NEW - Experimental Forecasts for Pacific Island Rainfall:

In the September 1995 issue of the Experimental Long-lead Forecast Bulletin, a new statistical forecasting technique (developed by researchers at NOAA's Climate Prediction Center in Washington D.C.) for rainfall in the Pacific Islands was introduced. It includes experimental forecasts for U.S.-Affiliated Pacific Islands out to a year in advance for various seasons (3-month periods). An additional piece of information, which measures the skill, or expected accuracy, of these forecasts, is included. An adaption of the original figures, with an indication of minimum skill levels (discussed below), are shown below (forecasts for Wake, Kwajalien, and Johnston have been excluded, but are available on request):

The forecasts shown are for the various 3-month periods indicated by an abbreviation using the first letter in each month and the year (example: "OND95" refers to October-November-December 1995). The lines give a measure of the expected skill for the various forecasts. In general terms, these kind of forecasts are thought to be of useful skill if the value (the statistical "correlation") is greater than 0.3. Higher skills correspond to greater expected accuracy of the forecasts. Skill levels greater than 0.5 are thought to be quite good, and skill levels greater than 0.8 are thought to be excellent. Forecasts with greater than minimum skill are highlighted with arrows. For most cases, the forecasts are of fairly modest skill. However, these forecasts can be useful in identifying an expected trend in seasonal rainfall, and higher skills are expected for forecasts produced at different times of the year.

The actual forecasts are indicated by the bars, and give an outlook for the "standard anomaly" expected to occur in rainfall, compared to average rainfall for the 3-month period indicated using a climatology based on records for the years 1961-1990. "Standard anomalies" are a statistical measure which, together with the skill described above, can be converted to an expected range of rainfall when the average and the range of actual rainfall during the base period (1961-1990) are known. As an example, this information has been determined for the Guam forecasts, and appears with an explanation in Figure 1 below:


FIGURE 1 (below) is an interpretation of the Guam forecast (above), based on actual rainfall records. The LIGHT (red) LINES in the figure show the range of rain that fell on Guam about 20 out of 30 years (or 68% of the time) that was closest to the median or "normal" rainfall for the 3-month seasons indicated (based on data from the 30 years of 1961-1990). The median rainfall is shown with the light dotted line, and the 68% range around it with the light solid lines. This can be though of as the "range of certainty", without a forecast, that rainfall will be seen within these limits about 2 out of 3 times (or 68%) in the 3-month seasons shown. Note that in some seasons, above-normal or "wet" conditions can be more extreme, or farther from normal, than dry conditions - this is called skew in a rainfall distribution. The HEAVY (blue) LINES show the actual forecasts (dashed line) and the 68% "range of certainty" that rainfall will be seen within the limits of the heavy solid lines, based on the expected skills of the forecasts - note that "useful" skills (marked with arrows above) give "better" (smaller) ranges of certainty compared to the 1961-90 climatology. However, even if forecasts are for below-normal rainfall, there is still some chance, as can be seen, that rainfall could be above-normal.


When available, these forecasts will be included in future issues of Pacific ENSO Update.

Similar research on this prediction technique is underway at the Pacific ENSO Applications Center. In general, the skill levels from this experimental work are comparable with those shown in these forecasts. By exploring the differences in skill, we expect to contribute to the development of more accurate predictions of seasonal rainfall in the Pacific Islands in the near future. Progress of this work will be reported in future issues of Pacific ENSO Update.