A spatial pattern analysis of the halophytic species distribution in an arid coastal environment

1 23

Environmental Monitoring andAssessmentAn International Journal Devoted toProgress in the Use of Monitoring Datain Assessing Environmental Risks toMan and the Environment ISSN 0167-6369Volume 187Number 5 Environ Monit Assess (2015) 187:1-15DOI 10.1007/s10661-015-4403-z

A spatial pattern analysis of the halophyticspecies distribution in an arid coastalenvironment

Nasem Badreldin, J. Uria-Diez, J. Mateu,Ali Youssef, Cornelis Stal, Magdy El-Bana, Ahmed Magdy & Rudi Goossens

1 23

Your article is protected by copyright and

all rights are held exclusively by Springer

International Publishing Switzerland. This e-

offprint is for personal use only and shall not

be self-archived in electronic repositories. If

you wish to self-archive your article, please

use the accepted manuscript version for

posting on your own website. You may

further deposit the accepted manuscript

version in any repository, provided it is only

made publicly available 12 months after

official publication or later and provided

acknowledgement is given to the original

source of publication and a link is inserted

to the published article on Springer's

website. The link must be accompanied by

the following text: "The final publication is

available at link.springer.com”.

A spatial pattern analysis of the halophytic species distributionin an arid coastal environment

Nasem Badreldin & J. Uria-Diez & J. Mateu &

Ali Youssef & Cornelis Stal & Magdy El-Bana &

Ahmed Magdy & Rudi Goossens

Received: 29 July 2014 /Accepted: 4 March 2015# Springer International Publishing Switzerland 2015

Abstract Obtaining information about the spatialdistribution of desert plants is considered as a seri-ous challenge for ecologists and environmentalmodeling due to the required intensive field workand infrastructures in harsh and remote arid environ-ments. A new method was applied for assessing thespatial distribution of the halophytic species (HS) inan arid coastal environment. This method was basedon the object-based image analysis for a high-resolution Google Earth satellite image. The integra-tion of the image processing techniques and fieldwork provided accurate information about the spatialdistribution of HS. The extracted objects were basedon assumptions that explained the plant-pixel rela-tionship. Three different types of digital image pro-cessing techniques were implemented and validatedto obtain an accurate HS spatial distribution. A total

of 2703 individuals of the HS community werefound in the case study, and approximately 82 %were located above an elevation of 2 m. The micro-topography exhibited a significant negative relation-ship with pH and EC (r=−0.79 and −0.81, respec-tively, p<0.001). The spatial structure was modeledusing stochastic point processes, in particular a hy-brid family of Gibbs processes. A new model isproposed that uses a hard-core structure at very shortdistances, together with a cluster structure in short-to-medium distances and a Poisson structure forlarger distances. This model was found to fit thedata perfectly well.

Keywords Arid environment . Digital imageprocessing . Google earth . Spatial pattern analysis

Environ Monit Assess (2015) 187:224 DOI 10.1007/s10661-015-4403-z

N. Badreldin (*) : C. Stal :R. GoossensDepartment of Geography, Ghent University, Krijgslaan 281S8, 9000 Ghent, Belgiume-mail: [email protected]

N. Badreldine-mail: [email protected]

J. Uria-DiezDepartment of Environmental Biology, University of Navarra,Irunlarrea s/n, 31008 Pamplona, Spain

J. Uria-DiezDepartment of Ecosystem Modeling, Faculty of ForestSciences and Forest Ecology, University of Goettingen,Buesgenweg 4, 37077 Goettingen, Germany

J. MateuDepartment of Mathematics, Universitat Jaume I, CampusRiu Sec, 12071 Castellón, Spain

A. YoussefDivision M3-BIORES: Measure, Model & ManageBioresponses, Kasteelpark Arenberg 30, box 2456,3001 Leuven, Belgium

M. El-BanaDepartment of Botany, Faculty of Science, Port SaidUniversity, Port Said, Egypt

A. MagdyCentral Laboratory for Agricultural Climate (CLAC),Agricultural Research Center (ARC), Ministry of Agricultureand Land Reclamation (MALR), Dokki, Giza, Egypt

Author's personal copy

Introduction

The spatial distribution of the natural plants in an aridenvironment, such as the halophytic species (HS), isusually not random, and there are a variety of factorsthat influence this distribution. These factors are classi-fied into two general classes: (1) environmental factorsthat are spatially heterogeneous, such as soil propertiesand micro-topography; and (2) genetic factors that pro-vide the natural plants with the ability to survive inadverse environmental conditions (Dale 1999). Manystudies found a relationship between the natural plants’distribution and the environmental heterogeneity of fac-tors such as the soil depth (Kershaw 1959), topography(Greig-Smith 1961), soil nutrients (Galiano 1985), po-sitions of the subsurface rocks (Usher 1983), and soilsalinity, which are considered as the most abiotic factorsassociated with HS distribution (Ungar et al. 1977;Rubio-Casal et al. 2001).

