The IOD forecast is a multi-model product. Each contributing prediction system is run as an ensemble from the observed state of the ocean and atmosphere, and each ensemble member's Dipole Mode Index is computed from its predicted sea-surface temperatures – the same index definition used for the observations (see About). The figure on the Forecast page shows those per-model predictions alongside the category probabilities formed from them.
Contributing models
Six of the models are members of the North American Multi-Model Ensemble (NMME), retrieved from the IRI Data Library; CMCC publishes its IOD forecast directly. Every model is shown out to six months ahead of the initialization month.
| Model | Centre and system | Hindcast climatology |
|---|---|---|
| CanESM5 | Environment and Climate Change Canada – CanESM5 (CanSIPS) | 1990–2024 |
| CESM1 | COLA / University of Miami RSMAS – CESM1 | 1982–2020 |
| CFSv2 | NOAA NCEP – Climate Forecast System v2 | 1982–2020 |
| CMCC | Euro-Mediterranean Center on Climate Change – SPS4 | 1993–2020 (applied by CMCC) |
| GEM-NEMO | Environment and Climate Change Canada – GEM5.2-NEMO (CanSIPS) | 1990–2024 |
| GFDL-SPEAR | NOAA GFDL – SPEAR | 1991–2020 |
| NASA | NASA GMAO – GEOS-S2S | 1981–2017 |
The set of models available changes from month to month, since a centre may skip an issuance or retire a system. The model count and labels shown on each forecast figure are therefore the authoritative record of what went into that particular issuance.
How the models are combined
Each model's DMI is an anomaly from that model's own hindcast climatology, taken by target month and lead time, so that a model with a persistent warm or cold drift in either pole does not bias the combined forecast. (GFDL-SPEAR's climatology is taken by initialization month; CMCC's anomalies arrive already computed.) The multi-model mean in the plume is the equal-weighted average of the models' ensemble means.
The category probabilities are computed for each model first – the share of its ensemble members below −0.4 °C, between −0.4 and +0.4 °C, and above +0.4 °C – and those probabilities are then averaged across models, so each model counts equally regardless of its ensemble size. CMCC publishes only an ensemble mean, which gives no spread to turn into probabilities, so it appears in the plume and the multi-model mean but not in the probabilities. The climatological probabilities shown alongside are from observations over 1991–2020.
Reading the plume
Each line on the figure is one model's ensemble-mean DMI prediction by target month; the spread between lines is the disagreement between prediction systems, which is generally the larger source of uncertainty at longer leads. Agreement between models is a necessary but not sufficient condition for confidence – the systems share observational inputs and, in places, components, so their errors are not independent.