You are on page 1of 20

Physics and Chemistry of the Earth 110 (2019) 31–50

Contents lists available at ScienceDirect

Physics and Chemistry of the Earth


journal homepage: www.elsevier.com/locate/pce

Soil salinity mapping in Everglades National Park using remote sensing T


techniques and vegetation salt tolerance☆
Fahad Khan Khadima, Hongbo Sub,∗, Lina Xuc, Jing Tiand
a
Department of Civil and Environmental Engineering, University of Connecticut, 38 W Park Street, Willimantic, CT, 06226, USA
b
Department of Civil, Environmental and Geomatics Engineering, Florida Atlantic University, Boca Raton, FL, 33433, USA
c
Institute of Geophysics and Geomatics, China University of Geosciences, Wuhan, China
d
Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, 100101, China

ARTICLE INFO ABSTRACT

Keywords: Everglades National Park (ENP) is a hydro-ecologically enriched wetland with varying salinity contents, which is
Everglades national park a concern for terrestrial ecosystem balance and its nearby urban sustainability. In this study, spatio-temporal soil
Remote sensing salinity maps are created using remote sensing techniques, coupled with literature review to understand vege-
Geographic information system tation salt tolerance properties, and field assessments entailing insitu electric conductivity (EC) measurements.
Soil salinity mapping
The mapping first entailed the execution of a supervised machine learning technique—the maximum likelihood
Land cover classification
classification algorithm—to delineate seven vegetation-based land cover classes for the area, namely, mangrove
forest, mangrove scrub, low-density forest, sawgrass, prairies and marshes, barren lands with woodland ham-
mock and water, for the years 1996, 2000, 2006, 2010 and 2015. The classifications for 1996–2010 yielded
accuracies of 82%–94%, and the 2015 classification was supported through ground truthing. Afterwards, EC
tolerance thresholds for each vegetation class were established, which yielded soil salinity maps comprising four
soil salinity classes—i.e., the non- (EC = 0–2 dS/m), low- (EC = 2–4 dS/m) and high-saline (EC 4 dS/m)
areas. The soil salinity maps visualized the spatial distribution of soil salinity with no significant temporal
changes. Furthermore, insitu EC measurements carried out at 23 sampling sites covering all land cover classes
mostly validated (91% samples were tested within range) the estimated soil salinity ranges for the latest dis-
tribution. The approach of using land cover classes to sense salinity ranges in an urban-proximal ecosystem is
pragmatic and application oriented, attributing to novel and useful study upshots considering the diversifying
ecological context.

1. Introduction mangroves near the coastline, centering the sawgrass like freshwater
indicative species. Driven by these, it was attempted in this research to
The South Florida Everglades is a vast subtropical wetland with a create a local soil salinity dataset for the Everglades National Park
globally unique hydro-ecology, and is designated as an International (ENP) area using a passive-to-active approach in which comprehensive
Biosphere Reserve and a wetland of International Importance (Davis vegetation-based land cover classes were first delineated, and then
and Ogden, 1994). Since the start of 20th century, many anthropogenic using literature driven understanding on threshold soil salinity ranges
activities and natural disturbances have led to significant alterations to for each vegetation, salinity maps were developed and validated
existing natural hydrologic patterns, ecosystem balance, biogeochem- through insitu sensor deployment.
ical and urban environmental sustainability, which attributed to con- Making an assessment of land cover is a core field in remote sensing
tinuous saltwater intrusion from bay areas, yielding soil salinity pol- applications (King, 2002; Foody, 2002), and it is regarded as a funda-
lution. Soil salinity often causes other soil degradation phenomena such mental variable of the physical environment's impact on ecological
as soil dispersion, soil nutritional change, soil erosion, soil structure systems (Foody, 2002; Vitousek, 1994). Several studies have shown that
destruction, and some urban environmental issues (Metternicht and SPOT and Landsat images provide good distinctions of vegetative land
Zinck, 2003; Shober, 2009). Fluctuating salinity inside the Everglades cover classes (Ramsey and Jensen, 1996; Laba et al., 1997;
have shaped up variations in vegetation type—e.g., morphing a cover of Rasolofoharinoro et al., 1998; Aschbacher et al., 1995). There have also


Research Focus: Surface and sub-surface hydrology, geomatics and remote sensing.

Corresponding author.
E-mail addresses: fahad.khadim@uconn.edu (F.K. Khadim), suh@fau.edu (H. Su), silvaxu@sina.com (L. Xu), Tianj.04b@igsnrr.ac.cn (J. Tian).

https://doi.org/10.1016/j.pce.2019.01.004
Received 25 September 2018; Received in revised form 3 December 2018; Accepted 3 January 2019
Available online 07 January 2019
1474-7065/ © 2019 Elsevier Ltd. All rights reserved.
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

Fig. 1. Base map of the study area, Everglades National Park (ENP).

been attempts to use aerial photography to monitor high-resolution Karmaker (Mitra and Karmaker, 2010) carried out mangrove mapping
vegetation succession (Lucas et al., 2000). The use of radar technologies in the Sundarbans and attempted to draw linkages of mangrove extent
has been a useful innovation in delineating hydrogeological and and salinity ingress. However, these studies have focused more on
structural features in the ENP area (Aschbacher et al., 1995; Wdowinski modelling specific salinity properties with vegetation characteristics,
et al., 2004; Kasischke et al., 2003; Bourgeau-Chavez et al., 2005; rather developing spatio-temporally distributed soil salinity maps.
Proisy et al., 1996; Held et al., 2003). Radar data is also used in deli- The detrimental effects of salt on plants are a consequence of both
neation of flood zones in different forest canopies; however, the process water deficit resulting in osmotic stress and effects of excess sodium
falls short in dense areas (Held et al., 2003; Proisy et al., 2000; Wang ions on critical biochemical processes (Wyn Jones and Johnson, 1981).
and Imhoff, 1993). Models to identify scattering mechanisms and di- Typically, plant root cell membranes allow water to pass through,
rectly estimate the physical parameters of vegetation classes have also however, prevent salt from entering, making it difficult for saline soils
been developed (Proisy et al., 2000; Wang and Imhoff, 1993). More- to hydrate plant roots. Most landscape plants’ salt tolerance depend on
over, the mapping of vertical vegetation depths and structure has also the amount and duration of exposure as well as the concentration of
been carried out using radars (Mougin et al., 1999) and digital elevation salt, in a way that the higher the amount of salt in the soil, the greater
models (Simard et al., 2006). In addition to inspect land covers, there the impact on the plants (Appleton et al., 2009). Upon realization of
are also studies using remote sensing to relate vegetation cover changes this linkage between soil salinity and associated vegetation cover, the
with soil salinity (Allbed et al., 2014a, 2014b, 2017; Allbed and Kumar, novel concept of using vegetation classification of the ENP area to
2013). Allbed et al. (2014b) carried out a modelling of the spatial produce soil salinity maps was framed. The objective was to carry out
variation of salinity, using regression techniques between soil salinity soil salinity mapping for the ENP area using remote sensing techniques
indices and IKONOS image's band reflectance. In another study, Allbed through coupling vegetation information with insitu salinity values. Up
et al. (Allbed et al., 2017) analyzed normalized difference vegetation to the timeline of this research, no study has specifically engaged re-
index (NDVI) values and salinity index properties to monitor changes in mote sensing techniques to construct salinity maps for such an urban
soil salinity and vegetation cover from multispectral images. Mitra and influenced ecologically sensitive region covering a diversified series of

32
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

Fig. 2. Schematic flow diagram illustrating key activities of the research.

