The University of Arizona

What Self-Driving Cars Can Teach Us About the Future of Weather Forecasting

February 25, 2014
Author Profile: 

The Consumer Electronics Show (CES) takes place in Las Vegas every year. This show often features the latest advancements in various forms of technology, from computers to televisions to smart phones and even robot pets. This year the CES unveiled technology that will undoubtedly change your life: BMW and Audi demonstrated their completely autonomous cars, and Ford showcased a communication system that allows cars to talk to each other. 

Some auto industry and technology experts suggest the first true self-driving cars will hit the market by 2020.  Others say they will make up the majority of the vehicles on the road by 2035, replacing all others by 2050. Amazing!

At the 2014 Consumer Electronics Show, BMW unveiled a new vehicle with highly automated driving capabilities (image courtesy of BMW Group).Sure, making a car that drives itself is not easy, but believe it or not, you have participated in the development of this new technology. If you have purchased a car in the last 20 years or so, it likely includes some form of computer control. Anti-lock brakes, automatic traction control, rear or front sonar, rearview cameras, park assist, automatic windshield wipers, cruise control, and even an automatic transmission are all forms of driver-assisted technologies. These advancements were made incrementally, and the transition to fully autonomous vehicles likewise will not take place overnight.     

So what do autonomous vehicles have to do with weather and climate? Well, there are several parallels between self-driving cars and weather forecasting. Before I make my argument, I want you to think about self-driving cars and how you feel about being removed from the driving process. Carnegie Mellon University has one of the world’s premier autonomous vehicle programs and professor Raj Rajkumar leads the university’s effort. He argues that “there are technical issues, but also issues of social acceptance…” when it comes to self-driving cars. So do you like the idea? Do you have any reservations about autonomous transportation?             

I’m willing to bet that most people have serious concerns about a computer driving them around town. It’s terrifying. What if the car drives off a cliff? What if the car hits a pedestrian crossing the street? What if a hacker compromises the network controlling the cars? What if I don’t want to lose my sense of freedom and my joy of driving? On the other hand, some people consider human drivers to be equally terrifying. We get drowsy, distracted, intoxicated, or fail to process information fast enough to make good driving decisions. The U.S. Department of Transportation accident statistics show 93 percent of car accidents result from human error. In 2010 alone, there were roughly 10 million car accidents across the United States leading to nearly 33,000 fatalities.   

Now let’s talk weather. The world of weather forecasting is going through a very similar transition. One hundred years ago, meteorologists used rudimentary methods or rules of thumb to predict the future state of the atmosphere. We had our successes, but we also made serious errors and could only forecast the weather out a few hours to a few days. As technology advanced, so did weather forecasting. Observation equipment became more reliable, cheaper, and capable of sensing minute details of the atmosphere. Satellites and radar allowed us to peer into weather like never before. Computers became faster and able to produce accurate 7- to 10-day forecasts in just a few hours. In other words, our forecast accuracy is highly correlated to technology.

Currently, the National Weather Service verifies the accuracy of our own forecasts. We compare our official forecast to a computer model as a way of gauging the accuracy of our forecast. For years, we have routinely outperformed the computer. The reason for this is due to the inherent errors in a computer model. Some models have a warm bias, or a wet bias, or a windy bias.  We can learn these biases over time and correct them in our official forecast. We can also compare several models at one time, giving us the ability to assess the range of possible future outcomes. In other words, the human has assumed that the computer model is inherently flawed and must be corrected. The human drives the forecast and performs exceptionally well.

In February 2013, the National Weather Service office in Tucson began experimenting with a different method to generate forecasts known as a blending approach. Instead of basing a forecast on the performance of one model, we combined all the models and took an average. This averaged forecast is called a consensus. As it turns out, this consensus forecast is very good. In fact, it outperforms any single model in almost every case. This means we are receiving a better forecast simply because we started blending models together. The human is still involved in this process. Even the consensus forecast has limitations and can be prone to error. However, the number of times this forecast does well far outweighs the times when it doesn’t.

This is a potentially scary idea for weather forecasters. Remember, we didn’t make any single model better by improving its resolution or adding computer power. We simply combined existing models. Some of our models will go through a significant upgrade this year, while others will be upgraded through 2018 (and beyond of course). Forecasters are now asking, “Where do we fit in?” 

The truth is humans must continue to be involved in the forecast process. We cannot simply allow the blended forecast to run without oversight. However, our roles in the process are certainly changing. In the near future, we may be less concerned with whether the high temperature tomorrow will be 75 or 77 degrees F and more concerned with what that temperature means for society. We’ll work more closely with members of the emergency response community to better communicate what certain conditions mean for their operations. 

In forecasting, we are nearing the time when the computer can take more control over some of the forecasting responsibilities, just as your car automatically will make corrections to your driving, giving you more of an opportunity to focus on reading a book, writing the Great American Novel, video chatting with a friend, or even taking a nap. Forecasters will then take more control over how that forecast is communicated to the public, fire officials, emergency responders, or elected officials. This is not the end of the human forecaster. Instead, this is simply the next evolution in using forecast information to make decisions. And just to be safe, we will still have an override button in case the forecast starts to veer off a cliff.