In many areas of the Western United States, atmospheric rivers (ARs) and associated patterns of large-scale atmospheric circulation determine average and extreme precipitation. Managing the impacts of these climate phenomena on communities and ecosystems requires the understanding of their natural variability and potential for long-term change, so that natural resource managers can better tailor adaptation strategies to address long-term climate shifts without over-adapting to natural fluctuations that could revert to average conditions. However, limited instrumental records complicate the quantification of this natural variability. This dissertation seeks to address this challenge across three separate studies employing novel data-driven analytics, networks of tree-ring chronologies across the Western US, and large climate model ensembles. In the first study, I develop a six hundred year-long reconstruction of extreme precipitation in the Sacramento River Watershed in California based on penalized regressions against a tree-ring based gridded standardized precipitation index (SPI). Results show a high prevalence of extreme precipitation events in the 1600s and the late 1800s, with a decadal frequency prominent from the mid-1500s onward. In the second study, I use neural networks to estimate the daily probability of AR occurrences throughout the first half of the 20th century, and then use the tree-ring based gridded SPI to reconstruct the annual frequency of AR landfalls in three regions of US west coast back to 1400 CE. Results highlight how the extensive network of precipitation gages and tree-ring chronologies across the Western US can help capture key features of past AR activity, including two modes of low-frequency variability in AR landfalls and an undocumented, multi-decadal signature in the latitudinal variability of AR landfalls across the west coast. In the third study, I explore the past variability and potential future change in large-scale patterns of atmospheric circulation (i.e., weather regimes), some of which are directly related to the frequency of ARs. I quantify natural variability, model bias, and future changes in multiple features of WRs over the Pacific North American sector using a large ensemble climate simulation. Results suggest little evidence for long-term change in most WRs features, except for an intensification of their spatial patterns.