vegetation. Another salient significance of the study lies in its simplistic said years were collected from USGS database (Earth Explorer Websi,
approach in addressing and overcoming of complex challenges asso- 2017), which followed some preprocessing and supervised machine
ciated with remote sensing-based salinity mapping at such spatio-tem- learning (maximum likelihood classification algorithm) techniques, in
poral resolution (spatial: 30 m; and temporal: every 5–6 year starting the ENVI-IDL image processing program. The classification was carried
from 1990). The mapping indicates existing spatio-temporal trends in out using one image at a time rather than selecting a cluster of images
soil salinity at the ENP and may be useful owing to its greater geo- because of the limited availability of clear, cloud-free Landsat data.
ecological impacts, and the link with urban sustainability; also, the Upon inspection, only two Landsat images from each year of 1996,
methodological novelty tied with the process may further interest sci- 2000 and 2006 were found to have cloud covers of less than 10%,
entific community. The motivation of this research also serves ongoing whereas for 2010 and 2016, there was only one image from each year.
national interest, as the U.S. government has already allocated 7.8 Because of this and the primary objective of the study to eventually map
billion dollars for restoring the ecosystems of the Everglades, which is soil salinity, a property which takes longer time to be altered naturally
expected to increase the quantity, quality and timing of the freshwater (Wu et al., 2008), the decision of selecting only one image for land
reaching the Florida coast and, among other objectives, stabilize or cover classification was made. To further complement the process and
even increase the freshwater terrestrial habitats in the coming years assess the accuracies of the estimations, the derived land cover maps
(Simard et al., 2006; CERP Comprehensive Everglades Restoration Plan, were compared against the existing macroscale land cover maps from
2015). Nevertheless, wetland protection and soil salinity of the ENP the National Land Cover Database (NLCD) for 1992, 2001, 2006 and
area can be considered as an important environmental issue for the 2011 (MRLC, 2015). The NLCD maps are also useful because they
metropolitan area of Miami. provide a spatio-temporal understanding on land use changes, however,
the scope of this research is to delineate land cover classes which are
2. Materials and methods ecologically sensitive, rather than using the macroscale NLCD maps that
represent changing land use patterns from a mere surficial context.
2.1. Study area Afterwards, comprehensive literature review on potential soil sali-
nity clustering was made and the salt tolerance thresholds for different
The spatial domain of the study is confined within the ENP vegetation classes within the ENP area were identified. A two-tier field
boundary (Fig. 1). The ENP has been designated a World Heritage Site, investigation was carried out during the study, the first tier included an
International Biosphere Reserve, and Wetland of International Im- initial examination at selected field sites to perform groundtruthing
portance, which are significant to all people of the world (ebsite following the supervised land cover classification. The second tier
Everglades Na, 2015). From satellite images (Earth Explorer Websi, comprised two distinct field surveys under which a total number of 23
2017), the area enclosed by the ENP was estimated as 2410 square soil samples were collected and tested at site to measure the electric
miles (sqm). The area is bounded by Florida Bay and the Florida Keys in conductivity (EC) to validate the estimated soil salinity ranges of the
the south, the Gulf of Mexico in the west, the upper Everglades regions latest time frame and its spatial distribution. Fig. 2 shows a schematic
and Lake Okeechobee in the north and the Atlantic Ocean in the east. flow diagram that illustrates the key activities of the research.
The ENP area is also contiguous to the metropolitan area of Miami and
therefore possess substantial urban environmental significance. Fig. 1 2.3. Data
provides a base map of the study area, as well as 23 soil sampling lo-
cations randomly selected for insitu EC measurement while validating The Landsat TM 5 and Landsat OLI 8 images were collected from the
the latest soil salinity map. USGS data portal (Earth Explorer Websi, 2017; Data and Tools:, 2016)
for different dates—namely, February 18, 1996; February 14, 2000;
2.2. Approach May 4, 2006; November 7, 2010; and February 6, 2015. The TM images
contained six spectral bands—i.e., band 1: blue (0.45–0.52 μm), band 2:
The approach of the study is two-fold: first, the development of land green (0.52–0.60 μm), band 3: red (0.63–0.69 μm), band 4: near-in-
cover classification maps for 1996, 2000, 2006, 2010 and 2015, and frared (0.76–0.90 μm), band 5: near-infrared (1.55–1.75 μm), and band
second, the derivation of soil salinity maps by spatially integrating the 7: mid-infrared (2.08–2.35 μm) —as well as a distinct thermal band
information of the salt tolerance thresholds of each identified vegeta- (band 6). The Landsat OLI sensor has 11 bands in total, of which bands
tion class with the developed land cover maps. The vegetation classes 2–7 are similar in wavelength and resolution to the Landsat 5 TM sensor
used in the study for land cover mapping are inspired and infused from (What are the band de, 2017). The existing 30 m—resolution land cover
the detailed vegetation map developed by the Florida Coastal classifications were collected for an accuracy assessment from the
Everglades and Long Term Ecological Research (FCELTER) for 2000 NLCD, developed by the Multi-Resolution Land Characteristics Con-
(lorida Coastal, 2016). Landsat Thematic Mapper (TM) images for the sortium (MRLC) for years 1992, 2001, 2006 and 2011 (MRLC, 2015).

33
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

The 1992, 2001 and 2006 classifications have level—I accuracies of density scrub vegetation. Mangrove scrubs were distinguished from
83%, 85% and 84%, respectively (Stehman et al., 2003; and Cover mangrove forests in this study as having relatively lower densities and
https://cf), whereas the 2011 classification's accuracy assessment was heights below 5 m (MRLC, 2015). Simard et al. (2006) carried out a
still underway during the time of this research (and Cover https://cf). mangrove vegetation height mapping and inferred that the short
mangroves (termed “mangrove scrub” in this study) usually range
2.4. Land cover mapping process between 2 and 6 m in height and are located inward compared to the
high mangroves (termed “mangrove forest” in this study). The word
2.4.1. Identifying vegetation classes “scrub” has been used in this study to denote reduced development
As discussed above, the land use classes used in the NLCD classifi- under marginal conditions (Schaeffer-Novelli et al., 2000). There have
cations are more generic. In addition to these maps, an integrated effort not been studies to distinguish the salt tolerance of scrubs from large
from South Florida Natural Resources Center, ENP authorities, the mangrove trees; however, there are studies (Pezeshki et al., 1990;
Center for Remote Sensing and Mapping Science at the University of McKee, 1996) indicating that there are influencing factors that limit the
Georgia, Big Cypress National Preserve and the South Florida Water growth of mangrove scrubs into trees. The most dominating factors are
Management District developed vegetation classifications containing the reduced nutrient availability and other marginal conditions
forest covers, scrubs, savanna, prairies and marshes, shrublands, exotics experienced in scrubs. Laboratory studies have suggested that
and others (Jones et al., 2002). Both this integrated classification and mangrove growth differs in response to nutrient enrichment (Pezeshki
the NLCD maps are useful, however, for this study a finer and ecolo- et al., 1990; McKee, 1996). Flowers et al. (1977) carried out a detailed
gically discernible land cover classification was required. The FCELTER investigation, based on laboratory tests, on the relationships of plant
(lorida Coastal, 2016) map is the most detailed vegetation map found growth and salinity concentrations, and inferred that the salinity
for the ENP area covering a broad set of 12 vegetation classes, a concentrations inversely affect plant growth for a particular plant
modification of which led to the selection of seven final vegetation species in optimum surroundings, meaning that the more the increase
classes for this study—namely, the mangrove forest, mangrove scrub, in the optimum salinity, the lower the plant growth. Based on this
low-density forest, prairies and marshes, sawgrass, barren lands with information, it has been inferred that one of the major reasons for the
hammocks and water. The FCELTER map was carried out for a parti- dominance of mangrove scrub, with its reduced density and size, inside
cular time, which motivated this study to develop time-series maps the ENP boundary may be the reduction in salinity, which, as discussed
showcasing the temporal variation in addition to the spatial variation of above, limits nutrients and restricts plant growth. Another notable
selected landcover classes. The selected vegetation classes are discussed observation was made about the landscape of scrub plants and the
in the following paragraphs. surrounding physiography. The scrub regions were dominated by
inundated sloughs and water bodies, which provide substantial
freshwater recharge into the nearby areas during rainfall. This may
2.4.1.1. Mangrove forest. The mangroves of South Florida are mainly
be another factor limiting the soil salinity contents in these regions.
located inside national parks and other protected areas and are
Because of these issues, the optimum salinity condition for mangrove
therefore protected from the direct impact of urbanization (Simard
scrub may be slightly less than the high-salinity conditions considered
et al., 2006). Mangrove trees typically expand their ranges when
for mangrove trees.
natural and manmade events decrease freshwater flow from the
Everglades and increase salinities. During the classification, the
2.4.1.3. Sawgrass. Sawgrass (Cladium jamicense), which is a common
mangrove forest coverage was defined as high-density, salt-tolerant
characteristic plant species covering the largest extent of the ENP area,
stands of trees, typically with heights over 5 m (MRLC, 2015). The
constitutes about 65–70% of the total freshwater marshes (Steward and
mangroves in the ENP area include mangrove forest areas covering
Ornes, 1975). Sawgrass is a rhizomatous, perennial sedge, rather than a
black mangrove (Avicennia germinans), white mangrove (Laguncularia
grass, and can be of both long (1–3 m) and short (< 1 m) types (Davis
racemosa), red mangrove (Rhizophora mangle), mixed mangrove and
and Ogden, 1994). The areas identified as sawgrass included areas
buttonwood (Conocarpus erectus). Mangroves are facultative
covered with freshwater—indicative grass vegetation along the central
halophytes—i.e., salt water is not a physical requirement (Bowman,
and north-central regions of ENP, areas dominated by perennial
1917; Egler, 1948). In fact, most mangroves are capable of growing
herbaceous vegetation (vegetative cover of greater than 80%) and
quite well in freshwater (Teas, 1979). However, mangrove ecosystems
areas with periodically saturated or water covered soil extents.
do not develop in strictly freshwater environments, and salinity is
Sawgrass is a freshwater indicator and cannot grow on saline soils
important in reducing competition from other vascular plant species
(Miao, 2004).
(Kuenzler et al., 1974). Mangroves accommodate fluctuations and
extremes of water and soil salinity through a variety of mechanisms,
2.4.1.4. Prairies and marshes. Prairies (both wet and marl types) were
although not all mechanisms are necessarily present in the same species
another species considered along with the sawgrass, that constitute the
(Odum et al., 1982). Mangroves can sustain very high salinities, and
freshwater marshlands in the Everglades. Prairies exist in the east and
several studies indicate that the red mangroves can sustain salinities
west margins of the Shark River Slough and Taylor Slough, where the
around 65 parts per thousand (ppt)1 (Teas, 1979), whereas the black
bedrock elevations are higher and the hydroperiods are shorter (Davis
mangroves can grow at salinities greater than even 90 ppt (MacNae,
and Ogden, 1994). This land cover class included both prairies and
1969). Moreover, there are other factors affecting the plants’ salt
marshes, which are areas frequently exchanged with the areas of
tolerance—such as the soil type, temperature, rainfall, humidity, etc.
sawgrass cover—with dominating extents (over 80% vegetation
(Odum et al., 1982).
cover) of emergent herbaceous vegetation—and marsh wetlands.
Similar to the sawgrass type discussed in the previous paragraph,
2.4.1.2. Mangrove scrub. Mangrove scrub areas include the red prairies are also freshwater indicators and cannot be sustained in saline
(Rhizophora mangle), black (Avicennia germinans), white (Laguncularia environments.
racemosa) and buttonwood (Conocarpus erectus) types as well as high-
2.4.1.5. Barren lands with woodland hammocks. Upon a visual
1
unit ppt (parts per thousand) refers to the salinity of the surrounding water inspection of satellite images, a series of rough patches were observed
condition, which is identical to the grams per liter (g/l) unit—and is defined as along the north-central regions of the ENP. These areas are mainly
the amount of salts dissolved in a thousand units of water (e.g., 1 g of salt barren lands without much vegetation, and they are occasionally
dissolved in 1000 g of water yields a solution with 1 ppt salinity). covered by woodland forest. The hammock areas, both native and

