Potential and limits of vegetation indices compared to evaporite 1 mineral indices for soil salinity discrimination and mapping

. The study aims to analyze the ability of the most popular and widely used vegetation indices (VI’s), 8 including NDVI, SAVI, EVI and TDVI, to discriminate and map soil salt contents compared to the potential of 9 evaporite mineral indices such as SSSI and NDGI. The proposed methodology leverages on two complementary 10 parts exploiting simulated and imagery data acquired over two study areas, i.e. Kuwait-State and Omongwa salt-pan 11 in Namibia. In the first part, a field survey was conducted on the Kuwait site and 100 soil samples with various 12 salinity levels and contents were collected; as well as, herbaceous vegetation cover canopy (alfalfa and forage 13 plants) with various LAI coverage rates. In a Goniometric-Laboratory, the spectral signatures of all samples were 14 measured and transformed using the continuum removed reflectance spectrum (CRRS) approach. Subsequently, 15 they were resampled and convolved in the solar-reflective spectral bands of Landsat-OLI, and converted to the 16 considered indices. Meanwhile, soil laboratory analyses were accomplished to measure pHs, electrical conductivity 17 (EC- Lab ), the major soluble cations and anions; thereby the sodium adsorption ratio was calculated. These elements 18 support the investigation of the relationship between the spectral signature of each soil sample and its salt content. 19 Furthermore, on the Omongwa salt-pan site, a Landsat-OLI image was acquired, pre-processed and converted to the 20 investigated indices. Mineralogical ground-truth information collected during previous field work and an accurate 21 Lidar DEM were used for the characterization and validation procedures on this second site. The obtained results 22 demonstrated that regardless of the data source (simulation or image), the study site and the applied analysis 23 methods, it is impossible for VI’s to discriminate or to predict soil salinity. In fact, the spectral analysis revealed 24 strong confusion between signals resulting from salt-crust and soil optical properties in the VNIR wavebands. The 25 CRRS transformation highlighted the complete absence of salt absorption features in the blue, red and NIR 26 wavelengths. As well as the analysis in 2D spectral-space pointed-out how VI’s compress and completely remove 27 the signal fraction emitted by the soil background. Moreover, statistical regressions ( p ˂ 0.05) between VI’s and EC- 28 Lab showed insignificant fits for SAVI, EVI and TDVI (R 2 ≤ 0.06), and for NDVI (R 2 of 0.35). Although the 29 Omongwa is a natural flat salt playa, the four derived VI’s from OLI image are completely unable to detect the 30 slightest grain of salt in the soil. Contrariwise, analyses of spectral signatures and CRRS highlighted the potential of 31 the SWIR spectral domain to distinguish salt content in soil regardless of its optical properties. Likewise, according 32 to Kuwait spectral data and EC- Lab analysis, NDGI and SSSI incorporating SWIR wavebands have performed very 33 well and similarly (R 2 of 0.72) for the differentiation of salt-affected soil classes. These statistical results were also corroborated visually by the maps derived from these evaporite indices over the salt-pan site, as well as by their

5 radiation (APAR), production rate of the biomass, etc. Moreover, their interest lies in the detection of changes in 146 land use and the monitoring of the seasonal dynamics of vegetation on local, regional and global scales (Leeuwen et 147 al., 1999). Based on the red and NIR bands, the NDVI was proposed by Rouse et al. (1974) at the dawn of remote 148 sensing. Since these two spectral bands are generally present on Earth observation and meteorological satellites, and 149 often contain more than 90% of the information relating to vegetation canopy (Baret, 1986;Bannari et al., 1995), the 150 NDVI had taken a privileged place in the NASA/NOAA Pathfinder project (James and Kalluri, 1994). Thus, it was 151 daily derived from NOAA-AVHRR data at the Earth scale. Subsequently, it was also derived every day from 152 MODIS and SPOT-Vegetation data to produce a time series products for global vegetation assessment and 153 monitoring at the regional and global scales (Chéret and Denux, 2011;Hameid and Bannari, 2016;Liu et al., 2021).

