Sunday, March 15, 2015

When a forecast goes wrong

The forecast made on Thursday and Friday for today has gone terribly wrong, as moderate to heavy rain falls over western Washington.  The composite radar image at 10:13 AM Sunday tells the story:


Two to four inches of rain have fallen on the western sides of the Olympics and Cascades.

The National Weather Service forecast released at 3 PM on Friday said:


SEATTLE AND VICINITY-
INCLUDING THE CITIES OF...SEATTLE...SHORELINE...FEDERAL WAY...KENT 330 PM PDT FRI MAR 13 2015

.SATURDAY...BREEZY...RAIN. RAINFALL AMOUNTS A HALF INCH TO ONE INCH POSSIBLE. HIGHS IN THE MID TO UPPER 50S. SOUTH WIND 15 TO 25 MPH. 
.SATURDAY NIGHT...RAIN AT TIMES IN THE EVENING...THEN MOSTLY CLOUDY WITH A CHANCE OF RAIN AFTER MIDNIGHT. RAINFALL AMOUNTS A TENTH TO A QUARTER OF AN INCH POSSIBLE. LOWS IN THE 40S. SOUTHWEST WIND 10 TO
20 MPH BECOMING NORTHEAST TO 10 MPH AFTER MIDNIGHT. 
.SUNDAY...MOSTLY CLOUDY WITH A CHANCE OF RAIN. HIGHS IN THE LOWER TO MID 50S. NORTH WIND 10 TO 20 MPH. 

And they were not alone, the WeatherChannel forecasts were similar as was the forecast I gave on KPLU.   

Why did we get it wrong?   In fact, this is a good example of one of the most problematic and difficult forecasts: when an atmospheric river is hanging around our coast for an extended period, and waves/disturbances on this river of moisture moves the precipitation around, often in poorly predicted ways.

Let me begin by showing you an amazing image:  the water vapor image at 10 AM on Sunday (in this image the NOAA GOES satellite is looking at wavelength in which water vapor is a very good emitter of infrared radiation).  The brighter areas indicated more water vapor.  You can see a plume of water vapor from the deep tropics.   This is not a pineapple express, it is a tropical express.

An analysis of the total water vapor in a column of air (below) shows the narrow plume of high moisture content reaching our region (5 AM Sunday).

Let me show you the forecasts of one modeling system (the University of Washington's), but they all had the same issues.  In this images, I will show you the three-hour precipitation totals ending 11 AM Sunday.  

The latest run (6hr forecast that began 5 AM Sunday) is quite good with precipitation over the region.  You see the undulation in the precipitation offshore?  That is the result of a low pressure system (or wave) on the atmospheric river. 


Similarly, a 30-hr forecast began 5 AM Saturday has the right story.

But the forecast begun 5 AM Friday has the band of precipitation over Oregon.


The forecast begun 5AM Thursday has the rain south of  Puget Sound.

We can get a better view of a relatively strong low center that is moving along the atmospheric river, causing the precipitation band to ripple and undulate.  Here is the UW WRF model forecast pressure pattern (solid lines) and temperatures at 925 hPa (about 3000 ft above sea level) for 5 PM Sunday, based on a the model simulation started at 5 AM Sunday.   A strong low pressure center right over Puget Sound country.  This should be close to correct. Expect strong winds behind it!


The forecast started 5 AM Saturday also has the low, but farther south.  This forecast follows a similar track to the 5 AM Sunday prediction, but the low is weaker and slower.

However, the forecast starting 5 AM Friday is very different, with the low being weak and it never moves into our area.

This was a situation of substantial inherent uncertainty, since small undulations on atmospheric rivers can produce big changes in local precipitation.  And the inability of the models to lock into the position and intensity of the strong low center before Saturday morning added to this problem.

A tool that meteorologists use is ensemble forecasting, running models many times with slightly different initial states and model physics.   Was that useful in this case?

The NOAA/NWS SREF ensemble system started on Saturday at 8 AM, showed that most of the model runs indicated substantial precipitation on Sunday at Seattle (see below, shows 3 hr precipitation, time in GMT/UTC, black line is the average of the ensemble of forecasts.)


