The aim of this research is to map the salt-affected soil in an arid environment using an advanced semi-empirical predictive model, Operational Land Imager (OLI) data, a digital elevation model (DEM), field soil sampling, and laboratory and statistical analyses. To achieve our objectives, the OLI data were atmospherically corrected, radiometric sensor drift was calibrated, and distortions of topography and geometry were corrected using a DEM. Then, the soil salinity map was derived using a semi-empirical predictive model based on the Soil Salinity and Sodicity Index-2 (SSSI-2). The vegetation cover map was extracted from the Transformed Difference Vegetation Index (TDVI). In addition, accurate DEM of 5-m pixels was used to derive topographic attributes (elevation and slope). Visual comparisons and statistical validation of the semi-empirical model using ground truth were undertaken in order to test its capability in an arid environment for moderate and strong salinity mapping. To accomplish this step, fieldwork was organized and 120 soil samples were collected with various degrees of salinity, including non-saline soil samples. Each one was automatically labeled using a digital camera and an accurate global positioning system (GPS) survey (σ ≤ ± 30 cm) connected in real time to the geographic information system (GIS) database. Subsequently, in the laboratory, the major exchangeable cations (Ca 2+, Mg 2+, Na +, K +, Cl - and SO 4 2-), pH and the electrical conductivity (EC -Lab) were extracted from a saturated soil paste, as well as the sodium adsorption ratio (SAR) being calculated. The EC -Lab, which is generally accepted as the most effective method for soil salinity quantification was used for statistical analysis and validation purposes. The obtained results demonstrated a very good conformity between the derived soil salinity map from OLI data and the ground truth, highlighting six major salinity classes: Extreme, very high, high, moderate, low and non-saline. The laboratory chemical analyses corroborate these results. Furthermore, the semi-empirical predictive model provides good global results in comparison to the ground truth and laboratory analysis (EC -Lab), with correlation coefficient ( R 2) of 0.97, an index of agreement ( D) of 0.84 ( p < 0.05), and low overall root mean square error (RMSE) of 11%. Moreover, we found that topographic attributes have a substantial impact on the spatial distribution of salinity. The areas at a relatively high altitude and with hard bedrock are less susceptible to salinity, while areas at a low altitude and slope (≤2%) composed of Quaternary soil are prone to it. In these low areas, the water table is very close to the surface (≤1 m), and the absence of an adequate drainage network contributes significantly to waterlogging. Consequently, the intrusion and emergence of seawater at the surface, coupled with high temperature and high evaporation rates, contribute extensively to the soil salinity in the study area.
Soil salinity development occurs in the landscape in response to many factors, especially topographic attributes (altitude and slope) which contribute significantly to the flow paths and, therefore, the salinity of the soil. It is highly dynamic, varies considerably in time and in space, and modifies temporarily or permanently the state of the surface and of the soils below [
The measurement of electrical conductivity (EC) of saturation extract from a saturated soil paste is considered a standard and universally accepted way of measuring soil salinity [
Furthermore, some scientists have advocated the hypothesis that soil-salt occurs in many landscapes in response to the way water moves through and over the landscape. Indeed, terrain attributes contribute significantly to the flow paths and, therefore, to the soil salinity attributes [
The used methodology in this research is summarized in four steps. Firstly, the preprocessing step involves the OLI image corrections from the atmosphere, the radiometric drift of the sensor, and the topographic and the geometric distortions using a high spatial resolution (5-m) DEM. Secondly, the processing step addresses the soil-salinity map retrieval using an advanced semi-empirical model for salinity detection based on the Soil Salinity and Sodicity Index-2 (SSSI-2). In addition, vegetation cover was extracted based on the Transformed Difference Vegetation Index (TDVI) using EASI-modeling of PCI-Geomatica [
