Why Forecasting Winter Storms Can Be So Difficult
I often highlight how far forecasts of extreme weather events have come in the last ten years. Still, there are times when the forecast can be difficult and these generally involve small scale weather features. Take a look at this computer model forecast of moisture (in this case the water content of melted snow, not rain) for the next two days:
The northern band of snow shows heavy snow (dark blue) from southeast Nebraska and northeast Kansas across northern Missouri. The light blue is potentially more than eight inches. This forecast has good chance of being approximately correct.
Now, look at the southern band of snow. If you look closely you can see an area of dark blue, (4-8" of snow) about ten miles wide, on the Kansas-Oklahoma border. Then, just 30 miles south the forecast is for zero snow! The chance of this forecast being exactly correct? Poor to fair.
While the pattern is may be correct, the exact placement (location) of the narrow heavy snow band is smaller than our techniques can account for. If the system moves 40 miles north (and that is possible, but not likely), Sedgwick County's 500,000 residents get an unpleasant surprise with 4-8 inches as opposed to the 1-2 inches indicated by the model.
It is a fact of life for meteorologists that you measure our success by how much snow fell at your home and we don't get any "extra credit" for being correct 30 miles away. That is fine with us. I just want to let you know how some snow-producing weather systems are more "forecastable" than others.
Now, look at the southern band of snow. If you look closely you can see an area of dark blue, (4-8" of snow) about ten miles wide, on the Kansas-Oklahoma border. Then, just 30 miles south the forecast is for zero snow! The chance of this forecast being exactly correct? Poor to fair.
While the pattern is may be correct, the exact placement (location) of the narrow heavy snow band is smaller than our techniques can account for. If the system moves 40 miles north (and that is possible, but not likely), Sedgwick County's 500,000 residents get an unpleasant surprise with 4-8 inches as opposed to the 1-2 inches indicated by the model.
It is a fact of life for meteorologists that you measure our success by how much snow fell at your home and we don't get any "extra credit" for being correct 30 miles away. That is fine with us. I just want to let you know how some snow-producing weather systems are more "forecastable" than others.
Hi Mike,
ReplyDeleteInteresting find. Even though I firmly believe precipitation type and amount with winter weather is one of, if not, the hardest thing a meteorologist has to do, let me play devil's advocate.
How is this example any different from a meteorologist attempting to predict severe thunderstorms? Consider the case where a high-resolution, storm scale model forecast a severe thunderstorm to occur on the Kansas, Oklahoma border and in reality the storm occurred in Sedgewick County. Most meteorologists would classify that as a highly successful forecast. In fact, residents of Sedgewick County would, most likely, have been alerted to the severe weather threat ahead of time and would, again most likely, consider this a very successful forecasts -- especially two days in advance! Why the apparent contradiction?
My guess is that meteorologists have fallen into the habit of giving fairly specific snowfall forecasts because people demand them. If a meteorologist won't give a specific snowfall forecast, another will. This essentially forces the hold outs to either give in and provide a snowfall forecast or be put out of business. The problem is that there will always (or for the foreseeable future) be questions about snow-to-liquid ratios, how long till snow begins to accumulate on the ground, compression of snow due to the weight of accumulated snow, etc. All of these are outside the realm of what meteorologists can currently predict with any sort of regular certainty. (Un)Fortunately, throughout history meteorologists have gotten the snowfall forecast correct enough times to warrant people trusting them, only to be let down time and time again.
People are a fickle bunch...
Patrick,
ReplyDeleteThanks for an interesting comment. I'm sure blog readers will enjoy reading it.
We are making a great deal of progress with forecasting the larger extreme snow events (see: http://meteorologicalmusings.blogspot.com/2011/01/how-much-snow-fell-in-nyc.html ). The Atlanta snow storm was incredibly well forecast: http://meteorologicalmusings.blogspot.com/2011/01/lazy-reporting.html . These very small events are tough.
You say: "Consider the case where a high-resolution, storm scale model forecast a severe thunderstorm to occur on the Kansas, Oklahoma border and in reality the storm occurred in Sedgewick County. Most meteorologists would classify that as a highly successful forecast."
I agree that most meteorologists would consider it a successful forecast, but I would not. The general public and WeatherData's clientele, as I say in the post, do not give us credit for being right 30 miles away. They judge us by the weather they personally experience.
The other difference between snow and a severe thunderstorm is that no one would notice much difference between .2 inches of rain and .5 inches of rain where everyone notices the difference between 2 and 5 inches of snow.
Thanks for posting.
Mike
Mike,
ReplyDeleteA couple of comments.
You said, " ...I agree that most meteorologists would consider it a successful forecast, but I would not. The general public and WeatherData's clientele, as I say in the post, do not give us credit for being right 30 miles away. They judge us by the weather they personally experience. "
Do you mean to tell me that WeatherData can predict severe thunderstorms, down to the correct households, 2 days in advance as is the case with discussing snowfall totals 2 days in advance? If so, then you could easily put all other weather companies out of business (at least in terms of convection). Even then, you are implicitly couching your original snowfall totals by giving a range, not a specific amount.
Regarding the difference between .2 and .5 inches of rain and 2 and 5 inches of snow, I think you are missing my point. I've seen plenty of 5 inch snow falls end up being thought of (and reported) as 2 inch snowfalls by weather enthusiasts due to factors such as initial melting on contact, compression from weight of snow, melting from beneath, etc. In all of these events people will "think" they only had 2 inches of snow when in reality, what should have been 5 inches fell from the sky.
I can also think of situations where a difference between .2" and .5" of rain would impact a place more significantly than a difference of 2" and 5" of snow. Freezing rain is one. Mountainous areas another.
All weather is local...
In any event, I don't mean to detract from the post at hand. Winter weather forecasting is incredibly difficult. And as I said before, I generally think it is harder than severe storms forecasting. However, with that said, I do think the expectations are significantly higher (and at longer ranges) for winter weather forecasting than it is for severe storms forecasting. And I don't think meteorologists do a good enough job properly conveying the uncertainty and subtleties of winter weather forecasting. In that regard, this post is spot on.
ReplyDeletePatrick,
ReplyDeleteExcellent point about freezing rain, I hadn't thought of that.
You asked: "Do you mean to tell me that WeatherData can predict severe thunderstorms, down to the correct households, 2 days in advance as is the case with discussing snowfall totals 2 days in advance? "
We do not try to do this. We work really hard not to allow ourselves to be pushed past the state-of-the-art.
You are also correct that snow settled and compression causing people to mistake the correct amount of snow.
Thanks for a great comment.
Mike