Frequently asked questions (FAQ)
(select/deselect an answer by mouseclick on a specific question)

Why radar image forecasts?

These forecasts are the best in short time ranges (0-4 hours, also called the range of nowcasting). Major advantages vs. numerical forecast models are the short computing time and, consequently, frequent update rates (typically every 5-15 min). Forecasts of rainfall, rain accumulation, thunderstorms, hail, or (in winter) snowfall or freezing rain are possible and helpful for any activity in free air. Warning messages can be generated in a fully automated manner. Typical users of radar image forecasts and derived warnings are weather institutes/companies or traffic maintenance/water management organisations. But many others, also in private sector, use these forecasts for their comfort or protection of life.

Can I use your forecast technique for my radar images?

Yes. Please have a look to our links Server and Client at the top-left edge of this page.

Please note that this service is for demo only. You can upload and test as many examples as you like. But the capacity of our server for this demo is limited. You will have to wait as long as the server is busy.

Which are the costs for customers using trep?

The regular costs depend on

  • The size of your images
  • The time step between two consecutive forecasts
  • The type of your application (end-user, reseller)
  • Licensing, maintenance and support are included

Additional costs are unique and depend on our expences for
software modifications (if required) and installation.

Please do not hesitate ordering a proposal.


How is the radar forecast calculated?

First, we treat the radar images with a clutter filter. After that, we retrieve the motion of radar echoes, and we use the motion vectors for extrapolation into the future. Our technique is novel with respect to the fact that the advantages of two traditional methods, the box tracking and the cell tracking, are combined to one single tracking method. In this manner, the advantages of both methods contribute to the best possible extrapolation, leading to optimal forecasts for stratiform precipitation and convective storms as well.

What is a clutter filter?

Weather radar images are often affected by non-weather echoes, originating from ground (mountains, buildings, trees ...), or from birds, insects, aircrafts and so on. Filter techniques try eliminating these disturbing echoes, called 'clutter', without affecting weather echoes. Clutter filters are very often in use, but not always effective. Stationary radar echoes from ground may affect our extrapolation technique. Therefore, we use a novel filter which is especially designed for removing ground clutter still present in the radar images.

Why are you using 6 or 7 images for the forecast?

With six images, we have a good sequence for minimizing random variations in the calculated motion field. The last of the six images is typically extrapolated into the future. For some applications, it may be desirable using a modified radar image or even a non-radar image for extrapolation, e.g., an image showing lightning density. Therefore, a 7th image can be added in our demo. Obviously, this additional image must have the same size as the radar images.

Are you using a specific intensity scale?

No, the scale is free. Nevertheless, we recommend using a scale starting at 13 dBZ, rising in steps of 3 dB, and ending at 72 dBZ. This scale has 20 intervals, corresponding to levels 1 to 20 (out of 255) in your gif images. Please note that 20 intervals only are considered in our algorithm. All levels larger than 20 are set to zero before processing the images.

Who developed the technique?

Many persons have worked on the methodology since many years. Extrapolation of radar images into the future is a topic of radar meteorology since the sixties of the last century. One of several basic techniques is TREC ('tracking radar echoes by correlation'), introduced by US scientist Rinehart in the late seventies. COTREC ('continuity of TREC vectors') was developed in the nineties by two doctorands at ETH in Zurich. A modified version, called RAINCAST, is in operational use by meteoradar since 1999. A major update of the technique has recently been developed by meteoradar. We call it TREP since the basic idea of the old TREC has remained unchanged. Various in-house tests demonstrate a further improvement of the forecast quality of TREP compared with RAINCAST, especially for convective storms. Therefore, we upgrade our operational forecast applications to TREP in these days.

How does the forecast quality changes with forecast lead time?

Naturally, the forecast quality degrades with increasing forecast lead time. However, the degree of degradation depends on the size of the phenomenon you are interested in. A broad precipitation front is better predictable (up to several hours) than a small isolated thunderstorm cell (less than one hour). We developed a stochastic model for simulation of these properties. With this model, we can calculate the risk that an event such as hail or heavy rain may occur at a given location and time. Our warning system uses this model: a warning message is issued when a pre-defined risk value is reached for the first time.

Upload ZIP

© meteoradar 2008