The American Community Survey (ACS) is an ongoing survey administered by the US Census Bureau which collects social, economic, and other community data. ACS estimates are released annually, with varying spatial and temporal resolution: 5-year time periods refer to smaller municipal subdivisions, while 1-year time periods refer to larger areas. Although these estimates contain important community information and are often used in social and health studies, their spatial and temporal resolution pose various challenges: the 5-year ACS estimates might be temporally misaligned with finely resolved outcome data, whereas the coarser 1-year estimates are likely spatially misaligned with finely resolved outcome data. In this paper, we present a Bayesian hierarchical model that leverages both 1-year and 5-year ACS data and accounts for the survey sampling design to obtain estimates of community indicators at any given spatial and temporal resolution. The disaggregation is achieved by introducing a latent, point-referenced process, in turn modeled using a multi-resolution basis function expansion, which is linked to the ACS data via a stochastic model that accounts also for the survey design.