MULTIVARIANTE ANALYSES OF VEGETATION IN THE SALINE PLAIN OF LOWER-CHELIFF, ALGERIA تحليلات التغيرات المتعددة للنباتاتِ في السهلِ الملحيِ لمنخفض الشليف ، الجزائر

MULTIVARIANTE ANALYSES OF VEGETATION IN THE SALINE PLAIN OF LOWER-CHELIFF, ALGERIA

Ababou A.^*, Chouieb, M.^**,Khader, M.^***,
Mederbal, K.^*** and Mahmoud A. Essa^****

* Dept. of Biology, Faculty of Biology and Agronomy, University Hassiba Ben Bouali, Chlef, Algeria

E-mail: ab_adda@yahoo.fr

**Dept. Agronomy, Faculty of Engineer Science, University Abd El Hamid Ibn Badis, Mostaganem, Algeria

***Dept. of Biology, Faculty  of Science and Sciences of the Ground, University Mustapha Stambouli, Mascara, Algeria.

**** Dept. of Geology, Faculty of Science, Assiut University, Assiut, Egypt

INTRODUCTION:

        The vegetal association is "a plant community characterized by definite floristic and sociological features” which shows, by the presence of diagnostic species “a certain independence” (Braun Blanquet, 1928) and grows in uniform habitat conditions (Flahault & Schroter, 1910). These plant communities are recognized by diagnostic species as defined by Westhoff and Van der Maarel (1978). In this context, the term diagnostic species is an important concept in the vegetation classification. It is a plant of a high fidelity to a particular community which it serves as a criterion of the recognition of that community (Curtis, 1959). The relative constancy or abundance of the vegetation classification distinguishes one association from another (Whittaker, 1962; Chytry & Tichy, 2003). Vegetal association includes species which preferably occurs in a single vegetation unit (character species) or in a few vegetation units (differential species) (Chytry et al., 2002). The presence, abundance, or vigor of some species indicates certain site conditions (Gabriel & Talbot, 1984). To relate the species data to the site conditions, the direct gradient analysis is the most prominent method, where either redundancy analysis (RDA) or canonical correspondence analysis (CCA) could be applied (Leps & Smilauer, 2003; Zuur et al., 2007). Also statistical fidelity measures (Chytry et al., 2002) such as phi coefficient can be employed to characterize the floristic composition of that site independently of the environmental conditions. In this context, our study proposes to analyze the vegetation assemblage and community structure in the disturbed site of lower-Cheliff, Algeria, in relation and independently to the site conditions.

MATERIALS AND METHODS:

Study Area:

                Covering approximately 450 km², the lower-Cheliff (Fig. 1) is one of the largest Quaternary alluvial plains of the Northwestern Algeria. This plain, located between 35.750°-36.125° N of latitude and 0.5°-1°E of longitude, is about 35 km far from the Mediterranean Sea with an average altitude of 70 m. It is syncline framed in north by Dahra’s hills and Benziane’s hills in the south that both composed of clayey silt, schist and saline marls. These geological characteristics are accentuated by a semi-arid climate with an average annual temperature of 20°C and a weak annual pluviometry (approximately 250 mm/yr) that explain the high salinity of the plain.

Soil and Vegetation Sampling:

        Vegetation relevés were recorded during spring 2006, 2007 and 2008 (March 21st - May 21st) by using the Braun-Blanquet (Maarel Van der, 1975) seven-degree scale of abundance dominance. A total of 133 relevés were recorded with adding up 40 species, among them 11 rare species that were excluded from the analysis. Also a total of 133 soil samples were collected at a depth of 30 cm. The measured soil properties were of physical (granulometry, altitude, soil structure and soil color) and chemical nature (electrical conductivity, CaCO₃, pH, Ca⁺⁺, Na⁺, Cl^-, organic matter and CaMg).In order to use geostatistical analysis, the geographical position of each site was determined by using GPS.

Data Analysis:

        Initially, a colinearity test (Appendix 1) performed between environmental variables showed a strong correlation coefficient (R>0.9) between sand and silt as well as Na⁺ and Cl^-. Also, Cl^- and silt were eliminated. Then, the remaining variables were subject to a normality test (Shapiro-Wilk); those with non normal distribution were log transformed. For the determination of the most significant variables, an individual preselection (Okland & Eilertsen, 1994) was performed by using Monte Carlo test (999 permutations without restriction); except sand, they were all significant (p-value<0.05) with variance inflation factors (Erkel-Rousse, 1995; Besse, 2001) < 4 (Appendix 2).

        In order to establish the main links between environmental variables and vegetation assemblage, a redundancy analysis (RDA) (Ter Braak, 1994; Legendre & Legendre, 1998; Leps & Smilauer, 2003) was performed. First, a detrended correspondence analysis (DCA) (Hill & Gauch, 1980) was conducted to test if a model with unimodal (CCA) (Ter Braak, 1986) or linear (RDA) response curve should be used in ordination analysis. Results of the DCA showed that the gradient length was 3.95 for axis one to 2.53 for axis four (Table1); thus, both RDA and CCA may give correct results (Leps & Smilauer, 2003; Jongman et al., 1996). As the percentage of the total variance explained by RDA (21%) that was higher than CCA (17.2%), it was appropriately considered to perform an RDA, as linear responses. However, the presence of double zeros strongly affects the RDA with another potential problem due to the arch effect (Zuur et al., 2007). An alternative is to apply either chord (Orloci, 1967) or Hellinger (Rao, 1995) distance transformation. Legendre and Gallagher (2001) showed that this approach is less sensitive to double zeros and, therefore, to the arch effect. After several comparisons, Hellinger transformation was chosen followed by the RDA.

