1.3.6: Meteorology

The science of forecasting, or predicting, weather conditions is heavily dependent on statistical modeling and data science. Weather forecasting is essentially based on the wealth of historical regional weather data that has been collected for approximately the last one hundred years or so. For short term forecasts the process, from a very simplified viewpoint, is relatively straightforward. For example, one can look at the regional weather conditions today and look back through the historical data for days that had similar regional weather conditions. If we want to predict whether it will rain tomorrow, we can look at how many times there were the same conditions, and it rained the next day. This method may work to some extent, but modern weather forecasting is much more sophisticated.

The need for sophistication and advanced computer algorithms can be appreciated when one thinks about the five-day forecast that is quite common on weather reporting news services. As one seeks to predict conditions much later beyond the present time, the atmospheric conditions over a much wider area becomes important. One can imagine that if we want to predict what the weather will look like in an hour, then relatively local conditions should suffice for making that prediction. A storm system that is 500 miles away is unlikely to have a large immediate effect of the local weather. When one increases the prediction to five days, then even the potential formation of storms 500 miles away might be very important to a local weather forecast.

The potential for large errors will generally increase as the time frame becomes longer. It might be relatively easy to predict what the temperature might be in an hour, and in most circumstances even if there is some error, it might not be very great. This is simply because in most conditions the temperature does not make substantial changes in a brief amount of time. If it is 90 degrees Fahrenheit at 2:00 p.m., it is usually likely to be near that temperature at 2:30 p.m. Now consider the five-day forecast horizon. Predicting the temperature at 2:30 p.m. in five days would be very difficult unless you were able to take in the many possible changes in conditions that might occur in relatively distant regions. Once a prediction is made, any small changes in those remote regions can have a large effect on the outcome. For example, a meteorologist might see a cold front moving in a way that would cross the area in interest in four days. This front obviously has the potential to have a large effect on temperature predictions in local regions on the fifth day. In the meantime, it is entirely possible that the regional conditions may change sufficiently so that the front turns northward and misses the area of interest. Predictions made based on the front will crossing the region on the fourth day will be very far from the actual outcome.

Despite what some individuals might feel personally, the business of weather forecasting has made great technological strides over the past 100 years. The very thought of a five-day forecast one hundred years ago must have seemed impossible. The advent of satellite technology and the seeming ubiquity of weather stations across the globe has provided modern meteorologists with more data than they have ever had before. Sophisticated modeling algorithms based on mathematical and statistical models and advanced data processing algorithms have made predicting the weather more accurate than it has ever been. Of course, prediction is not always perfect, and meteorologists often make mistakes.

One important point that this book emphasizes is that errors based on data cannot be avoided. The key idea is to know how big the errors might be and how often they might occur. Armed with these two bits of information, one can still make informed decisions in the face of uncertainty.