile avoiding the spatial accuracy problems associated withspatial interpolation [52]

The results indicated that soil attributes were predictedwith acceptable accuracy using SRTM 90m DEM and also

provide a visual representation of the spatial distribution ofsoil attributes (Figure 3) This indicated that soil attributescan be predicted with low resolution DEMs These conclu-sions were similar to those of Thompson et al [53] andWechsler [54] who concluded that high-resolutionDEMs arenot always necessary for soil-landscape modeling Chabrillatet al (2002) reported that although higher spatial resolutionprovided a purer image in more heterogeneous sites it didnot identify new features that a lower spatial resolution dataset would miss in homogeneous terrain [55]

When only terrain attributeswithout satellite imageswereused to build multiple regression models (Table 5) the 1198772declined for all classes compared with those obtained usingboth terrain and satellite data (Table 3) For example therewere large decreases in 1198772 in class 5 from 038 (Table 3)to 018 (Table 5) and in class 2 from 08 (Table 3) to 068(Table 5) This indicated that satellite images could improvethe prediction of soil attributes when used together withterrain attributes in building regressionmodels Gerighausenet al found that the use of multiannual image data enhancedthe prediction of soil parameters [56] The percent of agree-ment between the predicted and observed clay content usingsatellite images and terrain attributes was compared withthose derived using terrain attributes only The predictionaccuracy of clay content (percent of correctly predicted obser-vations) increased as a result of using satellite images withterrain attributes compared to using terrain attributes onlyFor example 50 of the observations showed agreement inpredicting clay contents within a difference of plusmn7 betweenobserved and predicted values when only terrain attributeswere used to build the model (Table 6) The above agreement


Table 4 Root mean square error between field observations and predicted soil attributes for multiple regression and IDW and using differentdensities of observations

Predicted soil attributes R 180 (A) I 180 (B) R 60 (C) I 60 (D) B-A D-CSoil depth (cm) 264 3264 337 4337 624 967Clay () 126 1664 1270 2051 404 781Silt () 73 1025 859 1464 295 605Sand () 94 1147 1024 1544 207 52Organic matter () 139 153 155 182 014 027Bulk density (g cmminus3) 018 025 024 036 007 012pH 038 054 046 083 016 037Total N () 029 046 009 034 017 025Available P (mg kgminus1) 1941 2035 1712 2122 094 41Surface stone cover () 122 1453 1415 2062 233 647Stone in the soil () 124 1571 1713 2638 331 925R 180 (A) root mean square error calculated from those predicted using 180 observations by multiple regression models I 180 (B) root mean square errorcalculated from spatial interpolation (IDW inverse distance weighted) using 180 observations R 60 (C) root mean square error calculated from thosepredicted using 60 observations by multiple regression models I 60 (D) root mean square error calculated from spatial interpolation (IDW inverse distanceweighted) using 60 observations B-A the difference in root mean square errors between those interpolated from 180 observations and those predicted using180 observations by multiple regression models and D-C the difference in root mean square errors between those interpolated from 180 observations andthose predicted using 180 observations by multiple regression models

Table 5 Regression models to predict clay percent of the surface layer using terrain attributes only (without satellite images)

Class No Constant Slope () As CTI Aspect Curvature 1198772

1 1506 006 000001 023 027 061 0102 4618 minus053

lowastlowast

00001lowast

minus129 088 107 0373 1931 minus028

lowast 000006 029 454lowastlowast

minus455lowastlowast 068


minus000008 268 317 237 0465 097 minus011 minus00002 178 125 110 0186 6635 minus040

lowastlowast

00002lowastlowast


minus141 0457 346 minus022

lowast

minus000004 185 125 252 024lowastSignificant at the 005 probability level lowastlowastsignificant at the 001 probability levelAs upslope contributing area CTI compound topographic index

Table 6 Agreement between observed and predicted values of soilattributes using terrain attributes with and without satellite images

Predicted clay contentusing terrain attributes andsatellite images

Predicted clay contentusing terrain attributes only(without satellite images)

Dlowast sect Dlowast sect

0 50 0 03 275 3 2255 500 5 357 625 7 50lowastMagnitude of difference between observed and predicted value sectpercent ofobservations where predicted values agreed with observed values

was improved to 625 when satellite images were usedwith the terrain attributes The results showed that satelliteimages improved the prediction of soil attributes when usedwith terrain attributes to build multiple regression modelsThis is because satellite images provide more informationabout the factors controlling soil variability in addition to thetopographic information provided by DEM-derived terrainattributes Chabrillat et al (2002) indicated that imaging

spectrometry aids in the detection and mapping of expansiveclays [55] The images taken at two different dates provideddirect and indirect information The direct information wasthrough the variation in the soil surface reflectance whichis related to soil variability (when the soil surface is bareduring March) The indirect information was through thereflectance of the different vegetation cover which reflectsthe variation in soil type through the effect on vegetationcover characteristics during October Many researchers havesuggested that remote sensing helps in providing accurateprediction of soil properties [27ndash29] Bartholomeus et al(2012) showed that although variation of soil propertieswithin vegetation classes is large the vegetation compositionis useful to estimate soil properties [57]

