Forecasters, not tellers – Hudson Valley Weather
The previous website post isn’t the best if you’re looking for the latest forecasts, this is just the late night ramblings of a slightly tired, caffeinated weatherman living in the muddy fourth quarter of winter. Keep going if this is your cup of tea!
Weather forecasting is indeed an amazing science but it’s not without drama! When I see comments like “paid for error”, “hype for views”, “prediction was better 40 years ago” etc.
I’m just laughing.
Anyone would think that we stay awake for countless hours, devoting all this time away from family, friends and other obligations, and care deeply about just miles from where the lines on a snow map should be. While Also Don’t Sleep About Being More Accurate and Humble than any other resource out there, that’s just a mistake.
Keyboard Warriors, Armchair Steps, Monday Mornings, Google Scholars.
No sweat. There are also people who believe in flat Earth, so this is not surprising. Too many keystrokes have already been spent on these people, who are not being held in HVW against their will. There are plenty of multi-million dollar companies that are not located in the Hudson Valley that people can get weather forecasts from, and you are free to move on.
So let’s address the elephant in the room, or maybe it was expected to be an elephant and now it’s a mouse? Either way, there is something in the room to discuss. What happened??
Forecast modeling is actually very, very good when used as a tool for what it is. Knowing which tool to use, when, in what time frame, and what types of details are important. Knowing each model’s biases, strengths and weaknesses also adds to forecasters’ ability to interpret the data, which again is just a guideline. It rarely tells you exactly what will happen in all locations with complete accuracy. It helps identify patterns, and data is fed into the models from ground data, satellites, weather balloon data and even data probes dropped from Air Force reconnaissance flights. These sources of information are fed into models that in turn create their own output scenarios. Overall, these models produce very good results, especially in less than 3 days, and are actually light years ahead of anything that was available to forecasters even two years ago.
These same models have predicted hurricanes, blizzards and severe weather within more than a week. Based on portions of the atmospheric energy that remains over the Pacific Ocean and in some cases has not even developed into anything resembling a storm. However, they are doing it right, lives are being saved, travel decisions are being carefully controlled and no one is ignoring it. Sometimes, models get things wrong, and there’s nothing your favorite forecaster can do but bend the knee. As forecasters, we are in fact bound by the data provided by these models, and none of the forecasters appreciate this relationship in any way. Those who understand meteorology also understand the intimate relationship between forecaster and forecast models. Is this a crutch? Breakdown in weather forecasts? Absolutely not – there is simply no one who can better extrapolate millions of data points, every few hours, and generate a probability forecast at any single point, about 2,000 miles away with better accuracy, not now and not 40 years ago.
If you think this is not true, then your perception is deceiving you, the reality is the media cycle, access to information, and the sheer volume of predictions and ways to obtain them are light years in terms of quantity and coverage from 20 years ago. You didn’t know you missed a storm because you didn’t know it was coming at all. You’d think things were more accurate because forecasts out 4 days later weren’t published as often, and the drama of the news cycle wasn’t the main revenue stream. The prediction is no less accurate, you just pay more attention, because now 1″-3″ is exciting and 40 years ago 1″-3″ was not an inconvenience.
The reality is that dramatic swings of this magnitude in a major storm are very rare. While models can often be off by a few inches, or miss a change in timing, or type of precipitation, very large errors within 12 hours of a storm aren’t that common, but they do happen. The data for this storm always had some read marks which lowered our confidence, and Bill spends countless hours creating videos to explain this to those who care to know, and most of these people were not shocked by this result.
- The air mass before this storm was very warm, actually registering warm.
- The cold air filtering behind the front was mild at best, not an arctic air mass as required for the best snowmakers.
- The rain/snow line has always depended on the exact locations of the frontal boundary, not far enough south, and sleet south of 84. Farther south, heavy snow moves out of the area.
- The models always had a sharp gradient cut on the northern edge of the storm.
- The lack of high-latitude blocking and a shallow trough does not mean it will be a fast mover, and progressive systems tend not to deliver and not fully bloom until far east.
- The data was very inconsistent, with the axis of best snow shifting from south to north about 50-100 miles twice in a 36-48 hour time frame.
These were all red flags that had us already on our toes ahead of initial expectations. Unfortunately, it’s called a forecast for a reason, you can’t wait for better conditions. You should create forecasts based on the best information available and 36 hours later we release the primer. It’s a balance between accuracy, but also gives adequate notice of upcoming life and trading weather to make decisions, or at least take them into consideration. Within 24 hours, we always try to release final forecasts, and we have always adhered to the standard of releasing two maps and only two maps (preliminary and final). We do this to keep things consistent, so you don’t all need to carry what your apps do and most forecasts, change every 6-12 hours with each model run (most animated weather apps/forecasts are tied to a single model). When data throws us signals, we do our best to express uncertainty or complexity in forecasts, and we know that most of them are transparent. The idea here is to prepare for all possible outcomes rather than going down the boat with poor expectations. A good example of this is that we now have hundreds of comments from people saying that we are the only ones who lowered expectations. Schools are closing, businesses are losing revenue, and people are canceling things that matter to them, all based on outdated information that is no longer supported.
Everyone can have an opinion about how the weather should or should be predicted, or how this type of information should or should not be communicated. I promise we’ve thought about all of this already. We don’t exaggerate, we don’t take changing final forecasts lightly, we don’t compromise, we don’t sensationalize, and we don’t bombard you with irrelevant news or information. For the weather. We don’t have multi-million dollar budgets, or millions in advertising revenue, we all have other jobs to support our families and we do it out of our love of science, this community and the challenge of trying to predict the most powerful and unpredictable force in this. The entire planet.
These projections can change very quickly due to the release of latent heat from distant convection that can alter the upper level extending thousands of miles upstream in a delicate balance of thermodynamics, and the sea surface temperature that is all affected on a larger scale by several different ocean and Antarctic temperatures. North. Sea ice coverage, some of you think these predictions are inside fortune cookies and that computers aren’t smart. Only kids with homework should throw tantrums when the snow doesn’t come to fruition. Adults are allowed to be disappointed, because there’s something primal about the run-up to a storm, the preparation, the anticipation, the vocal dampening of a passing car when it’s snowing. Others will be cheerful and rejoicing, because they may be on a fixed income and can’t afford a plow, or just had hip surgery, or lost a loved one in an ice accident, or can’t afford heating. We all experience weather differently, but we can also experience it maturely and with a greater appreciation and understanding of science and those trying to muster it.
Below are images of the two most accurate short-range models, NAM and HRRR. Both images show data 12 hours apart and less than 24 hours after the start of the storm. This helps keep in mind how quickly and important this data is evolving. The best part is that there will still be surprises, as models can’t handle a mid-range setup well. This band could turn northwest and boom above current forecasts, or it could continue to shift southward and less snow could fall. Our predictions can be found in the previous post which is where we decided to plant our flag, if we waited any longer it would no longer be a prediction, but a narrative. Remember, even if the data and therefore the forecast changes, it happened before the storm started, and that’s what we call a forecast! ??
We are forecasters, not storytellers!
Stay tuned for more!