's Resilience Platform is Here. Learn More.

Ruth Favela
By Ruth Favela
Ruth Favela
Ruth Favela
Ruth Favela is's AI Marketer. She draws on over 5 years of experience as an editor, writer, and social media manager for AI startups, B2B SaaS, and B2C products. In her role, Ruth focuses on using AI tools to create customer-first content for the various industries has solutions for. She writes about weather innovations, AI/ML modeling, weather API applications, weather AI use cases, and much more.
Oct 26, 2023· 11 min, 56 sec

From Data to Decisions: How’s Accurate Forecasting Drives Impact

    Weather impacts every business and individual. Yet most struggle to understand how they can leverage weather data to inform decision making in a business setting.

    We recently had the chance to chat with Tyler McCandless, Director of Data Science at, to find out more about the current innovations in weather forecasting.

    In Tyler’s own words, “We do a lot of machine learning and data science at” His team focuses on improving forecasts using ML in post-processing and work to create derivative products to help customers make better decisions.

    By combining high-performance computing, machine learning algorithms, and new data sources, the weather industry is providing more accurate and customizable forecasts.

    Businesses can now tailor weather insights to their unique risk profiles.

    Read the interview to learn how advanced weather technology empowers organizations to incorporate forecast probabilities into strategic planning. With the right solutions, weather uncertainty can become an advantage rather than a liability.

    And if you’d like to watch the conversation check it out below.

    What Recent Advances In Weather Modeling Are Enabling More Accurate Forecasts?

    “We’re at a really interesting intersection right now of advancements in high performance computing and machine learning and weather observations and in numerical weather prediction.

    So combining all those together, right now we’re at the forefront of trying to combine machine learning in systematic ways to improve the weather forecasts from better observational data sets going into our numerical weather prediction models, to using machine learning to predict in very short timescales with faster latency or faster predictions than with numeric weather prediction models to can deliver to outputting more accurate forecasts and probabilistic forecasts  across all space and time.”

    How Does Generate Forecasts and Leverage Data To Improve Accuracy? 

    At, the ability to generate forecasts and leverage data to improve accuracy is driven by a combination of the best physical models in numerical weather prediction and machine learning. Tyler explains that, “We really try to match the forecast challenge or opportunity with the right inputs and the right models. If we are predicting [weather] for the U.S., we’re gonna take the appropriate high resolution models that are capturing the weather phenomenons of interest. So we might use a model that’s at three kilometer resolution that’s convective, allowing us to understand the formation of thunderstorms.”

    The team is then leveraging ML and AI to generate more accurate forecasts and probabilities.

    “You can imagine if you’re an event company and you’re interested in the potential for a weather event—for example, a summertime thunderstorm— to impact your operations, we wanna make sure we’ve trained on observations.

    We’ve married that up with the best numerical weather prediction models, and then we’re generating most accurate forecasts with probabilities so that a customer can make a better decision if there’s a 30% likelihood of high winds or 30% likelihood of precip above a threshold, they can make those decisions.”

    How Does Tomorrow.Io Use ML To Enhance Weather Models? utilizes machine learning to enhance weather models through a post-processing technique that learns from historical forecast errors and observations to correct systematic model biases.

    Tyler explains, “The way numerical weather prediction works is it has physical equations that govern the atmosphere that have been derived over the last many decades of scientific advancements. These numerical weather prediction models do a very good job of capturing how the atmosphere is gonna currently evolve or evolve from its current state.

    But there’s always biases and there’s challenges with different weather phenomena. So what we do is we use machine learning to learn the patterns between how the numerical weather prediction model expects a storm system to evolve. The observations of what actually occurred historically.

    Then the machine learning model can tailor that there tends to be a warm bias in this area of the models, or it tends to under forecast precip in this season, in this area, and ultimately producing a forecast that has the lowest error that we’ve quantified over a long historical record of the models as well as the observations and generating the probabilities because as the forecast lead time increases, there’s definitely a greater likelihood of error growth. So we wanna generate those probabilities with machine learning while still using a high resolution numerical weather prediction model.”