Since the spectral reflectance of both satellite images wassignificantly correlated with clay content (Table 3) we canconclude that two images taken during dry and rainy seasonscould be sufficient with terrain attributes to predict soilattributes usingmultiple linear regressionmodels During thedry period there was a correlation between soil properties(clay content) and spectral reflectance in the visible range[58] During the rainy season most of the area was coveredwith vegetation that reflects variation in soil properties such


(km)0 05 1 2

N

W E

S

Clay content ()0ndash1011ndash2526ndash3536ndash50gt50

Figure 3 Predicted clay content () of the surface soil using terrainattributes satellite images and 180 field observations

as clay content As a result soil properties could be indirectlypredicted by determining the spectral reflectance of thevegetation cover However the satellite image taken whenthe field was bare showed higher correlation with all soilattributes compared with the satellite image taken when thefield was covered by vegetation Hence the best time toacquire satellite images to improve the prediction of soilattributes would be during the dry period when most of thearea is bare Mulder et al [29] also reported similar findingsand successfully used airborne space borne and in situmeasurements using various instruments

Selecting Optimum Number of Field Observations to PredictSoil AttributesThe seven classes that were used in the earliermodeling (and using 180 observations) were amalgamated

Table 7 Percent of correctly predicted observations within plusmn50 cmof the observed soil depth RMSEs and MAEEs calculated usingdifferent field observation densities

Number ofobservationsused

Percent of correctlypredicted

observationswithin plusmn50 cm

RMSE (cm) MAEE (cm)

180 975 264 217150 95 297 255120 925 314 26590 875 317 26560 875 337 29440 725 409 31330 675 543 42525 65 575 46420 35 1260 882

first into five classes and several trials were done to predictsoil depth using 150 120 and 90 observations to buildregression models The analysis was repeated based on twoclasses to build regression models using only 60 40 3025 and 20 observations This was because when a limitednumber of observations were used (le60 observations) therewere insufficient observations if five or seven classes wereused (the number of observations for each class was toolow to build good regression models) The regression models(and 1198772) to predict soil depth at different densities of fieldobservations using terrain attributes and satellite imageswere used to get predicted soil-depth grids (Figure 4) Theproportion of predicted soil-depth of zero (no soil) increaseddramatically as the number of observations used reducedfrom 25 to 20 (Figure 4)

The optimum number of observations to produce anacceptable prediction of soil attributes was assessed usingpercent of observations where the soil depth was correctlypredicted within an acceptable range of agreement (usingRMSE and MAEE) The accuracy of predicting soil depthdecreased from 975 when 180 field observations wereused to 95 for 150 observations reached 650 using 25observations and then dropped drastically to 35 for 20observations (Table 7)Therewere gradual increases inRMSEandMAEE from the predictions using 180 observations (264and 217 cm resp) to the predictions using 25 observations(575 and 464 cm resp) then both increased sharply for 20observations (126 and 882 cm resp Table 7) Since there wasa high increment in RMSE andMAEE values at 20 comparedwith 25 observations it can be considered that 25 obser-vations (or one observation2 km2) were optimum to buildmultiple regression models in soil-landscape modeling forthis particular watershed The accuracy assessment methodsused for selecting the optimum number of observations topredict soil depth indicated that 25 observations were theminimum required for multiple linear regression models toproduce acceptable soil prediction The results also indicatedthat the prediction accuracy was the same when 60 or 90observations were used (875) However a close look at


Soil depth (cm)

1ndash5051ndash100101ndash150

180 observationsSeven classes

150 observationsFive classes



60 observationsTwo classes





le0

gt150

Figure 4 Predicted soil depth (cm) using terrain attributes satellite images and different density of field observations

Figure 4 revealed that the spatial distribution of the predictedsoil depth classes drastically changed when the number ofobservations used was lt60 Therefore it can be suggestedthat for this study area the optimumnumber of observationsto generate an acceptable prediction is somewhere between60 and 25 observations (1-2 observations2 km2) The userscan select between these observation densities to produce anoptimum results The approach indicated that soil-landscapemodeling could be used with a low number of observationsto produce accurate predictions of soil attributes with anacceptable representation of the spatial distribution of soilattributes over the landscape

