Monitoring real time rainfall over the Arabian Peninsula using
Meteosat Second Generation (MSG).
Mazen Assiri, D.I.F. Grimes
Department of Meteorology, University of Reading, Earley Gate, PO Box 243, Reading, RG6 6BB,
UK
Abstract
Water is essential for life. In the Arabian Peninsula, rainfall is irregular, infrequent and low. Climate studies
show that the Arabian Peninsula receives between less than 50mm per year and more than 250mm per
year depending on location. Therefore rainfall monitoring is very important to optimise use of this scarce
resource. Up till now, all rainfall monitoring has made use of raingauge observations, but the raingauge
distribution is irregular and the number of raingauges is inadequate for reliable results. Consequently,
there is a need for an approach which can deliver improved areal coverage in a timely manner. Recent
studies have shown that rainfall monitoring based entirely on Meteosat imagery provides adequate and
reliable monitoring in semi-arid regions of Africa such as Ethiopia and the West African Sahel and the aim
of this study is to investigate whether a similar approach will work for the Arabian Peninsula.
Two basic approaches have been tried. The first is the TAMSAT method which depends on estimating
rainfall using cold cloud duration based only on Meteosat thermal infra red imagery; the second is the UK
Met Office algorithm which uses four channels from MSG in the visible, infra red and near infra red bands
to produce rainfall estimates. A rigorous geostatistical approach has been used to validate the results of
the estimation against available raingauge data. This work will present the results of the comparison
between the estimates and the areal rainfall averages and discuss the feasibility of a real time operational
system for the Arabian Peninsula.
INTRODUCTION
Full area coverage rainfall monitoring is useful for water management, meteorological hazard warning and
public services. The ground based measurement of rainfall is usually used to monitor precipitation
amount. However, this method has many problems such as irregular distribution, the low number of gauge
in many areas and the high cost of its operation. The use of meteorological satellites to estimate and
monitor rainfall helps solve these problems. The advantages of using satellite observations are the full
spatial coverage, continuous observations and low cost of operation.
In this paper, the use of geostationary satellites is investigated to estimate real time rainfall over the
Arabian Peninsula (AP). The AP is covered by several satellites such as Meteosat-7, Meteosat-9 and
Indian satellite (INSAT). The main aim of this research is to estimate rainfall using Meteosat Second
Generation (MSG).
There are several methods for estimating rainfall using satellite. This research aims to compare two
different estimating methods over the AP. The first method is the TAMSAT (Tropical Application of
Meteorological Satellites) approach (Grimes et al., 1998). The second method is the UK met office
algorithm (Francis et al., 2006).
In this paper, the location and climate of the study area is presented. The use of the geostatistical
application to find out the areal averages of rainfall from raingauge data is given. The explanation of the
estimation methods is shown. Finally, the discussion of the results and the future work is given.
THE ARABIAN PENINSULA (LOCATION AND CLIMATE)
The Arabian Peninsula is located in Southwest Asia and connects Asia with Africa (figure 1). It is located
in the sub-tropical belt between 12°N, 35°N, 30°W and 60°W. The main feature of its climate is arid/semiarid nature (Edgell, 2006). In detail, the climate can be explained in terms of regional air masses.
There are four different types of air mass that influence the weather in the Arabian Peninsula. Firstly, the
polar continental air mass affects the area in the winter time (December – February, occasionally midMarch) and comes from the central of Asia as a result of the domination of the Siberian High. The second
air mass is the polar maritime air mass. It reaches the area as a result of western low depression
movement through winter season. If it is associated with an upper trough, then rainfall is expected over
the North of Saudi, Kuwait and Jordan. The third air mass is the tropical continental air mass which
influnces the AP during spring and early summer. This air mass is hot and dry causing mainly dry
convection. The last air mass is the tropical maritime air mass that comes from North of the Indian Ocean
and Arabian Sea during Summer. It affects mainly the Southwest of Saudi Arabia, Yemen and coastal line
of Oman bringing hot and humid air (Ghazanfar and Fisher, 1998, Abdullah and Al-Mazroui, 1998). Figure
1 (left) shows the source and direction of these air masses.
Figure 1 shows the direction and the source of the effecting air masses over the AP (Ghazanfar and Fisher, 1998). The
distribution of the surface pressure over the AP during winter time is shown in the middle map while the summer time is
shown in the right map (Abdullah and Al-Mazroui, 1998).
