There is a massive amount of data that is collected to assist in forecasting the weather:

• Thousands of trained weather observers worldwide report  the details of the weather at least daily, as do thousands of airplane pilots and ships at sea.
• 900 weather balloons are launched everyday from all over the globe – 92 of them by the U.S. Government.
• Thousands of buoys have been deployed by a number of governments to collect weather data from the ocean surface.
• NASA’s new GOES-16 satellite, which launched in November, can scan the Earth five times faster and with better image resolution than previous satellites.
• The National Weather Service maintains 159 high-resolution DOPLAR weather radars
• The National Weather Service is a massive data collecting machine with 122 offices over the U.S. and its territories. The National Weather Service is part of the NOAA (National Oceanic and Atmospheric Administration) within the Department of Commerce. Note that the majority of the Department of Commerce’s budged goes to the NOAA, and most of the NOAA’s budget goes to the National Weather Service.
• Everyday weather information collected by the National Weather Service is about 2x the amount of all the information contained in the Library of Congress!

Weather forecasts have dramatically improved over time. Particularly the prediction of temperature. Notably:

• The 5 day out weather forecast in 2016 was as accurate as the one day out forecast in 2005.
• The forecast of the temperature 9 days out is now better than just looking at long-term climate averages
• The one-day temperature forecast is now accurate within about 2 – 2.5 degrees. This means if you see a forecast high of 80, most of the time the actual high will be between 78 and 82 degrees.
• The biggest improvements over the past 10 years have been in the accuracy of five-day forecasts.

Source: Forecast Watch. The red line depicts the least skillful provider each month and the blue line depicts the most skillful provider each month. The black line is the simple average of the monthly minimum and maximum. The dotted green line is a computer-generated linear trendline that best expresses the average plotline. “RMSE” is “root-mean square error”

• These improvements are huge. According to Nate Silver writing in the New York Times, “In 1972, the service’s high-temperature forecast missed by an average of six degrees when made three days in advance. Now it’s down to three degrees. More stunning, in 1940, the chance of an American being killed by lightning was about 1 in 400,000. Today it’s 1 in 11 million.”

• Huge advancements have also been made in forecasting hurricanes and their direction. With respect to forecasting landfall location, in 2010, the mean track error at five days was 327 miles. That was reduced to 225 miles for the error rate between 2012 to 2016. In 2017, the five-day track error was about 180 miles. Go back to the 1990s and the error was well over 350 miles. Every mile of accuracy counts as It’s been estimated that it costs $1 million per mile of coastline on average to evacuate before a hurricane.

Even given all the improvements, weather forecasting is still an inexact science. The primary problem is that many weather factors are exponential with respect to each other. That means small changes in one factor can have huge impacts on the weather. Thus, weather is still very hard to predict, even with all the data and sophisticated computer models.

A key takeaway: forecasts of the weather more than about 7-10 days out are not useful. Seasonal outlooks – like whether it will be relatively cold vs. warm winter are also not very helpful.