Automatically correcting bad differential reflectivity data

Heavy rain fell for much of the past week across parts of the Gulf Coast, but by far the worst of the deluge was Thursday night through Saturday night in Louisiana, where rainfall over 12” was widespread and some areas near Lafayette and Baton Rouge approached or exceeded 24”. More general info about the flooding can be found here (

In an event like this, it’s natural to want to have a map that shows how much rain has fallen since it started. All Weather Service radars produce a Storm Total Accumulation product, which is exactly that, the amount of rain estimated by radar since the start of the event (“event” start/end usually based on the presence of sufficient radar echo). With the advent of dual-polarization, they now produce a dual-polarization-based Digital Storm Total product as well as the legacy reflectivity-based Storm Total Accumulation product. Some people may have noticed that the Digital Storm Total precipitation estimate for the New Orleans radar (KLIX) was way too high during this event. In the screen shot below on the left, the large area of white is at least 40.6”, which is the largest value that can be represented in the file format used for these data. Though it rained a lot this weekend, it didn’t rain that much, and certainly not that much in so many places. The legacy Storm Total Accumulation displayed on the NWS website has much lower values for the same time period, all less than 20” (below on the right). These are generally lower than actually reported by gauges, but much closer to reality. So why the huge difference?

The main difference between these products is that the legacy version only uses reflectivity to estimate rainfall, while the new digital version uses dual-polarization variables such as differential reflectivity (ZDR) and specific differential phase (KDP) in addition to reflectivity. These extra variables provide additional information about characteristics such as the shape of the drops and the total water content along the path of the radar beam, and this information generally improves the precipitation estimate. However, in this case, the ZDR was being severely underestimated by the New Orleans radar. In general, lower ZDR means higher rain rates for the same relatively high reflectivity (see for more discussion of this). This error in ZDR led to an error in the precipitation rate calculation with each scan, and when those errors are added up over a few days of heavy rain, the total error is very large.

The radar team at WDT has frequently observed this type of ZDR error at the New Orleans radar (and similar kinds of errors at other radars) and the effect on precipitation estimates. Many months ago, we devised and implemented a method to automatically correct for them to prevent gross errors in the precipitation estimates. Our method uses the routine data disseminated by the NEXRAD network to produce correction values that update every hour and are automatically applied to the raw ZDR data for each radar before it is used for hydrometeor type classification or precipitation rate estimation. During this event, the correction being applied to the New Orleans radar was approximately +1.2 dB (considering ZDR typically ranges from 0-5 dB in rain, an error of 1.2 dB is very large). Below are images of the ZDR before (left) and after the correction (middle), along with ZDR from the Jackson, MS radar for comparison (right).

WDT’s storm total precipitation estimate (that uses dual-polarization radar data, gauges, and satellite) for the event is shown below. The maximum rainfall amounts near Baton Rouge are up to almost 30”, which is in better agreement with the gauge values (multiple 25-31” amounts NE of Baton Rouge) than either estimate from the New Orleans radar shown earlier. Thanks to our ZDR correction process, we are able to obtain the benefit from the dual-polarization data without being hurt by these persistent, systematic errors.



Spring is fire season too

In addition to severe thunderstorm season, spring is also grass fire season in the plains. While many ranchers execute controlled burns through the spring to clear dead grass from the previous year and remove unwanted tree saplings, there are also many unintentional fires from other sources. This week was been warm, dry, and very windy across the plains. Because most grass has not yet greened up this spring and rain has been limited, the strong wind and very low humidity is leading to critical fire weather conditions in which fires spread quickly and are difficult to put out, as indicated by the Storm Prediction Center fire weather outlook for today.

national RHNational wind speedday1otlk_fire

One of the larger ongoing fires started yesterday in northwest Oklahoma and is still burning in southwest Kansas this afternoon. So far it is estimated to have burned over 40,000 acres. Both the burn scar (dark line in the red-outlined area) and the smoke plume (white patch in the white-outlined area) can be seen in the visible satellite imagery.

