Forecasting of weather began in the early 1860’s, and the world's first televised weather forecasts were broadcast by the British Broadcasting Corporation in November 1936. Since then, TV Weather people grace our favorite channels 365 days a year on all the networks. While we all like to have an indication of what’s the weather will be tomorrow, the accuracy of individual forecasts are not comprehensively measured, nor do professional weather people admit they simply got it wrong.
Indeed, weather and forecasting of it never stops. As failed predictions on the weather materialize, a new forecast quickly takes precedence. Does it matter? Unless your daughter’s outdoor wedding last weekend got rained out when a forecast by weather-guy ‘Steve’ said 24 hours earlier it would be perfectly clear, you probably don’t give it much thought. Unlike my job, and yours, performance measures are required, analyzed and a factor for pay increases and continued employment. Forecasting weather, on the other hand, has no accountability. TV personalities can be ‘wrong as wrain’, and it doesn’t matter. How about tomorrow?
Add to this is the corporate claim factor. Every local network’s weather team can’t be the best and most accurate; but they say that in one context or another, and no way to officially certify claims. The forecast now calls for an approaching cold front.
While there are a few organizations that quantify this challenging data and publish on the web (kudos to them), a new generation of analytics is coming that will track outcomes at a very data-centric, localized, and scientific level. Enter artificial intelligence or as commonly called, AI.
As computable weather objects (standardized metrics) are collected, characterized, classified, and aggregated, the algorithms will unambiguously report. One example of change in this domain is Google’s AI arm called DeepMind. This system called “Nowcasting” can predict the chance of rain within the next couple of hours with much higher precision than what we have now. It will be matter of time before the time horizon of DeepMind extends 24 hours or more. This will revolutionize not only forecast accuracy but also methods to identify weaknesses in local geographies.
TV weather people across the spectrum are in for a significant ‘whether event’ that will shine light on those who claim success, but the data will show whether a forecaster was right, wrong or somewhere in the middle. We do know this; the performance of Punxsutawney Phil is 39%.
Clearly, the television audience will continue to prefer a person talking about the weather versus a robot, but new metrics for accuracy will determine which weather forecasting team is in fact, the most accurate. You can hang your hat on that.