4 Conclusions

The use of satellite image data together with terrain attributesimproved the prediction of soil attributes Two satelliteimages taken at different dates one in the dry season andthe other in the rainy season were sufficient to capture

the variation in soil reflectance and improve the predictionaccuracy If one image is available the best acquisition timeto facilitate soil-landscape modeling is during the dry seasonwhen most of the soil surface is bare Generally the resultsindicated that soil attributes were predicted with acceptableaccuracy using multiple linear regression models from freelyavailable DEM (90m resolution) and satellite images with aminimum of field work In addition the prediction modelwas highly preferred due to comparable accuracy with thespatial interpolation using IDW especially when a limitednumber of observations were used which is usually the casein data-scarce areas For this study the optimum number ofobservations to predict soil attributes using multiple regres-sion models and using terrain attributes and remote sensingdata was between 60 and 25 observations (approximately 1-2 observations2 km2)The produced predictions will be veryuseful to provide information for detailedmodeling activitiesespecially for a country like Ethiopia in which informationon the detailed spatial distribution of soil attributes is very


scarce Further investigations should be made for applyingthe model over larger areas with diverse soils and landscapefeatures to validate the results and to out-scale their applica-tion

Conflict of Interests

The authors declare no conflict of interests with any trade-mark or software mentioned in this paper

Acknowledgments

The authors would like to acknowledge the financial sup-port by the Austrian Development Agency and the CGIARResearch Program onWater Land and Ecosystems (CRP-5)

References

[1] J Bouma ldquoThe role of soil science in the land use negotiationprocessrdquo Soil Use and Management vol 17 no 1 pp 1ndash6 2001

[2] A R Mermut and H Eswaran ldquoSome major developments insoil science since the mid-1960srdquo Geoderma vol 100 no 3-4pp 403ndash426 2001

[3] M H Salehi M K Eghbal and H Khademi ldquoComparison ofsoil variability in a detailed and a reconnaissance soil map incentral Iranrdquo Geoderma vol 111 no 1-2 pp 45ndash56 2003

[4] C-W Ahn M F Baumgardner and L L Biehl ldquoDelineation ofsoil variability using geostatistics and fuzzy clustering analysesof hyperspectral datardquo Soil Science Society of America Journalvol 63 no 1 pp 142ndash150 1999

[5] N J McKenzie P E Gessler P J Ryan and D A OrsquoConnellldquoThe role of terrain analysis in soil mappingrdquo in Terrain Anal-ysis Principles and Applications J P Wilson and J C GallantEds Chapter 10 John Wiley and Sons New York NY USA2000

[6] A-X Zhu ldquoMapping soil landscape as spatial continua theneural network approachrdquoWater Resources Research vol 36 no3 pp 663ndash677 2000

[7] P A Burrough P FM VanGaans and RHootsmans ldquoContin-uous classification in soil survey spatial correlation confusionand boundariesrdquo Geoderma vol 77 no 2ndash4 pp 115ndash135 1997

[8] A-X Zhu ldquoA similarity model for representing soil spatialinformationrdquo Geoderma vol 77 no 2ndash4 pp 217ndash242 1997

[9] J Balkovic G Cemanova J Kollar M Kromka and K Harno-va ldquoMapping soils using the fuzzy approach and regression-krigingmdashcase study from the Povazsky Inovec MountainsSlovakiardquo Soil and Water Research vol 2 no 4 pp 123ndash1342007

[10] A B McBratney M L Mendonca Santos and B Minasny ldquoOndigital soil mappingrdquoGeoderma vol 117 no 1-2 pp 3ndash52 2003

[11] D M Browning and M C Duniway ldquoDigital soil mapping inthe absence of field training data a case study using terrainattributes and semi automated soil signature derivation todistinguish ecological potentialrdquo Applied and EnvironmentalSoil Science vol 42 pp 1904ndash1910 2011

[12] S Lamsal S Grunwald G L Bruland C M Bliss and NB Comerford ldquoRegional hybrid geospatial modeling of soilnitrate-nitrogen in the Santa Fe River Watershedrdquo Geodermavol 135 pp 233ndash247 2006

[13] P Scull J Franklin and O A Chadwick ldquoThe application ofclassification tree analysis to soil type prediction in a desertlandscaperdquo Ecological Modelling vol 181 no 1 pp 1ndash15 2005

[14] E Giasson R T Clarke A V Inda Jr G H Merten and C GTornquist ldquoDigital soil mapping using multiple logistic regres-sion on terrain parameters in southern Brazilrdquo Scientia Agricolavol 63 no 3 pp 262ndash268 2006

[15] A K Stum Random forests applied as a soil spatial predictivemodel in arid Utah [MS thesis] Utah State University 2010

[16] X Lui J Peterson Z Zhang and S Chandra Improving SoilSalinity Prediction with Resolution DEM Derived for LIDARData Center for GIS School of Geography and EnvironmentalScience Monash University Melbourne Australia 2006

