This week, the final curtains are closing on one of my absolute favorite online resources for weather data, Dark Sky. A startup originally founded to provide high-resolution rain forecasts for customers like airlines and event coordinators, Dark Sky had created a number of tools that were readily available for other app developers and weather enthusiasts. The most remarkable of these, in my opinion, was a worldwide map interface that would display highly detailed colormaps of temperature, humidity, precipitation, wind, and other meteorological variables. Their data accounted for topography more rigorously than any other resource I had seen, adjusting temperature and pressure for elevation differences, even estimating wind channeling effects over complex terrain. Individual point values within these datasets were made available for use in other programs through the Dark Sky API. After the company was bought by Apple Weather in 2020, these free resources were no longer in the interests of the ownership and were gradually phased out. While the map interface has long since disappeared from their site, the Dark Sky API has lasted until its scheduled termination date of March 31st, after which developers must update their code to another source because this data will disappear for good.
I have some history with Dark Sky: in late 2019/early 2020, I was very interested in collaborating with (or even working for) the company, having several calls with team members over the course of a few months. I saw their terrain heatmap as an ideal base layer for my tornado prediction model and thought that my tornado risk data (instantaneous and eventually aggregate) could be beautifully displayed with the same mapping interface. I exchanged some code and sample data with a project manager. In our conversations, the developers behind Dark Sky repeatedly gave me the disclaimer that their algorithms were not grounded in meteorology or physics, emphasizing that they were simply programmers using purely numerical methods. As I interviewed for the title of Data Engineer, I proposed running some of the same analyses on forecast validation that I had applied to my tornado probability model, knowing that it takes years of testing and documentation to be accepted by meteorologists as a dependable predictive tool. Ultimately our conversations wound down as Dark Sky finalized their sale to Apple, a deal that immediately constrained what these guys could share or work on outside of their core products.
It’s a little saddening to look back on what could have been – instead of merging my spatial predictive algorithms for severe weather with the sophisticated mapping and testing interface of Dark Sky, I began working in Texas shortly thereafter and all but abandoned my efforts on the project. I am grateful for their interest in my work, and I completely understand their decision to take the multimillion-dollar opportunity with Apple. After all, a corporate buyout is the end goal for most tech startups, and private weather analytics is an increasingly lucrative industry valued at over 17 billion dollars. However, as the market becomes saturated with products from the big players like Apple, Accuweather, Baron, and IBM (parent of The Weather Company), it has become more challenging for individual developers to break through without IP challenges. Coupled with the lobbying efforts that have weakened NOAA in favor of public-private partnerships, I worry that this general trend will have a negative impact on the amount of scientifically useful information that will be accessible for independent developers and scientists going forward.