Assessing yearly transition probability matrix for land use / land cover dynamics

Assessing yearly transition probability matrix for land use / land cover dynamics
Jean-François Mas 1, Ernesto Vega 2 1

1.Universidad Nacional Autónoma de México (UNAM), Centro de Investigaciones en Geografía Ambiental (jfmas@ciga.unam.mx)
2.Universidad Nacional Autónoma de México (UNAM), Centro de Investigaciones en Ecosistemas (evega@oikos.unam.mx)

Abstract: In order to generate land cover projections and model land use and cover changes (LUCC), probability-based transition matrices are obtained through the overlaying of two maps of two different dates. However, the observation interval may differ because maps are not available every year or at a constant time interval and a matricial algorithm is commonly used to adjust the matrix. However, although the obtained yearly matrix is mathematically correct, it does not necessary represent adequately the yearly transitions due to the spatial coincidence of various transitions over the entire period of time and because some transitions that are observed over a large period are not possible over a shorter (e.g. yearly) period. In this paper, a novel approach, based on a genetic algorithm (GA), is applied to adjust yearly transition matrices taking into account criteria to produce more realistic transitions. A LUCC model was used to produce land cover maps at different dates using transition rules defined by the user. Yearly matrices were then obtained from these maps by both matricial and genetic algorithms and were compared. For certain periods of time, the matricial algorithm was unable to find a yearly matrix or found a matrix with impossible transitions. The GA approach was able to find realistic matrices from all the periods.

Keywords: Markov matrix, modelling, land cover change.

AttachmentSize
MasAccuracy2012.pdf79.93 KB