object-based classification of high resolution sar images ......soil properties, and other stresses...

AbstractDelineating management zones is important in agriculturefor implementing site-specific practices. We delineatedwithin-field homogeneous zones over a corn and a wheatfield using high spatial resolution multi-temporal airborneC-band synthetic aperture radar (SAR) imagery with anobject-based fuzzy k-means classification approach. Imageobjects were generated by a segmentation procedure imple-mented in eCognition® software, and were classified as basicprocessing units using SAR data. Results were evaluatedusing analysis of variance and variance reduction of soilelectrical conductivity (EC), leaf area index (LAI), and cropyield. The object-based approach provided better resultsthan a pixel-based approach. The variance reduction in LAI,and soil EC varied with SAR acquisition time and incidenceangle. Although the variance reduction of yield was not assignificant as that of LAI and EC, average yield among thedelineated zones were different in most cases. The SAR dataclassification produced interpretable patterns of soil andcrop spatial variability, which can be used to infer within-field management zones.

IntroductionThere is a growing interest in delineating managementzones for implementing site-specific practices in agriculture.Management zones are delineated by classifying the within-field spatial variability of yield limiting factors (Doerge,1999). Different inputs, such as nutrient, seed rate, water,tillage, and soil management, can then be applied accord-ingly for optimal profitability and environmental sustain-ability. To avoid the uncertainty related with interpolationprocedures and achieve robust zone delineation, data shouldbe densely sampled within a field. This is an expensivepractice using conventional survey methods. High-resolutionsatellite remote sensing potentially provides a cost-effectiveand non-invasive way to obtain crop field informationcontinuously over space and frequently through a season.

As reviewed by Moran et al. (1997), the variabilityof many field descriptors could be mapped with remotesensing techniques by measuring short-wave (0.4 to 2.6 mm)reflected radiance, long-wave (3 to 16 mm) emitted radiance,

Object-based Classification of High ResolutionSAR Images for Within Field Homogeneous

Zone DelineationJiangui Liu, Elizabeth Pattey, and Michel C. Nolin

or synthetic aperture radar (SAR; 0.9 to 25 cm) backscatter.While single remote sensing observation at an optimal timeis useful in mapping seasonally-stable field descriptors,e.g., soil-based management units, multi-temporal observa-tions are especially important to capture unexpected envi-ronmental impacts, map temporal variation, and obtainseasonal profiles of crop descriptors that could be used asinputs to crop models (Wiegand et al., 1986). Furthermore,since the crop canopy integrates the effects of the weather,soil properties, and other stresses (e.g., disease or nutrientdeficiencies) (Wiegand and Richardson, 1984), stable anddynamic soil properties could also be mapped by multi-temporal observations when the crop is present. Thus,classification of high spatial resolution remote sensingimages could provide useful information for within fieldmanagement zone delineation. The all-weather acquisi-tion capability of radar makes it advantageous over opticalremote sensing, since obtaining multi-temporal optical datacan be challenging due to cloud cover. The availability ofdata from RADARSAT-2, with its multi-polarization or fullypolarimetric modes, and higher spatial resolution (Fox et al.,2004), could boost the application of SAR data in agriculture(McNairn and Brisco, 2004). The objective of this study wasto exploit the variability in multi-temporal, multi-polarizedhigh spatial resolution SAR imagery for within field homo-geneous zone delineation for site-specific agriculture using aclassification approach.

Classification of land-cover using SAR data was mainlybased on the sensitivity of radar backscatter to target struc-tural information (Dobson et al., 1995). A few general land-cover types can thus be differentiated. A widely used methodfor unsupervised classification of polarimetric SAR datathrough target decomposition was devised by Cloude andPottier (1997), in which pixels were divided into eightfeasible zones in the entropy (H) and anisotropy (a) planeaccording to different scattering mechanisms. By usingthis method for an initial unsupervised classification toderive training sets, Lee et al. (1999 and 2004) developed asupervised classifier based on a complex Wishart distribution.These kinds of methods require spatial averaging to obtain arobust estimation of a coherency matrix; hence, they tend todegrade the spatial resolution. Furthermore, it could bedifficult to segment an agricultural field according to the

PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING Sep t embe r 2008 1159

Jiangui Liu and Elizabeth Pattey are with Agricultureand Agri-Food Canada, 960 Carling Avenue, Ottawa, ON,Canada, K1A 0C6 ([email protected]).

Michel C. Nolin is with Agriculture and Agri-Food Canada,140#-979 De Bourgogne Avenue, Quebec, QC, Canada,G1W 2L4.

Photogrammetric Engineering & Remote Sensing Vol. 74, No. 9, September 2008, pp. 1159–1168.

0099-1112/08/7409–1159/$3.00/0© 2008 American Society for Photogrammetry

and Remote Sensing

06-098.qxd 8/9/08 12:33 AM 159

1160 Sep t embe r 2008 PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING

scattering mechanisms, since a single mechanism usuallydominates within a field. In this study the three linearlypolarized backscatter coefficients, HH, HV, and VV, weresynthesized and used for classification.

