AI Weather Forecasting for a Resilient World: Salient Predictions

Dunja Jovanovic: Could you share a bit more about yourself and what Karma Metrix is all about?

Fabio Mecarone: My academic background is in economy management. After graduating, I ventured into marketing analysis, eventually leading to the digital sector. There, I joined the initial team behind what would become Karma Metrix. Initially, we were focused on SEO services, but a serendipitous inquiry from one of our clients about the environmental impact of our SEO practices prompted a significant shift in our direction. We hadn't previously considered the environmental impact, and this question led us to research and discover that digital activities indeed have a tangible ecological footprint. This discovery was pivotal, and it led to the foundation of Karma Metrix, which is dedicated to enhancing digital sustainability. We've developed an innovative algorithm that measures the CO2 emissions generated by websites, aiming to help companies understand and mitigate their digital environmental impact.

DJ: How do digital activities impact the environment and why it's important?

FM: The digital environment is often misleadingly perceived as 'clean' because it's not directly emitting smoke or pollutants like traditional industries. However, the infrastructure that powers the digital world—data centers, servers, and the networks that connect them—consumes a significant amount of energy. If the digital world were classified as a country, it would rank as the fourth largest in terms of CO2 emissions and third for energy consumption. Most of this energy still comes from non-renewable sources like fossil fuels, which are responsible for a large portion of global CO2 emissions. Our work at Karma Metrix aims to shed light on this issue and provide solutions that help reduce the digital carbon footprint.

DJ: What specific technologies and methodologies does Karma Metrix employ to assess and improve the eco-friendliness of digital products?

FM: Our primary tool is an algorithm we developed to estimate the carbon footprint of websites. It analyzes various factors, such as the server types, the energy sources, and the efficiency of the website’s design and operational code. For instance, we've found that websites with darker color schemes consume less energy on user devices than those with bright, white backgrounds. We also evaluate the efficiency of the website's backend processes, like CSS and JavaScript implementations, and the size and format of images, which can significantly affect the site’s energy consumption. We provide detailed reports and recommendations to help companies optimize their websites for lower carbon emissions.

 

DJ: How do companies generally respond to the concept of digital sustainability? Are they proactive in seeking solutions?

FM: It varies. While there is growing awareness about environmental sustainability, digital sustainability is still a relatively new concept for many. We often find ourselves in the role of educators, introducing companies to the idea that their digital operations can significantly impact the environment. Once companies understand the scope of their digital footprint, they are usually very interested in finding ways to reduce it. We’ve seen a positive shift towards more sustainable practices once the impacts are clearly understood and actionable strategies are provided.

DJ: What has been the most surprising or challenging aspect of your work at Karma Metrix?

FM: One of the most surprising findings was the significant impact of AI and big data on water consumption. It's well-known that AI requires substantial computational power, but the associated water use for cooling these massive data centers was startling. For example, the water footprint for training some advanced AI models can rival that of small cities or even nuclear reactors. This highlights a less-discussed aspect of digital sustainability: the vast amounts of resources required beyond just energy.

DJ: If you could change one thing about global sustainability practices with a magical wand, what would it be?

FM: I would immediately transition all energy production to renewable sources. This would drastically reduce global emissions and slow the pace of climate change. Although this wouldn’t solve all environmental issues, such as those related to resource consumption and waste management, it would address one of the largest contributors to global warming and environmental degradation.

