Overview
Each month CCSR issues a forecast of the El Niño–Southern Oscillation (ENSO) for the next nine overlapping three-month seasons. The first season starts in the month the forecast is issued, so a forecast issued in February covers February–April (FMA) through October–December (OND).
ENSO is tracked with the Niño 3.4 index: the sea surface temperature (SST) anomaly averaged over the central-eastern equatorial Pacific (5°N–5°S, 170°W–120°W). Seasons fall into three categories:
| Category | Niño 3.4 anomaly |
|---|---|
| La Niña | −0.5 °C or below |
| Neutral | between −0.5 °C and +0.5 °C |
| El Niño | +0.5 °C or above |
The Model Plumes
Traditional Oceanic Niño Index (TONI) Forecast Plume
The forecast starts from a collection of Niño 3.4 predictions made by modeling centers around the world, known as the ENSO plume. The models are of two kinds:
- Dynamical models simulate the physics of the coupled ocean–atmosphere system.
- Statistical models forecast from relationships found in the historical record.
Some models predict monthly values. These are converted to three-month running means so that every model forecasts the same nine seasons. If a model doesn't provide a forecast for a particular season, it's left out of that season's calculations.
Relative Oceanic Niño Index (RONI) Forecast Plume
The Relative Oceanic Niño Index (RONI) measures Niño-3.4 SST anomalies relative to the average SST anomaly across the tropical oceans, offering a complementary perspective on ENSO strength. The RONI plume combines dynamical and statistical model forecasts, displaying individual predictions and the multi-model mean across three-month seasons.
Why RONI? When tropical ocean temperatures warm or cool broadly, RONI can tell a different story from conventional Niño-3.4 about the relative strength of El Niño or La Niña.
The RONI plume is shown separately from the Niño-3.4 plume and is based on a limited number of models, providing an additional perspective on ENSO’s expected evolution.
Multi-Model Average
For each season, the forecasts from all available dynamical and statistical models are averaged with equal weight. This multi-model mean is the forecast's best estimate of the most likely Niño 3.4 value. Averaging many independent models usually does better than any single model, because their separate errors partly cancel out.
The plume also shows separate averages for the dynamical and statistical models.
From a Forecast Value to Probabilities
A single number doesn't say how confident the forecast is. To produce probabilities for La Niña, Neutral and El Niño, each season's forecast is treated as a normal (bell-curve) distribution:
- Center: the multi-model mean for that season.
- Width (uncertainty): depends on how well the models have predicted that season in the past at that lead time, and on how much Niño 3.4 naturally varies in that season.
The width is calculated as:
σ = σobs × √(1 − r²)
where:
- σobs is the observed year-to-year standard deviation of Niño 3.4 for the target season. ENSO is much more variable in late fall and winter than in spring. The standard deviation ranges from about 0.6 °C for April–June to about 1.3 °C for November–January.
- r is the historical correlation between the multi-model forecasts and what was later observed, for that issue month and lead time.
When past skill is high (r close to 1), the curve is narrow and the probabilities are decisive. When skill is low, the curve widens and the probabilities move toward each other. Skill is generally lowest:
- at longer lead times, toward the end of the nine-season outlook, and
- for forecasts made in spring that look past the Northern Hemisphere spring. This is the well-known spring predictability barrier: ENSO forecasts made before or during March–May are noticeably less reliable.
The probability of each category is the share of the bell curve that falls in that category's range:
- La Niña: the part below −0.5 °C
- Neutral: the part between −0.5 °C and +0.5 °C
- El Niño: the part above +0.5 °C
Probabilities are rounded to whole percentages. If the rounded values don't add up to 100%, the most likely category is adjusted so the total is exactly 100%.
ENSO Strength Outlook
To show how strong an El Niño or La Niña might be, the individual dynamical and statistical model forecasts are also sorted into nine Niño 3.4 strength ranges:
| Strength | Niño 3.4 anomaly |
|---|---|
| Very Strong La Niña | −2.0 °C or below |
| Strong La Niña | −2.0 to −1.5 °C |
| Moderate La Niña | −1.5 to −1.0 °C |
| Weak La Niña | −1.0 to −0.5 °C |
| Neutral | −0.5 to +0.5 °C |
| Weak El Niño | +0.5 to +1.0 °C |
| Moderate El Niño | +1.0 to +1.5 °C |
| Strong El Niño | +1.5 to +2.0 °C |
| Very Strong El Niño | +2.0 °C or above |
For each season, the chart shows the percentage of models that fall in each range. This is a count of how the models are distributed, not a calibrated probability. It shows how much the models agree on strength, but isn't adjusted for past skill the way the category probabilities above are.
Prediction models
These are the models currently used in the Niño-3.4 SST plume. The traditional models are grouped as dynamical or statistical; the relative models forecast the Relative Oceanic Niño Index (RONI) and are shown separately on the Plumes page. Each model links to a description of the model. Please contact forecast-ccsr-help@columbia.edu with any questions.
Dynamical models
| Model | Description |
|---|---|
| AUS-ACCESS | CSIRO Australian Community Climate and Earth System Simulator |
| BCC DIAP | Beijing Climate Center Dynamical Inter-Annual Prediction System |
| CMC CANSIP | Government of Canada CanSIPS |
| CMCC SPS4 | Centro Euro-Mediterraneo sui Cambiamenti Climatici |
| COLA CCSM4 | University of Miami |
| CS-IRI-MM | University of Pretoria, Willem Landman |
| GFDL SPEAR | NOAA GFDL |
| IOCAS ICM | Institute of Oceanology, Chinese Academy of Sciences |
| JMA | Tokyo Climate Center |
| KMA | Korean Meteorological Administration |
| LDEO | Lamont-Doherty Earth Observatory, Columbia University |
| MetFRANCE | Météo-France Seasonal Forecast |
| NASA GMAOv3 | NASA GMAO GEOS-S2S |
| NCEP CFSv2 | NOAA Climate Prediction Center NCEP CFSv2 |
| SINTEX-F | Japan Agency for Marine-Earth Science and Technology |
| UKMO | UK Met Office GloSea6 |
Statistical models
| Model | Description |
|---|---|
| BCC_RZDM | Beijing Climate Center RZDM |
| CPC MRKOV | NOAA CPC Markov |
| CSU CLIPR | Colorado State University |
| IAP-NN | Institute of Applied Physics of the Russian Academy of Sciences |
| JAMSTEC CNN | Japan Agency for Marine-Earth Science and Technology (JAMSTEC) |
| NTU CODA | Taiwan Computational Oceanography and Dynamics of Air-Sea Interaction Lab |
| TONGJI-ML | Mean of three Tongji University deep learning models: ENSO-ASC, ENSO-GTC and ENSO-MC |
| UCLA-TCD | UCLA Theoretical Climate Dynamics |
| UW PSL-CSLIM | University of Wisconsin |
| UW PSL-LIM | University of Wisconsin |
| Wyrtki-CSLIM | University of Hawaii Sea Level Center |
| XRO | University of Hawaii, School of Ocean and Earth Science and Technology (Nature article); statistical/dynamical |
Relative (RONI) models
| Model | Description |
|---|---|
| AUS-RELATIVE | Australian Bureau of Meteorology |
| CMCC CM3 RELATIVE | Centro Euro-Mediterraneo sui Cambiamenti Climatici |
| Wyrtki-CSLIM RELATIVE | Relative (RONI) version of Wyrtki-CSLIM |
| XRO RELATIVE | Relative (RONI) version of XRO |