The mechanism of rainfall production over the AP is governed by two major factors; the synoptic systems
and the topography of the AP. The influence of the topography on the weather of the AP is considered
because there are three different types; coastline, desert and mountains. The mechanism of producing
rainfall can be described clearly through the interaction of the topography with seasonal pattern. In winter
time, the rainfall is produced from two mechanisms. The First source is the movement of the westerly
upper troughs which are associated with surface depressions. These systems bring fronts to the area. The
second source is the penetration of the Sudan trough which advects warm and humid air. This system is
associated with the Siberian ridge which brings cold and dry air to the area. This mechanism is active over
the western region and southwest highlands of Saudi Arabia (see figure 1, middle). In spring time, the
rainfall still occurs over the whole area. The north of the Arabian Peninsula is under the influence of North
African depressions which produce thunderstorms and rainfall. The thunderstorms reach a peak during
this season. The southwest mountains receive rainfall from convective systems that are the result of the
strong temperature contrast between land and sea, and mountains and valleys. In the interior of Oman,
the Oman convergence zone is active and produces deep convective clouds. In summer, the study area is
affected by thermal lows which combine with the Indian monsoon to become one system. It causes dust
storms over the Northern part of the area and brings warm and humid air to the southern coast of Yemen
and Oman which can produce rainfall in the mountains. Also, southwesterly winds affect the southwest
mountains of Saudi Arabia and Yemen producing orographic convective rainfall (see figure 1, right). In
autumn, the southwest monsoon starts to change to northeast monsoon. Occasionally, the shower events
occur over north of Oman and Emirates due to the passage of cold northeasterlies over the warm water of
the Arabian Gulf (Ghazanfar and Fisher, 1998, Abdullah and Al-Mazroui, 1998).
THE DATA
Raingauge data have been collected from the operators of the raingauge networks that are working in the
AP. Figure 4 (A) shows the distribution of the available raingauges. The satellite data is provided by the
TAMSAT project at the department of Meteorology in University of Reading and by the UK met office. In
this paper, monthly data were used.
GEOSTATISTICAL PROCESSING OF GAUGE DATA
In order to compare point raingauge measurements with pixel based satellite data, it is first necessary to
convert the raingauge data to the same spatial scale as the satellite estimates. This has been done by
block kriging (Webster and Oliver, 2001). Block kriging is usually regarded as the best approach because
of spatial interpolation or because it is based on weighted averages which take account of the variation of
spatial correlation with separation distance (Subyani, 1997, Kitanidis, 1997, Goovaerts, 1997, Chilès and
Delfiner, 1999). The variation of correlation with distance is usually represented by a variogram as
described below.
Variogram
The variogram is defined as
2
γ (h ) = E ⎡( R ( x ) − R ( x + h ) ) ⎤ for all pairs of gauges where E is ⎣
⎦
expectation operator, R is rainfall, x is location and h is separation distance. It is usually described by three parameters sill, range and nugget respectively the semivariance between uncorrelated points, the distance over which rainfall is correlated and the variance at a distance less than the smallest separation distance (see figure 2). In this case it was necessary to divide the AP into different zones and produce a variogram for each month. Examples are shown in figure 3. In calculation of variogram for this work, climatological variogram (Grimes et al., 1999) were calculated by normalizing the values of γ(h) with respect to the variance for each month. In the rest of this paper raingauge data refers to block kriged values at pixel scale. Figure 2 shows the variogram parameters that are displayed in the variogram graph. Sill is the Semivariance value at which
the variogram levels off. Range is the spatial correlation range. Nugget is the variance near the origin.
TAMSAT APPROACH
The TAMSAT approach is the first method applied to estimate rainfall over the AP. This method assumes
rainfall occurrence when the top cloud temperature is below a threshold temperature. Therefore, the
amount of rainfall is related to the Cold Cloud Duration (CCD) linearly. CCD is the length of time for which
a cloud top is below a threshold temperature. The relationship is given by
Rain = a0 + a1CCD
Where a0 and a1 are the estimation parameters. These parameters are determined through climatic
calibration against historical raingauge data.
A)
B)
Monthly rainfall variogram for January over Southwestern region- 20km lag
2.4
2.2
Exponential Model
2.0
SemiVariance
1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0
50
100
150
200
250
300
Distance (km)
C)
D)
M onthly rainfall variogram for August over Southw estern region- 20km lag
Monthly rainfall variogram for January over Middle and Southern region- 20km lag
2.4
2.0
2.2
Exponential model
1.8
Exponential Model
2.0
1.6
1.4
SemiVarianc e
SemiVariance (mm2)
1.8
1.2
1.0
1.6
1.4
1.2
1.0
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0.0
0.0
0
100
200
300
400
0
100
200
300
400
500
600
700
800
900
1000
Distance (km)
Distance (km)
E)
Monthly rainfall variogram for January over Northern region - 20km lag
2.4
2.2
Exponential Model
2.0
SemiVariance
1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0
50
100
150
200
250
300
Distance (km)
Figure 3 gives the division of the AP into zones based on the rainfall seasonality, the analysis of the variogram for each
zone for January and the results of variogram parameters. A) represents the location of zones and the gauge distribution
for each region. Zone-A is the mountainous region. Zone-B is the central and southern region. Zone-C is the northern
region. B) is the variogram analysis for zone-A for January. C) is as B) but for zone-A for August. D) is as B) but for zone-B
for January. E) is as B) but for zone-C for January.
