I assume this may be caused by the METRo model using the previous
hours' forecast as the observation and continuing from there, almost
like a self-feeding problem. It causes forecast generated in the
evening to get progressively colder as new model runs come out and
forecasts generated in the morning to get progressively warmer as new
model runs come out.
Any ideas on how to resolve this?
Hi Jeremy,
Like you said, the problem arise from the fact that the roadcast is
forced to match the first observation at first hour of roadcast. The
later in the night you initiate the forecast, the colder it will start.
This is what you observed in the provided graph, each line is below the
previous one.
I understand that you aim to improve the consistency between 2 roadcasts
initiated at different time. (By the way, for our non-US friends, the 20
degrees difference in question where Fahrenheit, not Celsius).
If you want to achieve a better consistency, you could simply use the
output of the previous forecast (say the orange line) to initialize the
next one (the green line) instead of the same old road forecast for all
of them. This would most probably give more consistency but, since
there is no more information added into the system, the road forecast
would probably not be more accurate.
In any case, I think that starting METRo with no observations cannot
give a good accuracy in the first forecast hours, especially during the
night. The accumulated energy into the soil is too important in the
energy balance to simply guess it at start and hope to have consistent
result.
However, after a full daylight of roadcast, METRo is probably more
accurate and consistent with the road surface temperature since it has
taken in account the total energy that went into the soil. You can see
this in your graph: the roadcast of day 2 are more consistent one with
another.
Maybe you can try, when there is no observations, to use day 2 for the
roadcast, not the day 1. The atmospheric forecast won't be as good as
day 1, but you will gain in term of consistency between the roadcasts,
and the uncertainty will be averaged in part during day 1.
What do you thing of this suggestion?
Miguel
--
Miguel Tremblay
Coordonnateur national des services commerciaux de données-
National coordinator, commercial data services
Centre météorologique canadien - Canadian meteorological centre (CMC)
Environnement Canada - Environment Canada
http://www.ec.gc.ca/
2121 Trans-Canada N. Suite 201 Téléphone/Phone: 514-421-4729
Dorval, Québec Fax: 514-421-4679
CANADA H9P 1J3 courriel/email: [email protected]
_______________________________________________
METRo-developers mailing list
[email protected]
https://mail.gna.org/listinfo/metro-developers