An intrinsic shortcoming of a conventional pixel-based classifier is that it is based on global behavior of thefeature set without considering the spatial information.It ensures optimal segmentation in the feature domain,but not necessarily in the spatial domain. This leads tofragmentation within field spatial patterns and fractalboundaries that cannot be managed by farmers. Moreover,in order to derive meaningful spatial patterns, coherentspeckle noise in SAR data has to be reduced by specklefilters, which reduces the effective spatial resolution.Baring these factors in mind, an object based classificationscheme was proposed in this study to classify the synthe-sized polarization data. The field was first segmented intosmall homogeneous areas, referred to as image objects orimage primitives, using eCognition® software (Baatz et al.,2002). Both spectral and spatial information were utilizedat this preliminary stage. Average values of SAR data inthese image objects were then extracted and classifiedusing an unsupervised Fuzzy k-means algorithm imple-mented in FuzMe 3.0 (Minasny and McBratney, 2002). Theclassification was evaluated using three field descriptors:(a) soil electrical conductivity (EC), which was related to afew important soil properties, (b) green LAI, an importantindicator for crop development status, and (c) the cropyield. This provided an assessment of SAR data classifica-tion for potential management zone delineation.

MaterialsThe Study SiteThe study site was located in the former Greenbelt Farm inOttawa, Ontario, Canada (45°18� N, 75°45� W). Two neigh-boring fields, one planted with corn and another with wheatin the 2001 growing season were considered in the study.The field planted with corn (72 ha) is characterized with ahigh pedodiversity level, with six soil series combinationsand moderately to poorly drained conditions. The fieldplanted with wheat (32 ha) is characterized with a lowerpedodiversity level, with two soil series combinations andmainly a poorly drained condition (Marshall et al., 1977).

Three intensive field campaigns (IFC) were carried out tomonitor crop growth conditions. Remote sensing data wereacquired on 13 June, 26 June, and 19 July 2001 for the threecampaigns. Airborne C-Band (5.3 GHz) polarimetric SAR datawere acquired with the Environment Canada Convair-580SAR system (CV-580). Airborne hyperspectral data wereacquired with the CRESTech Compact Airborne Spectro-graphic Imager (CASI), with 2 m resolution and 72 bands inthe visible-near infrared range. Crop biophysical descriptors,including phenological stage, height, biomass, leaf areaindex (LAI), and crop fraction, were also collected during thethree campaigns. During the first campaign (IFC1), wheat wasin the tillering phase with a few areas at the stem elongationstage. The crop fraction of cover ranged from 40 percent to80 percent in response to different soil conditions andnitrogen treatments. Three to six leaves were expanded forthe corn plants, and the crop fraction was approximately10 percent. During the second campaign (IFC2), wheat wasin between the stem elongation and heading phases. Forcorn, six to nine leaves were expanded and the fraction ofcoverage ranged between 30 percent and 70 percent. Theheight of the corn canopy exceeded 1 m in the productiveareas. During the third field campaign (IFC3), portions of thewheat field were already senescing, while the corn reached

full coverage with eight to thirteen expanded leaves andemerging tassels in the productive areas.

CV-580 SAR DataCV-580 polarimetric SAR data were acquired at two nominalincidence angles, about 35° and 55° at field center. Radio-metric calibration was accomplished at the Canada Centerfor Remote Sensing (CCRS). A radiometric accuracy of 0.8 dBand phase accuracy of 10° were achieved. Geocoded prod-ucts were also processed at CCRS, with a 4 m pixel size.

Speckle noise was reduced using Kuan filter with aminimum window size, i.e., 3 pixels by 3 pixels. The threelinearly polarized backscatter coefficients HH, VV, and HV

were synthesized using the software package PolarimetricWorkstation (PWS) (Touzi and Charbonneau, 2004). Dueto difficulties during the acquisition, SAR data were notacquired on 26 June with the 35° incidence angle. Conse-quently, five radar data sets were available for analysis:three acquisitions with a 55° incidence angle for each of thefield campaigns, and two acquisitions with a 35° incidenceangle for the first and the third field campaigns.

Biophysical MeasurementsMaps of soil EC, yield, and multi-temporal green LAI weregenerated for the two fields. The spatial variability of soilEC is dominated by static factors and exhibits relativelyconstant patterns, while its magnitude is affected by thedynamic soil properties (Corwin and Lesch, 2005). There-fore, spatial and temporal measurements of soil EC arepowerful tools in precision agriculture studies (Corwin andLesch, 2003; Corwin et al., 2003). Soil EC was measured inNovember 2002 using the VERIS-3100 at 0 to 30 cm (EC30)and 0 to 100 cm (EC100) depths, with a sampling rate ofapproximately 150 points ha�1. Soil EC data measured inthese two fields was correlated with soil texture, drainageproperty, and some soil quality indicators (Perron et al.,2003). Maps of multi-temporal green LAI were generatedfrom CASI hyperspectral data for the three field campaignsusing vegetation index MTVI2 (Haboudane et al., 2004).The LAI maps revealed the within-field crop growth condi-tion variability (Haboudane et al., 2004; Liu et al., 2005).Yield in 2001 was measured when the wheat was harvestedon 22 August and the corn on 10 October. Data from a yieldmonitor (GreenStar Combine Yield Systems, Deere and Co.,Moline, Illinois) and a global positioning system (GPS) wererecorded with a 1 Hz rate. The final yield was processed asdry grain mass in kg/m�2. The measured point data of soilEC and crop yield were interpolated to a 4 by 4 m rasterformat. The estimated green LAI was mutually co-registeredwith the SAR data for further analysis.

