June 18, 2009

Updated NAME/TRMM Comparisons

I have updated key NAME/TRMM comparison plots. Cutoff Z is now 18 dBZ for S-Pol, S-Pol Z is adjusted to Ku-modeled values, and rain rates are calculated using 0.04-deg Z (non-Ku-adjusted), ZDR, and KDP data.

Rain PDF
Rain CDF

Rainfall values aren't too much different than before, except that annoying kink is back in S-Pol data at nearly all elevations, due to the switchover between methods in the CSU blended algorithm. As before agreement improves as we move up in elevation, but agreement is much better than before at the highest terrain band - almost dead on.


The agreement at low-altitudes is gone, except for the highest terrain band. Similar to rainfall, going up in elevation improves the agreement. Note that the low-Z tail for S-Pol is now gone, due to the change in rain rate calculation.

Z vs. Rain

Match-up here is good for the Gulf and 1500m+, but is bad elsewhere. Except now it is bad in a different way than before. For a given rain rate, S-Pol has a higher Z than TRMM for the 0-1500m bands. But recall that S-Pol has a higher dBZ than TRMM in these bands, so if we adjusted downward for that the agreement would be much better. Meanwhile, doing that would wreck the nice agreement over the Gulf. The 1500m+ band looks the best of all. TRMM matches up with both rainfall and reflectivity, and so plotting against the two maintains that agreement.

Given the results of Wang and Wolff (2009) and Lang et al. (2009), which suggest that the TRMM PR attenuation correction does pretty well (even in NAME), my current working hypothesis is that these differences are best explained by 2004 being an anomalous year in terms of precipitation. In particular, storms were just more intense on average during 2004 compared to the 10-yr TRMM climatology. A secondary possibility is that there are issues with the TRMM DSDs.

Posted by Timothy at 02:58 PM

June 01, 2009


Here are updated NAME S-Pol/TRMM CFADs. I now require Z ge 17 dBZ to count S-Pol data, to better match TRMM. The major changes occur aloft, where such a requirement would be expected to have the most effect. Now the shapes of the two radars' CFADs match much better, although S-Pol has the wider distribution aloft (i.e., more high-dBZ outliers), and so its mean profiles actually increase their differences with TRMM there. So S-Pol is greater than TRMM at most altitudes other than near-surface (where the two radars match well).


Posted by Timothy at 01:34 PM

April 14, 2009

TRMM vs. S-Pol Plots

Here are plots to be added to the EGU presentation this month, featuring TRMM vs. S-Pol.

TRMM vs. S-Pol CFAD RR > 10 mm/h
TRMM vs. S-Pol dBZ PDF RR > 10 mm/h
TRMM vs. S-Pol dBZ CDF RR > 10 mm/h
TRMM vs. S-Pol Mean Z vs. RR

The first three plots are updated versions of older plots. In the CFADs, in the core high-percentage region (5%+) TRMM matches up with S-Pol well, at least up to 10 km MSL. It is in the tails of the distributions, the 0.1-2.5% range, that the two radars diverge, and this causes S-Pol's mean profile to be more intense. But still, fairly good matchup near the surface in all regions.

In the near-surface dBZ PDFs/CDFs, at a given % we see that TRMM is well within 1-2 dBZ of S-Pol. The worst agreement occurs over the Gulf of California, where TRMM is consistently below S-Pol, but still things are remarkably good here given the vastly different spatial and temporal sampling between the two radars. Note in the PDFs how TRMM is clipped around 34-35 dBZ. Because TRMM is beholden to a Z-R relationship, any RR threshold (10 mm/h here) necessarily leads to a Z threshold. S-Pol can bypass this problem due to its ability to invoke polarimetric measurands to estimate rainfall.

The new plot is mean Z in each RR bin (every 2 mm/h). The relationships are at best well behaved out to only 60-80 mm/h, where sampling issues come into play. Over the Gulf, S-Pol and TRMM are in remarkable agreement, although recall here that TRMM had slightly lower dBZs than S-Pol, so that may be helping the agreement slightly. But still, it appears that TRMM may be getting the right DSD over the water. However, over land things are much different. At low rain rates, S-Pol tends to have a higher reflectivity. This, again, may be due to polarimetric effects, where S-Pol can separate out ice contamination and estimate RR more directly. That allows S-Pol to see low RR in relatively high Zs. Only a few pixels would need to be this way to skew the average up, due to Z being averaged in linear (not logarithmic) space.

But at high rain rates, TRMM has consistently higher Zs over the land. This means that S-Pol can get more rain from less reflectivity via polarimetric techniques - in essence, capture the effect of lots of small drops, while TRMM can only use a Z-R. And if the Z-R's DSD assumptions are off, then the RRs will be off. This problem persists in roughly the same form regardless of elevation band.

Could somehow the different attenuation-correction techniques by TRMM PR over land vs. ocean cause the DSD to be mis-estimated over land, but not over the ocean, even though TRMM's near-surf reflectivities seem to be OK? What could be the influence of the different vertical profiles in this scenario?

Posted by Timothy at 12:22 PM

April 06, 2009


Here are CFADs of S-Pol reflectivity for several different categories.

All raining profiles
All profiles where RR > 20 mm/h
All profiles where RR > 80 mm/h

These are basically followup plots to the mean profiles presented earlier, giving a full perspective of the reflectivity profile statistics. Compared to TRMM PR, what CFAD differences that exist are relatively subtle and need to be examined more carefully. Even the extreme profiles (e.g., top 0.1% profiles) are actually captured pretty well by TRMM aloft, despite the fact that S-Pol's mean profile aloft is more intense than TRMM.