The TAMSAT approach is based on Infrared channel which gives continuous estimation during day and
night. However, the limitations of this method are; 1) it works only with convective systems and 2) it does
not distinguish between cirrus and thick clouds. For rainfall estimates of convective rain and accumulation
of one day and longer, these limitations appear to be relatively unimportant (Grimes et al., 1999).
Operationally, TAMSAT approach is applied over the Sahel and Southern of Africa. The details of applying
this method may be found in Grimes et al (1998).
In brief, the estimation of rainfall using TAMSAT approach should follow few steps. The first step is
defining the agreement level between ground based data and CCD data. For more robust calibration, the
pixels that have at least one gauge are used. The second step is determining the calibration zones which
are assumed to be climatically homogeneous. The third step is the regression of raingauge values against
the CCD values to find out the parameters of the regression. The regression is performed by regressing
median rainfall against mean CCD for specific bins based on the CCD data. This is important to reduce
the effect of the outliers. Then, the estimation is carried out using the calibrated parameters.
UK MET OFFICE ALGORITHM
The UK met office algorithm has been developed to estimate rain rate using several channels from the
MSG satellite data; Visible, Infrared, near Infrared and water vapour (Francis et al., 2006). The algorithm
classifies instantaneous rainfall amount into a number of bins depending on the different combinations of
radiances from the various channels. Different algorithms are used for day and night over Europe. Real
time calibration is carried out by comparison with radar data. To generate the rainfall estimates using this
method, the data of MSG have been calibrated using the rainfall radar data from European networks.
Operationally, the estimates are produced for Europe and the North Atlantic. Francis (2006) has shown
that the algorithm works well over the UK by validating the estimates against the UK rainfall radar data.
RESULTS AND DISCUSSION
Geostatistics results
The seasonality analysis of rainfall over the AP shows that the mountainous region (zone-A) has two main
seasons; winter and summer while the rest of region (zone-B and zone-C) receive rainfall only in winter
season. Then, these zones were analyzed independently (see figure 3, A). The spatial structures of
rainfall for all zones have been defined clearly by variogram analysis (see graph 3, B, C, D and E). B and
C are the variogram for zone-A for January and August. It can be seen that the variation in particular
range is very large.
Zone-A shows that the greatest correlation range is 91km in April which is the wettest month of the year.
The shortest correlation range is 15km in September which is the end of the summer season. The wettest
month in zone-B is March. Its spatial structure shows that the correlation range is 221km with the lowest
nugget effect of 0.23. Variograms were not defined for September for zone-B because of the low rainfall
amount. The spatial structure of rainfall for zone-C has been defined for the period between October and
February. The correlation range of rainfall in December is 111km as the longest distance during the period
(see table 1).
Calibration results
The calibration zones for the rainfall estimation process have been defined through analysis of the
agreement level between the rainfall data and the CCD data. The AP can be divided into four major
zones; central and northern region, western coast line, south-western mountainous region and southern
and eastern region. Also, each zone may be divided into sub-zones based on the results of calibration.
The regression has been done for all calibration zones for all month. A range of threshold temperature
(Tt), -20 °C, -30 °C, -40 °C, -50 °C, -60 °C and -70 °C, has been examined for each zone and month.
It is noticed that the most zones give strong correlation with threshold temperature of -40 °C and -50 °C.
However, there is few zones give good correlation with colder threshold temperature. On the other hand, a
warmer threshold temperature is required to get correct rainfall estimates for few zones. In summer for
some zones, it was not possible to carry out a calibration because of the low rainfall amount.
The calibration also shows that the topography influence is strong specifically in the South-western region
of the AP (Asir highland and Yemen mountains). Thus, this region has been considered as one zone for
all months. The coastal line in the west of the AP is defined as a main zone. Because of the low number of
gauges, the Eastern and Southern part has been defined as one sub-zone with significant variation in its
boundary with time.