Fire satellite zoom

The smoke plume of this fire as well as another one to the south are also visible on radar. The first image is from the new dual-pane mode of RadarScope 3.0 and shows Reflectivity on the left and Correlation Coefficient on the right. Correlation coefficient is a measure of how similar the particles being sampled are to each other in size, shape, orientation, and composition, and because smoke consists of a mixture of small aerosol particles and larger more diverse pieces of ash, extremely low values of correlation coefficient is one of the distinguishing characteristics of smoke. That signature is apparent here, as almost all the higher reflectivity between Freedom, OK and Pratt, KS has correlation coefficient below 0.3 (“high” values for precipitation are > 0.95, “low” values for clutter or insects are 0.5-0.8).

The second image shows the same reflectivity on the left, but differential reflectivity on the right. Differential reflectivity is a measure of the average shape and orientation of the particles being sampled. Values close to zero indicate particles that are either spherical or have no preferred orientation. Positive values indicate particles that are wider than they are tall, and negative values indicate particles taller than they are wide. In the center of the smoke plumes, differential reflectivity is near zero. Given the very low correlation coefficient, this likely indicates larger irregularly-shaped pieces of ash that are tumbling or fluttering in the turbulent airflow above the active part of the fire and have no consistent preferred orientation. On the edges of the plumes, differential reflectivity is higher. This likely results from these same ash particles falling back to earth with a preferred horizontal orientation where vertical air motion is weaker away from the fire itself.


This is another example of interesting applications of dual-polarization radar data (and an opportunity to play with dual-pane mode in RadarScope), as well as a reminder to pay attention to fire weather advisories issued by the National Weather Service, and be very careful of outdoor fires on excessively windy days such as this.


Weather Scoffs at Calendar

Though many of us took time off over the holiday weekend, the weather certainly did not. A series of storm systems affected large areas of the country with strong tornadoes, hail, flooding rainfall, and/or blizzard conditions from the 23rd through today, and will continue through mid-week.

The first event of note was a severe weather outbreak on the 23rd over the Mississippi and Ohio valleys. There were 51 preliminary tornado reports from several states (the final number of confirmed tornadoes will likely be different) and over 250 reports of wind damage. This included the first documented December tornado in Michigan (an EF-1 near Canton). The hardest-hit areas were northern Mississippi and southern Tennessee, with a single thunderstorm producing nearly continuous tornado damage for over 2 hours and approximately 140 miles through several communities in northern Mississippi into Tennessee, including Holly Springs. Preliminary survey information indicates two separate tornadoes with a short gap in the damage path, the first rated EF-3 and the second EF-4. Several other strong tornadoes occurred across Arkansas, Tennessee, and Alabama, with a total of 7 being rated EF-2 or higher (subject to change as further damage surveys are completed). Below is an image of WDT’s radar-based rotation track product over some areas most affected by the outbreak. The strongest wind shear is shown in red, with decreasing strength in yellow and green. Preliminary tornado reports are shown as black triangles. The paths of the strongly-rotating thunderstorms responsible for the tornadoes appear as long parallel swaths.

WDT’s Rotation Track product for the 24 hours ending at 6 am CST on Dec 24, 2015.

Additional severe weather occurred across the South on Christmas Eve and Christmas Day, with a few reports of tornadoes, wind damage and hail (some up to baseball size). On the evening of the 26th, more intense tornadoes occurred, this time in the Dallas metro area. Damage in Garland, TX was preliminarily rated as EF-4.

Another facet of the warm side of the storm systems was very heavy rain and flooding. Many areas from east Texas to southern Illinois and parts of Mississippi and Alabama were inundated with over 6” of rain between 6 am Christmas Day and 6 am on the 28th. Some locations received over 10”, which is a tremendous amount of rain for this time of year. The widespread heavy rain led to significant, in some cases record, flooding on many rivers.

WDT’s radar-estimated precipitation for the 72 hours ending at 6 am CST Dec. 28, 2015.


Christmas flooding
USGS flood gauges showing many locations with major flooding in progress on the afternoon of Dec 28, 2015.

Even though much of the extreme weather this week looked more like spring than winter, the storm also had a wintry side, with eastern New Mexico and western Texas receiving a major blizzard from the 26th through early on the 28th. Snow amounts exceeded a foot in some places. Persistent strong northerly winds frequently gusting above 50 mph piled the snow into drifts of 5-10 feet and closed numerous roads.

TX snow drift
Snow drift near Abernathy, TX. Via NWS Lubbock Twitter account.

The storm is still producing heavy rain, ice, and snow from Alabama and Georgia to the Great Lakes and Midwest.

