In the Department of Meteorology and Atmospheric Science at Penn State.

This is Weather world.

Good evening and welcome to Weather World.

On this Thursday, April 23rd.

I'm Marisa Ferger alongside May 2025 graduate Gavin Sandell.

He's back in town to be a guest forecaster this evening.

Gavin, you were last year in October.

Tell us what you've been up to.

Well, first of all, it's great to be back here.

And yeah, I'm a ten month veteran now of WFMJ TV, a TV station in Youngstown, Ohio.

If you're in Mercer or Lawrence counties, you might be a familiar face.

But of course, it's great to be back here.

Well, we're thrilled to have you here.

And what a day to be in Pennsylvania.

Yeah, really an A-plus day.

But we do have more unsettled weather on the way.

Rain chances are here for Saturday.

Okay, that's actually some good news.

And Gavin will tell you why in just a couple of moments.

Also tonight, Rob Lydick is going to spotlight the work that graduate student Christian Spallone did for his master's thesis.

But first, here's Gavin.

Thanks very much, Marissa.

If you were a teacher, you would certainly greet this day in a plus, because we've got lots of warmth and those blue skies on our Commonwealth cameras, all nine of them to lots of fair weather, cumulus clouds.

And we'll see in just a moment that these cumulus clouds could even spur up some rain showers, especially in southwestern Pennsylvania.

But we are coming from a couple of cool days in the early parts of this week, but it actually has been so warm during the spring, so warm, in fact, that Pittsburgh and Johnstown actually their warmest start to meteorological spring, that being March 1st to yesterday's date on record.

All the other places on our map except Erie.

The warm year of 2012 supersedes that, but there have been lots of warm days during the period of spring.

And as we go through the next couple of days, especially in southwestern Pennsylvania, that warmth will really stick around and sticking around.

Right now, lots of 70s, especially in the southern tier of Pennsylvania 60s, kind of hanging tough.

Mount Pocono, Scranton, both in the upper 60s, 66, obviously near that cooler lake shore in Erie.

But all of us still feeling like spring right there and lots of blue skies to hear those puffy cumulus clouds to our south and west.

And I'm going to zoom out now to my state of Ohio.

And you can see that little boundary setting up some rain and thunderstorms.

And again, that could be just a little bit of a presence, especially in far southwestern Pennsylvania.

Can't rule it out in the Pittsburgh metro if you are going to the NFL draft tonight.

But after that, we look towards the West and see this shield of rain and clouds, especially in Minnesota, going down to Iowa and the rest of the Midwest, as you.

That is going to start to progress towards our area.

By the time we get into the weekend, we'll have lots of rain by the time we get to Saturday.

But before that happens, look at all of this warmth.

We're expecting that warm, warm air and warm conditions to continue before we could see some 50s.

By the time we get to Saturday.

Now pay attention to this stationary front right here.

This is going to be a problem by time we get into tonight.

Maybe tomorrow morning as well.

Can't quite rule out the chance for some light pop up showers, especially in southwestern and maybe even central Pennsylvania when we wake up tomorrow morning.

But the big bad cold front starting to move through by the time we get into Friday evening and really starting to push into the Commonwealth, by time we get into Saturday, much cooler, especially if you are in eastern Pennsylvania, because we'll have winds coming off of the east, winds coming off of the cold Atlantic Ocean as well.

Get to that in a moment.

But you can see again, just a couple of showers from this stationary boundary in southwestern Pennsylvania, mainly everybody dry, but I couldn't say that we wouldn't have showers in central Pennsylvania.

Could have that as we wake up tomorrow morning.

But you can see lots of those pop up showers starting to blossom as we do have that heat coming through tomorrow.

Temperatures in the 60s and 70s for most of the area.

And here comes that cold front starting to move through.

By the time we get to Friday evening.

Clouds especially increasing in the western side of Pennsylvania, west of I 79 before that happens.

And then you could see lots of rain moving through by the time we get into Saturday.

