2002 Annual

Monday, 14 January 2002: 1:29 PM
Diagnostic Verification of the IRI Net Assessment Forecasts, 19972000
Daneil S. Wilks, Cornell Univ., Ithaca, NY; and C. M. Godfrey
The International Research Institute (IRI) for Climate Prediction produces, four times per year, operational outlooks for seasonal (3-month periods) average temperature and for total precipitation over global land areas, at lead times of zero and three months. These outlooks are probabilistic in nature, and the forecast quantities are 3-dimensional vectors specifying probabilities of temperature or precipitation outcomes at particular locations falling in the lower, middle, and upper thirds of the climatological distributions for those locations.

The performance of these forecasts for the period 1997 through 2000 are assessed through a diagnostic verification, which involves examination of the full joint frequency distributions of the forecasts and the corresponding observations, rather than relying on a distillation of forecast performance in terms of simple scalar scores. In particular, the calibration-refinement factorization of these joint distributions will be examined graphically, through reliability diagrams. These allow determination of the degree of forecast calibration (correspondence between forecast probabilities and event relative frequencies), forecast sharpness, and any conditional (consistently over- or underconfident) or unconditional (e.g., consistently too warm) biases that may be present. Detailed analyses such as these are necessary for fully optimal use of such forecasts in decision making. In addition to overall verification results for each of the two forecast projections, results will be presented for geographic stratifications of the data.

Supplementary URL: