P1.4

**An evaluation of the 10 years of seasonal climate outlooks issued for Wisconsin**

**Steven J. Meyer**, University of Wisconsin, Green Bay, WI; and K. A. Hemauer, K. L. Karl, C. Sonnabend, and C. S. Spannagle

Since a considerable share of Wisconsin's economy is contingent upon weather and climate particularly as it relates to agriculture, energy use, and tourism, accurate predictions of future climate could prove extremely useful. Ten years ago the Climate Prediction Center (CPC) began issuing long lead seasonal climate outlooks. Issued on the third Thursday of each month, these outlooks are seasonal in that they cover overlapping three-month seasons (JFM, JMA, …, DJF) and are long lead in that they are issued 0.5 to 12.5 months in advance (thus 13 outlooks are issued each month). The outlooks are expressed as the probability that the seasonal mean temperature and accumulated precipitation will fall into one of three categories – above normal (AN), normal (N), or below normal (BN) – relative to climatic normal. The CPC-assigned probability may be skewed toward a greater likelihood in one of the three categories or may indicate “equal chances” (EC); i.e., a 33.3% probability that temperature/precipitation will fall into any of the three categories: AN, N, BN.

Any prediction, especially a long-range one, is accompanied by uncertainty. Thus, potential users need to be aware of the extent of the uncertainty so they can base their decisions accordingly. If the CPC's seasonal climate outlooks are to be useful in the public or private sectors, their skill on a spatial and temporal basis must be determined. In this study, we evaluated the CPC's temperature and precipitation outlooks for the three “outlook divisions” in which Wisconsin is located. Our evaluation addressed three questions: 1) Do certain seasons yield more accurate outlooks than others?, 2) Does lead time (shorter vs. longer) influence the accuracy of the outlook?, and 3) Does increased confidence in an outlook (as expressed by the assignment of a higher probability) translate into a more accurate outlook?

From the Climate Prediction Center's website we obtained the seasonal climate outlooks (i.e., the probabilities assigned to the temperature and precipitation outlooks), the observed seasonal values, and the season-specific threshold values used to differentiate the AN, N, and BN categories. These data were entered into a spreadsheet then sorted by forecasted season, lead time, and magnitude of the CPC-assigned probability anomaly. We used a chi-squared analysis to compare the expected number of AN, N, and BN outcomes (based on the probability anomaly assigned to the temperature/precipitation outlooks) to the observed number of outcomes. The resulting p-value (significant at the 0.05 level) indicated the level of agreement between the expected number of AN, N, BN outcomes and the observed number of AN, N, BN outcomes. In all, 4368 temperature and 4368 precipitation outlooks were evaluated.

When examined by season, the P-values obtained from the chi-squared analysis of outlooks showed expected outcomes based on the AN temperature outlooks made for the JFM, MAM, and JAS seasons matched the observed outcomes at the 0.05 level. Expected outcomes based on the BN temperature outlooks made for the MJJ season matched the observed outcomes. Expected outcomes based on the EC temperature outlooks made for the JJA season matched the observed outcomes, indicating that JJA temperatures were quite random and unpredictable. For precipitation, expected outcomes based on the AN outlooks made for the JFM, JJA, and NDJ seasons matched the observed outcomes. And expected outcomes based on BN outlooks for the FMA, ASO, SON, and OND seasons matched the observed outcomes. There were no EC outlooks whose expected outcomes matched the observed outcomes.

Due to the methods and tools the CPC uses to generate the outlooks, the assigned probability anomalies change little from the time a particular season is issued its 12.5 month lead time outlook to the time it is issued its 0.5 month lead time outlook (i.e., the probability anomalies assigned in their outlooks remain consistent/persistent over time). Therefore, the expected number of outcomes changes little over the course of the 13 lead times (0.5 to 12.5 months). In the case of temperature outlooks, the P-values indicate expected outcomes based on AN outlooks match observed outcomes at all lead times for outlooks with greater confidence (probability anomalies of 0.400 to 0.499). For AN outlooks with lesser confidence (probability anomalies of 0.334 to 0.399), expected outcomes matched observed outcomes at all lead times of 0.5 months to 8.5 months. For BN outlooks with lesser confidence (probability anomalies of 0.334 to 0.399), expected outcomes matched observed outcomes at the following lead times: 0.5, 1.5, 5.5, 6.5, 7.5, 8.5, 10.5, and 12.5 months. For precipitation, expected outcomes based on the AN outlooks matched observed outcomes at all lead times, however, beyond a lead time of 4.5 months there were fewer than 30 outlooks to analyze. Expected outcomes based on the EC outlooks matched observed outcomes at all lead times also. This is a good indication of the randomness and unpredictability of seasonal precipitation in Wisconsin.

When all of the temperature outlooks are pooled and examined, only in AN outlooks with high confidence (probability anomalies of 0.500 – 0.599) did the expected outcomes match the observed outcomes. However, there were only 21 outlooks with probability anomalies at this level. When all of the precipitation outlooks are pooled and examined, only in AN outlooks with high confidence (probability anomalies of 0.400 – 0.499) did the expected outcomes match the observed outcomes. However, there were only 30 outlooks with probability anomalies at this level.

Wisconsin, with its continental climate, makes for a challenging location to attempt prediction of seasonal climate. This was observed in the analysis of long lead seasonal climate outlooks the CPC has been issuing for Wisconsin the last 10 years.

Poster Session 1, Probability and Statistics

**Monday, 30 January 2006, 2:30 PM-4:00 PM**, Exhibit Hall A2** Previous paper Next paper
**