And those easterly winds, especially east of I 99, moving through by time again Saturday, a very cloudy day if it isn't raining.

So now in terms of tonight's conditions, very calm.

Could have a couple of drops in southwestern Pennsylvania.

Temperatures there hang in the 50s, so it'll be a good night to go outside over there.

If you are in any scattered showers, increasing clouds in the northeast sector of the state.

Temperatures a little bit higher in the 40s.

Now, by the time we get into tomorrow, you can see a bit of a temperature gradient here, 80 near Pittsburgh and just 60 in northeastern Pennsylvania.

And again, just that unsettled nature, especially west of Interstate 81 temperatures feeling quite warm out there in the 60s, 70s and even 80s for most of us.

Now, by the time we get to Saturday again, you can see these winds coming out of the west and western Pennsylvania coming out of the east again, that ocean really limiting that warmth.

By the time we get to Saturday, especially in places west of I-79, we could see those temperatures going up just a couple of degrees before they finally fall down and rain will start to clear our area.

By the time we get through the rest of the morning and afternoon hours on a very cold Saturday, temperatures not going much up.

By the time we get into Sunday, we'll have a feature on grad student Christian Stallone coming right up.

And I think one of the most promising ventures is the ability to combine kind of the best of both worlds of these systems.

Graduate school presented Christian Stallone with a unique opportunity mixed together his fascination of forecasting with his programing prowess.

So Christian was an undergraduate student at Penn State, and he took an AI course in meteorology.

And through that, we found we both had an interest in seeing how we could use artificial intelligence to better predict the weather.

As I started to look into researching different graduate programs, I. I knew more about the different role and the increasingly important role that AI was playing within the atmospheric sciences.

But because it was so new, there was still a lot of uncertainty about it.

As Christian's research project came together, it became clear he would need to combine evaluation with creation to explore artificial intelligence methods for weather prediction.

It all started with understanding how AI models work.

In terms of a simple explanation, we can kind of describe the way that AI models make a prediction by essentially looking back at all the previous data that we know about the weather.

And so based on, say, 40 years of past weather information, if we can teach a model to learn all the different patterns and all of these different years and these different variables that exist through these reanalysis data sets, right, that reconstruct that history.

If we simply teach a model those relationships, we can allow it to be trained on that data.

And then as we apply it into the future, we can allow it to make predictions using some sort of just initial state of the atmosphere.

While these AI models can describe how the atmosphere will evolve with time, they are fundamentally different from the traditional models you're used to seeing on Weather World.

All of our traditional models utilize those initial conditions to make a prediction, but they do it by using all these physics based equations, right?

So really complicated numerical systems that take a really long time to run, but maybe very accurate.

So with the AI model it's doing is cutting out all of that interchange with these equations that take a while to run and simply using that initial state and then connecting it to past patterns that it's been trained on in order to extrapolate that, say, ten days into the future.

Right.

And so that's able to make a prediction in a much more, much more efficient time frame and sometimes a more accurate prediction as well.

The first part of Christian's research focused on the accuracy of these different AI systems for specific types of high impact weather, like winter storms.

Once I had a lot of these high impact winter events that were actually selected for a database, I can go about kind of analyzing the accuracy of these different models.

So in order to do that, we have to find some sort of actual ground truth data that represents what actually occurred in these different storms.

We usually use reanalysis data for that specific circumstance.

And then we can compare different forecasts from different modeling systems.

So these are things like the European model or the American GFS model.

Um two different AI based models.

So we can take a collection of these different systems and we can use those to see how accurate they are and predicting that specific reanalysis data set, for example.

And then that tells us how accurate they were in predicting that specific storm case.

Sticking with high impact weather, the second part of Christian's research involved thunderstorm forecasting.

But instead of evaluating existing AI models for accuracy, Christian created his own.

We had to process a lot of different satellite information radar data in order to teach a model to predict these specific storm cases in real time.