MethodsImage SegmentationFigure 1 shows an object-based unsupervised fuzzy k-meansclassification procedure for within field homogeneous zonedelineation. The fields were first segmented into imageobjects or image primitives using eCognition® software(Baatz et al., 2002). Because it is strongly correlated with afew soil properties, and its spatial patterns are relativelystable over time (Corwin et al., 2003; Perron et al., 2003),soil EC was used for the first level segmentation to obtainmeaningful image objects. In the second level, the obtainedimage objects were further segmented in three consecutivesteps using the linear polarization data acquired for thethree campaigns.

Segmentation in eCognition® is a bottom up region-merging procedure. As pairs of image objects merge, the

06-098.qxd 8/9/08 12:33 AM Page 1160


Figure 1. The study procedure. (LAI: leaf area indexderived from the Compact Airborne Spectral Imager;PWS: Polarimetry Work Station).

resulted object becomes more inhomogeneous, and theheterogeneity increases. Starting from single pixels, smallerregions are iteratively merged into bigger ones, until theincrease of heterogeneity of any mergence does not exceed agiven threshold. Heterogeneity of an image object defined ineCognition® is split into spectral heterogeneity, which isrelated with the variance of data within the object, andspatial heterogeneity, which is related with the shape of theobject. The spatial heterogeneity consists of “smoothness”and “compactness.” Weighting factors control the relativecontributions between spectral and spatial heterogeneity,and between smoothness and compactness to the overallheterogeneity (Benz et al., 2004). Thus, both the spectral andspatial information contained in the feature sets are utilizedin the segmentation stage. The heterogeneity threshold isreferred to as the scale factor, and is an important parameterfor segmentation in eCognition®. A bigger scale factor resultsin larger objects. Determination of an optimal scale factor isaffected by scene characteristics, data dynamic range, andthe objectives. Preprocessing such as calibration, scaling andmathematical conversion (e.g., conversion of digital numbersto dB) that changes the data dynamic range may influencethe value of an optimal scale factor.

In this study, the weighting factors for spectral andshape heterogeneity was set to 0.7 and 0.3, respectively, andthat for compactness and smoothness were set to 0.8 and0.2, respectively. These parameters were approximate to thedefault settings. The scale factors for the first and the second

level segmentations were set to 10 and 4, respectively,so that the image objects are not too big to keep the impor-tant scene details. The averages of the backscattering inthe HH, VV, and HV polarizations were calculated for theobjects derived from the above segmentation procedure, andwere exported from eCognition® to an ASCII format file forclassification.

Fuzzy k-means ClassificationFuzzy k-means classifier implemented in FuzMe (Minasny andMcBratney, 2002) was used to classify the derived imageobjects using SAR data. Mahalanobis distance was chosen asthe measure of difference between objects, since it standard-izes data and accounts for the correlation between the fea-tures. The fuzzy exponential was set to 1.25 to allow adequatefuzziness while achieving convergence of classification. Imageobjects were classified into 2 to 10 classes. The fuzzinessPerformance Index (FPI) and Modified Partition Entropy (MPE)were reported as functions of the number of classes. Theoptimal number of classes into which the image objects can begrouped was determined by the minimum FPI and MPE values(Roubens, 1982). For comparison purpose, classification wasalso applied on a pixel bases.

Evaluation of the ClassificationTo evaluate the potential of classification as a tool for withinfield homogeneous zone delineation, a method proposedby Fridgen et al. (2000) and used by Liu et al. (2005) wasadopted here. This method is effective to evaluate the delin-eation of field descriptors with non-categorical variability.Basically, for the classification of an agricultural field anda selected field descriptor, the sum of within class variancesof this descriptor can be calculated. Compared with thetotal within field variance, the percent variance reduction isan evaluation criterion. The more variance reduction, thebetter this descriptor is delineated by the classification. Fourspatially continuous field descriptors, green LAI, EC30, andEC100, and the final crop yield, were selected as evaluationvariables to assess the classification. Another measure ofevaluation is the Analysis of Variance (ANOVA). It assesseswhether the average field descriptors between different classesare significantly different. This was done using Statisticasoftware (StatSoft, Inc., 2005).

Results and DiscussionStatistics of Field Descriptors and SAR Backscattering CoefficientsAverage, standard deviation (STDEV), and coefficient ofvariation (CV) for the selected field descriptors were reportedin Table 1. Here CV refers to the ratio between the standarddeviation and the average value. CV of soil EC in the corn-field was much higher than that in the wheat field. This was

TABLE 1. STATISTICS OF FIELD DESCRIPTORS. LAI WAS ESTIMATED FROM CASI HYPERSPECTRAL DATA;SOIL ELECTRICAL CONDUCTIVITY WAS MEASURED USING VERIS-3100 TECHNOLOGY AT 0 TO 30 CM