The Kingdom of Bahrain (26˚00'N, 50˚33'E) is a group of islands located in the Arabian Gulf, east of Saudi Arabia and west of Qatar (
Geologically, Bahrain is characterized by Eocene and Neocene rocks, which are partly covered by Quaternary sediments and a complex of Pleistocene deposits. The dominant rocks are limestone and dolomitic-limestone with subsidiary marls and shales. The leading structure is the north-south axis of the main dome, with minor cross folds predominantly tilting from northeast to southwest. The beds are gently inclined towards the coast from the center of the main island. The fringes of Bahrain are covered by more recent marine and Aeolian sand dunes, which were derived from the Arabian land connection across the present Arabian Gulf [
Since 1972, the Landsat satellite program, involving NASA, the USGS and other agencies, has provided a continuous record of the Earth’s surface reflectivity from space. Indeed, the Landsat satellites series supports more than four decades of global moderate resolution data collection, distribution and archives of the Earth’s surface [
Prior to launch, the OLI sensor was subject to rigorous radiometric and spectral characterization and calibration [
L * ( λ ) = G ( λ ) D N ( λ ) + O ( λ ) (1)
ρ * ( λ ) = [ π ⋅ L * ( λ ) ⋅ D 2 ] / [ E 0 ( λ ) ⋅ cos ( θ s ) ] (2)
ρ * ( λ ) = t g ¯ ( λ ) ⋅ [ ρ a ( λ ) + ρ G ( λ ) ⋅ T ( λ ) θ s ⋅ T ( λ ) θ v 1 − ρ G ( λ ) ⋅ S ] (3)
where:
L * ( λ ) = Apparent equivalent spectral radiance at TOA [Watts. (m2 sr μm)−1],
G ( λ ) = Radiance multiplicative rescaling factor from the metadata (gain),
O ( λ ) = Radiance additive rescaling factor from the metadata (offset),
D N ( λ ) = Digital number values,
E 0 ( λ ) = Irradiance [Watts. (m2 sr μm)−1],
D = Earth-Sun distance [astronomical units],
ρ * ( λ ) = Apparent reflectance at the TOA, with a correction for solar zenith angle,
ρ a ( λ ) = Atmospheric at-sensor reflectance,
ρ G ( λ ) = Ground reflectance,
t g ¯ ( λ ) = Average total gaseous transmittance,
T ( λ ) θ s = Total descending scattering transmittance,
T ( λ ) θ v = Total ascending scattering transmittance, and
θ s = Solar zenith angle,
θ v = Sensor zenith angle,
S = Spherical albedo.
Parameter | Value |
---|---|
Terrain elevation (ASL) | 0.065 km |
Sensor elevation | 705 km |
Time of over-flight (GMT) | 10:45 |
Date of over-flight | April 5, 2015 |
Solar zenith angle | 30.452˚ |
Solar azimuth angle | 126.625˚ |
Atmospheric model | Dry |
Aerosol model | Desert |
Horizontal visibility | 30 km |
Ozone content | 0.319 cm-atm |
Water vapor | 0.75 g・(cm2)−1 |
CO2 mixing ratio | 357.5 ppm (as per model) |
During the 90-day period following the OLI launch, three types of geometric calibrations were performed on-orbit including updating the OLI-to-spacecraft alignment knowledge, refining the alignment of the sub-images from the multiple OLI sensor chips, and refining the spatial alignment of the OLI spectral bands. The results showed that the considered aspects of geometric performance met the system accuracy requirements [
To exploit remote sensing for salinity soil mapping, different spectral salinity indices have been proposed in the literature [
EC- Predicted = C s t ⋅ [ 4521 ( SSSI-2 ) 2 + 125 ( SSSI-2 ) + 0.41 ] (4)
SSSI-2 = ( ρ OLI-6 ⋅ ρ OLI-7 − ρ OLI-7 ⋅ ρ OLI-7 ) / ( ρ OLI-6 ) (5)
where:
C s t : Scaling factor,
EC- Predicted : Predicted EC from remote sensing semi-empirical model,
ρ OLI-6 : Reflectance in OLI SWIR-1 channel, and
ρ OLI-7 : Reflectance in OLI SWIR-2 channel.
The scaling factor ( C s t ) enables an up-scaling between the spatial information measured in the field (fine scale) and its homologous information derived from the image (coarse scale). In the literature, several methods exist to calculate this factor depending on the remote sensing applications [
Salt-affected landscape inhibits vegetation cover growth and agricultural productivity. However, diverse halophytic plants grow in an arid environment according to their tolerance to salinity and the alkalinity of the soil [
TDVI = 1.5 [ ( ρ OLI-5 − ρ OLI-4 ) / ρ OLI-5 2 + ρ OLI-4 + 0.5 ] (6)
where:
ρ OLI-4 = Reflectance in OLI red channel, and
ρ OLI-5 = Reflectance in OLI near-infrared channel.
