This chapter considers transferring some of the most relevant among the frameworks for a specific spatio-temporal change of state analysis. It also considers as a case study image time series observations associated with the Argentiere glacier. The chapter presents a coarse- to fine-grained change of state dataset associated with the Argentiere glacier observed by several ground-based cameras during the period 2016-2020. It provides a library of deep learning networks and compares their main characteristics and argues that a significant part of the parameters issued from the original network should be forced to be non-learnable, a strategy called transfer learning. The experimental tests performed involve testing the sensitivity of the learners with respect to training hyperparameters that are the learn rates for weights and biases for the last fully-connected layer associated with the test networks.