Seamless seasonal to multi-annual climate predictions by constraining transient (CMIP6) climate model simulations

Seamless climate predictions combine information across different timescales to deliver information potentially useful for sectors like agriculture, energy, and public health. Seamless operational forecasts for periods spanning from sub-annual to multi-annual timescales are currently not available throughout the year. We show that this gap can be closed by using a well-established climate model analog method. The method consists in sampling model states from the CMIP6 transient simulation catalog based on their similarity with the observed sea surface temperature as a means of model initialization. 

Here we present the methodology and basic skill evaluation of the analog-based temperature and standardized precipitation index retrospective predictions with forecast times ranging from 3 months up to 4 years. We additionally compare these predictions with the non-initialized CMIP6 ensemble and with two operational benchmarks produced with state-of-the-art dynamical forecasts systems: one on seasonal timescales and the other on annual to multi-annual timescales.

The analog method provides skillful climate predictions across the different timescales, from seasons to several years, offering temperature and precipitation forecasts comparable to those from state-of-the-art initialized climate prediction systems, particularly at the annual to multi-annual timescales. However, unlike operational decadal prediction systems that provide only one or two initializations per year, the analog-based system can generate seamless predictions with monthly initializations, offering year-round climate information. Additionally, analog predictions are computationally inexpensive once the multi-model transient climate simulations have been completed. We argue that these predictions are a valuable complement to existing operational prediction systems and may improve regional climate adaptation and mitigation strategies.