Atmospheric water vapor is a key climatological variable because of its
capacity to drive energy exchanges between the ocean and the atmosphere
and within the atmosphere by releasing latent heat. Moreover, of all atmospheric
gases, water vapor exhibits the largest spatial and temporal variations
and has the potential to drive a large positive feedback in global warming
scenario (Rind et al., 1991). Global determination of the Total
Precipitable Water vapor (TPW) distribution is important in boosting the
understanding of the physical processes such as hydrological cycle, biospheric-atmospheric
interactions, the energy balance and for monitoring climate change due
to the greenhouse gases effects (Menzel et al., 2002).
On the other hand, the complex interaction between water vapor, aerosol
and clouds renders quantification of water vapor feedback without further
systematic measurements of water vapor, aerosol and clouds very difficult
(Menzel et al., 2003). It is also of particular importance to monitor
seasonal and annual changes in the precipitable water on local to regional
scales in order to monitor drought conditions and decertification processes
as well as floods.
Accurate global mapping of atmospheric water vapor is an important objective
of the global energy and water cycle assessment. However, the local scale
assessment can assist in improved weather forecasting as well as nowcasting
The sparse location of ground monitoring stations, especially in arid
areas, merits the need for an accurate remote sensing technique that can
provide water vapor information on a daily basis with an adequate spatial
resolution (e.g., 1-5 km from MODIS).
Currently available infrared sounder sensors are capable of retrieving
water vapor profiles as a byproduct of remote sensing of the atmospheric
temperature profiles (Mobasheri, 2007). The derived water vapor profile
depends in part on the initial guess for the temperature and moisture
profiles assumed in the inversion equations and consequently being sensitive
to these initial profiles especially close to the surface region (Menzel
et al., 2003).
Kleespies and McMillin (1990) developed a split window technique that
utilizes temperature variation in a heterogeneous terrain. They found
a correlation better than 0.7 between the in situ and retrieved
water vapor. In arid regions where the difference between the surface
and the boundary layer temperature is large (e.g., central region of Iran
during the afternoon), a high correlation was found between the apparent
temperature difference in the 11 and 12 μm bands and total precipitable
A technique based on differential absorption of water vapor was suggested
by Frouin et al. (1990) and Frouin and Middleton (1990) where in
this technique, two water vapor channels centered at the same wavelength
of 0.94 μm but having different band widths (of 17 and 45 nm) were
used. These channels show different sensitivities to the variation in
the amount of atmospheric water vapor but no (or small) sensitivity to
the surface reflectance. Consequently a ratio of the radiances measured
in these two channels could almost be independent of the surface reflectance.
Also the ratio of a water vapor absorption band to a nearby non-absorbing
band can be calibrated for TPW determination (Chylek et al., 2003).
This is the objective of this research.
MATERIALS AND METHODS
Methodology of this work is based on the regression between two TPW values
estimated through Remote Sensing and Radiosonde techniques. From theoretical
points of view, it follows that if the apparent surface temperature (not
corrected for the emissivity effects) is about equal to the mean temperature
of the boundary layer, where usually most of the water vapor resides,
the emitted radiance in infrared and microwave region will not be sensitive
to the boundary layer water vapor (Kaufman and Gao, 1992). In this case,
any emission in the IR or microwave from the surface will be absorbed
by water vapor in the boundary layer and will be remitted from this boundary
layer, thus having little effect on the upwelling radiance (Mobasheri,
In this research, it is decided to base our work on the use of an optical
technique offered by Kaufman and Gao (1992) that utilizes the solar radiation
reflected by the surface for remote sensing of water vapor from the Moderate
Resolution Imaging Spectrometer (MODIS) onboard of Earth Observing platform
Terra and Radiosonde technique simultaneously. Although the procedure
is the same but the selected technique is different. The technique is
applicable to the cloud free images acquired over land.
Although MODIS instrument has a low spatial resolution (1 km), but on
the other hand this deficiency make it capable of having a global coverage
within a period of 2 to 3 days.
