Adv. Geosci., 30, 23–29, 2011
www.adv-geosci.net/30/23/2011/
doi:10.5194/adgeo-30-23-2011
© Author(s) 2011. CC Attribution 3.0 License.
Advances in
Geosciences
Estimation of spatio-temporal distribution of precipitable water
using MODIS and AVHRR data: a case study for Cyprus
D. Hadjimitsis1, Z. Mitraka2, I. Gazani2, A. Retalis3, N. Chrysoulakis2, and S. Michaelides4
1Cyprus University of Technology, Department of Civil Engineering & Geomatics, Lemesos, Cyprus
2Foundation for Research and Technology – Hellas, Institute of Applied and Computational Mathematics, Greece
3National Observatory of Athens, Institute for Environmental Research and Sustainable Development, Greece
4Cyprus Meteorological Service, Nicosia, Cyprus
Received: 16 September 2010 – Revised: 3 February 2011 – Accepted: 10 March 2011 – Published: 9 May 2011
Abstract. In this paper, the atmospheric precipitable water
(PW) over the area of Cyprus was estimated by means of
Advanced Very High Resolution Radiometer (AVHRR) ther-
mal channels brightness temperature difference (1T ). The
AVHRR derived 1T was calculated in a grid of 5× 5 km
cells; the corresponding PW value in each grid cell was ex-
tracted from Moderate Resolution Imaging Spectroradiome-
ter (MODIS) Level 2 product (near-infrared algorithm).
Once the PW – 1T relationship coefficients corresponding
to the area of Cyprus were calculated, the relationship was
applied to AVHRR data for one month period. Radiosonde
derived PW values, as well as MODIS independent PW val-
ues were used to validate the estimations and a good agree-
ment was noted.
1 Introduction
Atmospheric water vapor is the principal contributor to the
greenhouse effect and plays a key role in our understanding
of the Earth’s climate. Precipitable water (PW) is the amount
of vertically integrated water vapor and can be expressed in
g cm−2, or as the height of an equivalent column of liquid
water in cm. PW is an important component of the hydro-
logical cycle and it is adopted as an input variable in global
climatological studies. Moreover, it has the potential to sup-
port hydrological, biospheric and atmospheric modeling ef-
forts, both at the local and regional scales, since it is widely
used in energy budget and evapotranspiration studies. PW
is also of an essential requirement in atmospheric correction
of high spatial resolution satellite data and it also necessary
Correspondence to: D. Hadjimitsis
(d.hadjimitsis@cut.ac.cy)
for the enhancement of the precision of land surface tem-
perature estimates obtained from satellite data. The satellite
derived high-resolution PW retrievals can be useful in antici-
pating the distribution of precipitation patterns. Thus, it is of
particular importance to monitor PW variability on regional
scales in order to monitor drought conditions and desertifica-
tion processes.
Retrieving PW using Moderate Resolution Imaging Spec-
troradiometer (MODIS) near-infrared records is based on de-
tecting the water vapor absorption of the reflected solar radi-
ation, after it has been transmitted down through the atmo-
sphere, reflected at the surface and subsequently transmitted
up through the atmosphere to the sensor. The equivalent to-
tal vertical amount of water vapor is derived from a com-
parison between the reflected solar radiation in the absorp-
tion channel and the reflected solar radiation in nearby non-
absorption channels (Gao and Kaufman, 2003). Three near-
infrared channels were implemented on MODIS for water
vapor sensing. PW is estimated over clear land areas, clouds
and oceanic areas with sun glint (King et al., 2003). Typical
errors for MODIS derived PW over land range between 5%
and 10% (Gao and Kaufman, 2003).
A number of algorithms have been proposed to derive PW
from Advanced Very High Resolution Radiometer (AVHRR)
thermal infrared observations and can be classified into four
main categories: simple split-window of thermal channels
(Dalu, 1986), variance ratio (Jedlovec, 1990), regression
slope (Goward et al., 1994), covariance – variance method
and look-up table approach (Czajkowski et al., 2002). It
should be noted that emissivity over land surfaces presents a
significant complication in interpreting split window temper-
ature differences, however accuracies of better than 0.5 cm
water can be achieved (Eck and Holben, 1994).
Published by Copernicus Publications on behalf of the European Geosciences Union.