In ecological studies, the explicit considerations ofthe spatial structures play an important role in under-standing and managing ecological processes. If we in-clude the spatial component of the studied community,we could characterize the nature and intensity of thespatial relationships between organisms and their envi-ronment (Fortin and Dale 2009). Normally, natural eco-systems are not spatially homogeneous and show aspecific spatial structure. Investigation of the spatialstructure using point pattern analysis can therefore be afirst step toward providing important insights into theprocesses affecting the community ecology (Perry et al.2006). Spatial point processes (e.g., Gibbs models) ex-hibit spatial interactions at only a single spatial scale,whereas natural processes exhibit dependence at multi-ple scales (Delicado et al. 2010). Thus, with the use ofspatial statistics, like new hybrid spatial point processes(Baddeley et al. 2013), we are able to show the jointeffects of biotic processes and abiotic heterogeneity onthe spatial pattern of the halophyte community.

In the past 10 years, a useful approach in the ecolog-ical and environmental sciences has been the use ofdigital image processing to improve image quality(Schowengerdt 2007). Such an improvement can beestablished using standard image processing techniques,which can represent the case study as a digital datasetwith greater than 90 % accuracy (Liu and Mason 2009;Quets et al. 2013). Developing appropriate and powerfulimage processing algorithms depends on the transfor-mations of the functionalities using neighborhood

processing techniques, which allowed the parallel pro-cessing of the images in real time with conceptuallyhigh-level operations (Shih 2010). Satellite images areoften used as datasets for environmental monitoring,and the main approach is to numerically capture thecoordinates of the studied points or habitats. The digitalimage processing techniques deliver additional informa-tion about the environment status, which requires datavalidation from the field survey (Vincent and Soille1991; Winkler 2006; Illian et al. 2008; Blaschke2010). Google Earth provides high-resolution satelliteimages and is freely accessible to the public. Moreover,many researchers have found that these images arevaluable for environmental studies because of their ex-ceptional quality (Sheppard and Cizek 2009; Sadr andRodier 2012; Taylor and Lovell 2012; Yang et al. 2012;Quets et al. 2013; Frankl et al. 2013).

The specific objectives for this investigation were (1)to improve Google Earth satellite images using differentsimple digital image processing operators to obtain theHS spatial distribution and to understand the local envi-ronmental variables’ impact, and subsequently, (2) toanalyze the spatial distribution of the community ofhalophyte species in the El-Zaraniq Protected Area,North Sinai, Egypt, to provide a combination of modelsthat best reflects the observed structures.

Materials and methods

Case study description and field survey

Approximately 95 % of the total area of Egypt is cov-ered by desert, while approximately 85 million inhabi-tants depend on the remaining 5 % of the land (Shalabyand Tateishi 2007; Abd El-Kawy et al. 2011). The casestudy was conducted in the northern side of the El-Zaraniq Protected Area in the North SinaiGovernorate, Egypt, between the latitudes 31° 7′12.38″, 31° 7′ 1.46″ N and the longitudes 33° 30′18.44″, 33° 30′ 29.99″ E. The study area is approxi-mately 10.25 ha, as shown in Fig. 1. Based on theclassification system (soil taxonomy) that was devel-oped by the United States Soil Conservation Services(USSCS), the major soil group that was found in thecase study is Entisols (Greenwood 1997), and theNatural Resources Conservation Service NRCS (1999)described the Entisols soils as having little or no devel-opment of pedogenic horizons that consist mainly of

224 Page 2 of 15 Environ Monit Assess (2015) 187:224

Author's personal copy

quartz. According to the climate classification of Kotteket al. (2006) and Peel et al. (2007), the Sinai is generallyclassified as (Bwh) arid-desert-hot climate; however, theclimate of El-Zaraniq is a Mediterranean arid type; thetemperature is above 35 °C in the summer and ∼15 °C inthe winter, and the mean annual precipitation is approx-imately 150 mm with ±50 mm. The vegetation inZaraniq is composed of over 150 plant species (El-Bana et al. 2002).

The field survey and data collection were performedduring January 2012. At that time of the year, thevegetation covers is at its maximum growth stage.Additionally, the minimum accumulation of salts at thesoil surface offered better living conditions to the HSthat were investigated in the case study.

The data collection is based on unbiasedness andrepresentativeness to control the sampling error. Thesoil surface characteristics, HS distribution, and avail-able measuring tools are considered carefully for furtherstatistical analysis and data mapping. The ground con-trol points and soil samples were selected by means of asystematic sampling technique at the corners of an ap-proximately 1000-m2 grid.

Ninety soil samples were obtained from the top soilsurface (25 cm depth) during the field work. The soilsamples were analyzed in the laboratory using 1/5 di-luted extracts for soil salinity and 1/2.5 for soil pHanalysis (U.S. Salinity Laboratory 1954). The soil salin-ity refers to the total concentration of dissolved salts perunit volume, with the salt concentration indicated in theelectrical conductivity (EC), which is expressed in unitsof decisiemens per meter (Rhoades 1992). The pH levelis a parameter used for indicating the acidity status of the

soil; the pH range is from 0 to 14, where 7 is the neutralstatus, below 7 is acidic, and above 7 is basic (Blanco-Canqui and Lal 2010). Elevation of the case study isdelivered from the high-resolution topographic maps.

The data analysis was conducted in two parts: animage processing step and a spatial pattern analysis stepaccording to the framework in Fig. 2. This investigationwas based on a GeoEye satellite image with resolutionof 0.5 m, which was provided via Google Earth. Thesatellite image was acquired for a date of January 2012in a clear atmospheric condition without the presence ofclouds or sandstorms, which could affect the imagevisualization. ENVI 4.8, Matlab, and R software wereused for the digital image and the spatial pattern analy-ses (spatstat package).