(EC30) AND 0 TO 100 CM (EC100); YIELD WAS MEASURED WITH YIELD MONITOR; STDEV REFERS

TO STANDARD DEVIATION, AND CV REFERS TO COEFFICIENT OF VARIATION

EC30 EC100 Yield LAI (m2 m�2)(mS m�2) (mS m�1) (kg m�2) 13 June 26 June 19 July

Mean 7.12 15.63 0.70 0.32 1.39 3.90Corn Stdev 4.69 8.67 0.13 0.09 0.78 0.79

CV (%) 65.9 55.5 18.7 26.5 56.1 20.3

Mean 18.18 24.86 0.36 1.11 1.86 1.01Wheat Stdev 4.03 4.41 0.06 0.66 0.60 0.26

CV (%) 22.2 17.8 15.5 59.2 32.4 25.9

06-098.qxd 8/9/08 12:33 AM Page 1161


in conformity with the fact that the cornfield had a higherpedodiversity than the wheat field (Marshall et al., 1979).From IFC1 to IFC3, average LAI of the corn increased steadily,while CV reached maximum at IFC2. Crop growth conditionsin the wheat field were quite different due to the differencesin soil conditions and nitrogen treatments (Liu et al., 2005).Average LAI reached maximum at IFC2, whereas the CV

value was largest at IFC1, and dropped significantly at thelast campaign. CV of the yield was smaller than the otherdescriptors in these two fields.

Field averages and standard deviations of HH, VV, and HV

polarizations and the green LAI for the three campaigns wereshown in Figure 2. Relative to the wavelength of C-Bandradar, corn and wheat are typical of broad and narrow leafcrops, respectively. The temporal variation in backscatterdemonstrated the differences in radar responses to these twokinds of crops. For the corn, radar backscatter was the lowestwhile its variance was the most significant during the firstcampaign. The variance in backscattering was mainly due tothe variability of soil surface conditions, e.g., moisture androughness. As the crop developed in the following twocampaigns, the backscattering approached a saturation levelwhile the variance decreased significantly. This is a typicalphenomenon for wide leaf crops (Ferrazzoli et al., 1997).Backscatter from the crop became the dominant factor in theradar response, whereas soil surface information was increas-ingly attenuated. For instance, the standard deviations ofbackscattered HH polarization over the corn canopy at 55°incidence angle were 2.1, 1.4, and 1.2 dB, with within fieldaverages of �14.5, �9.6, and �10.8 dB for the three cam-paigns, respectively. Similar trends can be observed for the

other two polarizations and at both incidence angles. Thebackscatter recorded at a 35° incidence angle was alwaysstronger than that at a 55° incidence angle.

For the wheat, the magnitude and the variance ofradar backscatter did not change as much as that observedfor the corn canopy. For instance, the average HH polarizedbackscatter at a 55° incidence angle were �12.9, �11.2,and �11.0 dB, and the standard deviations were 1.2,1.2, and 1.1 dB for the three campaigns, respectively.A variation of angular dependency of VV polarization wasobserved in the wheat field at IFC3. A stronger VV backscat-ter was observed at a 55° incident angle than at a 35°incidence angle, which was consistent with the observa-tion by Mattia et al. (2003). This may be explained by agreater interaction of the incident microwaves with thevertical crop structures of the wheat, e.g., stem and head(Mattia et al., 2003). For the 55° SAR acquisitions, HH andVV polarizations had the greatest difference in the wheatfield at IFC2, when green LAI was the greatest. Whereas inthe cornfield, the difference between HH and VV polariza-tions was almost unchanged when green LAI increasedfrom the lowest level at IFC1 to the highest level at IFC3.

Image SegmentationThe first level segmentation using soil EC produced 264image objects in the cornfield and 86 objects in the wheatfield. In the second level, the sequential segmentation ofthe linear polarization data generated 841 image objectsin the cornfield, and 248 in the wheat field. These imageobjects were supposed to have captured the field variabilityrevealed by the spectral and spatial information contained in

Figure 2. Average and standard deviation of multi-temporal SAR backscatteringcoefficients and green LAI for corn (a) and (b), and wheat (c) and (d). There was no35° incidence angle SAR data acquisition on 26 June 2001, i.e., Calendar day 177.(Ave.: Average; STDEV: Standard deviation).

06-098.qxd 8/9/08 12:33 AM Page 1162


Figure 3. Segmentation results of (a) the corn and (b) the wheat fields. Theupper and the lower rows show the polygons of the first and the second levelsegmentation, respectively.

the soil EC data and the multi-temporal SAR data, and willbe further grouped in the next step. Figure 3 shows thepolygons derived for the two segmentation levels overlainon EC100 map.

The Number of ClassesFPI and MPE as functions of the number of classes wereplotted in Figure 4 for the 55° incidence angle data. FPI andMPE decreased significantly when the fields were classifiedinto three classes. When the number of classes increased, adecreasing trend could be observed without a well-definedminimum value. Similar results were observed for the 35°acquisitions (data not shown). Therefore, both of the fieldswere classified into three classes for all the three campaigns.However, this is determined on the synthesized linearpolarization data only. It does not mean that three classesare best for all the field descriptors. The effectiveness fordelineating those field descriptors requires independentevaluation.

Variance ReductionPercent variance reduction of green LAI, yield, and soilEC due to the classifications was calculated to assess theperformance of the delineations. The results were reported inTable 2. To serve as a comparison, the results from the pixel-based approach were also included. For classification of thecorn at IFC1 with the 55° SAR data, although a higher variancereduction of green LAI was achieved with the pixel-based

approach (21.9 percent) than the object-based approach(10.5 percent), this was probably related to the significantlylower variability in LAI (standard deviation 0.09). However,generally a greater variance reduction was achieved with theobject-based approach than with the pixel-based approach,which supported the use of the object-based method in thisstudy. The following sections only present the results fromthe object-based approach.

