INTRODUCTION
Soil salinity is the most important constraint to agricultural sustainability,
but accurate information on its variation and its impact on agricultural regions
are difficult to obtain (Lobella et al., 2007).
Monitoring soil salinity requires knowledge of its magnitude and its spatial
and temporal variability. The method used depends on the data availability and
the aim of the study. The joint use of temporal stability and temporal mean
shift and spatial shift tests could result in a drastically reduced sampling
effort (Douaika et al., 2007). So, spatial changes
of various soil traits such as its salinity may be one of the main error sources
in the estimation of the non-measured data. Corwin et
al. (1992) have implied to the necessity of the application of Geostatistical
methods in the compensation for the defects encountered with the above-mentioned
methods. Accordingly, geostatistical methods like non-parametrical statistical
estimators such as weighed moving average and or parametrical Geostatistical
methods such as Kriging and Co-Kriging attract attentions. Because of taking
in account of spatial correlation of data, Geostatistical methods are of high
importance in the studies related to the distribution of earth data (Goovaerts,
1999).
The results obtained by Alaeddin et al. (2007)
suggested that sampling cost can be reduced and estimation can be significantly
improved using cokriging. Walter and McBratney (2001)
used Kriging method in order to predict soil surface salinity in his analytical
studies on the spatial distribution of soil salinity. Triantafilis
et al. (2001) resulted to more accuracy of Co-Kriging method than
ordinary Kriging or other univariate predictors.
Mohammadi (1998) concluded that the calculated variograms
were mainly in accordance with spherical and exponential models.
Corwin et al. (1992) applied GIS and successfully estimated salinity
of the lands under irrigation. Lesch et al. (1995)
mapped soil salinity distribution using calibrated spectrometric data. Mohammadi
(2000) compared the efficacy of different Geostatistical methods including
Co-Kriging, Kriging and linear regression methods and found that the Geostatistical
estimators were relatively superior to linear relations and introduced Kriging
method as the superior method for the estimation of soil spatial data.
Geostatistics offers a collection of deterministic and statistical tools aimed
at understanding and modeling spatial variability (Goovaerts,
1999; Triantafilis et al., 2001).Therefore
the present research was conducted to analyze spatial changes in soil salinity
distribution as an aspect of soil degradation and to compare the efficacy of
different Geostatistical methods in its estimation and the preparation of maps
of the positional distribution of soil salinity.
MATERIALS AND METHODS
Area identification: The investigation was performed in Southern
part of Orumieh plain in West Azerbaijan province, Iran. Geographically,
this area is longitudinally located 45°05’ 00" E and 45°20’ 00"
E and between 37°15’ 00" N and 37°35’ 00" N. Figure
1 shows the location of the region in the country and province and
shows the position of the profiles used in the plain. Based on the relatively
detailed pedological studies made in Water and Soil Research Institute
(2000), the regional soils are classified in the class Inceptisols and
belong to one of two main subgroups, typical Calcixerepts and typical
Haploxerepts. The distance between the profiles in the studied region
varied between 1300 and 4700 m. Extensive soil salinity is observed around
the Uromieh Lake and in parallel to its coasts.
Research framework: Geostatistical methods including Kriging, Weighed
Moving Average and Co-Kriging methods (Metternicht and Zinck,
2003) were used in GIS medium and GS+ and ARCVIEW8
software (Goovaerts, 1999) were applied to investigate
the spatial changes and the estimation of superficial soil salinity. The general equation for
these methods is as follow:
Where:
| Z* (xi) |
= |
The estimated amount |
| Z (xi) |
= |
The observed amount around the assumed point |
| (xi) |
= |
The position of the observed points |
| λ i |
= |
The amount of weights of the observed points |
| N |
= |
No. of the measurement points |
To evaluate interpolation methods, the cross validation technique and
two MAE and MBE statistical parameters were used. MAE is an indicator
of errors in the results and MBE indicates the biass of the results obtained
through the applied method. When MAE and MBE are 0.00 or near to naught,
the applied method simulates the fact well. However, as far as its amount
is farer than 0.00, it implies to less precise and more biass. How the
parameters MAE and MBE are calculated, has been indicated as follow:
|
| Fig. 1: |
Location of the study area and soil profiles |
Where:
| Rs |
= |
The estimated amount |
| Ro |
= |
The measured amount |
| N |
= |
No. of the data |
Spatial distribution of surface soil salinity was mapped after the selection
of the appropriate interpolation model. This map was compared with the photos
taken with the satellite through multi-spectral combination method (Metternicht
and Zinck, 2003) and salinity borderlines and the trend of regional salinity
variations were reviewed and tested.
