We developed Bayesian statistical approaches to assess non-perennial stream network connectivity. Our new methods allow: 1) consideration of changes to both local (stream segment) and global (stream network) connectivity over time, 2) incorporation of prior information from different data sources, and 3) straightforward computation of the posterior distributions of both active stream length and a new metric called communication distance. Communication distance measures the effective stream length for the movement of materials, including water and solutes, from upstream to downstream sites. Communication distance posteriors require the inverse-beta probability density function whose form had not been previously derived. The inverse-beta distribution can be used to represent the rarity of surface water presence compared to a perennial stream, thus clarifying bottlenecking propensities for stream segments. As an application, we considered Murphy Creek, a simple stream network in southwestern Idaho, USA. Our models used surface water presence/absence data from 2019, and priors based on existing regional USGS model predictions for surface water. Murphy Creek probabilities for surface water presence were heterogeneous in space and time, and were likely driven by fine-scale spatial variations in shallow subsurface hydraulic conductivity. Strong seasonal (spring, summer, fall) temporal differences were evident in network-level posterior distributions of both stream length and communication distance. Specifically, stream lengths were shorter and more variable in the summer and fall than in the spring. The novel communication distance posteriors were multimodal, platykurtic, and negatively skewed for spring, summer and fall, respectively, revealing bottlenecking effects that varied over time.