IRI@AGU: Capturing ENSO Predictability
This post is the first in a series of Q&As with scientists from the International Research Institute for Climate and Society who will be presenting their work at the annual meeting of the American Geophysical Union in San Francisco December 9 to 13.
Many researchers focus on the ability to predict El Niño-Southern Oscillation (ENSO) conditions up to 12 months in advance. Such research improves seasonal prediction forecasts in many regions, which in turn are used for applications like disease early warning systems, agriculture planning, and reservoir management. Scientists have long been discussing the limits that prevent ENSO prediction at multiannual to decadal timescales. Prediction at this scale could allow for even more preparedness by decision makers.
IRI’s Paula Gonzalez and co-author Lisa Goddard are studying the possibility of longer-term ENSO prediction using a new set of model predictions. Read the Q&A with Gonzalez below to learn more, and check out her poster if you’re going to AGU.
What are some of the limiting factors to increasing the lead-time of ENSO prediction?
The limits to ENSO predictability are still debated. Two main factors are thought to limit the skill of long-lead model predictions: the growth of initial errors and the role of the stochastic forcing or ‘random weather noise’ in the initiation of ENSO events.
If the latter plays a greater role, we can’t expect the models to predict this triggering noise many years in advance. If, however, the limit is controlled by the growth of initial errors, there is room for longer-lead predictions. A better initialization of the models, with a more complete representation of the current state of the climate system, can potentially result in better predictions. The production of decadal predictions from a collection of state-of-the-art dynamical climate models is a new activity, so examination of the processes that could lead to multi-year prediction of big ENSO events is only now possible.
Why does the longer forecast matter to decision makers?
It is well known that ENSO events (both El Niño and La Niña) yield significant impacts for society throughout the world, through increased frequency of floods and droughts, epidemics, loss of crops and the subsequent impact on food security, etc. In this context, being able to predict an event more than six months in advance would be extremely useful. And even if current models wouldn’t allow one to pinpoint the exact year an event occurs but could assess whether the forthcoming decade is likely to experience intense/weak ENSO activity, that information has the potential to be incorporated in near-term decision-making. Most of my work has focused in southeastern South America, one of the largest grain-exporting regions in the globe. This region typically exhibits large crop and cattle losses during La Niña events, and increased production during El Niño events. Since the economies of most of the region’s countries rely largely on the success of agricultural production, being able to project drier or wetter years in the next decade would allow the farmers and the government to plan ahead and minimize negative impacts.
What are the new models you’re using, and how are they different from previous studies?
This study has been focused on analyzing the properties of ENSO variability in a set of new experiments developed for the Coupled Model Intercomparison Project Phase 5 (CMIP5), known as decadal hindcasts. In contrast with the long-term ‘climate change’ projections that we are more used to, the ocean conditions in these experiments have been initialized every year starting with 1960 and are each run for 10 years, with the aim of exploring the potential for decadal prediction. The model simulations used for ENSO prediction are usually not longer than 12 months, which has limited the scope of the ENSO predictability studies. In this context, these new set of 10-year long hindcast provides a unique opportunity for extending beyond that temporal boundary.
Could you briefly explain how you use models to test the long-term predictability?
This initial stage of our analysis focused on exploring the variability of sea surface temperature (SST) in a region of the tropical Pacific normally used to describe ENSO activity. The study compares the properties of the regional time series for each model with the ones in observational datasets. The simulated time series requires some pre-processing known as bias correction, which accounts for the fact that some models might drift towards the model’s climatology instead of the actual observations. To assess how well the models recreate past observations, we concentrate on certain parameters such as the average magnitude of the prediction error and the hit rates and false alarm rates in the detection of ENSO events.
Based on your results, what is your outlook for decadal predictability of ENSO conditions?
Our results suggest that ENSO predictability might extend beyond the suspected time horizons. We found that some models showed signs of the major ENSO events up to four or five years in advance. However, a lot more research is required to assess whether this potential predictability can be translated into actual prediction skill in operational ENSO forecasting. Nonetheless, we are hopeful that given the tremendous beneficial impact that this type of prediction could have, there will be room for the required research in near future.
What will be your next steps in this research?
Since our study has been based solely on the variability of SST in the tropical Pacific, our next step is to analyze the properties of the sub-surface oceanic variability in the models that show some evidence of multi-year predictability and provide a physical explanation for the multi-year prediction ability for certain events. Finding reasonable precursors and evolutions in the sub-surface life of these events would provide greater confidence that there is potential for these multi-year predictions.