It is well-known that TMI rainfall rate is 24% larger than PR rainfall
rate in the zonal average over tropical region at the algorithm version
5 (Kummerow et al. 2000). We compared TMI and PR rainfall rates
over the ocean, land and coast, in order to identify the origins of
this difference. TMI rainfall rate is larger than PR rainfall rate over
tropic or midlatitude summer ocean. On the contrary, PR rainfall rate
is larger than PR over the midlatitude winter ocean. Over the
midlatitude winter land, PR rainfall rate is twice larger than TMI. We
investigated the vertical profile of TMI and PR rainfall rate, TMI
freezing height, PR bright band height, PR storm height, TMI vertical
and horizontal brightness temperature.
In this web page, we show the results over tha ocean written in a paper
of Ikai & Nakamura (2003).
In this analysis, the calibrated brightness temperature of TMI
(TMI-1B11), rainfall rate of TMI (TMI-2A12), rainfall rate of PR
(PR-2A25), path integrated attenuation (PIA) of PR (PR-2A25) and
gridded statistical three months averaged rainfall rate of PR
(PR-3A25). Toolkit is version 5. We produced global data sets by
averaging in 0.5 degrees (longitude) by 0.5 degrees (latitude) grid
boxes. Three months from December 1998 to February 1999 (hereafter
called DJF1999) and from June 1999 to August 1999 (hereafter called
JJA1999) are used to see the seasonal change of the characteristics.
We found three plausible reasons
for these differences. One is a problem in the freezing level assumption in the
TMI algorithm for midlatitude region in the winter. That results in underestim
ation of TMI-derived rain rates. Second is inadequate Z-R or k-R relationships f
or convective and stratiform types in the PR algorithm. Third is wrong interpret
ation of rain layer when freezing level is low and the rain type is convective.
The strong bright band echo seems to be interpreted as rain and too strong rain
attenuation correction is applied. This results in too strong rain rate by PR al
gorithm.
Results
1. Comparison of TMI with PR
Global comparison of rain rates at surface level
we compared the three-month averaged TMI-rain with PR-rain on a global
scale. Figure 1 shows global maps of the ratio
of PR-rain to TMI-rain averaged for DJF1999 and JJA1999. In this
figure, the ratios were plotted except for the grids where either the
three-month averaged TMI-rain or PR-rain was below 0.05 mm/h. This is
because TMI rain rate can easily become zero due to the TMI sensitivity
while the minimum detectable rain rate of PR is about 0.5 mm/h. Though
the very weak rain is generally negligible for total rain and ratios,
elimination of very weak rain might have significant effect for weak
rain regions. Some characteristics which could be noticed from Fig. 1
are as follows:
a) Generally TMI-rain is greater than PR-rain.
b) PR-rain is greater than TMI-rain in the north Atlantic and western part
of the Pacific Ocean north of 30 deg N for DJF1999, and also PR-rain is
also greater than TMI-rain in western part of Pacific Ocean south of
30 deg S for JJA1999.
c) TMI-rain is significantly greater than PR-rain
in western part of the north Pacific Ocean for JJA1999, and south of
30 deg S for DJF1999 in the Pacific Ocean. TMI-rain is also significantly
greater than PR-rain in tropical regions of central to the eastern
Pacific Ocean for both season,.
d) PR-rain is greater than TMI-rain in
mountainous regions over land, for example , Rocky, Andes, and
Himalayan Mountain Ranges.
2. Case Study
To investigate these characteristics, we selected five typical regions
where either of two characteristics appears clearly. We additionally
selected western part of tropical Pacific Ocean as a reference region
where TMI-rain was generally greater than PR-rain, and these regions
are shown in a lower panel of Fig. 1. We show
results of events over ocean, east of Japan in Feb. 11th on 1999 (MWNP) and
over ocean, east of Philippine in June 13th on 1999 (TWP). In those cases
rainfall area extended horizontally over 40 km in along track
direction, and a heavy convective rainfall area existed. Those case s
were selected for the reason that the wide raining area existed in the
narrow PR swath. Figure 2-1 shows scatterplots
of TMI-rain versus PR-rain.
In case of TWP, TMI-rain is greater than PR-rain by about 100%, and the
slope is correspondingly 0.49. In contrast to case of winter MWNP,
PR-rain is greater than TMI-rain by about 63%, and the slope is
1.64. This characteristic is similar to those in corresponding regions
mentioned in the previous section on the global characteristics. We can
also notice that the tendency of the overestimation of PR-rain is
remarkable in grids where the convective rainfall defined by the
PR-2A25 algorithm spreads widely, which are plotted by open circles in
this figure.