1228 US radar
National precipitation-typed reflectivity mosaic showing widespread snow (blue), ice/sleet (pink), and rain (green/yellow/red) on the afternoon of Dec. 28, 2015.

-Noah Lock

That’s Not Rain

Oklahoma has gotten a lot of rain this year, and it looks like it is raining again this afternoon along the Red River near Frederick.

kfdr ref kfdr vis

However, there is hardly a cloud in the sky, as shown by the visible satellite image (right).

So it’s not raining there. What is the radar seeing, then? The answer is lots of grasshoppers, beetles, other bugs, and probably some birds, collectively referred to as “biological scatterers”. This type of radar return is very common during the warm months of the year, though today’s echoes are considerably stronger than normal. The yellow areas in the reflectivity image represent values between 30 and 40 dBZ, about the same as moderate rain.

We can use the dual-polarization variables to differentiate these biological echoes from actual rain. Return from bugs and birds is characterized by very high differential reflectivity (indicating objects that are wide but not very tall) and low correlation coefficient (objects of different size/shape/material mixed together) relative to typical values for rain or snow. The following images confirm that the radar echoes are the result of biological scatterers.

kfdr zdr kfdr cc

The differential reflectivity values (left) are near the top of the color scale, with most greater than 7 dB. Typical values for rain are 0 – 5 dB. The correlation coefficient values range from 0.7 (blue) to 0.95 (red). Typical values for rain are 0.97 – 1.0.

For the most common purpose of weather radar, knowing where and how hard it’s raining, it is often desirable to remove these biological echoes because they give a false indication of rain. The standard approach to this using dual-polarization data is through a Hydrometeor Classification Algorithm (HCA), which decides the most likely type of object present (rain, hail, biological, etc.) by comparing the observed values of the dual-pol variables to typical values for each classification. Regions that are identified as biological scatter or ground clutter are then removed from the final radar images. Below are the outputs of the National Weather Service HCA (left) and WDT’s POLARIS algorithm (right).

kfdr hca (nws) kfdr hca (wdt)2

The gray areas in both images are correctly identified as biological scatter. The green and yellow areas are misidentified as light rain and “big drops”, respectively. The “big drops” classification generally has the highest differential reflectivity and lowest correlation coefficient. It is most often observed in specific parts of thunderstorms. The misidentification by the NWS algorithm was likely a result of higher reflectivity values than are considered typical for biological echo. In these two images, it is clear that WDT’s algorithm did much better at correctly identifying the echoes as biological and filtering them out.

What’s so Special about Tropical Rainfall?

Last week produced the second landfalling tropical storm of the season in the United States, as Tropical Storm Bill came ashore along the central Texas coast. Bill moved slowly north and eventually northeast and weakened to a tropical depression, producing copious amounts of rain from Texas through Oklahoma and into Missouri. The figure below shows PQPE (WDT’s radar-estimated precipitation) for the 72-hour period ending at 7 am CDT on June 19. The black line shows the approximate track of the center of Bill during that time.


Tropical storms and hurricanes are known for producing both tremendous amounts of rain due to both the long duration of rainfall as well as torrential hourly rainfall rates that are commonly under-estimated by conventional radar-based estimation methods. The rain rates calculated using various common relationships for reflectivity between 30 and 50 dBZ are shown below. Note how much higher the tropical rain rate (green line) is for the same reflectivity.

rain rate comparison

The main reason tropical rainfall is different is that the process that produces the rain drops inside the clouds is different, as illustrated very well in the figure below, from the COMET program. Most non-tropical rain actually starts as snow high up in the cloud where the temperature is below freezing. As the snow flakes fall, they melt into rain drops. This is known as a “cold rain” process because the particles grow mainly as snow. Heavier precipitation rates corresponding to reflectivity above 30 dBZ generally mean larger snow flakes that will melt into relatively large drops.


Tropical rainfall is generally formed by a different process that does not involve ice, and is known as “warm rain”. In this process, some liquid rain drops grow just large enough to start falling (still relatively small for rain drops). As they fall, they collide with other droplets that are smaller and not falling as fast (possibly moving upward with the wind). Often the two drops involved in the collision merge together into a larger drop that falls even faster and collides with more small drops. Collisions between drops can also cause the drops to break up into two or more smaller ones. Though this process can result in some large drops, it also results in very many small drops. These small drops contribute quite a bit to the rain rate but not very much to the reflectivity. The tropical rainfall formula accounts for these small drops, which is why it yields much higher rain rates for the same reflectivity.