24 D. Hadjimitsis et al.: Precipitable water using MODIS and AVHRR over Cyprus
Theoretical analysis proposes an almost linear relationship
between PW and 1T , with slope and intercept being deter-
mined by the surface emissivity. Spectral variation of emis-
sivity is found to be the major factor that causes deviation
of the intercept from a near-zero value (Choudhury et al.,
1995). Assuming that the surface is Lambertian and the land-
atmosphere system is spatially homogeneous, a split win-
dow algorithm can be applied to AVHRR thermal infrared
channels (Choudhury and Di Girolamo, 1995). As is has
been shown in past studies (Dalu, 1986; Eck and Holben,
1994; Choudhury and DiGirolamo, 1995; Czajkowski et al.,
2002), if the radiative transfer equation is linearized, the PW
is proportional to the difference between the two AVHRR
thermal channels brightness temperature. This split-window
technique is not a universal solution to PW retrieval due to
the dependence of the difference 1T on air temperature and
surface characteristics. Linear relationships characterized by
different coefficients are expected to hold in different loca-
tions. The coefficients of the linear relationship between PW
and 1T are site specific and can be derived as output from
a radiative transfer model or as observational data at each
location.
Let T4 and T5 denote the brightness temperatures of
AVHRR channels 4 and 5, respectively. Analysis of the
split window temperature differences 1T = T4−T5, for land
surfaces in several different climatic regions with differing
soil compositions and vegetative covers, revealed a linear
relationship between 1T and PW (Eck and Holben, 1994;
Choudhury et al., 1995; Chrysoulakis et al., 2008). The
accuracy of the split window technique is uncertain in arid
environment during daytime due to strong near-surface air
temperature gradient (Choudhury and Di Girolano, 1995).
Provided that the channels 4 and 5 radiances are accurately
calibrated, the AVHRR split window technique of PW esti-
mation has the potential to provide consistent estimates of
PW. This technique can be referenced to a given time pe-
riod of independent data or analyzed PW fields and thus pro-
vide consistent estimates for each location (Chrysoulakis et
al., 2008). As discussed by Choudhury et al. (1995), spatial
and temporal changes of 1T and PW can also be associated
with the changes of land surface characteristics and hence
the emissivity. In this paper, the MODIS precipitable wa-
ter Level 2 product is related to the AVHRR derived 1T .
All spatial calculations were performed in a grid of 5× 5 km
cells over the study area. Data examination confirmed a lin-
ear relationship between adjusted PW and 1T . The PW –
1T relationship was applied to the AVHRR data to provide
a PW spatial distribution. The distribution was validated
against radiosonde derived PW time series at one synoptic
station and against independent MODIS PW measurements.
AVHRR data acquired during July 2006 were used. In a pre-
vious study, Chrysoulakis et al. (2008) have calculated a 1T
– PW relationship, corresponding to a different geographic
area. AVHRR PW values were estimated using the relation-
ship derived from their study for comparison.
  4
In this paper, the MODIS precipitable water Level 2 product is related to the AVHRR 
derived ΔΤ. All spatial calculations were performed in a grid of 5 x 5 km cells over 
the study area. Data examination confirmed a linear relationship between adjusted PW 
and ΔΤ. The PW – ΔT relationship was applied to the AVHRR data to provide a PW 
spatial distribution. The distribution was validated against radiosonde derived PW 
time series at one synoptic station and against independent MODIS PW 
measurements. AVHRR data acquired during July 2006 were used. In a previous 
study, Chrysoulakis et al. (2008) have calculated a ΔΤ – PW relationship, 
corresponding to a different geographic area. AVHRR PW values were estimated 
using the relationship derived from their study for comparison.  
2   Study area and datasets 
2.1 The study area 
The study area is the island of Cyprus. The area is shown in Fig. 1 and a grid 
comprising 5 x 5 km cells is superimposed. The PW – ΔΤ relationship was calibrated 
at 5 x 5 km cell level; for this reason, the 1 x 1 km AVHRR and MODIS derived data 
were spatially aggregated in 5 x 5 km cells.  
In particular, ΔΤ was estimated at AVHRR pixel level and the mean value in each 
grid cell was computed. Likewise, the PW values extracted at 1.1 x 1.1 km level from 
the MODIS Level 2 product were aggregated in 5 x 5 km cells. To calibrate the PW – 
ΔT relationship, 16 MODIS and AVHRR acquisitions were used.  
 
Fig. 1. The study area. The MODIS derived PW and the AVHRR
derived 1T were spatially averaged in 5× 5 km. For this rea-
son a grid of 5× 5 km cells was developed (Projection: Geo-
graphic/WGS84). The location of the Athalassa synoptic station
is shown.