[17] E Aksoy G Ozsoy and M S Dirim ldquoSoil mapping approachin GIS using landsat satellite imagery and dem datardquo AfricanJournal of Agricultural Research vol 4 no 11 pp 1295ndash13022009

[18] A Al-Shamiri and F M Ziadat ldquoSoil-landscape modeling andland suitability evaluation the case of rainwater harvesting ina dry rangeland environmentrdquo International Journal of AppliedEarth Observation vol 18 pp 157ndash164 2012

[19] D G RossiterDigital Soil Mapping Towards aMultiple-Use Soilinformation System Santa fe Bogota Colombia 2005

[20] J D Phillips ldquoSpatial structures and scale in categorical mapsrdquoGeographical and Environmental Modelling vol 6 no 1 pp 41ndash57 2002

[21] I DMoore P EGessler GANielsen andGA Peterson ldquoSoilattribute prediction using terrain analysisrdquo Soil Science Societyof America Journal vol 57 no 2 pp 443ndash452 1993

[22] P E Gessler I D Moore N J McKenzie and P J Ryan ldquoSoil-landscape modelling and spatial prediction of soil attributesrdquoInternational Journal of Geographical Information Systems vol9 no 4 pp 421ndash432 1995

[23] H-T Jiang F-F Xu Y Cai and D-Y Yang ldquoWeathering char-acteristics of sloping fields in the Three Gorges Reservoir AreaChinardquo Pedosphere vol 16 no 1 pp 50ndash55 2006

[24] X-Y Zhang Y-Y Sui X-D Zhang K Meng and S J HerbertldquoSpatial variability of nutrient properties in black soil of north-east china1 1 project supported by the National Basic ResearchProgram (973 Program) of China (No 2005CB121108) and theheilongjiang provincial natural science foundation of China(No C2004-25)rdquo Pedosphere vol 17 no 1 pp 19ndash29 2007

[25] I Esfandiarpoor Borujeni M H Salehi N Toomanian JMohammadi and R M Poch ldquoThe effect of survey density onthe results of geopedological approach in soil mapping a casestudy in the Borujen region Central Iranrdquo Catena vol 79 no1 pp 18ndash26 2009

[26] S L Kuriakose S Devkota D G Rossiter and V G JettenldquoPrediction of soil depth using environmental variables in ananthropogenic landscape a case study in the Western Ghats ofKerala Indiardquo Catena vol 79 no 1 pp 27ndash38 2009

[27] UMishra Predicting storage and dynamics of soil organic carbonat a regional scale [PhD thesis] Ohio State University 2009

[28] AMarchetti C Piccini S Santucci I Chiuchiarelli and R Fra-ncaviglia ldquoSimulation of soil types in Teramo province (CentralItaly) with terrain parameters and remote sensing datardquoCatenavol 85 no 3 pp 267ndash273 2011

[29] V LMulder S de BruinM E Schaepman andT RMayr ldquoTheuse of remote sensing in soil and terrain mappingmdasha reviewrdquoGeoderma vol 162 no 1-2 pp 1ndash19 2011


[30] Food and Agriculture Organization (FAO) Guidelines for SoilDescriptions FAO Rome Italy 2006

[31] P R Day ldquoHydrometer method of particle size analysisrdquo inMethods of Soil Analysis C A Black Ed Agronomy Part II No9 pp 562ndash563 American Society of Agronomy Madison WisUSA 1965

[32] G R Blake ldquoBulk densityrdquo inMethods of Soil Analysis Agron CA Black Ed Part I No 9 pp 374ndash399 American Society ofAgronomy Madison Wis USA 1965

[33] AWalkley and C A Black ldquoAn examination of different meth-ods for determining soil organic matter and the proposedmodification by the chromic acid titrationmethodrdquo Soil Sciencevol 37 pp 29ndash38 1934

[34] S Sahlemeden andB TayeProcedure for Soil and PlantAnalysisNational Soil Research Center Ethiopian Agricultural ResearchOrganization Addis Ababa Ethiopia 2000

[35] S R Olsen C V Cole F S Watanabe and L A Dean Estima-tion of Available Phosphorus in Soil by Extraction with SodiumBicarbonate vol 939 of Circular USDA 1954

[36] J Murphy and J P Riley ldquoA modified single solution methodfor the determination of phosphate in natural watersrdquoAnalyticaChimica Acta vol 27 pp 31ndash36 1962

[37] F M Ziadat ldquoAnalyzing digital terrain attributes to predictsoil attributes for a relatively large areardquo Soil Science Society ofAmerica Journal vol 69 no 5 pp 1590ndash1599 2005

[38] F M Ziadat ldquoPrediction of soil depth from digital terrain databy integrating statistical and visual approachesrdquoPedosphere vol20 no 3 pp 361ndash367 2010

