Winter Season Climate Prediction for the UK Health Sector Glenn R McGregor, The University of Birmin

1
Winter Season Climate Prediction for the UK
Health SectorGlenn R McGregor, The University of
Birmingham, UKg.r.mcgregor_at_bham.ac.uk NOAA 29th
Annual Climate Diagnostics and Prediction
Workshop October 2004, Madison, Wisconsin
Background To date the UK health sector has not
considered incorporating seasonal level climate
information into the winter planning process.
Aim To assess the utility of long-lead winter
season climate forecasts as a basis for
developing climate based health forecasts (HF)
for the West Midlands region England.
Methodology Daily de-trended all cause mortality
(total and 65yrs per 100,000) and daily Tmean,
Tmax and Tmin temperature data were averaged into
monthly (D,J,F) and seasonal (DJF) values and
then converted to standardised anomalies for the
period 1974/75 1989/99. Daily temperature data
were also used to count the number of days per
month/season below a range of standardised
thresholds (1.0, 0.5, 0.0, -0.5, -1.0 standard
deviations). Delete-1 jack-knife regression was
used to find the best fit linear (HF) models
describing winter temperature - mortality
relationships. The HF models were then applied to
the problem of making retrospective forecasts of
D.J, F and DJF mortality. The HF model input
variables were constructed from 186 day (November
start 1987-1999) 15 member ensemble hindcasts of
temperature for 9 grids points across the UK
produced by the Met Office GloSea seasonal
climate prediction system.
Results Significant associations were found
between mortality and Tmean, Tmax and Tmin and
the number of threshold days for most months,
but most notably for February (Figs 1a 1b). ROC
analysis revealed that the GloSea hindcasts
possessed skill for February only. Further the
skill for Tmean, Tmax and Tmin was greater than
that for the number of threshold days. Only
Tmean and Tmax based hindcasts of total (r2.41)
and 65yr(r20.35) mortality for February were
found to possess any skill predictability based
on the number of threshold days was low.
February Tmean vs Mortality
Days February Tmean lt -1.0 sd vs mortality
Fig 1a
Fig 1b
Discussion and Conclusions Clearly the extended
range predictability of mortality depends on the
degree to which input data for climate based HF
models can be skilfully forecast. Winter
mortality in the UK is related to periods of
anomalously cold weather, which are associated
with distinct atmospheric circulation modes or
regimes such as blocking. Therefore, the
challenge for health related seasonal prediction
is to estimate the future occurrence of
circulation regimes associated with anomalously
cold conditions. Accordingly for the wider UK
region, the extent to which coupled O- A
prediction systems, like GloSea, can resolve the
North Atlantic SST tripole pattern is critical as
this, through O- A interaction, has a discernable
impact on the European winter circulation regime
and blocking occurrence.