Remote sensing result validation is a crucial step and requires independent protocols to collect updated and accurate information as ground truth. Pre-existing data are often used for validation [
The soils of Bahrain are characterized by five different soil classes of moderate to shallow depth, and are closely related to the geology and geomorphology of the terrain [
Soil sample collection was carried out between 2 and 7 April, 2016. The Bahrain soil map was used as a reference for the sampling data, and 120 samples were selected based on the spatial representativeness of the major soil classes as discussed above and by considering various degrees of salinity and the non-saline soil. Samples were taken from the upper layer (5 cm deep) of the soil, in an area of about 50 × 50 cm2. Observations and remarks about each sample (color, brightness, texture, etc.) were noted. The location of each point was automatically labeled and recorded using a 35 mm digital camera equipped with a 28 mm lens and accurate GPS survey (σ ≤ ±30 cm) connected in real time to the GIS database. Then, each sample was analyzed in the laboratory in order to extract the major exchangeable cations (Ca2+, Mg2+, Na+, K+, Cl− and SO 4 2 − ), the EC-Lab, pH, and the SAR from a saturated soil paste. These elements were analyzed using methods that meet the current international standards in soil science [
Statistical analyses were computed using “Statistica” software. Various statistics were calculated for both EC ground sampling points obtained from the laboratory analysis (EC-Lab) and the predicted values derived from OLI data (EC-Predicted) using the semi-empirical model. Standard deviation statistics enabled the evaluation of data variability. This parameter was reported in all cases as an error percentage of the average values extracted from the ground sampling point (ECG-Lab) and image data (EC-Predicted). For validation purposes, EC-Lab and EC-Predicted were compared using the 1:1 line. Ideally, observed and predicted values should have a correspondence of 1:1. The index of agreement D reflects the degree to which the observed value is accurately estimated by the predicted value. This measure was calculated as follows [
D = 1 − [ ∑ i = 1 n ( P i − O i ) 2 ∑ i = 1 n ( | P ′ i | + | O ′ i | ) 2 ] (7)
where Pi is the predicted value at sample i, Oi is the observed value at sample i, P ′ i is the difference between Pi and the average of the predicted values, and O ′ i is the difference between Oi and the average of the observed values and n is the number of values. This index provides a measure of the degree to which a model’s predictions are error-free. The index ranges between 0 and 1, with 1 indicating a perfect match between observed and predicted values. The observed values were those calculated from each sampling point at the laboratory (EC-Lab) and the predicted values were from the salt-affected soil map using the semi-empirical model and OLI image (EC-Predicted). The root mean square error (RMSE) was used as an overall error to supplement the index of agreement described above. This error also quantifies the 1:1 relationship between observed and predicted values. It was calculated as follows [
RMSE = ∑ i = 1 n ( P i − O i ) 2 n (8)
The relationships between observed and predicted values were also analyzed using a linear regression model. The correlation coefficient (R2) of the regression model was also used to evaluate the strength of the linear relationship between observed (EC-Lab) and predicted values (EC-Predicted). Systematic linear overpredictions or underpredictions generate characteristic variations in the slope and intercept values, which can help to interpret the major sources of error and the potential of the semi-empirical model for salinity mapping and prediction using OLI data.
In this section, we present and evaluate the soil salinity maps and relate their accuracy directly to our field observations, laboratory analyses, and statistical validation. Globally, the results show that the semi-empirical model based on the SSSI-2 index provided a satisfactory result in comparison to the ground truth, laboratory analyses, as well as good agreement with spatial distribution of vegetation cover derived with TDVI index, ancillary data (soils, geology, and geomorphology maps), and topographic attributes.
Based on the histogram analysis of the derived salt-affected map, the spatial variability of soil salinity was characterized by six classes (
The very high salinity class presents other crustal features including a bare level soil surface (class 2, red color in
According to the field visit and ancillary data (soil, geological and geomorphological maps), we find that these first four saline classes (extreme, very high, high and moderate) are globally characterized by Eocene, Neocene and Miocene rocks. They are partly covered by Quaternary sediments and a complex of Pleistocene deposits that are often rich in calcium sulfate and sodium chloride and associated with shales and marls, limestone and dolomitic-limestone. Their gypso-saline characteristics cause the formation of salt strata that are thrust upward to the surface from the underlying salt bed. These saline formations are often associated with anhydrite, gypsum, sulfur, and paleo-lagoonary sedimentary rocks.
