You are on page 1of 12

Environmental Management (2012) 50:478–489

DOI 10.1007/s00267-012-9889-3

Spatial Variation of Soil Salinity in the Mexicali Valley, Mexico:


Application of a Practical Method for Agricultural Monitoring
Gabriel Judkins • Soe Myint

Received: 24 September 2010 / Accepted: 22 May 2012 / Published online: 29 June 2012
Ó Springer Science+Business Media, LLC 2012

Abstract The degradation of irrigated lands through the applicability of this methodology to the study of broad
process of soil salinization, or the buildup of salts in the regional patterns of degradation most appropriate for use
soil, has hampered recent increases in agricultural pro- by regional resource managers.
ductivity and threatens the sustainability of large-scale
cultivation in critical agricultural regions of the world. Keywords Remote sensing  Soil salinization 
Rapid detection of soil salinity on a regional basis has been Land degradation  Agriculture  Produce  Onion  Mexico
identified as key for effective mitigation of such land
degradation. The ability to detect regional patterns of soil
salinity at an accuracy sufficient for regional-scale resource Introduction
management is demonstrated using Landsat 5 Thematic
Mapper (TM) imagery. A case study of the Mexicali Valley The contribution of irrigation to the expansion and increased
of Baja California, Mexico was selected due to the region’s productivity of the world’s agriculture has been significant as
agricultural significance and concern for future soil salinity over 40 % of the food and fiber crops are currently cultivated
increases. Surface soil salinity was mapped using geore- on irrigated lands occupying only 17 % of the world’s agri-
ferenced field measurements of electrical conductivity cultural land area (Rhoades 1990; Scheumann 1997). Meeting
(EC), collected concurrently with Landsat 5 TM imagery. future demands of a growing and increasingly affluent pop-
Correlations between EC measurements and common ulation for year-round access to fresh produce will require an
indices derived from the satellite imagery were used to even greater contribution from irrigation, particularly in arid
produce a model of soil salinity through regression analy- regions as their climates are favorable for winter produce
sis. Landsat band 7, TNDVI, PCA 1, Tasseled Cap 3 and cultivation (Redman 1999). However, irrigated agricultural
Tasseled Cap 5 were found to offer the most promising lands face a number of challenges including rising capital
correlations with surface soil salinity. Generally low levels costs, relatively low food prices, and the degradation of land
of soil salinity were detected, however, distinct areas of and water resources (Scheumann 1997). One of the most
elevated surface salinity were detected at levels potentially significant threats to the future productivity of irrigated agri-
impacting sensitive crops cultivated within the region. The culture is the degradation of irrigated soils through the process
difficulty detecting low levels of salinity and the mid-range of soil salinization. Already, one out of five hectares of irri-
spatial resolution of Landsat 5 TM imagery restrict the gated land is affected by soil salinization, stagnating the
productivity of the world’s agriculture since as much land is
lost to degradation as is being gained through the opening of
G. Judkins (&)
Department of Geoscience, University of Nevada Las Vegas, new irrigated lands (Rhoades 1990; Postel 1999). Soil sali-
4505 Maryland Parkway, Las Vegas, NV 89154-4010, USA nization is a land degradation problem specifically affecting
e-mail: gabriel.judkins@unlv.edu; gabejudkins@gmail.com the critical agricultural regions of the Lower Colorado River
area of the U.S. and Mexico, where two-thirds of the agri-
S. Myint
School of Geographical Sciences and Urban Planning, cultural land within this region is believed to be afflicted to
Arizona State University, Tempe, AZ, USA varying degrees (Postel 1999).