2 Study area and datasets
2.1 The study area
The study area is the island of Cyprus. The area is shown in
Fig. 1 and a grid comprising 5× 5 km cells is superimposed.
The PW – 1T relationship was calibrated at 5× 5 km cell
level; for this reason, the 1× 1 km AVHRR and MODIS de-
rived data were spatially aggregated in 5× 5 km cells.
In particular, 1T was estimated at AVHRR pixel level and
the mean value in each grid cell was computed. Likewise, the
PW values extracted at 1.1× 1.1 km level from the MODIS
Level 2 product were aggregated in 5× 5 km cells. To cal-
ibrate the PW – 1T relationship, 16 MODIS and AVHRR
acquisitions were used.
The time difference between the NOAA 17/AVHRR
or NOAA 18/AVHRR and Aqua/MODIS or Terra/MODIS
satellites was found to be between 7 and 37 min, for each
one of the selected 16 days. The time lag between the two
satellites is small enough, allowing us to consider the satel-
lites overpasses as concurrent. Near nadir, AVHRR acqui-
sitions over the study area were used, therefore, the effect
of satellite zenith angle was ignored. Furthermore, based on
MODIS Level 2 aerosol product, it was concluded that the
atmospheric aerosol content was low for the selected days
and thus the effect of aerosols in thermal infrared radiation
absorption was ignored.
2.2 The MODIS data
The MODIS radiometer has 36 channels that cover the spec-
tral region between 0.4 and 15 µm. Five near-infrared chan-
nels centered at 0.865, 0.905, 0.936, 0.940 and 1.240 are
used for PW sensing. The channels at 0.865 and 1.24 µm
positioned to avoid atmospheric gaseous absorption are used
for remote sensing of vegetation and clouds. The channels
at 0.936, 0.940, and 0.905 µm are water vapor absorption
Adv. Geosci., 30, 23–29, 2011 www.adv-geosci.net/30/23/2011/
D. Hadjimitsis et al.: Precipitable water using MODIS and AVHRR over Cyprus 25
channels with decreasing absorption coefficients. The strong
absorption channel at 0.936 µm is most useful for dry con-
ditions, while the weak absorption channel at 0.905 mm is
most useful for very humid conditions, or low solar elevation
(Gao and Kaufman, 2003).
The MODIS Level 2 precipitable water product consists
of column water vapor amounts. The results of the near-
infrared retrieval algorithm over land were used in this study;
this algorithm (Gao and Kaufman, 2003) is applied over clear
land areas of the globe and above clouds, over both land and
ocean, during daytime. Over clear ocean areas, water va-
por estimates are provided over the extended glint area. The
Level 2 data are generated at the 1-km spatial resolution of
MODIS using the near-infrared algorithm during the day. In
this study, the Aqua and Terra MODIS PW retrievals for the
study area and for 16 cases in July 2006 (Table 1) were se-
lected; the selection criteria are explained in Sect. 3.1.
2.3 The AVHRR data
Records of AVHRR onboard NOAA 17 and NOAA 18 ac-
quired over the study area for July 2006 were used to de-
rive the spatio-temporal distribution of PW. The AVHRR is
a five channel instrument, with three of the spectral chan-
nels located in the visible, near-infrared and mid infrared
regions of the spectrum, while the remaining two are lo-
cated in the thermal infrared with effective wavelengths cen-
tered around 10.8 µm (channel 4), and 12 µm (channel 5). It
has the same spatial characteristics with previous versions of
AVHRR: spatial resolution of 1.1 km at nadir and swath cov-
erage of 2700 km. For the present study, the AVHRR data
were available from both NOAA Satellite Active Archive and
the HRPT ground receiving station of the Foundation for Re-
search and Technology – Hellas (Chrysoulakis et al., 2008).
For the PW – 1T relationship calibration, data correspond-
ing to the 16 cases displayed in Table 1 were used.
3 Methodology
3.1 Overview
In this study, PW is estimated using MODIS as an indepen-
dent dataset. Therefore, the MODIS derived PW can be re-
lated to the AVHRR derived 1T by:
PW(x,y)= a1T (x,y)+b (1)
where, PW(x,y) is the total column precipitable water in the
(x,y) cell of the grid x,y are the coordinates of the center of
the cell, a and b are coefficients that need to be estimated for
the study area, 1T = T4−T5 is the brightness temperatures
for channels 4 and 5 in the (x,y) cell. The data for AVHRR
channels 4 and 5 are given in digital numbers (Level 1b for-
mat), which are converted to radiance at the sensor and fi-
nally to brightness temperatures to the allowable accuracy of
0.5 K (Chrysoualkis et al., 2008).
Table 1. Dates and time passes of Terra, Aqua, NOAA 17 and
NOAA 18 acquisitions used in this study. The pass time difference
is between 7 and 37 min.