Furthermore, the spatial distribution of low salinity class was located in the northwest part of Bahrain island (class 5, blue-green color in
Likewise, the major exchangeable cations in the considered soil samples (Ca2+, Mg2+, Na+, K+, Cl− and SO 4 2 − ), pH, EC-Lab, and SAR values were determined in the laboratory from saturated soil paste extract (
Salinity class | EC-Lab (dS・m−1) | pH | Ca2+ (mg・l−1) | K+ (mg・l−1) | Mg2+ (mg・l−1) | Na+ (mg・l−1) | SAR | Cl− (mg・l−1) | SO 4 2 − (mg・l−1) |
---|---|---|---|---|---|---|---|---|---|
Extreme | 507.0 | 7.6 | 1276 | 843 | 672.0 | 154,700 | 874.0 | 170,715 | 11,275 |
Very high | 170.0 | 7.2 | 1878 | 1454 | 2874.0 | 76,373 | 258.9 | 100,281 | 28,020 |
High | 67.0 | 7.5 | 1905 | 651 | 1581.0 | 24,171 | 99.2 | 48,546 | 5488 |
Moderate | 7.4 | 8.6 | 531 | 67 | 181.0 | 1324 | 12.7 | 2480 | 881 |
Low | 4.4 | 8.2 | 284 | 44 | 96.0 | 782 | 10.3 | 1329 | 754 |
Non-saline | 2.6 | 7.9 | 154 | 28 | 58.4 | 530 | 9.2 | 886 | 63 |
With reference to ground truth, the vegetation cover map derived using TDVI (
The topography of Bahrain generally has low variability. Over half of the surface lies below 20 m, and is composed mainly of low angle slopes. It is possible to identify five major physiographic regions from our map products. These occur as concentric units of variable width (
mimic the major physiographic areas: the coastal lowland, the upper Dammam back-slope, the multiple escarpment zones, and the interior-basin and central plateau. Consequently, it is apparent that the topographic attributes have a significant relationship with the spatial distribution of soil salinity. Indeed, topography has a strong impact on controlling flow accumulation and, consequently, the development of various levels of soil salinity, as well as more complex relationships over larger areas that can be discerned from spatial analyses. Terrain aspect controls the flow direction and accumulation, whereas the magnitude of slope controls the speed of groundwater and mechanical erosion. The groundwater accumulation and groundwater salinity are strongly controlled by flow direction, slope percentage, aspect, relief catchment area and distance between upstream and down streams. In general, the areas of low topography and low slope (
When the elevation and slope maps (
catchments. Moreover, the absence of an adequate drainage system and soil-water management policy contribute significantly to waterlogging. Therefore, the intrusion and emergence of the seawater at the surface coupled with the very high temperature and evaporation rate contribute extensively to soil salinity crust. This situation characterizes the sabkha formation, where evaporation exceeds water influx. Nonetheless, according to the field visits and ancillary data (geology and geomorphology maps), the extreme salinity zones located in the southeast of the island (~20 m altitude and moderate slope) are not related to the coastal deposits or to waterlogging. Instead, they are attributed to the carbonate formation due to the presence of evaporate salts dissolved from local bedrock, geological structures and chemical dissolution. These findings are also in agreement with other scientists’ results elsewhere in the world [
In this research, we carried out salt-affected soil mapping in an arid environment using Landsat-OLI data, DEM, field soil sampling, a semi-empirical predictive model, and laboratory and statistical analyses. The OLI data were preprocessed from the atmosphere, the sensor radiometric drift, the geometric distortions, and topographic variabilities. Then, the soil salinity map was derived using a semi-empirical predictive model based on the SSSI-2, as well as the vegetation cover map being extracted from the TDVI. Moreover, topographic attributes were derived using accurate DEM of 5-m pixel in size. For statistical analysis and validation purposes, fieldwork was organized and 120 soil samples, as well as non-saline soil samples, were collected with various degrees of salinity. Each sample was automatically labeled using a digital camera and accurate GPS survey (σ ≤ ±30 cm) connected in real time to the GIS database, and was then analyzed in the laboratory both to measure the major exchangeable cations in the considered soil samples (Ca2+, Mg2+, Na+, K+, Cl− and SO 4 2 − ), pH, EC-Lab, and to calculate the SAR from a saturated soil paste extract.
