Quantifying Sources of Subseasonal Prediction Skill in CESM2 Within a Perfect Modeling Framework

The success of numerical weather prediction depends on accurate atmospheric initialization, but at subseasonal lead times, land and ocean initial states become increasingly important. Predictability on these timescales arises from slowly evolving land surface conditions such as soil moisture and snowpack, convectively coupled waves such as the Madden–Julian Oscillation and from oceanic variability including the El Niño–Southern Oscillation. While operational systems provide skillful subseasonal-to-seasonal forecasts, it remains uncertain whether this skill can be extended or if it reflects the intrinsic predictability limit. Using the Community Earth System Model in a perfect modeling framework, we estimate the theoretical limit of subseasonal-to-seasonal predictability from initialization. We find that over land, land initialization is the dominant source of predictability beyond week four, while ocean initialization plays a secondary role. Although the perfect modeling framework has limitations, our results suggest substantial potential to advance prediction through improved land initialization and representation of land–atmosphere coupling.