        The most significant variables were determined by using the method of Wilk's lambda (Butler & Wood, 2004;Marques de Sa, 2007).

        To cluster samples into similar groups and to identify the characteristic vegetation unit of each group, relevés were separated in groups, according to their contribution and coordinates according to the first two canonical axes obtained by RDA. Finally, four predefined groups were used because they showed major ecological relevance and were easily interpretable. Then, the phi coefficient of association was used (Sokal & Rohlf, 1995; Chytry et al., 2002) to identify species discriminating among the four groups. This coefficient is a statistical measure of association which can be used as a measure of fidelity. It can be defined as:

        The same notation as that used by Bruelheide, (2000) and Chytrý et al., (2002) is used in this study:

N = number of relevés in the data set;

N_p=number of relevés in the particular vegetation unit;

n= number of occurrences of the species in the data set;

n_p= number of occurrences of the species in the particular vegetation unit.

Fig. (1): Location of the study area, showing the Lower Cheliff plain and the surrounding hills of Dahra in the north and Benziane in the south

        Traditionally, the phi coefficient considers only the presence/absence of vegetation, so that fidelity values calculated using this coefficient are not influenced by species cover or abundance. The advantage of phi coefficient is its independence of data set size. The phi coefficient values range from –1 to 1. The highest phi value of 1 is achieved if the species occurs in all relevés of the vegetation unit and is absent elsewhere. A positive value that lowers than 1 means that the species is absent from some relevés of the vegetation unit. A value of 0 indicates no relation between the target species and the target vegetation unit.

        Finally in order to establish the relation between vegetation and salinity, a cartography is carried out using Kriging (Krige, 1951; Matheron, 1963;Journel & Huijbregts, 1978; Legendre & Legendre 1998; Stein et al., 2002).

RESULTS AND DISCUSSION:

Important Variables:

        The marginal effects (Appendix 3) indicate that altitude, conductivity and soil structure are best explanatory variables, followed by Ca⁺⁺, pH, and Na⁺, whereas the remaining variables play a secondary role. The highly significant (p < 0.01) increases in the total sum of eigenvalues during the forward selection indicated by the conditional effects (Appendix 4) are successively shown by altitude, conductivity, Na⁺ and soil structure according to Monte Carlo test (999 permutations). Calcium carbonate (CaCO₃), Ca⁺⁺ and CaMg point out significant increases (p<0.05), whereas the contribution of the remaining variables are non significant.

Best Predictors of Vegetal Distribution:

        The variance of species data explained by each variable according to the partial RDA is in the order of altitude (8.8%)>conductivity (7.7%)>soil structure (6.5%)>Ca⁺⁺ (4.6%)> pH (4.5%)> Na⁺ (3.5%)> Clay (2.3%)> OM (1.7%) > CaCO₃ (1.4%) = RGB (1.4%)> CaMg (1.2%) > Sand (0.9%). This means that vegetation distribution of Lower-Cheliff is highly related to altitude, conductivity and soil structure.

Redundancy Analysis:

        The total variance of species data explained by the first four axes of the RDA is 21.0% (Table 1). Although the Monte Carlo permute-tion test indicates that all canonical axes are highly significant with the set of variables used, only the first two canonical axes are used because they include the maximum variability expressed by the environmental variables. Almost all variables that are significant on axes 3 and 4 are also significant on axes 1 and 2. The first axis (13.6% of the variation explained) is mainly negatively correlated to the conductivity then to Ca⁺⁺, Na⁺ and clay, while itis positively correlated mainly to altitude (Fig. 2), then soil structure, pH, organic matter and CaCO₃. This means that sampling sites situated to the right of the first axis are characterized by higher altitudes, and lower conductivity values. On the left, the sampling sites are those with higher conductivity values. Thus, axis 1 shows a gradient of decreasing altitudes and increasing conductivity values. This axis could be interpreted as a conductivity environmental gradient. So, the relative high altitude observed in the lower-Chelif has a considerable effect on the spatial distribution of the salinity and consequently on the vegetation. Low altitudes are accompanied by salt deposits worsened by a high percentage of clay which prevents salt drainage. This process leads to the degradation of soil structure and high conductivity with appearance of halophytes species such as Chenopodiaceae and Caryophyllaceae that characterize the extreme conditions (Fig. 2). Relative high altitudes are characterized by healthy soils with low conductivity and a slightly high organic matter level, which improve the soil structure and appear to be more diversified floristic composition (Asteraceae, Fabaceae, Géraniaceae, Apiaceae, Brassicaceae, Primulaceae, Plantaginaceae). The second axis (only 4.2% of the variation explained) is negatively correlated to RGB, pH and sand, but positively correlated mainly to the conductivity and Na⁺, with a notable appearance of Chenopodiaceae and Caryophyllaceae.