In the cornfield, variance reduction of soil EC andgreen LAI due to classification decreased considerably fromIFC1 to IFC3. The fully developed corn canopy during IFC3caused two effects. First, direct backscattering from the soilwas reduced by the significant interaction between theincident microwaves and the corn canopy. Second, radarbackscatter from the mature canopy saturated so that thesignal was not sensitive to the variability of crop biomass.At IFC1, classification of SAR data with the steeper inci-dence angle (35°) was more effective than the shallowerincidence angle (55°) with respect to the reduction in thevariance of soil EC. The percent variance reductions of EC30and EC100 were 22.9 percent and 25.1 percent for the 35°acquisition, and 18.2 percent and 20.6 percent for the 55°acquisition, respectively.

In the wheat field, the variance reduction of soil EC dueto classification did not show a decreasing trend from IFC1to IFC3. The canopy development was likely responding tothe soil variability. In addition, particularly as the wheatcanopy senesces, the microwaves can penetrate the canopy

06-098.qxd 8/9/08 12:33 AM Page 1163


TABLE 2. VARIANCE REDUCTION OF THE SELECTED FIELD DESCRIPTORS DUE TO CLASSIFICATION OF SAR DATA. RESULTS

OF OBJECT-BASED AND PIXEL-BASED CLASSIFICATIONS WERE REPORTED AS A COMPARISON; THE VALUES IN THE TABLE

REPRESENTED PERCENT OF VARIANCE REDUCTION RELATIVE TO THE TOTAL WITHIN FIELD VARIANCE

Object-based Pixel-based

IFC1 IFC1 IFC2 IFC3 IFC3 IFC1 IFC1 IFC2 IFC3 IFC355° 35° 55° 55° 35° 55° 35° 55° 55° 35°

EC30 (mS m�1) 18.2 22.9 13.2 7.1 1.1 17.2 19.4 9.6 4.2 2.6EC100(mSm�1) 20.6 25.1 17.4 7.3 1.3 20.9 22.3 15.1 3.7 2.1

Corn LAI (m2 m�2) 10.5 15.2 15.1 3.3 4.8 21.9 10.9 17.3 1.3 2.3Yield (kgm�2) 4.6 6.6 6.9 1.6 1.4 3.3 2.9 2.0 0.1 0.1

EC30 (mS m�1) 11.1 15.4 21.8 16.2 10.0 10.3 10.6 13.3 11.6 5.5EC100(mSm�1) 7.9 12.0 15.3 12.7 7.0 8.1 8.2 9.0 8.7 4.0

Wheat LAI (m2 m�2) 19.3 23.7 17.8 8.5 1.6 17.1 17.8 10.4 1.4 0.1Yield (kgm�2) 5.9 7.2 8.4 7.0 6.3 4.1 4.1 3.8 1.8 0.1

Figure 4. Fuzziness performance index (FPI)and modified partition entropy (MPE) asfunctions of the number of classes in (a)the corn, and (b) the wheat fields, for the55° incidence angle SAR acquisitions.

to the underlying soil. Variance reduction of LAI at the firsttwo IFCs was significantly higher than that at IFC3. At IFC3,variance reduction of green LAI showed a clear dependenceon incidence angle, which was contrary to the observationsin the cornfield. This demonstrated that, at later develop-ment stages, SAR backscatter from a wheat canopy was moresensitive to incidence angle than that from a corn canopy.

The variance reduction of yield in both fields was limited,generally less than 10 percent. This was probably because thatyield limiting factors were not well detected by the radar atthe date of acquisition. This was especially the case for thecornfield, since the last remote sensing observation was at

least two months before harvesting, leaving a large intervalwithin which the crop growth conditions were not monitoredby remote sensing.

Evaluation of the ClassificationsThe potential of classification as a tool for homogeneouszone delineation was evaluated using the selected fielddescriptors. The classification maps of the two fields wereshown in Figures 5 and 6, respectively. The average valuesof the field descriptors and radar backscatter for each classwere reported in Tables 3 and 4. Analysis of Variance(ANOVA) was performed using Statistica, and the results werealso included in the two tables in order to evaluate if theaverage values were different between the classes. Valuesthat do not differ at a 95 percent significance level wereshaded.

In the cornfield, spatial patterns were well delineatedfrom the classifications for the first two campaigns. Visualcomparison between Figures 5a and 5d showed that SAR

incidence angle had a slight influence on the spatial pat-terns during IFC1. Spatial patterns derived at IFC2 (Figure 5b)were also similar to that derived at IFC1 (Figure 5a), althoughsome small areas were dissolved from Class 1 into Classes 2and 3, and some others altered class attributes betweenclasses 2 and 3. The areas of Class 1 are well to imperfectlydrained sandy soil associations of deep and shallow sandysoils over clay materials. Class 2 represented an imperfectlydrained to poorly drained conditions of sandy loam to finesandy loam soil, and Class 3 represented poorly drained finetextured clay loam to silty clay loam soil. The field descrip-tors were significantly different in these three classes. BothEC30 and EC100 were significantly lower in Class 1 than theother two classes. Either because of a drier soil condition, ora slower canopy development (i.e., lower LAI) in response tothis dry condition, the HH and VV backscatter were theweakest in this class, with more than 2 dB lower at IFC1and 1.4 dB lower at IFC2 when compared with the other twoclasses. At the first two IFCs, HV polarization was posi-tively related with LAI in that, both LAI and HV backscatterincreased in the order of classes 1, 3, 2. This was becauseHV polarization mostly represented a volumetric scatteringcomponent of canopy, which increased with the increasingof canopy biomass. The spatial patterns of the differentclasses obtained at IFC3 (Figure 5c and 5e) were spatiallyfragmented. Due to the reduced variability of SAR data atthis stage, the discriminant ability was limited, leading todecreased differences between classes (Table 3).