It would be nice to have the TRMM and S-Pol CFADs overplotted together, to examine potential differences in more detail.

Posted by Timothy at 11:34 AM

April 03, 2009

Rainfall CDFs

Here are NAME-region rain "volume" CDFs for S-Pol alone and then TRMM vs. S-Pol. The way these are computed is by multiplying a particular rain rate by the number of observations of that rain rate, then computing PDFs and CDFs as normal by dividing by the total rain "volume" of all relevant grid points (i.e., within a particular terrain band). Technically, for volume one should also multiply by the time frame of the observation (usually a constant, like 15 min for S-Pol), but this is not straightforward to do for TRMM PR so I neglected that step. Hence, "volume" and not volume.

S-Pol only
S-Pol vs. TRMM

The first plot shows that the 80-mm/h soft cap on TRMM rain rates causes it to potentially only miss 0.5-3.5% of the total rain "volume," depending on terrain, but as the second plot shows this is not so simple.

While TRMM and S-Pol have nearly the same mean conditional rain rates for the Gulf (~7.5 mm/h) and 0-500m (~8.5 mm/h), this is clearly coincidental as the distributions for these terrain bands are fundamentally different. For example, over the Gulf, S-Pol has over 50% of the rain "volume" within rain rates above 20 mm/h, while TRMM only has about 25%. The 1500m+ CDFs match the best, but here S-Pol's mean rain rate is nearly half of TRMM's (5.2 mm/h vs. 9.4 mm/h).

Posted by Timothy at 04:06 PM

April 02, 2009

NAME S-Pol Rain Rates and Vertical Profiles

Here are some plots of NAME region rain rates and vertical profiles from S-Pol, compared with rain rates from TRMM PR.

Rain rate PDFs
S-Pol at 0.02-deg lat/lon resolution, against TRMM
S-Pol at 0.04-deg lat/lon resolution, against TRMM

Reducing S-Pol's resolution causes TRMM to exceed S-Pol at the highest elevations, probably due to resolution issues (this terrain band starts more than 50 km away from S-Pol. But the high S-Pol rain rates at low elevations continue to be unmatched by TRMM.

S-Pol vertical profiles
All rain rates greater than 20 mm/h
All rain rates below 20 mm/h
All rain rates above 80 mm/h

Grid columns with greater than 20 mm/h rain rates have 6x the ice and water mass of lesser rain rate columns, while 80+ mm/h columns have 2-3x more ice and water mass than the 20+ mm/h columns. From the rain rate PDFs, it is clear that TRMM PR doesn't see much above 80 mm/h. Perhaps this is due to all the ice mass in these columns, messing up the attenuation correction?

Rain rates above 20 mm/h, by terrain band
Rain rates above 80 mm/h, by terrain band

In the heaviest rainers (80+ mm/h), it appears that ice microphysics plays a bigger role at the higher elevations than at the lower (esp. over the Gulf), which is unsurprising. However, there are far fewer grid columns with these sorts of rain rates at high elevations, so these results need to be examined with caution.

Posted by Timothy at 10:29 AM

March 17, 2009

TRMM vs. S-Pol in NAME

Here are some frequency plots for S-Pol and TRMM Precipitation Radar rain parameters in the NAME domain. S-Pol is from summer 2004 and TRMM is summer 1998-2007.

TRMM Rain Rate
S-Pol Rain Rate
TRMM a in Z=aR^b (b almost exactly 1.5)
S-Pol a in Z=aR^1.5

In rainfall frequency, it appears that in extreme rain rates (30+ mm/h) TRMM has a different distribution than S-Pol, with S-Pol more likely to see extreme rain events over the coastal plain and Gulf of California, while TRMM sees these events over the Sierra Madre Occidental. At lower rain rates, while absolute percentages differ, relative positioning of the different terrain bands roughly match between S-Pol and TRMM (e.g., rain rates of a given value more frequent over coastal plain and Gulf).

The a parameter in Z=aR^b is more difficult to interpret. The clearest difference is that higher values of a are more frequent at 1500+ m MSL via S-Pol than via TRMM. The other obvious point is that TRMM seems to agree well with S-Pol on the relative positioning of the Gulf a distribution.

Posted by Timothy at 04:28 PM

April 24, 2008

AMS Tropical Conference Talk

Here is a NAME-related talk I will give at next week's AMS conference on tropical meteorology.

Tropical conference presentation

Posted by Timothy at 09:03 AM

November 12, 2007

Latest NAME Posters

These posters review much of my most recent NAME research.

AMS 2007 Radar Conference
Taiwan ICMCS-VI Conference

CSU Radarmet's NAME research page

Posted by Timothy at 03:47 PM

November 16, 2004

Kdp in the 05 August 2004 NAME Storm

I am currently testing out Kdp calculation and thresholding with the NAME S-Pol radar data. Here is a nice example from the 05 August 2004 mesoscale convective system.

Here we see large Kdp values aloft, below the melting level - over 5.5 deg/km in fact! This is likely due to the presence of melting hail, and represents either the wet hailstones or shed water drops. Note the negative Zdr and high LDR values above the Kdp maximum, indicating large hail above the melting level. Interestingly, the Kdp values decline toward the ground. This could be partly due to some evaporation, but more likely there are some size distribution effects on Kdp. That is, a distribution of large water-coated ice particles may have inherently larger Kdp values than the same amount of water in drop form. Thus, as the water-coated hailstones further melt and form regular rain, Kdp values would be expected to decrease, as observed. This is a hypothesis that will be investigated using a T-matrix scattering model.

Posted by Timothy at 11:56 AM