Table (1): Eigenvalues and percentage of variance explained by RDA, with Pearson correlations (r) between environmental variables and the four canonical axes, and results from the Monte Carlo test checking for axis significance in RDA

** p < 0.01; * p<0.05

Fig. (2): Constrained ordination biplot (RDA) of vegetation, 133 sites, 29 species and 12 environmental variables selected using foward selection and Monte Carlo permutation test. For species names see code tab.4, letters between bracket represent the families (Ch: Chenopodiaceae, Ap: Apiaceae, As: Astéraceae, Br: Bromeliaceae, Ca: Caryophyllaceae, Fa: Fabaceae, Ge: Geraniaceae, Pl: Plantaginaceae, Po: Poaceae et Pr: Primulaceae)

Vegetation Units According to Fidelity Coefficient:

        The two first canonical axes are used in the classification of sites cluster to all samples, because they include the maximum variability expressed by environmental variables. As a result, four distinct groups of sites with similar floristic composition were identified. Group A is composed of 43 sites. These sites are differentiated by the presence of 6 diagnostic species, that on the basis of phi coefficient results are Plantago coronopus (L), Bellis perennis (L), Hypochaeris glabra (L), Phalaris arundinacea (L), Spergula sp (L)and Beta vulgaris maritima (L)(Table 2). The diagnostic species that are present in this group are generally associated with the lowest altitudes and saline soilsas indicated by the normal distribution (Fig. 3).Group B includes 25 sites. This group is characterized by four diagnostic species of Torilis nodosa (L), Erodium cicutarium (L), Onopordum acanthium(L) and Lolium multiflorum (Lam). The diagnostic species of this group are associated with the highest altitudes and lowest conductivity (Fig. 4).Group C with the highest diversity according to Shanon-Weiner index (Table 2) is composed of 30 sites that are slightly saline (Fig. 5) with intermediate altitudes and characterized by eight diagnostic species such as Sinapsis arvensis (L), Plantago lanceolata (L)andScolymus hispanicus (L). Finally, group D includes 35 sites of highly saline soils (Fig. 6). The five diagnostic species of this group exclusively belong to Chenopodiaceae and Caryophyllaceae such as Suaeda maritima (L) Dumort, Arthrocnemum macrostachyum Moric, Suaeda fruticosa (L) Forssk, Atriplex halimus (L), Spergularia marina (L) Griseb. Group A and D are the less diverse groups according to Shanon-Weiner index (Table 2); indeed, a great number of species sensitive to high salinity is eliminated with the exclusive appearance of halophytes.

Table (2): Synoptic table of 133 relevés and 29 species, based on fidelity coefficient. Diagnostic species (values grey-shaded) are those with significant phi value according to the test of Fisher ranked by decreasing value of Φ

*** p<0.001; ** p<0.01; *p<0.05

Electrical Conductivity:

        All theoretical variograms used to estimate the electrical conductivity show a high nugget
effect, high variance and a strong variation coefficient (Table 3). The electrical conductivity is consequently unpredictable and slightly structured variable, with a strong variability even at very short distance. This supposes a high density of sampling to predict the electrical conductivity.

        On the basis of variographic analysis (Table 3 and Fig. 7) and cross validation (R²=0.73 and S.E.E = 4 dS/m) (Fig. 8), the exponential model is the best theoretical one to estimate the electrical conductivity (Fig. 9).

The conductivity map (Fig. 9) first shows an increasing conductivity from east to west and from the periphery to the center of the plain. This corresponds to a decreasing altitude from east to west and from the periphery to the center. Secondl, a very large surface of the lower-cheliff plain belongs to the range of non-saline to highly saline soils.

Fig. (7): Exponential model

Fig. (8): Cross validation of electrical conductivity estimated by exponential model

Fig. (9): Electrical conductivity map of the Lower-Cheliff obtained by kriging and estimated by the exponential model

Table (3): Summary of theoretical variograms

Spatial Distribution of Vegetation Units According to Conductivity:

        Transformations of species abundance from Braun Blanquet scale to 0-9 Van der Maarel scale enable us to cartography the spatial distribution of the vegetation units by using kriging. As a result, Figures 10, 11, 12 and 13 show a close relation between the vegetation units and the electrical conductivity values. Vegetation unit A (Fig. 10) is distributed throughout the saline soils according to the same gradient of salinity, from the periphery to the center and from the west to the east. Vegetation unit D (Fig. 13) occupies the highly saline soils, whereas vegetation unit B (Fig. 11) is completely absent from the western side of the plain that is characterized by saline soils and low altitudes. Vegetation unit C (Fig. 12) is especially localized at the periphery of the plain and characterized by slightly saline soils.

Fig. (10): Distribution map of vegetation unit A

Fig. (11): Distribution map of vegetation unit B

Fig. (12): Distribution map of vegetation unit C

Fig. (13): Distribution map of vegetation unit D

CONCLUSION:

        The plain of lower-Chelif is an ecosystem weakened by particular edaphic constraints and hard climatic adversities. These constraints strongly reduce the vegetal diversity. Thus, during three years of this study, we counted only 40 species through 133 relevés.

        With an aim of reasonable management of this ecosystem, traditional methods of evaluation of site conditions are expensive and time consuming, especially in areas as large as the lower-Cheliff. Thus, recognition of vegetation ecology and biology is the easiest way limiting cost and time to assess environmental conditions. The present study provides baseline information about the environmental variables affecting the distribution patterns of vegetation assemblages in one of the largest plain in Algeria. It is clear that understanding the local scale factors is needed to assess the importance of factors structuring vegetal communities. The key factors that determine the distribution of vegetation in lower-Cheliff are altitude, electrical conductivity, sodium and soil structure. However, in this study, one of the most important factors that influence the composition of vegetation assemblages is the electrical conductivity. Indeed, it was able for us to differentiate among the four vegetation units in relation to the influence of the main environmental variable. Less diverse vegetation units composed of halophytes species and distributed throughout the saline soils as well as more diverse vegetation units that are very sensitive to salinity occupying the non saline to slightly saline soils were distinguished. Thus, the assessment of vegetal communities is a useful tool to classify salinity, especially in terms of revealing the spatiotemporal changes of this variable. It would be interesting to compare these results with other vegetation studies at the same conditions.