But the story is very different for the ensembles started on Friday AM, 24 hr before. The first peak in the average was for Saturday and the mean (black line) was clearly dropping on Sunday.  But there was some hints of possible disaster:  a few of the runs were going for heavy precipitation on Sunday.


Anyway, is a good example of a very difficult forecast challenge.  We had the right story the day before, but predictability was low for longer ranges.  With more data over the Pacific and better use of the data we have, hopefully such failures will decline in the future.


11 comments:

Joel said...

Example of how failure can often be the best time to learn :)

ryamkajr said...

But that is what I am curious about. Is there actually anything to learn in this case? Is this a case of not understanding what COULD happen vs what WOULD happen.

They knew what the general conditions were going to be, but variations in the air current/water vapor content ended up with a different result. In these cases, they can only take so much input without specific, precise, immediate measurements, and there will always be natural variabilities (randomness) due to the general "waves" of the weather.

John McBride said...

Cliff, is it accurate to suggest that the activity in the Seattle vicinity on Saturday was indicative of the probability that the forecast was inaccurate? I offer that simply because I live in NE Seattle, about 2 miles west of the Sandpoint NOAA offices, and observed that the conditions Saturday weren't what I'd expected based on the forecasts I'd read. At most we had a few tenths of inches of rain from Friday night through Saturday morning. Saturday was dry, really very pleasant, but leading me to conclude that the system that was plain on the maps was either not behaving as anticipated, or had changed.

After the Sounders game Saturday night it began to rain the rain that I had anticipated, but was still raining in the morning when we did the St Patrick's Day Dash, and is still raining copiously here, in NE Seattle, as I type at 2:15.

Interesting information you've posted. Thanks very much.

John F McBride
or, for, St Patrick's Day
Sean Proinsias Mac Giolla Brighde
.

Rod said...

Indeed, Cliff. This is one of the bigger misses that I have seen in a very long time. The rain has parked over West Seattle all day...

Not complaining, though. I am happy that the weather forecasts are not perfect...they have improved...big time...but not perfect.

Matches life...

-Rod

Dude Diligence said...

I signed up for the McClinchy Mile bicycle ride based on the original forecast, then came the flip-flop on Saturday. I still did the ride in steady rain. Weather Underground had the rain in the forecast about a day before the NWS finally came around.

David B. said...

It has not stopped raining since daybreak here on Bainbridge Island (and it rained a good part of the night, too). Now standing at 1.92" since it began and 2.66" for the weekend.

Front yard is more like a marsh than a lawn at the moment.

Ron Muzzall said...

We have farmed on Whidbey Island for a hundred years now. We have had .82 today and it appears to be about done. The challenge with PNW weather prediction is and always has been unpredictability. Models or no models it is still the institutional knowledge that makes the difference. In weather forecasting and farming.

Eric Deutsch said...

I just noticed on my home LaCrosse weather station that the pressure just shot up a lot in the last 3 hr. I thought it might be an error, so looked at wunderground, and sure enough, their plot shows the predicted pressure zooming up an inch of Hg in the next 15 hr. That seems like to huge fast jump. What's the cause of that? Is that unusual?

Dick on Whidbey said...

Dr Mass. For your records: I live on the West side of Whidbey Island, GPS 48.1439, -122.6042. As of this time the 24 hr rainfall total on my rain gauge is 1.41 inches. Dick on Whidbey.

Gary said...

Here in Olympia I had some yard work to do this (past) weekend. The mid-week forecast had pegged Sat as the wetter day, but, by 1pm the showers had ended and didn't resume until 10pm. I was able to get a lot done.

About 9pm Sat I checked the forecast, which showed 90% rain through 6am Monday, with total precip about 2". This was more or less consistent with the Sat 8AM ensemble forecast Cliff showed (for Seattle).

It did pour Sunday but ceased abruptly at 5pm (and has been dry since), just when the ensemble had predicted a peak in precip (around 00Z Mon 16th).

So, even with 18H lead time or less, the timing of this storm couldn't be accurately forecast. Nevertheless, with a measured precip since Sat 10pm of slightly > 2", the prediction was remarkably close.

richard583 said...

.. One for the record books certainly. A few more stories like this one would probably make a pretty good read.