We chose the hidden Markov random field (HMRF) to model the road network (in NRI imagery) over time. The HMRF emerged as the natural means to leverage spatio temporal contextual information in our goal to improve the road network predicted by our neural network model, and define “change” in the road network over time. We start the seminar by defining the HMRF and then review the main estimation methods. Estimation is typically via the EM algorithm, preceded by either pseudolikelihood or mean-field-like model approximation. We culminate with an update of our ongoing implementation. Notably, our implementation is — to the team’s current knowledge — the first of its kind in two key ways: each node in the underlying undirected graph is a road segment, as opposed to a pixel; we hope incorporating such topological information will improve performance, and the observed variable y will likely be multivariate and at least include (1) a preliminary binary estimate of the road network graph and (2) conformal prediction output, likely in functional form. We hope to eventually submit this methodology for publication, at least as a general means for extracting network-like features from image time series.