Zone-A
Range (km) Nugget
41
0.39
81
0.65
25
0.25
91
0.64
40
0.55
77
0.7
60
0.68
90
0.74
15
0.12
44
0.67
45
0.69
69
0.81
Month
January
February
March
April
May
June
July
August
September
October
November
December
Zone-B
Range (km) Nugget
190
0.2
169
0.24
221
0.23
235
0.25
189
0.2
363
0.69
209
0.43
140
0.13
No Vario.
94
0.11
251
0.22
200
0.44
Zone-C
Range (km)
Nugget
19
0.43
10
0.43
No Vario.
No Vario. No Vario. 92
0.36
129
0.36
No Vario.
No Vario.
186
0.65
12
0.32
111
0.42
Table 1 gives the variogram parameters for all zones in the AP for all months. No vario. means that the variogram was not
calculated due to insufficient data.
For example, in January the rainfall is received in most areas in the AP. The calibration zones have been
defined (see figure 4, left map). The number of rainy events is sufficient to carry out the calibration for all
sub-zones (see table 2). The middle and North-Eastern region (1a) give reliable calibration with Tt of -40
°C. The northern and North-Western region (1b) can be calibrated with -20 °C. The Western coast (Saudi
and Yemen) has shown that the optimum threshold temperature is -40 °C. The Saudi and Yemen
highlands (Southwest) are calibrated with Tt of -30 °C. -40 °C is the optimum threshold temperature for
calibrating the Southern and Eastern region of the AP (Yemen, Oman and Emirates).
Arabian Peninsula Calibration Regions
Arabian Peninsula Calibration Regions
35
35
34
34
33
33
32
32
31
31
30
29
30
29
1b
28
27
27
26
26
25
25
Latitude
Latitude
28
24
1a
23
22
21
2
23
22
21
20
20
19
19
18
17
16
1
24
2
18
4
3
4
17
16
15
15
14
14
13
3
13
12
12
33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Longitude
Longitude
Figure 4 illustrates the location of calibration zones for January (left) and August (right).
In August, the Northern and middle (desert) part of the AP has no calibration based on dry condition, zone
1 (see figure 4, right map). The western coast gives calibration at -40 °C. The Southwest highlands
calibration results show that dividing the region into smaller areas may give better results but it cannot be
done because of low number of stations. Then -30 °C has been chose to be the optimal threshold
temperature. Although the Eastern of Yemen, Oman and Emirates receive low amount of rainfall during
this month except Oman’s highlands, the calibration results seem to be reasonable for the whole zone.
The optimum threshold temperature is -30 °C or warmer (see table 3).
Zone
1a
1b
2
3
4
N
238
80
245
168
161
Nr
216
71
170
124
91
Tt (°C)
-40
-20
-40
-30
-40
a1 (mm h-1)
0.62
0.21
0.39
0.34
0.71
a0 (mm)
5.68
-1.63
2.52
1.79
-1
R2
0.94
0.97
0.81
0.84
0.95
Table 2 shows the results of the calibration for each zone for January. The names of zones are corresponding to the names
in figure 4.N is the number of available data. Nr is the number of rainy pixels. Tt is the threshold temperature. a0 and a1 are
the estimation parameters.
Zone
1
2
3
4
N
350
238
189
147
Nr
4
117
176
54
Tt (°C)
No Cal.
-40 (c)
-30
-30 (w)
a1 (mm h-1)
No Cal.
0.21
0.95
0.12
a0 (mm)
No Cal.
-0.33
-6.11
-0.42
R2
No Cal.
0.94
0.96
0.87
Table 3 illustrates the results of calibration for August. No Cal. means no calibration for the zone. The rest of symbols are
as table 2.
A)
B)
C)
y = 0.804x + 1.8129
R2 = 0.692
0
2
4
6
8
30
28
26
24
22
20
18
16
14
12
10
8
6
4
2
0
10 12 14 16 18 20 22 24 26 28 30
Estimates-mm
D)
y = 0.6616x + 2.1588
R2 = 0.5119
0
2
4
6
8
y = 0.5532x + 1.3379
R2 = 0.697
0
6
8
10 12 14 16 18 20 22 24 26 28 30
Estimates-mm
Med. Rainfall against Med. estimates for -30C based on
estimates bins.
55
y = 0.7821x - 1.2211
R2 = 0.5649
16
y = 1.0507x - 0.2167
R2 = 0.9474
50
45
14
40
Rainfall-mm
Rainfall-mm
4
60
18
y = 0.8555x - 0.3886
R2 = 0.9479
2
F)
Med. Rainfall against Med. estimates for -40C based on
estimates bins.