    What a great overview! By leveraging ML algorithms in this post-processing role, can generate weather forecasts with lower error and customized probabilities, critical capabilities for providing actionable intelligence to customers.

    The Difference Between a Deterministic and a Probabilistic Forecast

    The difference between deterministic and probabilistic forecasts lies in how they characterize forecast uncertainty. Tyler defines the two below:

    Deterministic Forecast: A deterministic forecast is when we’re predicting the temperature to be 34 degrees at 8:00 AM on October 18th.

    Probabilistic Forecast: The probabilistic forecast is really getting at, ‘What’s that full likely outcome gonna be at 8:00 AM on October 18th?’ Making this a full distribution around that 34 degrees.

    Tyler continues by explaining, “There’s gonna be different uncertainty or different variability in different areas in different times of years and different locations. The probability that we’re generating is a full forecast pdf. We can quantify that in different ways. It could be the inner-quantile range, which is where 50% of the observations fall.

    And say that for that prediction of 34, we expect 50% of the time it to be between 33 and 35. And it’s an easy decision to make that more than likely it’s gonna be above freezing. But what happens if somebody cares about there being a 10% chance that it’s below 32 degrees?

    We’re gonna show that full distribution ’cause we’re outputting multiple realizations, what we call ensemble members of what the atmosphere is gonna be, and we can compute that.”

    By conveying the full spectrum of potential scenarios and their probabilities, probabilistic forecasts empower advanced risk analysis and data-driven decision making, advancing the value of what weather forecasting and weather intelligence is.

    How Does’s Forecast’s ‘Likely Range’ Empower Business Leaders To Account For Uncertainty When Making Decisions That Impact Their Businesses?

    “The likely range is a really nice quantification of that relative uncertainty.

    Referencing the example of predicting 34 degrees, you can see that range around it. And if you’re looking at a timeline for the next week and you’re a gardener and you’re trying to understand, ‘Am I gonna bring my plants inside or not?’ you could see where that range often falls outside of that 32 to degree threshold and give you a visual representation of you know what, this is a week where I’m gonna have to be covering my plants every night. Or, oh, the risk is fine. It’s mostly gonna be above, you know, 38 degrees and we don’t really need to take caution.

    So the likely range is a really nice way to get an understanding of the likely outcomes for a given forecast variable for a period of time.”

    How Do Customizable Probability Thresholds Allow Customers To Match Forecasts To Risk Tolerance?

    According to Tyler, “Instead of taking an approach of, we’re just going to predict certain quantiles or certain aspects of the distribution. Our machine learning based model does not have a set distribution beforehand. It allows the model to learn the relationships between what’s been observed. And the underlying numerical weather prediction model forecast.

    Therefore, we generate ensemble numbers, or in this case, 51 potential forecasts of temperature at every hour at every given location. So with that anybody can customize to within two percentage points of what the likely probability is gonna be.

    So, if there’s an airline who’s making a decision on deicing, they can say above 34% likelihood that the temperature gets below freezing – they’re gonna have to get deicing crews out. And then we can see how many of those 51 ensemble members are below the freezing mark and quantify the probability that way. So it’s an extremely high level of customizability rather than just outputting like the fifth and 95th percentile, or the 25th and 75th .

    That gives a good understanding of relative ranges, but here we’re allowing much more precise forecasts because we’re outputting each of those ensemble members.”

    Why Is Displaying Past Forecast Trends Valuable For Teams Who Are Assessing Reliability?

    Tyler tells us that viewing past forecast trends helps assess uncertainty in current predictions. “In meteorology, operational meteorologists like to say that, ‘The trend is your friend.’ So when numerical weather prediction models start trending in one direction, maybe it’s a snowstorm in the mid-Atlantic and the, the average snowfall in a given area was two inches, and then it became three inches and four inches.

    And the last model runs have been trending that way. Oftentimes, the numerical weather prediction models continue in that same direction. Not always the case, but it gives a good understanding of where that trend is likely to go. Somebody could say, oh, there’s probably a higher risk of this being a higher than normal snowfall event compared to what’s forecast.”