Hello, friends, you are watching a brand new episode of the Green New Perspective Podcast. Your go-to place when you want to learn more about innovative tech aimed at combating climate change. In today's episode, my guest is Laura Fieselman from the company called Salient Predictions.
And Laura is going to tell us how accurate weather predictions can help businesses to stay resilient in the future. If you want to learn more, stay tuned and enjoy!
Hi, Laura, and welcome to the Green New Perspective Podcast.
Hi, Dunja, thanks, it's great to be here.
For starters, can you briefly introduce yourself to our audience, and tell us a bit more about your role and interest in climate solutions?
I'm Laura Fieselman, I lead operations here at Salient, which means that I hold finance, people, business operations, commercial operations, non-dilutive funding, marketing, all company admin, all that to say it's never a dull day. And I come to this work initially from the world of sustainability. I started my career as a sustainability officer, back when you could do that just out of college.
And I spent about a decade opening and managing sustainability offices for colleges and universities. And then I pivoted to mission driven startup world in 2018. I've always been in COO and Chief of Staff type roles and mission driven startups are my jam.
I've always cared about sustainability and climate and I'm delighted to continue to be in this space.
Can you give me a brief overview of the company and what climate related problems are you trying to solve?
Salient is a weather forecasting company that's focused on the S2S time horizon. And let me tell you what S2S means for the listeners that aren't familiar. It's seasonal to sub-seasonal weather forecasting.
So about a week to a year ahead. And everyone always says, OK, Salient, interesting. But how do you forecast the weather nine months ahead?
And the answer is that the short-term weather, like in the next day to two weeks, is generally governed by the atmosphere and in the longer term, weather is governed by the ocean because oceans hold the bulk of both the heat and the water on our planet. What we are doing is taking a whole bunch of ocean data and also land data together with the atmospheric data of shorter-term weather forecasts and feeding all of that through a machine learning engine to spit out S2S, or Seasonal Subseasonal Weather Forecasts. And I'll brag on my team a little bit.
I can do that because I'm not on the R&D side of the house. But Salient's got the most accurate, reliable S2S weather forecast commercially available today. Generally, our accuracy is 10-plus percent over the next best available model.
Season, photography, lead time, etc. And just to briefly orient you and the listeners to the product, we're forecasting temperature, precipitation, wind and solar irradiance as our core variables, plus a whole bunch of derived variables things like growing degree days, heating degree days, relative humidity, soil moisture, etc. One week to a year ahead, anywhere that there's land.
That's what we do.
What challenges are you trying to solve with your tech?
The core Salient is helping businesses profit and organizations create positive impact in the world by using advanced long-range weather forecasts and analytics. And Dunja, I'll say that our current core verticals where we work are energy, agriculture and insurance. There's a lot of other potential verticals where good weather forecasts far in advance are helpful.
Things like retail, shipping, defense, etc. There's so many. But right now, we are focused on energy, ag and insurance.
And there are a handful of use cases inside each of those verticals. I'll give you just some examples to orient you and the listeners. Inside the world of energy, we've got power and gas traders using our forecasts.
We've got hydropower hedging, utility planning, everything from vegetation management ahead of wildfires to putting storm crews in place ahead of storms. Demand forecasting, renewable supply forecasting. So many corners of the grid are weather dependent at this point.
In the world of agriculture, we've got commodity traders using our forecasts and supply chain planners. And then in insurance, use cases include things like pricing but also customer alerting. So yeah, a wide variety of different ways in which Salient is helping to solve challenges in the business world.
Can you tell me more about the complexities of weather systems and how changes in one area can affect conditions globally?
Let me briefly tell you the company's origin story, which is that our co-founder, Dr. Ray Schmidt, who's an oceanographer by training, originally had this hypothesis that we can correlate ocean salinity to precipitation, because the ocean is salty or abnormal. Water has evaporated. That's going to come up somewhere else.
So he's at the tail end of his science career, but he had his sons build a machine learning model of things, salinity, precipitation, and they entered the US Bureau of Reclamation Forecasting Rodeo with their model, delivered a precip forecast week on week for a year, won in the precipitation category, took the prize money, eventually hired the person that won in the temperature category, and started the company. That's the heart of the answer to your question, which is how and why does ocean over here affect weather over there?
How salience tech helps different sectors to manage the risks associated with unpredictable weather patterns?
There are so many applications. Let me just give you a couple examples. One, the Bill and Melinda Gates Foundation is working with Salient to make our S2S weather forecasts available to smallholder farmers in East Africa, who are making all kinds of decisions like what to plant, when to plant, which variety to plant, when to harvest.
