This work stands apart from its neighbors: nearby lie studies of space weather, rainfall seasonality, solar cycles, and El Niño forecasts. All of them touch on the predictability of natural systems, but none attempts to revise the predictability limit for a system as complex as the Indian monsoon. Until this work, the subseasonal variability of the monsoon was thought to be chaotic noise that caps seasonal forecasts at a correlation of about 0.65. The authors break through this barrier by showing that part of the 'noise' is actually signal.
They used the global coupled atmosphere-ocean model CFSv2 in four versions: standard (CONT), with improved snow physics (SNOW), with improved cloud microphysics (MPHY), and with both improvements (SNMP). For each version, they ran retrospective forecasts (hindcasts) over 30 years (1981–2010) with 16 ensemble members, totaling 1920 model-years. The predictability limit was estimated in three ways: 'perfect model' correlation (comparing one ensemble member against the rest), analysis of variance (ANOVA), and a new method based on exhaustive search over all possible subsets of the ensemble to find the maximum correlation with observations.
Key result: the correlation between forecast and observations for the best model version (MPHY) reached 0.71, exceeding the previous predictability limit of 0.65. The new method for estimating the limit, based on subset search, gave a maximum correlation of 0.82. This means the model can explain about 70% of the interannual variability of Indian monsoon rainfall, whereas previous estimates suggested 45%. They found that the variance (spread) of rainfall in the 3–5 day band (synoptic disturbances) has the strongest link to the seasonal rainfall anomaly (correlation 0.63) and to sea surface temperature in the Niño-3 region (correlation -0.51). This implies that the 'fast noise' is actually driven by the slow El Niño signal.
This work differs from its neighbors in that it does not merely improve the forecast but revises the predictability limit itself. Neighboring studies on rainfall seasonality, El Niño, and droughts use standard methods and do not question fundamental limitations. Here, the authors show that previous estimates of the limit were too low because they ignored the predictable part of subseasonal variability. This changes the view of the monsoon as a system with inherent unpredictability and paves the way for more ambitious goals in seasonal forecasting.
If the contribution of subseasonal anomalies (especially 10–20-day oscillations) to the seasonal anomaly is better represented, the predictability limit could rise to 0.9, corresponding to about 80% of explained variance. This requires continued improvements in cloud physics and ocean-atmosphere coupling. It is also important to investigate whether the minimum predictability grows when the ensemble size is increased beyond 16 members.
It will affect seasonal monsoon forecasting in India and neighboring countries, and planning for agriculture and water resources. It will also give impetus to revisiting predictability limits in other tropical regions.
Test whether the high predictability holds with other models and longer data records. Investigate the contribution of non-ENSO sources of predictability, which, judging by the correlation of 0.71 versus 0.6 for Niño-3, are also important.
For climate science, this is a step toward solving a long-standing problem: why the tropics as a whole are more predictable than extratropical latitudes, yet the Indian monsoon was an exception. The work shows that the exception was apparent, due to a misestimation of the role of subseasonal processes. An open question remains: how universal is this mechanism for other monsoon systems?
The authors compare predictable climate modes to an octopus that keeps its tentacles on the statistics of subseasonal oscillations, thereby controlling monsoon predictability. This metaphor even made it into a schematic in the paper.
The theme of monsoon prediction echoes Kim Stanley Robinson's novel 'The Ministry for the Future', where climate models and forecasts play a key role in preventing disasters. Here, though, we have a more modest but real step toward reliable forecasting.