Since water vapor has a high spatial and temporal variability; therefore,
a revisiting time of 2 to 3 days make MODIS products ideal for the study
of the biosphere-atmosphere interaction, its relation to global change
and producing global maps of water vapor distribution. In this regards,
two channels of MODIS-N specifically designed for monitoring the global
distribution of water vapor over the land in cloud-free conditions, were
The technique is based on detection of the absorption of solar radiation
by water vapor as it is transmits down to the surface and up to the sensor
through the atmosphere.
Ground-based transmission measurement of water vapor by sunphotometers
from in and around the near IR absorption bands has been reported (Menzel
et al., 2002; Chylek et al., 2003). These measurements were
carried out via transmitted sun light in a channel that corresponds to
the water vapor absorption (0.94 μm) as well as nearby channels in
atmospheric windows (0.87 and 1.03 μm).
A comparison between the reflected solar radiation in the absorbing bands
and the reflected solar radiation in nonabsorbing bands can quantify the
total vertical amount of water vapor. The main uncertainty in the determination
of water vapor comes from evaluation of the surface reflection for heterogeneous
surfaces in the absorption band (Kaufman and Gao, 1992). In these cases,
more non-absorbing (window) channels are required to "predict" the surface
reflectance in the water absorption channel and consequently less accuracy
can be expected.
The good MODIS water vapor spectrum offers a variety of possibilities,
from a strong absorption in a narrow channel around 0.935 μm (used
for detection of water vapor in clouds), to more moderate absorption around
0.95-0.97 μm and to a weaker absorption around 0.91 μm.
Kaufman and Gao (1992) based their technique of remote sensing of water
vapor on a ratio of absorbing to non-absorbing channels (e.g., a ratio
of the measured radiation at 0.94 μm to that of 0.86 μm), or
alternatively on a ratio of a strongly absorbing channel (e.g., a narrow
channel centered at 0.94 μm) to that of a moderately absorbing channel
(e.g., a wide channel centered at 0.94 μm) where the latter technique,
proposed by Frouin et al. (1990) significantly reduces the effect
of surface reflectance on the channel ratio. Of course this also significantly
reduces the sensitivity of the channel ratio to water vapor. Kaufman and
Gao (1992) then related this ratio to the total precipitable water W through
some empirical relationship of the form:
based on reflectance values of different surfaces in absorbing channels
(ρab) and non-absorbing channels (ρnon-ab).
α and β are coefficient that can be determined through some
in situ measurements. For a mixture of all surfaces, values of α
= 0.020 and β = 0.651 can be used (Kaufman and Gao, 1992).
Tw can also be defined as a function of the precipitable water
vapor along the optical path, W* which is related to the total precipitable
water vapor W by Kaufman and Gao (1992):
here θ and θo are view zenith angle and solar zenith
angle, respectively. The coefficients for the best fit to Tw,
as a function of W* for the MODIS water vapor channels suggested by Kaufman
and Gao (1992) is given in Table 1.
Three different band ratios can be assigned to Tw for three
different states of the atmospheric water vapor contents:
Dry atmosphere: In this case, the reflectance ratio of channel
18 to channel 2 as a function of the amount of water vapor is recommended.
In the case of very small water vapor content (W < 0.5 cm water), the
main error in the remote sensing technique may result from uncertainty
in the spectral surface reflectance. To minimize this effect, a ratio
of the narrow channel of 18 to the wide channel 19 can be used. Then Tw
can be found from:
Low to moderate water vapor: For TPW of around 2 cm in nadir or
even less than 2 cm in off-nadir viewing angle, ratio of channel 19 to
channel 2 is suggested (Kaufman and Gao, 1992).