The northeastern corner of the wheat field was charac-terized with sandy clay loam to fine sandy loam, whereas

06-098.qxd 8/9/08 12:33 AM Page 1164


the other portions were characterized by silty clay loam toclay loam. At IFC1, all the field descriptors were significantlydifferent among the delineated classes with the 55° inci-dence angle data (Figure 6a). LAI was highest in Class 2 (tothe eastern portion of the field) and lowest in Class 1 (to thewestern portion of the field). For the 35° incidence angle,LAI in Class 2 was distinctively higher than the other twoclasses (Figure 6d). At IFC2, LAI among the three classes weresignificantly different (Figure 6b). Class 2 delineated at thistime was in conformity with the sandy soil distribution area.

Soil electrical conductivity was significantly lower, andthe wheat canopy developed faster in this region than theother regions. Among the three classes obtained at IFC3, soilelectrical conductivity was lower in Class 2 than the othertwo classes, but LAI was not well differentiated. Due to alarger portion of vertical structure of wheat canopy, the VV

polarization was higher at 55° than at 35° incidence angleat this time. Compared with HH and HV polarizations, VV

polarization was the most distinctive among the delineatedclasses in all the cases.

TABLE 3. AVERAGES OF THE SELECTED FIELD DESCRIPTORS AND RADAR POLARIZATIONS IN EACH

DELINEATED HOMOGENEOUS ZONE IN THE CORNFIELD. “C” REPRESENTS CLASSES. FOR A GIVEN

CLASSIFICATION, THE SHADED VALUES WERE NOT DIFFERENT AT 0.05 SIGNIFICANCE LEVEL AS

DETERMINED BY ANALYSIS OF VARIANCE (ANOVA)

Field Descriptors SAR Polarizations (dB)

Yield LAI EC30 EC100Corn C (kg m�2) (m2m�2) (mSm�1) (mSm�1) HH VV HV

1 0.65 0.29 3.08 7.36 �16.6 �18.1 �24.9IFC1 2 0.70 0.36 7.61 17.22 �13.2 �15.6 �21.555° 3 0.73 0.32 9.02 19.15 �14.1 �15.7 �23.0

1 0.64 0.28 2.48 6.10 �14.5 �14.8 �24.4IFC1 2 0.69 0.37 7.51 17.11 �12.2 �12.6 �20.735° 3 0.74 0.31 9.31 19.76 �12.1 �12.1 �21.8

1 0.63 0.82 2.77 6.02 �10.8 �12.7 �20.1IFC2 2 0.69 1.72 7.48 17.09 �9.4 �10.5 �18.155° 3 0.73 1.25 8.33 17.84 �9.0 �10.9 �18.5

1 0.68 3.95 5.17 12.08 �10.7 �11.9 �20.0IFC3 2 0.69 4.01 7.14 16.17 �10.9 �11.9 �19.255° 3 0.72 3.64 8.59 18.17 �10.9 �12.6 �19.9

1 0.67 3.95 6.07 13.60 �8.7 �9.7 �17.9IFC3 2 0.70 4.05 7.29 16.14 �8.9 �9.9 �17.235° 3 0.72 3.61 7.59 16.75 �8.7 �10.6 �17.9

TABLE 4. AVERAGES OF THE SELECTED FIELD DESCRIPTORS AND RADAR POLARIZATIONS IN EACH

DELINEATED HOMOGENEOUS ZONE IN THE WHEAT FIELD. “C” REPRESENTS CLASSES. FOR A GIVEN

CLASSIFICATION, THE SHADED VALUES WERE NOT DIFFERENT AT 0.05 SIGNIFICANCE LEVEL

AS DETERMINED BY ANALYSIS OF VARIANCE (ANOVA)

Field Descriptors SAR Polarizations (dB)

Yield LAI EC30 EC100Wheat C (kg m�2) (m2m�2) (mSm�1) (mSm�1) HH VV HV

1 0.33 0.71 20.45 26.93 �13.5 �14.4 �20.6IFC1 2 0.38 1.49 15.87 22.69 �12.7 �13.5 �19.555° 3 0.35 1.06 18.86 24.98 �12.5 �14.1 �19.9

1 0.32 0.76 20.76 27.27 �10.7 �11.6 �19.0IFC1 2 0.38 1.49 16.04 22.55 �9.8 �11.0 �17.435° 3 0.36 0.83 19.44 25.89 �10.3 �10.5 �18.3

1 0.35 1.86 19.27 25.57 �11.0 �14.5 �21.7IFC2 2 0.40 2.40 12.83 19.79 �11.5 �13.3 �22.355° 3 0.34 1.46 19.48 25.99 �11.3 �14.2 �21.1