This indicates the problems are due to a relations between radar
refrectivity factor and the rainfall rate, that is, Z-R and/or
a relations between radar refrectivity factor and
attenuation coefficient of rainfall, that is, k-Z, and/or a method for
a correction of an attenuation for the precipitation, etc.
We compared the brightness temperature directly measured in TMI-10GHz
vertical polarization channel (TMI-10GHzV-BT) with path integrated
attenuation (PIA) derived from PR. The frequency of TMI-10GHzV is close
to that of PR (13.8 GHz). The increase of TMI-10GHz-BT due to rain
emission is nearly proportional to the path attenuation. The linearity
also results in a small beam filling effect which is caused by the
non-linear relationship between excess brightness temperature and rain
rate. Because linear PIA (absolute value instead of value
in dB) also relates nearly linearly to column total rain amount along
the radar beam path, we converted PIA in dB to linear value for
comparison. Figure 2-2 shows the scatterplots
of TMI-10GHzV-BT versus linear PIA in the same cases in Fig. 2-1. In both cases,
TMI-10GHzV-BT is strongly related with linear PIA in contrast to the
rain rate case. This strongly suggests that generally TMI and PR both
measure or estimate the path attenuation correctly regardless of
regions and seasons. Though the surface condition affects the
brightness temperature, we assumed that it is not crucial for case
studies when we are interested only in excess temperatures.
Investigating more details, the minimum of linear PIA of about 0.55 in
the case of winter MWNP is smaller than in the case of TWP (about 0.65),
although excess of TMI-10GHz-BT are almost
the same.
This indicates the differences of rainfall rates between TMI and PR is
caused by their differences on the vertical profiles of the precipitation.
3. Comparison of Vertical rain profiles (effects of TMI 0℃ height)
The fact that the relations between TMI-10GHzV-BT and PIA do not show
significant differences suggests that the path-integrated liquid water
content from TMI and path-integrated rain should generally be correct
and that differences in rain retrieval algorithms in TMI and PR yield
the disagreement of rain rates at the surface level. PR can observe
directly the vertical rain profile and detect the bright band height
(PR-BBH), which is an indicator of freezing height. However, TMI
measures only the brightness temperature related with vertical
integrated attenuation or water content. Therefore, in estimating the
rain rate at surface from brightness temperatures of TMI, the vertical
rain profile and freezing level are very important. TMI 2A12 algorithm
estimates the freezing height (TMI-FH) by using a lookup table of the
brightness temperatures of 19 GHz and 21 GHz and retrieves the vertical
profiles of liquid water content etc. Figure 3
shows monthly averaged rain profiles from TMI (TMI-profile) and PR (PR-
profile), and the profiles of ratio of TMI-profile to PR-profile
(TMI/PR-profile) for the regions shown in Figure 1 except for TCP. These
profiles were averaged for January 1999 and July 1999 including
no-rain pixels. In these figures, the rain rate at the lowest layer
(500 m) of PR-profile almost always corresponds to the near surface
rain rate. In the regions of tropical and midlatitude summer except for
MCNP where rain amount is very small, TMI-profiles have similar
characteristics. There, each freezing level (TMI-FH) is expected
between 4 km and 5 km in height and rain rate is decreasing downward
slightly below the weak peak at 2 km, although the intensity of rain
rate is different. In these regions, PR-profiles are almost uniform
below the bright band height (PR-BBH). PR-BBH seems to be between about
4 km and 5 km in height, and is roughly the same as TMI-FH in each
region. For these four regions, TMI- and PR-profiles are approximately
uniform below PR-BBH, while the ratios are different. This indicates
that the shapes of TMI-profiles are similar to those of PR-profiles.
In the four midlatitude regions in winter (winter MNA, MWNP, MCNP, and
MWSP), TMI -profiles increase rapidly downward from 4.5 km to 2 km and
decrease towards surface below 2 km, while PR-profiles gradually
increase downward from 4 km to 1.5 km or 2 km. Therefore, for these
four regions very large overestimation of TMI (the ratio is at most
about 4) between 3 km and 4 km appears. The ratio decreases towards
surface rapidly, and represents only a weak overestimation of PR or TMI
(the ratio is 0.8 to 1.5) at the surface. These characteristics mean
that TMI-2A12 al gorithm distributes the vertically integrated water
content into a tall column and the surface water content becomes
less. In other words, it is suggested that TMI-2A12 algorithm uses an
incorrect estimate of freezing level which is too high for midlatitude
regions in winter, and it leads to the underestimation of TMI-rain .