The difficulty for meteorologists is often in knowing when to apply the tropical formula and when to apply one of the “normal” ones, because tropical rainfall can occur far from oceans or locations that would normally be considered tropical if there is a warm, humid air mass present (like one would often expect to find in the tropics). The most important characteristics of the air mass that supports tropical rain are high relative humidity through a deep layer, low cloud bases, and a high freezing level. In years past, analysis of these environmental factors was the key to knowing whether to switch to the tropical rain rate formula or not.

However, the recent dual-polarization upgrade allows for better characterization of the drop size distribution by using reflectivity and differential reflectivity (Zdr) together. The differential reflectivity essentially measures the average shape of the particles. Spherical particles have Zdr that is near zero, and particles that are “pancake-shaped” have positive Zdr. A fortunate property of rain drops is that they are spherical when small, but become increasingly “pancake-shaped” as they get bigger, so the Zdr can be used to infer the size of the drops as well. Methods to estimate precipitation that incorporate this information (such as PQPE) can handle both tropical rainfall and “normal” rainfall without having to switch from one formula to another.

Because we know that tropical rainfall has unusually large numbers of small drops, and small drops have low Zdr, we would expect that tropical rainfall would have unusually low Zdr for a given reflectivity. Below are representative reflectivity images from Tropical Depression Bill while it was over Oklahoma (left) and a typical non-tropical convective system over Nebraska (right).

ktlx_Reflectivity-20150618-040343-1 kuex_Reflectivity-20150618-131322-1

From several times near the times shown for each case, I calculated the mean Zdr for each reflectivity value between 30 and 50 dBZ (the dark green, yellow, and orange colors). These are shown in the figure below. The typical range of Zdr at those reflectivity values is between the thin black lines. As expected, the mean Zdr is considerably lower in the Oklahoma data. Unlike most cases where the Zdr increases as reflectivity increases, in the Oklahoma tropical rain data Zdr is nearly constant between 0.5 and 0.9 dB.

zdr comparison

This is another example of how data from dual-polarization radars helps meteorologists identify physical processes occurring in the atmosphere. When this information is used skillfully, it can lead to more flexible algorithms for estimating precipitation.

-Noah Lock

Water Water Everywhere

One of the applications of radar data is rainfall estimation, which is very important for agriculture and hydrology. In fact, the opportunity to improve these estimates was one of the driving motivations for the recent dual-polarization upgrade. Because of the relatively good spatial coverage and high resolution of the data, radars are particularly well-suited for quantitative precipitation estimation (QPE).

Over the winter, WDT has developed POLARIS Quantitative Precipitation Estimation (PQPE), which utilizes the dual-pol variables differential reflectivity (Zdr) and specific differential phase (Kdp) in addition to reflectivity (Z) to estimate precipitation amounts, in contrast to standard techniques that only use reflectivity.

Within the past week, very heavy rain fell in several parts of the United States, providing an excellent opportunity to compare traditional reflectivity-only QPE to PQPE. This blog post will highlight five of those areas.

Area 1: Localized rainfall near and just east of Austin, TX on May 5-6

The following image shows the 24-hour precipitation totals ending near 7 am local time reported near and just east of the Austin, TX metro area by a dense network of volunteer observers. The central and northern Austin metro area generally saw 3-5” of rain, with the highest totals near Elgin, including a maximum of 7.09”.


The following images are the corresponding radar-based precipitation estimates for the same area and time. PQPE is on the left and traditional QPE is on the right. Note that both images use the same color scale, with shades of red indicating 3-6” and pink indicating 6-8”. In this case, PQPE has maximum values between 6” and 8” that correspond well with the highest gauge observations near Elgin. Over the Austin metro, PQPE has values mostly between 3” and 5”, with a few pockets of 5-6”, which are also consistent with gauges. The traditional QPE has values that are much lower in these regions, with generally 1.5-2.5” over Austin, and maximum amounts near Elgin of only 3-4”, only about half of what was observed.

dualpol_qpe1440min_smoothed-20150506-120000_tx opqpe_qpe1440min-20150506-120000

Area 2: Southern Nebraska on May 6

During the evening of May 6, numerous thunderstorms tracked over the same narrow strip of southern Nebraska, resulting in tremendous rainfall and several reports of severe weather. The following image shows the National Weather Service precipitation analysis (using radar, gauges, and expert quality control) for the 24 hour period ending at 7 am local time on May 7 over the state of Nebraska. Maximum amounts exceeded 10”.

