Identifiability of total effects from abstractions of time series causal graphs