Wishcasting versus Forecasting

I started talking about this last week on Facebook about the coming cold pattern coming to the Southeast. It’s also a favorable pattern for a storm to develop but as of right now there isn’t anything specific that points to one. Here in lies the problem. Too many people like to talk about potential as reality or worse yet talk about one model run of the deterministic model as something that is going to happen. I’ll get into what a deterministic model is versus ensemble in a second. First let address a fundamental aspect of being a good weather forecaster. Job #1 of any good and experienced forecaster is to forecast what is “most” likely to happen not what “could” happen. I see entirely too much time spent on the “could” forecast and not the most likely forecast. While it can be fun to talk about the what ifs it doesn’t make for a very good forecast.

Deterministic versus Ensemble:

The majority of the model data people post on-line and that you see is deterministic. Meaning it’s just one model solution or a single model forecast from the Numerical Weather Model. Although the “deterministic” approach has served us well, it is steeped in error and will NEVER provide a perfect forecast.  That’s right – NEVER – due to 4 primary reasons: (via WPC training manual)

Summary of primary sources of error using the deterministic method of modeling
1. Equations used by a model do not fully capture processes in the atmosphere
2. Model resolution is not sufficient to capture all features in the atmosphere
3. Initial observations are not available at every point in the atmosphere
4. The observational data can not be measured to an infinite degree of precision

When doing Ensemble forecasting you are using many model solutions and trying to eliminate error or the outliers and come to the most probable solution. We call this the Probability density Function. Which is illustrated below. You want to be forecasting in the peak of the curve not the edges.

So how does this apply to next week?:

A cold pattern is depicted in the ECMWF ensemble mean. Which has 51 members versus the GFS which has 22. The ECMWF mean 500mb anomalies shows a persistent trough or batch of cold air over the eastern U.S. all of next week.  Below is a graphical depiction of the mean 500 mb trough over the eastern 2/3rd of the U.S. Notice the below average heights over the eastern half the country. Plus notice poor California continues to stay bone dry.


Now this means the likelihood of cold air is high but and with a trough it’s logically to think we may see a few storms form in the base of the trough. Maybe even finally a gulf low that rides up the east coast. Right now though there is nothing that points to a specific storm. Unless you go crazy and just look at the ECMWF deterministic output for next weekend.


Right now every snow lover looking at this is like, YEAH!!!!!! and with the cold pattern in the Ensemble forecast one could be excused for getting excited for snow. Small problem though when looking at the ensembles for this storm they tell a far different story. Below is a graphically output for Charlotte, NC of the 51 members that make up the ECMWF ESP.


Notice that only 3 members out of the 51 have significant snowfall. Then at the bottom of the chart you can see the Deterministic snowfall(blue) versus Ensemble mean in green. So while there “could” be 4” of snow in Charlotte next week the probability is more like 0.3” and even that is a 7 day output which further diminishes the confidence in the outcome.

More reason to doubt a snow forecast this far out is that there is little backing from the GFS deterministic model output as well as it’s 22 ensemble members.


So what’s the forecast?

Based on using the most probable forecasting technique plus my experience as a forecaster. It’s easy to say it will be very cold next week and that we need to watch next weekend for a possible storm. Though right now there is no storm to talk about because the likelihood of no storm is about 9 times higher than the likelihood of a storm at this forecast range. Once the confidence in the storm grows to about 30% then we can start talking.

These are all things to take into account when seeing people just post model output without explaining it or using their own experience to make an actual forecast. Forecasting the weather is a skill acquired through time and with experience. This skill is built on the back of many failures which all makes you a better forecaster over time.

I’m an unapologetic snow lover myself, but as much as I love snow I like getting the forecast right more!

  • linklike

    Love this article!

    • wxbrad

      Thank you!

  • Mike

    So you’re saying theres a chance? =}

  • Edward Butler

    Fantastic job on this article Brad!

  • Pingback: Deterministic and Ensemble Forecasts - WeatherConnector()

  • Laura Todd Cooper

    Wonderful explanation….even if I did not understand it all!!!! 🙂

  • guinsmike

    I read this whole article, and here’s what I think you’re trying to say: We’re almost certainly going to get a huge snow storm next weekend, and you’re really happy about that because you love snow. But at the same time, you are very sensitive to disappointment, having been let down before, so you want to maintain cautiously optimistic while keeping your expectations low. That way, it’ll be like a big surprise when it definitely happens! 😉

  • sezwhom

    I know this is from last year but it’s still spot on with regard to a Forecaster #1 priority. I know other forecasters who constantly say “chance” or “could” all the time. I always go with “most likely”! Never base a forecast on the GFS past seven days. Our private joke is that the GFS equals “Good For S&%#”. The resolution has gotten better but the GEM and ECMWF are usually much more consistent than the bi-polar, if you will, GFS.