Hidden Tropical Pacific Sea Surface Temperature States Reveal Global Predictability for Monthly Precipitation for Sub‐Season to Annual Scales.

We present the first global precipitation predictability estimates corresponding to the recently discovered flavors of El Niño Southern Oscillation (ENSO) that are encoded in the hidden states of Tropical Pacific sea surface temperatures identified using a non‐homogeneous hidden Markov model. For each calendar month and for each hidden state, we assess future precipitation predictability through the conditional standardized anomaly of the average and the standard deviation of monthly precipitation, at 1, 3, 6, 9, and 12 months lead times. We find statistically significant potential predictive skill for key regions for each hidden state and calendar month, even for 12‐months in the future. We apply the algorithm sequentially over the period of record to identify regions that can be consistently predicted for different lead times and calendar months. The cross‐validated correlation skill is demonstrably superior to that of regression with an ENSO index used in the same way. Plain Language Summary: El Niño Southern Oscillation (ENSO) information is routinely used by hydrologists, agriculturalists and others as a useful prognosticator of upcoming precipitation. Since the characterization of ENSO has recently been broadened to consider different "flavors," the NINO3.4 correlations used by many people may not be the most informative. Our machine learning model identified five hidden states of the tropical Pacific sea surface temperatures that correspond to the different ENSO flavors. We explore how future monthly precipitation may be predictable knowing which hidden state is currently identified and show that this leads to a higher predictability for up to almost a year in the future for many more calendar months in more regions of the world than NINO3.4 correlations. Key Points: Predictability of monthly global rainfall using hidden states of Sea Surface Temperatures in the tropical Pacific Ocean is demonstratedHighly significant skill for many global regions and months is identified by sampling of means and deviations given hidden stateA sequential application of the method leads to much higher cross‐validated correlations than those based on regression on an El Niño Southern Oscillation index [ABSTRACT FROM AUTHOR]