Those are really difficult decisions in an environment where agriculture is primarily rain-fed. Or let's take the Texas freeze as an example of 2021, when winter storm Uri knocked out so much of the grid in Texas. Salient caught a signal on that event four weeks ahead of time and think about helping to solve that kind of challenge with a bunch more lead time, very helpful.
And then another example that I can give you where we're helping companies handle natural disasters is our work with Zurich Santander in Brazil, where big rain events create mudslides which negatively affect their customers and a lot of other people, of course. And Zurich Santander is working with Salient to put together alerts for their commercial and industrial customers ahead of those major rain events to prevent disaster and losses.
Can you share a success story on how you helped a client to make a significant difference, like disaster preparedness or resource management?
All of our customers are managing weather-dependent enterprises in some form or another. And basically, what our product does is either help them make money or lose less money based on what happens with the weather. To go a level deeper there, Salient is creating probabilistic weather forecasts that inform decision-making.
And when I say probabilistic weather forecasts, what I mean is that we're not saying in July the average will be X. We're saying here's the range of potential outcomes for temperatures in July, for example. And making decisions off a probabilistic range of outcomes is a difficult thing to do and requires a sophisticated decision maker.
And so Dunja, I was thinking about how to help your listeners understand the core. How does one do that? And the cost loss framework is a tool that comes to mind.
I'll briefly explain it. It's a pretty basic two by two, but it's a powerful one where what we're thinking about is across the top of the matrix, does the weather event occur or not? And across the down the side of the matrix, was the preventative action taken or not taken?
And to use a cost loss matrix, what you do is you figure out the cost of taking a preventative action, the cost of a loss if the weather event occurs without the action, the probability of the weather event, and then you do some math. And what a matrix like this ultimately spits out is the threshold probability above which it makes sense to take a preventative action. That's the ratio of cost to the loss, cost loss ratio.
And yeah, this is the critical question that decision makers have to make. Is the cost of taking an action less than the expected loss from the event, given the event's probability? So quickly, you know, it sounds simple.
We make probabilistic weather forecasts, but making decisions off those forecasts isn't easy. So, yeah, requires decision making models and frameworks. And this is just one.
I'm happy to point you and your listeners, if you're able to add show notes to a written post, where we explain this in a lot more detail. Another client that I can speak about publicly is Salient's work with Anheuser-Busch or ABN Dev, who makes so much of our beer. And the question, well, basically, beer is made from barley.
And in a wet year, barley quality degrades. And in a dry year, barley quantity degrades. And so the sort of deep dive on what Salient and ABN Dev have been doing together is looking at how, you know, longer range weather forecasts up to a year in advance can help with barley supply chain sourcing.
So maybe you know this, maybe you don't, but weather actually causes more variability in crop production than anything else, not soil, not land, not farmer skill, not plant genetics. The weather accounts for 40 to 80 percent of crop yield variability. And in this day, when the range of weather outcomes is far greater than the historical range of possibilities, just relying on historical data is less and less helpful.
So what we did was run a pilot with ABNBEV's North American Agronomy Group with attention to procurement and trading as well to look at volume, quality and risk as pieces of the puzzle. And I'll just give you some examples of what's possible in a pilot like this. So in 2021, there was a major drought in the Midwest where fields just didn't have enough moisture to support barley and oat crops.
And Salient caught that signal in August of the prior year. And then again, in 2023 in the spring, we caught the signal on a below normal precipitation summer, two to five months ahead. And those kind of predictions are really valuable if we can turn them into business value.
So let me just speak briefly to how we take a forecast like that and translate it into business value. What that looks like is the procurement team or the group at AB and Bev that buys grain, accounting for decreased yields in their planning, contracting for more grain, shifting regional allocations. That's things like the Agronomy Group or the group that advises farmers, advising farmers to buy less nitrogen to improve grain quality and acceptance rate.
And it's things like the Trading Group or the group that hedges prices in advance to be able to anticipate price and supply risks and improve their hedges and forward contracting. But Dunja, I want to flag it's not just all about the economics. When we factor in the community implications of farmer incomes and the ecosystem implications of the environment, projects like these are win-win-win collaborations.
I would like to steer this conversation in a bit of a different direction at this point. So I know that Salient is a fully remote company. And New Perspective, a marketing agency sponsoring this podcast, is also a fully remote operating agency.