Google Earth image processing

Image pre-processing is an essential treatment forobject-based image analysis research to establish thedirect link between the image and the case study char-acteristics (Coppin et al. 2004; Badreldin and Goossens2013a). Various image pre-processing operations wereconsidered, including image enhancement and geomet-ric correction.

The purpose of the image enhancement was to sub-jectively improve the satellite images for better analysis.Manipulating the image histogram was performed toincrease the highest possible contrast and remove thenoise (Petrou and Bosdogianni 1999). The point opera-tion for modifying the image histogram is considered asthe most powerful technique to optimize the best bright-ness and contrast for highlighting information, in

Fig. 1 The geographical location and dimension of the case study

Environ Monit Assess (2015) 187:224 Page 3 of 15 224

Author's personal copy

particular the digital number (DN) ranges, withoutchanging the image size, as shown in Fig. 3.

Object-based image studies required geometric cor-rections to exploit image usefulness and obtain accurateresults (Knorn et al. 2009). The Google Earth imageused was geo-referenced using a set of seven groundcontrol points (GCPs) by GPS, where the root meansquare error (RMSE) <0.015 pixels.

The extracted objects from the image were based onthe assumptions that explained the spectral response for

each plant within that pixel. We assumed that each plantlocation is represented in the image with a distinctivespectral response and that most of the HS individualswere observed on the top of the sand accumulationcalled Bphytogenic hummocks^ with an average heightof 0.5 m (El-Bana et al. 2007). The schematic of thisphenomenon shown in Fig. 4a, b explained the differ-ence in the solar radiation reflectance between individ-uals of HS and the hummocks’ edges. The plant positionexhibited more reflectance than the surrounding area,

Fig. 2 Overview of data analysissteps and layouts, starting fromfield survey and sampling,Google Earth image analysis,target object extraction, andspatial pattern analysis

224 Page 4 of 15 Environ Monit Assess (2015) 187:224

Author's personal copy

which is exposed Bbarren soil^ that causes scattering ofmost of the incident solar radiation that is received fromthe sun (Fig. 4c).

We implemented three different types of operators inimage processing and spatial filtering to extract thetarget objects by image convolution. These operatorsare the standard operators in digital image processingand are based on moving identified kernels over all ofthe pixels within the image boundaries. In the context ofimage processing, a 3×3 kernel size was used as aninput array and was multiplied with gray-level scaleimage. The resultant image presented both characteris-tics, as shown in Fig. 5.

Roberts cross edge detector (RCED)

RCED is a non-linear filter that is similar to the Sobeloperator and has the advantages of being a standard andquick method to process a grayscale image.Additionally, this method is based on 2D differencingfor sharpening edges and object isolation, and it is not

possible to change the filter size and edit kernels. Intheory, the operator consists of a pair of 3×3 convolu-tion kernel masks; one kernel is simply the other rotatedby 90°. Themasks were applied separately in the studiedimage to produce separate measurements of the gradientcomponent in each orientation (denoted Rx and Ry);these masks can then be combined to determine theabsolute magnitude of the gradient. The gradient mag-nitude is given by

Rj j ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiRx2 þ Ry2

pð1Þ

An approximate gradient magnitude was computedusing

Rj j ¼ Rxj j þ Ryj j ð2Þ

Dilate operator (DO)

DO was used to expand the bright pixels and enhancethe dark spots in the image. The output was built on the

Fig. 3 The radiometriccorrection of the satellite image ofthe case study. a The originalsatellite image. b After the imageenhancement

Fig. 4 Illustration draws of the target objects BHS^. a Single object position. b Group of objects positions. c The reflectance and absorbedsolar radiation pattern

Environ Monit Assess (2015) 187:224 Page 5 of 15 224

Author's personal copy

principle of filling the holes smaller than the structuralkernel in our panchromatic image. Every pixel in thestudied image is brought into a line with the darkestpixels of the eight neighboring in a 3×3 matrix.

d ¼ i⊕k ð3Þ

Erode operator (EO)

EO was used to reduce the clustered bright pixels andexpand the dark ones. The values of the output pixels arethe minimum values of all of the pixels in the inputpixel’s neighborhoods.

e ¼ i⊖k ð4Þ

Data validation

Random HS positions were counted and located in thefield using GPS. To evaluate the results of the objectextraction analysis, the field survey data were used as areference. If this experimental dataset totally correlateswith the estimate generated by the edge detectors, theEuclidean distance between a field measurement and anestimated point should be equal to zero. Because the

Fig. 5 Three different types of operators (RCED, Dilate, and Erode) for image processing and spatial filtering to extract the plant positionstarget objects by image convolution

Table 1 Summary statistics of the field survey data and theestimated object’s location

Roberts Erode Dilate

Count 87 87 87

Mean (m) 7.70 23.31 12.60

St. dev (m) 5.96 17.56 8.35 Fig. 6 Cumulative frequencies distribution and distance betweenthe measured estimated points

224 Page 6 of 15 Environ Monit Assess (2015) 187:224

Author's personal copy

estimated locations were determined using a prior kernelsize and the edge detectors are sensitive to noise, therewill be deviation between the measured and the estimat-ed locations. For each measured point, the nearest loca-tion from each edge detector was selected. The distancesbetween this measured point and its nearest counterpartwere analyzed, resulting in amean distance and standarddeviation, as demonstrated in Table 1.

