Accurate year‑to‑year rainfall prediction remains an important yet unresolved challenge with direct implications for water management. Here, we explore rainfall persistence — the tendency for wet or dry years to follow one another—as a benchmark for evaluating climate models. Comparing observations of annual-mean rainfall and temperature against outputs from state‑of‑the‑art climate model simulations, we find that observations exhibit greater interannual persistence than do models.

We then link differences in persistence at the point-scale to larger-scale climate variability. Through statistical techniques, we attribute roughly one‑third of how temperature and rainfall co-vary to large‑scale factors: anthropogenic warming, the El Niño‑Southern Oscillation, and north-south shifts in the position of the tropical rain belt. After removing these covarying components, the residual time series exhibit near-zero temperature autocorrelation and a weakly-negative precipitation autocorrelation —yet it does explain the offset in rainfall persistence between models and observations.

These findings suggest that real‑world rainfall may possess greater inherent predictability than current climate models simulate. Key open questions include identifying the physical mechanisms (e.g., cloud organization, soil moisture feedbacks?) responsible for enhanced persistence, understanding the drivers of the residual negative autocorrelation in rainfall, and evaluating whether higher‑resolution climate models better simulate rainfall memory.