Computers and Electronics in Agriculture 99 (2013) 101–107

Contents lists available at ScienceDirect

Computers and Electronics in Agriculture

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

A near-infrared reflectance sensor for soil surface moisture

measurement
Zhe Yin a, Tingwu Lei a,b,⇑, Qinghong Yan a, Zhanpeng Chen a, Yuequn Dong a
a
College of Water Resources & Civil Engineering, China Agricultural University, Beijing 100083, PR China
b
State Key Laboratory of Soil Erosion and Dryland Farming on the Loess Plateau, Institute of Soil and Water Conservation, Chinese Academy of Science and Ministry of
Water Resources, Yangling, Shaanxi Province 712100, PR China

a r t i c l e i n f o a b s t r a c t

Article history: Soil moisture is an important soil property that has important functions in various studies and applica-
Received 29 November 2012 tions, such as agricultural practices, hydrological processes and ecological issues. A near-infrared (NIR)
Received in revised form 28 August 2013 reflectance sensor designed for moisture measurement at the soil surface is based on the reflectance of
Accepted 31 August 2013
two light-emitting diodes (LEDs) of different wavelengths, in which one has a wavelength of 1940 nm
and a strong water absorption band, whereas the other has a wavelength of 1800 nm and a weak water
absorption band, as related to soil moisture reflectance. The algorithm is designed for estimating soil
Keywords:
moisture using the relative absorption depth from reflectance data of the 1800 and 1940 nm wave-
Soil moisture
Near-infrared reflectance
lengths. Laboratory experiments investigate the relationship between soil moisture and surface reflec-
Light-emitting diode tance by studying four different soils. The results indicate a strong linear correlation between soil
Relative absorption depth moisture and relative absorption depths for the different soils tested, and the reflectance models are
dependent on soil type. The soil moistures predicted using the proposed method agree well with mea-
surements obtained by gravimetric method, indicating the feasibility of the proposed method for soil
moisture measurement. The NIR LED reflectance sensor developed in this study is potentially useful
for soil surface moisture measurement in the laboratory and field.
Ó 2013 Elsevier B.V. All rights reserved.

1. Introduction Gravimetric method is currently the gold standard, most com-

Soil moisture is an important component of plant growth that is
closely related to irrigation and exchange of mass and energy be-
tween the soil and the atmosphere. Soil moisture is also the pri-
mary vehicle for chemicals, such as fertilizers and pesticides, to
be transported to the soil surface, inside the soil body, and through
plant organs (Loan and James, 2008). Therefore, soil moisture mea-
surement is a critical aspect in many scientific studies. Soil mois-
ture is also closely related to hydrological processes, such as
rainfall and runoff, soil erosion and ecological issues (Fitzjohn
et al., 1998; Wei et al., 2007)
Only a few direct and indirect methods are available for deter-
mining soil moisture content, including gravimetric methods
(Gardner, 1986), frequency and time domain reflectometry (FDR
and TDR) (Clarke Topp and Reynolds, 1998; Gaskin and Miller,
1996), neutron probes (Chanasyk and Naeth, 1996), capacitance
probes (Eller and Denoth, 1996) and microwave techniques
(Jackson, 1993).
⇑ Corresponding author at: College of Water Resources & Civil Engineering, China
Agricultural University, Beijing 100083, PR China. Tel./fax: +86 10 62736367.
E-mail address: leitingwu@cau.edu.cn (T. Lei).
monly accepted, and most reliable method for soil moisture deter-
mination and calibration of all indirect measurement methods.
Despite its advantages of accuracy and high reliability, the gravi-
metric method is time and resource consuming, destructive, and
unrepeatable.
TDR and FDR can be used to measure volumetric soil moisture
after proper calibration, with automatic logging capability. Soil
salinity has unpredictable effects on the measured moisture. The
spatial resolution depends on soil moisture and probe length. The
neutron probe has a wide range of measurement capability with
reasonable accuracy. However, the instrument requires a trained
operator due to the use of a radioactive source, which is potentially
hazardous to health and the environment (Tarantino et al., 2008).
In addition, the neutral probe cannot make reliable measurements
at the soil surface. The capacitance method has long been studied
but has become commercially available only in recent years. Its
resolution also depends on the size of the sensor. The gamma ray
attenuation method is capable of determining the moisture con-
tent at soil surface layers (up to 1–2 cm), but high cost and diffi-
culty of use limit the applicability of this technique (Dobriyal
et al., 2012). The radioactive source also poses a big risk to human
health and the environment. Remote sensing of soil moisture is
dependent on the electromagnetic energy that is either reflected