[39] P Roudgarmi M Monavari J Feghhi J Nouri and N Kho-rasani ldquoEnvironmental impact prediction using remote sensingimagesrdquo Journal of Zhejiang University A vol 9 no 3 pp 381ndash390 2008

[40] M L Kunkel A N Flores T J Smith J P McNamara and S GBenner ldquoA simplified approach for estimating soil carbon andnitrogen stocks in semi-arid complex terrainrdquo Geoderma vol165 no 1 pp 1ndash11 2011

[41] K Sumfleth and R Duttmann ldquoPrediction of soil propertydistribution in paddy soil landscapes using terrain data andsatellite information as indicatorsrdquo Ecological Indicators vol 8no 5 pp 485ndash501 2008

[42] J Van de Wauw G Baert J Moeyersons et al ldquoSoil-landscaperelationships in the basalt-dominated highlands of TigrayEthiopiardquo Catena vol 75 no 1 pp 117ndash127 2008

[43] P E Gessler O A Chadwick F Chamran L Althouse andK Holmes ldquoModeling soil-landscape and ecosystem propertiesusing terrain attributesrdquo Soil Science Society of America Journalvol 64 no 6 pp 2046ndash2056 2000

[44] I V Florinsky R G Eilers G R Manning and L G FullerldquoPrediction of soil properties by digital terrain modellingrdquoEnvironmental Modelling and Software vol 17 no 3 pp 295ndash311 2002

[45] P Burrough ldquoSoil variability a late 20th century viewrdquo Soils andFertilizers vol 56 pp 529ndash562 1993

[46] B P Umali D P Oliver S Forrester et al ldquoThe effect of terrainand management on the spatial variability of soil properties inan apple orchardrdquo Catena vol 93 pp 38ndash48 2012

[47] E Dobos E Micheli M F Baumgardner L Biehl and THelt ldquoUse of combined digital elevation model and satelliteradiometric data for regional soil mappingrdquo Geoderma vol 97no 3-4 pp 367ndash391 2000

[48] Y A Pachepsky D J Timlin andW J Rawls ldquoSoil water reten-tion as related to topographic variablesrdquo Soil Science Society ofAmerica Journal vol 65 no 6 pp 1787ndash1795 2001

[49] Y Sulaeman andH Subagyo ldquoModeling soil landscape relation-shipsrdquo Jurnal Ilmu TanahDan Lingkungan vol 5 no 2 pp 1ndash142005

[50] S Zhang Y Huang C Shen H Ye and Y Du ldquoSpatial predic-tion of soil organic matter using terrain indices and categoricalvariables as auxiliary informationrdquo Geoderma vol 171-172 pp35ndash43 2012

[51] A Bayer M Bachmann A Muller and H Kaufmann ldquoA com-parison of feature-based MLR and PLS regression techniquesfor the prediction of three soil constituents in a degraded southafrican ecosystemrdquoApplied and Environmental Soil Science vol2012 Article ID 971252 20 pages 2012

[52] T Selige J Bohner and U Schmidhalter ldquoHigh resolution top-soil mapping using hyperspectral image and field data inmultivariate regression modeling proceduresrdquo Geoderma vol136 no 1-2 pp 235ndash244 2006

[53] J A Thompson J C Bell and C A Butler ldquoDigital elevationmodel resolution effects on terrain attribute calculation andquantitative soil-landscape modelingrdquo Geoderma vol 100 no1-2 pp 67ndash89 2001

[54] S Wechsler ldquoUncertainties associated with digital elevationmodels for hydrologic applications a reviewrdquo Hydrology andEarth System Sciences vol 3 pp 2343ndash2384 2006

[55] S Chabrillat A F H Goetz L Krosley and H W Olsen ldquoUseof hyperspectral images in the identification and mapping ofexpansive clay soils and the role of spatial resolutionrdquo RemoteSensing of Environment vol 82 no 2-3 pp 431ndash445 2002

[56] H Gerighausen G Menz and H Kaufmann ldquoSpatially explicitestimation of clay and organic carbon content in agriculturalsoils using multi-annual imaging spectroscopy datardquo Appliedand Environmental Soil Science vol 2012 Article ID 868090 23pages 2012

[57] H Bartholomeus G Schaepman-Strub D Blok R Sofronovand S Udaltsov ldquoResearch article spectral estimation of soilproperties in siberian tundra soils and relations with plantspecies compositionrdquo Applied and Environmental Soil Sciencevol 2012 Article ID 241535 13 pages 2012

[58] J A Thomasson R Sui M S Cox and A Al-Rajehy ldquoSoilreflectance sensing for determining soil properties in precisionagriculturerdquoTransactions of the American Society of AgriculturalEngineers vol 44 no 6 pp 1445ndash1453 2001

Submit your manuscripts athttpwwwhindawicom

Forestry ResearchInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Environmental and Public Health