34
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

nonnative, remain interspersed by freshwater sawgrass as well as with the highest probability (Equation (3)). The method is a well-
prairies. During classification, the barren lands were identified as known parametric classifier, with the advantage of taking into account
land without dominant vegetation or open canopy. The areas the variance–covariance within the class distributions (Erdas Inc. Erdas
included the existence of occasional woodland hammock trees, with Field Gu, 1999; Otukei and Blaschke, 2010).
less than 15% of total cover. The lands are freshwater regions, typically 1 1 1
at a higher elevation, and are not subjected to any tidal flooding gi (x ) = ln p ( i ) ln i (x mi)T i (x mi ).
(3)
2 2
inundations.
Here, i is the class, x is the n-dimensional data (n is the number of
2.4.1.6. Low-density forest. The low-density forest classes included the bands), p ( i ) is the probability that class i occurs in the image and is
pines (Pinus elliottii var. densa), palms, mixed cypress and exotics with assumed the same for all classes, i is the determinant of the covariance
heights above 5 m. The cluster was comprised of deciduous, evergreen matrix of the data in class i , and mi is the mean vector.
and mixed forest covers with more than 75% vegetation but with higher The selection of the maximum likelihood classifier, over some other
canopy coverage in comparison with mangroves. The low-density sophisticated classifier algorithms (support vector machine, decision
exotic species are likely to be sustained in minor salinity ranges, most trees, etc.); was preferred because the objective of the land cover
likely in between the freshwater indicator species (prairies and classification was to come up with a set of realistic and acceptable
marshes, sawgrass, hammocks) and marine coastal mangroves. vegetation classes that were to be used for soil salinity mapping. The
use of decision trees for classification requires accurate input in-
2.4.1.7. Water. The class entitled “water” included areas of all formation of spectral separability thresholds, also known as the deci-
distinguishable open water sources. As per the definition in NLCD sion boundaries. In this study, the decision tree classifier could also
(MRLC, 2015), water contents may be accompanied by less than 25% accurately identify water, light vegetation (grass) and dense vegetation
cover of vegetation and soil. The major share of water inside the ENP (forest areas) as these areas had significant differences in their surface
area comprises the Florida Bay area as well as other water courses in the reflectance thresholds (e.g., water has a very low reflectance in visible
form of lakes, sloughs and ponds The multispectral Landsat images bands, and by using this information, it can be classified easily in a
provide good accuracy in detecting the aerial extents of water bodies decision tree algorithm). However, no finer classifications could be
(Frazier and Page, 2000), and the maximum likelihood algorithm has carried out between the different types of forests in consideration
been identified as a useful tool for the process in many studies (mangrove trees, scrubs, hammocks) or grass species as the surface
(Manavalan et al., 1993; Lee et al., 1995; Blackman et al., 1995; reflectance characteristics of different bands for these areas were si-
Kingsford et al., 1997; Brady et al., 1999). The identified areas of water milar. This is a reason why the decision trees algorithm was eliminated
content were removed from the study area during the soil salinity from consideration. On the other hand, the support vector machine
mapping. algorithm performs better for hyperspectral images (Deilmai et al.,
2014). For multispectral images, both the maximum likelihood classi-
fication and support vector machine algorithms can provide good ac-
2.4.2. Land cover classification process
curacies when trained properly (Otukei and Blaschke, 2010). However,
Before the actual land cover classification, the images were pre-
because of the limited availability of ground information, established
processed through the radiometric calibration and atmospheric cor-
rule images, and compliance with the objective of the study, which is to
rection tools. The radiometric calibration was performed in ENVI-IDL2
infer salinity variations rather than depict finer land cover character-
to convert the spectral radiance into top-of-atmospheric (ToA) re-
istics, the use of the maximum likelihood classifier algorithm was em-
flectance (Equation (1)) for the Landsat TM sensors. The Fast Line-of-
phasized.
sight Atmospheric Analysis of Spectral Hypercubes (FLAASH) atmo-
To begin with the maximum likelihood classifier algorithm, eight to
spheric correction algorithm was applied to transfer ToA reflectance
ten training polygons were manually selected to best represent each
into surface reflectance properties (ModuleAtmospher, 2009). FLAASH
vegetation class. The selection was made from the best representative
starts from a spectral radiance equation at a pixel L to the solar wa-
areas of pixels, where a judgment on existing land cover could be made
velength range and flat, Lambertian materials or their equivalents
from visual observations. The areas were selected in a way so that they
(Equation (2)).
were sparsely distributed and represented pixels from all around the
Ld 2 Landsat image. The potential areas with clouds or other atmospheric
= .
ESUN sin (1) disturbances were avoided. Following the training polygon selection,
the maximum likelihood classifier algorithm was executed with the
A B e testing imageries. The classification was carried out using different
L= + + La .
1 eS 1 eS (2) probability thresholds, and a relatively fair number of classified pixels
were found using a threshold of 5%. The classified image still contained
Here, L is the radiance, d is the earth-sun distance, ESUN is the solar
some unclassified pixels, seemingly because the spectral characteristics
irradiance, is the sun elevation in degrees, is the pixel surface re-
of the training pixels were different from those pixels. In order to
flectance, e is an average surface reflectance for the pixel and a sur-
perceive information for such pixels, one field visit was carried out for
rounding region, S is the spherical albedo of the atmosphere, La is the
visual ground truthing, and contemporary information extracted from
radiance back scattered by the atmosphere, A and B are the coefficients
Google Earth images were also used.
that depend on atmospheric and geometric conditions but not on the
surface.
2.4.3. Groundtruthing
The land cover classification was then attempted using a supervised
Groundtruthing was performed to complement the machine learning
machine learning technique – the maximum likelihood classification
based image classification process with field based information and va-
method (Erdas Inc. Erdas Field Gu, 1999; Lu and Weng, 2007). Max-
lidation. To verify the land cover assessment and infer information on
imum likelihood classification algorithm (Richards, 1999) assumes
unclassified and misclassified pixels, five groundtruthing points were
normal statistical distribution for each class in each band, and then
selected, of which four locations were for visual observations with the
calculates a probability for each pixel to belong to the specific class
other one being for drone observation. As a follow-up, a field in-
vestigation was carried out by the researchers’ team on November 8,
2
More information can be obtained from https://www.harrisgeospatial.com/ 2016, in the ENP area. One major issue encountered during the field
docs/RadiometricCalibration.html. investigations was the limited accessibility and transportation options

35
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

Fig. 3. A snap from the study domain showing


the land cover classification from the Landsat
TM image through (a) a visual interpretation of
the color infrared image, (b) initial classification
using an image processing algorithm and (c)
incorporation of the final classification of
ground truthing information. (For interpretation
of the references to color in this figure legend,
the reader is referred to the Web version of this
article.)

Fig. 4. Photographs collected from the field investigation of the ENP area showing the (a) low-density forest (cypress), (b) mangrove scrub roots, (c) drone testing
and (d) an aerial photograph captured from the drone observations showing a portion of the mangrove area.

Table 1 unclassified pixels, as illustrated in Fig. 3. The example figure (Fig. 3) is


The vegetation classes and soil salinity ranges defined in this study. used to demonstrate how the process was carried out. The figure does not
No. Land cover class Salt Tolerance EC (dS/m) Soil salinity class
show the entire ENP area, rather a portion of the study area where five of
the total seven land cover classes were found. Fig. 4 shows some pictures
1 Mangrove Forest Tolerant 4 High-saline captured during the field investigation for the ground truthing purpose of
2 Mangrove Scrub the land cover assessment.
3 Low Density Forest Moderately Sensitive 2–4 Low-saline
4 Prairies and Sensitive 0–2 Non-saline
Marshes
5 Sawgrass 2.4.4. Post-classification process
6 Barren lands A series of post-classification steps were carried out in ENVI-IDL3 to
7 Water N/A N/A N/A
clean and soothe the classified images so as to remove noise and iso-
lated pixels. First, the images were sieved and filtered using the “sieve
analysis” and “majority filter” post-classification tools. Afterwards, the
inside the ENP area. To address this problem, a DJI Phantom 3
Professional Quadcopter 4K UHD Video Camera Drone was used at one
location to capture a high-resolution image. The information collected 3
More information can be obtained from https://www.harrisgeospatial.com/
through drone and visual observations helped in classifying most of the docs/ClassificationTools.html.