0168-1699/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.compag.2013.08.029
102 Z. Yin et al. / Computers and Electronics in Agriculture 99 (2013) 101–107
or emitted from the soil surface, and is most suitable for determin-
ing the average soil water situations over large areas. However, the
uncertainty in determining the relationships among brightness,
temperature, and soil moisture limits its accuracy (Wang and Qu,
2009).
Most of the above mentioned techniques essentially measure
the average water content over a certain depth of soil layer, and
are therefore not suitable for measuring moisture at the soil
surface. For example, the neutron probe has been characterized
as inaccurate at the top 30 cm soil layer because of the loss of neu-
trons to the atmosphere at this shallow depth (Evett et al., 2003).
Soil moisture near the soil surface is one of the essential variables
for understanding water transfer at the soil–atmosphere interface
(Brutsaert, 1983). Inoue et al. (2001) made a good attempt to
develop a TDR probe for ‘‘non-penetration’’ and ‘‘on-the-move’’
measurement of moisture condition near the soil surface. Their
K-type and S-type probes provided reasonable estimates of soil
moisture condition near the surface at depths of 0–5 cm of soil
layer. This new TDR probe awaits further operational examination
under different specific conditions.
The near-infrared (NIR) reflectance technique can measure the
reflectance from the surface of objects and has been widely used
for detecting soil attributes, such as soil organic matter, minerals,
water content, pH and heavy metal content. Different soil constit-
uents have unique absorption features due to the vibrations in
molecular bonds (Wang et al., 2012). Most studies have shown that
reflectance in certain spectral bands can be correlated with soil
properties (Ben-Dor and Banin, 1995; Bowers and Hanks, 1965;
Daniel et al., 2004; Viscarra Rossel and McBratney, 1998; Zhu
et al., 2010). Soil moisture has absorption bands of 970, 1200,
1450, and 1940 nm in the NIR spectrum. Two strong absorption
bands have been reported at 1450 and 1940 nm due to the first
overtone of the OH-bending band and the combination of the
OH-stretching band and the OH-bending band (Büning-Pfaue,
2003; Baumgardner et al., 1985; Liu et al., 2003). A higher moisture
content is correlated with deeper water absorption band, and vice
versa.
Meaningful attempts have been made to investigate the rela-
tionship between soil reflectance to NIR and soil moisture by full
spectrum and light filters that capture narrow bands.
Lobell and Asner (2002) acquired reflectance measurements of
soil moisture in a laboratory setting for four different sieved soil
samples at various moisture contents using a full-range (350–
2500 nm) spectrometer (Analytical Spectral Devices). Soil reflec-
tance was found to be nonlinearly correlated with soil moisture,
which was well defined by an exponential model between the
1100 and 2500 nm wavelengths, suggesting that longer
wavelengths are more suitable for measuring volumetric moisture
contents above 20%. Measurements of reflected radiation (350–
2500 nm) were obtained in the field for four Israeli soils, and they
showed moderately successful results, with the R2 value at the
calibration stage between 0.9 and 0.96 and the RMSE ranging from
4.9% to 5.4% (Ben-Dor et al., 2008). Zhu et al. (2010) predicted the
moisture content for three types of soils using the wavebands
1400, 1940, and 2250 nm, including one artificially constructed
sample, one natural soil core, and one natural surface soil sample.
A total of 21 water content levels varying from 0.001 cm3 cm3 to
0.373 cm3 cm3 were used. The results indicate that the most sig-
nificant correlation between soil moisture and reflectance was
identified at 1400 nm for disturbed soil samples (R2 = 0.996,
RMSE = 0.010 cm3 cm3) and at 1940 nm for both cored
(R2 = 0.969, RMSE = 0.019 cm3 cm3) and surface soil samples
(R2 = 0.895, RMSE = 0.059 cm3 cm3).
All of the above studies were conducted using commercial spec-
trometer or spectral devices to investigate the responses of NIR
