| You might think that weather forecasting requires | | | | at the statistics could answer this. One could |
| a degree in meteorology. Well, perhaps a degree | | | | check the forecast highs and lows for the last |
| in statistical analysis would be more useful. Here is | | | | 365 days, and check the actual temperatures for |
| a new way to forecast the weather with less | | | | those days. One could also see what the |
| knowledge, but greater accuracy. | | | | predicted probabilities of rain or snow were, and |
| Here in Canon City, Colorado, on Friday, February | | | | then note what actually occurred. |
| 2, 2007, I brought in my Newspaper from the | | | | For example, suppose a forecaster predicted a |
| porch when it arrived, at about 3 in the afternoon. | | | | 50% chance of rain 24 times, but it actually rained |
| I opened the newspaper to the page with the | | | | 18 times. Perhaps he had the best data, but he |
| weather forecast. I was wondering how cold it | | | | was too conservative in its application. This may |
| would be on Saturday. | | | | not be a one-time problem. This can be |
| 13 degrees Fahrenheit was the high temperature | | | | determined by doing more statistical analysis. If his |
| forecast for Saturday. This was way too low, I | | | | error was consistent, you could know nothing |
| figured. I checked the forecasts on television and | | | | about weather forecasting and provide a more |
| on the internet. They said we would reach 23 or | | | | accurate forecast simply by saying "A 75% |
| 27 degrees the following day. These forecasts | | | | chance of rain tomorrow" every time he said |
| were also too low, I knew. I told my wife it would | | | | there was a 50% chance. |
| be in the 30s or higher. What was the actual high | | | | That's the essence of how this new forecasting |
| temperature the next day? 53 degrees | | | | idea works. You first gather statistical information |
| Fahrenheit. | | | | on the forecasts of several meteorologists or |
| That's not a typo, by the way (how do they | | | | weather forecasting services. Then you compare |
| keep their jobs?). Weather forecasting "experts" | | | | these forecasts to the actual weather that |
| were off by as much as 40 degrees - for a | | | | happened, and look for any consistent errors. |
| simple 24-hour forecast of the high temperature. | | | | Ideally you would want to create a computer |
| Why were they so far off, and how could I be | | | | program, the idea being that as you enter each |
| better than them at forecasting the weather for | | | | of these forecasts into it, they are adjusted for |
| Saturday? | | | | any known tendencies. The result should be more |
| Well, I don't have an answer for the first part of | | | | accurate weather predictions. |
| the question. Weather here is certainly more | | | | An example might make this clearer. Suppose |
| unpredictable than in many places. Perhaps the | | | | that over the last year Forecaster A has been |
| meteorologists follow there computer models too | | | | forecasting a high temperature that averages 4 |
| slavishly, even when experience and intuition tell | | | | degrees over the actual high. The computer would |
| them to adjust a forecast up or down. | | | | adjusts his forecast down four degrees. Perhaps |
| I can answer the second part of the question. My | | | | a more sophisticated analysis shows that |
| forecast was closer because the "experts" were | | | | Forecaster B is consistently predicting too high of |
| so consistent in the errors they made. I had | | | | a probability of rain in the fall, but too low of a |
| counted something like 15 out of 20 days when all | | | | probability of rain in the summer. Once this is |
| the various weather forecasts predicted a high | | | | discovered and programmed in, the computer can |
| temperature that was 5 degrees or more lower | | | | adjust the forecast for these factors. For greater |
| than the actual temperature. Seeing that, all I had | | | | accuracy, the adjusted forecasts of three or |
| to do was take the forecast (the one predicting | | | | more sources could be averaged. |
| the highest temperature) and add five or six | | | | You wouldn't need to know anything about |
| degrees. | | | | weather forecasting. The underlying idea is that |
| A New Weather Forecasting Idea | | | | even when experts have the best knowledge and |
| The consistency in the direction of their errors | | | | data, they can apply it incorrectly, and do so |
| was the key to my better forecast. In other | | | | consistently. Perhaps some television stations will |
| words, they weren't forecasting too high one day | | | | soon get rid of their meteorologists and take |
| and too low the next. They were making their | | | | advantage of this new weather forecasting idea: |
| errors in the same ways repeatedly. | | | | "And now it's time for your electronic weather |
| The next logical question is whether errors are as | | | | forecast, from our Statistical Analysis Weather |
| consistent in other parts of the country. Looking | | | | Machine... |