We assessed the estimation of freezing level by TMI (TMI-FH) in
comparison with the bright band height from PR by three-month averaged
global data. We produced averaged TMI-FH from TMI-1B11 product, and
compared TMI-FH with PR-BBH. TMI-FH is higher than PR-BBH by about 300
m to 500 m when TMI-FH or PR-BBH is high. The freezing height estimated
from TMI is reasonable in tropical regions. However, when TMI-FH or
PR-BBH is low, however, this three-month averaged TMI-FH is higher by
about 1500 m to 2000 m than the three-month averaged PR-BBH. This
difference is not reasonable. It suggests that TMI algorithm estimates
the freezing level too high in the midlatitude winter hemisphere. This
disagreement has also been shown in comparisons of freezing level
derived from SSM/I. Figure 4 shows global maps
of the ratio of TMI-FH to PR-BBH averaged for DJF1999 and JJA1999. The
ratio is small (less than 1.2) in tropical and midlatitude summer
regions, but it is great (over 1.2) in midlatitude winter
regions. TMI-FH/PR-BBH has large values for the regions where PR-rain
is greater than TMI-rain as shown in Fig. 1. Next, we compared the
ratio of the three-month averaged PR-rain to TMI-rain
(PR-rain/TMI-rain) with the ratio of the three-month averaged TMI-FH to
PR-BBH (TMI-FH/PR-BBH) for DJF1999 and JJA1999. We show the average and
standard deviation of PR-rain/TMI-rain against TMI-FH/PR-BBH of every
0.05 in Fig. 10. Though the standard deviation is large, these figures
indicate that the larger TMI-FH/PR-BBH is, the larger PR-rain/TMI-rain
is. For example, if TMI-FH is twice the PR-BBH, the surface TMI-rain is
underestimated approximately by half compared with PR-rain
(PR-rain/TMI-rain is about 0.7 to 1.5) by linear regression
method. This suggests that the inadequate estimation of rain height for
TMI causes the TMI-rain underestimate. In general, the TMI-rain is
smaller than PR-rain in midlatitude regions in winter can be explained
by the inadequate freezing level in TMI algorithm.
However, three-month averaged PR-rain is not always greater than
TMI-rain in midlatitude winter regions. For example, in winter MCNP and
over the midlatitude central south Pacific Ocean TMI-rain is greater
than PR-rain. Also in summer MWNP and MWSP and TCP, the overestimation
of TMI-rain was greater than other summer and tropical regions. To
explain this zonal variation, we considered the Z-R relationship, rain
type and attenuation correction in the following sections.
4．Effects of the rain types defined by PR algorithm
(Z-R and k-Z relationships)
PR-rain is derived by the Z-R relationship after attenuation correction depending on the k-Z relationship, while TMI-rain is translated from water content of each layer (R-W relation). The Z-R, k-Z and R-W relationships depend on the drop size distribution (DSD) and terminal velocity of rainfall drops. In PR-2A25 algorithm Z-R and k-Z relationships are changed according to the rain types (convective, stratiform, and others) defined by the vertical and horizontal rainfall structure. Convective type rain has a greater rainrate than stratiform rain by about 1.3 when both have the same radar reflectivity. Furthermore, the difference of rain rate at surface level between the convective and stratiform rain types may be emphasized through attenuation correction using the different k-Z relationship according to rain types. In addition to this, PR-2A25 algorithm uses different profiles of the Z-R relationships according to rain types. The fitting of the k-Z profile and real microphysical one are very important to calculate at tenuation. Therefore, we investigated the relationship between rain type and the PR-rain/TMI-rain. Figure 5-1 and figure 5-2 are the scatterplots showing ratio of convective rainfall amount of PR to total (PR-Ratio-conv) versus PR-rain/TMI-rain in oceanic regions between 5°N and 10°N for DJF1999 and JJA1999 and for the regions between 30°N and 35°N for DJF1999, and between 30°S and 35°S for JJA1999, respectively. We used only grids where both TMI-rain and PR-rain were over 0.1 mm/h. Figure 5-3 and figure 5-4 show the mean and standard deviations of Figs. 5-1 and 5-2, respectively. In the midlatitude winter, the bright band height has a lot of variation and it seems that the characteristics are different for high and low bright band cases. Thus, in midlatitude regions we separated the cases in which bright band height is below 2400 m or above 2400 m. Assuming that the temperature decreases at 6 K/km, the criterion corresponds the sea surface of 15 degrees