Higher-order internal modes of variability imprinted in year-to-year California streamflow changes

Climate internal variability plays a crucial role in the hydroclimate system, and this study quantifies its predictability on streamflow in California using historical observations, climate simulations, and various machine learning (ML) models. Here we demonstrate that while 5% of the year-to-year variability in seasonal peak streamflow can be attributed to the well-known climate variability indices, the explained variance surpasses 30% when higher-order empirical orthogonal functions of these indices are retained in the analysis. Notably, the results highlight the significant influence of the 5th empirical mode of the Pacific North American pattern and of the Pacific Decadal Oscillation in shaping the streamflow variability, which is consistent across all the tested ML models. A deeper investigation reveals a clear and monotonic quasi-linear response of streamflow to these dominant patterns, emphasizing the substantial role played by higher-order internal modes of variability in shaping regional hydroclimate systems, which contributes to bridging the gap between the well-known variability domains and local climate systems. Machine learning techniques informed by CMIP6 models and historical observations suggest that higher order internal climate variability had a significant influence on streamflow, and therefore water resources, in California between 1981 and 2015.