REFERENCES:

Besse, P. (2001): Pratique de la modélisation Statistique, Publications du laboratoire de statistique et probabilité, université Paul Sabatier, Toulouse III, France

Bruelheide, H. (2000): A new measure of fidelity and its application to defining species groups. J. Veg. Sci., 11: 167-178

Braun Blanquet, J. (1928): Pflanzensoziologie. Gründzuge der Vegetationskunde. Springer-Verlag, Berlin, Germany. 631 p.

Butler, R.W. & Wood, A.T.A. (2004): A Dimensional CLT for Non-Central Wilks’ Lambda in Multivariate Analysis, Scandinavian Journal of Statistics, Volume 31, Number 4, December 2004, pp. 585-601(17).

Chytry, M., Tichy, L., Holt, J. & Botta-Dukat, Z. (2002): Determination of diagnostic species with statistical fidelity measures J. Veg. Sci. 13: 79-90.

Chytry, M. & Tichy, L. (2003): Diagnostic, constant and dominant species of vegetation classes and alliances of the Czech Republic: a statistical revision. Folia Fac. Sci. Nat. Univ. Masaryk. Brun., Biol., 108: 1–23.

Curtis, J.T. (1959): The vegetation of Wisconsin: an ordination of plant communities. University of Wisconsin Press, Madison, Wisconsin, USA. 657p.

Erkel-Rousse, H. (1995): Détection de la multicolinéarité dans un modèle linéaire ordinaire : quelques éléments pour un usage averti des indicateurs de Belsley , Kuh et Welsh, revue de statistique appliquée, tome 43, n°4, p. 19-42.

Flahault, C. & Schroter, C. (1910): Rapport sur la nomenclature phytogeographique. Proceedings of the 3^rd. International Botanical Congress, Brussels, 1:131–164.

Gabriel, H.W. & Talbot, S.S. (1984): Glossary of landscape and vegetation ecology for Alaska. Alaska Technical Report 10. Bureau of Land Management, U.S. Dept. of the Interior, Washington, D.C.

Hill, M.O. & Gauch, H.G. (1980): Detrended correspondence analysis, an improvised ordination technique. Vegetatio, 42, 47-58.

Jongman, R.H.G., Ter Braak, C.J.F. & Van Tongeren, O.F.R. (1996): Data analysis in community and landscape ecology. Cambridge University Press. Cambridge. 299 pp.

Journel, A.G. & Huijbregts, C.J. (1978): Mining Geostatistics. Academic Press, 600 p.

Krige, D.G. (1951): A statistical approach to some basic mine valuation problems on the Witwatersrand. 1951. J. of Chem., Metal and Mining Soc. of South Africa, 52, 119-139.

Legendre, P. & Legendre, L. (1998): Numerical Ecology (2nd English edn). Amsterdam, The Netherlands: Elsevier, 853 pp.

Legendre, P. & Gallagher, E. (2001): Ecologically meaningful transformation for ordination of species data. Oecologia 129: 271-280

Leps, J. & Smilauer, P. (2003): Multivariate analysis of ecological data using CANOCO. Cambridge, UK: Cambridge University Press. 269 pp.

Maarel Van der, E. (1975): The Braun-Blanquet approach in perspective. Vegetatio, 30 (3), 213-219.

Marques de Sa, J.P. (2007): Applied Statistics Using SPSS, Statistica, Matlab and R. Springer-Verlag Berlin Heidelberg.

Matheron, G. (1963): Principles of Geostatistics. Economic Geol., 58, 1246-1268.

Okland, R.H. & Eilertsen, O. (1994): Canonical correspondence-analysis with variation partitioning-some comments and an application. Journal of Vegetation Science, 5, 117-126.

Orloci, L. (1967): An agglomerative method for classification of plant communities. J. Ecol 55:193–205.

Rao, C.R. (1995): A review of canonical coordinates and an alternative to correspondence analysis using Hellinger distance. Qüestiió 19:23–63.

Sokal, R.R. & Rohlf, F.J. (1995): Biometry the Principles and Practice of Statistics in Biological Research. Third Edition. W. H. Freeman and Company, New York.

Stein, A., Van Der Meer, F. & Gorte, B. (2002): Spatial statistics for remote sensing, kluwer academic publishers, 301 p.

Ter Braak, C.J.F. (1986): Canonical corres-pondence analysis: a new eigenvector technique for multivariate direct gradient analysis. Ecol 67:1167-1179.

Ter Braak, C.J.F. (1994): Canonical community ordination. Part I: Basic theory and linear methods. Ecosc. 1:127-140.

Westhoff, V. & Van Der Maarel, E. (1978): The Braun-Blanquet approach. In: Whittaker, R.H. (editor). Classification of plant communities, pp. 287-399. Dr. W. Junk,  The Hague, The Netherlands.

Whittaker, R.H. 1962. Classification of natural communities. Bot. Rev. 28: 1-239.