Intuitively, the distribution of these deviations willclearly not be normal but will follow a gamma distribu-tion. This distribution is illustrated by Fig. 6, where thecumulative frequencies are plotted against the distancebetween a measured point and the estimated point. Eachedge detector is represented by a curve. Based on thisfigure, as well as Table 1, it becomes clear that theRCED is the best operator that represents the HS loca-tions in this particular case study.

The obtained soil samples that represent soil salinityand pH are denoted z(x1),z(x2),…,z(n), where xi is thelocation vector and n is the number of observations thatwere used for predicting the uncollected soil samplesinside each grid over the entire case study area using theordinary kriging (OK) method (Webster and Oliver2007).

Z x0ð Þ ¼X m

i¼1λiz xið Þ ð5Þ

where m is the number of neighbors that have beenobserved and λi is the weights.

Ninety soil samples were used for cross validation,and 22 independent soil samples were collected ran-domly from the case study for validating the EC andpH maps. Two accuracy assessment parameters wereused: (1) the RMSE was used to indicate how far thepredicted values from the data measurement, and (2) theroot mean square standardized error (RMSSE) was usedto identify whether the predicted values are under/overestimations (Webster and Oliver 2007).

RMSE ¼ffiffiffi1

n

r X m

i¼1Z x0ð Þ−z xið Þf g2 ð6Þ

RMSSE ¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiX m

i¼1Z x0ð Þ−z xið Þf g

.σ xið Þ

h i2n

vuutð7Þ

where σ(xi) is the estimated error variance and n is thesample size.

The validation results of the OK method confirmedthe accuracy of the predicted EC and pH maps, with thecorrelation coefficient calculated between measured andestimated EC and pH values presented in Table 2.

Table 2 Statistical parameters (RMSE and RMSSE) for soil EC and pH validation, based on cross validation (n=90) and separate datasets(n=22)

Statistical parameter EC r2 r pH r2 r

Cross validation RMSE 0.3993 n=90 0.8323 0.4199 n=90

RMSSE 0.7678 0.912 0.9526 0.3647 0.607

Separate dataset RMSE 0.4053 n=22 0.2459 n=22

RMSSE 1.0840 0.8614 0.928 0.6034 0.7929 0.882

Fig. 7 The spatial distribution of the soil salinity (EC) and pH in the case study

Environ Monit Assess (2015) 187:224 Page 7 of 15 224

Author's personal copy

The spatial distribution of the soil salinity and the pHare essential to understand the spatial distribution pat-terns of HS in the case study. As shown in Fig. 7, theeastern side of the case study is more affected by highsoil salinity (>4 dS/m) and alkalinity (pH>9), whichrepresent a serious land degradation condition in thestudy site.

Spatial analysis and statistical modeling

A planar point process is a random collection of pointsin the plane. Point processes are useful statistical toolsthat can be applied in a variety of scientific fields, suchas forestry, epidemiology, or ecology. Modeling andinference for spatial point processes is an issue that hasbeen investigated broadly in the last few years. The widerange of application fields has been the main enginedriving such increased interest.

In the statistical analysis of the spatial point patterndata, a parametric modeling approach requires a supplyof stochastic models for the point patterns. One impor-tant source of the models is the class of finite Gibbspoint processes. Many models of this class can be fittedrapidly to real datasets containing large numbers ofpoints (Geyer 1999; Diggle 2003; Møller andWaagepetersen 2003; Baddeley and Turner 2006).

However, in practice, the Gibbs models currentlyavailable for use in data analysis are few in numberand limited in scope. The most commonly used Gibbsmodels exhibit spatial interaction at only a single spatialscale, whereas most natural processes exhibit depen-dence at multiple scales. The key problem is that theconstruction of new Gibbs models ab initio is not trivial.Gibbs models are usually defined using the un-

Fig. 8 The relationship betweenthe soil salinity (EC) distributionsalong with the micro-topographyof the case study

Fig. 9 HS spatial distribution and density per square meter in thecase study

224 Page 8 of 15 Environ Monit Assess (2015) 187:224

Author's personal copy

normalized probability density h. While it is easy towrite down a new functional form for h, it is not alwaysobvious whether h is integrable, so that the normalizingconstant is finite and the process is well defined.

This paper considers a practical technique for con-structing new Gibbs models from the existing ones. Theun-normalized densities (h1h2) of two existing modelsare multiplied together to form a new hybrid un-

Fig. 10 HS distributions over different micro-topography and its relation with a pH and b EC

Environ Monit Assess (2015) 187:224 Page 9 of 15 224

Author's personal copy

normalized density h(x)=h1(x), h2(x). This process iscalled hybridization, and it defines a way to constructnew point process models to be fitted to data. Hybridsare particularly useful for modeling interactions at mul-tiple scales. This method has been demonstrated to beuseful on a real dataset on human social interaction(Baddeley et al. 2013).

