Combining the strengths of statistical and dynamical modeling approaches for forecasting Australian seasonal rainfall

Abstract

[1] Forecasting rainfall at the seasonal time scale is highly challenging. Seasonal rainfall forecasts are typically made using statistical or dynamical models. The two types of models have different strengths, and their combination has the potential to increase forecast skill. In this study, statistical-dynamical forecasts of Australian seasonal rainfall are assessed. Statistical rainfall forecasts are made based on observed relationships with lagged climate indices. Dynamical forecasts are made by calibrating raw outputs from multiple general circulation models. The statistical and dynamical forecasts are then merged using a Bayesian model averaging (BMA) method. The skill and reliability of the forecasts is assessed through cross-validation for the period 1980-2010. We confirm that the dynamical and statistical groups of models give skill in different locations and seasons and the merged statistical-dynamical forecasts represent a significant improvement in terms of maximizing spatial and temporal coverage of skillfulness. We find that the merged statistical-dynamical forecasts are reliable in representing forecast uncertainty.Citation: Schepen, A., Q. J. Wang, and D. E. Robertson (2012), Combining the strengths of statistical and dynamical modeling approaches for forecasting Australian seasonal rainfall,

Using subseasonal-to-seasonal (S2S) extreme rainfall forecasts for extended-range flood prediction in Australia

“…addressing issues such as model selection and calibration, observed catchment wetness, lagged and real-time forecasts, uncertainties around extremes, bias correction, requirements for high-resolution rainfall forecasts, integration with communication and response systems -is an active area of research in Australia (e.g. Cuo et al, 2011;Tuteja et al, 2011;Wang et al, 2012;Schepen et al, 2012b).…”

The challenge of forecasting high streamflows 1–3 months in advance with lagged climate indices in southeast Australia

“…In addition, we note that the skill of statistical forecasts may complement that of dynamical rainfall forecasts (e.g. the statistical rainfall forecasts may exhibit skill in different seasons or locations to dynamical forecasts; Schepen et al, 2012b), and that merging forecasts of high rainfalls from dynamical and statistical models may improve overall skill. Using climate indices derived from SST forecasts from coupled ocean-atmosphere dynamical climate models shows promise in improving forecasts of monthly rainfall totals at lead times of more than six months (Hawthorne et al, 2013), and avoids the use of lagged climate indices for forecasting.…”

“…At present dynamical climate models do not necessarily exhibit more skill than statistical forecasts of seasonal precipitation (e.g. Schepen et al, 2012b). Future improvements in dynamical climate models used for forecasting weeks to months advance (e.g.…”

Does objective cluster analysis serve as a useful precursor to seasonal precipitation prediction at local scale? Application to western Ethiopia

“…Additionally, statistical models are often linear by construction and may not well capture non-linear complex interactions and feedbacks. The physical nature of dynamical models, however, allows for prediction under non-stationary conditions and also when insufficient historical data are available, whereas statistical models, by construction, typically rely on stationary relationships (Schepen et al, 2012).…”