36
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

Fig. 5. Vegetative land cover classification map for 1996.

smaller, adjacent class regions were aggregated using the “classification so, 15 random pixels were identified from each of the seven vegetation
aggregation” tool with a threshold pixel number of 200. This removed classes (mangrove forest, mangrove scrub, low-density forest, prairies
the isolated pixels with linear dimensions of less than 6 km and marshes, sawgrass, barren lands and water) from each of the
(=200*30 m). Therefore, even though the mapping resolution was aforementioned years’ images. The identified classes for those pixels
30 m (due to the Landsat images’ spatial resolution of 30 m), the were then tested against the existing classifications in the same pixel
minimum size of the adjacent land cover classes identified in the final locations from the NLCD classes for 1992, 2001, 2006 and 2011, re-
land cover maps was 6 km, and any adjacent classes with lower size spectively. Due to the unavailability of information pertaining to any
were ignored. The reason for such an aggregation process was that the rule image classification for 2015, or any other relevant years, the ac-
primary objective of the study was to produce soil salinity maps and not curacy of the land cover classification for 2015 could not be assessed.
to produce high-resolution land cover maps. Therefore, the spatial de- However, as the land cover classification for 2015 was supported by
tails for some locations were compromised in order to generate spatially ground truthing attempts laid in the field, as well as through a Google
consistent soil salinity maps. Soil salinity is not a phenomenon that Earth satellite image observation, the issue could be ignored.
fluctuates drastically over time and space but rather a gradually evol-
ving natural process. 2.5. Salinity assessment framework

2.4.5. Accuracy assessment The key component of the study was to delineate soil salinity maps
Following the image classification and post-classification processes, for the years 1996, 2000, 2006, 2010 and 2015. To do so, the developed
the accuracies of the land cover classifications for 1996, 2000, 2006 land cover classification features and their soil salinity thresholds were
and 2010 were tested against a set of randomly selected points. To do used as inputs. There are some standards for soil salinity classifications,

37
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

Fig. 6. Vegetative land cover classification map for 2000.

mostly based on the varying extent of electric conductivity (EC) existing Total salinity (ppt) = EC (dS/m) * 0.80; for EC > 5 dS/m (5)
in soils. EC is a common field level indicator of soil salinity measured in
Lamond and Whitney (1992) classified soil into five salinity classes-
units of dS/m, which is correlated with the other conventional total
namely, low (EC = 0–2 dS/m), moderate (EC = 2–4 dS/m), high
salinity unit, ppt. In order to convert values of dS/m to ppt, a site-
(EC = 4–8 dS/m), excessive (EC = 8–16 dS/m) and very excessive
specific relationship needs to be developed either using deterministic or
(EC > 16 dS/m). Miller and Donahue (1995), VRO (Victoria Resources
stochastic approaches (Rhoades and Chanduvi, 1999). The United
Online, 2012) and FAO (Abrol et al., 1988) all used the same classifi-
States Department of Agriculture (USDA) in one of their field manuals4
cations but changed the titles for each classification. Miyamoto et al.
provides following two empirical relationships to convert dS/m into ppt
(2004) and Shober (2008) provided different classification thresholds
(Equations (4) and (5)), the latter unit is a representation of total dis-
of soil salinity, providing more emphasis on plant tolerance, and the
solved salts. In this study, EC was used as the soil salinity indicator
thresholds were more locally suited for South Florida. These researchers
considering its better convenience in field measurements (Rhoades and
classified plant tolerance of salinity into five categories, and the EC
Chanduvi, 1999).
thresholds for each category were also different from the other classi-
Total salinity (ppt) = EC (dS/m) * 0.64; for EC between 0.1 and 5 dS/m fications discussed above. The latter classification entailed plant toler-
(4) ance clusters termed as sensitive (EC < 3 dS/m), moderately sensitive
(3–6 dS/m), moderately tolerant (6–8 dS/m), tolerant (8–10 dS/m) and
highly tolerant (> 10 dS/m).
As the soil salinity mapping carried out in this study depended on
4
More information can be obtained from https://prod.nrcs.usda.gov/ the spatial pattern of the existence of different vegetation, the salt
Internet/FSE_DOCUMENTS/nrcs144p2_067096.pdf.

38
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

Fig. 7. Vegetative land cover classification map for 2006.

tolerance-based classifications of different plants were used to map the (1992) were merged together to form the high salinity class ( 4 dS/m)
soil salinity. And because of the types and numbers of classes identified in the study.
in the ENP area, the sensor limitation and the limited accessibility in Mitra and Karmaker (2010) carried out laboratory tests for the
the ENP areas, the selection of salinity classes in this study were re- Sundarban area to infer the EC ranges for mangrove scrub and marshes
stricted to three, e.g. non-saline (EC = 0–2 dS/m), low saline of around 3.5–5 dS/m. Because of this, and the proclivity of mangroves
(EC = 2–4 dS/m), and high saline (EC 4 dS/m) soils. The salinity to be sustained in extremely saline conditions (Mitra and Karmaker,
level classes used in the study have been modified from the classifica- 2010; Teas, 1979; MacNae, 1969), the mangrove forest vegetation and
tions used by Lamond and Whitney (1992), Miller and Donahue (1995), mangrove scrub land cover classes has been classified in the high-saline
VRO (Victoria Resources Online, 2012) and FAO (Abrol et al., 1988), (EC 4 dS/m) category. Due to the existence of exotics in the low-
only with some exceptions. The first two salinity classes in this study, density forest groups and their proclivity to grow in relatively more
namely, non-saline (EC = 0–2 dS/m) and low saline (EC = 2–4 dS/m) saline conditions, as compared to the freshwater sawgrass species, their
classes are in accordance with the other established salinity classes as salinity class has been assigned as having low-saline values
discussed before. However, due to the poor accessibility inside the (EC = 2–4 dS/m), between the high- and non-saline areas. Last, the
mangrove areas which hampered the insitu data sampling to a con- sawgrass, prairies and marshes and barren lands with woodland ham-
siderable extent, and the limitation of the salinity sensor which had an mocks are known to either be or contain freshwater species, and as
upper threshold of 4 dS/m, more detailed salinity classifications within such, their salinity class has been assigned as non-saline (EC = 0–2 dS/
the mangrove areas could not be attempted. Therefore, all the high, m). The final information on vegetation and soil salinity classes has
excessive and very excessive classes as labelled by Lamond and Whitney been summarized in Table 1, which also shows the relative salt

39
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

Fig. 8. Vegetative land cover classification map for 2010.

sensitivity for each vegetation class as well as the EC levels selected for HI98331” sensor was injected to measure the bulk EC. The upper range
the study. Moreover, all the soil salinity estimates in this study indicate of the sensor was 4 dS/m, with measurement accuracies ranging
the top-soil salinity. Chimney and Goforth (2006) indicated that in the from ± 0.05 mS/cm (0.00–2.00 mS/cm), ± 0.30 dS/m (2.00–4.00 dS/
ENP area the well-drained topsoil may contain layer thickness of up to m). Three measurements were taken at each sampling location, and the
30-inch. In order to represent this, soil sampling during field validation mean value was taken into consideration. The soil salinity sampling
in this study was carried out at 10-inch depth from surface. locations are already shown in Fig. 1.

2.6. Field validation for estimated salinity 3. Land cover mapping

Two separate field investigations dated October 1, 2017, and June 3.1. Land cover classifications for 1996–2015
17, 2018, were carried out in the ENP area which yielded the collection
of 23 soil samples and their insitu measurement for EC. Considering the The spatial distribution of the different categories of selected ve-
limited accessibility inside the ENP, the locations were selected pri- getation for the study area, as identified in section 2.4.1, is shown in
marily based on their proximity from either the nearby roads or the Figs. 5–9 below. The maps were produced while undergoing a series of
coast. Also, emphasis was given so as to make sure that the spatial supervised classification and post-classification steps, thus maintaining
distribution of the sampling locations covers the entire ENP area, spe- a balance between technical accuracy and pragmatic applicability. As
cifically the spatial variability of vegetation. During all the measure- already discussed, the objective of the study is to carry out indicative
ments, a uniform 10-inch hole was dug through the moistest soil spots mapping of soil salinity for the ENP area, for which the spatial existence
available, inside which a portable “Soil Test™ “Direct Soil EC Tester - of isolated pixels having linear dimensions below 6 km (200 pixels),

40
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

Fig. 9. Vegetative land cover classification map for 2015.

Table 2 located within the Florida Bay. Of all of the distinguishable soil vege-
List of NLCD classes and subsequently matched subclasses (vegetation classes). tation classes, sawgrass-covered areas were found to entail the largest
NLCD Classes within ENP Vegetation classes
vegetation extent among all, covering 440 sqm in areas in the north-
central and center regions of the ENP area. The second most dominant
Emergent Herbaceous Wetlands Barren lands, prairies and marshes, sawgrass, vegetative land cover was the mangrove scrubs, covering around 370
low-density forest sqm. The layer of mangrove scrubs was surrounded by a coastal layer of
Woody Wetlands Mangrove forest, mangrove scrub
Deciduous Forest Barren lands, low-density forest
dense mangrove forests, which covered 281 sqm. Furthermore, there
Evergreen Forest Barren lands, low-density forest were around 199 sqm of barren lands, 113 sqm of prairies and marshes
Shrub Prairies and marshes, sawgrass, low-density and around 29 sqm of savanna pine, palm, cypress and other types of
forest low-density forests.
Grasslands Prairies and marshes, sawgrass, low-density
The vegetation map for February 2000 (Fig. 6) produced similar
forest
Water Water results to Fig. 5, with a distinguishable increase of up to 559 sqm in
Developed Space N/A sawgrass extents. This increase was complemented by the reductions of
the spatial extents of barren land and prairies and marsh classes by up
to 126 sqm and 69 sqm, respectively. The low-density forest extents
representing discrete land cover classes, were compromised. remained nearly unchanged (30 sqm). The other salient land cover
For the period of February 1996 (Fig. 5), approximately 959 sqm features-i.e., the mangrove scrub and mangrove forest extents-experi-
(39.8% of total area) of coastal water bodies were found inside the ENP enced moderate increases of up to 393 sqm and 291 sqm, respectively.
area, and the largest share of which was the shallow coastal embayment The land cover mapping for 2006 was carried out for a wet season