Date Satellite AVHRR Satellite MODIS Pass
Pass Time Pass Time Time
(UTC) (UTC) Difference
2 July NOAA 18 10:55 Aqua 10:40 0:15
3 July NOAA 17 08:37 Terra 08:05 0:32
3 July NOAA 18 10:45 Aqua 11:20 0:34
8 July NOAA 17 08:22 Terra 08:25 0:02
10 July NOAA 18 11:14 Aqua 11:25 0:10
11 July NOAA 18 11:04 Terra 10:30 0:34
12 July NOAA 17 08:30 Terra 08:00 0:30
12 July NOAA 18 10:53 Aqua 11:15 0:21
13 July NOAA 17 08:10 Terra 08:45 0:34
13 July NOAA 18 10:43 Aqua 10:20 0:23
14 July NOAA 18 10:33 Aqua 11:05 0:31
17 July NOAA 17 08:18 Aqua 08:20 0:01
21 July NOAA 17 08:23 Terra 07:55 0:28
21 July NOAA 18 11:02 Terra 11:10 0:07
22 July NOAA 18 10:52 Aqua 10:15 0:37
23 July NOAA 18 10:41 Terra 10:55 0:13
In this study, the calculation of the coefficients a and b was
based on MODIS and AVHRR data acquired during the se-
lected days of June 2006 (Table 1). The selection criteria for
these days were: (a) as much cloud free land area as possible
and b) near nadir satellite acquisitions over the study area.
The PW – 1T relationship was applied to the AVHRR
dataset to provide PW spatial distributions for a summer
month (July 2006) and was validated using radiosonde de-
rived PW values from the synoptic station of Athalassa
(35.15◦ E 33.40◦ N), as well as MODIS PW retrievals.
3.2 MODIS derived precipitable water
As noted before, MODIS PW retrievals for July 2006 were
used. The data were available as images (PW scenes from
collection 5) in Hierarchical Data Format (HDF). A Matlab
script was used to read the HDF files, to extract the PW val-
ues for each pixel and to assign a flag value to all the cloudy
pixels of each PW scene.
Information about cloud cover was included in HDF for-
mat. A cloud mask in bit-level representation provided cri-
teria that guide one to decide if a pixel of the PW scene was
cloudy or not. Pixels that considered being less than 5%
cloudy were used.
Navigation parameters were also included in HDF for-
mat; these parameters were used to correct the panoramic
distortion and to project each PW scene in a common pro-
jection system (Geographic/WGS84). The nearest neigh-
bor resampling method was used to geometrically correct
the PW scenes. Subsequently, the MODIS derived PW was
transformed to geometrically corrected scenes at 1 km spatial
www.adv-geosci.net/30/23/2011/ Adv. Geosci., 30, 23–29, 2011
26 D. Hadjimitsis et al.: Precipitable water using MODIS and AVHRR over Cyprus
resolution. Real PW values were assigned to non cloudy
pixels whereas the cloudy pixels were masked. Finally, all
scenes were spatially averaged to the 5× 5 km cell grid cov-
ering the study area. It should be noted that the spatial aver-
aging was performed if all pixels corresponding to each grid
cell were non-cloudy. If one or more cloudy pixels were in-
cluded, the corresponding cell was not taken into account in
calculations.
3.3 AVHRR derived brightness temperature difference
Standard pre-processing steps were employed to convert the
raw AVHRR records to calibrated values. The calibration
procedure is based on the conversion of digital numbers of
each image to radiance at the sensor values for visible and
near-infrared channels (channels 1 and 2) and to brightness
temperature values for the thermal infrared channels (chan-
nels 3, 4 and 5) with the use of the non-linear conversion
equations given in the NOAA KLM Polar Orbiter Data Users
Guide (Chrysoulakis et al., 2008). The intercept and gain
for each AVHRR channel, as well as the parameters related
to the central wave number and the constants needed to pa-
rameterize the calibration equations were extracted from the
NOAA Level 1b header of each image file. The platform
flight altitude was also extracted from the same source and
it was used for the geometric correction and the correc-
tion of the panoramic distortion in calibrated AVHRR im-
ages. The standard NOAA/AVHRR projection system (Ge-
ographic/WGS72) was used. The images were finally repro-
jected to Geographic/WGS84 to be combined with the pro-
jected MODIS PW products. The nearest neighborhood re-
sampling method was used in all cases.
