• A friendly and periodic reminder of the rules we use for fostering high SNR and quality conversation and interaction at Stormtrack: Forum rules

    P.S. - Nothing specific happened to prompt this message! No one is in trouble, there are no flame wars in effect, nor any inappropriate conversation ongoing. This is being posted sitewide as a casual refresher.

Chat GPT forecasting checklist

Joined
Jun 17, 2017
Messages
76
Location
Sherwood, Arkansas (Little Rock area)
I’d like your opinions on this forecast process. It’s what chatgpt responded with when I was looking for a detailed guide of a forecast process.

There’s great educational information on StormTrack, YouTube, SPC website, etc. When sitting down with a blank sheet of paper and opening up a computer to begin making a forecast, I wanted an organized way to apply everything I’ve learned.

What do the chasers here think about what chatgpt came up with and what alterations would you make that are things you’ve found to be important in the forecast process and what order do you do those things in?

Here’s the chatgpt version:



Here’s a chronological checklist for making a tornado forecast, organized by the order in which you would typically analyze each parameter, along with guidance on how each parameter informs subsequent analysis:

1. Synoptic Scale Patterns:
• Begin by analyzing large-scale weather patterns, such as troughs, ridges, and frontal boundaries, which can provide insights into areas of potential atmospheric instability and lift conducive to severe weather.
2. Surface Observations:
• Check current surface observations for temperature, dew point, wind speed, and wind direction to assess the current state of the atmosphere. High moisture content (e.g., dew points above 60°F) and directional wind shear are favorable for severe storms.
3. Instability Parameters (CAPE):
• Evaluate Convective Available Potential Energy (CAPE) to gauge the amount of energy available for thunderstorm development. Higher CAPE values indicate greater instability and a higher potential for severe storms, including tornadoes.
4. Shear Parameters (Wind Shear and Helicity):
• Assess wind shear and helicity to evaluate the atmospheric conditions favorable for storm organization and rotation. High values of wind shear, both in the low-levels and deep-layer, along with significant helicity, increase the likelihood of supercell thunderstorms capable of producing tornadoes.
5. Lift Mechanisms (Triggering Mechanisms):
• Identify potential lift mechanisms, such as boundaries, outflow boundaries, orographic lifting, and convergence zones, which can initiate storm development in areas with favorable instability and shear.
6. Mesoscale Features (Boundary Interactions, Drylines, etc.):
• Analyze mesoscale features, such as boundaries (cold fronts, warm fronts, drylines), to identify areas of enhanced convergence and potential storm initiation or intensification.
7. Satellite and Radar Data:
• Utilize satellite imagery to observe cloud patterns, convective initiation, and storm development. Additionally, monitor radar data for evolving storm structures, including signs of rotation, supercells, and mesocyclones.
8. Model Guidance and Short-Term Forecast Updates:
• Consult various weather forecast models, such as numerical weather prediction models (e.g., NAM, GFS, HRRR), for updated guidance on storm evolution and potential tornado threats. Pay close attention to high-resolution models for short-term forecasts.
9. Real-Time Observations and Reports:
• Continuously monitor real-time observations, reports from storm spotters, and official tornado watches and warnings issued by the National Weather Service for critical updates and verification of forecasted conditions.

By following this chronological checklist and carefully analyzing each parameter in the forecast process, you can effectively assess the potential for tornado formation and make informed decisions to mitigate risks and protect lives and property.
 
Some days I look at a lot of things and on others I’ll only look at a few to get pointed in the right direction. Before the day of a given chase I’ll usually have a pretty good idea how things are laying out, absent what transpired during the overnight hours.
My personal MVP of data is how much moisture is there, where is it and is there enough? I’m really happy with dew points in the fifties in Eastern Colorado, but really push for mid sixties in Oklahoma. I’m of the opinion that if you have an abundance of moisture it doesn’t take a whole lot to get something going.
If the moisture is doable, then it’s on to wind shear. I like to see at least 30 knots in the mid and upper levels to vent things off, but personally attach a lot of value in the lower levels. I’ve seen robust lower level winds seemingly save the day when dealing with mediocre mid levels.
Diving into the mesoscale features next is my favorite part. I can tie in today with what happened overnight and start picking out things like outflow boundaries and dryline bulges while comparing those to what I see on the satellite. Is cloud cover going to be a problem somewhere? Is there a two boundaries intersection bonus out there?
I may already have picked out a target by this point, but if I haven’t it’s onto things like laspse rates, helicity etc…I’ll dig into how ever much I need to to settle disputes in my head. I click on the Skew-T charts for an easy look at things like if the winds are veering or backed and the hodo,but prefer models for a bigger picture view of what’s going on.
I neglected to mention, probably with a few other things, that I run composite reflectivity loops to see if models even have storms going up. I try to simplify things so I don’t get to paralyzed by over analyzing. I realized it could help one time about eight years ago. I went down for breakfast at a hotel and noticed there was a group of four other chasers surrounded by their laptops going through all the data they could get their hands on. I finished eating and they were still at it. After going to my room and getting ready, I went downstairs to leave and they were still pouring over the details. This was one of those no brainer days and they were killing themselves. I don’t know how much longer they spent at it, but I was parked off the road waiting later in the day and they pulled up right next to me. It turned out to be a good day, but got the feeling they weren’t enjoying chasing as much as they could be by over analyzing.
 
Overall that is a pretty good list. I do have a few comments. Be skeptical of statements like "parameter X above Y"( for example "dew points above 60°F"). Not a bad guideline, but Mother Nature loves to show us there are exceptions to every rule. Often something can compensate for something else. Likewise there is a lot of nuance to sheer. Cameron Nixon's online presentations on Hodos provide a lot of info in that area.
 
Commenting just on how good a job ChatGPT did in putting together that list, without typing up my own process - I think it’s a decent start but misses a lot of nuance. Some random comments -

- It would have been helpful if it provided threshold values for key parameters. It did so only for dewpoint. Randy’s point is absolutely correct that no individual parameter threshold should be considered a binary criteria. My point here is just a critique of ChatGPT in being less than comprehensive.
- It fails to address *when* the forecast step is performed and makes no distinction between forecasting and nowcasting. For example radar and satellite are useful when doing the initial assessment of large scale patterns, and again later in the day closer to initiation time. Many mesoscale features (step #6) may not take shape until later in the day.
- It lists model guidance as step #8, but models are a tool for evaluating parameters in many of the other steps. For example, step #4 implies an analysis of current shear, whereas it’s the shear later in the day that matters as a mid-level jet noses in or the LLJ gets cranking. Models would help assess that. Similar to my comment above, another indication that time considerations are not really factored into ChatGPT’s output - it’s important to see how parameters are now, *and* how they will evolve later. I don’t think it adequately makes this distinction.
- It fails to specifically mention the different levels of the atmosphere (personally, I like to start at 500mb for a preliminary big picture view, then check upper levels and then go back down to 700 and 850).
- There’s no mention of assessing cap strength

Overall the ChatGPT output comes across as somewhat of a hybrid between a checklist of ingredients and a forecasting workflow. As such, I don’t think it’s really does a great job with either!
 
ChatGPT's output basically looks like it was ripped out of meteorological textbooks that cover forecasting severe weather. It mentions all useful things to consider and goes into little or no detail into each of them. I'd expect nothing more from a text regurgitator machine.

That checklist is already stuff that all seasoned forecasters and chasers already implicitly do. Many are able to do it in very little time.
 
Back
Top