In addition to PRIMA’s core models (i.e. agent-based and microsimulation models) spatial regression techniques qualify as relevant complementary tools for defining realistic ranges of future landcover transitions on PRIMA’s municipalities. This interest is also based on other shortcomings: CORINE Land Cover does not capture well local incremental changes, municipality clusters are not necessarily homogeneous in terms of landcover change patterns. While refinement is possible for specific transitions (forest), it remains difficult to ascertain generalization effects, commission errors and structural/contextual changes Considering transitions over 1990-2000 and 2000-2006 as learning data that could be used for inferring future trends, spatial autoregressive models have been introduced in order to answer the following questions: •Does any spatial smoothing compensate for the grainy nature of CLC data? •To what extent are past trends related to more recent changes? •Are there any categories of landcover transitions associated with specific spatiotemporal patterns? Three kinds of models have been developed: •Spatial lag model :y = ρWy + Xβ + ε •Spatial error model:y = Xβ + ε, where: ε = λWε + u •Spatial Durbin model:y = ρWy + Xβ + σWx + ε where X and y are respectively associated to 1990-2000 and 2000-2006 broad categories of landcover transitions. W are spatial weight matrices that describe the neighbouring relationships of municipalities - and can be considered as a ‘degraded’ form of spatiality. The procedure was then the following: 1.Run OLS regressions, run tests for spatial dependence 2.Run autoregressive models 3.Compare outputs with ANOVA, AIC. It appeared that for some landcover transition associated with agriculture or forest, the added value of autoregressive models -especially the spatial Durbin (mixed)- was clear. For others, such as internal urban transitions or changes related to waterbodies, basic OLS models were not outperformed. In other words, regional trends make sense for some landcover change processes. But for others, such a quest is much more delusional, requiring a longer-term and/or landscape perspectives, as well as more complex models.