reflectance to soil moisture.
Various attempts have been made to develop devices that can
measure soil moisture in the laboratory with NIR reflectance. Skid-
more et al. (1975) developed a reflectometer to measure the reflec-
tance of near-infrared radiation from a soil surface at four bands.
The integrating sphere and filter resulted in cumbersome proce-
dure and slow response. Narrow-band pass filters had to be man-
ually positioned before the reflected light could reach the
integrating sphere. Kano et al. (1985) designed a near-infrared
reflectance moisture meter to obtain soil moisture readings using
NIR reflectance at 1800 and 1940 nm. The entire meter was housed
in an aluminum tube with diameter of 38.1 mm and length of
381 mm, with a data acquisition circuitry outside the tube. The
large size limits its applicability as a sensor. Given the lack of prop-
er LED light sources, two narrow-band interference filters of 10 nm
band pass had to be used as interference filters to isolate the mea-
suring band at 1940 nm and the reference band at 1800 nm. The
output voltage of the meter showed good linear correlation with
gravimetric soil moisture content, from 2.5% to 35%, with a stan-
dard error of ±1.9% moisture units for clay and loam soils. The out-
put voltage failed to respond when the moisture content was
higher than 35%. The results also indicated that the meter may re-
quire a different calibration for sand and sandy loam. Belisle et al.
(1996) discussed an optical reflectance technique with a flashlight
and a photodiode, which has small volume and low energy con-
sumption. However, an interference filter had to be used in front
of the detector, which made the structure more complex.
The present study has the following objectives: (1) to design a
near-infrared sensor for measuring soil surface moisture; (2) to de-
velop a mathematical model for soil moisture determination based
on soil surface reflectance data; and (3) to compare the moisture
contents measured by the NIR approach with measurements ob-
tained using the oven-dry method.
2. Sensor description
Compared with the tungsten lamp and laser LED, infrared LED
has certain advantages, such as low energy consumption, instant
response and long lifetime. Therefore, infrared LED was selected
as the light source in this study.
The NIR reflectance sensor constructed for soil surface moisture
measurement consisted of LED light sources for generating NIR
light at 1940 and 1800 nm, which were the same wavelengths used
by Kano et al. (1985), photodiodes for receiving the reflectance,
sensor case, amplifier circuit, and analog-to-digital (A/D) converter
(Fig. 1). The sensor case is made of black polyvinylchloride (PVC),
with dimensions of 2 cm in diameter at the upper end, 5.5 cm in
diameter at the lower end, and 2.5 cm in height.
The small volume of light source is beneficial when incorporated
into a shield case to ensure that more light reaches the photodiode
while minimizing ambient light impact when the shield case is
placed at the soil surface. The light detector is installed at the top
of the sensor case, and six LEDs are arranged circularly around the
light detector. All LEDs are angled at 45° to the normal so that a beam
of light from a LED is focused on a plane, resulting in a central irradi-
ated area of approximately 8 mm in diameter at the soil surface.
During operation, the radiations from the LEDs are directed to
the soil surface at a fixed distance of 15 mm. After interacting with
the soil surface, a fraction of the reflected light is reflected back to
the receiving detector, and the rest is lost due to scattering, absorp-
tion, and transmission. The detector has a wide angular response to
incident radiation, and no light reaches the detector directly from
the LEDs. To generate maximum average power, the infrared LED
emits a pulse signal to produce 600 mA current for 700 ls
(Fig. 2). Normally, all the LEDs at both wavelengths of 1940 and
1800 nm are under off status. When measurement is started, the
 Z. Yin et al. / Computers and Electronics in Agriculture 99 (2013) 101–107 103