1 0.34 0.94 18.69 25.12 �10.5 �11.5 �22.5IFC3 2 0.41 0.95 12.18 18.52 �11.0 �10.8 �20.955° 3 0.36 1.12 19.15 25.75 �11.2 �11.8 �22.7

1 0.34 0.98 18.83 25.33 �8.8 �13.1 �20.7IFC3 2 0.38 1.02 15.10 22.02 �9.5 �12.3 �20.135° 3 0.36 1.08 19.05 25.27 �9.0 �13.5 �20.2

06-098.qxd 8/9/08 12:33 AM Page 1165


The yield of both corn and wheat was significantlydifferent among the delineated classes for most of thecases. As observed from the class statistics, the two cropsresponded differently to the sandy soil. In the cornfield,there was a slower canopy development and a lower yieldin the sandy soil areas, whereas in the wheat field, thecrop developed faster and the final yield was the highest inthe sandy soil area.

ConclusionsAn object-based Fuzzy k-means unsupervised classificationapproach was proposed in this study for within field homog-enous zone delineation using C-band multi-polarization SAR

data. Multi-temporal linear polarization data over a cornfieldand a wheat field were synthesized from CV-580 C-Bandpolarimetric SAR data and were classified using the proposedapproach. Classification was evaluated by inspecting the

variance reduction and analysis of variance (ANOVA) of theselected field descriptors, green LAI, yield, and soil electricalconductivity.

In most of the cases, the object-based classificationachieved better results than the pixel-based classification forthe selected field descriptors, as evaluated by variancereduction of field descriptors. Although a priori knowledge,such as soil properties represented by the electrical conduc-tivity measurements, can be helpful to define meaningfulimage objects, the advantages of the algorithm implementedin eCognition® reside in the integration of spatial hetero-geneity with spectral heterogeneity in the segmentationstage. Thus, information in both the spatial and spectraldomain of remote sensing data could be exploited for betterimage analysis.

Classification of multi-temporal multi-polarization SAR

data produced spatial patterns that were interpretable usingcrop and soil information. SAR is capable of delineating

Figure 5. Fuzzy k-means classification of the cornfields into three classesusing HH, VV, and HV polarizations: (a): IFC1 (55°), (b): IFC2 (55°), (c): IFC3(55°), (d): IFC1 (35°), and (e): IFC3 (35°).

06-098.qxd 8/9/08 12:33 AM Page 1166


Figure 6. Fuzzy k-means classification of the wheat field into three classes using HH, VV, and HV polarizations. (a): IFC1 (55°), (b): IFC2 (55°), (c): IFC3 (55°),(d): IFC1 (35°), and (e): IFC3 (35°).

within field spatial patterns of relatively stable soil propertieswhen crop fraction is low. With the crop development, theclassification can map the crop growth conditions as repre-sented by green LAI. Different behaviors in seasonal variationof backscatter and its variability were observed for wheat andcorn canopies. Although yield could be discriminated amongdifferent classes, the achieved variance reduction is limited.

It should be noted that SAR backscatter is an integrationof scattering and attenuation within a tightly coupled soil-crop system. Further studies are expected to explore thepolarimetric information to decouple this system for quanti-tative retrieval of field descriptors.

AcknowledgmentsThe study was done with the financial support from theGovernment Related Initiative Program (GRIP) managed by theCanadian Space Agency (CSA) and from Agriculture and Agri-Food Canada. We thank D. Dow, D. Meredith, M. Hinther, aswell as the CRESTECH team led by Dr. J. Miller (York Univer-sity), Dr. D. Haboudane (University of Qubec at Chicoutimi),the SAR team led by Dr. Bob Hawkins (CCRS) for data collec-tion and analyses (ECORC contribution number: 06-694).

ReferencesBaatz, M., U. Benz, S. Dehghani, M. Heynen, A. Holtje, P. Hofmann,

L. Lingenfelder, M. Mimler, M. Soplbach, M. Weber, andG. Willhauck, 2002. eCognition Object Oriented Image Analysis,User Guide 3, Definiens Imaging Gmbh, Trappentreustrasse 1,80339 Munchen, Germany.

Benz, U.C., P. Hofmann, G. Willhauck, L. Lingenfelder, and M. Heynen, 2004. Multi-resolution, object-oriented fuzzy analysisof remote sensing data for GIS-ready information, ISPRS Journalof Photogrammetry and Remote Sensing, 58:239–258.

Cloude, S.R., and E. Pottier, 1997. An entropy based classifica-tion scheme for land applications of polarimetric SAR, IEEETransactions on Geoscience and Remote Sensing, 35(1):68–78.

Corwin, D.L., S. M. Lesch, P.J. Shouse, R. Soppe, and J.E. Ayars,2003. Identifying soil properties that influence cotton yieldusing soil sampling directed by apparent soil electrical conduc-tivity, Agronomy Journal, 95(2):352–364.

Corwin, D.L., and S.M. Lesch, 2003. Application of soil electricalconductivity to precision agriculture: theory, principles, andguidelines, Agronomy Journal, 95(3):455–471.

Corwin, D.L., and S.M. Lesch, 2005. Apparent soil electricalconductivity measurements in agriculture, Computers andElectronics in Agriculture, 46:11–43.