May 6 NE Stage IV

Below are the corresponding PQPE (left) and traditional QPE (right) images. In this case, both methods agree quite well with each other and with the gauge observations (maximum amount of 10.47”, several observations above 8” within the heavy rain swath).

dualpol_qpe1440min-20150507-120000_ne opqpe_qpe1440min-20150507-120000_ne

Area 3: Oklahoma and neighboring states May 5-11

Heavy rain occurred over much of Oklahoma and parts of Texas and Arkansas on multiple days during the week, leading to widespread flooding. Rainfall amounts observed by the Oklahoma Mesonet for the 7 days ending at about 9 am local time on May 11 exceeded 10” at several locations, with a maximum of 12.57” at Minco, just west of Oklahoma City, as shown in the image below.

May 5-10 Mesonet Rainfall

The corresponding 7-day rainfall estimates from PQPE (left) and traditional QPE (right) are shown below (note the different color scale for the longer accumulation interval, with shades of red now representing 5-10” and pink showing 10-15”). The two techniques agree reasonably well regarding widespread 5”+ rainfall in southern Oklahoma. PQPE appears to be perhaps a bit closer on the very high amounts along I-40 east of Oklahoma City, though there are some underestimates in SW Oklahoma (this likely had to do with the raw data feed from the Frederick, OK radar having some outages/delays during the period).

dualpol_qpe10080min-20150511-120000-ok opqpe_qpe10080min-20150511-120000_ok

Area 4: Coastal North Carolina on May 11 associated with TS Ana

This past week also saw the first named tropical storm of the year in the Atlantic basin, with Ana making landfall in the Carolinas and producing some localized heavy rain, with some areas of eastern North Carolina getting up to 5”. The following image shows the National Weather Service rainfall estimate for the 24 hours ending at 8 am local time on May 12.

May 11 NC Stage IV

Rainfall with tropical characteristics is one of the situations in which traditional QPE often significantly underestimates rainfall, and dual-polarization methods perform better, and this was no exception. Below are the PQPE (left) and traditional QPE (right) images for the corresponding time period. The maximum amounts and areal coverage of amounts above 2” are clearly better represented by PQPE.

dualpol_qpe1440min-20150511-120000 opqpe_qpe1440min-20150511-120000_nc

Area 5: Southeast Wyoming on May 9-10

Even the high plains of southeastern Wyoming got in on the heavy rain action in the past week. As the following image shows, some volunteer observers east of Cheyenne saw over 3” on May 9-10, an unusually heavy amount of rain for an area with an average annual rainfall of around 16”.


In this case, the precipitation mostly fell as a cold rain, with the freezing level not far above the surface. Because the radar beam gets higher as it moves away from the radar, a low freezing level means it will be observed relatively close to the radar (located close to Cheyenne). In addition to the reduced correlation coefficient seen near the freezing level that was described in a previous blog post, the area near the freezing level often contains higher reflectivity than other areas. Again, because the radar beam gets higher off the ground as it goes out, this often shows up as a ring of higher reflectivity known as the “bright band”. An example of this is shown below, with a ring of higher reflectivity between the black rings on the 3.3° scan from KMPX. The outer ring, representing the location where snowflakes are beginning to melt, is at a height of about 3 km above the radar, which is very close to the environmental freezing level.

kmpx bright band annot

If these enhanced reflectivity values are persistent over the same areas on the lowest radar scan, rainfall can be considerably overestimated by traditional QPE methods. That appears to have happened in this Wyoming case, because traditional QPE (right) is severely overestimated just east of Cheyenne. Because PQPE (left) takes advantage of dual-polarization data to identify areas that are likely part of the “bright band” and apply a compensating adjustment to the rain rate, it provides a more realistic estimate.

dualpol_qpe1440min_smoothed-20150510-120000_wy opqpe_qpe1440min-20150510-120000_wy

These examples illustrate the potential for improved estimates of heavy rainfall by using dual-polarization variables in addition to reflectivity. These improved estimates would be beneficial for hydrology and flood warnings/forecasts, as well as agriculture.