    Consistent model shifts in one direction indicate the forecast may change further that way. High volatility between model runs signals low confidence due to tricky weather patterns.

    “Another thing could be that there’s good volatility in the forecast and the models are not resolving it really well. It’s a really tricky pattern. There’s multiple storm systems that are nearby, and depending on how they expect ’em to evolve, it has a big impact on the forecast. And someone could see that in the past trend of, it seems like the forecast trending 10 degrees swings because of this uncertainty.”

    This trend insight allows users to double-check potential forecast changes based on historical reliability.

    How Do Spatial Probability Maps Help Users Visualize Potential Geographic Impacts?

    One thing that helps business and organization leaders with is visualizing geographic weather impacts through spatial probability and supporting them in understanding when weather systems will evolve.

    He uses NYC as an example, “If you think about a forecast for New York City and you’re looking at the specific point forecast for Newark airport. Seeing it is great. You can see a timeline, you can see what the forecast is, and maybe it shows that there’s a 25% chance of a weather event, a thunderstorm, impacting that area on that given day.

    But it doesn’t show how the storm system is likely to evolve, and maybe there’s a greater chance of storms to the north and less chance of storms to the south. Or maybe New York is just east of where the expected thunderstorms are supposed to impact, and that’s why there’s only a 25% chance. So when you look at a spatial map of those probabilities, you may gain understanding that there’s actually a 90% chance that Eastern Pennsylvania is gonna get thunderstorms, and it would only take a shift of tens to hundreds of miles for those thunderstorms to then impact New York.”

    This process allows for more spatial awareness and understanding, especially for high-impact weather events where they may be in relation to a specific point of interest that someone the platform has.

    How Can Probabilistic Forecasts and Forecast Data Improve Business Decision Making? 

    Different individuals and businesses have varying risk tolerances when it comes to weather.

    Tyler gives us a running example to better understand how this could improve decision making.

    “As an individual, I run every morning and my threshold for temperature for wearing a jacket is different than my wife’s and is different than my friends, right? For me, I look at the probabilities of a forecast being below a certain threshold where I would want that jacket and I make a different decision than somebody else would for high impact events.”

    By providing a comprehensive forecast probability, users can tailor their actions according to their unique risk thresholds. This personalization leads to better, more informed decisions, whether it’s choosing appropriate attire or making significant business choices.

    What Innovations Are You Excited About at 

    “We’re really on a literal rocket ship towards improving forecast capabilities and machine learning being foundational to that.” 

    With the launch of’s first two pathfinder satellites and the announcement of their preliminary satellite data, a lot is in the works to advance weather intelligence and weather modeling.

    Tyler tells us he’s excited about continuing to leverage machine learning as his team continues forward, “We’re really at an inflection point of bringing advanced R&D into operations across the spectrum from very short term. Nowcasting in the minutes to hours ahead and looking at that on the precipitation side globally, all the way out to improving forecasts and having probabilities out to 14 days —all of these fundamentally using machine learning and various components.

    And then of course, we’re in space and we’re gonna be even more in space, and that data’s gonna be extraordinarily valuable. Better data and allows for better quality models out improving the forecast capabilities. We’re really on a literal rocket ship towards improving forecast capabilities and machine learning being foundational to that.”

    What’s Next for Weather Intelligence?

    As Tyler illustrates, the future of weather forecasting will be driven by leveraging advances in high-performance computing, new data sources like satellites, and machine learning algorithms. These innovations enable more granular, accurate, and probabilistic modeling of the atmosphere and with’s growing satellite constellation, our data will support teams across sectors. For businesses, this means weather intelligence that quantifies uncertainties and forecast reliability.

    Armed with these data-driven insights, organizations can confidently incorporate weather risks into strategic planning. supports teams in customizing  forecasts to their unique risk tolerance and helping them leverage AI-backed insights to make smarter decisions.

    As modeling and data collection continues to improve, weather forecasts will become even more valuable tools for mitigating operational impacts. The future of weather intelligence promises actionable insights that give businesses a competitive edge.

    Ready to get started?

    See in Action!

    This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.


    Sign Up for the Weather API for Free

    Try Now