Its All in the Timing

How many times does this happen?  The National Weather Service forecasts rain coming in at a certain time and the forecast is off by a few hours, either early or late.  The truth is that this happens all the time and such timing error increase with the length of a forecast.  And variations in timing of consecutive forecasts provide valuable information on forecast uncertainty.  Consider the rain over Saturday night.

Here is a radar image from the new Langley radar at 7 PM on Saturday night. 

A frontal band is moving in, crossing the north-central WA Coast. Now here is the 74 forecast valid this time (1-h precipitation at ending 7 PM).  Got the right idea of frontal system...but too far south.

 Here is a forecast for the same time from the next cycle (62h forecast).  Whoa!  The frontal band has shifted way north...too north.

Next forecast cycle and the 50 hr forecast...the band is shifted south...not too bad.

The 38 hr forecast?  Pretty similar but a tad stronger and farther south.

The 26hr forecast? Similar, but slightly weakened.

And finally the 14h forecast. Very much the same, but perhaps a bit more diffuse.

These are very typical forecast sequences.  Go far enough out in time and the forecasts are shifting in a big way, particularly in the time that features pass a location.  As you get closer in time, the solutions tend to stabilize.  On occasions when the solution stabilizes late or not at all, forecast confidence is low, and vice versa.    But just because a solution stabilizes does not mean it HAS to be right, and sometimes the forecast models equilibrate on  a solution that is not correct.

Forecasters have a special name for the variation in forecast for the same time (e.g., 10 PM August 5) but for different forecast intervals (12, 24, 36 hr forecast, etc):  dmodel/dt.  If any of you have calculus you know why.  For those who haven't, this translates as the change in model solution over a specific time interval... a.k.a. the time  derivative.   Large dmodel/dt--low confidence in the forecast, small dmodel/dt--more confidence.

Ten years I did a paper with Brian Colle and David Ovens that showed that models  were typically fast by an hour or so.  I haven't repeated this study, but I suspect that although the errors are probably less now, errors in timing are not usual.  That is why the coastal radar is so important.  When our models are off by a few hours--too fast or slow, the radar will give us a heads up when weather systems are 6-9 hrs out.

Dog Alert
There have been repeated sightings of our dog in and near Terrace Creek Park in Mountlake Terrace.  If you are in that area, please let us know if you see her.