One objective of the European PROVOST experiment (Prediction Of climate Variations On Seasonal and interannual Timescales) is to estimate potential dynamical seasonal predictability given ideal surface boundary conditions on a global scale. To this end, four European models (the UKMO Unified Model at climate resolution, the ECMWF T63model and the ARPEGE model run at T 42 by Mätäo-France andT63 by EDF) have been integrated in 9-member ensembles initialised at 24-hour intervals for four months for each season over 15 years from 1979 to 1993 (climatologies are calculated over the same period). Common initial conditions and verifying analyses obtained from the ECMWF reanalysis as well as common SST anomalies from the UKMO GISST and Reynold's OI data sets were used in all experiments. All initializations were at 0000Z finishing on the day prior to the start of the season. The north-eastern region of South America has been identified through these PROVOST experiments using the UKMO model as an area of relatively high predictability and in this paper a real-time seasonal forecast at one month lead is provided for the region based on these assessments. The area with higher predictability broadly stretches through much of eastern Brazil, the Amazon Basin, French Guiana, Surinam and Guyana, even covering the southeastern parts of the Caribbean. Within this area correlations over the 15 years between ensemble-mean rainfall anomalies and anomaly values obtained from the gridded observed land-surface rainfall data set of Hulme (1994) exceed 0.5, with correlations in excess of 0.8 in the vicinity of the Nordeste region of Brazil (Fig. 1). The observed data set is gridded to the same resolution as the model (2.5°0x3.75°), but in order to reduce noise 4°x4° blocks have been joined together to produce Figure 1; only blocks with adequate data were retained. Time series of ensemble mean rainfall and the Hulme data for selected regions illustrate the fact that the dynamical model, although able to capture the

interannual variability reasonably well, has insufficient variability (Harrison et al, 1997). Hence a variance inflation has been calculated using ensemble means for each gridded area and applied to the forecasts from both the members and the ensemble means given below. There are high correlations between ensemble mean predictions and the common rainfall indices for the Nordeste (see Harrison et al., 1997 and Colman et al., 1997, for index details); again the model has a high level of predictability (correlations approach 0.9) but requires variance inflation (for example against the FQ index, but results for other indices are almost identical). It is found that observed anomalies frequently lie within the ranges of the inflated ensembles or are close outliers to those ranges.

Forecasts as produced for the 1998 March to May season are derived from nine-member ensemble runs, but with two major design differences from the predictability experiments outlined above. First, initialization for the predictions is from 7-9 February rather than late February as in the PROVOST runs. Secondly the real-time experiments use persisted SST anomalies (from January) throughout. Neither is thought likely to have significant negative impacts on the model's ability to provide real-time predictions. Skill over the region as deduced from the PROVOST runs remains high throughout the year, whether for months 1 to 3 or 2 to 4 of the simulations. Indeed equivalent levels of skill tend to be present on a monthly time scale, although with some drop-off into the fourth month. Hence the shift in start date is considered unlikely to affect potential predictability for the region from that deduced for the 'standard' seasons. start date is considered unlikely to affect potential predictability for the region from that deduced for the 'standard' seasons. Experiments for 12 boreal spring seasons have been carried out to test the impact of using persisted anomalies. While there is some inevitable loss of predictability associated with the use of persisted anomalies this appears to be minimal in areas of relatively high predictability such as considered here, and certainly does not eliminate predictability in terms of the levels normally associated with seasonal forecasts (Fig. 2). Use of persisted anomalies fails, of course, during seasons in which there is a substantial readjustment of SST anomalies over ocean areas related to a given region's rainfall; experience has been gained of such failed forecasts for the Nordeste in preliminary work with the model. Currently there is no solution to this problem of rapid intraseasonal SST anomaly distribution changes: the forecasts given below are conditional on the continuity of the January anomalies.

The ensemble provides a consistent prediction of below-average rainfall during March to May 1998 across the region except over the north-east of Brazil (area D - above average in nearly all members) - Fig. 3 and Table 1). A southward displacement of the ITCZ from its model climatological location is suggested. Inflated ensemble-mean, together with maximum and minimum, rainfall anomalies for each of the gridded areas depicted in Fig. 1, are listed in Table 1 along with the Fortaleza-Quixeramobim (FQ) rainfall index. FQ predictions are also expressed as quintiles (five equi-probable categories relative to normals), with all members in the above-and well-above normal categories. Note that the gridded prediction for the Nordeste is somewhat less than for the indices, mainly because the block includes signal from the oceanic dry area representing the shift in ITCZ location (Fig. 3).

For the third time since this research began the dynamical model forecasts for the Nordeste rainfall indices are entirely inconsistent with those from the empirical techniques developed at UKMO (Colman et al., 1998), techniques with an extended history of high skill. Examination of the empirical methods indicates that most of the information this season is being extracted from the ENSO predictor, with a sign associated with dry Nordeste conditions. An Atlantic dipole predictor is also included in the empirical model, but this is currently weak and favors above average rainfall. Scrutiny of Atlantic SST anomalies in January 1998 reveals a region of warm anomalies extending eastward from the Nordeste/Bahia coastal regions. Brief analysis suggests that anomalies of this sign in this region are typically associated with above-average Nordeste rainfall. It is hypothesized, therefore, that the difference in the empirical and dynamical predictions may result from the dynamical model's ability to respond to SST anomalies off Bahia where as the UKMO empirical methods are tuned only to basin-scale anomalies and he nce do not give much weight to the relatively localized effects from the Bahia region.

References 

Colman, A. et al., 1997: Multiple Regression, Discriminant Predictions of Mar-Apr-May 1997 in Northeast Brazil. NOAA Long-Lead Bulletin, March 1997, 6,29-32. 

Colman, A. et al., 1998: Multiple Regression, Discriminant Analysis and AGCM Predictions of Mar-Apr-May 1998 Rainfall in Northeast Brazil, this volume.

Harrison, M. et al., 1997: A Dynamical One-Month Lead Seasonal Rainfall Prediction for March to May 1997 for the North-Eastern Area of South America. NOAA Long-Lead Bulletin, March 1997, 6,25-28. 

Hulme, M., 1994: Validation of large-scale precipitation fields in general circulation models. Global Precipitation and Climate Change, M. Desbois and F. Desalmand, Eds., NATO ASI Series, Vol. 23, Springer-Verlag, 387-406.

Table 1. March to May 1998 seasonal forecast rainfall percentages of normal for the unmodified Ensemble Mean (E Mean - with respect to the model 1979-1993 climate) and inflated Mean (with respect to the Hulme data for 1979-1983 - I Mean) and for the highest and lowest (inflated) ensemble members for each of the 6 areas depicted in Figure 1. Also shown are equivalent predictions for the Fortaleza-Quixeramobim (FQ) rainfall index; (see Colman et al., 1997), inflated using the Hulme observed dataset, estimated from the closest two and four model grid points; the equivalent Hulme area is D. In the FQ columns, the forecasts are expressed as quintiles (five equi-probable categories relative to normals): Q1:well-below, Q2:below, Q3:normal, Q4:above, Q5: well-above.