Humid atmosphere: For total precipitable water vapor amount larger
than 4 cm in nadir view condition or even less than this value but for
slant view and illumination conditions, the strong absorption in the proposed
0.915-0.965 μm channel may partially saturate, resulting in lower
sensitivity to water vapor (Table 2). In this case,
a water vapor absorption band in a spectral range corresponding to lower
absorption i.e., 0.890-0.920 μm for remote sensing of water vapor
in humid conditions is preferred (Kaufman and Gao, 1992). This band is
located in spectral region with minimum values of change in Tw.
Table 2 shows the particulars of three water vapor absorption
MODIS bands of 17, 18 and 19.
TPW calculation using radiosonde data: Radiosonde is a telemetric
system that can measure atmospheric parameters such as air temperature,
pressure and humidity as it ascends through the air. Then this system
transmits the collected data to the ground stations. Moreover, wind speed
and direction will also be measured using balloon`s speed and direction
relative to the surface below.
To calculate TPW using radiosonde data the following steps were adopted:
Step 1: Calculation of water vapor partial pressure e: To calculate
vapor partial pressure e which is equal to the saturated partial pressure
at dew point temperature, we used the dew point temperature and the equation
offered by Rogers and Yau (1996):
||Values of α and β for different absorbing
and non-absorbing band and for two viewing angles of nadir (θ
= 0, θo = 40) and off-nadir (θ = 60, θo
= 60) (Kaufman and Gao, 1992)
||Characteristics of the absorption bands used in this
||Dew point temperature
||2.5106 (j kg-1)=Latent heat of evaporation
||461.5 (j k-1 kg-1)=Universal gas constant
Step 2: Specific humidity: Specific humidity can be calculated
using Eq. 8 suggested by Hurly (1994).
Where, q would be in g kg-1. Td and P are from
nearby weather station and es(Td) from Eq. 1 in
Step 3. Calculation of total water vapor: The TPW was worked out
using all calculated q values at different level with pressure P and equation
offered by Carlson et al. (1991):
||1000 kg m-3 is water density
||Specific humidity in g kg-1 and TPW in mm
||Mean acceleration due to gravity in m s-2 which varies
with latitude and height
Site and data selection: Temperature, pressure and dew points
are parameters that are collected and transmitted to the surface by radiosonde
sensors directly. A relatively complete bank of radiosonde data are in
access at the Wyoming University site as well as Iran Meteorological Organization.
These data are being collected on a routine basis two times a day on the
synoptic hours, i.e., 0000Z and 1200Z (GMT). The selected site for this
work was Mehrabad Airport synoptic station located in Southern part of
Tehran capital city of Iran where the location is shown in Fig.
1. This station is located at 51°, 21`E and 35°, 41`N at an
altitude of 1191 m from sea level. The index of this station is OIII and
its number is 40754.
A DEM of the Mehrabad synoptic station region located
at south of capital city of Tehran southern slopes of Alborz Mountains.
White codes are for those stations where radiosondes collect data
on a routine basis
MODIS images are only available from 1999. To select proper data, we
first selected cloud free MODIS images and then selected the radiosonde
data as close as possible to the satellite acquisition time. To carry
out weather analysis for the time of study, the thermodynamic graphs such
as Stuve or Skew-T were used. In these graphs, temperature, pressure and
dew point profiles as well as condensation temperature and pressure were
This helped extraction of information regarding atmospheric stability
as well as the possibility of the presence of small patches of clouds
that can not be detected in the low resolution MODIS images, otherwise.
Presence of small patches of clouds within a pixel increases the uncertainty
of TPW assessment using satellite imageries (Jeffery and Austin, 2003).
Geometrical processing on MODIS images were carried out by Iran Space
Agency. Radiometric correction by pixel to pixel method was carried out
using IMMAP software and parameters retrieved from image header file.
Finally using Scan Magic software, geometrical corrections based on Lambertian
system with elliptical base of WGS84 was done applying nearest neighbourhood
method and parameters retrieved from image header file. Finally using
Scan Magic software, geometrical corrections based on Lambertian system
with elliptical base of WGS84 was done applying nearest neighbourhood
Image selection: Upon thorough investigation and analysis of weather
condition, four MODIS images were selected particulars of which are shown
in Table 3.
