Forecasts
of Southern African DJF Rainfall Using Model Output Statistics
contributed by W. A.
Landman, L. Goddard and A. Barnston
Numerical model global
predictions from the ECHAM3.6 AGCM (Deutches Klimarechenzentrum 1992) are
recalibrated statistically to produce regional-scale rainfall forecasts. Using
this dynamical-statistical approach suggests that rainfall over much of the
southern Africa region is likely to be near-normal to above-normal during the
DJF 2001/02 season.
Method:
Statistical correction
of GCM output is often necessary when significant systematic biases occur
between the real world and its modeled presentation. Such an approach can be
utilized to construct relationships between the desired forecast quantity such
as rainfall and GCM-simulated variables, such as large-scale circulation. The
statistical approach used here to develop the equations relating the GCM
large-scale to southern African summer rainfall is called model output
statistics (Wilks 1995). Canonical correlation analysis (CCA) is the
mathematical technique used to set up the MOS recalibration equations. EOF
analysis is performed on the predictor and predictand sets, and the number of
modes to be retained in the CCA eigenanalysis problem is determined using
cross-validated skill sensitivity tests. The GCM large-scale patterns
considered for the MOS model are the DJF GCM fields of geopotential heights
(850, 700, 500 and 200 hPa), thickness (850 to 500 and 500 to 200 hPa) and
moisture (700 hPa specific humidity). Regional rainfall indices form the
predictand set, and are computed for nine homogeneous rainfall regions (Mason
1998; Landman et al. 2001).
To select the best
predictor(s) from the array of GCM fields, the screening process called forward
selection (Wilks 1995) is employed, resulting in the selection of the 850 hPa
geopotential field as the only predictor in the MOS model. In addition, the
number of retained predictor and predictand EOF modes that produced the highest
average cross-validated correlations resulted in the use of two CCA modes in
the MOS model. Details of the MOS model, its skill comparison with the
GCM-generated rainfall and forecasts from a simple statistical forecast
algorithm relating SSTs and rainfall, as well as the MOS model’s performance
over an independent 9-year retro-active forecast period, are described in
Landman and Goddard (2001).
The dynamical model
predictions are generated for DJF 2001/02 using the ECHAM3.6 AGCM, by
persisting observed November 2001 SST anomalies on top of the monthly varying
annual cycle of climatological SSTs. An ensemble of 10 realizations is
produced; 5 of those are used in the MOS model. At initialization, ensemble
members differ from each other by one model day integration. An identically
formulated set of retrospective forecasts using persisted SST anomalies from
November were used for DJF 1970/71 – 1999/2000, and these were used to build
the MOS model and determine its forecast skill.
Large-Scale Circulation and Rainfall over
Southern Africa:
Pattern analyses of
the dominant CCA mode reveal strong ENSO-like variability between the 850 hPa
geopotential heights and the rainfall. Anomalously high (low) 850 hPa
geopotential heights over the central and western interior of the subcontinent
are associated with dry (wet) conditions over the region. The second CCA mode,
although it explains considerably less of the rainfall variability than the
first, shows that anomalously low 850 hPa geopotential heights over central
southern Africa, reminiscent of tropical easterly waves at about 15 to 20°S,
are associated with wet conditions over land. Wet conditions have been observed
during a number of years within the 30-year climate period that are associated
with such a tropical wave. In contrast, anomalously dry conditions have been
observed in the absence of such a wave system. Notwithstanding, this second CCA
mode makes only a small contribution to the MOS forecast skill.
The Forecast:
Near-normal rainfall
is forecast for the most part over southern Africa, except over the semi-arid
western region and the southern coastal strip where above-normal rainfall is
likely to occur (Figure 1). The forecast DJF 2001/02 850
hPa geopotential anomaly field shows a pattern that is in close agreement with
the spatial pattern of the second canonical mode 2 (Figure 2).
However, while there is pattern agreement, the amplitude is small owing to the
weak positive height anomalies over the central parts of the region and the
southwestern Indian Ocean. Moreover, even if the observations had higher
amplitude for this pattern, mode 2 exerts only a minor influence on the
forecast due to its small role in terms of variance explained. Mode 1 therefore
dominates this forecast, and this year’s weakly negative SST anomalies in the
eastern equatorial Pacific Ocean are not sufficient to produce an above-normal
rainfall recalibrated MOS forecast in most of the regions, as seen in Figure 1.
References:
Deutches Klimarechenzentrum (DKRZ), 1992: The ECHAM3 Atmospheric General Circulation
Model. Technical Report 6, Hamburg, Germany. pp 184.
Landman, W. A., and L. Goddard, 2001: Statistical
recalibration of GCM forecasts over southern Africa using model output
statistics. Journal of Climate, accepted.
Landman, W. A., S. J. Mason, P. D. Tyson, and W. J. Tennant,
2001: Retro-active skill of multi-tiered forecasts of summer rainfall over
southern Africa. International Journal of
Climatology, 21, 1-19.
Mason, S. J., 1998: Seasonal forecasting of South African
rainfall using a non-linear discriminant analysis model. International Journal of Climatology, 18, 147-164.
Wilks, D. S., 1995: Statistical
Methods in the Atmospheric Sciences. Academic Press, San Diego, 467 pp.
Figure captions:
Figure
1. Categorized rainfall MOS forecasts for DJF 2001/02. A refer to
above- and N to near-normal equi-probable rainfall categories. LEPS
scores are shown with the predicted categories, calculated from 30 years of
cross-validated rainfall forecasts. LEPS scores significant at the 95% level of
confidence are indicated with a “*”, and those significant at the 99% level,
with “**”.
Figure
2(a). CCA mode 2 predictor map of the 30-year training period used in the
MOS equations relating GCM predicted 850 hPa geopotential heights to DJF
regional rainfall. Shaded regions depict areas of significant loadings at the
95% level of confidence.
Figure
2(b). DJF 2001/02 850 hPa geopotential height anomalies in gpm, based on
the 30-year climate period of 1970/71 to 1999/2000.