The analysis and construction of hybrid models wererun through several modeling steps. We first investigat-ed the spatial inhomogeneity and then the evidence ofany interaction between points. Next, with this previousanalysis result, we started building models of increasingcomplexity with the aim of describing the observedbehavior of the data. The spatial trend was modeledusing a parametric estimate with a log-linear combina-tion of the EC covariates and a log-cubic function of theCartesian coordinates. This log-linear function wasfitted by a maximum likelihood approach using theBerman-Turner device (Berman and Turner 1992;Baddeley and Tuner 2000). We verified the goodness-of-fit of the linear function using smoothed partial re-sidual diagnostics, which is a useful diagnostic for thetransformation of a covariate in a Poisson point processmodel. Then, considering the estimated degree of inho-mogeneity using the fitted parametric first-order inten-sity function, we estimated the inhomogeneous paircorrelation function (Baddeley et al. 2000). The avail-able Gibbs processes are able to model inter-point inter-action and tend to have a simple mathematical structurethat cannot model complex patterns. In this paper, wetake three existing (and widely used) models (Hard-core, Geyer, and Strauss spatial point processes (Illian

et al. 2008)). We used the method of maximum profilepseudolikelihood to calculate the a priori irregular pa-rameters from the three models. Then we multiplied theprobability densities together and renormalized to obtaina new probability density. The resulting model is ahybrid that is particularly useful for point patterns thatexhibit interactions at different spatial scales.

Results and discussions

Understanding the abiotic factor distribution

Soil salt accumulation in an arid climate condition is adominant land degradation process in the case study,where the evaporation is higher than the precipitation.This arid condition along with the seepage ofMediterranean Sea water is causing water logging tosalty ground, which results in the formation of surfacesalt crust and increased soil salinity (Goossens and VanRanst 1998; Badreldin and Goossens 2013b). Themicro-topographic characteristics are influencing thedistribution of the soil salinity in the case study. Thelower micro-elevation holds up the higher salt concen-tration and vice versa, as shown in Fig. 8. Water loggingis found in lower elevation <2 m (Fig. 8, point 1) andbetween phytogenic hummocks (Fig. 8, points 2); thesepoints are the spots that are most affected by theupraising of salty ground water and most likely willbecome more salinized in the dry seasons. The soilsalinity in Fig. 8 (points 3) is <3.8 dS/m, which provides

Fig. 11 Spatial inhomogeneity of the halophyte community data.a Kernel smoothing estimates of ρ at different values of EC.Intensity units are individuals’ shrubs per square meter in thehalophyte community in El-Zaraniq Protected Area, North Sinai,

Egypt. b Parametric estimate of intensity described by the spatialtrend of EC and log-cubic function of coordinates fitted by max-imum likelihood

224 Page 10 of 15 Environ Monit Assess (2015) 187:224

Author's personal copy

better environmental conditions for HS growth andsurvival.

HS spatial distribution along with EC and pH

HS are distributed in different densities over the entirestudy site, as shown in Fig. 9. The southeastern side ofthe case study exhibited a low plant density (<3 plants/m2), unlike the northern side, which had a higher density(>5 plants/m2).

Soil salinity and pH are considered to be the localenvironmental variables that influenced the HS spatialdistribution in the case study with the association of the

micro-topography. As shown in Fig. 10, approximately82% of HS in the case study were found in elevations of>2 m. The micro-topography exhibited significantlynegative correlations with pH and EC (r=−0.79 and−0.81, respectively): the lower the elevation, the higherare the soil salinity and soil alkalinity at each HS loca-tion. The spatial pattern of the extracted objects indicat-ed that the HS avoided the severe abiotic spots thatcontained higher soil salinity and alkalinity in the casestudy.

Modeling the HS spatial pattern

The analysis of the spatial structure and construction ofthe hybrids starts studying the spatial inhomogeneity byproviding a model for the first-order intensity function.Figure 11a shows the parametric estimate of the inten-sity modeled through a log-linear function of the EC anda log-cubic function of the Cartesian coordinates, fittedby the maximum likelihood using the Berman-Turnerdevice (Berman and Turner 1992; Baddeley and Tuner2000), and Fig. 11b shows the goodness-of-fit of thisproposed model to the halophyte community intensity.

Figure 12 shows the estimation of the inhomoge-neous pair correlation function (Baddeley et al. 2000)for the halophyte community using the fitted intensityfunction shown in Fig. 11b. This function exhibits clearevidence of the Bhard-core^ effect at distances less thanapproximately 1.3 m, meaning that two individualscannot come within the minimum distance betweenthese two points. This effect is combined with a strongattraction between any two individuals for distancesbetween 1.3 and 10 m. Finally, for distances larger than10 m, we observed no spatial interaction (random pat-tern) in the HS community.

