With better weather information, farmers can plan their farm related activities in advance, fishermen won’t venture out unawares and get caught in sea storm, airline pilots will know which route to take to avoid turbulence, and shopkeepers would know when to stock up extra rations. The accuracy of the forecasts has always been the key to all this. IBM owned The Weather Company is counting on their new forecasting systems that can even predict something as small as thunderstorms. In a telephonic interaction with Mint, Kevin Petty, Director of Science and Forecasting and Head of Public-Private Partnerships, IBM and Himanshu Goyal, India Business Leader, The Weather Company elaborate on their position in India, recent partnerships and the new forecasting system.
Q. The Weather Company recently tied up with SGT University. What is the nature of this tie up and what is the company trying to achieve with it?
Himanshu Goyal: There is a lot of weather-related situational analysis, from a research point of view in terms of how it affects health, engineering work, or agriculture, and various other things. In order to build on the research capability, we collaborated with SGT University which has a research team within their university, where they encourage students to do a lot of this point of view research based on available data sets.
With our association, this particular lab will focus on getting the data and getting the students to triangulate all the data sets in these verticals and work with internal researchers and faculties. The idea is to encourage such partnerships and get more research activity in the country, so that we can pull out some noteworthy analysis and empirical formulas on how these things coordinate and look together.
Q. How is The Weather Company doing in India?
Himanshu Goyal: We have been in India for two and half years. In the B2B segment, we use the data and solutions for clients in logistics industry, media, agriculture and various other verticals. In the B2C segment, we have the Weather Company app, which is consumed well in India. So now being part of IBM, we are exposed to product capabilities internally and broader market externally. India is a very important market for us and we see making value additions in our clients’ business.
Q. How do you ensure the weather predictions stay accurate in case of unpredictable weather events like slow moving cyclones?
Kevin Petty: We are working to improve forecasting and prediction in a number of different ways. We are working to release a new global high resolution atmosphere forecast (GRAF). The numerical models that are being driven by mathematical equations simulate what the atmosphere is going to do in the future. There are multiple global models that are running right now. But those models are running at 10 kilometers to 15 km resolution, and you're getting updates, maybe four times a day. We are pushing the envelope when it comes to science and innovation in the modelling domain when we release the GRAF model, which will be running at 3 km resolution per scale. That is really the key for the ability to pick up things like high impact events like tropical cyclones and thunderstorms, which are having a significant impact on the end user.
Q. How are you taking advantage of AI (artificial intelligence) to improve your solution and the user experience?
Kevin Petty: Weather itself is universal. However, the way we respond to it is very personal and might be slightly different. So we're really looking at how we invest AI and ML into facilities that will specifically target end users requirements, be it someone responsible for airport operations, or running a farm, or operating an electric utility. By using things like AI, we're coupling weather data with operational data, so we can better assess the impact of weather on each of these operations.
Q. How has weather prediction changed over the years? What are the new parameters that are being used to measure forecast?
Kevin Petty: The advancements that have been seen are in the areas of observations. So if we look at observations, they have grown significantly over the last few decades. In order to predict what's going to happen in the future, you have to know what's happening right now. So observations are fundamental. Second to that is computational power. If we go back a few decades, the computational power just wasn't there to be able to run the computer models in a way that we could run them, for example, at the resolution that I was just describing down to 3km. So now, with the computational power, that utilization of cloud architecture along with taking advantage of things like GPU (graphical processing units), we are able to fully recognize and put into play some of the power available on the computation side. These computational resources have allowed us to apply machine learning on these very large data sets.
Finally, in numerical weather prediction model itself, we continue to see a lot of advancements and a lot of them have come through community involvement and partnerships.
Q. How do you ensure the weather forecasts are accessible in times and areas when there is no internet connectivity?
Kevin Petty: Yes, that is a really challenging problem. Which is why we are trying to take advantage of every available type of communication protocol or capability that's out there, whether that be directly through internet or cell phone connections, or satellite connections. I don't think we're going to necessarily solve, all the last mile types of issues tomorrow, but it's something that we have to do. And this is where, it is important to truly partner and make sure that no one group is trying to do all of these things itself. So we are looking to partner with individuals who are responsible for last mile connectivity, so that we can make sure the products that we develop, truly can reach into the hands of the end users.
Q. How can government trying to connect and help people in cyclone like situations take advantage of platforms like The Weather Company?
Kevin Petty: When we talk about serving the general public or other businesses, we are in a much better position to do that if we partner the government or companies. Each one of these sectors have various unique skills and capabilities and also limited resources. And a great example is the fact that the Weather Company, the app that we have developed to provide a concrete dissemination platform where we can get information out to the public quickly. So if we're able to partner with government and the government is producing different types of alerts, we can leverage our platform to get those government alerts out to the public. So the government doesn't necessarily need to invest in putting its resources on the development of the app. Now it can utilize its resources in a much more effective way and target those things that it's really focus on. It has actually played out very well, in other parts of the world, such as the US, where NOAA’s National Weather Service has looked to the private sector, to really be that part of the machine that is positioned to disseminate key weather and then also addressed specific verticals, such as the aviation domain, or energy or other verticals, where the National Weather Service just doesn't have the resources to do that.
Q. The performance of a machine learning model depends on the amount of data it is being fed. Often the data that might be more relevant may not be available. How is Weather Company dealing with the data challenge?
Kevin Petty: In the meteorological domain there's a considerable amount of data that is going on unused. We are sending large amounts of logical data that we just haven't taken advantage. So I think there's plenty of opportunity there, right now that that we can take to move forward. Where we really need to focus is on how to bring the logical data together with other illogical datasets, because that's where you're, you're going to create value. So in some instances, depending on the application that you're trying to develop, you might have a shortage of the type of data that you might be looking for. But you can still develop a model and get it in place, knowing that it might not be optimized just yet to the level of accuracy that you're looking for. But the thing about machine learning is this aspect of continual learning. So every day, we're collecting more data, the model learns and continues to get better.