41
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

Table 3
Confusion matrix showing the accuracies (%) of the land cover classifications for different years, using random training points.
Selected Class Sample Points Accepted Classes Rejected Classes Accuracy

WW EHW DF/EF S/G W WW EHW DF/EF DS W

Year 1996 82%


Barren lands 15 3 8 4 73%
Prairies and marshes 15 3 10 1 1 87%
Sawgrass 15 7 5 3 80%
Low density forest 15 4 4 3 4 73%
Mangrove scrub 15 12 2 1 80%
Mangrove forest 15 13 2 87%
Water 15 14 1 93%
Year 2000 85%
Barren lands 15 15 100%
Prairies and marshes 15 14 1 93%
Sawgrass 15 11 4 73%
Low density forest 15 4 4 3 4 73%
Mangrove scrub 15 12 3 80%
Mangrove forest 15 15 100%
Water 15 12 2 1 80%
Year 2006 94%
Barren lands 15 15 100%
Prairies and marshes 15 14 1 93%
Sawgrass 15 15 100%
Low density forest 15 5 5 5 100%
Mangrove scrub 15 10 4 1 67%
Mangrove forest 15 13 1 1 87%
Water 15 14 1 93%
Year 2010 92%
Barren lands 15 15 100%
Prairies and marshes 15 14 1 93%
Sawgrass 15 14 1 93%
Low density forest 15 6 4 2 2 1 80%
Mangrove scrub 15 14 1 93%
Mangrove forest 15 14 1 93%
Water 15 15 14 1 93%

WW = Woody Wetlands, EHW = Emergent Herbaceous Wetlands, DF = Deciduous Forest, EF = Evergreen Forest, S = Shrub, G = Grassland, DS = Developed
Space, W = Water.

Table 4
Changes in land cover classes over the years.
LC Class Area in sqm (and in %)

Feb 1996 Feb 2000 May 2006 Nov 2010 Feb 2015

Barren lands 199 (8.3%) 126 (5.3%) 219 (9.1%) 165 (6.9%) 179 (7.4%)
Prairies and marshes 113 (4.7%) 69 (2.9%) 109 (4.5%) 94 (3.9%) 70 (2.9%)
Sawgrass 440 (18.3%) 559 (23.2%) 454 (18.8%) 502 (20.8%) 490 (20.3%)
Low density forest 29 (1.2%) 30 (1.2%) 23 (1.0%) 35 (1.5%) 37 (1.5%)
Mangrove scrub 370 (15.4%) 393 (16.3%) 351 (14.6%) 406 (16.9%) 393 (16.3%)
Mangrove forest 281 (11.7%) 291 (12.1%) 314 (13.0%) 245 (10.2%) 278 (11.5%)
Water 959 (39.8%) 937 (38.9%) 932 (38.7%) 955 (39.6%) 907 (37.6%)
Unclassified 19 (0.8%) 5 (0.21%) 8 (0.3%) 8 (0.3%) 56 (2.3%)
Total 2410 (100%) 2410 (100%) 2410 (100%) 2410 (100%) 2410 (100%)

image (May 2006), in consideration of the overall spectral quality (low ones observed in 2006, as prairies and marshes reduced to 94 sqm,
cloud cover and high visual extent) of the image. The land cover classes barren lands reduced to 165 sqm and mangrove forests reduced to 245
for the year 2006 are shown in Fig. 7. The figure shows a significant sqm. In the latest land cover map produced for 2015 (Fig. 9), minor
conversion of sawgrass areas into barren lands (219 sqm) and prairies changes were observed in comparison to 2010. The dense mangrove
and marshes (109 sqm). The sawgrass areas for 2006 were reduced to forest extent increased up to 278 sqm, whereas the mangrove scrub
454 sqm. There were significant reductions in low-density forest (23 areas dropped to 393 sqm. The extent of prairies and marshes reduced
sqm) and mangrove scrub (351 sqm) areas. Such differences may be due to 70 sqm, whereas the other land cover features remained almost the
to the changes in the deciduous types of vegetation. Unlike the man- same (barren lands covered 179 sqm, sawgrass covered 490 sqm and
grove scrub areas, the other saltwater intrusion indicator species-i.e. the low-density forests covered 37 sqm).
dense mangrove forest coverage-increased up to 314 sqm.
The vegetation map produced for 2010 (Fig. 8) shows a significant 3.2. Accuracy assessment: confusion matrix
increase in mangrove scrub areas (406 sqm), as well as freshwater
sawgrass coverage (502 sqm), compared to 2006. An increasing extent Following the delineation of the land cover maps for different time
was also found for low-density forest areas (35 sqm). All of the other periods over the last 20 years, an attempt was made to assess the ac-
land cover classes decreased in their spatial extents compared to the curacy of the supervised classification process in quantifiable terms.

42
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

Fig. 10. Top-soil (validated at 10-inch depth) salinity map for 1996.

The process entailed comparing the existing land cover classifications of classification for the ENP area differed for this study, compared to the
the NLCD, developed by MRLC (MRLC, 2015). The MRLC produced NLCD classifications. Therefore, the comparison of two land cover
land cover classifications for the entire United States, for the years classifications was not straight forward, as the classification inferred in
1992, 2001, 2006 and 2011. The data sets were downloaded and this study yields more emphasis on the ecological adaptability and
clipped for the ENP area, and this study's land cover classes were then salinity tolerance of vegetation species. On the other hand, the NLCD
compared with the aforementioned classifications from the NLCD. The classification was more generic and inclusive and was better suited for
land cover classifications developed in this study for 1996, 2000, 2006 describing the land cover patterns of the entire country. Considering the
and 2010 were compared with the NLCD's classifications for 1992, ecological adaptability and salinity tolerance characteristics of the land
2001, 2006 and 2011, respectively. The land cover classification for cover classes selected in this study, coupled with the visual inter-
2015 could not be compared, as no such land use classifications or pretation of the NLCD's classifications, a number of the NLCD's classes
vegetation maps for the same temporal and spatial domain exist. were considered to encompass a set of subclasses, which were indeed
In order to assess the classification accuracy, a set of 105 sample the seven land cover classes developed in this study. Table 2 shows that
locations were randomly generated for each land cover classification, the sets of vegetation classes matched with the existing NLCD classes.
precisely selecting 15 random points for each of the seven land cover The table infers that, for example, if the randomly selected training
classes (sawgrass, low-density forest, mangrove forest, mangrove scrub, point falls in the emergent herbaceous wetlands class from the NLCD
barren lands, prairies and marshes and water). For the same locations, classification and, at the same time, falls in either the barren lands,
representative land cover samples were extracted from both this study's prairies and marshes, sawgrass, or low-density forest classes from this
land cover classification products as well as from the NLCD's products. study (Table 2), then the location-specific classification will be accepted
However, the vegetation classes used to define the existing land cover as correct. If the point infers any class other than those included in the

43
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

Fig. 11. Top-soil (validated at 10-inch depth) salinity map for 2000.

table, the classification will be rejected for that location. vegetation classes, post-processing and refining of an image, etc.).
Table 3 provides a confusion matrix that shows the accuracy of each However, going into that detail is beyond the scope of this research as
classified land cover class. The accuracy here is quantitatively defined the primary objective of the study was to carry out soil salinity map-
as the percentage of total random training points from each land cover ping, a phenomena which is typically not as much spatially detailed as
class that yielded acceptable classes from the NLCD classes. The table land cover classifications used as the indicative inputs. As such, the
shows that the land cover classification carried out in this study yielded obtained land cover classification accuracies for each year's maps were
accuracies in the range of 82%–94%, as compared to the NLCD classes. accepted for the purpose of the study.
Albed et al. (Allbed et al., 2017) in their study incurred an overall ac-
curacy of 93% while developing soil salinity maps in the Al Hassa Oasis 3.3. Trend of land cover variation
in Saudi Arabia, which falls within the range of this study. However, the
mapping accuracies obtained in this research is relatively broad To investigate the trends of land cover changes over past 20 years,
(82–94%) it is difficult to reason to what extent this accuracy metric is the spatial extents for each class observed during different time periods
useful to validate the salinity mapping efforts carried out in this study, were compared (Table 4). The comparisons showed a stable land cover
as there might be additional propagation of uncertainties especially due characteristic. In each classification, changes were found in spatial
to the consideration of NCLD classifications having 83%–85% ac- extents, as compared to previous year's extents; however, with no de-
curacies as references to this study's validation. This error in landcover finite trends. ENP is a naturally conserved area, where no major land
mapping could be better investigated with selection of more soil sam- cover changes were observed for past 20 years, with the total amount of
ples, and the sensitivity testing of land cover classification performance salt-tolerant mangrove areas remained around 26–27% in the last 20
with partially restraining some of the study assumptions (aggregation of years. The freshwater indicative vegetation covers also remained stable.

44
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

Fig. 12. Top-soil (validated at 10-inch depth) salinity map for 2006.

