Detection of Summer Hail using Radar
Iwan Holleman / Hans Beekhuis
Contents:
Introduction:
On May 1. 2001 the first part of the new radar-based warningproduct, the detector of summer hail, will become operational. From then on, an overview of detected summer hail will be displayed on the Meteorological Work Station every 15 minutes. Last summer the prototype of the hail detection product has already been displayed via the intranet. The data, collected during this semi-operational run, has been used for an extended verification of the hail detection product. The results of this extended verification show that the new hail detection product detects summer hail in a reliable way, even in semi-operational context.
Method of Waldvogel:
At the end of the seventies Waldvogel et al. developed a method for the detection of summer hail which is based on a combination of radardata and information on the temperature distribution of the atmosphere. In the figure below a vertical cross-section through the three-dimensional scandata of the Radar in De Bilt is shown. Such a cross-section is often called a Range Height Indicator (RHI).
The azimuth at which this vertical cross-section is taken is 187 degrees, so this cross-section runs from De Bilt, via Noord-Brabant, and finally to Belgium. This figure gives a good insight into the vertical structure of a thunderstorm passing Baarle-Nassau on august 8 1999 at 16:34 UTC. A number of altitude levels have been indicated in the figure as well: the height where the pCAPPI-cross-section, the "normal radarimage" is taken, the height of the freezing level (H_T0), and the maximum height at which a reflectivity of 45 dBZ (corresponding to about 25mm/h) is observed (H_Z45). It is quite obvious that the pCAPPI-cross-section contains hardly any information on the vertical structure of the thunderstorm. The method for detection of hail from Waldvogel is based on the observation that the presence of high radar reflectivities, i.e, the presence of large amounts of liquid and solid waterparticles, at high altitudes points to the presence of a strong updraft. When radar reflectivities of 45 dBZ or higher are observed above the freezing level, and thus large amounts of undercooled or solid water is present at high altitudes, the presence of summer hail is likely. Finally, Waldvogel et al. reached the conclusion that the presence of hail is likely when radar reflecitivties of 45 dBZ are observed 1.4 km or more above the freezing level and that probability increases when this high reflecticty is higher above the freezing level.
Verification:
An extensive comparisson has been conducted between the results of the radar-based hail detection methods and reported hail by on-ground observers. The radar scan data of a few selected days within the summer of 1999 and of all days between May and September 2000 has been used for this purpose. The collection of on-ground hail observations for these periods turned out to be quite a challenge. Because summer hail is usually a small scale phenomenon, it is almost never observed by the network of synop stations. In the summer of 1999, for instance, the 19 synops-stations have reported only 14 cases of hail and during the summer of 2000 a mere total of 26 cases. In order to obtain a more dense observer network, hail reports of the 325 voluntary rainfall observers have been taken into account as well. In addition, cases of damage caused by hail as reported to three large insurance companies during the summers of 1999 and 2000 have been used. The reports of hail damage to agricultural property will surely not be distributed homogeneously across the Netherlands. This inhomogeneity can, however, be taken into account by looking at the distribution land use. Despite this extensive search for groundtruth data, there still will be cases of hail that have not been reported. Eventually this will result in an underestimation of the performance of the hail detection product.
This figure shows the results from the extended verification during the summer of 2000. Only days with a freezing level higher than 2 km have been taken into account in order to exclude possible events of winter hail (small hail on a large scale). The green curve shows the Probability-Of-Detection (POD) as a function of the warning threshold, and the red curve shows the False-Alarm-Rate (FAR). It is obvious from this figure that a high POD is accompanied by a high FAR, and a low FAR by a low POD. The blue curve shows the Critical-Success-Index (CSI) which is an index of the overall performance of the method. The CSI has a maximum value at a warning threshold of about 1.75 km
Implementation:
In May of 2001, the hail detection product based on the method of Waldvogel became operational. For this, an radar composite, including both the De Bilt and Den Helder radars, with the 45dBZ echotops is produced every 15 minutes. In addition, a map of the height of the freezing level is calculated from Hirlam forecasts (+3h,+6h) every 3 hours. When a 45dBZ-echo is present in the radar image, the difference between the height of this echo and height of the freezing level is calculated. Making use of the verification results of the summer of 2000, this height difference can be converted to the Probability-Of-Hail (POH). The POH is coupled to the FAR.
This figure shows the curve of the POH as a function of the warning threshold, i.e., the difference between the height of the 45dBZ-echo and height of the freezing level. A straight line has been fitted to the POH curve. The coefficients of this line (offset: 0.319, slope: 0.133) are used in the operational hail detection product to convert a height difference into the Probability-Of-Hail.
Day Bin of Detected Hail:
For referencing by the Klimatological Department (WM/KD) and for verification purposes, the 96 hail detection images are overlayed into a single images every day (day bin). For this, all images are collected and the maximum Probability-Of-Hail that has occurred that day is determined for each pixel.
This figure is an example of a day bin image of detected hail for August 8 1999. It is evident that there are several tracks of thunderstorm cells visible in this image. Due to the 15 minute timespacing between the subsequent hail detection images used to compile this day bin, there is a clear jump-like motion of the hail cells. For general use of the day bins, it is desirable to interpolate between subsequent images to obtain a continues field of detected hail.
The construction of the hail day bin using interpolation is a three-step
process. First of all, a pattern recognition technique is used to detect hail
cells in each of the 96 hail images per day. Hail cells are detected using a
recursive algorithm wich looks for neigboring pixels above a certain threshold.
For each of the detected cells, the average position and the number of pixels
is stored. Then, detected hail cells in subsequent images are matched. For each
pair of cells in subsequent images, the displacement vector/velocity is
calculated. The displacement vector (Vd) is compared with the (predicted)
Hirlam wind vector at 700 hPa (V700). The quality of the match is expressed in
a quality index:
| | Vd - V700 | | |
| Q = | ---------- |
| | V700 | |
Hail cells are matched when the quality index Q is smaller than 1. In addition, a match with a larger cell is preferred above a match with a smaller cell. Finally, the interpolation between pairs of matched cells is performed by displacing both cells is small steps towards each other while keeping for each pixel the maximum value that has occurred. An example of the interpolated day bin is shown above. Cells that have not been match to another one are just shown as is.
Examples:
In the figure below the outcome of the hail detection method of Waldvogel is
shown for the same time and date as the vertical cross-section shown above. In
Baarle-Nassau and the surrounding area, a hail storm has indeed been observed
and it produced hail with a diameter of maximum 3.5 cm.
A more recent example is shown below. The end of the warm, almost tropical period in the first half of may was accompanied by severe thunderstorms. At several locations hail was reported, south of Dordrecht hail was reported around this time with a diameter of 2-4 cm.
Back to the index
Hans Beekhuis