Journal of



EcosystemsJournal of


MeteorologyAdvances in

EcologyInternational Journal of


Marine BiologyJournal of


Hindawi Publishing Corporationhttpwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2014

Advances in


Environmental Chemistry

Atmospheric SciencesInternational Journal of



Waste ManagementJournal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal of

Geophysics


Geological ResearchJournal of

EarthquakesJournal of


BiodiversityInternational Journal of


ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

OceanographyInternational Journal of


The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Journal of Computational Environmental SciencesHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


ClimatologyJournal of


fuzzy logic linear and multiple regression regression treesand neural networks have been used to develop soilmaps [1013] Ongoing research in digital soil mapping indicates thatquantitative prediction models are promising tools to pro-duce soilmapswith acceptable accuracy [14 15] Research hasprovided optimistic results and some researchers obtainedbetter results than traditional soil surveys [16ndash18] The useof satellite data to complement topographic informationimproves the mapping of natural resources [10 13 19]These investigations are based on the relationship betweenlandscape pattern and the differentiation of soil types andattributes [20] The spatial distribution of specific soil prop-erties has been predicted using soil-landscapemodelingThisincluded sand silt and organic matter contents A-horizonthickness solum depth extractable phosphorus and pH [2122] Using different terrain attributes the empirical modelsdeveloped in these studies explained 41ndash68 of the variationin soil properties [21 22]

GIS is used to quantify the relationships between soilsand topographic and remote sensing variables and to aid thedigital soil mapping efforts [2 3 10 23 24] Topographicparameters are derived from DEMs and used to predictthe distribution of soil characteristics [5 22 25 26] Theincreasing availability of high-resolution remote sensing dataprovides a new opportunity for predicting soil character-istics with acceptable accuracy [11] Many researchers havereported on the contribution of remote sensing data inproviding acceptable prediction of soil characteristics [27ndash29] Nevertheless there are many issues that warrant furtherresearch the conjunctive use of satellite data and topographicdata to improve the prediction of soil characteristics and thebest time of year to acquire satellite data The use of variousnumbers (densities) of observations and how that affectsaccuracy will help researchers decide on a reasonable numberof observations to maintain acceptable accuracy with anappropriate amount of field work The use of soil-landscapemodels to predict soil attributes is theoretically better thaninterpolation techniques because the former uses topogra-phy and satellite data as a background to guide predictionwhile interpolation relies solely on the spatial relationshipsbetween adjacent observations without considering the vari-ations in factors that drive soil differentiation Howeverthe merits of using soil-landscape modeling compared withinterpolation techniques need verification and quantificationespecially when a limited number of observations are avail-able The objective of this study is to investigate the utilityof a soil-landscape model using DEM and remote sensing topredict the spatial distribution of soil characteristics with anoptimum number of field observations

2 Materials and Methods

The study area is located in the Lake Tana basin in the Gum-ara-Maksegnit watershed of the Amhara region in Ethiopiawithin 12∘241015840ndash12∘311015840N and 37∘341015840ndash37∘371015840E It is centered45 km southwest of Gondar town and covers an area of56 km2 Altitude within the watershed ranges from 1923 to2851m above sea level with topography ranging from gentle

Observations for predictionObservations for validationStream network

SubwatershedsMain watersheds

N

(km)0 05 1 2

343500 345500 347500 349500

343500 345500 347500 349500

1383

500

1381

500

1379

500

1377

500

1375

500

1373

500

1371

500

1383

500

1381

500

1379

500

1377

500

1375

500

1373

500

1371

500

Figure 1 Distribution of field observations watershed boundarystream network and subwatershed divisions for Gumara-Maksegnitwatershed

to sharp steep slopes The mean annual rainfall is 1052mmThe mean minimum and maximum temperatures are 133and 285∘C respectivelyThe study area is characterized by anustic moisture regime [30] The natural resource base hasbeen depleted to a great extent due to soil erosion andimproper land use which resulted in reduced tree coverbiodiversity agriculture range and pasture productivity

21 Soil Survey and Field Observations The area was dividedinto square grids (500m times 500m) The total area of thewatershed and the heterogeneity of topography and vegeta-tion cover and consequently the soils were used to guidethe selection of this grid size Larger grid size (1 km) wouldresult in a limited number of observations and would limitthe representation of soil variability Soil samples were takenwithin each grid in the watershed (Figure 1) This was a com-promise between grid and free sampling techniques Whilethe surveyor was free to take the sample within the definedgrid to represent the area the distribution of sampling sites








2O) The






(km)0 05 1 2

Class1234567



N

W E

S








(1)



















RMSE = 1119899

119899

sum

119894=1

radic(119894minus 119910119894)

2

MAEE = 1119899

119899

sum

119894=1

10038161003816100381610038161003816119894minus 119910119894

10038161003816100381610038161003816

(3)












Claycontent()