123
Environmental Management (2012) 50:478–489 479

Elevated levels of salts in the soil inhibit the uptake of focusing primarily on the high soil salinity levels that have
water and nutrients by plants, may result in the deteriora- damage the periphery of the agricultural area. Benefits
tion of the soil structure and can lead to increased toxicity from such inconsistent monitoring efforts are diminished
experienced by these plants (Keren 2000). Soil salinity by the fact that soil salinity is a dynamic threat, typically
generally stunts the growth of vegetation and results in experiencing significant spatial and temporal variations.
decreases in agricultural yields that are dependent on the Such variability may even occur seasonally depending on
level of salinity, the sensitivity of the plant, and other environmental conditions such as precipitation events that
influential environmental conditions (Gerrard 2000). mobilize salts and physical disturbances such as tilling that
Human-induced or secondary soil salinization is prevalent tend to disperse salt crusts through the physical mixing of
in irrigated agricultural lands as high rates of evapotrans- the soil. What is needed to assist in addressing this land
piration and low annual precipitation allows mineral sol- degradation threat to productive agricultural lands is the
utes such as dissociated cations Na?, Ca?, Mg?, and K?; capability to monitor, on a regional scale, low levels of
and anions Cl-, SO42-, CO32-, HCO3-, and NO3-, to salinity with sufficient consistency to facilitate the study of
collect and concentrate within the soil through the appli- temporal variations and identify trends. Satellite-based
cation of irrigation water (Hillel 2008). The solubility of remote sensing platforms and their multi-spectral data
such salts means that they are commonly dissolved within offers the best chance of tackling this difficult land deg-
irrigation water and the act of irrigating arid soils has the radation problem. The opportunity for deploying remotely
tendency to mobilize any salts that are already present sensed imagery for the direct detection and monitoring of
(Rhoades 1990). Additionally, excessive watering and surface soil salinity is particularly enhance in environ-
leaky irrigation systems lead to the movement of salts into mental settings where there is sparse vegetation and at
the water table, which can later result in the salinization of times following the high evaporation rates of the summer
surface soils in the case of a rising water table that but prior to the arrival of rains or other factors that might
approaches the surface (Rhoades 1990; Keren 2000). disturb salt crusts and other surface salt deposits (Howari
Salinity values can vary with soil depth, directly affecting 2003b; Nield and others 2007).
plants based on their root zone exposure and presence of Remotely sensed imagery has been utilized to study soil
groundwater, however the shallow roots of most seasonal salinization for several decades due to its ability to detect
crops means that their exposure to surface soil salinity can changes in surficial conditions as a result of differences in
be of greater concern. Furthermore, the timing of salinity the electromagnetic energy reflected or emitted by the
exposure can be even more critical of a factor to consider ground and vegetation (Huete 2004; Howari and Goodell
as the sensitivity of plants varies at different life stages 2009). Salt affected soils typically exhibit surface salt
during its growth (Rhoades 1990). Surface soil salinity, crusts or depositional efflorescence altering the reflectance
referring to that within the average root zone of cultivated of the exposed soil surface based on the specific absorption
plants, is frequently advanced as the typical soil reference features of the salts present and their arrangement or other
for monitoring soil salinity due to its obvious applicability characteristics (Howari and others 2002; Howari 2003a).
for agriculture (Keren 2000). However, the surface salinity The molecular vibration process is the basis for the dis-
of the soil may also refer to the formation of visible, typ- tinctive spectral reflectance and absorption features of
ically bright white, salt crusts or non-visible concentrations various salt crusts, as limited wavelengths of electromag-
of salts present within the first few cm of the soil. It is this netic energy are absorbed in the vibrational stretching and
usage of the term surface soil salinity, referring to the bending of certain chemical bonds (Howari and others
surficial deposition of salts with other soil constituents, that 2002; Howari 2003a). Howari and others (2002) examined
is relevant to a discussion of employing multi-spectral the spectral reflectance and absorption features for various
remotely sensed data for the monitoring of salt affected types of salt crust, demonstrating the variability in spectral
soils. responses of different salt crusts as well as identifying the
Despite the magnitude of land degradation from soil impact of other factors such as grain size that could affect
salinization and its implication for future agricultural pro- the resulting spectral reflectance curves. In addition, the
ductivity, insufficient attention has been paid to the likelihood of having mixed salt crusts within a single pixel
regional mapping of salinity affected agricultural soils. The of a remotely sensed image has driven the need to explore
expense and time required for conventional field mapping the use of spectral endmembers to unmix such pixel values,
based on soil samples, as well as a lack of concerted while acknowledging that other physical factors may also
attention to the problem, has inhibited monitoring of this affect the observed spectral signatures (Dehaan and Taylor
growing land degradation problem. For instance, the 2003; Howari and Goodell 2009). The complexity of
monitoring of soil salinity within the Mexicali Valley of dealing with the variability among salt types and their
Mexico has occurred irregularly over the last twenty years, occurrence amid the influences of other soil properties has

123
480 Environmental Management (2012) 50:478–489

Fig. 1 Study area, locating the


area of interest (92,072.11 ha)
within the Mexicali Valley of
Baja California and its
borderlands environment

led some to declare that there are no known narrow this free and widely available imagery can be used effec-
absorption bands distinctive to salinity in general (Huete tively to assist in monitoring regional patterns of surface
2004). This ignores the advances that have been made soil salinity impacting productive agricultural lands.
using hyperspectral sensors to hone in on the various
absorption features of specific salt crusts (Dehaan and
Taylor 2003; Ghrefat and others 2007; Howari and Goodell Methods
2009). However, practical limitations including the avail-
ability, complexity and costs associated with hyperspectral Study Area Background
imagery still limit its potential to serve as the basis for
widespread, iterative, regional monitoring of surface soil The Mexicali Valley is located in the Municipality of
salinity in affected critical agricultural regions of the Mexicali and the Mexican State of Baja California, with
world. the international border defining its northern boundary with
A practical method for detecting soil salinity for regional Southern California and the Colorado River delineating its
agricultural resources managers must provide sufficiently eastern border with Southern Arizona. The Colorado River
accurate measurements across a large agricultural area and serves not only to define the international border with
must be easily repeated without excessive costs or com- Arizona but also is critical to the agricultural success of the
plexity given the high temporal and spatial variability of region as it supplies the majority of its irrigation water.
soil salinization. This research explores one such method- Centered at 31° North latitude, the valley’s climate is
ology through a case study on salinity affected soils within defined by clear skies and plenty of sunlight, with little
the agricultural region of the Mexicali Valley of Baja precipitation given the atmospheric dominance of high
California, Mexico (Fig. 1). An innovated approach to pressure. Average monthly high temperatures in July are
quantifying soil salinity on agricultural lands at the regional 42 °C and the lowest average monthly high of 21 °C
scale is proposed through the use of Landsat 5 Thematic occurs in December (Weather 2011). The yearly average
Mapper (TM) imagery, concurrent field measurements of precipitation is 72.5 mm, with monthly average precipita-
soil salinity, and a regression analysis of several commonly tion values averaging 0 mm in June and 12.4 mm in
available spectral indices to model the extent of soil sali- January (Weather 2011). This examination of surface soil
nization in the Mexicali Valley. While this methodology is salinity utilizes a study area of approximately 92,000 ha,
limited by the spectral and spatial resolution of Landsat 5 centered on the prime agricultural lands in the center of the
TM imagery and the fundamental difficulty in detecting low Mexicali Valley. The specific dimensions of the study area
levels of surface soil salinity, this study demonstrates that were designated based on the known extent of produce

