At the height of 1997's very strong El Nino episode, IRI's prediction division began generating seasonal climate forecasts on a real-time basis. To produce these forecasts an extensive suite of forecasting tools was developed, including a multimodel ensemble approach that required thorough validation of each model's accuracy level in simulating interannual climate variability. The IRI's forecast system was developed as two-tiered, in which forecasts of global tropical SSTs are generated first and then used as boundary forcing for the suite of atmospheric models that respond to the patterns of SST anomaly in terms of resulting surface temperature and precipitation anomaly patterns. Forecasts of precipitation and air temperature are expressed as probabilities of each of three climatologically equally likely (i.e., tercile-based) categories called below normal, near normal, and above normal. Between 1997 and 2000, forecasts were based on three atmospheric models (ECHAM3.6 model from Max Planck Institute in Germany; MRF9 model from National Weather Service of U.S. NOAA; and CCM3.2 model from NCAR/UCAR in Boulder, Colorado). Updated versions of these models, and additional models, allowed the suite of models to expand in the early 2000s, and since late 2004 seven atmospheric models are being used. Objective multi-model ensembling, in which weighting of the forecasts of the several models is allowed to vary by season and location based on the models' historical hindcast performance, was developed and refined.
In this part of our web site we provide a range of current climate prediction information, including global probability forecasts for seasonal temperature and precipitation, the predictions of the individual tools that contribute to the final forecasts, and forecasts of the sea surface temperature, including specifically the ENSO state that is so critical to the climate forecasts. Additional, more specialized products include a dynamical model-based forecast of tropical cyclone activity in several ocean basins, a merging of the climate forecasts with recently observed climate events, and a product that provides probabilities for more detailed, user-defined precipitation outcomes for selected locations. Most of our forecasts are issued near the middle of each month (on the third Thursday of the month), and extend to the next 6 months into the future. Our final forecasts are expressed in probabilistic terms, because of their considerable uncertainty. The "shift in the odds" away from the average odds, when present in the forecasts, can be useful information for agricultural, hydrological, public health and food security planning purposes.