Table 3: Heidke Skill Score SS2 for
experimental forecasts for summer 1998, based on antecedent soil moisture conditions. Element w
stands for instantaneous soil moisture at the last day of the target month/season, while P and T
are time averages of temperature and precip over the 30/90 days. [The smaller #s in ( ) are SS2 for
official CPC forecast]
Target
| W
| T
| P
|
|
June 1998
| 30 (NA)
| 14 (13)
| 13 (10)
| (data thru April 30; fcst released mid-May)
|
July 1998
| 31 (NA)
| 18 (17)
| 09 (02)
| (data thru May 31; fcst released mid-June)
|
Aug. 1998
| 34 (NA)
| 34 (15)
| 11 (08)
| (data thru June 31; fcst released mid-July)
|
JJA 1998
| 29 (NA)
| 28 (25)
| 17 (07)
| (data thru April 30; fcst released mid-May)
|
|
The results for Summer 1998 show that
1. Monthly forecasts for w have rather high skill (compared to say T and P, which has
SS2=6.4 and 0.4 respectively, averaged over Jan95-May97). This is generally true and is
helped greatly by the fact that soil moisture anomalies are, in general, much more
persistent than temp and precip anomalies, thus making it an easier forecast target, even
at a one, two or even four month lead.
2. Forecast skill for monthly T and especially P are high by the above-mentioned
standards ( (expected is only SS2=6.4/0.4 for T/P). The scores for P are in fact
record-like, certainly for summer, certainly for a short averaging period like a month,
and as good as they were during the El Nino winter of 97/98.
3. On the whole the scores of this tool for summer 1998 were very good. Especially the
scores on Precipitation in the summer (the most difficult target of all) SS2 ~ 9-17 are
outstanding, given scores that normally hover near zero.
4. In spite of the boasting above, the scores are only modest by absolute standards
(100 is perfect). The scores for T and P were much better than persistence, but the scores
for w did not outperform persistence of the initial state's anomaly. This is a bit
mysterious, and raises a research issue of a hybrid model, in which the physical soil
model is integrated in time to produce the w forecast with the driving P and T coming from
an empirical forecast like CAS.
|