To derive the spatial distribution of 1T at 5× 5 km level,
a cloud mask (Chrysoulakis and Cartalis, 2003) was used
to detect cloudy pixels. Following, 1T was calculated at
AVHRR pixel level (1.1 km) for non-cloudy pixels. 1T val-
ues were spatially averaged at the 5× 5 km level assigning a
mean 1T value in each grid cell. As for the MODIS data, the
spatial averaging was performed if all pixels corresponding
to each grid cell were non-cloudy. The above steps were re-
peated for all NOAA 17 and 18 acquisitions of July 2006.
Cloudy acquisitions, as well as acquisitions of significant
time difference with Terra or Aqua over the study area, were
considered worthless and were not included in the analysis.
3.4 Cross-calibration, application and validation
The coefficients in Eq. (1) were derived by applying a lin-
ear regression procedure. The derived PW – 1T relation-
ship was applied to produce PW spatial distributions over the
study area using AVHRR thermal infrared measurements. As
noted previously, 1T was estimated for cloud free pixels of
geometrically corrected scenes acquired over the study area
for July 2006. Daytime acquisitions were used.
  11
Table 2. Estimates for the coefficients of Eq. (1). 
 Coeff. S.E. t-value P-value 
Constant 0.747 0.054 13.869 <0.0001 
ΔΤ 0.583 0.016 
R2= 0.691 
36.131 <0.0001 
 
 
 
 
Fig. 2. Scatterplot of MODIS derived PW versus AVHRR derived ΔT. 
The derived PW – ΔΤ relationship was applied to the dataset for July 2006 and PW 
istributions derived from AVHRR records. AVHRR acquisitions were chosen to be 
between 10:00 and 12:00 UTC, because the radiosonde measurements were taken at 
11:00 UTC. The resulted AVHRR derived PW values were validated using 
radiosonde measurements for the Athalassa synoptic station. As mentioned before, a 
nu ber of radiosonde measurements at some heights were missing and linear 
interpolation methods were used to fill in the missing values. This may have resulted 
in not so accurate radiosonde derived PW values.  
A relatively good agreement between radiosonde and AVHRR derived PW values 
was observed (RMSE= 0.76 cm). Figure 3a shows the radiosonde derived PW values 
versus the AVHRR derived PW values. AVHRR derived PW values seem to be 
overestimated, but this is the result of underestimated radiosonde PW values. This 
underestimation may have been resulted by the interpolated radiosonde measurements 
Fig. 2. Scatterplot of MODIS derived PW versus AVHRR derived
1T .
Radiosonde derived PW values at Athalassa synoptic sta-
tion were used (35.15◦ N, 33.40◦ E) for validation. The
radiosonde data (pressure, temperature and humidity mea-
surements at several levels at 12:00 UTC) for the study pe-
riod (July 2006) were obtained from the UK Meteorolog-
ical Office (available online at: http://badc.nerc.ac.uk/data/
radiosglobe). The radiosonde measurements provided a set
of time series that was used as an independent dataset.
Radiosondes are expected to produce PW with an uncer-
tainty of 0.1–0.2 cm, which is considered to be the accu-
racy standard of PW for meteorological applications. The
PW was estimated using the appr ach described by Cartalis
and Chrysoulakis (1997). Missing records at some heights in
the radiosonde data were estimated using a linear regression
(Chrysoulakis et al., 2008).
It was assumed that the radiosonde measurements at the
synoptic station were representative for the 5× 5 km cell in
which t e statio was included. Therefore, the radiosonde
derived PW corresponding to the station location, was as-
signed to the respective grid cell. In this way, the radiosonde
derived PW and the AVHRR derived PW can be compared
at cell level. The Root Mean Squared Error (RMSE) error
was calculated and scatterplot diagrams were produced for
the visualization of the validation results.
To validate further the PW estimates, these were compared
to a MODIS PW independent dataset for July 2006. To mea-
sure how close PW estimates were to MODIS PW values,
the RMSE was calculated for the cloud free cells. Validation
was performed for 20 independent cases in July 2006. More-
over, to visualize biases in estimation, a scatterplot diagram
of the estimated PW values versus the MODIS PW values
was used.