Fig. 1. NIR sensor case schematic.

Fig. 2. Pulse pattern of the LED.

reference LED is turned on for a short period of 700 ls, and the
Table 1
reflectance of reference light is detected and collected by the Properties of the four soils used in this study.
control unit. Afterward, the measuring LED is turned on and
Soil no. Soil 1 Soil 2 Soil 3 Soil 4
the reference LED is turned off after 700 ls. The procedure is re-
peated once every second. Upon receiving the light within the Clay (%) 15 29 39 19
spectral response range, the receiving photodiode produces a Sand (%) 45 23 25 61
Silt (%) 40 48 36 20
voltage signal that is proportional to the amount of incoming light Dry bulk density (g/cm3) 1.33 1.31 1.26 1.41
energy. The output voltage is sent directly to the low-noise Moisture at 1.5 MPa (%) 13.6 11.1 6.0 9.5
amplifier before it is input into the data storage unit. The amplifier, CaCO3 (%) 0.64 7.68 9.64 0.92
serving as a high pass filter, can attenuate signals with lower Organic matter (%) 1.72 0.73 0.42 1.52
CEC (m mol/kg) 220.36 111.77 59.91 211.76
frequencies.
The InGaAsSb photodiode has a spectral response range of 900–
2400 nm and a 0.3 mm  0.3 mm square sensing area, with peak
sensitivity at 2100 nm. The InGaAsSb photodiode is very sensitive During the experiments, soil materials were air dried and
at wavelengths of 1800 and 1940 nm, which were used in the pres- passed through a 1 mm sieve before use. The Pressure Plate
ent study. Extractor was employed to adjust soil moisture to different levels
under different designed potential levels. The designed potential
3. Materials and methods levels used for this study are 0, 0.005, 0.01, 0.03, 0.05, 0.08, 0.1,
0.3, 0.5, 1.0, 1.3, and 1.5 MPa.
3.1. Soil samples The following test procedure was employed: (1) the saturated
samples were placed on the porous disk inside the chamber after
Four soil samples were collected from different regions in Chi- measuring the original weight; (2) the chamber was closed and pres-
na. The soils have different textures, varying organic contents, surization was started under the designed level of water potential;
and different colors. The soil textures were determined and data (3) the soil sample was drained for a sufficiently long period until
are listed in Table 1. Soil 1 was collected from Jilin, Jilin province equilibrium was reached; (4) the samples were weighed to deter-
in northeastern China; Soil 2 and Soil 3 were collected from Yan- mine the water content; (5) the reflectance of soil surface was mea-
gling and Yan’an, Shaanxi province in northwestern China, respec- sured; (6) the samples were subjected to a higher pressure, and steps
tively; Soil 4 was collected from Chongqing in southwestern China. (2–5) were repeated until a pressure of 1.5 MPa was obtained. Final-
The colors of the Soils 1, 2, 3, and 4 after oven drying were black, ly, the samples were oven-dried before soil bulk density and water
yellow, pale-yellow, and purple, respectively. contents at different water potential levels were determined. Three
104 Z. Yin et al. / Computers and Electronics in Agriculture 99 (2013) 101–107

4. Result and discussion

4.1. Influence of particle size on reflectance

Surface roughness is a major factor affecting soil reflectance. As

particle size increases, the surfaces became rougher, and vice versa.
Thus, the roughness of the surface is a function of the soil aggre-
gate size (Bowers and Hanks, 1965). The particle size is determined
as the midpoint of two successive sieves. As the particle size is in-
creased, the reflectance decreased at both bands (Fig. 4). The re-
sults agree with results previously obtained by Wu et al. (2009)
and Skidmore et al. (1975). The difference in reflectance values of
different sizes reaches 35% between the smallest and the largest
particles. The most evident decrease occurs at particle size less
than 1 mm. However, the reflectance value remains almost con-
stant when the particle size is greater than approximately 1 mm,
which indicates that particles have little influence on additional
Fig. 3. Normalized spectrum with or without the continuum removed. absorption of radiant energy when particle size exceeds a certain
value. The reflectance value of the 1940 nm wavelength is very
close to that of the 1800 nm wavelength. Larger particles reflect
to five soil samples were placed into the chamber at a time, as less energy due to bigger void spaces between soil particles, which
repetitions. The gravimetric moisture contents of soil samples were causes additional scattering and absorption of light. Based on cal-
determined using the oven-dry method and calculated using the culations, when the particle size increases from 0.18 mm to
following formula: 5 mm, at least 43.2% reduction in the reflectance occurs.
m  m0
w¼  100% ð1Þ
m0 4.2. Model for soil surface moisture estimation