Table 3 Parameters and comparison of values of Akaike’s information criterion (AIC) among the three multi-scale hybrid models used tounderstand the spatial structure of halophyte community in El-Zaraniq Protected Area, North Sinai, Egypt

Hard core Geyer Strauss AIC

r0 r1/sat Gamma ( ) r2 Gamma ( )

First model (a) 1.2 – – – – 23313.99

Second model (b) 1.2 1.4/2 2.128 – – 18927.55

Third model (c) 1.2 1.4/2 2.001 12 1.022 14245.2

Parameters are r0 radius of the hard-core process; r1 for interaction radius, sat is saturation threshold that is the number of close neighbors,and gamma ( ) is the interaction parameter of the Geyer process; r2 is the interaction radius and gamma ( ) is the interaction parameter of theStrauss process

Fig. 12 Graph showing pcf(r) functions with parametric kernelintensity with a spatial trend described by the EC and log-cubicfunction of coordinates, fitted by maximum likelihood in inhomo-geneous version The observed ginhom(r) values is expressed asblack solid line and the gray polygon represents the limits of the95 % confidence intervals generated by Monte Carlo simulations,where r represents the distance examined in the halophyte com-munity in El-Zaraniq Protected Area, North Sinai, Egypt

Environ Monit Assess (2015) 187:224 Page 11 of 15 224

Author's personal copy

The hybrid procedure consists of the addition ofdifferent Gibbs models to capture the three observedbehaviors (see Fig. 12, Table 3). We first fit an inhomo-geneous hard-core process with a spatial trend of the ECand the log-cubic function of the coordinates; as shownin Fig. 13a, the simulated pattern does not appear toexhibit the same strong clustering as the data (seeFig. 9). We analyzed the residual of the fitted modelsusing the K-residual function used in Baddeley et al.

(2013). The K-residual from the first model showed apositive association between the points, as it lies outsidethe confidence intervals at a range from 10 to 20 m. Thesecondmodel added to the previous hard-core process isa Geyer process (see Fig. 13b, Table 3). This hybridprocess has parameters of an interaction radius at 1.4 mand 2 HS individuals (number of neighbors), with agamma parameter of 2.001, indicating a strong aggre-gation for distances up to 1.4 m. However, after

Fig. 13 Diagnostics for models of the halophyte community datafor k-residual function and simulated realizations; plus symbols,Bextracted point pattern^; blue circle, Bsimulated point pattern^using a hard core, b hard core plus Geyer, and c hard core plus

Geyer and Strauss processes with parametric intensity estimationused in the halophyte community in El-Zaraniq Protected Area,North Sinai, Egypt

224 Page 12 of 15 Environ Monit Assess (2015) 187:224

Author's personal copy

including another Gibbs process, the second hybrid didnot capture the low level of clustering of the process atthese scales. The differences between the AIC values(AIC is a likelihood-based criterion to select the bestmodel) were significant (Akaike 1973), as shown inTable 3. This difference indicated that the addition ofthe Geyer clustering structure was significant. Our pro-cedure was completed by a third step of the addition of aStrauss process to the second hybrid model. This thirdmodel was able to capture the random pattern, as dem-onstrated by the gamma parameter value of 1.002 of thefitted Strauss model (see Table 3) for a range from 10 to70 m, indicating no interaction at these ranges of dis-tances. This hybrid built with the three models wasclearly the best among the previous ones according tothe AIC values.

Conclusion

This research contributes in understanding the spatialstructure of the halophyte community in an arid envi-ronment through the use of a freely accessible dataset.The integration between field survey, digital image pro-cessing, and point pattern analysis provided a betterinterpretation to the local environmental conditionsand described more accurately the HS spatial distribu-tion in the case study.

The RCED is proven to be the best object extractionoperator that provided accurate information about theHS distribution and met the assumptions that explainedthe local environmental conditions of the case study.About 82 % of HS are found at elevations >2 m, andthe micro-topography was found to be strongly corre-lated with both soil pH and EC (r=−0.79 and −0.81;p<0.001, respectively).

Using the new hybrid Gibbs point process model, wedemonstrated that the joint effects of biotic processes atdifferent scales and abiotic heterogeneity could explainthe spatial pattern of the halophyte community.

Soil salinity alone was not responsible for the distri-bution of the halophyte community (Naz et al. 2010). Inaddition, Pan et al. (1998) remarked on the relativeimportance of edaphic factors on the spatial structureof halophytic communities, which is scale dependent(Pan et al. 1998), while biotic interactions play an im-portant role in structuring these communities (Quetset al. 2013).

A medium-sized shrub (2.4 m in diameter) is found.The shrub canopy diameter is related to the presence ofnebkhas size, and these vegetation islands are a functionof the micro-topography (El-Bana et al. 2002) and arehighly correlated with soil salinity (Fig. 9b). There is apositive interaction at 1.4 m between two individuals,indicating a process of facilitation described for hardarid conditions (Bertness and Callaway 1994). As thereis no information on the species identity of each indi-vidual, we cannot interpret if the facilitation is betweenconspecifics or interspecific individuals. The spatialstructure exhibits a random effect at 12 m, indicatingthat these groups of two individuals are independentlydistributed. This structure reveals that the studied com-munity is in transition from aggregates at early stages toa regular pattern due to a self-thinning response.

Acknowledgments Part of this research was supported by Ag-ricultural Research and Development Fund (ARDF) in Egypt, andJ.U-D. was partly funded by German Research Foundation (DFG)research training group BScaling problems in statistics^ (RTG1644). We gratefully acknowledge Google Earth for providinghigh-resolution satellite data. Also, we thank the anonymous re-viewers for their criticism.

References

Abd El-Kawy, O. R., Rød, J. K., Ismail, H. A., & Suliman, A. S.(2011). Land use and land cover change detection in thewestern Nile delta of Egypt using remote sensing data.Applied Geography, 31(2), 483–494.

