Recently, the popular weather app Dark Sky launched a website version of its platform, and at first glance, it didn’t look much different than the mobile version. There are some advanced deep-dive features, but the main draw is still the gorgeous maps and hyper-local forecasts. Even the company’s cofounder, Adam Grossman, admits that its function is partially just to direct more people to the app store. “It’s a lower barrier to entry for using our platform than downloading a $4 app,” he says.
What’s most exciting about the site is not what you see in your browser. It’s what is going on behind-the-scenes at Dark Sky, which is experimenting with machine learning and UX that could radically alter how we view and understand the weather.
Companies like the Weather Channel have long been the butt of jokes about the inaccuracy and sensationalism of weather forecasting; when the anticipated enormous snowstorm turns out to be a light dusting, they’re the first to take the blame. But nature is unruly, and even in today’s data-driven world it’s impossible to predict it with 100% accuracy. In that way, communicating the weather forecast is a design problem: The way these predictions are presented influences the way they are understood by viewers.
Weather design typically falls into one of two camps: the ultra-minimal, embodied by Apple’s iOS weather app, or the overstimulating, sensational reporting by sites like weather.com. Dark Sky is unique not just in terms of its lovely, balanced aesthetic, but also in its dedication to presenting both the certainties and uncertainties of the forecast. It does this through big data analysis and digestible infographics that show, for example, the likelihood of rain in granular, 10- minute increments.
Being able to sift through and analyze big data is a relatively new concept, and Dark Sky is joined by other tech researchers–like this team from Microsoft–who want to use data and machine learning to make weather forecasting better. But Dark Sky’s focus on design and presentation make it stand out. The team views the website as a place for experimentation, since it’s an easier platform for developers to play around with than an app. DarkSky.net replaces Forecast.io, the company’s first foray into visualizing weather data, which Grossman and his cofounder, Jay LaPorte, launched three and a half years ago.
No one at Dark Sky is a meteorologist. Grossman studied physics and the rest of their team are web developers and backend engineers. They have a massive amount of government-provided, raw meteorological data at their fingertips, and increasingly, the skills to synthesize it. When they launched the Dark Sky app in 2011, the functionality was limited to showing the rain forecast only for the next hour. Still, as John Pavlus wrote for Co.Design that year, the app’s compelling visualizations and GPS-based location accuracy turned “the whole user experience of weather forecasting on its head.” Rather than a vague prediction, the app gave a more data-focused look about what was known and what wasn’t. While this approach introduces more uncertainty on the user’s end, it’s also refreshingly honest and informative.
Since then, the Dark Sky app has developed into a full-fledged weather app, and the new website displays even more information. Still, Dark Sky manages to maintain the middle ground between overloading the user with unhelpful data and cutting out helpful information for the sake of good design. For example, perhaps the most fascinating part of the new site from a UX standpoint is the feature that lets you zoom in on the map into a specific place–the Grand Canyon, for example–and get a detailed look into how weather data varies from the high altitude of the hills to the deepest valley. These are called microclimate adjustments, and it’s the technology that Grossman is most excited about–because it hints at how weather forecasting could function in the future.
Microclimate adjustments are based on a number of factors. Dark Sky has access to high-resolution data that shows the geophysical parameters at every point on Earth. It shows elevation, the steepness of slopes, the distance to the nearest water, and where the heat island effect alters temperatures. It also gets raw data from the various weather observing stations across the U.S. as well as national and local computer models that gather weather data.
This isn’t necessarily more data than what the meteorologist from your local TV channel has access to, but Grossman and his team have a secret weapon. They’ve developed a method of using machine learning to compare these data sets and automatically make adjustments to things like temperature, dew point, wind speed, and more at very specific locations within a single area. For example, it would adjust your forecast if it noted that you were on a lakefront, as opposed a half a mile inland, using the same basic raw weather data–combined with geophysical modeling.
Dark Sky is still working on developing these microclimate adjustments, but using machine learning could have exciting implications for how weather is displayed and understood. Human beings will never be able to predict the weather with 100% accuracy, but the power of big data allows us to look at a particular scenario from new angles, often uncovering insights previously hidden. Teaching a computer to analyze the data allows analysis to happen that much faster. The combination of cutting-edge AI and clever interface and infographic design could make weather, if not easier to predict, at least easier to understand.
For Grossman, the most exciting use of these capabilities would be offering a level of transparency to the accuracy (or inaccuracy) of forecasting. “The biggest thing that’s missing from weather forecasts is a sense of error and uncertainty,” he says. Dark Sky has always shown the percentage chance of its forecasts, rain or shine, which can be frustrating even as it’s enlightening. But by continuing to reframe the UX of weather forecasting, Dark Sky could eventually develop a way for people to understand why there’s a percentage of uncertainty.
The question, says Grossman, is “how do you present alternate possible histories? We have this data, so a big part of our back-end effort is how you display it to people.” The website might be mostly a front, but in the back end, it’s just the beginning.