Zuur, A.K., Ieno, E.N. & Smith, G.M. (2007): Analysing Ecological Data. Springer, USA.

MULTIVARIANTE ANALYSES OF VEGETATION IN THE SALINE PLAIN OF LOWER-CHELIFF, ALGERIA

Ababou A.^*, Chouieb, M.^**,Khader, M.^***,
Mederbal, K.^*** and Mahmoud A. Essa^****

* Dept. of Biology, Faculty of Biology and Agronomy, University Hassiba Ben Bouali, Chlef, Algeria

E-mail: ab_adda@yahoo.fr

**Dept. Agronomy, Faculty of Engineer Science, University Abd El Hamid Ibn Badis, Mostaganem, Algeria

***Dept. of Biology, Faculty  of Science and Sciences of the Ground, University Mustapha Stambouli, Mascara, Algeria.

**** Dept. of Geology, Faculty of Science, Assiut University, Assiut, Egypt

INTRODUCTION:

        The vegetal association is "a plant community characterized by definite floristic and sociological features” which shows, by the presence of diagnostic species “a certain independence” (Braun Blanquet, 1928) and grows in uniform habitat conditions (Flahault & Schroter, 1910). These plant communities are recognized by diagnostic species as defined by Westhoff and Van der Maarel (1978). In this context, the term diagnostic species is an important concept in the vegetation classification. It is a plant of a high fidelity to a particular community which it serves as a criterion of the recognition of that community (Curtis, 1959). The relative constancy or abundance of the vegetation classification distinguishes one association from another (Whittaker, 1962; Chytry & Tichy, 2003). Vegetal association includes species which preferably occurs in a single vegetation unit (character species) or in a few vegetation units (differential species) (Chytry et al., 2002). The presence, abundance, or vigor of some species indicates certain site conditions (Gabriel & Talbot, 1984). To relate the species data to the site conditions, the direct gradient analysis is the most prominent method, where either redundancy analysis (RDA) or canonical correspondence analysis (CCA) could be applied (Leps & Smilauer, 2003; Zuur et al., 2007). Also statistical fidelity measures (Chytry et al., 2002) such as phi coefficient can be employed to characterize the floristic composition of that site independently of the environmental conditions. In this context, our study proposes to analyze the vegetation assemblage and community structure in the disturbed site of lower-Cheliff, Algeria, in relation and independently to the site conditions.

MATERIALS AND METHODS:

Study Area:

                Covering approximately 450 km², the lower-Cheliff (Fig. 1) is one of the largest Quaternary alluvial plains of the Northwestern Algeria. This plain, located between 35.750°-36.125° N of latitude and 0.5°-1°E of longitude, is about 35 km far from the Mediterranean Sea with an average altitude of 70 m. It is syncline framed in north by Dahra’s hills and Benziane’s hills in the south that both composed of clayey silt, schist and saline marls. These geological characteristics are accentuated by a semi-arid climate with an average annual temperature of 20°C and a weak annual pluviometry (approximately 250 mm/yr) that explain the high salinity of the plain.

Soil and Vegetation Sampling:

        Vegetation relevés were recorded during spring 2006, 2007 and 2008 (March 21st - May 21st) by using the Braun-Blanquet (Maarel Van der, 1975) seven-degree scale of abundance dominance. A total of 133 relevés were recorded with adding up 40 species, among them 11 rare species that were excluded from the analysis. Also a total of 133 soil samples were collected at a depth of 30 cm. The measured soil properties were of physical (granulometry, altitude, soil structure and soil color) and chemical nature (electrical conductivity, CaCO₃, pH, Ca⁺⁺, Na⁺, Cl^-, organic matter and CaMg).In order to use geostatistical analysis, the geographical position of each site was determined by using GPS.

Data Analysis:

        Initially, a colinearity test (Appendix 1) performed between environmental variables showed a strong correlation coefficient (R>0.9) between sand and silt as well as Na⁺ and Cl^-. Also, Cl^- and silt were eliminated. Then, the remaining variables were subject to a normality test (Shapiro-Wilk); those with non normal distribution were log transformed. For the determination of the most significant variables, an individual preselection (Okland & Eilertsen, 1994) was performed by using Monte Carlo test (999 permutations without restriction); except sand, they were all significant (p-value<0.05) with variance inflation factors (Erkel-Rousse, 1995; Besse, 2001) < 4 (Appendix 2).

        In order to establish the main links between environmental variables and vegetation assemblage, a redundancy analysis (RDA) (Ter Braak, 1994; Legendre & Legendre, 1998; Leps & Smilauer, 2003) was performed. First, a detrended correspondence analysis (DCA) (Hill & Gauch, 1980) was conducted to test if a model with unimodal (CCA) (Ter Braak, 1986) or linear (RDA) response curve should be used in ordination analysis. Results of the DCA showed that the gradient length was 3.95 for axis one to 2.53 for axis four (Table1); thus, both RDA and CCA may give correct results (Leps & Smilauer, 2003; Jongman et al., 1996). As the percentage of the total variance explained by RDA (21%) that was higher than CCA (17.2%), it was appropriately considered to perform an RDA, as linear responses. However, the presence of double zeros strongly affects the RDA with another potential problem due to the arch effect (Zuur et al., 2007). An alternative is to apply either chord (Orloci, 1967) or Hellinger (Rao, 1995) distance transformation. Legendre and Gallagher (2001) showed that this approach is less sensitive to double zeros and, therefore, to the arch effect. After several comparisons, Hellinger transformation was chosen followed by the RDA.

