Tuesday, May 26, 2020 11:30 A.M.
More showers, t-storms today
Big disconnect today between the NWS forecast and the models’ forecasts. NWS has just a 20% chance for afternoon/evening showers and thunderstorms. Models, however, are the most favorable for showers/t-storms they’ve been for quite awhile. They indicate at least a 60% – 70% chance you’ll have rain at your house this afternoon/evening. Man vs. machine…we’ll see how it works out! Personally, I’ll go with the machines.
Tuesday, May 5, 2020 6 P.M.
Current weather: A strong upper air disturbance over upper midwest now will rotate southeast toward Kentucky tonight. This system will provide some reasonably potent lifting motions over the lower Ohio Valley late tonight. Not much moisture is available, but additional periods of light rain/drizzle are likely late tonight into tomorrow’s morning rush hour.
We’ll get to see some sunshine as tomorrow wears on, but temperatures will remain unseasonably cool. The below normal temperatures will likely continue through the next week or two.
Computer models, part 3
The past two posts have told the story how weather forecasting and computers have been wedded since the beginning of electronic computing. Today there are computer models/projections for just about everything. Even one that predicted Secretariat to win the Super Derby last Saturday.
We all know that weather forecasts certainly are still not perfect, even though meteorologists have been at it the longest. Other forecast models have the same problem, but do get better with age. That leads us to the Coronavirus Models, Unlike sports, weather, economics, etc., the available data on pandemics is pretty sparse. Luckily not many pandemics occur. Nevertheless, models have been built and put into action. As expected we are hearing a variety of conflicting reports. Pandemic modelling is a relative new field…its going to take some time for the model errors to shrink. But with more data rapidly becoming available, improvement will occur.
Early talk of millions of deaths possible in the U.S. were simply “potential”. They assumed no precautionary steps taken. When precautions/restrictions went into effect across the country, the oft-quoted University of Washington model predicted 100,000 to 240,000 fatalities. Our PIC (prevaricator in chief) just laughed that off. Meanwhile, as the volume of data escalated, a few weeks later, UW lowered its prediction to 60,000. Subsequent revisions went to 68,000 and then to 76,000. All the recent revisions have been ridiculously low. I’ve been watching the case/death numbers closely. They just didn’t mesh with the predictions. For example, Monday UW estimate was still in upper 70,000’s. Just following the daily data, it was obvious we’d exceed that this week. But, then, yesterday…
Kudos to the New York Times
For some (obvious) reason, it appears that PIC and his gang have been withholding information from an internal government forecast model. That model predicts total U.S. deaths at 135,000. To me, that number seems about right. But with the recent rush to reopen the country, even that number could be low. Thanks to the Times for breaking this story. PIC, however, will likely just discard it as “fake news.” (The only thing worse for PIC was if Jeff Bezos’ Washington Post had broken the story.)
‘
Monday, May 4, 2020 6 P.M.
First, our local weather
A weak low pressure system will work its way across the southern plains tonight and across Kentucky tomorrow. It’ll bring us some rain tonight (after Midnight) into tomorrow. Any rain after mid-morning tomorrow will be very light. Total rainfall is expected to be in the quarter inch to half inch range. Cooler air filters in tomorrow and we’ll see below normal temperatures likely through the weekend, at least.
Computer models and forecasting, part 2
After the Army built the first eniac (computer) from 1943-6, the Navy decided to build a second one in cooperation with the private sector. A group of mathematicians and meteorologists was chosen to complete the project. Why meteorologists? Two reasons: first, the earliest computers had no what we now call “software.” The machine had to be built to solve one single problem. It would have no other use since the “software” had to be built into the machine, Second, the project leader was John von Neumann, a mathematician. Von Neumann, however, was familiar with Richardson’s work from the 1920’s. (previous post)
He figured the team had an already-solved problem. All they had to do was build the machine to perform the calculations. Thus, he reasoned, meteorology had the problem and a pre-existing method to solve the problem. That would save a lot of time.
In reality, Richardson had made some mistakes and faulty assumptions. The meteorology team spent a lot of redoing the physics and methodology before the machine could be built.
Finally, in 1950, in a large lab at Princeton University, eniac produced the world’s first non-military computer “output” – a 24 hour numerical weather forecast. It took the machine 24 hours to produce it.
Within two years, the computer time dropped to two hours. The world of “computer weather forecasting models” accelerated from there. Improvement has been immense.
Other models:
Computer modelling has expanded over the years. Virtually anything you can think of is under the scrutiny of various models. Weather has had 70 years working on the problem and we still make mistakes. Same goes for other models you may hear mentioned. Which brings up today’s most talked about model – the coronavirus model. The conversation continues tomorrow.
Sunday, May 3, 2020 5 P.M.
Summary
Weekend forecasts for Louisville were reasonably good although temperatures were several degrees higher than predicted yesterday. If this had been a “business as usual” Derby Day, I wonder how many bad sunburns the infield crowd would have suffered. Today’s temperature forecast was far worse than yesterday’s. The expected showers finally arrived, but rain so far has hardly been worth the effort. We’ll still maintain a chance for a few more light showers until about 8-9 P.M.
Looks like a nice day tomorrow with mid 70’s highs, then a good shot at a more significant rainfall tomorrow night into Tuesday.
Below normal temperatures will prevail from midweek through the weekend.
Forecast Models
Homo Sapiens (that’s us) evolved, we believe between 200,000 and 300,000 years ago, but did not develop language skills until about 50,000 years so. So, we’ve been trying to predict the future for at least 50 thousand years. The first scientific numerical attempts, that I’m aware of, were by Lewis Richardson in the early 1920’s. He used equations developed by Wilhelm Bjerknes ( the Farther of modern meteorology) to create a numerical “model” of the atmosphere. He then extrapolated the input data forward in time. He produced a 6-hour forecast for two cities in Europe. Sounds pretty simple, but there are an enormous number of calculations required to move forecast data horizontally and vertically through the atmosphere, even for just six hours. Working off and on, Richardson took six MONTHS to complete his 6-hour forecast!
The numerical prediction did not produce realistic results. But, the concept was proven correct. But the huge number of calculations needed proved it was not feasible at the time.
Jump ahead two decades. During World War II the military wanted some way to speed calculations needed during battles. Started in 1943 the project ended in 1946 by calculating (very rapidly) trajectories for canon balls by a machine called eniac (electrical numerical integrator and computer).
Story continues tomorrow