-Noah Lock


It has been nearly 24 years since the first NEXRAD radar was deployed and today the network continues to be regarded as “state of the art” and the best S-band operational network of radars in the world. Simply having identical hardware (such as radar frequency, beam-width, manufacturer, etc.) and software builds across the network results in a high degree of reliability and usability of radar data. Commonality among all radars is typically not seen in many other radar networks. But the NEXRAD network also remains on the cutting edge of science through continual refreshes of radar technology. Scientific upgrades including advances in radar algorithms, the addition of dual-polarization technology, and new scanning strategies continually provide new operational uses of NEXRAD radar data.

Some of the more exciting upgrades to the network have come through increases in spatial and temporal resolution of the data. New Volume Coverage Patterns (VCP) have been introduced to shorten volume scan completion times such as with VCP 12 in 2004.  Radars in VCP 12 could provide a full volume scan for the first time in under five minutes, reaching an update time of about 4.2 minutes. Super-resolution data increased resolution to 250 m radially and half-degree azimuthally in 2008. But even further increases in temporal resolution could be made not just through the introduction of new VCPs, but by providing new dynamic scanning techniques.  In 2012, the Automated Volume Scan Evaluation and Termination (AVSET) method was implemented to terminate a current volume scan as a radar tilts upward once radar echoes are no longer observed. Thus, low level updates where precipitation is occurring are made available more often without sacrificing vertical sampling. AVSET can provide volume scan updates as fast as every 190 seconds (just over 3 minutes).

AVSET set the stage for further improvements through dynamic scanning in 2014 with the introduction of the Supplemental Adaptive Intra-Volume Low-Level Scan or SAILS. SAILS provides a supplemental elevation scan at the lowest tilt (0.5 degrees) for VCP 12 and 212 scanning strategies. The additional scan is inserted into the middle of the volume scan thereby halving the latency of data closest to ground. When not using SAILS, the 0.5 degree update time in VCP 12 is 253 seconds, but with SAILS the average update time would improve to approximately 147 seconds on average (in table below). When SAILS is combined with AVSET, the frequency of 0.5 degree scans could increase to a maximum of 101 seconds.


The video below compares radar returns for KICT (Wichita, KS) from a convective event on April 17, 2015.  The image on the left shows KICT scans at 0.5 degrees for a two hour period (0300-0500 UTC) while the radar was operating in SAILS mode using VCP 12.  The image on the right is what the KICT returns would have looked like without SAILS.

With SAILS, the average update frequency of the lowest elevation angle improves to 162 seconds, with some updates available as fast as 140 seconds.  Faster update times is reflected in a smoother looking transition between images. The figure below gives the update time of the lowest elevation angle in seconds for KICT over the two-hour time period.  Update times ranged from 140 to 181 seconds.


Believing you can never have enough of a good thing, the concept of SAILS has been pushed even further with MESOSAILS or the Multiple Elevation Scan Option for SAILS.  MESOSAILS is currently being tested operationally for several radars in 2015 with plans to deploy this capability throughout the whole network in 2016. In MESOSAILS mode, a radar returns to scan at 0.5 degrees 2 or 3 times within one volume scan. A study was performed to determine if the antenna/pedestal assembly of a NEXRAD radar could withstand the acceleration/deceleration strain in MESOSAILS mode.  It was found that performance characteristics were all within original design specifications for the NEXRAD radars. With MESOSAILS, a maximum of 4 total scans can be made at the lowest elevation angle within one volume scan. Original testing of this new method has shown that update times at 0.5 degrees could be reduced to as low as 89 seconds on average and potentially as low as 72 seconds a certain points within a volume scan.

SAILS Test VCP Definitions

One of the first radars to employ the MESOSAILS technique is KDDC (Dodge City, KS). KDDC was running in MESOSAILS mode for the same convective event shown above for April 17, 2015. A comparison of the KDDC lowest angle returns (shown on the left) with KICT (without SAILS, on the right) is made in the video below. The time period is over the same two hours as shown in the prior video. Clearly, the improvement in update time can be seen in the animation.