Adv. Geosci., 30, 23–29, 2011 www.adv-geosci.net/30/23/2011/
D. Hadjimitsis et al.: Precipitable water using MODIS and AVHRR over Cyprus 27
  12
at some heights, as it has been explained above. The RMSE between the radiosonde 
derived and MODIS derived PW values found to be 0.75 cm. 
As mentioned before, additional comparisons were performed to compare the results 
to the previous study (Chrysoulakis et al., 2008). AVHRR PW was estimated for the 
study area for July 2006 using the relationship proposed by Chrysoulakis et al. (2008). 
In Fig.3b the radiosonde derived PW values versus the AVHRR derived PW values 
are shown. Comparing Figs. 3a and 3b, it is observed that the relationship formulated 
in the present study gave more accurate results. This confirms the assertion that PW – 
ΔΤ relationship is site-specific and depends on land cover and emissivity.  
 
a) b) 
Fig. 3. Scatterplots of radiosonde derived PW versus AVHRR derived PW, estimated 
for cloud free cells corresponding to the synoptic station of Athalassa, a) using the 
relationship formulated for the study area and b) using the relationship formulated for 
the broader area of Greece. 
Fig. 3. Scatterplots of radiosonde derived PW versus AVHRR derived PW, estimated for cloud free cells corresponding to the synoptic
station of Athalassa, (a) using the relationship formulated for the study area and (b) using the relationship formulated for the broader area of
Greece.
Additional comparisons were performed to compare the
results to a previous study referring to a different geographic
area. Chrysoulakis et al. (2008) have estimated AVHRR
1T – PW linear relationship coefficients for the broader
area of Greece. These coefficients were also estimated for a
5× 5 km cell grid. Cyprus is geographically close to Greece
and has similar topographic characteristics. The equation
calculated for the area of Greece should result in relatively
good PW estimations for the study area. To test this assump-
tion, PW values were estimated for June 2006, using the re-
lationship of Chrysoulakis et al. (2008).
It is worth mentioning the attempt to perform all the cal-
culations in this study at a lower scale: the methodology de-
scribed above was applied to the initial available data but all
calculations were performed in a grid of 3× 3 km cells; how-
ever, the PW – 1T values referring to 3× 3 km cells were not
well correlated, so the 5× 5 km grid was finally selected.
4 Results and discussion
The PW – 1T relationship was calibrated using cloud free
cells from the 16 cases in Table 1. A significant correla-
tion was identified between 1T and PW (Pearson r = 0.83,
p < 0.01). The regression sample consisted of 587 PW –
1T pairs. A simple linear regression method was used for
the derivation of the coefficients in Eq. (1). The results of
the regression analysis are shown in Table 2. The t-statistic
together with the corresponding p-values indicate that all es-
timates are statistically significant. A PW – 1T scatterplot
depicting the regression analysis is shown in Fig. 2.
Table 2. Estimates for the coefficients of Eq. (1).
Coeff. S.E. t-value P-value
Constant 0.747 0.054 13.869 < 0.0001
1T 0.583 0.016 36.131 < 0.0001
R2= 0.691
The derived PW – 1T relationship was applied to the
dataset for July 2006 and PW distributions derived from
AVHRR records. AVHRR acquisitions were chosen to be
between 10:00 and 12:00 UTC, because the radiosonde mea-
surements were taken at 11:00 UTC. The resulted AVHRR
derived PW values were validated using radiosonde measure-
ments for the Athalassa synoptic station. As mentioned be-
fore, a number of radiosonde measurements at some heights
were missing and linear interpolation methods were used to
fill in the missing values. This may have resulted in not so
accurate radiosonde derived PW values.
A relatively good agreement between radiosonde
and AVHRR derived PW values was observed
(RMSE = 0.76 cm). Figure 3a shows the radiosonde
derived PW values versus the AVHRR derived PW values.
AVHRR derived PW values seem to be overestimated,
but this is the result of underestimated radiosonde PW
values. This underestimation may have been resulted by
the interpolated radiosonde measurements at some heights,
as it has been explained above. The RMSE between the
radiosonde derived and MODIS derived PW values found to
be 0.75 cm.
www.adv-geosci.net/30/23/2011/ Adv. Geosci., 30, 23–29, 2011
28 D. Hadjimitsis et al.: Precipitable water using MODIS and AVHRR over Cyprus
  13
 
Fig. 4. Scatterplot of MODIS derived PW versus AVHRR derived PW values 
calculated using the relationship formulated for the study area. 