        The most significant variables were determined by using the method of Wilk's lambda (Butler & Wood, 2004;Marques de Sa, 2007).

        To cluster samples into similar groups and to identify the characteristic vegetation unit of each group, relevés were separated in groups, according to their contribution and coordinates according to the first two canonical axes obtained by RDA. Finally, four predefined groups were used because they showed major ecological relevance and were easily interpretable. Then, the phi coefficient of association was used (Sokal & Rohlf, 1995; Chytry et al., 2002) to identify species discriminating among the four groups. This coefficient is a statistical measure of association which can be used as a measure of fidelity. It can be defined as:

        The same notation as that used by Bruelheide, (2000) and Chytrý et al., (2002) is used in this study:

N = number of relevés in the data set;

N_p=number of relevés in the particular vegetation unit;

n= number of occurrences of the species in the data set;

n_p= number of occurrences of the species in the particular vegetation unit.

Fig. (1): Location of the study area, showing the Lower Cheliff plain and the surrounding hills of Dahra in the north and Benziane in the south

        Traditionally, the phi coefficient considers only the presence/absence of vegetation, so that fidelity values calculated using this coefficient are not influenced by species cover or abundance. The advantage of phi coefficient is its independence of data set size. The phi coefficient values range from –1 to 1. The highest phi value of 1 is achieved if the species occurs in all relevés of the vegetation unit and is absent elsewhere. A positive value that lowers than 1 means that the species is absent from some relevés of the vegetation unit. A value of 0 indicates no relation between the target species and the target vegetation unit.

        Finally in order to establish the relation between vegetation and salinity, a cartography is carried out using Kriging (Krige, 1951; Matheron, 1963;Journel & Huijbregts, 1978; Legendre & Legendre 1998; Stein et al., 2002).

RESULTS AND DISCUSSION:

Important Variables:

        The marginal effects (Appendix 3) indicate that altitude, conductivity and soil structure are best explanatory variables, followed by Ca⁺⁺, pH, and Na⁺, whereas the remaining variables play a secondary role. The highly significant (p < 0.01) increases in the total sum of eigenvalues during the forward selection indicated by the conditional effects (Appendix 4) are successively shown by altitude, conductivity, Na⁺ and soil structure according to Monte Carlo test (999 permutations). Calcium carbonate (CaCO₃), Ca⁺⁺ and CaMg point out significant increases (p<0.05), whereas the contribution of the remaining variables are non significant.

Best Predictors of Vegetal Distribution:

        The variance of species data explained by each variable according to the partial RDA is in the order of altitude (8.8%)>conductivity (7.7%)>soil structure (6.5%)>Ca⁺⁺ (4.6%)> pH (4.5%)> Na⁺ (3.5%)> Clay (2.3%)> OM (1.7%) > CaCO₃ (1.4%) = RGB (1.4%)> CaMg (1.2%) > Sand (0.9%). This means that vegetation distribution of Lower-Cheliff is highly related to altitude, conductivity and soil structure.

Redundancy Analysis:

        The total variance of species data explained by the first four axes of the RDA is 21.0% (Table 1). Although the Monte Carlo permute-tion test indicates that all canonical axes are highly significant with the set of variables used, only the first two canonical axes are used because they include the maximum variability expressed by the environmental variables. Almost all variables that are significant on axes 3 and 4 are also significant on axes 1 and 2. The first axis (13.6% of the variation explained) is mainly negatively correlated to the conductivity then to Ca⁺⁺, Na⁺ and clay, while itis positively correlated mainly to altitude (Fig. 2), then soil structure, pH, organic matter and CaCO₃. This means that sampling sites situated to the right of the first axis are characterized by higher altitudes, and lower conductivity values. On the left, the sampling sites are those with higher conductivity values. Thus, axis 1 shows a gradient of decreasing altitudes and increasing conductivity values. This axis could be interpreted as a conductivity environmental gradient. So, the relative high altitude observed in the lower-Chelif has a considerable effect on the spatial distribution of the salinity and consequently on the vegetation. Low altitudes are accompanied by salt deposits worsened by a high percentage of clay which prevents salt drainage. This process leads to the degradation of soil structure and high conductivity with appearance of halophytes species such as Chenopodiaceae and Caryophyllaceae that characterize the extreme conditions (Fig. 2). Relative high altitudes are characterized by healthy soils with low conductivity and a slightly high organic matter level, which improve the soil structure and appear to be more diversified floristic composition (Asteraceae, Fabaceae, Géraniaceae, Apiaceae, Brassicaceae, Primulaceae, Plantaginaceae). The second axis (only 4.2% of the variation explained) is negatively correlated to RGB, pH and sand, but positively correlated mainly to the conductivity and Na⁺, with a notable appearance of Chenopodiaceae and Caryophyllaceae.