154
Due to this glorious history and its simplicity, the NDVI has become the most widely used to assess vegetation.

155
However, despite its popularity and its capability to reduce the sun illumination geometry and to normalize the 156 topographic variations (Kaufman and Holben, 1993;Bannari et al., 1995), the NDVI shows some sensitivity to the 157 atmosphere (scattering and absorption) and soil background artefacts (color, brightness, texture, etc.). To overcome 158 these limitations, more than fifty VI's have been developed and proposed for various applications and under specific 159 conditions (Bannari et al., 1995). However, despite these new development and innovative efforts, the use of VI's to

164
Cert, the majority of these limiting factors can be corrected on remote sensing imagery or in situ measurements 165 before the extraction of such index; except the impact of the optical properties of the soil background. This last 166 factor has been considered in the theoretical concept supporting many VI's development for minimising or removing 167 completely the contribution of the soil underlying the canopy on the remotely sensed signal and, therefore, to 168 enhance that resulting from the biomass. For instance, the soil adjusted vegetation index (SAVI) was proposed by 169 Huete (1988) to minimize the artefacts caused by soil background on the estimation of vegetation cover fraction by 170 incorporating a correction factor "L". Moreover, to overcome the limitations of linearity and saturation, to reduce 171 the noise of atmospheric effects, and to remove the artefacts of soil optical properties, the enhanced vegetation index 172 (EVI) was proposed also by Huete et al. (2002). Furthermore, the transformed difference vegetation index (TDVI) 173 was proposed by Bannari et al. (2002) to describe the vegetation cover fraction independently to the soil-174 background, to reduce the saturation problem, and to enhance the vegetation dynamic range linearly. These indices 175 (NDVI, SAVI, EVI, and TDVI) were developed and used to establish a close relationship between radiometric responses and vegetative cover densities. However, despite their particular mission of assessing and managing 177 vegetation covers, many users of remote sensing applied these indices for soil salinity detection and mapping above (NDVI, SAVI, EVI and TDVI) are considered and compared to the newly proposed evaporite mineral indices 183 (NDGI and SSSI). In this regard, a field survey was conducted for soil and vegetation cover sampling, soil 184 laboratory analysis, spectral measurements in a Goniometric-Laboratory, and Landsat-OLI image were used. Two  and LAI densities were transformed using the CRRS (Clark et al., 1987). Likewise, all measured spectra were 201 resampled and convolved in the solar-reflective spectral bands of OLI sensor using the Canadian Modified Simulation 202 of a Satellite Signal in the Solar Spectrum (CAM5S) radiative transfer code (Teillet and Santer, 1991)   Field survey was organized in the center and the east of Kuwait territory (Fig. 2), it includes irrigated 254 agricultural fields, desert land, urban areas, coastal zones, and low-land such as Bubiyan Island. Based on the 255 fieldwork and soil map, the following soil salinity classes represented by photos in Fig. 2

267
In the laboratory, the considered soil samples were air-dried, ground, and passed through 2 mm sieve. After the 268 spectral signatures measurements, the saturated soil paste extract method was utilized to measure the EC -Lab and pH 269 of saturated soil paste (pHs). Moreover, the major soluble cations (Ca 2+ , Mg 2+ , Na + , and K + ) and anions (Cland 270 SO 4 2-) were measured, and the sodium adsorption ratio (SAR) was calculated. These analyses have been carried out 271 at the soil laboratory using methods that meet the current international standards in soil science (Richards, 1954;

281
The bidirectional reflectance spectra were measured above each air-dry soil sample at nadir with a field of view 282 (FOV) of 25° and a solar (Halogen floodlights) zenith angle of approximately 5° by averaging forty measurements.

283
The ASD was installed at a height of 60 cm approximately over the target, which makes it possible to observe a     (1)

369
(3) Where: R and NIR are the ground reflectance in the red (OLI-4) and near-infrared (OLI-5) spectral bands, "L" is a 375 correction factor equal 0.5; SWIR1 and SWIR2 are the ground reflectance in shortwave infrared spectral bands,

396
The main sources of Clin the soil are from seawater (level rise and spray), precipitation, salt dust, irrigation, and   Table 1).

418
In fact, the sample "D" is a sandy soil with small amount of gypsum crystals and shells, and the beginning of salt 419 crust formation (light gray and white color), while the sample "H" is a pure salt-sabkha (bright florescent halite 420 crust). Similar confusion is also observed between the opposite samples "A" and "H", respectively, with 2.4 and 507 421 dS.m -1 values of EC-Lab . Moreover, the samples "A" and "G" are sandy soils with EC-Lab of 2.4 and 445.5 dS.m -1 , according to their color ( Fig. 5a and Table 1). Consequently, it is impossible to discern or to separate between "D" 424 and "H" or "A" and "G" samples in the VNIR. This affirmation was also reported by Metternicht and Zinck (1997),

425
who demonstrated that the soil textures can be a source of spectral confusion between soil salinity classes; as well as 426 the color and roughness of the soil crusts influenced the reflectance in VNIR and, therefore, causing confusion 427 among the salts contents in the soil.

