Real-time Verification of CPC Long-Lead Forecasts Using Heidke Skill Scores from Station Data All Stations Non-EC Only % Coverage Season T 1.7 4.9 35.0 MJJ 05 P -5.1 -20.0 25.6 Month (LL) T 10.7 43.0 24.8 JUL 05 P 4.3 29.4 14.5 Updated Mo. T 6.8 21.1 32.5 JUL 05 P 0.0 0.0 35.9 Change T -3.9 -21.9 +7.7 (U-LL Mo.) P -4.3 -29.4 +21.4 Brief Discussion of Forecasts and Verification ---------------------------------------------- The temperature forecast for MJJ 2005 had very modestly positive Heidke skill scores of 1.7 for all stations and 4.9 for non-EC only, with coverage of 35.0 %. These were all below the long-term averages of 8.6 for all stations and 18.4 for non-EC only, and the average coverage of 46.9% for the entire period of record. The MJJ 05 seasonal forecast was based on a variety of statistical and dynamical models, and multi-model ensembles from IRI and CDC, as well as CPC's CCA and OCN statistical tools. The NCEP CFS model was minimally considered, as it seems to have a cold and wet bias during the warm seasons. Indications over the equatorial Pacific suggested that ENSO would be in a neutral phase during summer 2005. The forecast was primarily based on the traditional CPC statistical tools, all of which favored warmth in the interior Southwest, parts of the Southeast, and along the Pacific Northwest Coast. Most of the models at least hinted at some relative warmth in these areas as well. Below normal temperatures were forecast over the upper Mississippi Valley on the basis of a consensus of most of the models with support from CCA. Only the SMT statistical tool correctly hinted at the warmth that was observed in the middle of the country, and most of the Southeast showed a mix of mostly near and below normal temperatures, despite the indications from most of the tools for warmth. The forecast of above normal temperatures in Alaska was very successful, scoring 89.1 for all stations and 93.2 for non-EC only. Every station in Alaska had above normal temperatures, with some record early season heat in late April and early May contributing to a rapid loss of snow cover. The good forecast was aided by the fact that all the tools, both statistical and dynamical, correctly indicated the warmth that occurred in Alaska. The precipitation forecast for the CONUS had negative skill overall. The area of above median centered over the upper Mississippi Valley and western Great Lakes had marginally negative skill overall, with the western part mostly correct due to frequent MCS activity, while the eastern part, in an area of worsening drought, was nearly all wrong. The area of dryness predicted in the Southeast was all wrong, most of it with two-class errors. The all-station score was -5.1 and the non-EC only score was -20.0, with a moderate coverage of 25.6%. These forecasts were based on almost unanimous indications from the dynamical models that it would be wetter than normal over much of the north-central part of the country, and dry over the Southeast. There were no significant signals from OCN or CCA, but CAS was largely incorrect, indicating dry in the Southwest, wet in the southern Great Plains and Rio Grande Valley, dry over the upper Mississippi Valley, wet over the upper Ohio Valley, and dry over the Southeast. The CFS model did correctly forecast much of the wetness in the West, and the northern Great Plains, but would not have scored very well overall because it indicated wet conditions over nearly all the CONUS. The CAS statistical soil moisture model did correctly indicate some of the wetness that was observed over the northern Great Basin and northern Rockies. Alaska was left with all EC, as there were no consistently strong or reliable forecast indications from any of the tools for that area. The 0.5-month lead temperature forecast for July was quite successful, with an all-station Heidke Skill score of 10.7 and a non-EC score of 43.0, with a modest coverage of 24.8%. The areas predicted to have above normal temperatures in the far West, the interior Southwest, and the Florida Peninsula were nearly all correct, but unfortunately the smaller area of below normal predicted over the northern Rockies had all two-class errors. Alaska again had a very good forecast of above normal for most of the state with an all-station score of 43.5 and a non-EC score of 46.5, with coverage of 95.7%. The areas forecast to be warm were based primarily on OCN with agreement from CCA and CAS in some areas. Pre-existing above normal SSTs around Alaska also contributed to the largely successful forecast there. The failed area of below normal temperatures forecast over the northern Rockies was based on CCA and the CFS dynamical model. The all-station skill of the 0.5-month lead precipitation forecast for July had modest skill, scoring 4.3 for all stations and 29.4 for non-EC with a modest coverage of 14.5%. The area of above median predicted primarily on the basis or the CFS model in the interior Pacific Northwest was only partially successful, while the dry area forecast in the interior Southwest on the basis of the expectation of a late summer monsoon onset was mostly correct, and for the right reason. The monsoon onset was one of the latest on record, more than two weeks later than usual. The same persistently strong trough near the West Coast that was one of the factors in delaying the monsoon also contributed to the above median precipitation totals observed over much of the Far West, including traces of rainfall at most of the California stations, which are normally rainless in July. Alaska was left as all EC due to lack of any statistically significant signals from any of the tools. The revised updated temperature forecast for July issued at the end of June, incorporating input from several dynamical models used for short- and medium-range forecasting resulted in a decline in the temperature forecast skill, lowering the scores to 6.8 for all stations and 21.1 for non-EC only. The area of above normal forecast for the Florida Peninsula was removed with the expectation that wet conditions and persistence of below normal SSTs would cancel the strongly significant long-term trends for warmth. However, precipitation over peninsular Florida was near to below median, and SSTs increased much faster than the normal seasonal trend, going from below to above normal not only in the vicinity of Florida but along most of the south and middle Atlantic Coast during the month. An extension of the southwestern area of predicted above normal temperature eastward through the Rio Grande Valley gained a few points, but the eastward extension of the forecast area of below normal temperatures through the upper Mississippi Valley to the western Great Lakes did not gain even one correct station. This change was done primarily on the basis of medium range model indications that there would be active precipitation and frontal activity in the north-central states. Model guidance for Alaska, which also indicated below normal temperatures, was wisely not followed with the expectation that the strong long-term trends for warmth would prevail, and the fact that the models often forecast middle tropospheric heights and temperatures to be too low over Alaska. The temperature forecast was left unchanged in Alaska, maintaining a good score there. The revision to the precipitation forecast in expanding the area of above median to include the central and northern Great Plains and the upper Mississippi Valley, along with the addition of an area of forecast above median in the Southeast, including Florida, resulted in the Heidke skill scores for both all stations and non-EC declining to exactly 0.0, although the coverage increased dramatically to 35.9%. Short- and medium-range model forecast indications overdid the amount of precipitation in the Great Plains and upper Mississippi Valley, although the forecast of heavier amounts of precipitation in the Southeast, aided somewhat by CAS, was largely successful, except for over the Florida Peninsula. The forecast of EC was retained for all of Alaska. --- EOF