Calibration and validation of a post-forest land use dynamic model based on timed automaton using a Markov model

摘要

Determining average transition rules to which farmer cropping succession practices tend after forest clearing is essential to deciding on a strategy for forest conservation. The bordering area of the Malagasy East forest is characterized by a heterogeneous mosaic landscape. Among six land use states for each plot –?forest, fallow, crop, herb, plantation and paddy field?– we justify the choice of the first four for modelling the dynamics of ecological and cultural practices. The transitions between these states have been governed by a set of land use change rules observed in inquiries conducted with farmers. These transitions are represented by a land-use model based on a timed automaton which is parameterized by the durations of fallow, crop and herb states and the number of cropping cycles since first forest clearing. The objective is to calibrate and validate the timed automaton with the help of a Markov model according to χ 2 metric from two samples of plots initially covered with Forest and considered independent and not spatially localized. The timed automaton is calibrated by simulating on 34?years since forest clearing with the first sample and then validated on 22?years with the second sample to minimize the distance between two Markov matrices which summarize the simulated and observed dynamics. According to the test of χ 2, the fitted parameters are accepted and on average in accordance with farmers’ practices. Some scenarios of deforestation are built with the validated model.