Regional climate is influenced by interactions between large-scale (LS) circulation and the underlying land surface. Such interactions are more complex in the coastal mountain regions where terrain is also critical in shaping the local hydroclimate conditions. Understanding the impact of LS drivers and terrain effects can help improve understanding of the hydroclimate predictability in these regions, but gaps remain due to the strong non-linearity embedded in their connections. In this study, we apply multiple machine learning models to the Puget Sound region of the US Pacific Northwest and reveal the predictive skills of LS drivers in this region. A variety of LS drivers are used to predict daily precipitation (P) and near-surface temperature (T2) conditions at various spatial scales (regional, elevation band, grid scale) during 1981-2020: indices of El Niño/Southern Oscillation (ENSO), Madden-Julian Oscillation (MJO), and Pacific Decadal Oscillation (PDO), and daily (IVT). Both random forest and deep learning models exhibit similar skills in this region, with R² of ~0.7 for daily T2 and accuracy of ~0.82 for extreme daily P prediction. The maxima of predictive skills are achieved at 1500m elevation. IWV and IVT are more relevant to local temperature and extreme precipitation, respectively, consistent with the role of atmospheric rivers on extreme precipitation. While IWV/IVT partially reflect the impact of ENSO/MJO/PDO, models still reveal additional predictive skills of these drivers. Analyses of these models indicate that ENSO has greater control over T2 while MJO has more controls on P, especially its extremes. Our results may be used to guide the construction of more specific and effective prediction models for the Puget Sound region. Meanwhile, the presented analysis framework can be used to explore such additional predictability from LS drivers in different geographic regions.