123
Environmental Management (2012) 50:478–489 481

cultivation and the characteristics of available satellite regional understanding of magnitude and pattern of soil
imagery, with the extreme southeast corner including a salinization affecting the valley due to the lack of consis-
portion of the Colorado River and the Mexican state of tent mapping of salt-affected soils. Mexico’s National
Sonora. Water Commission (CNA) has undertaken 2 years (1995
Soils in the valley are generally classified under the and 2004) of quantitative regional analysis of the extent
principle soil order of Aridisols, according to the soil and magnitude of soil salinization, however, these analyses
taxonomy developed by the United States Department of are dated and were conducted at a level of precision
Agriculture (USDA). The defining characteristics of such insufficient to accurately project the effects on specific
soils are their lack of sufficient moisture for mesophytic crops cultivated in the valley. Additionally, these results
plants and limited soil horizon development. Entisols are have not been widely disseminated among agriculturalists.
also present in the valley given its original formation as the A questionable collective perception of soil salinization has
flood plain of the Colorado River (Hillel 2008). While emerged from this lacuna of understanding, leading to
localized variations do exist across the valley, the soil various poorly substantiated assertions regarding the
texture is generally a sandy loam, with limited areas on the extent, severity and pattern of salinity-affected soils in the
western edge of the valley having a greater concentration Mexicali Valley (Judkins 2005, 2009).
of clay as a result of erosion associated with the Sierra
Cucapa and Sierra Mayor mountains. Remote Sensing Background
Agriculture within the Mexicali Valley has developed
over time to specialize in high value, fall/winter export- The complexity and cost of traditional field sampling
oriented produce cultivation. The two principal agricultural approaches for regional monitoring of soil salinity has
seasons include a summer (May to September) season driven the exploration of various remote sensing tech-
during which field grains, cotton and oil seed crops dom- niques and several studies have demonstrated their capa-
inate, and a fall/winter season (October to May) during bility for detecting and mapping variations in surface soil
which a range of very high-value produce crops are culti- salinity (Metternicht and Zinck 2003). Unfortunately, the
vated for export markets alongside some field grains. Far successful application of satellite imagery for the accurate
fewer hectares (averaging around 9,000 ha) of land are detection of salinity has proven difficult as the range of salt
cultivated in produce than in field grains during the fall/ types associated with soil salinity share no single identi-
winter season due to the limited supply of available irri- fying spectral signature and soil, mineral and land-cover
gation water during that time of year and the generally conditions can interfere with the spectral responses asso-
higher water demands of these crops. The month of Sep- ciated with surface salt deposits (Dwivedi and Rao 1992;
tember is a particularly favorable time to examine salinity- Howari and Goodell 2009). The complexity and difficulty
affected soils given its timing at the close of the summer in detecting surface soil salinity has lead some to focus on
agricultural season and the relatively little agricultural the detection of salinity affected soils at moderate to high
activity that occurs during this month. The fact that little EC levels in non-agricultural settings (Dwivedi and Rao
active cultivation is ongoing at this time of year is partic- 1992; Metternicht and Zinck 1997). Interest in assisting
ularly advantageous for remote sensing of surface soil individual cultivators has also motivated research on pre-
salinity as a majority of fields are bare and there is a cise sub-field level monitoring of surface soil salinity
minimal amount of physical disturbance to the surface of (Corwin and Leach 2005). Recently, the focus has centered
the soils as tilling has not yet commenced for the fall/ on the use of hyperspectral imagery and subpixel analysis
winter season. The end of the summer agricultural season is using image-derived spectral endmembers as promising
the optimal time to study the salinity content of surface improved accuracies (Dehaan and Taylor 2003). This
soils as they should be at their peak as a result of the imagery and processing technique also holds the potential
extensive irrigation undertaken during the heat of the for distinguishing among various salt types in the presence
summer agricultural season (Metternicht and Zinck 2003). of a few dominant salt crusts (Howari and others 2002). A
Interviews with produce operations in the Mexicali review of these approaches is provided by Metternicht and
Valley indicate that soil salinization is a prime concern Zinck (2003) and Howari and Goodell (2009). However,
influencing the land-use decisions of agriculturalists as the complexity of classification techniques and the limited
they select parcels to rent for the cultivation of valuable availability of hyperspectral imagery presents a challenge
produce crops. The concern for soil salinization has forced to their general use in applied agricultural research (Csillag
cultivators to move operations to lands perceived to be less and others 1993; Dehaan and Taylor 2003).
affected in order to minimize crop losses due declining Valuable information on surface soil characteristics for
harvest yields (Judkins 2005, 2009). However such deci- regional resource managers can yet be gleaned from tra-
sions on land-use have been undertaken with limited ditional medium-resolution multispectral imagery. While