4. Soil salinity mapping for ENP nearby land cover's salinity ranges were used.
The indicative maps visualized the spatial distribution of soil sali-
4.1. Soil salinity maps for 1996–2015 nity in the ENP area. The existence of high-saline regions encompassed
the entire ENP area, guarding the non-saline and low-saline regions in
Using the existing land covers-the following top-soil salinity maps the central ENP areas. The position of low-saline areas near the city
were produced for 1996–2015 (Figs. 10–14). During the mapping, the built-up areas can be explained by the significant amount of ground
extents of the prairies and marshes as well as the sawgrass and barren water abstraction so as to meet regular city dwelling requirements,
lands in the central and north-central portions of the ENP area were which can yield water salinity intrusion through the subsurface layers.
considered to have EC values less than 2 dS/m, which categorized the Moreover, the existence of the high-saline region can be justified by the
soil as non-saline. In comparison, the lands with low-density forest surrounding coastal water bodies and the internal lakes, sloughs and
(savanna palm, pines, cypress etc.) were selected as of having little salt marshy wetlands. The major source of soil salinity in the Everglades is
contents (EC values ranging from 2 to 4 dS/m); as such, forest covers the salt contents that are transported through the water courses, leaving
with extended canopy coverage can sustain these minor salinity levels. the peripheral soil areas as a high-saline region. The extents of fresh-
This land portion was classified as a low-saline area and was discretely water indicative plants and land covers-i.e., the sawgrass, prairies and
located in the eastern side of the ENP, adjacent to Miami Dade built-up marshes and barren lands-can be justified due to their reduced direct
area. The two mangrove classes-i.e., the mangrove scrub and the interference with the brackish water system, coupled with the sig-
mangrove forest areas typically indicate high salinity tolerances; nificant annual freshwater recharge received from the upper Everglades
therefore, their extents were converted to the high salinity (EC equal or areas.
above 4 dS/m) group. For the unclassified regions in Figs. 5–9, the most

45
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

Fig. 13. Top-soil (validated at 10-inch depth) salinity map for 2010.

4.2. Soil salinity trends for 1996–2015 areas’ coverage (low and high) increased from the observations made
for November 2010.
Following the production of the soil salinity maps shown in Moreover, the soil salinity trends observed in the ENP area have
Figs. 10–14, an assessment was carried out to analyze the temporal been stable for all of the observations. This is understandable as the
trends of the coverage of different soil salinity areas. Fig. 15 provides a study area experiences naturally preserved vegetation covers and is free
comparative assessment of the fluctuations of the total areas for the from abrupt anthropogenic disturbances that might have induced the
different soil salinity classes for the five mapping instances in higher salinity alterations within the area. Hydrological changes in the
1996–2015. The analysis shows that, in the last 20 years, the areas of last 100 years have significantly impacted the Everglades (Sklar et al.,
different soil salinity level classes have remained fairly stable, experi- 2002; Light et al., 1994). Furthermore, starting in the early 1950s, the
encing minor rises and drops in different temporal estimations. The U.S. Army Corps laid a series of construction programs (canals, levees,
values obtained for May 2006 indicate a difference in values from the dams and pump stations) that threatened the natural balance of the
other years' observations. This may be because the season in which the system (Douglas et al., 2007). Since the mid-1940s, the boundary of the
data was collected was different (wet season) compared to that of the mixed graminoid-mangrove and caladium communities has shifted in-
other observations (dry season). This might leave a portion of areas land by around 3.3 km, whereas the interior boundary of a low-pro-
covered in deciduous trees classified as either sawgrass, prairies and ductivity zone appearing white on both black-and-white and CIR photos
marshes or barren lands (which are non-saline area-indicators) by the moved inland by 1.5 km on average (Ross et al., 2000). The Everglades
satellite, thus marking a sudden rise in non-saline areas to 782 sqm. is a system that depends not only on the flow of water from Lake
This was complemented by the reductions of low- and high-saline areas Okeechobee into the park but also on the Kissimmee River, which feeds
for that year. For the latest observation in February 2015, the saline the lake. Based on this realization, the Army Corps of Engineers

46
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

Fig. 14. Top-soil (validated at 10-inch depth) salinity map for 2015.

the recent soil salinity maps.

4.3. Salinity validation results

Attempting to validate the latest spatial distribution (Fig. 14) of soil


salinity, the 23 soil samples were tested at site during the two field
surveys on October 1, 2017 and June 17, 2018. From the land cover
classifications derived in section 3.1, it can be found that 11 sampling
sites were located within the Non-Saline land cover classes (sawgrass,
prairies and marshes, barren lands with hammock); 3 sites were located
in the Low-Saline land cover classes (Low Density Forest); 7 sites were
located in the High-Saline land cover classes (Mangroves). The insitu
test results show that these 21 sites of the 23 sampling locations have
Fig. 15. Soil salinity variation from 1996 to 2015. EC values within the estimated salinity ranges. The two points (ID = 5
& 6) which do not comply with the estimated salinity ranges actually
changed its focus from structural developments to environmental sus- show a very large bias from the range threshold. These two points were
tainability, thus forming a stable ecological system in recent years. This initially chosen from google earth imageries, but upon field inspection
was visualized in the soil salinity maps developed in Figs. 10–14, as it was found that the points were located inside coastal wetlands and
there were not many significant salinity intrusion trends observed for the sampling sites were not dry, which might have affected its soil EC

47
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

Table 5
Field validation of latest estimated soil salinity.
ID Lat Lon Land cover class Measured Salinity (dS/m) Estimated Range (dS/m) Assessment Result

1 25.760 −81.059 Low Density Forest 2.1 2–4 Accepted


2 25.423 −80.679 Low Density Forest 2.3 2–4 Accepted
3 25.429 −80.772 Low Density Forest 2.2 2–4 Accepted
4 25.339 −80.811 Prairies and Marshes 0.4 0–2 Accepted
5 25.137 −80.933 Coastal Wetland 0.2 4 Rejected (observed salinity well below estimated range with about
95% bias)
6 25.141 −80.917 Coastal Wetland 0.5 4 Rejected (observed salinity well below estimated range with about
87.5% bias)
7 25.187 −80.894 Mangrove Forest 4 4 Accepted
8 25.215 −80.850 Mangrove Scrub 4 4 Accepted
9 25.254 −80.798 Mangrove Scrub 4 4 Accepted
10 25.411 −80.542 Sawgrass 0.09 0–2 Accepted
11 25.814 −81.396 Mangrove Forest 4 > =4 Accepted
12 25.832 −81.348 Mangrove Forest 4 > =4 Accepted
13 25.825 −81.333 Mangrove Forest 4 > =4 Accepted
14 25.692 −81.124 Sawgrass 0.15 0–2 Accepted
15 25.788 −81.224 Sawgrass 0.2 0–2 Accepted
16 25.601 −80.937 Barren lands with 0.33 0–2 Accepted
Hammock
17 25.589 −80.978 Barren lands with 0.25 0–2 Accepted
Hammock
18 25.668 −80.730 Barren lands with 0.04 0–2 Accepted
Hammock
19 25.445 −80.607 Prairies and Marshes 0.04 0–2 Accepted
20 25.532 −80.695 Sawgrass (grass land) 0.04 0–2 Accepted
21 25.492 −80.636 Sawgrass (grass land) 0.11 0–2 Accepted
22 25.388 −80.732 Prairies and Marshes 0.47 0–2 Accepted
23 25.214 −80.851 Mangrove Scrub 4 4 Accepted

All soil samples were collected from 10-inch depth, and each value indicates a mean average of three successive insitu measurements at the same location.

values. Nevertheless, about 91% of the samples show good agreement collected with the Landsat 1–3 Multispectral Scanner sensors could not
with estimated results which is a major finding. Table 5 shows detail be used in this study due to the inconsistent information associated with
information on the sampling locations and salinity assessment valida- them, and the Landsat 7 ETM sensors were avoided due to the existence
tion result. of occasional dark patches in the satellite images. This further limited
the availability and implications of satellite data, which is a funda-
5. Discussions and limitations mental input of the soil salinity mapping.
The confusion matrix produced classification accuracies of
The vegetation classes and their spatial extents were found to be 82%–94%, which can be a factor of the several pragmatic assumptions
consistent in comparison with the existing detailed vegetation map for made in the study, coupled with the execution of a post-classification
2000, which was developed by the Florida Coastal Everglades Long aggregation process to compromise finer details for considering rela-
Term Ecological Research (lorida Coastal, 2016). The soil salinity maps tively coarse threshold for feature detection (6 km). The post-classifi-
provided good understanding of the spatial distributions of non-, low- cation aggregation carried out in this study eliminated the identifica-
and high-saline areas in the ENP, even though no significant temporal tion of smaller (less than 6 km dimension) isolated feature, and hence
salinity variations were found. This phenomenon indicates the ecolo- compromised finer details during vegetation mapping. However, for
gically stable and semi-urbanized nature of the ENP area and the as- predicting the soil salinity information, which is the primary objective
sociated natural preservation attempts of the area by different line of the study, this limitation can be overlooked. Soil salinity is a rela-
agencies and developers. In the most recent salinity map developed for tively steady and inclusive feature compared to water salinity, and
2015, areas of about 708 sqm were found to be affected by different temporal undulations in soil salinity values are typically less.
ranges (low and high) of salinity, leaving around 739 sqm as non-saline In this study, the construction of salinity maps depended on the
areas. During the time series salinity mapping from 1996 to 2015, no extents to which the indicative ranges of soil salinity have been accu-
major salinity fluctuations were observed, stressing the importance of rately predicted and validated from the relative salt tolerance threshold
considering larger temporal scales to assess a long-term salinity fluc- ranges of each vegetation class. The method chosen to predict such
tuation. Usually, the soil salinity beside the coast is higher and Allbed thresholds is based on a comprehensive review of literature and is also
et al. (2014b) in their study found EC values above 16 dS/m at man- constrained by the limited scientific exposure of such salt-threshold
grove areas beside a gulf coast in Saudi Arabia – qualitatively, this ranges of different vegetation classes being considered, especially for
understanding of having high-saline mangroves along the periphery of the ENP area. Due to time-labor constraints, dry weather scarcity, and
the coastal area was in agreement with the findings of this study. overall a limited accessibility inside the ENP areas, field investigations
The remote sensing based analysis in this study included extensive were largely hampered and delayed. The method deployed in the study
processing of satellite images, for which data availability is always a can only predict and visualize salinity ranges and not gradient-based
substantial concern. For each of the years of 1996, 2000, 2006, 2010 values for different locations. The EC sensor used for insitu validation
and 2015, only one or two satellite images were found to be consistent was constrained with an upper measurement threshold of 4 dS/m,
in terms of acceptable cloud coverage, atmospheric noise and other for which further higher and finer salinity clusters could not be de-
detrimental effects. For 2006, no dry season observation could be car- veloped or evaluated. The scope of the study was also specifically fo-
ried out unlike the other years, as only one image in the month of May cused on developing a data-driven machine learning based method to
was found to be satisfactory. This is a reason why the use of Landsat imply remote sensing tools to capture indicative estimates for soil
images was limited to only one image per year. Moreover, data sets salinity, without scientific considerations for hydro-climatic

