Multi-year storage reservoirs must be managed in the face of weather and climate variability across time scales ranging from daily weather to interannual climate. While seasonal climate may contain a predictable component associated with the El Nino-Southern Oscillation (ENSO), longer time scales are not yet usefully predictable, nor is the interannual-to-interdecadal power spectrum well estimated from observed data. In addition, climate simulations from general circulation models (GCMs) are often lacking in their ability to generate realistic hydroclimate variability across time scales, especially at small spatial scales. These issues are critical for climate change adaptation planning in water management, where realistic estimates of climate and stream flow variability are required. For the Bhakra reservoir in northern India, we develop estimates of climate and stream flow variability, including the interannual-to-interdecadal power spectrum, based on (1) instrumental stream flow records of the Sutlej river, 1963-2010; (2) tree ring reconstructions of the Sutlej flow back to 1321; and (3) multi-century control simulations of precipitation-minus-evaporation made with several coupled ocean-atmosphere GCMs archived in the IPCC CMIP5 database. By comparing these observed, paleo-proxy, and GCM-based estimates, we shed light on the ability of GCMs to simulate realistic hydroclimate variability over the Indus basin, as well as on the nature of tree-ring based streamflow reconstructions. In addition to these estimates of the variability spectrum, we explore the use of a nonlinear, multi-level stochastic polynomial inverse model to bridge between these different datasets.