Additionally, AVHRR PW estimates, were compared to the MODIS PW independent 
dataset for July 2006. A good agreement was observed between the AVHRR and the 
MODIS derived PW value for 20 independent cases in July 2006 (RMSE = 0.71 cm). 
In Fig.4, MODIS derived PW values versus the AVHRR derived PW values are 
shown in a scatterplot of for the 20 independent cases in July 2006. Moreover, Fig. 5 
shows the scatterplot of the MODIS derived PW values versus the AVHRR PW 
values derived using the relationship formulated for the broader area of Greece 
(Chrysoulakis et al., 2008) for the same 20 independent cases in July 2006. 
Fig. 4. Scatterplot of MODIS erived PW versus AVHRR derived
PW values calculated using the relationship formulated for the study
area.
As mentioned before, additional comparisons were per-
formed to c mpare th results to the previous study
(Chrysoulakis et al., 2008). AVHRR PW was estimated for
the study area for July 2006 using the relationship proposed
by Chrysoulakis et al. (2008). In Fig. 3b the radiosonde e-
rived PW values versus the AVHRR derived PW values are
shown. Comparing Fig. 3a and b, it is observed that the rela-
tionship f rmulated in the present study gave more accurat
results. This confirms the assertion that PW – 1T relation-
ship is site-specific and depends on land cover and emissivity.
Additionally, AVHRR PW estimates, were compared to
the MODIS PW independent dataset for July 2006. A
good agreement was observed between the AVHRR and the
MODIS derived PW value for 20 independent cases in July
2006 (RMSE = 0.71 cm). In Fig. 4, MODIS derived PW
values versus the AVHRR derived PW values are shown in
a scatterplot of for the 20 independent cases in July 2006.
Moreover, Fig. 5 shows the scatterplot of the MODIS de-
rived PW values versus the AVHRR PW values derived us-
ing the relationship formulated for the broader area of Greece
(Chrysoulakis et al., 2008) for the same 20 independent cases
in July 2006.
Finally, in Fig. 6 the spatial distribution of AVHRR derived
PW is shown. Mean PW values were estimated by averag-
ing the daily PW values for the whole month (July 2006),
for cloud free cells. The effect of topography is obvious.
Lower PW values noted at Tilliria, Marathasa and Troodos
mountains, (mean PWmin = 1.61 cm), while higher PW val-
ues can be observed in north-eastern area of the island (mean
PWmax= 3.07 cm).   14
 
Fig. 5. Scatterplot of MODIS derived PW versus AVHRR derived PW values 
calculated using the relationship formulated for the broader area of Greece 
(Chrysoulakis et al., 2008). 
Finally, in Fig. 6 the spatial distribution of AVHRR derived PW is shown. Mean PW 
v lues were estimated by averaging the daily PW values for the whole month (July 
2006), for cloud free cells. The effect of topography is obvious. Lower PW values 
not d at Tilliria, Marathasa and Troodos mountains, (mean PWmin=1.61 cm), while 
higher PW values can be observed in north-eastern area of the island (mean 
PWmax=3.07 cm).  
 
Fig. 5. Scatterplot of MODIS derived PW versus AVHRR de-
rived PW values calculated using the relationship formulated for
the broader area of Greece (Chrysoulakis et al., 2008).
  14
 
Fig. 5. Scatterplot of MODIS derived PW versus AVHRR derived PW values 
calculated using the relationship formulated for the broader area of Greece 
(Chrysoulakis et al., 2008). 
Finally, in Fig. 6 the spatial distribution of AVHRR derived PW is shown. Mean PW 
values were estimated by averaging the daily PW values for the whole month (July 
2006), for cloud free cells. The effect of topography is obvious. Lower PW values 
noted at Tilliria, Marathasa and Troodos mountains, (mean PWmin=1.61 cm), while 
higher PW values can be observed in north-eastern area of the island (mean 
PWmax=3.07 cm).  
 
Fig. 6. The spatial distribution of AVHHR derived PW values esti-
mated at each grid cell by averaging the daily values for July 2006.
5 Conclusions
The relationship between PW and 1T was calibrated by sim-
ple regression using the MODIS derived PW and validated
using radiosonde measurements and independent MODIS
derived PW values. The application area was the island
of Cyprus and all spatial calculations were performed for a
5× 5 km cell grid. The PW – 1T relationship was calibrated
using cloud free cells from 16 cases in July 2006 (Table 1).
This equation was applied to daily 1T data for July 2006.
