Researchers used neural networks to investigate if subseasonal forecasts of opportunity for U.S. West Coast precipitation provided by tropical precipitation vary on decadal timescales. We confirm using explainable artificial intelligence (XAI) that when the neural network is confident, it is identifying physically relevant regions associated with specific modes of variability (e.g. MJO and ENSO), with decadal variability in subseasonal prediction skill throughout the historical simulation and recent observational record.
We find that tropically-driven subseasonal predictability varies on decadal timescales during forecasts of opportunity. These results further indicate that the neural networks are capable of identifying predictable decadal states of the climate system within CESM2 that are useful for making confident, accurate subseasonal precipitation predictions in the real world.
Identifying forecasts of opportunity is useful to quantify when and where we may experience enhanced subseasonal prediction skill. This research utilizes an explainable neural network to quantify subseasonal prediction skill of U.S. West Coast precipitation provided by tropical precipitation and further, identifies and explores decadal variability of this skill in the CESM2-LE and reanalysis. These results indicate that subseasonal predictability during forecasts of opportunity varies on decadal timescales, and therefore, it is important to identify when we may experience reduced or enhanced skill. We demonstrate that neural networks can be a useful tool to explore the long-term variability of predictability and through explainability techniques, explore the earth system conditions that may contribute to these changes in predictability. Further, the research reveals that the subseasonal skill provided by certain tropical modes of variability may vary due to decadal modulation in the teleconnectivity between the tropics and extratropics in both CESM2 and reanalysis.