So I would like to ask you, how do you feel in that kind of environment? And how do you manage collaboration and maintain team cohesion?
We can't talk about remote culture without talking about the pandemic. And at the top of the pandemic, Salient was about a five-person team. Everybody in Boston or Boston area.
And now we are about now being May 2024, we're about a 17-person team spread across the US and world. And I think we've got a pretty strong remote culture and I feel proud of it. And I would say, you know, there's the small things, but also the big things that add up to create, to build a team's culture.
And some of the small things that I think make a difference for us are we run synchronous hours for the company. So 11 a.m. to 4 p.m. Eastern are the hours in which we're all online. We host our internal meetings during that block and otherwise work whenever you want.
So that gets all of us spread across a bunch of different time zones. I think we've got Croatia to Alaska as our time zone spread, all working for five hours a day together. Another thing we do is meeting free Wednesdays or at least internal meeting free Wednesdays, which unlock deep work blocks.
And you're probably familiar with Cal Newport's deep work frameworks, but it's one of my favorite and the whole team loves no meeting free Wednesdays. We get so much done. And small way that we build cultures, we've got recreational Slack channels.
I know a lot of companies have these. Some of our favorites are Animals and Kids, Food and Drink, My Ride, and our CTO, Carl Kritz, made a really fun Jeopardy game for our last offsite off these recreational channels. So we like to have fun like that.
In the camp of bigger things that we do is run company retreats with a lot of thoughtfulness and care. We have them about three times a year and they're always a mix of work and play. And when we get together, the team is problem solving, we're building relationships, we're having fun.
The agenda is a mix of formal sessions, co-working, walk and talks, adventures together. And we've had these retreats at a number of different locations so far. Boston, where the company is headquartered, Cape Cod, where our co-founder Ray did his academic work at the Woods Hole Oceanographic Institute.
We've been to the Rocky Mountains. A shout out to YMCA of the Rockies Snow Mountain Ranch for a really cool facility. And we're actually headed to Tulsa, Oklahoma next week.
We will have been already by the time the podcast comes out. We're in partnership with the University of Oklahoma's New Urology School. One of our investors is based there.
We have an employee working out of the Rose Rock Bridge Incubator. So yeah, we're really forward to that as a new retreat adventure for us.
And what do you feel are advantages of remote work? And where do you see challenges?
One of the obvious advantages is remote work enables us to recruit some of the best talent, no matter where people are in the world. Shout outs to my colleagues, Fran, Arulik and Victor, who are in Croatia and PhDs in Machine Learning and Data Science. They're amazing contributors to the team and we wouldn't be able to get them if we were a strictly in-person US-based company.
One of the challenges to remote work is for people who struggle with self-motivation or have a bunch of distractions at home, it can be hard to show up remotely. And then I think maybe the bigger challenge is supporting our early career team members and growing as early career professionals. To be someone just starting out in the professional world and working remotely, you got to really push yourself to build relationships, have mentors, attend events, listen, learn, read, have introductory coffees.
And as a manager, I try really hard to support my team members in doing that. And we do all across Salient. So that's, I think, a challenge that company leadership has to attend to.
But I'll end on a positive note and just say that, of course, the win of remote culture or remote work is that we get to have these really fun retreats. I really enjoy them.
Since a marketing agency is behind this podcast, of course, I have to ask some of the marketing related questions. So how do you educate your market on the importance of accurate long range forecast? And why is that important?
There's a few lines on which we have to educate. One is on the science of what we do. We are tackling a complex scientific challenge of weather forecasting a week to a year in advance.
And some of our customers care a lot about how we do that and want to know the details. And other customers don't care at all. They just say, give me the answer.
That's what weather forecast is going to be. And so we have to have a range of education for those varying levels of interest in science. And then similarly, we have to do a fair bit of education on the ROI of deploying good, reliable, accurate, S2S weather forecasts.
Because a forecast can be as accurate and reliable as humanly possible or as technically possible, but doesn't do any good if you don't make decisions and if a business doesn't get an ROI off the forecast. And so we work closely with our customers to quantify that ROI. It's not always easy to do.
We've got a technically complex product. I mentioned earlier, you know, we're providing probabilistic weather forecasts and those are opposed to deterministic forecasts. So again, not here's the temp average for July, but rather here's the range of potential outcomes.
Each forecast comes with a reliability diagram. You know, here's the reliability of salience forecast at this geography for this lead time. For those reliability.