The resulted PW values were validated using radiosonde
measurements and independent MODIS derived PW values.
A good agreement between radiosonde and AVHRR derived
PW values was observed, as well as a good agreement be-
tween MODIS derived PW values and AVHRR derived PW
values (RMSE = 0.76 cm). The time lag between MODIS
and AVHRR observations was low enough to consider the
satellite overpasses concurrent, however the above RMSE
Adv. Geosci., 30, 23–29, 2011 www.adv-geosci.net/30/23/2011/
D. Hadjimitsis et al.: Precipitable water using MODIS and AVHRR over Cyprus 29
value would be smaller if this time lag was lower. This
RMSE value can be therefore improved if the above time dif-
ference is compensated, using PW rates of change, according
to Chrysoulakis et al. (2008). A correspondence of the spa-
tial patterns of MODIS derived PW and AVHRR derived 1T
was observed. The effect of topography was evident since
lower PW and 1T values were observed over the mountain-
ous area of Troodos (mean PWmin= 1.61 cm) and higher PW
values over the area of Famagusta (mean PWmax= 3.07 cm).
Acknowledgements. The authors would like to express their appre-
ciation to the Cyprus University of Technology (www.cut.ac.cy)
for the internal funding and the Remote Sensing Laboratory of the
Department of Civil Engineering and Geomatics for its support.
Edited by: S. Athanasatos
Reviewed by: two anonymous referees
References
Cartalis, C. and Chrysoulakis, N.: Estimation of Precipitable water
in Greece on the basis of radiosondes and satellite data, Toxicol.
Environ. Chem., 58, 163–171, 1997.
Choudhury, B. J. and Di Girolamo, N. E.: Quantifying the Ef-
fect of Emissivity on the Relations between AVHRR Split Win-
dow Temperature Difference and Atmospheric Precipitable Wa-
ter over Land Surfaces, Remote Sens. Environ., 54, 313–323,
1995.
Choudhury, B. J., Dorman, T. J., and Hsu, A. Y.: Modeled and
Observed Relations between the AVHRR Split Window Temper-
ature Difference and Atmospheric Precipitable Water over Land
Surfaces, Remote Sens. Environ., 51, 281–290, 1995.
Chrysoulakis, N., Kamarianakis Y., Xu, L., Mitraka Z., and Ding, J.:
Combined use of MODIS, AVHRR and radiosonde data for the
estimation of spatiotemporal distribution of precipitable water, J.
Geophys. Res., 113, D05101, doi:10.1029/2007JD009255, 2008.
Chrysoulakis, N. and Cartalis, C.: A new algorithm for the detec-
tion of plumes caused by industrial accidents, on the basis of
NOAA/AVHRR imagery, Int. J. Remote Sens., 24, 3353–3367,
2003.
Czajkowski K. P., Goward S. N., Shirey, D., and Walz, A.: Ther-
mal remote sensing of near-surface water vapor, Remote Sens.
Environ., 79, 253–263, 2002.
Dalu, G., Satellite remote sensing of atmospheric water vapour, Int.
J. Remote Sens., 7, 1089–1097, 1986.
Eck, T. F. and Holben, B. N.: AVHRR split window temperature
differences and total precipitable water over land surfaces, Int. J.
Remote Sens., 15, 567–582, 1994.
Gao, B.-C. and Kaufman, Y. J.: Water vapor retrievals us-
ing Moderate Resolution Imaging Spectroradiometer (MODIS)
near-infrared channels, J. Geophys. Res., 108(D13), 4389,
doi:10.1029/2002JD003023, 2003.
Goward, S. N., Waring, R. H., Dye, D. G., and J. Yang, Ecologi-
cal remote sensing at OTTER: macroscale satellite observations,
Ecol. Appl., 4, 322 - 343, 1994.
Jedlovec, G. J.: Precipitable water estimation from high-resolution
split window radiance measurement, J. Appl. Meteorol., 29, 863–
877, 1990.
King, M. D., Menzel, W. P., Kaufman, Y. J., Tanre´, D., Gao, B.-
C., Platnick, S., Ackerman, S. A., Remer, L. A., Pincus, R., and
Hubanks, P. A.: Cloud and Aerosol Properties, Precipitable Wa-
ter, and Profiles of Temperature and Water Vapor from MODIS,
IEEE Trans. Geosci. Remote Sens., 41, 442–458, 2003.
www.adv-geosci.net/30/23/2011/ Adv. Geosci., 30, 23–29, 2011