20
30
28
26
24
22
20
18
16
14
12
10
8
6
4
2
0
30
28
26
24
22
20
18
16
14
12
10
8
6
4
2
0
10 12 14 16 18 20 22 24 26 28 30
Estimates-mm
E)
Med. Rainfall against Med. estimates for -40C based on
estimates bins.
Rainfall-mm
Med. Rainfall against Med. estimates for -40C based on
estimates bins.
Rainfall-mm
30
28
26
24
22
20
18
16
14
12
10
8
6
4
2
0
Med. Rainfall against Med. estimates for -20C based on
estimates bins.
Rainfall-mm
Rainfall-mm
Med. Rainfall against Med. estimates for -40C based on
estimates bins.
12
10
8
35
30
25
20
6
15
4
10
2
5
0
0
2
4
6
8
10 12 14 16 18 20 22 24 26 28 30
Estimates-mm
0
0
2
4
6
8
10
12
Estimates-mm
14
16
18
20
0
5
10
15
20
25 30 35
Estimates-mm
40
45
50
55
60
Figure 5 gives the median rainfall against median estimates for bins based on the estimates values. A, B, C, D are for
January, A is for zone 1a, B is for zone 1b, C is for zone 2 D is for zone 4. Graph E and F are for July validation for zone 2
and 3, respectively.
Validating the estimates
Rainfall estimates have been validated using the raingauge data. The validation has been carried out by
producing cross-validation for each month for each zone. The comparison has been done by dividing the
estimates into bins. Then, it is illustrated by plotting the median of rainfall against the median of estimates
of each bin. The results show that the TAMSAT approach is performing good estimates for some zones.
Figure 5 gives the results of the cross-validation of rainfall estimates. In January, the validation shows that
rainfall estimates for zone 1a, 1b, 2 and 4 are reasonable. The validation of rainfall estimates over
mountainous region (zone- 3) shows that TAMSAT approach is not the appropriate method. In August, the
method gives reasonable rainfall estimates for zone 2 and 3. However, the validation of rainfall estimates
for zone 4 does not show reasonable results.
CONCLUSION
In conclusion, a geostatistics approach has been used to generate ground based data for calibration
purposes. The variogram analysis explained the spatial structure clearly.
The applying of TAMSAT approach has been done. The calibration results show that the AP should
divided into sub-zones. Also, it shows the optimum threshold temperature for each sub-zone. the
validation shows that the TAMSAT approach performs reasonable estimates for some zones.
The future plan is to carry out the validation of TAMSAT estimates and to find out the best estimates of
rainfall for each zone. Also, investigate the feasibility of estimate rainfall using the UK met office algorithm.
Then, compare between these two methods to get the opportunity of estimating rainfall over the AP
operationally.
ACKNOWLEDGEMENT
The authors are grateful to the presidency of Meteorology and Environment in Saudi Arabia, Met office in
Oman and Met office in Yemen provided rain gauge data. The UK met office provided the data of Met
office algorithm estimates.
RERERENCE
M. A. Abdullah and M. A. Al‐Mazroui (1998).Climatological study of the southwestern region of Saudi Arabia. I. Rainfall analysis.Climate Research, 9, 213 ‐ 223. J. Chilès and P. Delfiner (1999). Geostatistics Modeling Spatial Uncertainty. John Wiley & Sons, Ltd. H. S. Edgell (2006). Arabian Deserts: Nature, Origin, and Evaluation. Springer. P. N. Francis, D. Capacci and R. W. Saunders (2006) In the 2006 EUMETSAT Meteorological Satellite Conference. S. A. Ghazanfar and M. Fisher (1998). Vegetation of the Arabian Peninsula. Kluwer Academic Publishers. P. Goovaerts (1997). Geostatistics for natural resources evaluation. Oxford University Press, Inc. D. I. F. Grimes, R. Bonifacio and H. R. L. Loftie (1998). Rainfall estimation workbook. Chatham, UK: Natural Resources Institute. D. I. F. Grimes, E. Pardo‐Iguzquiza and R. Bonifacio (1999).Optimal areal rainfall estimation using raingauges and satellite data.Journal of Hydrology, 222, 93 ‐ 108. P. K. Kitanidis (1997). Intorduction to Geostatistics: Applications in Hydrology. The Press Syndicate of the University of Cambridge. A. M. Subyani (1997) In Department of Earth Resources, Vol. PhD Colorado State University, Fort Collins, Colorado, pp. 182. R. Webster and M. A. Oliver (2001). Geostatistical for Enviromental Scientists. John Wiley & Sons, Ltd.