News

Rain long foretold: the enigma of decadal forecasts

To prepare for climate change, planners and policy-makers need reliable local forecasts, for a decade or two ahead, of key variables such as inter-annual variability in rainfall, or length of the growing season. Photo: N. Palmer (CIAT)

    by Sonja Vermeulen

    Ten to twenty years is the typical timeframe for designing and implementing many of the interventions most critical to agriculture: new crop varieties, or catchment-wide infrastructure for irrigation and water storage, or siting and establishment of major new processing hubs.  Good design of any of these depends on knowing what the climate will be like once they are up and running.  What planners and policy-makers need are reliable local forecasts, for a decade or two ahead, of key variables such as inter-annual variability in rainfall, or length of the growing season. 

    The bad news is that impatient end-users are likely to wait some years before “good-enough” decadal forecasts are available for most regions. To understand why, Decadal prediction skill in a multi-model ensemble, by Geert Jan van Oldenborgh, Francisco Doblas-Reyes, Bert Wouters and Wilco Hazeleger, offers insights into the challenges and opportunities in the young field of decadal forecasting.  “Skill” is meteorologists’ term for how well a model performs compared to some defined alternative.  In this case, the paper appraises the skill of an ensemble of climate models to predict retroactively – or hindcast – temperature and precipitation between 1959 and 2009. 

    Our climate is changing continuously due to a combination of anthropogenic impacts, intrinsic variability, and natural events such as volcanic eruptions.  Different factors are more important to skill at different timescales.  For seasonal forecasts, in many areas variability outweighs any long-term trend, and so initial conditions matter most.  For centennial forecasts, where the trend is readily distinguishable from variability, skill depends mostly on prescribed boundary conditions, for example scenarios of emissions of greenhouse gases and aerosols.  Decadal forecasts fall halfway.  What makes them challenging is their dependence on both variability and trend – and hence both initial and boundary conditions.

    Not surprisingly then, van Oldenborgh and co-authors’ analyses demonstrate important but different roles for boundary conditions and for initial conditions in determining the skill of multi-model ensembles.  Externally forced boundary conditions (levels of human emissions of greenhouse gases and aerosols) provide a significant level of skill in predicting mean temperature, but the models do less well at hindcasting the observed variability around the trends.  Initial conditions, specifically surface temperatures, contribute some additional skill over oceans.  The models are more reliable in predicting the strength of the trend in the global mean than in simulating local trends.  For precipitation, the multi-model ensembles do not show statistically significant skill, though there is some promise for predictions of four-year mean rainfall in the Sahel. 

    The authors make clear that their study is no more than an early foray into evaluating the skill of decadal forecasts.  They note that they have made no attempt to separate forced variability from natural variability or to make forecasts that correct the biases and drifts in the models.  Nonetheless their work opens a window to a conceivable future for CORDEX and other research programs racing to provide planners and policy-makers with reliable, meaningful decadal climate forecasts at local levels. 

     

    Links

     

    This is the March 2012 installation of AgClim Letters, a monthly e-bulletin on science and policy written by Sonja Vermeulen, Head of Research for CCAFS. Sign up to receive AgClim Letters bulletin and read past bulletins. Your comments are welcome below.