Weather analysis: Application of one TPW retrieving algorithm
for two images with similar acquisition time (from seasonal point of view)
may result in completely different values partly due to weather condition
at the satellite passing time. Since water vapor retrieving algorithm
can only be applied in the cloudless sky, then the weather condition (cloudiness,
water vapor, presence of a front) around satellite passing time need to
be investigated using synoptic data. This may increase our confidence
about the retrieved TPW values.
To avoid ambiguities resulting from weather conditions, parameters such
as wind speed and direction, dry and wet bulb temperature, dew point,
soil temperature, relative humidity and partial vapor pressure were investigated
prior to the image selection in this work. Also Stuve graphs at the vicinity
of satellite passing time were plotted using Radiosonde data. Through
these thorough investigations, finally four images were selected (Table
3) where details of their Stuve graphs are shown in Fig.
||Sun and Sensor zenith angles along with conversion coefficients
for four MODIS images
Since MODIS passing time is 7:30 (GMT) then any profiles at the satellite
passing time must be interpolated from the two Radiosonde profiles of
00Z and 12Z. These interpolated profiles are only valid when the two profiles
of 00Z and 12Z show similar shape and behavior. Figure 2
shows that almost in all four different dates, except at the region close
to the surface, the temperature profiles at 12Z do not deviate much from
those of 00Z, although the dew point profiles in three of these dates
at 00Z and 12Z are notably different. This means that the interpolation
of temperature profile for the time of satellite passage would be acceptable.
On the other hand the abrupt deviation of dew point profiles at any altitude
may represent the occurrence of change in water vapor content at that
altitude. This could be due to the condensation, if dew point decreases
(possible cloud formation at 400 mb for Sep 15th and May 26th, 2002) or
entrance of water vapor to that altitude brought by possible fronts and/or
wind as in Sep15 and May 26, 2002 plots. To clarify this, one may use
other profiles such as water vapor density (Fig. 3),
wind speed (Fig. 4) and wind direction (Fig.
As an illustration of the analysis, the anomaly in dew point profile
of Sep15, 2002 and May 26, 2002 at 400 mb level was due to a decrease
in density (absolute humidity) of water vapor as can be seen in Fig.
3 by high westerly winds (Fig. 4, 5).
However this anomaly can not cause any problem in TPW assessment due to
the low water vapor content of the atmosphere at this height (Fig.
The anomaly of dew point profiles at around 700 mb in Fig.
2 show a decrease for Sep 15th and May 26th, 2002 and slight (ignorable)
increase at the same altitude for Sep 17, 2004 and May 26, 2003 between
00Z and 12Z. Figure 3 shows abrupt decrease in absolute
humidity on Sep. 15th and May 26th, 2002 for this time interval while
wind speed is much higher for May 26, 2002 compare to Sep. 15th. The decrease
in water vapor content could be mostly due to the southerly to westerly
wind blowing during this time interval.
||Dewpoint (bold lines) and temperature profiles (non-bold
lines) at 00Z (dashed) and 12Z (solid) in four different satellite
||Water vapor density (absolute humidity) at 00Z (dashed)
and 12Z (solid) in four different satellite passing date
||Wind speed profile at 00Z (dashed) and 12Z (solid) in
four different satellite passing date
||Wind direction at different altitudes (pressures) for
00Z (dashed) and 12Z (solid) in four different satellite passing date
Due to the fact that the great portion of the atmospheric water vapor
content resides close to the surface and upon the above analysis, day
Sep. 15, 2002 has the driest atmosphere and Sep. 17, 2004 has the wettest
atmosphere compare to the other two days where the humidity of the atmosphere
is more moderate. Between these two, May 26, 2002 seems more humid than
May 26, 2003. This also can be concluded from absolute humidity profiles
in Fig. 3. Since in all these profiles, water vapor
abruptly decreases with height, no condensation is likely to happen above
500 mb and consequently there is no chance for cloud formation to occur.