Celsius. If the Z-R relationship has no difference between convective and stratiform rain, PR algorithm which uses different Z-R relationships for rain types would give biased rain rates. In Figs. 5-1 and 5-2, the solid lines show regression lines. In Figs. 5-3 and 5-4, the solid lines have inclinations of 1.3. if Z-R relationships are not different for convective and stratiform rain types. The these regions in Figs. 5-1, 5-2, 5-3 and 5-4 show that the greater the three-month averaged PR-Ratio-conv is, the larger the three-month averaged PR-rain estimates in comparison with the TMI-rain. In zonal regions of 30°N-35°N and 30°S-35°S when PR-BBH is high, the regression line and the solid line nearly coincide. This coincide may suggest that effect of the difference between convective and stratiform rain type in the PR-2A25 algorithm yield s this disagreement. In the zonal region of 30°N-35°N and 30°S-35°S with high PR-BBHs, and 5°N-10°N, the small difference of slope between regression line and a line with inclination of 1.3 may be due to the effect of attenuation correction of only k-Z relationship. But we can not conclude whether the disagreement is originated by PR or TMI. On the other hand, in the zonal region of 30°N-35°N and 30°S-35°S when PR-BBH is low, the difference of slope between regression line and a line with inclination of 1.3 is more significant. This difference seems not to be explained from the effect of the k-Z relationships on rain types. This difference may be caused by inaccuracy of attenuation correction when PR-BBH is low.
Conclusions
Over the ocean, the general overestimation of TMI-rain was confirmed.
PR-rain is greater than TMI-rain in part of the midlatitude winter regions.
The discrepancy between TMI-rain and PR-rain also showed regional variation.
The systematic differences in the three-month averaged PR-rain and
TMI-rain seem to appear as a combination of the following causes: (1)
overestimation of TMI-FH, (2) inadequate Z-R and k-Z relationships for
convective and stratiform rain types, (3) inadequate estimation of
hydrometeor profiles for convective rain cases. The regions where
PR-rain is systematically greater than TMI-rain are over the
midlatitude north Atlantic Ocean, the midlatitude western north Pacific
Ocean, and the midlatitude western south Pacific Ocean in midlatitude
winter. In these regions, both TMI-FH/PR-BBH and PR-Ratio-conv were
high and PR-BBH was low. The regions over the midlatitude central
north Pacific Ocean and the central south Pacific Ocean had the
characteristics of the overestimation of TMI-rain, even though they
were located in midlatitude winter hemisphere. Above two regions
certainly have the tendency of high TMI-FH/PR-BBH and low
PR-Ratio-conv. This fact suggests that the effect of overestimation of
TMI-FH was compensated by the effect of small
PR-Ratio-conv. Therefore, those two regions have only small TMI-rain
overestimation. For the tropical region and midlatitude region in
summer hemisphere, the regions where TMI-rain is much greater than
PR-rain located in summer midlatitude western north Pacific and central
south Pacific Ocean and the tropical central
Pacific Ocean. In these three regions, TMI-FH/PR-BBH did not
differ much from TMI-FH/PR-BBH, but PR-Ratio-conv was much smaller than
other tropical regions and summer regions. This can be understood by
the effect of PR-Ratio-conv. We studied the general characteristics
of rain rate derived from TMI-2A12 and PR- 2A25 algorithms. This kind
of comparison is very valuable for the development of algorithms for
microwave instruments, precipitation radar, and combination of the both
onboard future satellites. For example, the microwave imager and the
precipitation radar could be the principle sensors in the future
Global Precipitation Measuring (GPM) mission. In this study we
restricted to oceanic region. Now, this kind of exercise has been
performed over land where microwave radiometers do not detect radiation
from liquid water directly.
This page is cited from one part of Ikai & Nakamura (2003). Please
read this paper if you want to know in detail. Please see Masunaga et al.
(2002), Furuzawa(Akimoto) & Nakamura (2005) etc. as the similar paper
as this Ikai & Nakamura (2003),
References
Ikai, J. and K. Nakamura, J. Atmos. Oceanic Technol., 20(12), 1709-1726, 2003.
Masunaga, H. et al., J. Applied Meteorology, 41, 849-862, 2002.
Kummerow, C. et al., J. Applied Meteorology, 39, 1965-1982, 2000.
Furuzawa, Akimoto F. and K. Nakamura J. Appl. Meteor., 44(3), 367-383, 2005.
