by: Gene Pache

Anyone concerned about weather forecasts, which is all of us, wants to know if it is possible to obtain more accurate forecasts faster. The National Weather service has made progress in the area of hail versus rain and the movement of thunderstorms. Noteworthy improvements have been made as a result of the 160 doppler radars located throughout the US. The dual polarization doppler radars are able to distinguish precipitation types. These radars enable more accurate forecasting of hail embedded in thunderstorms.

Doppler radar is an outstanding technology. But how do we achieve an order of magnitude improvement in accuracy and speed of weather forecasts?   To attain an order of magnitude improvement in speed and accuracy of a regional weather forecast three elements have to be improved.

The three elements are:

(1) The weather model used to develop the forecast

(2) The accuracy of the “Initial Condition” of the atmosphere – the starting point

(3) Computing capability – the number of calculations per second available

All three elements are necessary to create a forecast. The weather models comprised of lines of code enable physical laws governing atmospheric motion, chemical reactions and other relationships to be applied to the initial condition of the atmosphere. The atmosphere’s “initial condition” is derived from the sensor dated collected and assimilated prior to beginning the forecast run. The weather forecasting model, with the initial conditions, is then run on high performance super computers with the capacity to perform trillions of calculations per second. The hardest element to improve with a given operating budget is element number three, computing capability.

To accelerate forecasting requires greater and greater numbers of CPUs. This significantly increases hardware costs. The greater CPU numbers also increase the cost of power and cooling.

One solution to the computing calculation capability problem is to use NVIDIA™ GPUs, graphic processing units. NVIDIA™ GPUs provide much greater calculating capability at far lower hardware and operating cost.

It is the solution TempoQuest is adopting.