123
482 Environmental Management (2012) 50:478–489

lacking the fine spectral resolution of hyperspectral imag- Data Acquisition and Processing
ery, and limited to regional analysis by its medium-spatial
scale, Landsat 5 TM imagery is widely available Landsat 5 TM imagery from the USGS, path 38 and row
throughout the world and is now provided free for down- 38, was purchased shortly after its acquisition date on 13
load by the United States Geological Survey (USGS). September 2007 to allow for accurate field measurements
Additionally, several studies have identified promising of soil salinity concurrent with the satellite imagery. This
spectral bands, ratios and classification techniques utilizing imagery having 28.5 m resolution with 7 channels—blue
Landsat imagery for the detection of surface soil salinity (0.45–0.52 lm), green (0.52–0.60 lm), red (0.63–0.69 lm),
(Dwivedi and Rao 1992; Metternicht 2003; Metternicht near infrared (0.75–0.90 lm), mid infrared (1.55–
and Zinck 2003; Nield and others 2007). These studies 1.75 lm), thermal infrared (10.4–2.50 lm) and mid infra-
indicate that there remains much that can still be accom- red (2.09–2.35 lm) was used to generate a basic land-use
plished utilizing Landsat 5 TM imagery. This is particu- and land-cover (LULC) map of the study area based on a
larly true when one considers the longstanding call for an modified Anderson level one and two classification system
approach to monitoring regional patterns of surface soil (Anderson and others 1976; Table 1). An unsupervised
salinity appropriate for resource managers seeking to classification method was employed using the Iterative Self
address the socio-environmental drivers of soil salinization Organizing Data Analysis (ISODATA) algorithm of
at the regional scale (FAO 1995; Judkins 2009). Such an ERDAS Imagine with 70 clusters, a convergence threshold
approach to monitoring salt-affected soils may therefore of 0.97 and defined maximum number of iterations set for
tolerate the sacrifice of some spatial and spectral precision 15. The classes generated by the unsupervised classifica-
for a regional perspective, capable of identifying the extent, tion were then recoded into the desired general LULC
magnitude and the general patterns of salt-affected soils classes need for the research based on expert experience
and how they change through time. with the study area developed over the previous 3 years of
field work and reference to available aerial imagery
Research Design (Table 2). The selected LULC classes include: urban/built-
up, irrigated agricultural land, non-irrigated agricultural
The spatial variation of surface soil salinity in the Mexicali land, sandy areas, transitional areas, mixed barren land, and
Valley was mapped using multispectral imagery and a unclassified lands (Fig. 2). Extensive fieldwork experience
simple, yet effective method based on the combination of within the region informed the selection of LULC classes,
several common spectral indices that exhibited correlations which were intended to be general enough to reflect the
with field measurements of surface soil salinity. The dynamic nature of the system of short-term agricultural
research project comprised three stages: data acquisition land leases that predominate within the study area. An
and preparation, model generation and selection, and final accuracy assessment employing aerial imagery and field
image processing. The acquisition of new Landsat 5 TM verification performed during the field soil sampling con-
imagery from the USGS was necessary as both agricultural firmed that the classification had been successful in iden-
land-cover and soil salinity can be highly variable over tifying agricultural land-uses with an acceptable range of
time. We measured the level of soil salinity using a tem-
perature standardized Field Scout Electrical Conductivity
Table 1 Land-use and land-cover categories modified slightly from
(EC) meter and the Saturated Media Extract method pre- Anderson and others (1976)
scribed by Spectrum Technologies (2006). The final pre-
LULC class Definition
cise location of sample sites was recorded using a handheld
Magellan Meridian Gold Global Positioning Service (GPS) Urban or built-up Areas dominated by human settlements and
device. EC, in mS/cm, was used as the standard measure of structures
soil salinity given its prevalence in soil salinization studies Irrigated agricultural Agricultural lands with green vegetation,
(Corwin and Leach 2005). We used the ERDAS Image land indicating active irrigation
software package to perform the image processing work Non-irrigated Agricultural lands without green vegetation,
agricultural land indicating no active application of irrigation
and ESRI’s ArcGIS software package to bring together the water
GPS waypoints with the satellite imagery. Correlations Sandy areas other Areas dominated by sand and lacking
between the pixel values generated by various spectral than beaches significant vegetation
indices and the averaged EC measurements of the field Transitional areas Areas in transition from agricultural lands to
sample sites provided the basis for a regression analysis other land categories, including multiple
that produced a model of soil salinity characterized by a season fallow land
relatively high degree of sensitivity and an acceptable level Mixed barren land Areas defined by the presence of some natural
vegetation
of accuracy for regional mapping of surface soil salinity.

123
Environmental Management (2012) 50:478–489 483

Table 2 Land-use and land-cover classification distribution within accuracy, however, classification confusion between sandy
study area and barren lands is evident (Table 3). The lower accuracies
Class name Histogram Hectares Percent of these two classes were viewed as acceptable given their
similar broad classification as non-agricultural lands.
Unclassified 3,462 281.20 0.31
Within 8 days of imagery acquisition, soil salinity field
Urban/built-up 13,895 1128.62 1.23 measurements were collected over a two day period from a
Irrigated agricultural land 242,068 19661.97 21.35 stratified random selection of sites that avoided urban or
Non-irrigated agricultural land 513,829 41735.76 45.33 built-up areas and other restricted lands. Additionally, any
Sandy areas 56,454 4585.48 4.98 actively cultivated fields were also avoided, given the
Transitional areas 197,623 16051.93 17.43 sanitary restrictions on accessing cultivated fields due to
Mixed Barren Land 106,213 8627.15 9.37 public health concerns. Avoiding actively cultivated lands
Total 1133,544 92072.11 100.00 does introduce a limitation to the study, however, it was not
deemed significant as only actively cultivated lands were
avoided and this represented a very small fraction of those
lands classified as irrigated agricultural lands due to the
timing of the study between the major agricultural seasons.
We have high confidence that significant changes in sur-
face soil salinity did not occur during the 7–8 days prior to
field work as the region did not experience any precipita-
tion or other atmospheric instability and little agricultural
work is typically undertaken at the close of the summer
agricultural season. Specifically, no major tilling opera-
tions, nor active irrigation of agricultural fields were wit-
nessed during the field work portion of this study.
Salinity measurements from 50 sites, were taken using
the Field Scout Soil & Water EC Meter and the Saturated
Media Extract (SME) method of standardization, following
the protocol dictated by Spectrum Technologies (2006).
Each sample site consisted of a column of surface soils
approximately 10 cm in diameter and 20 cm in depth that
was mixed and saturated with distilled water. Following ten
minutes of time, four temperature standardized measure-
ments of EC were taken, each at a depth of approximately
ten cm. The average of these readings was then recorded as
the official surface soil salinity of the field site. Locations
of the sample sites were mapped using a Magellan
Meridian Gold global positioning device with an accuracy
of around one meter, and ArcGIS was used to generate a
single point Geographic Information System (GIS) layer
with the associated EC attribute data.