Globally, the visual validation step demonstrated a very good conformity between the derived soil salinity map and the ground truth, highlighting six major salinity classes: extreme, very high, high, moderate, low and non-saline. It shows the efficacy of the semi-empirical model used to discriminate accurately among different and complex salinity classes, from sabkha to non-saline soils. The chemical laboratory analyses corroborate these remote sensing results and field observations. They reveal a very high concentration of sodium (Na+), which generally exceeds the sum of Ca2+ and Mg2+, and dominant Cl− that exceed SO 4 2 − for the six salinity classes taken into consideration. Moreover, the values of EC-Lab, Na+, and SAR increased gradually and very significantly from the non-saline soil to the extreme soil salinity (sabkha). Indeed, the non-saline and low soil salinity classes, which support the agricultural system in Bahrain, are characterized with low EC (2.6 ≤ EC-Lab ≤ 4.4 dS・m−1) and SAR (≤10.3). The moderate salinity class with EC-Lab of around 7.4 dS・m−1 and SAR ≤ 12.7 is the dominant soil class in Bahrain, allowing for the growth of halophytic plants. Contrariwise, the other three soil salinity classes with extreme, very high and high salinity content show very strong and extreme EC (67 ≤ EC-Lab ≤ 600 dS・m−1) and very high SAR (≥99.2) values. Obviously, these results are in agreement with the six classes predicted by the remote sensing method. Furthermore, statistical validation of the semi-empirical predictive model used provides satisfactory results in comparison to the ground truth and the laboratory analyses (EC-Lab), with correlation coefficient (R2) of 0.97 and an index of agreement (D) of 0.84, at significance level p < 0.05. Overall, the RMSE is approximately 11%, if we consider the 120 soil samples in totality. However, this error varies between 5 and 21%, respectively, if we consider only low-moderate salinity or only strong- extreme salinity. Although this model was developed for moderate and slight salinity in irrigated agricultural land in semi-arid regions, it also showed its ability to be applicable to different extreme salinity conditions in arid lands.
Visual analysis and field visits corroborated these results with respect to the spatial distribution of vegetation cover, ancillary data (soil, geology and geomorphology maps) and topographic attributes. Additionally, the TDVI confirmed its high performance for vegetation cover discrimination in arid lands, and provided supplementary results for mapping the spatial distribution of salinity over the study site. Furthermore, it is clear that topographic attributes have a significant impact on salinity spatial distribution. Our results showed that areas at a relatively high altitude and/or with hard bedrock are less susceptible to salinity, except for some areas where the evaporate salts were dissolved from local bedrock. However, areas at a low altitude and with Quaternary soil are prone to salinity, since the water table is very close to the surface at low elevations (≤1 m a.s.l.) where slopes are insignificant (≤2%). The water table is rarely horizontal, but reflects the surface relief due to the capillary effect in soils and sediments, but does not always mimic the topography due to variations in the underlying geological structure (e.g., folded, faulted, fractured bedrock). Moreover, the absence of an adequate drainage network contributes significantly to waterlogging. Consequently, the intrusion and emergence of seawater at the surface, coupled with poor irrigation water quality (in agricultural areas), a very high temperature and high evaporation rate, contribute extensively to soil salinity in The Kingdom of Bahrain. In this region, the derived salinity maps also showed important terrain-salinity relationships. For example, soils are saline and calcareous (to highly calcareous with high gypsum and calcium carbonate content) and alkaline (or neutral) on the surface horizon; coarse to medium texture on upland sites; fine texture in closed basins; are usually only weakly developed morphologically; poor in organic matter; deficient in micronutrients; and have low fertility potential. Furthermore, besides its salinity, the soil in Bahrain shows sodic characteristics, since the soil structure is completely absent or poorly defined. The preponderance of carbonate and the presence of bicarbonate in the soils are responsible for high pH values (7.1 to 8.6) and, consequently, contribute significantly to the alkalinity aspect of the soils.
The authors thank the NASA-GLOVIS-GATE for the Landsat-8 OLI data. Our gratitude goes to Professor Ramesh P. Singh of Chapman University (Orange, CA, USA) for his critical feedback of this paper. We would also like to thank the anonymous reviewers for their constructive rectifications and comments.
The authors would like to thank the Arabian Gulf University for the financial support of the soil-salinity mapping project accorded to Professor A. Bannari.
Bannari, A., El-Battay, A., Hameid, N. and Tashtoush, F. (2017) Salt-Affected Soil Mapping in an Arid Environment Using Semi-Empirical Model and Landsat-OLI Data. Advances in Remote Sensing, 6, 260-291. https://doi.org/10.4236/ars.2017.64019