48
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

variabilities and trends of sea level rise which might have influenced Bourgeau-Chavez, L.L., Smith, K.B., Brunzell, S.M., Kasischke, E.S., Romanowicz, E.A.,
the time series soil salinity mapping. As for future research, information Richardson, C.J., 2005. Remote monitoring of regional inundation patterns and hy-
droperiod in the greater everglades using synthetic aperture radar. Wetlands 25,
pertaining to climate change and subsequent predictions regarding sea 176–191. https://doi.org/10.1672/0277-5212(2005)025[0176:RMORIP]2.0.CO;2.
level rise can be incorporated with the present analysis to investigate Bowman, H.H.M., 1917. Ecology and physiology of red mangroves. Proc. Am. Phil. Soc.
future soil salinity ingress phenomena, as it is believed that sea level 56, 589–672.
Brady, A., Shaikh, M., King, A., Sharma, P., 1999. Remote sensing and the Great Cumbung
rise and its potential acceleration with global warming possess the Swamp. Wetl. Aust. 7, 596–606.
threats of pushing mangrove forests further inward. CERP Comprehensive Everglades Restoration Plan U.S. Army Corps of Engineer and South
Florida Water Management District, Florida. http://www.evergladesplan.org/,
Accessed date: 23 December 2015.
6. Conclusions Chimney, M.J., Goforth, G., 2006. History and description of the Everglades Nutrient
Removal Project, a subtropical constructed wetland in south Florida (USA). Ecol. Eng.
Salinity ingress has widespread effects on terrestrial vegetation, but 27 (4), 268–278.
USGS USGS Data and Tools: GIS Data. https://www.usgs.gov/products/data-and-tools/
it also impacts ecosystem balance and contemporary integrated water
gis-data (accessed Nov 3, 2016).
resources management pursuits, the latter of which is an integrated tool Davis, S., Ogden, J.C., 1994. Everglades: the ecosystem and its restoration. CRC Press,
to counter climate induced disasters (Khadim et al., 2013). In this re- Boca Raton, Florida.
search, attempt was made to develop ways to gain more understanding Deilmai, B.R., Ahmad, B.B., Zabihi, H., 2014. Comparison of two Classification methods
(MLC and SVM) to extract land use and land cover in Johor Malaysia. In: IOP
and develop solutions for these issues by, framing a remote sensing- Conference Series: Earth and Environmental Science, vol. 20. IOP Publishing, pp.
based mapping approach for soil salinity; constructing maps of soil 12052.
salinity every five years for ENP area; and reflect on the spatio-temporal Douglas, M.S., Fink, R., Grunwald, M., 2007. The Everglades: river of grass. Pineapple
Press Inc.
patterns of soil salinity in the area. Literature shows that remote sensing USGS Earth Explorer Website. https://earthexplorer.usgs.gov/, Accessed date: 1 March
has the potential to determine soil salinity, which is sometimes more 2017.
convenient compared to traditional measures (Allbed and Kumar, ENP Website Everglades National Park. U.S. National Park Service. https://www.nps.
gov/ever/index.htm, Accessed date: 1 September 2015.
2013). The results shown in this study showcase detailed vegetation Egler, F.E., 1948. The dispersal and establishment of Red Mangrove Rhizophora in
classifications for nearly every five-year interval from 1996 to 2015, Florida. Caribb. For. 9, 299–320.
which indicate spatial variability of different land cover features in the Erdas Inc. Erdas Field Guide; Atlanta, Georgia, 1999.
Flowers, T.J., Troke, P.F., Yeo, A.R., 1977. The mechanism of salt tolerance in halophytes.
ENP area. The vegetation classes were used to infer indicative ranges of Annu. Rev. Plant Physiol. 28, 89–121.
soil salinity values based on an innovative “vegetation-to-salinity con- Foody, G.M., 2002. Status of land cover classification accuracy assessment. Remote Sens.
version” approach proposed in this study. The soil salinity maps can be Environ. 80, 185–201. https://doi.org/10.1016/S0034-4257(01)00295-4.
Frazier, P.S., Page, K.J., 2000. Water body detection and delineation with Landsat TM
used as the basis of future research in the field of vegetation mapping as
data. Photogramm. Eng. Rem. Sens. 66, 1461–1468.
well as soil salinity investigations in the ENP area. Held, A., Ticehurst, C., Lymburner, L., Williams, N., 2003. High resolution mapping of
tropical mangrove ecosystems using hyperspectral and radar remote sensing. Int. J.
Conflicts of interest Rem. Sens. 24, 2739–2759. https://doi.org/10.1080/0143116031000066323.
VRO Victoria Resources Online http://vro.agriculture.vic.gov.au/dpi/vro/vrosite.nsf/
pages/water_spotting_soil_salting_class_ranges#s1 (accessed Jun 3, 2016).
The authors declare no conflict of interest. Jones, D., Madden, M., Snyder, J., Rutchey, K., 2002. Draft Report on Vegetation
Classification System for South Florida. National Parks.
Kasischke, E.S., Smith, K.B., Bourgeau-Chavez, L.L., Rmanowicz, E.A., Brunzell, S.,
Acknowledgment Richardson, C.J., 2003. Effects of seasonal hydrologic patterns in south Florida
wetlands on radar backscatter measured from ERS-2 SAR imagery. Remote Sens.
This work was supported by the Strategic Priority Research Program Environ. 88, 423–441. https://doi.org/10.1016/j.rse.2003.08.016.
Khadim, F.K., Kar, K.K., Halder, P.K., Rahman, M.A., Morshed, A.K.M.M., 2013.
of Chinese Academy of Sciences (Grant No. XDA20010302), the Integrated Water Resources Management (IWRM) Impacts in South West Coastal
National Natural Science Foundation of China (Grant No. 41571356). Zone of Bangladesh and Fact-Finding on Tidal River Management (TRM). J. Water
Research was also supported by the Walter and Lalita Janke Resour. Protect. 5, 953–961. https://doi.org/10.4236/jwarp.2013.510098.
King, R.B., 2002. Land cover mapping principles: a return to interpretation fundamentals.
Innovations in Sustainability Science Research Fund of 2016, through
Int. J. Rem. Sens. 23, 3525–3546. https://doi.org/10.1080/01431160110109606.
Florida Center of Environmental Studies, and the National High-tech R Kingsford, R.T., Thomas, R.F., Wong, P.S., Knowles, E., 1997. GIS Database for Wetlands
&D Program of China (863 Program Grant 2012AA121303). of the Murray Darling Basin; Sydney, Australia.
Kuenzler, E.J., 1974. Mangrove swamp systems. In: In: Odum, H.T., Copeland, B.J.,
McMahon, E.A. (Eds.), Coastal ecological systems, vol. 1. Conservation Foundation,
References Washington, D.C., pp. 346–371.
Laba, M., Smith, S.D., Degloria, S.D., 1997. Landsat-based land cover mapping in the
Abrol, I.P., 1988. In: Yadav, J.S.P., Massoud, F.I. (Eds.), FAO Salt-Affected Soils and their lower Yuna River watershed in the Dominican Republic. Int. J. Rem. Sens. 18,
Management. FAO, Rome. 3011–3025. https://doi.org/10.1080/014311697217170.
Allbed, A., Kumar, L., 2013. Soil Salinity Mapping and Monitoring in Arid and Semi-Arid Lamond, R.E., Whitney, D.A., 1992. Management of Saline and Sodic Soils.
Regions Using Remote Sensing Technology: A Review. Adv. Rem. Sens. 2, 373–385. Lee, K.H., Lunetta, R.S., 1995. In: Lyon, U.G., McCarthy, J. (Eds.), Wetland detection
https://doi.org/10.4236/ars.2013.24040. methods, Wetland and Environment Applications of GIS, Boca Raton, Florida.
Allbed, A., Kumar, L., Aldakheel, Y.Y., 2014a. Assessing soil salinity using soil salinity and Light, S.S., Dineen, J.W., 1994. In: Davis, S.M., Ogden, J.C. (Eds.), Water control in the
vegetation indices derived from IKONOS high-spatial resolution imageries: Everglades: A historical perspective, The Everglades—The Ecosystem and its
Applications in a date palm dominated region. Geoderma 230–231, 1–8. https://doi. Restoration. St. Lucie Press, Delray Beach, Florida.
org/10.1016/j.geoderma.2014.03.025. FCE LTER Florida Coastal Everglades: Long Term Ecological Research. http://fcelter.fiu.
Allbed, A., Kumar, L., Sinha, P., 2014b. Mapping and modelling spatial variation in soil edu/data/GIS/, Accessed date: 1 February 2016.
salinity in the Al Hassa Oasis based on remote sensing indicators and regression Lu, D., Weng, Q., 2007. A survey of image classification methods and techniques for
techniques. Rem. Sens. 6, 1137–1157. https://doi.org/10.3390/rs6021137. improving classification performance. Int. J. Rem. Sens. 26, 823–870.
Allbed, A., Kumar, L., Sinha, P., 2017. Soil salinity and vegetation cover change detection Lucas, R.M., Milne, A.K., Mitchell, A., Donnelly, B., Ellison, J., 2000. Use of stereo aerial
from multi-temporal remotely sensed imagery in Al Hassa Oasis in Saudi Arabia. photography for assessing changes in the extent and height of mangrove canopies in
Geocarto Int. 6049, 1–17. https://doi.org/10.1080/10106049.2017.1303090. tropical Australia. In: IGARSS Proceedings, vol. 5. pp. 1880–1882.
Appleton, B.L., Greene, V., Smith, A., French, S., Kane, B., Fox, L., Downing, A., Gilland, MacNae, W., 1969. A general account of the fauna and flora of mangrove swamps and
T., 2009. Trees and shrubs that tolerate saline soils and salt spray drift. forests in the Indo-West-Pacific region. Adv. Mar. Biol. 6, 73–270.
Aschbacher, J., Tiangco, P., Giri, C.P., Ofren, R.S., Paudyal, D.R., Ang, Y.K., 1995. Manavalan, P., Sathyanath, P., Rajegowda, G.L., 1993. Digital image analysis techniques
Comparison of different sensors and analysis techniques for tropical mangrove forest to estimate waterspread for capacity evaluations of reservoirs. Photogramm. Eng.
mapping. In: IGARSS Proceedings, vol. 3. pp. 2109–2111. Rem. Sens. 59, 1389–1395.
Blackman, J.G., Gardiner, S.J., Morgan, M.G., 1995. Framework for biogeographic in- McKee, K.L., 1996. Growth and physiological responses of neotropical mangrove seed-
ventory, assessment, planning. In: Finlayson, C.M. (Ed.), and management of wetland lings to root zone hypoxia. Tree Physiol. 16, 883–889.
systems: the Queensland approach, Workshop Proceedings, Wetland Research in the Metternicht, G.I., Zinck, J.A., 2003. Remote sensing of soil salinity: potentials and con-
WetlDry Tbopics of Australia, Supervising Scientist Report No. 101. Jabiru, NT, straints. Remote Sens. Environ. 85, 1–20. https://doi.org/10.1016/S0034-4257(02)
Australia. 00188-8.
Miao, S., 2004. Rhizome growth and nutrient resorption: mechanisms underlying the