Akaike, H. (1973). Information theory and an extension of themaximum likelihood principle. In B. N. Petrov & F. Csaki(Eds.), Second international symposium on information the-ory (pp. 267–281). Budapest: Academia Kiado.

Baddeley, A., & Tuner, R. (2000). Practical maximumpseudolikelihood for spatial point patterns. Australian andNew Zealand Journal of Statistics, 42, 283–322.

Baddeley, A. J., & Turner, R. (2006). Modelling spatial pointpatterns in r in case studies in spatial point process model-ling. USA: Springer.

Baddeley, A., Moller, J., & Waagepetersen, R. (2000). Non- andsemiparametric estimation of interaction in inhomogeneouspoint patterns. Statistica Neerlandica, 54, 329–350.

Baddeley, A., Turner, R., Mateu, J., & Bevan, A. H. (2013).Hybrids of Gibbs point process models and their implemen-tation. Journal of Statistical Software, 55(11), 1–43.

Badreldin, N., & Goossens, R. (2013a). Monitoring land use/landcover change using multi-temporal Landsat satellite imagesin an arid environment: a case study of El-Arish, Egypt.Arabian Journal of Geosciences. doi:10.1007/s12517-013-0916-3ui.

Environ Monit Assess (2015) 187:224 Page 13 of 15 224

Author's personal copy

Badreldin, N., & Goossens, R. (2013b). A satellite-based distur-bance index algorithm for monitoring mitigation strategieseffects on desertification change in an arid environment.Mitigation and Adaptation Strategies for Global Change.doi:10.1007/s11027-013-9490-y.

Berman, M., & Turner, T. R. (1992). Approximating point processlikelihoods with GLIM. Journal of the Royal StatisticalSociety, 41, 31–38.

Bertness, M. D., & Callaway, R. (1994). Positive interactions incommunities. Trends in Ecology & Evolution, 9, 191–193.

Blanco-Canqui, H., & Lal, R. (2010). Principles of soil conserva-tion and management. Dordrecht: Springer. doi:10.1007/978-1-4020-8709-7.

Blaschke, T. (2010). Object based image analysis for remotesensing. ISPRS Journal of Photogrammetry and RemoteSensing, 65(1), 2–16.

Coppin, P., Jonckheere, I., Nackaerts, K., Muys, B., & Lambin, E.(2004). Digital change detection methods in ecosystem mon-itoring: a review. International Journal of Remote Sensing,25(9), 1565–1596.

Dale, M. R. T. (1999). Spatial pattern analysis in plant ecology (p.337). Cambridge: Cambridge University Press.

Delicado, P., Giraldo, R., Comas, C., &Mateu, J. (2010). Statisticsfor spatial functional data: some recent contributions.Environmetrics, 21, 224–239.

Diggle, P. J. (2003). Statistical analysis of spatial point patterns(2nd ed.). London: Edward Arnold.

El-Bana, M., Khedr, A., Van Hecke, P., & Bogaert, J. (2002).Vegetation composition of a threatened hypersaline lake(Lake Bardawil), North Sinai. Plant Ecology, 1937, 63–75.

El-Bana, M. I., Li, Z. Q., & Nijs, I. (2007). Role of host identity ineffects of phytogenic mounds on plant assemblages andspecies richness on coastal arid dunes. Journal ofVegetation Science, 18, 635–644.

Fortin, M.-J., & Dale, M. (2009). Spatial autocorrelation in eco-logical studies: a legacy of solutions and myths.Geographical Analysis, 41, 392–397.

Frankl, A., Zwertvaegher, A., Poesen, J., & Nyssen, J. (2013).Transferring Google Earth observations to GIS-software:example from gully erosion study. International Journal ofDigital Earth, 6(2), 196–201.

Galiano, E. F. (1985). The small-scale pattern of Cynodondactylon in Mediterranean pastures. Vegetatio, 63(3), 121–127.

Geyer, C. J. (1999). Likelihood inference for spatial point process-es. In O. E. Barndorff-Nielsen, W. S. Kendall, & M. N. M.van Lieshout (Eds.), Stochastic geometry: likelihood andcomputation (pp. 79–140). London: Chapman andHall/CRC.

Goossens, R., & Van Ranst, E. (1998). The use of remote sensingto map gypsiferous soils in the Ismailia Province (Egypt).Geoderma, 87(1–2), 47–56.

Greenwood, N. H. (1997). The Sinai: a physical geography (p.148). Austin: University of Texas Press.

Greig-Smith, P. (1961). Data on pattern within plant communities: I.The analysis of pattern. Journal of Ecology, 49(3), 695–702.

Illian, J., Penttinen, A., Stoyan, H., & Stoyan, D. (2008). Statisticalanalysis and modelling of spatial point patterns (p. 557).Chichester: Wiley.

Kershaw, K. A. (1959). An investigation of the structure of agrassland community: II. The pattern of Dactylis glomerata.

Lolium Perenne and Trifolium Repens: III. Discussion andconclusions. Journal of Ecology, 47(1), 31–53.

Knorn, J., Rabe, A., Radeloff, V. C., Kuemmerle, T., Kozak, J., &Hostert, P. (2009). Land cover mapping of large areas usingchain classification of neighboring Landsat satellite images.Remote Sensing of Environment, 113(5), 957–964.