where w is mass-based moisture content (%); m is the weight of wet To formulate a calibration equation, a simple model was used to
soil (kg); and m0 is the weight of dry soil (kg).
correlate soil surface moisture with the relative absorption depth
Gravimetric moisture contents were then transformed into vol- for the four soils. For each soil type, the dataset is composed of dif-
umetric moisture content using the following formula:
ferent soil moisture levels. The dataset is divided into two groups.
qb One group of data is used for calibration and the other is used for
h¼w  100% ð2Þ validation (Table 2). The calibration data are selected randomly out
qw
of the dataset under different soil moisture levels to calibrate the
prediction model. The validation data are then used to verify the
where h is volumetric soil moisture content (%); qb is the soil bulk
calibrated model. The measured volumetric moisture contents cov-
density (kg/m3); and qw  1000 is the water density (kg/m3).
er a wide range from completely dry soil to saturation. The soil
The reflectance of a soil depends on soil particle sizes (Wu et al.,
moisture is correlated with the relative absorption depth using:
2009). Therefore, Soil 2 was used as an example to verify the effects
of particle sizes on reflectance. The soil sample was air-dried and
h ¼ aR þ b ð4Þ
passed through six sieves in a range of mesh sizes 5, 4, 2, 0.5,
0.3, and 0.18 mm.
where h is the volumetric soil moisture content (%); and a and b are
the regression coefficients.
3.2. Relative absorption depth Two statistical indices, the coefficient of determination (R2) and
root mean squared error (RMSE), are used to estimate the accuracy
The most commonly used procedure for determining absorp- of the measured data. A lower RMSE value indicates a smaller
tion depth is continuum removal (Van der Meer, 2004). Continuum
removal calculates normalized spectra by dividing the reflectance
within an absorption region by an interpolated reflectance
straight-line datum between the two local maxima that bound
the region (Whiting et al., 2004). As shown in Fig. 3, after contin-
uum removal is applied, the absorption band depth at each wave-
Reflectance (%)

length in the absorption is calculated by subtracting the

continuum removed reflectance from 1 (Kokaly and Clark, 1999).
To calculate the relative absorption depth at around 1940 nm,
the reflectance of 1800 nm is selected as reflectance datum due
to its closeness to 1940 nm. The relative absorption depth is calcu-
lated using the following equation, which is equivalent to Kano’s
algorithm (Kano et al., 1985):

Rb
R¼1 ð3Þ
Rc
Particle Size ( mm )
where R is the relative absorption depth, Rb is the reflection inten-
sity at 1940 nm, and Rc is the reflection intensity at 1800 nm. Fig. 4. Reflectance of dry soil with different particle sizes.
 Z. Yin et al. / Computers and Electronics in Agriculture 99 (2013) 101–107 105

Table 2
Calibration set and validation data sets.

Sample set Sample no. Number of samples Moisture content (%) Mean (%) Standard deviation (%)
Calibration Soil 1 45 0–50.1 33.29 11.15
Soil 2 53 0–50.0 31.53 10.49
Soil 3 41 0–52.0 32.72 13.39
Soil 4 43 0–47.1 30.48 11.21
Validation Soil 1 29 16.6–48.5 33.34 8.51
Soil 2 28 16.7–49.0 31.60 8.57
Soil 3 28 8.4–51.3 32.04 11.37
Soil 4 28 15.1–45.0 31.88 7.59

difference between the measured and predicted values. The R2 and Fig. 6 shows a strong linear correlation between volumetric
RMSE values are calculated using the following formulas: moisture content and the relative absorption depth for all soils
, combined, with R2 = 0.642. However, the RMSEs of linear models
X
n
2 Xn
2
2
R ¼1 ð^hi  hi Þ ð^hi  hÞ ð5Þ for individual soils are lower than the RMSEs for all soils combined
i¼1 i¼1 (Table 3). The results show that the linear prediction model for all
soils combined is not as reliable as the linear prediction models ob-
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi tained for individual soils. These results suggest that individual cal-
1 Xn 2
RMSE ¼ i¼1 i
ðh  h^i Þ ð6Þ ibration models must be constructed for different soils.
n
Soil reflectance can be influenced by a number of factors, such as
where hi and ^
hi are the measured and predicted volumetric soil soil texture, surface roughness, organic matter content, and
moisture contents, respectively (%); 
h is the average of the mea- moisture content (Levitt et al., 1990; Liu et al., 2003; Zhu et al.,
sured soil moisture values (%); and n is the number of 2010). Therefore, each soil should utilize its own set of calibration
measurements. parameters. The best models are defined as those which produce
the lowest RMSE and highest R2. Among the four soil types, Soil 4
4.3. Relationship between soil surface moisture and relative absorption has the highest R2 value (0.863) and Soil 1 has the lowest calibration
depth RMSE value (2.88%). All the four soils, using the same form of
calibration model, have reasonably good regression results even if
Fig. 5 shows a linear correlation between volumetric soil mois- different regression parameters were produced. The slopes and
ture content and the relative absorption depth for the four soils. intercepts varied depending on the soil. The organic matter content,
Considering that the results differ among the different soils, the color, and other factors may potentially affect the reflectance. The
general trend for all four soils is shown in Fig. 5. Four regression linear models of Soil 2 and Soil 3 show the same slopes and
models are obtained from the calibration data sets. The coefficients approximately the same intersects because both soils are
of determination and RMSE are listed in Table 3. The volumetric loess-descended soils with approximately the same mineral
soil moisture content was found to increase with increasing composition. Soil 1 and Soil 4 generally show a slower increase in
absorption depth. soil moisture with increasing absorption depth compared with the
Volumetric Moisture Content (%)