428
On the other hand, the Fig. 5a shows that when the EC-Lab values increase, also the difference among the salt-429 affected soil spectra's increase significantly and progressively from 1100 to 2500 nm region of the spectrum. In this 430 SWIR domain, the spectral signatures of soil samples from "A" to "H" changed progressively in amplitude and 431 shape according to EC-Lab contents (from 2.4 to 507 dS.m -1 , see Table 1), as well as a function of SAR (from 1.6 to 432 444.7 (mmoles/l) 0.5 ). The ambiguity between "D" and "H" or "A" and "G" samples observed in the VNIR, is

448
Indeed, in this specific electromagnetic window we observe that the sample "H" which is 10 time more saline than

451
are presenting comparable absorption features (Fig. 5b). This similarty is automatically related to the texture, 452 raughness, color and brightness of soil samples and not for their salinity content degrees. Infact, "B" and "H" 453 samples have the same color (white, 10YR 8/1), while the samples "A", "C" and "E" are presenting a very slight

493
In this section, the analysis of VI's capability for soil salinity discrimination was undertaken in two different ways.

494
The first involves a 2D spectral-space analysis (scatter-plot) relating each index to the reflectance in the red band ranging from 2.4 dS.m -1 (non-saline soil) to 635 dS.m -1 (pure salt, sabkha), and their characteristics are summarized 497 in Table 1. The 2D spectral-space illustrates how the fraction of vegetation cover is perfectly highlighted by the VI's 498 (Fig. 7), and predicted correctly and gradually from 50% to 95% proportionally to the increased LAI rates. Whereas    In the PCI-Geomatica image processing system, the histograms of the derived salinity maps applying SSSI and 579 NDGI (Fig. 11) were thresholded based on the major salinity classes including non-saline (blue), low (cyan or sky-580 blue-green), moderate (clear green), high (yellow), very high (orange-red) and extreme salinity (red-purple). Indeed,

581
the values of the centroids of the clusters representing these classes were considered; as well as, the standard 582 deviation value was chosen to limit the overlap between the classes considered and to reduce the chance of a pixel 583 being classified into more than one class. Fig. 11 shows the spatial distribution of salinity classes across the study 584 area and in the outer-peripheral regions of the pan. In general, it is observed that the both indices (SSSI and NDGI) 585 mapped the salinity patterns almost similarly by reflecting the results of the statistical fits discussed above.  Fig. 3), followed by the gypsum as a second most abundant crust (Table 2). Their EC-Lab values are ranging between 599 17.6 and 129.7 dS.m -1 , and the pH is greater than 8.2 reflecting a strong sodicity coupled with salinity.

600
Superimposed on salinity maps derived by NDGI and SSSI indices (Fig. 11), most of these points coincide perfectly the playa edges due to wind action and erosion, as well as by human movements (Bryant, 1996). Moreover, it is also 610 observed that high, very high and extreme salinity classes are associated with slightly high elevation.

611
According to mineralogical ground truth, the validation points P63, P64, P65 and 172 in the southern region of 612 the pan are dominate by the gypsum crust (33% to 83%) associated with a small amount of halite (5% to 36%). This 613 terrain truth is detected and well mapped by the two indices, but NDGI highlighted more the gypsum belt in south 614 and southwest (Fig. 11). In this region the topography is slightly high and decreases toward the centre-east of the 615 pan, and then it becomes relatively higher in the north and north-west. Points P66 and P67 located on a small 616 circular ridge in the south-central part of the pan with a slight elevation, have almost similar contents of halite and 617 gypsum (45%). However, the salt content in these two points is more stressed in the SSSI map. As well as, nearby 618 points 143 dominated by halite (52%) followed by gypsum (38%) and point 171 with 50% of halite and 27% of 619 quartz, the SSSI map shows more sensitivity to this class than that of NDGI (Fig. 11). The zone surrounding sample

674
VNIR according to their color and texture ( Fig. 5a and Table 1). Consequently, it is impossible to discern or to 675 separate between soil salinity classes in the VNIR. While when the EC-Lab values increased also the difference 676 among the salt-affected soil spectra's increased significantly and progressively in the SWIR (Fig. 5a). In this increasing values of EC-Lab (from 2.4 to 507 dS.m -1 ). Nevertheless, the noted ambiguity between "A" and "G"

711
Over the Omongwa salt-pan site, although the higher albedo of the site centre in the image is principally due to 712 halite crust developed and accumulated during many years as illustrated by true color composite RGB of Landsat-

713
OLI image (Fig. 3), the derived soil salinity maps using VI's are completely unable to detect the slightest grain of spectral, CCRS, 2D spectral-space, and statistical fits) based on simulated data. Obviously, these results were

763
The results show that the soil spectral signatures are very sensitive to soil salinity contents.