Time space stochastic modelling of agricultural landscapes for environmental issues

Since the initial point of Langran (1993) saying that Geographic Information Systems (GIS) were poorly equipped to handle temporal data, many researchers have sought to integrate the time dimension into GIS (Roddick et al., 2001). We present a time space modelling approach - and a generic software named ARPEnTAge - capable of clustering a territory based on its pluri-annual land-use organization. By adding the ability to represent, locate and visualize temporal changes in the territory, ARPEnTAge provides tools to build a Time-Dominant GIS. One main Markovian assumption is stated: the land-use succession in a given place depends only on the land-use successions in neighbouring plots. By means of stochastic models such as a Hierarchical Hidden Markov Model and a Markov random field, ARPEnTAge performs an unsupervised clustering of a territory in order to reveal patches characterized by time space regularities in the land-use successions. Two case studies are developed involving two territories carrying environmental issues. Those territories have various sizes and are parameterized using long term surveys and/or remote sensing data. In both cases, ARPEnTAge detects, locates and displays in a GIS the temporal changes. This gives valuable information on the spatial and time dynamics of the land-use organization of those territories.