60 60
Volumetric Moisture Content (%)

Soil 1 Soil 2
50 50

40 40

30 30

20 20
y = 209.8x - 5.5 y = 250.6x - 6.1
10 R² = 0.701 10 R² = 0.603

0 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0 0.05 0.1 0.15 0.2 0.25 0.3
Relative Absorption Depth
Relative Absorption Depth

60 60
Volumetric Moisture Content (%)
Volumetric Moisture Content (%)

Soil 4
Soil 3
50 50

40 40

30 30

20 20
y = 290.3x - 11.9 y = 204.2x - 6.1
10 R² = 0.785 10
R² = 0.863

0 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0 0.05 0.1 0.15 0.2 0.25 0.3
Relative Absorption Depth
Relative Absorption Depth

Fig. 5. Relative absorption depth as a function of volumetric moisture content of different soils.
106 Z. Yin et al. / Computers and Electronics in Agriculture 99 (2013) 101–107

Table 3
Parameters of fitting relative absorption depth to soil moisture content by linear model (h = aR + b) for four individual soils and all soils combined.

Sample no. Individual soil All Soils

2
a b R RMSE (%) a b R2 RMSE (%)
Soil 1 209.8 5.5 0.701 2.88 6.74
Soil 2 250.6 6.1 0.603 4.67 204.9 2.1 0.642 7.97
Soil 3 290.3 11.9 0.785 5.27 8.62
Soil 4 204.2 6.1 0.863 3.58 9.26

60 4.4. Model evaluation

Volumetric Moisture Content (%)

50 Using the calibration models developed, moisture is predicted

for each soil in the validation set. RMSE values are calculated for
40
all models under the predicted soil moisture levels. The models
are compared by looking at the R2 and RMSE values. A high R2 value
30
and low RMSE is the best model. The scatter plots of the predicted
20 moisture are also generated for each model, which illustrate the
model performance. The data points of the measured moistures
10 y = 204.9x - 2.1 and the predicted are scattered close to the 1:1 line to indicate
R² = 0.642
good model performance.The volumetric moisture contents of
0 the four soils measured by the NIR reflectance sensor and by the
0 0.05 0.1 0.15 0.2 0.25 0.3
gravimetric method are compared, as shown in Fig. 7. The two data
Relative Absorption Depth
sets are highly correlated and all the data points are located close
to the 1:1 line (R2 = 0.902, 0.736, 0.890 and 0.885; RMSE = 2.88%,
Fig. 6. Relative absorption depth as a function of volumetric moisture content of all
soils combined. 4.67%, 5.27%, and 3.58% respectively for the four soils). Soil 4 shows
the best reflectance model, with a calibration R2 value of 0.863, cal-
ibration RMSE of 4.15%, and validation RMSE of 3.58%. The plot of
other two soils. Soil 1 and Soil 4 have higher sand contents than the the predicted versus the measured volumetric moisture contents
other two soils (Table 1). Larger particles with smaller specific for Soil 2 and Soil 3 shows that most predicted values underesti-
surface area require less water to cover all the particles when mated the volumetric moisture content, as the data points are
compared with smaller particles (White, 2006). Therefore, at the mainly located above the 1:1 line. The average relative error is
same absorption depth, Soil 2 and Soil 3 require more water to cover about 10% for the two models, which may have originated from
all the particles to enhance the forward scattering and absorption of uncertainty during calibrating models and/or systematic error be-
radiation light, when compared with soils of bigger particle tween the two groups of experiments. The scatter plots indicate
sizes. that the sensor produces highly linear correlations between the