On the other hand according to the classification mentioned in previous
section (Kaufman and Gao, 1992) all of these four states of the atmosphere
can be classified in Humid atmosphere and according to (Kaufman and Gao,
1992) the ratio of band 17 to band 2 must be applied for TPW assessment.
RESULTS AND DISCUSSION
Applying Eq. 8-10 to radiosonde data and Eq. 4- 6 to MODIS reflectance
images resulted in TPWs which are shown in Table 4.
As can be seen in Table 4, none of the four days Sep.
15, 2002; May 26, 2002; May 26, 2003 and Sep. 17, 2004 have possessed
the conditions mentioned for dry atmosphere but as a comparison, as mentioned
before Sep. 15, 2002 has the driest and Sep. 17, 2004 has the wettest
atmosphere among these four days.
According to Kaufman and Gao, (1992) classification reflectance ratio
of band 17 to band 2 was expected to be the most suitable one for TPW
assessment in all these four atmospheric states, but in practice the worst
correlation between Radiosonde and satellite TPW was found with this ratio
The possible explanation for this could be the special geography of the
region where all of the atmospheric water vapor contents are trapped in
the height between 1200 to 3000 m and it is not distributed logarithmically
in height as is expected for a normal atmosphere. A regression between
satellite derived TPW with the Radiosonde TPW were ran using least square
method results of which is shown in Table 5.
In Table 5, the poorest correlation is for 17/2 (0.00)
and the best one is for 18/19 (0.93). It is believed that the ratio of
the two water vapor channels centered at the same wavelength (such as
MODIS channel 19 (0.94 μm) and channel 18 (0.936 μm) having
different band widths (50 and 10 nm, respectively) may renders the effects
of surface reflectance on the results, minimum. This is because these
channels show different sensitivity to the variation of the amount of
atmospheric water vapor but no (or small) sensitivity to the surface reflectance
(Frouin et al., 1990; Frouin and Middleton, 1990). Consequently
a ratio of the reflectance measured in these two channels could almost
be independent of the surface reflectance.
This might be the case for Mehrabad urban region where variety of surface
covers is present in each pixel. The second suitable ratio is 18/2 (0.84)
that is suggested for moderate atmospheric humidity content by Kaufman
and Gao, (1992). This could be suitable for the region because of the
geometry where the Alborz Mountains with height of more than 4000 m have
surrounded the region.
Figure 6 shows the regression between shaded boxes
in Table 5 and radiosonde TPW data with a correlation
coefficient of 0.81 where we call it Combination Method. Although the
correlation coefficient is a little lower in Combination Method compare
to 18/19 ratio but since we are using three band ratios (including 18/19
which minimizes the surface reflectance effects) rather than one band
ratio (18/19), the uncertainties due to the threshold values suggested
by Kaufman and Gao (1992) might decrease. However in practice Combination
Method has its own difficulties and needs more in situ measurement
to determine the coefficients such as a, b, c and d in the following equation:
TPW(Radiosonde) = a.TPW18/19+
Of course these coefficients may depend on the region`s topography and/or
geography where these authors are working on it.
Then the suggested model for the region under study would be of the form;This
may increase the accuracy of satellite TPW assessment up to 20% which
is important for the forecasting of natural disasters such as flood in
mountainous regions like Tehran.
||Results of calculated TPW using radiosonde and satellite
data. Shaded data are selected for regression. Shaded boxes are for
a combination of band ratios
||Regression coefficients between satellite and radiosonde
||Regression between radiosonde estimated TPW and satellite
derived TPW (shaded boxes in Table 5), R2
We would like to acknowledge valuable assistance of Mr. Yusef Rezaie
in image processing and also helpful cooperation of Iran Space Agency
in providing MODIS images for this research. Also we would like to appreciate
I.R. of Iran Meteorological Organization for providing Radiosonde and
synoptic data. Finally this research has been funded by Water Resource
Management of I.R. Iran`s Energy Ministry and we appreciate their thoughtful