Diagnostic Verification of the IRI Net Assessment Forecasts, 1997–2000

The International Research Institute (IRI) for Climate Prediction produces operational outlooks for seasonal (3-month periods) average temperature and for total precipitation, at lead times of 0 and 3 months (Mason et al. 1999). These outlooks are probabilistic in nature and subjectively produced. During the 1997–2000 period considered here, two seasonal forecasts, at 0and 3month lead times, were produced 4 times per year. The forecast quantities are three-dimensional vectors specifying probabilities of temperature or precipitation outcomes falling in the lower (‘‘below normal’’), middle (‘‘near normal’’), and upper (‘‘above normal’’) thirds of the respective climatological distributions appropriate to particular seasons and locations; at global land (excluding Antarctica) and nearby ocean locations. For precipitation, these IRI forecasts began in late 1997, and are available for October–November–December (OND) 1997 and January–February–March (JFM) 1998 onward, for the 0and 3-month leads, respectively. The temperature forecasts began one season later, and are available from JFM 1998 and April–May–June (AMJ) 1998. The underlying information is from handdrawn maps of ‘‘probability anomalies’’ (e.g., Mason et al. 1999, p. 1864), that were subsequently digitized to latitude–longitude grids appropriate to available verification data. The precipitation forecasts are analyzed in the following after projection onto a global 2.58 3 2.58 grid, consistent with the format of the Xie and Arkin (1997) precipitation data, which begins in 1979. Similarly, the temperature forecasts were projected onto a global 28 3 28 grid to match the Ropelewski et al. (1985) temperature dataset. The climatological reference period, on the basis of which individual seasons were classified