Examination of the long-term climate records in some regions around the globe reveals persistent trends and periods of below-average rainfall extending over years to a decade or more, while other regions exhibit episodic, shorter droughts. Hence it is useful to consider the prediction of droughts on seasonal to interannual timescales and, separately, on longer decadal timescales.
B. Seasonal to Interannual Prediction
Our theoretical ability to make an explicit, reliable prediction of an individual weather event reduces to very low levels by about 10-15 days (this is called the "weather predictability barrier"), so forecasts with lead times longer than this should be couched in probabilistic terms. Consequently, a forecast with a lead time of a month or more requires a statistical basis for arriving at a set of probability estimates for the ensuing seasonal to interannual conditions. Two approaches allow us to derive these estimates. The first is based on statistical analyses of the climatic record and assumptions about the degree to which the statistics of the future record will differ from the past record. The second, and more recent, approach is based on the generation of statistics from multiple, explicit predictions of weather conditions using computer models of the climate system.
1. Forecasts Based on Empirical Analysis of the Climate Record
The fact that the earth's climate system is driven primarily by the regular rotation of the earth around the sun led to many efforts during the last two centuries to link the recurrence of droughts with cycles observed in the movements and features of heavenly bodies. Notable among these efforts were schemes based on the phases of the moon and the occurrence of sunspots. These purported linkages have been proven to be statistically insignificant, evanescent, or of little practical value. Nonetheless, there are recurring climate patterns, caused by the interacting dynamics of the earth's atmosphere and oceans, that provide some scope for prediction. The development of comprehensive climate records and the growth of computing power over the past 20 years or so have enabled a wide range of powerful statistical tools to be brought to bear to tease out these patterns and incorporate them into empirical algorithms for predicting future seasonal patterns.
One of the earliest identified and most powerful of these rhythms, apart from the annual cycle itself, is the El Niño/Southern Oscillation phenomenon, often referred to as ENSO. The robustness of ENSO-related patterns over time in the distribution of rainfall, air and sea temperatures, and other climatic variables, and the fact that the phenomenon is caused by slowly varying components of the ocean-atmosphere system, renders it useful as a predictor. ENSO-based indices (e.g., Troup, 1965; Wolter and Timlin, 1993) are the dominant predictors for statistically based seasonal prediction schemes over many parts of the globe, although other indices are now being combined with ENSO for different regions—for example, North Australia/Indonesia (Nicholls, 1984), the Indian Ocean (Drosdowsky, 1993), and the North Atlantic (McHugh and Rogers, 2001).
One of the simplest of the statistical prediction methods is based on the underlying premise that the behavior of a dominant pattern in the future climate will continue to replicate the behavior observed in the past record. A systematic scan of the record of the Southern Oscillation Index (SOI), for example, can reveal occurrences, or "analogs," when the track of the index over recent months was "similar" to the track in corresponding months in several past years (Stone and Aul-iciems, 1992).
More complex approaches for deriving empirically based forecasting schemes have been implemented in several operational forecasting centers throughout the world. A typical example is the methodology developed for the scheme used by the Australian National Climate Centre for forecasting probability ranges of seasonal (3-month) rainfall and temperatures (maximum and minimum). This methodology (Dros-dowsky and Chambers, 1998) involves:
1. Identification of predictands (e.g., rainfall and temperature) and possible predictors (sea surface temperatures representative of one or more areas).
2. Construction of the statistical model, including procedures for the optimum selection and weighting of predictors.
3. Verification or estimation of forecast skill.
Improvements in the forecast skill of such statistical schemes likely will plateau, because they are generally constrained by a limited number of useful predictors and relatively short periods of data. Most statistical methods also exhibit large variations in their skill level throughout the year—because of seasonal variations in statistical relationships between climate variables—and for particular regions. Further, if there are slow or even rapid changes of climate underway that are not adequately captured in the past record (as has indeed occurred in recent decades), it is possible that the skill of the forecasts may be lower than would be the case in a more stable climate. Despite these problems, statistically based schemes will likely remain useful and sometimes potent weapons for forecasting meteorological droughts.
2. Explicit Computer Model Predictions
Between about 1970 and 1980, the basis for generating daily weather forecasts moved from sets of empirical, observation-ally based rules and procedures to explicit predictions made by computer models of the three-dimensional structure of the atmosphere. However, in order to make similar progress in computer-based forecasting on longer time scales, it was essential to incorporate the slower contributions to variability from ocean circulations and variations of the land surface. In the last two decades, there have been significant improvements in the understanding of processes in the atmosphere and the ocean and in the way in which the atmosphere interacts with, or is coupled to, the various underlying surfaces. These advances in knowledge, combined with an expanded range of data and a massive increase in computer power, have made it possible to develop prediction schemes based on computer models that represent the entire earth/ocean/atmosphere system (e.g., Stockdale et al., 1998).
Although such schemes are still in their infancy, rapid developments are underway. For example, it is now evident that the details of a season's outcome are modulated by processes occurring on shorter, intraseasonal timescales, which may affect, for example, the timing and intensity of patterns of decreased or increased rainfall (Slingo et al., 1999; Schiller and Godfrey, 2003). Hence, efforts are being made to ensure that computer models of the coupled system can simulate and predict such short-term modes of variability. It is likely, too, that improvements in predictive skill on seasonal to interan-nual timescales, and hence improvements in prediction of droughts, will be realized from further expansions in the observational base, especially from the oceans (e.g., Smith, 2000); from the ability to generate larger prediction ensembles from individual computer models (Kumar and Hoerling, 2000); and from combined ensembles from several different computer models (Palmer et al., 2004).
Work is also underway to improve the spatial resolution at which seasonal forecasts can be made, through statistical "downscaling" techniques, through the nesting of high-resolution regional-scale climate models within coarser resolution global-scale models, and by increasing the resolution of the global models.
Despite these developments, it will never be possible to consistently generate forecasts of individual events beyond the 10-15-day weather predictability barrier. What these developments promise, however, is the generation of reliable short-term model-based "forecast climatologies" from which one can then generate probabilistic assessments of likely climate anomalies over a month, a season, or longer—for example, of conditions conducive to the onset, continuation, or retreat of drought.
Continue reading here: Time Scales
Was this article helpful?