Table (1): Eigenvalues and percentage of variance explained by RDA, with Pearson correlations (r) between environmental variables and the four canonical axes, and results from the Monte Carlo test checking for axis significance in RDA

** p < 0.01; * p<0.05

Fig. (2): Constrained ordination biplot (RDA) of vegetation, 133 sites, 29 species and 12 environmental variables selected using foward selection and Monte Carlo permutation test. For species names see code tab.4, letters between bracket represent the families (Ch: Chenopodiaceae, Ap: Apiaceae, As: Astéraceae, Br: Bromeliaceae, Ca: Caryophyllaceae, Fa: Fabaceae, Ge: Geraniaceae, Pl: Plantaginaceae, Po: Poaceae et Pr: Primulaceae)

Vegetation Units According to Fidelity Coefficient:

        The two first canonical axes are used in the classification of sites cluster to all samples, because they include the maximum variability expressed by environmental variables. As a result, four distinct groups of sites with similar floristic composition were identified. Group A is composed of 43 sites. These sites are differentiated by the presence of 6 diagnostic species, that on the basis of phi coefficient results are Plantago coronopus (L), Bellis perennis (L), Hypochaeris glabra (L), Phalaris arundinacea (L), Spergula sp (L)and Beta vulgaris maritima (L)(Table 2). The diagnostic species that are present in this group are generally associated with the lowest altitudes and saline soilsas indicated by the normal distribution (Fig. 3).Group B includes 25 sites. This group is characterized by four diagnostic species of Torilis nodosa (L), Erodium cicutarium (L), Onopordum acanthium(L) and Lolium multiflorum (Lam). The diagnostic species of this group are associated with the highest altitudes and lowest conductivity (Fig. 4).Group C with the highest diversity according to Shanon-Weiner index (Table 2) is composed of 30 sites that are slightly saline (Fig. 5) with intermediate altitudes and characterized by eight diagnostic species such as Sinapsis arvensis (L), Plantago lanceolata (L)andScolymus hispanicus (L). Finally, group D includes 35 sites of highly saline soils (Fig. 6). The five diagnostic species of this group exclusively belong to Chenopodiaceae and Caryophyllaceae such as Suaeda maritima (L) Dumort, Arthrocnemum macrostachyum Moric, Suaeda fruticosa (L) Forssk, Atriplex halimus (L), Spergularia marina (L) Griseb. Group A and D are the less diverse groups according to Shanon-Weiner index (Table 2); indeed, a great number of species sensitive to high salinity is eliminated with the exclusive appearance of halophytes.

Table (2): Synoptic table of 133 relevés and 29 species, based on fidelity coefficient. Diagnostic species (values grey-shaded) are those with significant phi value according to the test of Fisher ranked by decreasing value of Φ

*** p<0.001; ** p<0.01; *p<0.05

Electrical Conductivity:

        All theoretical variograms used to estimate the electrical conductivity show a high nugget
effect, high variance and a strong variation coefficient (Table 3). The electrical conductivity is consequently unpredictable and slightly structured variable, with a strong variability even at very short distance. This supposes a high density of sampling to predict the electrical conductivity.

        On the basis of variographic analysis (Table 3 and Fig. 7) and cross validation (R²=0.73 and S.E.E = 4 dS/m) (Fig. 8), the exponential model is the best theoretical one to estimate the electrical conductivity (Fig. 9).

The conductivity map (Fig. 9) first shows an increasing conductivity from east to west and from the periphery to the center of the plain. This corresponds to a decreasing altitude from east to west and from the periphery to the center. Secondl, a very large surface of the lower-cheliff plain belongs to the range of non-saline to highly saline soils.

Fig. (7): Exponential model

Fig. (8): Cross validation of electrical conductivity estimated by exponential model

Fig. (9): Electrical conductivity map of the Lower-Cheliff obtained by kriging and estimated by the exponential model

Table (3): Summary of theoretical variograms

Spatial Distribution of Vegetation Units According to Conductivity:

        Transformations of species abundance from Braun Blanquet scale to 0-9 Van der Maarel scale enable us to cartography the spatial distribution of the vegetation units by using kriging. As a result, Figures 10, 11, 12 and 13 show a close relation between the vegetation units and the electrical conductivity values. Vegetation unit A (Fig. 10) is distributed throughout the saline soils according to the same gradient of salinity, from the periphery to the center and from the west to the east. Vegetation unit D (Fig. 13) occupies the highly saline soils, whereas vegetation unit B (Fig. 11) is completely absent from the western side of the plain that is characterized by saline soils and low altitudes. Vegetation unit C (Fig. 12) is especially localized at the periphery of the plain and characterized by slightly saline soils.

Fig. (10): Distribution map of vegetation unit A

Fig. (11): Distribution map of vegetation unit B

Fig. (12): Distribution map of vegetation unit C

Fig. (13): Distribution map of vegetation unit D

CONCLUSION:

        The plain of lower-Chelif is an ecosystem weakened by particular edaphic constraints and hard climatic adversities. These constraints strongly reduce the vegetal diversity. Thus, during three years of this study, we counted only 40 species through 133 relevés.

        With an aim of reasonable management of this ecosystem, traditional methods of evaluation of site conditions are expensive and time consuming, especially in areas as large as the lower-Cheliff. Thus, recognition of vegetation ecology and biology is the easiest way limiting cost and time to assess environmental conditions. The present study provides baseline information about the environmental variables affecting the distribution patterns of vegetation assemblages in one of the largest plain in Algeria. It is clear that understanding the local scale factors is needed to assess the importance of factors structuring vegetal communities. The key factors that determine the distribution of vegetation in lower-Cheliff are altitude, electrical conductivity, sodium and soil structure. However, in this study, one of the most important factors that influence the composition of vegetation assemblages is the electrical conductivity. Indeed, it was able for us to differentiate among the four vegetation units in relation to the influence of the main environmental variable. Less diverse vegetation units composed of halophytes species and distributed throughout the saline soils as well as more diverse vegetation units that are very sensitive to salinity occupying the non saline to slightly saline soils were distinguished. Thus, the assessment of vegetal communities is a useful tool to classify salinity, especially in terms of revealing the spatiotemporal changes of this variable. It would be interesting to compare these results with other vegetation studies at the same conditions.