Kottek, M., Grieser, J., Beck, C., Rudolf, B., & Rubel, F. (2006).World Map of the Köppen-Geiger climate classification up-dated.Meteorologische Zeitschrift, 15(3), 259–263.

Liu, J. G., & Mason, P. J. (2009). Essential image processing andGIS for remote sensing (p. 462). Chichester: Wiley.

Møller, J., & Waagepetersen, R. (2003). Statistical inference andsimulation for spatial point processes. Chapman and Hall/CRC Press.

Natural Resources Conservation Service (NRCS). (1999). Soiltaxonomy: a basic system of soil classification for makingand interpreting soil surveys (Second Edi., p. 871).Washington, DC: United States Department of Agriculture(USDA).

Naz, N., Hameed, M., Ahmad, S. A., Ashraf, M., & Arshad, M.(2010). Is soil salinity one of the major determinants ofcommunity structure under arid environments? CommunityEcology, 11, 84–90.

Pan, D., Bouchard, A., Legendre, P., & Domon, G. (1998).Influence of edaphic factors on the spatial structure of inlandhalophytic communities: a case study in China. Journal ofVegetation Science, 9, 797–804.

Peel, M. C., Finlayson, B. L., & Mcmahon, T. A. (2007). Updatedworld map of the Köppen-Geiger climate classification.Hydrology and Earth System Sciences, 11, 1633–1644.

Perry, G. L. W., Miller, B. P., & Enright, N. J. (2006). A compar-ison of methods for the statistical analysis of spatial pointpatterns in plant ecology. Plant Ecololgy, 187, 59–82.

Petrou, M., & Bosdogianni, P. (1999). Image processing: thefundamentals (p. 347). Chichester: Wiley.

Quets, J., Temmerman, S., El-Bana, M. I., Al-Rowaily, S.,Assaeed, A. M., & Nijs, I. (2013). Unravelling landscapeswith phytogenic mounds (nebkhas): an exploration of spatialpattern. Acta Oecologica, 49, 53–63.

Rhoades, J. D. (1992). Instrumental field methods of salinityappraisal. Advances in measurement of soil physical proper-ties: bringing theory into practice (Vol. SSSA, pp. 231–248).Soil Science Society of America.

Rubio-Casal, A. E., Castillo, J. M., Luque, C. J., & Figueroa, M. E.(2001). Nucleation and facilitation in salt pans in Mediterraneansalt marshes. Journal of Vegetation Science, 12(6), 761–770.

Sadr, K., & Rodier, X. (2012). Google Earth, GIS and stone-walled structures in southern Gauteng, South Africa.Journal of Archaeological Science, 39(4), 1034–1042.

U.S. Salinity Laboratory. (1954). Diagnosis and improvement ofsaline and alkali soils (Handbook 6) (p. 172). Washington,DC: U.S. Government Printing Office.

Schowengerdt, R. A. (2007). Remote sensing: models andmethods for image processing (3rd ed., p. 558).Amsterdam: Elsevier.

Shalaby, A., & Tateishi, R. (2007). Remote sensing and GIS formapping and monitoring land cover and land-use changes inthe Northwestern coastal zone of Egypt. Applied Geography,27(1), 28–41.

Sheppard, S. R. J., & Cizek, P. (2009). The ethics of Google Earth:crossing thresholds from spatial data to landscape

224 Page 14 of 15 Environ Monit Assess (2015) 187:224

Author's personal copy

visualisation. Journal of Environmental Management, 90(6),2102–2117.

Shih, F. Y. (2010). Image processing and pattern recognition:fundamentals and techniques (p. 552). Hoboken: Wiley.

Taylor, J. R., & Lovell, S. T. (2012). Mapping public and privatespaces of urban agriculture in Chicago through the analysis ofhigh-resolution aerial images in Google Earth. Landscapeand Urban Planning, 108(1), 57–70.

Ungar, I. A., Benner, D. K., & McGraw, D. C. (1977). Thedistribution and growth of Salicornia europaea on an inlandsalt pan. Ecology, 60(2), 329–336.

Usher, M. B. (1983). Pattern in the simple moss-turf communitiesof the Sub-Antarctic and Maritime Antarctic. Journal ofEcology, 71(3), 945–958.

Vincent, L., & Soille, P. (1991). Watersheds in digital spaces: anefficient algorithm based on immersion simulations. IEEETransactions on Pattern Analysis and Machine Intelligence,13(6), 583–598.

Webster, R., & Oliver, M. A. (2007). Geostatistics for environ-mental scientists (2nd ed., p. 333). Chichester: Wiley.

Winkler, G. (2006). Image analysis, random fields and Markovchain Monte Carlo methods: a mathematical introduction(stochastic modelling and applied probability). Retrievedfrom http://dl.acm.org/citation.cfm?id=1206577.

Yang, X., Jiang, G.-M., Luo, X., & Zheng, Z. (2012). Preliminarymapping of high-resolution rural population distributionbased on imagery from Google Earth: a case study in theLake Tai basin, eastern China. Applied Geography, 32(2),221–227.

Environ Monit Assess (2015) 187:224 Page 15 of 15 224

Author's personal copy