Model Generation and Selection

Fourteen indices, including the seven Landsat 5 TM bands


and several types of image algebra and transformations
(hereafter referred to as indices), were selected for analysis
as part of the model generation process. The indices
employed were selected in consultation with the relevant
Fig. 2 An unsupervised classification of the area of interest derived literature based on their likelihood for exhibiting correla-
from the Landsat Thematic Mapper 5 imagery, scene Path 38 and tions with variations in surface soil salinity (Dwivedi and
Row 38, acquired 13 September 2007. Image width is approximately
24.9 km. The land-use and land-cover classes include agricultural,
Rao 1992; Csillag and others 1993; Metternicht and Zinck
natural, and urbanized areas within this largely rural agricultural 1997; Metternicht 2003; Metternicht and Zinck 2003;
region Huete 2004). These indices included some common

123
484 Environmental Management (2012) 50:478–489

Table 3 Accuracy assessment of the land-use and land-cover classification


Class name Reference Classified Number Producers Users Kappa
accuracy (%) accuracy (%)

Urban/built-up 9 4 3 33.33 75.00 0.7423


Irrigated agricultural land 63 64 52 82.54 81.25 0.7627
Non-irrigated agricultural land 138 136 116 84.06 85.29 0.7277
Sandy areas 18 15 9 50.00 60.00 0.5745
Transitional areas 44 52 28 63.64 53.85 0.4591
Mixed barren land 27 28 8 29.63 28.57 0.2151
Overall accuracy 72.24 %
Kappa coefficient 0.61

vegetation indices, principle components one through process confirmed the selection of the previously identified
three, tasseled cap index one through six, and the common preferred model, yielding the final model as follows:
iron oxide and clay mineral indices. Following their gen- y ¼ 2:326 þ ð0:039  Band 7Þ
eration, these indices were then combined with the seven
þ ð0:287  Band 9Þ  ð0:014  Band 11Þ
spectral bands of a Landsat 5 TM image into a single image
file comprised of 21 bands (Table 4). The correlation þ ð0:016  Band 16Þ þ ð0:036  Band 18Þ
between the pixel value of each band and the average of the
The continuous data of the selected model output, showing
salinity measurements falling within that pixel informed
increasing soil salinity with brightness, was converted to a
the selection of the nine bands with the strongest correla-
thematic map using the Subset function of ERDAS Image.
tion with soil salinity. Various permutations, comprising
This process divided the data into thematic categories by
sets of two, three, four, five and six bands (avoiding
rounding down all salinity measurements to their integer base.
combining highly correlated independent variables) were
Salinity readings were then categorized into half EC ranges,
undertaken using linear regression analysis to study which
roughly corresponding to the RMS accuracy of the selected
model would yield the strongest correlation to the sampled
model, producing a map of soil salinity for the Mexicali
surface soil salinity values. Models were generated for the
Valley appropriate for agriculturalists (Fig. 3). The predicted
ten combinations yielding the best R statistic and their
soil salinity data were also used to examine variations across
accuracies were determined using the root mean square
the study area that correspond to the LULC classification
(RMS) errors between their predicted salinity values and
(Table 6). The results of this last step yielded some interesting
the sample surface soil salinity (Table 5). The band com-
insights that serves to demonstrate the usefulness of this
bination 7, 9, 11, 16, 18 was ultimately selected based on it
regional approach to surface soil salinity detection for
having the highest R statistic and yielding the most realistic
resource managers.
map of predicted surface soil salinity values. The RMS
error of this model (0.631) implies that the average surface
soil salinity values predicted for each pixel should be
within approximately two thirds of one unit of EC. Results

Surface soil salinity values across the study area were


Final Image Processing found to be generally low with distinct areas of elevated
salinity appearing in clusters throughout the Mexicali
ERDAS Imagine’s Spatial Modeler was used to implement Valley (Fig. 3). Variations in predicted soil salinity across
the selected model and produce a map of predicted average the study area were limited mostly to the low end of the EC
soil salinity. The image generated by this model displayed spectrum, with no areas qualifying as having truly saline
predicted salinity levels in EC units for every pixel soils (those with EC readings above four). This finding was
according to the constants and band coefficients produced not unexpected given the yet productive nature of agri-
as part of model generation process. In addition to the culture in the valley and confirmed the previous soil
selected model, several other well performing models were salinity mapping work conducted by Mexico’s irrigation
also tested. While the variation in predicted values was water authority. Little to no salinity (0.0–0.4 EC) was
small, some generated small negative EC values and were attributed by the model to only 0.1 % of the total land
therefore disqualified from consideration. Importantly, this study area. Lands with some minor salinity (0.5–0.9 EC)

123
Environmental Management (2012) 50:478–489 485

Table 4 Indices and assigned


Band Index variables
band numbers
number

Band 1 Landsat band 1


Band 2 Landsat band 2
Band 3 Landsat band 3
Band 4 Landsat band 4
Band 5 Landsat band 5
Band 6 Landsat band 6
Band 7 Landsat band 7
Band 8 NDVI
Band 9 TNDVI
Band 10 Vegetation
index
Band 11 PCA 1
Band 12 PCA 2
Band 13 PCA 3
Band 14 Tasseled cap 1
Band 15 Tasseled cap 2
Band 16 Tasseled cap 3
Band 17 Tasseled cap 4
Band 18 Tasseled cap 5
Band 19 Tasseled cap 6
Band 20 Iron oxide
Band 21 Clay mineral

were mapped across 29.6 % of the study area. Most of the


land in the study area showed slightly elevated levels of
salinity (1.0–1.4 EC) as the model attributed this range to
68.3 % of the area, echoing concerns about soil salinization
as voiced by agriculturalists. Elevated levels of salinity
(1.5–1.9 EC) were modeled over 2.0 % of the study area, Fig. 3 Modeled average soil salinity values based on the area of
however the clustered pattern of these salt-affected soils interest derived from Landsat Thematic Mapper 5 imagery, scene
presents an increased likelihood for adverse impacts on Path 38 and Row 38, acquired 13 September 2007. Image width is
approximately 24.9 km. Each 28.5 m2 pixel is classified according to
individual farming operations while also suggesting that
its projected average surface soil salinity and showing the total area
these elevated levels of salinity are most likely the result of within each class
specific resource management practices. It is important to
note that the soil salinity values represent average predicted
values estimated for each pixel, likely concealing fine-scale higher soil salinity readings than those mapped by this
variations in the location and magnitude of salt-affected model, no areas were modeled to be large enough to
soils within each pixel. While there are certainly areas with influence average pixel values above those discussed.

