IntermtnClimateSummaryJan2006 - Laserfiche WebLink

Home Browse Search

FEATURE ARTICLE FROM INTERMOUNTAIN WEST CLIMATE SUMMARY, JANUARY 2006 (Continued from p.2) and 33 percent considered below-average. For example, a forecast that calls for a 40 percent probability of above-average temperature is less certain than a forecast that calls for a 70 percent probability of above-average temperatures. In both cases the projection is for temperatures to fall into the above-average tercile as compared to the actual conditions observed from 1971 through 2000. White space on the map indicates Equal Chances (EC) of falling into any of the three terciles (i.e., no forecast). Only rarely does the CPC issue a forecast predicting near-average temperatures, indicated by gray shading. Climate forecast performance On the FET home page, you'll also see options to "Explore the Forecasts," to consider "How do the forecasts relate to my specific situation?" and to evaluate "Forecast Performance." Select "Forecast Performance" to follow the example here. This is where you can test and com- pare how CPC forecasts have performed in the past, based on the forecasts issued since 1994. Here we take a step-by-step approach to testing a seasonal forecast's success: 1. The "National Weather Service Cli- mate Prediction Center" option is auto- matically selected, so there's no need to do anything. (In the future, other options will become available.) 2. Select NWS CPC seasonal climate outlooks (contiguous states). 3. Select precipitation. 4. Select a forecast season, in groups of three months, by sliding the shaded box with your cursor and then clicking on it. The months are listed by their first initial only. Choose DJF to get the three-month seasonal outlook for December, January, and February. The selected grouping will show up below the shaded area as DJF. (If you want to do more than one three-month period, click your mouse upon each selec- tion and you'll see the selected months listed below.) 5. Select the month or months during which the forecast was issued. Click in the boxes for each year you want. We'll select N (November) for each available year (1994-2004). The three-month sea- sonal forecasts are issued up to a year in advance and updated every month. 6. You now have the opportunity to select the type of statistical test you'd like to apply to the forecasts. Select the "False Alarm Rate" option. Brief descriptions of the other options (e.g., Probability of Detection, Brier Score) are included at the end of this article. 7. Once you have made your choices, hit "Submit" to launch the program. When the results appear, read the box at the top under "You Chose" to make sure the computer accurately recognized all your choices. (For example, if you did not click on your season selection, the default" All Seasons" \vill appear.) 8. The results will include national maps color-coded by division and a color bar below that explains the legend (Figure 1 b). For these comparisons, the 344 NOAA climate divisions have been grouped into 102 larger divisions. Colo- rado Wyoming, and Utah have eleven total di visions under this system, with some di- visions that overlap other states. You can see the actual value for a climate division by holding your cursor over it. Frequency of Forecast Results Regardless of which category you select, you will first see a map indicating the Frequency of Forecast Results. This shows how often a forecast was actu- ally made about the season of interest by climate division. A value of 0.322 means a forecast covered some or all of the divi- sion about 32.3 percent of the time since 1994, when forecasts were available more than one month ahead. Scroll down to see the results you were seeking. False Alarm Rate This comparison considers how often the projected forecast turns out to be wrong, using the category that was predicted to be most likely. To convert the resulting climate division score into a percentage, just multiply the value by 100. So if forecasters called for wet condi- tions three times, but they only occurred twice, the false alarm rate would be 0.333 or 33 percent. Note that, in this case, low scores are good. To consider how often an issued forecast was accurate, just subtract the False Alarm Rate score from 1 (or the percentage from 100). In this theoretical example, the forecast was accurate 66 percent of the time. In the actual example tested here, scores ranged from 0.5 to 0.857 for "wet" conditions and from 0 to 0.75 for "dry" conditions (Figure Ib). Wa- ter managers have indicated they find the False Alarm Rate particularly relevant. Show Data Behind the Map If you want to see the forecasts that were considered for the evaluation, click on a climate division of interest and then click on the "Show the Data Behind the Map" option. First you'll see a description of how to interpret bubble plots, including a sample bubble plot. Then you'll see the data used for the cli- mate division of interest for the season(s) and years indicated. Besides the False Alarm Rate, there are a number of other options available for evaluating forecasts. To try other tech- niques, return to the Climate Forecast Per- formance page. (If you can't find it, return to the FET homepage and select "Forecast Performance. ") Modified Heidke Score This selection is intended for use by the National Weather Service (NWS) forecast- ers who have historically used this ap- proach to evaluate forecasts. It is included on the FET site because NWS forecasters receive instruction in use of this tool as part of their ongoing climate training courses, as explained by NWS Climate Services Chief Robert Livezey. He feels that for those not familiar with the Heidke system, the other methods provided (e.g. Frequency of Forecasts, Probability of Detection, False Alarm Rate, Brier Score, FEATURE ARTICLE I 3 (Continued on p. 4)