RESULTS
Shapiro-Wilk test conducted to see if the data were of normal distribution,
indicated that the data related to soil salinity were normal and of a
coefficient less than 0.05 and their skew ness coefficient was less than
1 (df = 26; calculated test coefficient = 0.035 and skewness coefficient
= 0.568).
To follow Kriging and Co-Kriging methods, calculation of semivariogram
is a pre-requirement. The resulted model for Kriging method has been shown
in Fig. 2. The performed studies indicated that the
Gaussian Model is an appropriate model for this semi- variogram. The effect
radiance of this semivariogram is equal to 8000 m, nugget effect equals
1.4 and its sill is equal to 41.28 m2. The correlation coefficient
for the fitted model has been calculated as 0.98.
Also, there was a significant correlation between soil salinity and lime
content (R = 0.74); therefore, lime content was used as a co variable
with the Co-Kriging method. The empirical semi-variogram model obtained
through this method indicated that Gaussian model was a suitable model
for semi-variogram. The effect radiance of this semi-variogram is equal
to 10000 m, its nugget effect is 0.26 and its sill equals to 4.53 m2
with a correlation coefficient of 0.58.
As Fig. 3-5 shows the results from the cross-validation
test of selected geo-statistical methods, the Kriging method-based fitted
line obtained from the estimated data is of more fitness with the measured
data. The rates of preciseness and biass with Kriging, Co-Kriging and
weighted moving average have been shown in the Table 1.
|
| Fig. 2: |
Empirical semi-variogram model of soil salinity using
the Kriging method, Gaussian model (Co = 1.40000, Co+C = 41.28000,
Ao = 2900.00, R 2 = 0.980, RSS = 11.5) |
|
| Fig. 3: |
The cross validation of soil salinity estimation following
the Co-Kriging method, Regression co-efficient = 1.614 (SE = 0.871,
R 2 = 0.125, y-intercept = -1.44, SE prediction = 6.555) |
|
| Fig. 4: |
The cross validation of soil salinity estimation following
the weighted moving average method, Regression co-efficient = 1.236
(SE = 0.209, R 2 = 0.594, y-intercept = -0.70 SE prediction
= 4.468) |
|
| Fig. 5: |
The cross validation of soil salinity estimation following
the Kriging method, Regression co-efficient = 0.941 (SE = 0.304, R 2
= 0.285, y-intercept = 0.35, SE prediction = 5.924) |
|
| Fig. 6: |
Comparing the regions classified as saline land with
their position in the false color TM satellite image |
| Table 1: |
The error and biass values of the selected geostatistical
methods used in the estimation of soil salinity levels |
|
Based on the information presented in Table 1, it is
confirmed that Kriging method with an error rate of 1.31 dS m-1 is more preciseness for the estimation of soil salinity. While Co-Kriging
method is less biass compared to the Kriging method (-0.09 dS m-1),
this difference is only at the rate of 0.20 dS m-1. Therefore,
considering the preciseness and biass rates and the Kriging method would
be the most superior method.
Thus, because of its more preciseness, Kriging method was chosen as the appropriate
model for the spatial estimation of soil salinity and the salinity levels were
estimated for various regional points and their regional distribution maps were
prepared in GIS medium (Fig. 6). The comparison of the points
classified as saline lands shown in Fig. 6 are in good accordance
with the satellite image. Based on this, the regional trend of salinity variation
is in a way that saline lands have been extended paralleled to Orumieh coastal
lines (Region A). The breadth of these areas depends on the land topography
and the level of lake brine penetration inside coastal lands. In better words,
the rate of brine penetration is of more extension in the lands of mild slopes
(Region B). The lands located in the central and west southern and classified
as the relatively saline lands are under the effects of physiographic situation
and are considered as low lands (Region C).
DISCUSSION
The results obtained with the present investigation, namely the selection and
recommendation of Kriging method are in agreement with those published by Mohammadi
(1998), Walter and McBratney (2001) and apposite
with Triantafilis et al. (2001) results. This
indicate the spatial regularity of soil salinity data but it differs in different
regions. The fitted model in this research was Gaussian model for semi-variogram;
however, in a research by Mohammadi (1998) exponential
and spherical models have been resulted. Also, using the satellite-based calibrated
spectral and numerical data, Lesch et al. (1995)
and Mohammadi (2000) have made a TM digital data, however,
the correctness of the prepared maps have not been tested and in no case of
the reviewed literature on the comparison of different geo-statistical methods,
statistical methods have been applied in comparisons.
Subscribe Today