60 60
Soil 1
50 Soil 2
50
Predicted Moisture ( %)
Predicted Moisture (%)

1:1 1:1
40 40

30 30
y = 0.988x y = 0.91x
20 R² = 0.902 20
R² = 0.736
RMSE=2.88% RMSE=4.67%
10 10

0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Measured Moisture (%) Measured Moisture (%)

60 60
Soil 3
Soil 4
50 50
Predicted Moisture (%)
Predicted Moisture (%)

1:1 1:1
40 40

30 30
y = 1.053x
y = 0.877x
20 20 R² = 0.885
R² = 0.890
RMSE=3.58%
10 RMSE=5.27%
10
0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Measured Moisture (%) Measured Moisture (%)

Fig. 7. Measured versus predicted volumetric moisture content of the four soils tested.
 Z. Yin et al. / Computers and Electronics in Agriculture 99 (2013) 101–107
predicted and measured volumetric moisture contents with high
precision and relatively low accuracy. To improve the accuracy,
the method may require further modifications. More investigations
are needed to further understand the sensor’s performances. The
performances of the four models are very similar, and all RMSE val-
ues are lower than 6% to indicate that the predicted results of four
linear models are acceptable. Therefore, the soil surface moisture
can be predicted over the full moisture range from dry to
saturation.
5. Conclusion
The NIR reflectance method for measuring soil moisture is pre-
sented. A portable optical sensor is designed and used to measure
the reflectance of soil surface. Four soils are collected and tested to
construct the relationship between the soil moisture content and
relative absorption depth. The predicted soil moisture shows good
linear correlation with relative absorption depth. The measured
soil moistures agree well with the predicted, with an R2 value
above 0.7. The results of the present study indicate the feasibility
of estimating surface soil moisture by measuring the reflectance
from soil surface by the LED light source of 1800 nm and
1940 nm wavelengths. However, the results also show that reflec-
tance model for soil moisture measurement depends on soil type.
The reflectance is measured in the laboratory when the soil surface
is clean and soil materials are sieved beforehand. Given that actual
field conditions are different, more work is required to develop a
more efficient method for soil moisture determination in the field.
References
Baumgardner, M.F., Silva, L.F., Biehl, L.L., Stoner, E.R., 1985. Reflectance properties of
soils. Advances in Agronomy 38.
Belisle, W.R., Sharma, A., Coleman, T.L., 1996. An optical reflectance technique for
soil moisture measurement – Part 1: Theory, description and application.
Geoscience and Remote Sensing Symposium 2, 1315–1319.
Ben-Dor, E., Banin, A., 1995. Near-infrared analysis as a rapid method to
simultaneously evaluate several soil properties. Soil Science Society of
America Journal 59, 364–372.
Ben-Dor, E., Heller, D., Chudnovsky, A., 2008. A novel method of classifying soil
profiles in the field using optical means. Soil Science Society of America Journal
72, 1113–1123.
Bowers, S.A., Hanks, R.J., 1965. Reflection of radiant energy from soils. Soil Science
100, 130–138.
Brutsaert, W.H., 1983. Evapotranspiration into the Atmosphere. D. Reidel Publishing
Company, London, p. 299.
Büning-Pfaue, H., 2003. Analysis of water in food by near infrared spectroscopy.
Food Chemistry 82, 107–115.
Chanasyk, D.S., Naeth, M.A., 1996. Field measurement of soil moisture using
neutron probes. Canadian Journal of Soil Science 76, 317–323.
Clarke Topp, G., Reynolds, W.D., 1998. Time domain reflectometry: a seminal
technique for measuring mass and energy in soil. Soil and Tillage Research 47,
125–132.
Daniel, K.W., Tripathi, N.K., Honda, K., Apisit, E., 2004. Analysis of VNIR spectral
signatures for measurement of soil organic matter in tropical soils of Thailand.
International Journal of Remote Sensing 25, 643–652.
107
Dobriyal, P., Qureshi, A., Badola, R., Hussain, S.A., 2012. A review of the methods