Table 5 Band combinations with best statistical performance in modeling soil salinity
Bands employed R R2 Adjusted R2 SEE F RMS error Bias

7, 9, 11, 16, 18 0.402 0.162 0.057 0.675 1.543 0.631 -2.010


5, 9, 11, 21 0.400 0.160 0.078 0.668 1.950 0.630 -0.370
5, 9, 11 0.400 0.160 0.100 0.660 2.663 0.631 1.190
5, 9, 18, 21 0.400 0.160 0.078 0.668 1.948 0.631 1.300
5, 9, 11, 18 0.400 0.160 0.078 0.668 1.951 0.631 -1.470
5, 9, 16, 18, 21 0.400 0.160 0.055 0.676 1.521 0.632 -1.640
5, 9, 11, 18, 21 0.400 0.160 0.055 0.676 1.523 0.632 -1.940

123
486 Environmental Management (2012) 50:478–489

Table 6 Soil salinity (EC) distribution by land-use and land-cover classification for the winter 2007 fresh produce agricultural season
Land-use/land-cover 0.0–0.4 EC 0.5–0.9 EC 1.0–1.4 EC 1.5–1.9 EC
% Area (ha) % Area (ha) % Area (ha) % Area (ha)

Irrigated agricultural land 1.2 94.2 18500.3 5.8 1147.1


Non-irrigated agricultural land 29.8 12406.1 68.6 28626.0 1.6 673.0
sandy areas 2.2 98.3 97.9 4483.1
Transitional areas 17.7 2839.1 82.3 13204.7
Mixed barren land 84.8 7302.8 15.2 1310.3
Total 0.1 98.3 29.6 27032.3 68.3 61641.3 2.0 1820.1

The elevated levels of surface soil salinity mapped in Discussion


Fig. 3 are clear evidence that surface soil salinity occurs
throughout the study area at levels sufficient to cause Directly detecting salinity-affected soils utilizing satellite
immediate impacts to produce cultivation. These findings imagery is founded on the premise that surface salt crusts,
contradict the assertion that soil salinization is only a as well as other non-crustal salt deposits, would influence
concern for the future due to its relatively low EC level at the spectral reflectance of the soil. Several researchers have
the present time, as the clusters of elevated EC coincide undertaken studies of the spectral response of salt crusts
with irrigated lands considered prime lands for produce and other deposits and identified specific, narrow bands
cultivation (Table 6). Summarizing the occurrence of soil that yield distinctive spectral signatures that would allow
salinity by LULC reveals the propensity of agricultural for the discernment of salt-affected soils (Dwivedi and Roa
lands and those most recently cultivated to have elevated 1992; Csillag and others 1993; Howari and others 2002;
levels of surface soil salinity, likely as a result of the Howari 2003a; Howari and Goodell 2009). While not
application of irrigation water during the previous agri- directly applicable to Landsat 5 TM imagery, given the
cultural seasons (Table 6). Irrigated agricultural lands have broader spectral range of its image bands, these findings
the largest percentage of its area in the second highest underpin the justification for why certain Landsat bands
category of soil salinity (1.0–1.4 EC) at 94.2 %. Addi- and indices of bands have the potential for being respon-
tionally, these lands account for the majority of the area sive to variations in surface soil salinity.
experiencing elevated levels of soil salinity (1.5–1.9 EC), Our study identifies Landsat 5 TM band 7, TNDVI, PCA 1,
with almost double the frequency of non-irrigated lands. Tasseled Cap 3 and Tasseled Cap 5 as exhibiting the most
Non-agricultural lands, including sandy areas and mixed promising correlations with surface soil salinity. These
barren land, had the lowest level of soil salinity as pre- findings are in line with the work of other researchers that
dicted by the model employed in this research. The likewise found similar bands and indices as possessing
majority of both of these lands registered at the 0.5–0.9 EC promising spectral responsiveness to variations in the EC of
level, indicating a tendency toward lower surface soil surface soils. In studies published by both Dwivedi and Rao
salinization than those lands defined by their agricultural (1992) and Metternicht and Zinck (1997), Landsat TM band
land-use. The LULC category of transitional areas also 7 in combination with other bands was shown to be par-
showed increased levels of soil salinity similar to the ticularly helpful in identifying salt-affected soils. Addi-
agricultural lands, with 82.3 % of this LULC category tionally, Metternicht and Zinck (2003) cite several studies
falling in the range of 1.0–1.4 EC. As this LULC category that demonstrate high correlations between vegetation
represents agricultural lands not recently cultivated but in indices and soil EC, lending support to our inclusion of
the process of transition either from or to agricultural land, TNDVI within our model of soil salinization. In the same
it is to be expected that increased levels of soil salinity paper, the authors also note the ability of principal com-
would still be detectable in those lands previously ponents analysis to be used in the separation of saline and
employed as irrigated agricultural land. Lastly, the distri- nonsaline soils (Metternicht and Zinck 2003). Lastly, our
bution of soil salinity by land-use and its clustered pattern findings regarding the correlation of the tasseled cap index
is consistent with the anticipated characteristics of irriga- with soil salinity is also generally supported by the pub-
tion-induced soil salinization and does not seem to support lished literature on the usefulness of image transformations,
the notion that the low levels of surface soil salinity are the although we found no specific reference to its use in pre-
result of its natural occurrence in a low lying delta region. vious studies (Metternicht and Zinck 2003). Our approach