REFERENCES:

Besse, P. (2001): Pratique de la modélisation Statistique, Publications du laboratoire de statistique et probabilité, université Paul Sabatier, Toulouse III, France

Bruelheide, H. (2000): A new measure of fidelity and its application to defining species groups. J. Veg. Sci., 11: 167-178

Braun Blanquet, J. (1928): Pflanzensoziologie. Gründzuge der Vegetationskunde. Springer-Verlag, Berlin, Germany. 631 p.

Butler, R.W. & Wood, A.T.A. (2004): A Dimensional CLT for Non-Central Wilks’ Lambda in Multivariate Analysis, Scandinavian Journal of Statistics, Volume 31, Number 4, December 2004, pp. 585-601(17).

Chytry, M., Tichy, L., Holt, J. & Botta-Dukat, Z. (2002): Determination of diagnostic species with statistical fidelity measures J. Veg. Sci. 13: 79-90.

Chytry, M. & Tichy, L. (2003): Diagnostic, constant and dominant species of vegetation classes and alliances of the Czech Republic: a statistical revision. Folia Fac. Sci. Nat. Univ. Masaryk. Brun., Biol., 108: 1–23.

Curtis, J.T. (1959): The vegetation of Wisconsin: an ordination of plant communities. University of Wisconsin Press, Madison, Wisconsin, USA. 657p.

Erkel-Rousse, H. (1995): Détection de la multicolinéarité dans un modèle linéaire ordinaire : quelques éléments pour un usage averti des indicateurs de Belsley , Kuh et Welsh, revue de statistique appliquée, tome 43, n°4, p. 19-42.

Flahault, C. & Schroter, C. (1910): Rapport sur la nomenclature phytogeographique. Proceedings of the 3^rd. International Botanical Congress, Brussels, 1:131–164.

Gabriel, H.W. & Talbot, S.S. (1984): Glossary of landscape and vegetation ecology for Alaska. Alaska Technical Report 10. Bureau of Land Management, U.S. Dept. of the Interior, Washington, D.C.

Hill, M.O. & Gauch, H.G. (1980): Detrended correspondence analysis, an improvised ordination technique. Vegetatio, 42, 47-58.

Jongman, R.H.G., Ter Braak, C.J.F. & Van Tongeren, O.F.R. (1996): Data analysis in community and landscape ecology. Cambridge University Press. Cambridge. 299 pp.

Journel, A.G. & Huijbregts, C.J. (1978): Mining Geostatistics. Academic Press, 600 p.

Krige, D.G. (1951): A statistical approach to some basic mine valuation problems on the Witwatersrand. 1951. J. of Chem., Metal and Mining Soc. of South Africa, 52, 119-139.

Legendre, P. & Legendre, L. (1998): Numerical Ecology (2nd English edn). Amsterdam, The Netherlands: Elsevier, 853 pp.

Legendre, P. & Gallagher, E. (2001): Ecologically meaningful transformation for ordination of species data. Oecologia 129: 271-280

Leps, J. & Smilauer, P. (2003): Multivariate analysis of ecological data using CANOCO. Cambridge, UK: Cambridge University Press. 269 pp.

Maarel Van der, E. (1975): The Braun-Blanquet approach in perspective. Vegetatio, 30 (3), 213-219.

Marques de Sa, J.P. (2007): Applied Statistics Using SPSS, Statistica, Matlab and R. Springer-Verlag Berlin Heidelberg.

Matheron, G. (1963): Principles of Geostatistics. Economic Geol., 58, 1246-1268.

Okland, R.H. & Eilertsen, O. (1994): Canonical correspondence-analysis with variation partitioning-some comments and an application. Journal of Vegetation Science, 5, 117-126.

Orloci, L. (1967): An agglomerative method for classification of plant communities. J. Ecol 55:193–205.

Rao, C.R. (1995): A review of canonical coordinates and an alternative to correspondence analysis using Hellinger distance. Qüestiió 19:23–63.

Sokal, R.R. & Rohlf, F.J. (1995): Biometry the Principles and Practice of Statistics in Biological Research. Third Edition. W. H. Freeman and Company, New York.

Stein, A., Van Der Meer, F. & Gorte, B. (2002): Spatial statistics for remote sensing, kluwer academic publishers, 301 p.

Ter Braak, C.J.F. (1986): Canonical corres-pondence analysis: a new eigenvector technique for multivariate direct gradient analysis. Ecol 67:1167-1179.

Ter Braak, C.J.F. (1994): Canonical community ordination. Part I: Basic theory and linear methods. Ecosc. 1:127-140.

Westhoff, V. & Van Der Maarel, E. (1978): The Braun-Blanquet approach. In: Whittaker, R.H. (editor). Classification of plant communities, pp. 287-399. Dr. W. Junk,  The Hague, The Netherlands.

Whittaker, R.H. 1962. Classification of natural communities. Bot. Rev. 28: 1-239.

Zuur, A.K., Ieno, E.N. & Smith, G.M. (2007): Analysing Ecological Data. Springer, USA.