available for estimating soil moisture and its implications for water resource
management. Journal of Hydrology 458–459, 110–117.
Eller, H., Denoth, A., 1996. A capacitive soil moisture sensor. Journal of Hydrology
185, 137–146.
Evett, S.R., Tolk, J.A., Howell, T.A., 2003. A depth control stand for improved accuracy
with the neutron probe. Vadose Zone Journal 2, 642–649.
Fitzjohn, C., Ternan, J.L., Williams, A.G., 1998. Soil moisture variability in a semi-arid
gully catchment: implications for runoff and erosion control. CATENA 32, 55–
70.
Gardner, W.H., 1986. Method of soil analysis. American Society of Agronomy 9,
493–544.
Gaskin, G.J., Miller, J.D., 1996. Measurement of soil water content using a simplified
impedance measuring technique. Journal of Agricultural Engineering Research
63, 153–159.
Inoue, Y., Watanabe, T., Kitamura, K., 2001. Prototype time-domain reflectometry
probes for measurement of moisture content near the soil surface for
applications to ‘‘on-the-move’’ measurements. Agricultural Water
Management 50, 41–52.
Jackson, T.J., 1993. Measuring surface soil moisture using passive microwave
remote sensing. Hydrological Processes 7, 139–152.
Kano, Y., McClure, W.F., Skaggs, R.W., 1985. A near infrared reflectance soil moisture
meter. ASAE 28, 1852–1855.
Kokaly, R.F., Clark, R.N., 1999. Spectroscopic determination of leaf biochemistry
using band-depth analysis of absorption features and stepwise multiple linear
regression. Remote Sensing of Environment 67, 267–287.
Levitt, D.G., Simpson, J.R., Huetu, A.R., 1990. Estimates of surface soil water content
using linear combinations of spectral wavebands. Theoretical and Applied
Climatology 42, 245–252.
Liu, W.D., Baret, F., Gu, X.F., Zhang, B., Tong, Q.X., Zheng, L.F., 2003. Evaluation of
methods for soil surface moisture estimation from reflectance data.
International Journal of Remote Sensing 24, 2069–2083.
Loan, P., James, L.S., 2008. Soil water measurement: capacitance. Encyclopedia of
Water Science 1, 1054–1057.
Lobell, D.B., Asner, G.P., 2002. Moisture effects on soil reflectance. Soil Science
Society of America Journal 66, 722–727.
Skidmore, E.L., Dickerson, J.D., Schimmelpfennig, H., 1975. Evaluating surface soil
water content by measuring reflectance. Soil Science Society of America Journal
39, 238–242.
Tarantino, A., Ridley, A., Toll, D., 2008. Field measurement of suction, water content,
and water permeability. Geotechnical and Geological Engineering 26, 751–782.
Van der Meer, F., 2004. Analysis of spectral absorption features in hyperspectral
imagery. International Journal of Applied Earth Observation and
Geoinformation 5, 55–68.
Viscarra Rossel, R.A., McBratney, A.B., 1998. Laboratory evaluation of a proximal
sensing technique for simultaneous measurement of soil clay and water
content. Geoderma 85, 19–39.
Wang, L., Qu, J., 2009. Satellite remote sensing applications for surface soil moisture
monitoring: a review. Frontiers of Earth Science in China 3, 237–247.
Wang, Q., Li, P., Chen, X., 2012. Modeling salinity effects on soil reflectance under
various moisture conditions and its inverse application: a laboratory
experiment. Geoderma 170, 103–111.
Wei, L., Zhang, B., Wang, M., 2007. Effects of antecedent soil moisture on runoff and
soil erosion in alley cropping systems. Agricultural Water Management 94, 54–
62.
White, R., 2006. Principles and Practice of Soil Science: The Soil as a Natural
Resource. Blackwell Publishing, Berlin.
Whiting, M.L., Li, L., Ustin, S.L., 2004. Predicting water content using Gaussian model
on soil spectra. Remote Sensing of Environment 89, 535–552.
Wu, C.-Y., Jacobson, A.R., Laba, M., Baveye, P.C., 2009. Accounting for surface
roughness effects in the near-infrared reflectance sensing of soils. Geoderma
152, 171–180.
Zhu, Y., Weindorf, D.C., Chakraborty, S., Haggard, B., Johnson, S., Bakr, N., 2010.
Characterizing surface soil water with field portable diffuse reflectance
spectroscopy. Journal of Hydrology 391, 133–140.