49
F.K. Khadim et al. Physics and Chemistry of the Earth 110 (2019) 31–50

replacement of two clonal species in Florida Everglades. Aquat. Bot. 78, 55–66. Schaeffer-Novelli, Y., Cintrón-Molero, G., Soares, M.L.G., De-Rosa, T., 2000. Brazilian
Miller, R.W., Donahue, R.L., 1995. Soils in our environment. Prentice hall. mangroves. Aquat. Ecosys. Health Manag. 3.
Mitra, D., Karmaker, S., 2010. Mangrove Classification in Sundarban using High Shober, A.L., 2008. Soils and Fertilizers for Master Gardeners: Tackling Soil Salinity
Resolution Multi-Spectral Remote Sensing Data and GIS. Asian J. Environ. Disaster Problems in the Home Landscape.
Manag. 2, 197–207. Shober, A.L., 2009. Soils and Fertilizers for Master Gardeners: Tackling Soil Salinity
Miyamoto, S., Martinez, I., Padilla, M., Portillo, A., 2004. Landscape Plant Lists for Salt Problems in the Home Landscape1.
Tolerance Assessment. Simard, M., Zhang, K., Rivera-Monroy, V.H., Ross, M.S., Ruiz, P.L., Castañeda-Moya, E.,
Module, FLAASH. Atmospheric correction module: QUAC and FLAASH user's guide. pp. Twilley, R.R., Rodriguez, E., 2006. Mapping Height and Biomass of Mangrove Forests
44 Version, 4. in Everglades National Park with SRTM Elevation Data. Photogramm. Eng. Rem.
Mougin, E., Proisy, C., Marty, G., Fromard, F., Puig, H., Betoulle, J.L., Rudant, J.P., 1999. Sens. 3, 299–311.
Multifrequency and multipolarisation radar backscattering from mangrove forests. Sklar, F., McVoy, C., VanZee, R., Gawlik, D.E., Tarboton, K., Rudnick, D., Miao, S., 2002.
IEEE Trans. Geosci. Rem. Sens. 37, 94–102. In: Porter, J.W., Porter, K.G. (Eds.), The effects of altered hydrology on the ecology of
MRLC, M.-R.L.C.C. National Land Cover Database (NLCD). https://www.mrlc.gov/index. the Everglades, The Everglades, Florida Bay, and Coral Reefs of the Florida Keys,
php, Accessed date: 24 September 2015. Boca Raton, Florida.
Odum, W.E., McIvor, C.C., Smith III, T.J., 1982. The Ecology of the Mangroves of South Stehman, S.V., Wickhamb, J.D., Smithc, J.H., Yangd, L., 2003. Thematic accuracy of the
Florida. A Community Profile. 1992 National Land-Cover Data for the eastern United States: Statistical methodology
Otukei, J.R., Blaschke, T., 2010. Land cover change assessment using decision trees, and regional results. Remote Sens. Environ. 86, 500–516.
support vector machines and maximum likelihood classification algorithms. Int. J. Steward, K.K., Ornes, W.H., 1975. The autecology of sawgrass in the Florida Everglades.
Appl. Earth Obs. Geoinf. 11, S27–S31. Ecology 56, 162–171.
Pezeshki, S.R., DeLaune, R.D., Patrick Jr., W.H., 1990. Differential response of selected Teas, H.J., 1979. Silviculture with saline water. In: The biosaline concept. Springer US,
mangroves to soil flooding and salinity: gas exchange and biomass partitioning. Can. pp. 117–161.
J. For. Res. 20, 869–874. Vitousek, P.M., 1994. Beyond global warming: ecology and global change. Ecology 75,
Proisy, C., Mougin, E., Fromard, F., 1996. Investigating correlations between radar data 1861–1876.
and mangrove characteristics. In: IGARSS Proceedings, vol. 1. pp. 733–735. Wang, Y., Imhoff, M.L., 1993. Simulated and observed L-HH radar backscatter from
Proisy, C., Mougin, E., Fromard, F., Karam, M.A., 2000. Interpretation of polarimetric tropical mangrove forests. Int. J. Rem. Sens. 14, 2819–2828. https://doi.org/10.
radar signatures of Mangrove forests. Remote Sens. Environ. 71, 56–66. https://doi. 1080/01431169308904311.
org/10.1016/S0034-4257(99)00064-4. Wdowinski, S., Amelung, F., Miralles-Wilhelm, F., Dixon, T.H., Carande, R., 2004. Space-
Ramsey III, E.W., Jensen, J.R., 1996. Remote sensing of mangrove wetlands: Relating based measurements of sheet-flow characteristics in the Everglades wetland.
canopy spectra to site-specific data. Photogramm. Eng. Rem. Sens. 62, 939–948. Geophys. Res. Lett. 31. https://doi.org/10.1029/2004GL020383.
Rasolofoharinoro, M., Blasco, F., Bellan, M.F., Azipuru, M., Gauquelin, T., Denis, J., 1998. USGS What are the band designations for the Landsat satellites? Landsat Missions.
A remote sensing based methodology for mangrove studies in Madagascar. Int. J. https://landsat.usgs.gov/what-are-band-designations-landsat-satellites, Accessed
Rem. Sens. 19, 1873–1886. https://doi.org/10.1080/014311698215036. date: 3 January 2017.
Rhoades, J.D., Chanduvi, F., 1999. Soil salinity assessment: Methods and interpretation of Wu, J., Vincent, B., Yang, J., Bouarfa, S., Vidal, A., 2008. Remote Sensing Monitoring of
electrical conductivity measurements. Food & Agriculture Org. 57. Changes in Soil Salinity: A Case Study in Inner Mongolia, China. Sensors 8,
Richards, J., 1999. Remote Sensing Digital Image Analysis. Springer-Verlag, Berlin 240. 7035–7049.
Ross, M.S., Meeder, J.F., Sah, J.P., Ruiz, P.L., Telesnicki, G.J., 2000. The southeast saline Wyn Jones, R.G., 1981. In: Johnson, C.B. (Ed.), Physiological processes limiting plant
Everglades revisited: 50 years of coastal vegetation change. J. Veg. Sci. 11, 101–112. productivity. Elsevier, Butterworths, London.

50

You might also like