Monday, November 18, 2013

Cloud Warming

Valley-level temperatures in and around Fairbanks are being held above the precipice of radiative cooling (so to speak) by low clouds, as seen in the infrared satellite image below: relatively dark shading in parts of the southern and western interior indicates relatively warm temperatures at the top of low clouds.  In northern areas, a mix of high and low clouds is also keeping temperatures from dropping as low as they otherwise would; the cold temperatures (white shading) in the satellite image show areas of cold high cloud in the northeast, not surface-level temperatures.  However, skies are generally clear in the southern Yukon Territory, and there temperatures have dropped to the -20s F and colder.

The low-level clouds have destroyed yesterday morning's surface-based inversion in Fairbanks, as shown in the two soundings below, and so Fairbanks airport has warmed by 15 °F despite continued cooling aloft.  When the clouds break up in the next day or two, the inversion will return and valley-level temperatures will drop to their lowest levels of the season.


  1. If you compare various weather venues they all have different forecasts. For instance, Weather Underground is 10 -15 F below NOAA. I wonder if the biggest differences between the various models is how they handle clouds.

  2. Eric,

    I'm sure others can speak more knowledgeably about this, but I suspect most of the online weather forecasts for Fairbanks are created solely from computer output with no human intervention - and therefore will err badly even for the short-range in some cases. The NWS forecasts include expert subjective adjustments from the local professionals, so are likely to perform much better in these more "remote" parts of the country.

  3. Regardless of the model and methods I've given up on generic Internet forecasts for the most part, and rely on the local forecasters. It's fun to look out weeks and play what and why, but given their recent quality of predictions the local staff are to be monitored first I believe.

    Clouds are an important WX modifier. Knowing that, I wonder how well the various WX models derive and handle their presence, altitude, and degree of cover?

    What was was, and is is. What's going to happen...well that's another matter. Not sure if tea leaves, thrown bones, or asking my dog is the best source.


  4. I agree, Gary. Weather prediction is a lot more art than usually thought. The weather models have become quite good but the human touch is still needed.
    I once heard, perhaps antedotely, that any forecast past 7 days is as good as the tea leaves and asking a dog.

    1. One thing I as an outsider to the forecast community have often wondered...are post event reanalyses employed with some sort of a feedback loop(s) (not sure of the correct word) to improve forecasting?

      Something like 'we thought this would happen some days out, but the models mutated due to some known or unknown cause. Maybe next time we should throw some salt over the other shoulder, or make a different call.' Often that's experience exercised through the human touch, but it would seem there should be a reanalysis of the models if the actual play out differs by some designated magnitude.

      I see it (the maybe factors and oops!) verbally discussed to some degree on the Forecast Discussions which I follow ( These to me are the real work of the forecasters art in Alaska.


    2. Gary,

      Efforts are made on both a subjective and objective level to improve forecasts using experience. Good forecasters can become quite skilled at detecting biases in the models or unrealistic forecasts, and there are also many ways of automatically removing bias from models or "calibrating" the output. Unfortunately (in my view) model calibration is not performed as often as it ought to be, so there is still plenty of room for improvement in modern forecasting practice.

      If you're up for some reading, the GFS Reforecast project is a nice example of statistical postprocessing of the model forecasts.

    3. Thank you Richard for yet another interesting link. I scanned the page some of the products and will read further tomorrow.

      I suppose it's easy to assume if interest, $, and computing power are available via Supercomputers then post processing and refinement to Model outputs would be feasible.

      Is it possible we may be reaching the point of data saturation for forecasters? I'm not sure when but at some point may there not be automated forecasts versus human chosen products? That may already be happening perhaps.

      Think about someday unmanned drones collecting and calibrating data feeding computers who publish a forecast for certain aspects of the weather product.


    4. Gary,

      Without a doubt, the process of automation will continue and eventually replace human-generated forecast content. Already we see that many meteorologists either function as computer programmers to develop the algorithms that turn the computer forecasts into a useful product, or as expert advisers to specific needs and industries. There will always be a useful role for human interpretation of what the computers are telling us - and for some, the human element is just as important as the forecast itself.