123
Environmental Management (2012) 50:478–489 487

takes these findings further by demonstrating that mixing A final determination of the degree to which soil salinity
these bands and indices yields meaningful results when is driving land degradation in the Mexicali Valley must
combined into a model of surface soil salinity. While many consider the mix of agricultural crops cultivated in addition
combinations of varying numbers of bands and indices were to variations in EC and correlations with prevalent LULC
possible, the model selected both maximizes the strength of classifications. The predominant produce crop cultivated
the correlation with surface soil salinity and minimized its immediately following the summer agricultural season is
RMS error to yield a map deemed to best predict average green onions with other crops including carrots, cilantro,
surface soil salinity values at a regional scale. radish, garlic and lettuce. Green onions occupy an area that
Application of this practical method of detecting soil is around three times greater than the next most prevalent
salinity through remote sensing confirms the widespread type of produce, averaging around 3,500 ha (Judkins
concern that this area of the lower Colorado River is 2005). While only a fraction of the total irrigated agricul-
experiencing some problems with soil salinization. It tural land during the fall/winter season, green onions
should be noted that much of the study area (29.7 %) was account for approximately 13 % of the total harvested
found to be either unaffected or having only minor levels of value of crops during the agricultural season (Judkins
detectable salinity (0.0–0.9 EC) in the soil. While this 2009). All types of produce when combined typically
result might be construed to dissuade concern for salinity- accounts for around 31 % of the total harvest value of all
driven land degradation in the Mexicali Valley, one must crops cultivated during the fall/winter season, demon-
recall that even low levels of salinity can result in a loss of strating the outsized importance of produce crops to the
crop productivity and thus impact the economic viability of viability of agriculture within the Mexicali Valley.
an agricultural region (Postel 1999; Hillel 2000). The Increasing levels of soil salinity, while low in absolute
general prevalence of slightly elevated levels of salinity terms, endangers current pattern of agricultural activity as
(1.0–1.4 EC) and the existence of wide-spread clusters of green onions and many of the other types of fresh winter
elevated levels of salinity (1.5–1.9 EC) confirms the produce cultivated in the area are known to be sensitive to
assertion of agriculturalists that soil salinization is a real salinity. With regard to green onions, it is known that all
problem for agricultural operations. While it is interesting types of onions are highly sensitive to soil salinity and
to note that no pixels registered above two EC, this result is therefore it is asserted to hold true for green onions as well
understandable given the current productive nature of as bulb onions (Shannon and Grieve 1999). In tests on
agriculture in the valley. Salinity EC readings above three several types of onions, Berstein and Aers (1953) found a
or four have the potential to seriously impact the vitality of consistent measurable decline in yield (above 15 %) if soil
a wide range of produce and would thus quickly endanger salinity levels reached even 1.4 EC. This sensitivity to
the agricultural viability of winter produce in the entire salinity also resulted in a 50 % decline in onion yield when
region. Additionally, the values mapped are averaged over salinity reached 4.1 EC. The fact that declines in yield
a pixel covering an area of 812.25 m2, almost certainly occur at even the low levels of soil salinization (those
concealing higher levels of soil salinity occurring over above 1.4 EC), predicted to occur in clusters throughout the
small areas such as a section of one agricultural field. study area, suggests that crop losses may already be
Clusters of elevated salinity (1.5–1.9 EC) occur impacting produce cultivation within the Mexicali Valley.
throughout the study area, raising the question of what It is, however, notoriously difficult to determine the exact
biophysical and/or land management practices might be impact of salinity on crops as other factors, not the least is
tied to their occurrence. Additional investigation of these the timing and harvest practices of each crops, can impact
clusters is needed in order to explain their occurrence rel- the total agricultural losses (Chhabra 1996). We contend
ative to land and water management and the types of that the magnitude and spatial extend of salt-affected soils
produce cultivated in these areas. Examining the occur- is demonstrated to exist at a level appropriate to consider
rence of soil salinity by LULC offers strong evidence of the the surface soils of the Mexicali Valley as degraded by soil
agricultural impact on soil salinity in the Mexicali Valley, salinization.
suggesting that active cultivation is a primary factor in the
salinization of the soils in the valley. This finding is con-
sistent with the process of soil salinization as cultivated Conclusions
agricultural lands in this arid environment requires repe-
ated irrigation for produce cultivation, during which time This research successfully demonstrates an innovative
the lands experience little to no tillage of the surface soils. method for mapping the spatial variations in soil salinity in
Salts thus accumulate in the top layers of the soil as a result the Mexicali Valley utilizing Landsat 5 TM satellite
of the very high rates of evapotranspiration during the imagery and concurrent field sampling of soil salinity. The
summer season. regression analysis of field salinity measures and imagery

