You may notice that this post is coming at an unusual time. That’s because it’s discussing an unusual topic. Instead of looking at the weather for the day ahead, I wanted to take a look back at yesterday’s forecast. That’s because, to be perfectly honest, it was a bad forecast. After forecast ‘busts’ like this, I take a step back and evaluate what went wrong and why. What could I have seen that would’ve led me to a better forecast? What could I do better next time? This process of self reflection is key to becoming a better forecaster, and I wanted to be a little more transparent about it this time around.
What was expected?
In my morning update here, I called for a chance of afternoon thunderstorms that would be strong to severe, “including damaging winds, heavy rain, and lightning”.
#MEwx/#NHwx: Comparing KPWM to KCON soundings forecast for this afternoon is interesting. More ML dry air KCON but both cities have moist low levels/low LCL’s. Hodograph comparison shows more low level turning at PWM N of warm frontal boundary. A tornado is definitely psbl today! pic.twitter.com/nUvduNOMid
— Jack Sillin | weather.us (@JackSillin) May 20, 2019
Later in the morning, I posted this more technical discussion on twitter mentioning the potential for tornadic activity in addition to the damaging wind threat.
I wasn’t the only one concerned about the potential for severe weather. Despite having their hands full with the potential for a historic tornado outbreak in their backyard, the Storm Prediction Center in Norman Oklahoma issued a discussion for the region indicating that they also expected severe storms. About an hour later, they issued a severe thunderstorm watch which means that conditions were ripe for the development of severe thunderstorms.
Here’s what one of our forecast models was thinking the radar might look like yesterday evening (6PM). Big lines of severe storms were expected to be moving into both Portland and Waterville. Other storms were forecast to develop in the mountains and farther south in New Hampshire. It should be noted that these models aren’t the only tools we use to forecast the weather, nor are they accurate enough to pin down the exact towns that might be hit by storms several hours down the line. That said, the forecast models, the Storm Prediction Center, and I were all expecting big thunderstorms to develop yesterday afternoon.
What Did Happen?
Here’s what the radar looked like last night at 6 PM (compare to the model forecast map for the same time above). There were indeed some thunderstorms up near Waterville, but the vast majority of us saw just light showers, if anything. We knew that not everyone was going to get a big storm, but we sure as heck didn’t think that the coverage and intensity of storms would be that low. Only the cell near Waterville produced any lightning, and no storms produced any reports of wind damage or hail.
Why The Discrepancy?
The short answer is that not many storms actually formed, and those that did never ended up developing robust updrafts (currents of air moving rapidly upward in a thunderstorm). So why did that happen? That answer is a bit more complicated, but I’ll try to dig into it a bit here.
Below is a comparison of two model forecasts versus the actual weather balloon observations over Gray at 2 PM yesterday. The bright red line is the actual temperature and the bright green line is the actual dew point. The purple lines represent the morning model forecast and the tan lines represent the previous night’s model forecast.
The first thing to note is that the model forecasts were too warm in the lowest half mile of the atmosphere. This was most obvious at the surface where the overnight forecast was too high by 8 degrees. As we know, warmer air is less dense and thus more buoyant, so if the model overestimated the temperature it also overestimated the instability available for thunderstorms to tap into.
The other key factor is the dew point (moisture) forecast in the low to mid levels (500m to 2500m above the surface). Here, the model was overestimating how much moisture the atmosphere would contain. It overestimated the dew point by almost 20 degrees at times. That’s a big difference! Low-level moisture is crucial for thunderstorms in two ways. First, clouds are made of condensed water vapor. If you don’t have water vapor (moisture) to condense, you can’t get any clouds, and thus no thunderstorms. Additionally, moisture is also one ingredient that contributes significantly to instability.
Mid level moisture is important in a different way. Even if you get an updraft to develop, fueled by low level moisture and instability, mid level dry air can still kill off a wannabe storm. Water droplets begin to forming from condensation due to rising/cooling will start to evaporate if the mid levels are dry enough. That evaporation changes the water from a liquid to a gas, which requires energy from the surrounding air. The air loses energy as a result of this process, which means it cools down. The new cold air is very dense, and begins to sink back towards the surface, which replaces the updraft with a downdraft, and shuts down the storm. The animation above demonstrates the consequences of this process. Note that many clouds are trying to bubble up, but they never punch high enough up into the atmosphere to become fully fledged thunderstorms.
Another contributing factor in the lack of thunderstorm development can be found in the comparison of forecast surface wind directions with actual observed surface wind directions (click to enlarge if you want a closer look) shown below.
In the model forecast, there was expected to be an area (red) of strong convergence between winds blowing in different directions (SW vs W in NY, S vs SW in VT/N NH). This is what would force air upwards so that thunderstorms could form. What did we actually see? The same red area is highlighted in pink on the observed wind map (right panel). Aside from a few stations that were dealing with terrain influences, winds were almost uniformly out of the WSW in the area we were looking for storms to develop. This meant that there was less convergence than anticipated, which resulted in less air moving upwards, which made it harder for storms to form especially given the inhibiting mid level dry air.
Could I Have Seen This Coming?
Here’s a look at the data I had when I was making my forecast around 8 AM. With the benefit of hindsight, I’m going to try to see what I missed that would’ve led me to the correct forecast of scattered showers with a low chance of a non-severe thunderstorm.
Here’s a comparison of the model forecasted wind speeds versus observed wind speeds at 8 am. Was the lack of convergence problem evident? The answer is no. The model did a good job forecasting the wind directions over the area of interest to our west, and there was no sign that the model was suffering from a systemic convergent bias in surface wind fields.
It’s also interesting to note that the model only forecast an area of sharp convergence to develop after several hours of daytime heating. Is there some problem with how the model deals with the boundary layer in daytime heating situations that can explain this mistake? I don’t know. If I did know, perhaps I would’ve been able to account for that bias like I do many others. If you do know, drop me a line either in the comments, via email (email@example.com), or on twitter @JackSillin.
Here’s a comparison of the 8 AM weather balloon from Gray versus the overnight model forecast data I was looking at. The model’s low level warm bias is evident in this data, but here the model is actually too dry. This is largely due to a band of showers that moved through in the morning that wasn’t captured by the model, but if I had attempted to correct for a moisture bias based on that information, I would’ve been left incorrectly thinking mid level moisture wouldn’t be that big of an issue.
Here’s a comparison of the upstream environment in Albany NY at 8 AM (forecast = purple vs observed = bright lines) which I think holds some clues that shouldn’t have been ignored. First, the model is too moist in the mid/upper levels around 500mb (6km). That would’ve been as good a sign as was available that there might be more dry air problems than the models were letting on. Additionally, there was sinking air (blue bars at the far left of the image = sinking air, red = rising air) in this layer, which would hint at further drying over the coming hours.
If you’ve read this far, wow nice job! This ended up being much longer than I had originally planned for, but hopefully it was an interesting insight into a tricky forecast that ended up not working out. Despite all the tools and expertise weather forecasters have access to, the atmosphere always has a trick or two up its sleeve at the end of the day. Taking a step back after particularly notable forecast mistakes to see what was missed and what might have been done differently is an important part of becoming a better forecaster.
I think the biggest takeaway for me is to pay more attention to the upstream environment. Albany’s morning air is often our afternoon air, and paying more close attention to that would likely lead to improved forecast performance going forward.
I’ll be back with regular updates starting tomorrow morning. Thanks for following along, and for supporting me despite the occasional blown call!