Also, each forecast comes with a skill score. I'll briefly pull you into the world of skill score metrics by telling you that my team's favorite skill score metric is CRPSS or continuous ranked probability skill score, which is a measure of both accuracy and reliability at the same time. And every forecast comes with both that skill score and a suite of others.
All this to say, it's a technically complex product, and we have to really help our customers understand, depending on their level of sophistication and education, what all is in the product and how to use it. The other thing I'll say is that world of AI and weather is evolving really rapidly. There's a lot of new players.
Some of the science is strong. Much of it is somewhat lacking in the field, and it can be hard for customers to parse the difference between all the players and the range of science, and so we have to work at that with our customers.
Can we talk more about how you build long-term relationships with clients and partners to advance climate resilience?
Most of our clients are large multinationals, not all of them, but many of them, and an S2S weather forecast is relevant for a lot of different business units across the company. There's an ROI to be had off of S2S weather forecasts across a range of different use cases, and we get excited about helping a bunch of different corners of the business. But what we have to do is stay focused, start small, build a relationship within one business unit on a team, improve value, and then we can expand.
Our conversation is slowly coming to an end, and one of my last questions for you is, how do you see, what's your long-term vision for weather forecasting?
At the heart of our vision is that the private sector, the public sector, and society at large will all benefit from the improving resilience to a climate that's increasingly volatile. And again, our vision is not just excellent forecasts, but empowering decision makers across those sectors, public and private, to translate effective, excellent forecasts into decisions that drive business value and or create impact.
And what are some of the exciting innovations and advancements in forecasting tech?
Again, this world of AI and weather forecasting is evolving really rapidly. Let me just point you and the listeners to some things that are changing. One is that the skill of forecast on this S2S, seasonal, sub-seasonal, one-week, one-year time horizon is evolving really rapidly.
And the value of superior S2S forecasting is becoming really clear to decision-makers who have been relying just on climatology, historical averages. Another thing that's changing where this industry is going is that AI-based S2S forecasting is becoming a really important tool for sophisticated weather data users. And the meteorologists who are overly skeptical of AI and S2S are unfortunately falling behind their more forward-thinking counterparts.
And then the other thing that's changing in the business world is that decision-making models that have taken deterministic weather forecasts are being refactored into decision-making models that can accept probabilistic forecasts as this evolution towards probabilistic weather forecasting happens. And then beyond the business world, government agencies are increasing their commitment to AI and S2S forecasting. We see this both in the US and Europe and beyond, and partnerships between government and private sector are becoming core to progress in this space.
And then the other thing we see is the scientific research community continuing to push the envelope on what's possible in this space. And then, so sort of that's industry overview altogether. I'll say close to home, what Salient is able to do is collaborate really closely with our customers to hear what geographies are most important, what lead times are most important, what seasons are most important.
And then we focus our model skill improvements against those customer priorities. So we love getting to do that and we'll continue doing it.
My last question for you is pretty much the same one that I ask all of my guests here in the podcast. So can you share some of the resources for people who are listening and want to join the climate community? Resources like Slack channels, Discord channels, different kind of groups or webpages.
And also can you share your social media links, social media of Salient so they can people learn more about weather predictions and why that matters?
Yeah, sure. A couple resources in the climate community that I often point people to, they've been really valuable for me, are the two Slack communities. One, work on climate.
And two, new energy nexus. Both have tens of thousands of people in them and very robust conversations across all kinds of channels. I'm sure you get those recommendations a lot.
To get in touch with Salient, you can book a call through our website or a demo directly. And I would say follow us on LinkedIn. It's the social channel where we're most active.
And of course, please connect with me directly on LinkedIn. I'd love to be connected to more people in this space.
Thank you for watching another Green New Perspective episode and sticking with us for the second season of our podcast. As before, this episode is proudly sponsored by New Perspective, a Boston-based marketing agency working with cleantech clients only. So if you want to learn how our sponsor is helping cleantech businesses to grow, click out the link in the description of this episode and find out a lot more on their website and on their social media.
And if you want to support Green New Perspective podcast, subscribe to our podcast on all the streaming platforms. We are everywhere and we love you and we love your feedback. So thank you once again for joining this podcast episode and hopefully I'll see you in the next one.
Bye.