123
488 Environmental Management (2012) 50:478–489

allows surface soil salinity to be modeled across a large Berstein L, Aers AD (1953) Salt tolerance of five varieties of onions.
agricultural region with an adequate level of sensitivity to Proceedings of the American Society of Horticultural Science
62:367–370
varying levels of salinity for agricultural applications. Chhabra R (1996) Soil salinity and water quality. AA Balkema,
Through this study, we identified Landsat band 7, TNDVI, Brookfield
PCA 1, Tasseled Cap 3 and Tasseled Cap 5 as possessing Corwin DL, Leach MS (2005) Apparent soil electrical conductivity
the strongest correlation to variations in surface soil measurements in agriculture. Computers and Electronics in
Agriculture 46:11–43
salinity. By combining these indices into a single model, Csillag F, Pásztor L, Biehl LL (1993) Spectral band selection for the
we were able to generate a map of predicted average sur- characterization of salinity status of soils. Remote Sensing of
face soil salinity values, possessing an error value of Environment 43(3):231–242
around two thirds of one unit of EC across a pixel covering Dehaan R, Taylor GR (2003) Image-derived spectral endmembers as
indicators of salinisation. International Journal of Remote
812.25 m2. While of limited value to individual agricult- Sensing 24(4):775–794
uralists seeking to address variations in salinity occurring Dwivedi RS, Rao BRM (1992) The selection of the best possible
across a single cultivated parcel, such results present an Landsat TM band combination for delineating salt-affected soils.
opportunity for regional resource managers to better International Journal of Remote Sensing 13(11):2051–2058
Food and Agriculture Organization (FAO) of the United Nations and
monitor variations in soil salinity across a large agricultural French Institute of Agricultural and Environmental Engineering
area over time. This finding is all the more important as we Research (1995) Use of remote sensing techniques in irrigation
are beginning to realize that many of the driving factors of and drainage. In: Proceedings of the Expert Consultation,
land degradation are best addressed at the regional scale Montpellier, France, 2–4 November 1993, FAO, Rome, Italy
Gerrard J (2000) Soil degradation. Fundamentals of soils, Chapt 8.
rather than through remediation actions focused on indi- Routledge, New York, pp 177–199
vidual situations (Judkins 2009). Ghrefat HA, Goodell PC, Hubbard BE, Langford RB, Aldouri RE
While this approach and the model it generates are not (2007) Modeling grain size variations for aeolian gypsum
without limitations, including spectral and spatial resolu- deposits at white sands, New Mexico, using AVIRIS imagery.
Geomorphology 88:57–68
tion, its simplicity and degree of acceptable accuracy makes Hillel D (2000) Salinity management for sustainable irrigation:
it a promising tool for continuous region-wide salinity integrating science, environment, and economics. The Interna-
monitoring. The reliance on field sampling of soil saliniza- tional Bank for Reconstruction, Washington
tion may be viewed as a significant weakness by those Hillel D (2008) Soil chemical attributes and processes. Soil in the
environment: crucible of terrestrial life, Chapt 10. Academic
seeking a remote sensing solution that does not require Press, San Diego, pp 135–150
extensive field sampling, however, to date that approach has Howari FM (2003a) Comparison of spectral matching algorithms for
struggled to yield the level of detail necessary for land identifying natural salt crusts. Journal of Applied Spectroscopy
managers, without resorting to the complexity and costs 70(5):782–787
Howari FM (2003b) The use of remote sensing data to extract
associated with hyperspectral remote sensing approaches. It information from agricultural land with emphasis on soil salinity.
is anticipated that the application of this innovative method Australian Journal of Soil Research 41:1243–1253
for quantifying soil salinization at the regional scale has the Howari FM, Goodell PC (2009) Characterization of salt-crust build-
potential to assist resource managers to better understand up and soil salinization in the United Arab Emirates by means of
field and remote sensing techniques, Chapt 8. In: Metternicht GI,
this widespread form of land degradation and ultimately Zinck JA (eds) Remote sensing of soil salinization: impact on
facilitate the development of improved policies and land management. CRC Press, Boca Raton, pp 141–154
practices for the mitigation of salinity-driven land Howari FM, Goodell PC, Miyamoto S (2002) Spectral properties of
degradation. salt crusts formed on saline soils. Journal of Environmental
Quality 31:1453–1461
Huete A (2004) Remote sensing of soils and soil processes, Chapt 1.
Acknowledgments This project’s field work was funded by the In: Ustin SL (ed) Remote sensing for natural resource manage-
Melvin G. Marcus Scholarship Award administered by the School of ment and environmental monitoring. Wiley, New York, pp 3–52
Geographical Science and Urban Planning at Arizona State University. Judkins GL (2005) Mexican produce agriculture: an investigation of
Thanks to the anonymous reviewers for their helpful comments to NAFTA’s influence. Masters Thesis, Arizona State University
improve this manuscript. Judkins GL (2009) Mexican produce agriculture and land degrada-
tion: a study of soil salinization in the Mexicali Valley, Mexico.
PhD Dissertation, Arizona State University
Keren R (2000) Salinity, Chapt 1. In: Sumner ME (ed) Handbook of
References soil science. CRC Press, New York, pp G3–G25
Metternicht GI (2003) Categorical fuzziness: a comparison between
Anderson JR, Hardy EE, Roach JT, Witmer RE (1976) A land use and crisp and fuzzy class boundary modeling for mapping salt-
land cover classification system for use with remote sensor data. affected soils using landsat tm data and a classification based on
Geological survey paper 964: a revision of the land use anion ratios. Ecological Modeling 168:371–389
classification system as presented in the U.S. geological survey Metternicht GI, Zinck JA (1997) Spatial discrimination of salt-and-
circular 671. United States Government Printing Office, sodium-affected soil surfaces. International Journal of Remote
Washington, DC Sensing 18(12):2571–2586

123
Environmental Management (2012) 50:478–489 489

Metternicht GI, Zinck JA (2003) Remote sensing of soil salinity: Scheumann W (1997) Managing salinization: irrigation analysis of
potentials and constraints. Remote Sensing of Environment public irrigation systems. Springer, New York
85:1–20 Shannon MC, Grieve CM (1999) Tolerance of vegetable crops to
Nield SJ, Boettinger JL, Ramsey RD (2007) Digitally mapping gypsic salinity. Scientia Horticulturae 78:5–38
and natric soil areas using Landsat ETM data. Soil Science Technologies Spectrum (2006) Field scout soil & water EC meter:
Society of America Journal 71(1):245–252 catalog #2265FS. Spectrum Technologies, Plainfield
Postel S (1999) Pillar of sand: can the irrigation miracle last?. WW Weather.com (2011) Monthly Weather for Mexicali, Mexico. http://
Norton & Company, New York www.weather.com/outlook/travel/businesstraveler/wxclimatology/
Redman CL (1999) Human impacts on ancient environments. monthly/graph/MXBC0004. Accessed 24 Sept 2011
University of Arizona Press, Tucson
Rhoades JD (1990) Soil salinity—causes and controls, Chapt 4.
In: Goudie AS (ed) Techniques for desert reclamation. Wiley,
New York, pp 109–134

123

You might also like