The following figure shows the update time for KDDC at 0.5 degrees for a 20-hour time period encompassing the two hours in the video. KDDC was operated in SAILSx3 and SAILSx2 modes on April 16th and 17th and the transition is easily seen in the graph.  In SAILSx3, 0.5 degree scans are updated every 100 seconds on average and 116 seconds in SAILSx2.  It is interesting to note that the update time approaches 1 minute at several points with some updates provided after only 70 seconds.


Obviously approaching 1-minute updates for the lowest level scan could prove crucial in severe weather scenarios especially for the detection and tracking of tornadoes. But the MESOSAILS technique should benefit a host of radar products ranging from rotation tracks, storm cell identification and tracking, hail detection and tracking as well as providing more accurate QPE products. It will be intriguing to determine the extent of improvement in such algorithms once MESOSAILS is deployed across the entire network.

But MESOSAILS is just one of a very long line of improvements made to the NEXRAD network. The continual refresh of new technologies and operational control of radars keep the network at the forefront of scientific radar development. This also ensures that users from NWS forecasters to private sector companies will have plenty of new tools and opportunities to provide cutting edge products for the benefit of its clients.

-Chris Porter

Cold fronts, Couplets, and Debris Oh My

Radar can be an incredibly useful tool when used in nowcasting or real-time situations.  It has become an indispensable source of data to National Weather Service forecasters as well as researchers.  The events of March 25th across Oklahoma showed how radar can be used to detect several different types of phenomena all occurring within a short time span of one another.

By 2300 UTC March 25th (6 pm local time), storms had erupted along the cold front from west-central Oklahoma into northeastern Oklahoma.  One tornado had already occurred in Sand Springs near Tulsa 30 minutes ealier, unfortunately killing one person. The cold front can be seen in the image below as a thin line of relatively enhanced reflectivity south of several convective storms.


Zooming into the central Oklahoma region, a storm over El Reno is shown just behind the cold front with an associated mesocyclone, a low-level region of rotation within a supercell. The mesocyclone is apparent when examining radial velocity data, with green shading representing wind directions coming towards the radar, and red shading showing where winds are moving away from the radar. Radial velocities are one of six raw data fields that NEXRAD radar provides.  Interpreting radial velocity signatures can be very challenging because you have to account for where the radar data bin is located relative to where the radar is positioned.  The Warning Decision Training Branch has produced a document that provides an in depth discussion on interpreting radial velocity signatures.



The Storm Prediction Center issued a mesoscale discussion concerning storms in central Oklahoma, stating that they would “likely …become quickly undercut by the front.” Indeed, that appeared to be the case as the cold front was already south of the El Reno storm.  Many times when storms are undercut by a cold front or outflow, the storms can , weaken, or pose a reduced threat of tornadic activity.

However, by 2317 UTC, a noticeable kink in the cold front was apparent on radar, showing that the storm, now east of El Reno, was drawing the cold front back to it. By this time, a coworker who was chasing this particular storm saw a partially condensed funnel with a large dust cloud underneath it on the ground a few miles southwest of Yukon.


At 2328 UTC, a noticeable vortex couplet was easily seen in the radial velocity field indicative of a possible tornado.   The couplet shows azimuthally adjacent wind velocities in a direction away from the radar (in red) immediately next to strong wind velocities coming towards the radar (in green).


About 7 minutes later, a new area of rotation rapidly began to develop within the same storm.  However, this new area developed further south, at SE 119th and Pennsyvlania Ave, located along the border between Oklahoma City and Moore.  The local Oklahoma City media were able to capture the unfolding events through video from helicopters.  At 2341 UTC, the tornado had reached I-35 and SE 4th in Moore where it was able to overturn a semi going northbound on I-35.  The enhanced area of reflectivity curled up into a ball is a tornadic debris ball showing an area of lofted debris, not heavy rain.


The same area had extremely low correlation coefficients, well below 0.8.  As noted in a previous blog, correlation coefficient (CC) is a dual-pol parameter that represents how uniform radar scatterers are within a data bin.  Previously, CC had been used in determining the location of the melting layer above the surface.  In this example, CC can also be useful in determining the location of tornadic debris.  Tornadic debris will contain all types of materials of different shapes and sizes, driving the correlation coefficient well below 1.0.