Dobson, M.C., F.T. Ulaby, and L.E. Pierce, 1995. Land-cover classifi-cation and Estimation of terrain attributes using syntheticaperture radar, Remote Sensing of Environment, 51:199–214.

Doerge, T., 1999. Defining management zones for precision farming,Crop Insights, 8(21):1–5.

Ferrazzoli, P., S. Paloscia, P. Pampaloni, G. Schiavon, S. Sigismondi,and D. Solimini, 1997. The potential of multifrequencypolarimetric SAR in assessing agricultural and arboreousbiomass, IEEE Transactions on Geoscience and RemoteSensing, 35(1):5–17.

Fox, P.A., A.P. Luscombe, and A.A. Thompson, 2004. RADARSAT-2SAR modes development and Utilization, Canadian Journal ofRemote Sensing, 30(3):258–264.

Fridgen, J.J., N.R. Kitchen, and K.A. Sudduth, 2000. Variabilityof soil and landscape attributes within sub-field managementzones, Proceedings of the Fifth International Conference onPrecision Agriculture (P.C. Robert, R.H. Rust, and W.E. Larsen,editors), 16–19 July, Bloomington, Minnesota, ASA-CSSA-SSSA.,Madison, Wisconsin, unpaginated CD-ROM.

Haboudane, D., J.R. Miller, E. Pattey, P.J. Zarco-Tejada, andI. Strachan, 2004. Hyperspectral vegetation indices and novelalgorithms for predicting green LAI of crop canopies: Model-ling and validation in the context of precision agriculture,Remote Sensing of Environment, 90:337–352.

06-098.qxd 8/9/08 12:33 AM Page 1167


Lee, J.S., M. Grunes, T.L. Ainsworth, L.J. Du, D.L. Schuler, andS.R. Cloude, 1999. Unsupervised classification using polari-metric decomposition and the complex Wishart classifier,IEEE Transactions on Geoscience and Remote Sensing,37(5):2249–2258.

Lee, J.S., M. Grunes, E. Pottier, and L. Ferro-Famil, 2004. Unsuper-vised terrain classification preserving polarimetric scatteringcharacteristics, IEEE Transactions on Geoscience and RemoteSensing, 42(4):722–731.

Liu, J., J.R. Miller, D. Haboudane, E. Pattey, and M.C. Nolin, 2005.Variability of seasonal CASI image data products and potentialapplication for management zone delineation for precisionagriculture, Canadian Journal of Remote Sensing,31(5):400–411.

Marshall, I.B., J. Dumanski, E.C. Huffman, and P. Lajoie, 1979. SoilsCapability and Land Use in the Ottawa Urban Fringe, LandResource Research Institute, Agriculture Canada, Ottawa,Ontario, 59 p.

Mattia, F., T.L. Toan, G. Picard, F.I. Posa, A. D’Alessio, C. Notarnicola,A.M. Gatti, M. Rinaldi, G. Satalino, and G. Pasquariello, 2003.Multitemporal C-Band radar measurements on wheat fields,IEEE Transactions on Geoscience and Remote Sensing,41(7):1551–1560.

McNairn, H., and B. Brisco, 2004. The application of C-bandpolarimetric radar for agriculture: A review, Canadian Journalof Remote Sensing, 30(3):525–542.

Minasny, B., and A.B. McBratney, 2002. FuzMe, version 3.0,Australian Centre for Precision Agriculture, The University ofSydney, NSW 2006.

Moran, M.S., Y. Inoue, and E.M. Barnes, 1997. Opportunities andlimitations for image-based remote sensing in precision cropmanagement, Remote Sensing of Environment, 61:319–346.

Perron, I., M. Nolin, E. Pattey, J.L. Bugden, and A. Smith, 2003.Comparison de l’utilisation de la conductivité électriqueapparente (CEa) des sols et des données polarimétriques RSOpour délimiter des unités d’amenagement agricole, Proceedingsof the 25th Canadian Symposium on Remote Sensing – 11eCongrès de l’Association québécoise de télédétection, 14–16October, Montréal, Quebec, Canada, unpaginated CD-ROM.

Roubens, M., 1982. Fuzzy clustering algorithms and their clusteringvalidity, European Journal of Operational Research, 10:294–301.

StatSoft, Inc., 2005. STATISTICA (data analysis software system),version 7.1., URL: www.statsoft.com (last date accessed:23 May 2008).

Touzi, R., and F.J. Charbonneau, 2004. PWS: A friendly and effectivetool for polarimetric image analysis, Canadian Journal of RemoteSensing, 30(3):566–571.

Wiegand, C.L., and A.J. Richardson, 1984. Leaf area, light interception,and yield estimates from spectral components analysis, AgronomyJournal, 76:543–548.

Wiegand, C.L., A.J. Richardson, and P.R. Nixon, 1986. Spectralcomponent analysis: A bridge between spectral observationsand agrometeorological crop models, IEEE Transactions onGeoscience and Remote Sensing, GE-24:83–88.

(Received 01 September 2006; accepted 13 December 2006; revised10 January 2007)

September Layout.indd 1168September Layout.indd 1168 8/12/2008 11:51:43 AM8/12/2008 11:51:43 AM

object-based classification of high resolution sar images ......soil properties, and other stresses...

Documents