Siltcontent()

Sandcontent()

Organicmatter ()

Bulkdensity

(gm cmminus3)pH

Totalnitrogen()

Availablephosphorus

(ppm)

Stone atsurface()

Stone insoil ()

1 045 012 021 016 025 034 039 019 016 010 0132 045 050 026 053 032 021 018 009 007 037 0583 072 080 073 081 064 054 048 072 068 026 0324 076 054 042 057 085 065 037 078 071 069 0795 053 038 026 006 026 048 079 019 050 021 0546 034 052 024 052 018 026 014 026 018 007 0287 025 033 033 026 025 036 032 019 015 017 018




lowastlowast

00001lowast


3 3916 minus053lowast 00001 134 393

lowastlowast




lowastlowast

00001lowastlowast





minus004lowast 033











1 1506 006 000001 023 027 061 0102 4618 minus053

lowastlowast

00001lowast

minus129 088 107 0373 1931 minus028





lowastlowast

00002lowastlowast


minus141 0457 346 minus022

lowast











(km)0 05 1 2

N

W E

S









RMSE (cm) MAEE (cm)

180 975 264 217150 95 297 255120 925 314 26590 875 317 26560 875 337 29440 725 409 31330 675 543 42525 65 575 46420 35 1260 882




Soil depth (cm)











le0

gt150



4 Conclusions







Acknowledgments


References
































































Journal of












Volume 2014

Advances in









Geophysics





















2O) The






(km)0 05 1 2

Class1234567



N

W E

S








(1)



















RMSE = 1119899

119899

sum

119894=1

radic(119894minus 119910119894)

2

MAEE = 1119899

119899

sum

119894=1

10038161003816100381610038161003816119894minus 119910119894

10038161003816100381610038161003816

(3)












Claycontent()

Siltcontent()

Sandcontent()

Organicmatter ()

Bulkdensity

(gm cmminus3)pH

Totalnitrogen()

Availablephosphorus

(ppm)

Stone atsurface()

Stone insoil ()

1 045 012 021 016 025 034 039 019 016 010 0132 045 050 026 053 032 021 018 009 007 037 0583 072 080 073 081 064 054 048 072 068 026 0324 076 054 042 057 085 065 037 078 071 069 0795 053 038 026 006 026 048 079 019 050 021 0546 034 052 024 052 018 026 014 026 018 007 0287 025 033 033 026 025 036 032 019 015 017 018




lowastlowast

00001lowast


3 3916 minus053lowast 00001 134 393

lowastlowast




lowastlowast

00001lowastlowast





minus004lowast 033











1 1506 006 000001 023 027 061 0102 4618 minus053

lowastlowast

00001lowast

minus129 088 107 0373 1931 minus028





lowastlowast

00002lowastlowast


minus141 0457 346 minus022

lowast











(km)0 05 1 2

N

W E

S









RMSE (cm) MAEE (cm)

180 975 264 217150 95 297 255120 925 314 26590 875 317 26560 875 337 29440 725 409 31330 675 543 42525 65 575 46420 35 1260 882




Soil depth (cm)











le0

gt150



4 Conclusions







Acknowledgments


References
































































Journal of












Volume 2014

Advances in









Geophysics


















(1)



















RMSE = 1119899

119899

sum

119894=1

radic(119894minus 119910119894)

2

MAEE = 1119899

119899

sum

119894=1

10038161003816100381610038161003816119894minus 119910119894

10038161003816100381610038161003816

(3)












Claycontent()

Siltcontent()

Sandcontent()

Organicmatter ()

Bulkdensity

(gm cmminus3)pH

Totalnitrogen()

Availablephosphorus

(ppm)

Stone atsurface()

Stone insoil ()

1 045 012 021 016 025 034 039 019 016 010 0132 045 050 026 053 032 021 018 009 007 037 0583 072 080 073 081 064 054 048 072 068 026 0324 076 054 042 057 085 065 037 078 071 069 0795 053 038 026 006 026 048 079 019 050 021 0546 034 052 024 052 018 026 014 026 018 007 0287 025 033 033 026 025 036 032 019 015 017 018




lowastlowast

00001lowast


3 3916 minus053lowast 00001 134 393

lowastlowast




lowastlowast

00001lowastlowast





minus004lowast 033











1 1506 006 000001 023 027 061 0102 4618 minus053

lowastlowast

00001lowast

minus129 088 107 0373 1931 minus028





lowastlowast

00002lowastlowast


minus141 0457 346 minus022

lowast











(km)0 05 1 2

N

W E

S









RMSE (cm) MAEE (cm)

180 975 264 217150 95 297 255120 925 314 26590 875 317 26560 875 337 29440 725 409 31330 675 543 42525 65 575 46420 35 1260 882