The tornado continued to track southeastward for another 5 minutes, and eventually lifted at SE 34th and Sunnylane in Moore.  Another useful radar product is shown below that can capture the paths of tornadoes referred to as Rotations Tracks.  Rotation Tracks are created by calculating azimuthal shear fields, or the azimuthal derivative of radial velocity. The azimuthal direction is the direction perpendicular to a radar beam, with the radar beam emanating outward in all directions from the radar.  The KTLX radar location is shown by the yellow dot.  The northern-most path shows the primary mesocyclone path for the storm that had tracked from El Reno into southern Oklahoma City.  The second, smaller path further south denotes the path of the Moore tornado.  Rotation Tracks show areas where potential tornadoes may have formed, but a tornado did not necessarily exist across the entire path.  This type of product is useful when used in conjunction with other data to determine the most likely path of a tornado.  However, the Rotation Track product can also be very noisy, showing all types of shear patterns in the atmosphere including gust fronts and cold fronts.


As shown through the events of March 25th, many physical properties and storm attributes can be revealed by radar. However, even an experienced radar meteorologist can be challenged to understand the unfolding events of a severe weather event using radar data.  Interpretation of radial velocities, patterns in reflectivity and uses of dual-pol data can be difficult; especially when trying to interpret what is happening as it is actually occurring in real-time.  However, when several different radar data fields and derived products are used together, especially in hindsight, the progression of severe weather events can be illuminated.

-Chris Porter

Dual-Polarization and the Melting Layer

A slowly moving upper-level system produced a continuous area of light and moderate rain across central and eastern Oklahoma throughout the day Friday.  The reflectivity image below shows a rather innocuous system generating widespread rainfall.


Prior to the implementation of dual-polarization technology within the NEXRAD radar network beginning in 2011, only reflectivity, radial velocity and spectrum width information would have been available.  Dual-pol data provides three additional radar variables, and this newly available data can help to reveal physical processes occurring within the atmosphere.  Although Friday’s rain was rather benign, it afforded an opportunity for dual-pol technology to clearly locate the melting layer within the atmosphere over Oklahoma.

The image of correlation coefficient below is one of three dual-pol radar variables available. Correlation coefficient can provide information on the homogeneity of radar echoes within a radar volume.  A correlation coefficient near 1 indicates a uniform consistency of hydrometeors.  When a mixture of different hydrometeors is present, the correlation coefficient will typically drop below 0.98.  This transition can easily be seen below using data from the KTLX radar below.  The area closest to the radar has correlation coefficient values near 1.0 (red shading).  Beyond the white arc, values drop below 0.98 (within the yellow shading).

The transition from red to yellow marks the beginning of the melting layer.  The melting layer is a region of the atmosphere where snow and rain coexist as snow is falling aloft and melting before it reaches the ground.


While the image above is of the 0.5 degree elevation scan, the one below is from the 3.4 degree tilt.  Higher elevation scans show the melting layer region more distinctly as the radar beam passes upward through the melting layer region more quickly.


The white ring in the image shows the bottom of the melting layer located approximately 6500 feet above the surface.  The black ring denotes the top of the melting layer positioned somewhere from 9000 feet (to the west) to over 10,000 feet (to the east) above the surface.  The freezing level is located at the top of the melting layer, where temperatures above it are below freezing and temperatures below the freezing level (and within the melting layer) rise above freezing.

An upper-air sounding from central Oklahoma at approximately the same time is shown below. The thick black line is a line of constant temperature; in this case 0 degrees Celsius.  The red line shows the change in temperature vertically from the surface (bottom of the graph) to the upper levels of the troposphere (top of the graph). The red line passes across the black line at the freezing level.  In this sounding, the freezing level of the atmosphere is located at 700mb or about 10,100 feet.


Thus, the correlation coefficient did indeed provide a very accurate indication of where the freezing level was located in the atmosphere.  This also shows the power of using radar data, which can produce a  3D representation of atmospheric processes, while soundings are limited to providing information at only specific points as a weather balloon ascends.

There is great amount that can be said to explain both what dual-pol radar data is and how it can be used to determine what types of hydrometeors are present in various regions of the atmosphere surrounding a radar. And this rather unremarkable weather event in Oklahoma shows just one example of how atmospheric processes can be brought to light using dual-pol technology, which became available to us only within the last few years.

-Chris Porter