The Art of the Game

2016-04-09 11:26:55.000 – Michael Wessler, Summit Intern


While this post is in no means an exhaustive expose on weather forecasting, I hope it will provide some insight into where your forecasts are coming from, and provoke some thought as to where you get your weather. With the advance of technology, and most notably the prominence of the smartphone and an endless library of weather apps, how we get our weather has changed drastically. No longer do we find ourselves huddled around the radio waiting for our region’s forecast to cycle through on the weather band, or rush to the TV and flip to The Weather Channel for Local on the 8’s. Weather is at our fingertips whenever we want it, and for wherever we want it. While this change is overwhelmingly a good thing, it has made it so that just about anybody can produce a forecast and gather a large following, with or without the appropriate credentials.

While modern numerical forecast models such as the Global Forecast System (GFS), North American Mesoscale Model (NAM) and the ‘Euro’ (ECMWF) Model make the task of forecasting much easier than it would be without them, it’s important to understand their limitations. While models can produce a good starting point for a forecast, a trained human forecaster is still required to process this information and produce a quality product. Model Output Statistics (MOS) are products that help relate deterministic model output to specific stations based on climatology and statistical relationships, and often improve the model output, though still fall short of a “great” or even “good” forecast at times. The most noteworthy limitation with these numerical forecast models is that the further out the forecast, the less accurate the forecast is apt to be. Forecasts change frequently, and the accuracy of those forecasts will increase greatly with each update cycle.

Perhaps the greatest underlying issue is that the forecast models are only as good as the data they ingest. While the density of surface observations in the modern day is pretty good, especially along the east coast of the US, the weather balloons that take observations of upper-air conditions are still widely spaced and infrequent (balloons launch twice a day). Aircraft and data from remote sensing equipment such as satellites help improve coverage, though it is far from perfect. Models must then interpolate data to fill these gaps. The entire forecast is then calculated based on these initial conditions, and small assumptions made when filling these gaps may propagate into large errors in the forecast as early as 12 to 24 hours out. In short, if we had perfect understanding of the physics that govern the atmosphere and could also measure the state of every molecule in the atmosphere at a given moment, we could theoretically have a model calculate the perfect forecast. Obviously, achieving such a feat is practically impossible, and those are only a few pieces of a much larger puzzle. So what can we do?

Forecasters are responsible for assessing the current conditions, gathering information and guidance from numerical model output, and adding experience, empirical knowledge, and familiarity with nuances within their area of responsibility to produce an accurate and high quality forecast. For instance, mountain weather is complex and local phenomena occur on scales much smaller than operational models are able to resolve. In addition, terrain is ‘smoothed over’ in these numerical models, often reducing the overall height of the mountains, making forecasting things like rain versus snow, patchy or intermittent fog, and temperatures incredibly challenging for the models. It takes a human eye and experience to assess the model output for your area and translate it to something both useful and meaningful for the public. One of the most valuable additions a forecaster can make that the models cannot is the communication of forecast uncertainty in a discussion. Taking the time to read a forecast discussion in addition to glancing at the numbers will often give the consumer a much better picture of what to expect for weather in the coming days and how it may deviate from the posted numbers.

In an attempt to automate the process and provide an up to the minute point forecast for just about anywhere, on demand, some forecast sources have taken to developing ways to have computers interpret the model output. NOAA’s National Digital Forecast Database is able to downscale model output, extending what is essentially a clickable grid to the end user. However, realizing the limitations of such a product, regional forecast offices are constantly tweaking these numbers based on their own interpretation of current conditions and model output. The end result is a forecast product that is of high quality, but still not perfect, especially in areas of complex terrain like the White Mountains and the Presidentials.

Lesser forecasts are out there and often circulated through social media. These often take the raw model data and post them as-is, often advertising spectacular and improbable snowfall totals, striking fear or excitement across the internet. In other cases, some sources suggest they have developed a method to downscale the operational model data in a highly simplified manner, and produce accurate forecasts for specific locations and elevations. Often times, this is oversimplified, and may produce forecasts that are much less accurate than the raw model output and model output statistics! When looking at forecasts, it is always helpful to source your forecast from local forecasters with experience in that area, and consider the spread in the numbers among forecasts made by different individuals. If a forecast circulating on social media or elsewhere seems to be in left field with numbers much different than the consensus, it probably is worth taking with a large grain of salt.

For more on challenges forecasting winter weather, check out this post from the National Weather Service forecast office in State College, PA.

Forecasts and forecast discussions by the Mount Washington Observatory for the Higher Summits and Mount Washington Valley are produced twice daily and can be found on our website and in PDF format here.


Michael Wessler, Summit Intern

Overview of Lapse Rate Research

May 20th, 2024|0 Comments

Overview of Lapse Rate Research By Karl Philippoff As a weather observer and research specialist on top of Mount Washington, in addition to my usual observer duties such as taking hourly observations, releasing forecasts,

Deadline Driven: The 12-Hour Shifts that Power Weather Forecasting from the Northeast’s Highest Peak

May 9th, 2024|Comments Off on Deadline Driven: The 12-Hour Shifts that Power Weather Forecasting from the Northeast’s Highest Peak

Deadline Driven: The 12-Hour Shifts that Power Weather Forecasting from the Northeast's Highest Peak By Wendy Almeida  As a new member of the Mount Washington Observatory team, I wanted to gain a deeper understanding

Find Older Posts