And then this involved a lot of training of the different systems, which is a complicated task.

We were able to train for convection this system in only a matter of a couple hours, which is really great.

So some models that are more advanced and more complicated, like a lot of the global models that may use like 40 years of data, some of those tend to take on a magnitude of days or even weeks to be able to train using GPUs.

We could run this model that we've developed in under a second, which is able to produce pretty accurate predictions and then rapidly update those on a real time basis.

Combining both parts of his research project, Christian was able to prove just how useful AI models can be.

We were able to show how a prediction system like this might be practically used some day for making decisions, for example, at a big sporting event, whether to take action as a thunderstorm might approach.

So it was great to see the practical applications of the research that we developed together.

The big question is how will these AI tools change the way we forecast weather?

So that's of course, the million dollar question, right?

And I think, uh, we have the traditional models, which are really good at getting all of these physics right in the systems.

And then we have these AI models that are very efficient and also have all of this past information embedded within them because they've been trained on it.

Right.

And so if you can somehow take all the good pieces of both of those and merge them together into some sort of hybrid system that could be really powerful.

And I think that's really the way forward in terms of the next 5 to 10 years in this field.

While Christian was working on his graduate degree, digging through data and making forecast models of his own, he also continued to be a forecaster on Weather World.

This is the first time I've worked with a graduate student that has also been involved with Weather World, and it's a great opportunity for our students to be involved with the day to day weather, as well as to increase their ability to effectively communicate the research to a variety of audiences.

It's hard to overstate how much Weather World has helped me.

I think just from a presentation standpoint, uh, looking back at like the early days of my undergraduate experience to thinking about now, like the way I've communicated information has completely changed.

And I'm sure some of that's driven by having more practice actually communicating scientific information.

But on Weather World, that's what I'm doing essentially every time I go on air.

Right?

And so I'm getting experience with that through that process.

And so it's helped me so much by being confident in a lot of the things I'm saying.

With a master's degree in meteorology and atmospheric science now in hand, Christian is embarking on a new chapter once again looking to combine his experiences with research, forecasting and communications.

It's been great to watch Christian grow in his ability to be an independent researcher, whether it's developing AI code, writing scientific papers, presenting and developing and testing his own research ideas.

I'd like to apply a lot of this knowledge that I've gained, kind of not only in artificial intelligence, but really within this applied space of meteorology somewhere into the private sector.

And he'll do just that, working behind the scenes with the innovative folks at the Weather Channel for Weather World, I'm Rob Lydick.

Beautiful shot of Pittsburgh, where the NFL draft first round is going to happen later on this evening, and the weather looks rather nice.

We could have just a couple of scattered showers, especially in far southwestern Pennsylvania.

Those temperatures quite mild for this time of year, just in the 50s.

Looking further east, a little more dry by over there, but we'll have temperatures in the 40s.

But again, that activity starting to ramp up tomorrow, as we could see, very warm in the western portions of the state, a little bit cooler, mainly dry there, but we could have some scattered showers start to pop up, pop up by the time we get to Friday.

Rain really overspreading the region overnight.

As you can see, much colder by the time we get into Saturday.

These temperatures probably going to go.

Only going to go up five degrees before they fall during the day on Saturday.

Lots of rain overspreading the region, but that's going to start to clear as we head through the rest of the day and the afternoon.

But again, as we go through the day on Sunday, still feeling much, much colder, certainly a change compared to the warm spring that we've had in Marissa again, that the two cities, Pittsburgh and Johnstown, it's been so warm.

I think that warmth is finally going to come to an end by the time we end off the weekend.

Well and the rain on Saturday.

Much needed, especially in parts of eastern Pennsylvania, southeastern Pennsylvania with the drought there.

Okay.

Well, thank you for joining us today.

We really appreciate seeing you again and having you do this.

Thank you so much.

And thank you for joining us.

Come back tomorrow night.

We'll have the Friday forecast frenzy with forecast going through the first full week of May.