Soil depth (cm)











le0

gt150



4 Conclusions







Acknowledgments


References
































































Journal of












Volume 2014

Advances in









Geophysics


















RMSE = 1119899

119899

sum

119894=1

radic(119894minus 119910119894)

2

MAEE = 1119899

119899

sum

119894=1

10038161003816100381610038161003816119894minus 119910119894

10038161003816100381610038161003816

(3)












Claycontent()

Siltcontent()

Sandcontent()

Organicmatter ()

Bulkdensity

(gm cmminus3)pH

Totalnitrogen()

Availablephosphorus

(ppm)

Stone atsurface()

Stone insoil ()

1 045 012 021 016 025 034 039 019 016 010 0132 045 050 026 053 032 021 018 009 007 037 0583 072 080 073 081 064 054 048 072 068 026 0324 076 054 042 057 085 065 037 078 071 069 0795 053 038 026 006 026 048 079 019 050 021 0546 034 052 024 052 018 026 014 026 018 007 0287 025 033 033 026 025 036 032 019 015 017 018




lowastlowast

00001lowast


3 3916 minus053lowast 00001 134 393

lowastlowast




lowastlowast

00001lowastlowast





minus004lowast 033











1 1506 006 000001 023 027 061 0102 4618 minus053

lowastlowast

00001lowast

minus129 088 107 0373 1931 minus028





lowastlowast

00002lowastlowast


minus141 0457 346 minus022

lowast











(km)0 05 1 2

N

W E

S









RMSE (cm) MAEE (cm)

180 975 264 217150 95 297 255120 925 314 26590 875 317 26560 875 337 29440 725 409 31330 675 543 42525 65 575 46420 35 1260 882




Soil depth (cm)











le0

gt150



4 Conclusions







Acknowledgments


References
































































Journal of












Volume 2014

Advances in









Geophysics

















Claycontent()

Siltcontent()

Sandcontent()

Organicmatter ()

Bulkdensity

(gm cmminus3)pH

Totalnitrogen()

Availablephosphorus

(ppm)

Stone atsurface()

Stone insoil ()

1 045 012 021 016 025 034 039 019 016 010 0132 045 050 026 053 032 021 018 009 007 037 0583 072 080 073 081 064 054 048 072 068 026 0324 076 054 042 057 085 065 037 078 071 069 0795 053 038 026 006 026 048 079 019 050 021 0546 034 052 024 052 018 026 014 026 018 007 0287 025 033 033 026 025 036 032 019 015 017 018




lowastlowast

00001lowast


3 3916 minus053lowast 00001 134 393

lowastlowast




lowastlowast

00001lowastlowast





minus004lowast 033











1 1506 006 000001 023 027 061 0102 4618 minus053

lowastlowast

00001lowast

minus129 088 107 0373 1931 minus028





lowastlowast

00002lowastlowast


minus141 0457 346 minus022

lowast











(km)0 05 1 2

N

W E

S









RMSE (cm) MAEE (cm)

180 975 264 217150 95 297 255120 925 314 26590 875 317 26560 875 337 29440 725 409 31330 675 543 42525 65 575 46420 35 1260 882




Soil depth (cm)











le0

gt150



4 Conclusions







Acknowledgments


References
































































Journal of












Volume 2014

Advances in









Geophysics



















1 1506 006 000001 023 027 061 0102 4618 minus053

lowastlowast

00001lowast

minus129 088 107 0373 1931 minus028





lowastlowast

00002lowastlowast


minus141 0457 346 minus022

lowast











(km)0 05 1 2

N

W E

S









RMSE (cm) MAEE (cm)

180 975 264 217150 95 297 255120 925 314 26590 875 317 26560 875 337 29440 725 409 31330 675 543 42525 65 575 46420 35 1260 882




Soil depth (cm)











le0

gt150



4 Conclusions







Acknowledgments


References
































































Journal of












Volume 2014

Advances in









Geophysics















(km)0 05 1 2

N

W E

S









RMSE (cm) MAEE (cm)

180 975 264 217150 95 297 255120 925 314 26590 875 317 26560 875 337 29440 725 409 31330 675 543 42525 65 575 46420 35 1260 882




Soil depth (cm)











le0

gt150



4 Conclusions







Acknowledgments


References
































































Journal of












Volume 2014

Advances in









Geophysics















Soil depth (cm)











le0

gt150



4 Conclusions







Acknowledgments


References
































































Journal of












Volume 2014

Advances in









Geophysics


















Acknowledgments


References
































































Journal of












Volume 2014

Advances in









Geophysics
















































Journal of












Volume 2014

Advances in









Geophysics


















Journal of












Volume 2014

Advances in









Geophysics














Documents

Research Article Soil-Landscape Modeling and Remote ...downloads.hindawi.com/journals/aess/2013/798094.pdf · of a soil-landscape model using DEM and remote sensing to predict the