Insights Into the Accuracy of Weather Forecasts for Major US Cities

A century ago, daily weather forecasts were largely based on the experience each individual forecaster had accumulated over several years of practice, vaguely formulated rules of thumb, and poorly understood associations. Predictions were riddled with errors.

The 1900s brought major advancements in the science of weather and later widespread access to satellites and weather station data to aid in measurement. The 2010s yielded advancements in AI and ML technologies to produce weather forecasts based on decades of historical daily observations. And yet, errors remain and the cost to infrastructure, economies, and lives has only increased. The cost of extreme weather events, which in the United States alone amounts to hundreds of billions a year in lost revenue, underscores the price weather can impose.

Those in the weather business know forecasts become less reliable the further out you go from the date in question. But access to homogenous and statistically consistent weather data empowers organizations to create actionable business intelligence. To put this in context, below we consider a single day: July 4, US Independence Day. This is a federal holiday when many in the US have the day off of work to celebrate, including travel, cookouts, parades, boating, fireworks, and more. Accurate weather forecasts are important to a travel and outdoor-centric holiday, particularly one in which cities across the country dedicate significant budgets to outdoor festivities.

Using actual and historical forecast data from Weather Source, we examined the accuracy of modern weather predictions for the top five largest US cities on July 4, 2020. The common schemas in Weather Source data make it easy to join and compare disparate datasets via a single pane of glass. Following are our findings:

• As expected, the further you go out into the forecast period the forecast becomes less accurate with respect to certain parameters such as cloud cover, wind speed & wind direction. A forecast is generally extremely accurate out 72 hours and is relatively accurate out 5–7 days. Interestingly, air temperature forecasts made 10 days before July 4 were less accurate than predictions based on the historical average.
• The accuracy of the weather forecast, on average, did not improve greatly even within seven days of the holiday. The most accurate precipitation estimates were available within 3 days of July 4.
• On the day of July 4, cloud coverage estimates, which are rather important to fireworks displays, were often found to be incorrect.

View original infographics, live dashboard and download data at knoema.com