Approaching Uncertainties in Land-Use Change Modeling in the Amazon Rainforest with Bayesian Belief Networks

In recent years, modeling techniques to study land-use change and to develop scenarios for highlighting possible future pathways of land-use change have been increasingly applied and studied. While modeling results always incorporate uncertainties in terms of input data, model parameters, model boundaries, and model structure, up to now, only few land-use change modeling approaches have explicitly focused on uncertainties using e.g. Monte-Carlo simulation or Fuzzy logic. Decision makers, however, are not only interested in understanding possible consequences, but also in evaluating the reliability of these. Since land-use change is spatially dependent on its neighborhood, we especially address uncertainties associated with different settings of a spatial land use in neighborhood variable. We apply our land-use change model to Southern Amazon, Brazil, where rapid deforestation of large areas occurred in the past with high impacts on soil erosions, droughts, floods, and migration. To explicitly account for uncertainty, we apply a Bayesian Belief Network (BBN) approach. Uncertainty is integrated in terms of conditional probabilities between spatial variables; i.e. one state of one variable is conditionally dependent on the states of other variables. Our results highlight the capability of the BBN approach to investigate observed spatial dynamics of land-use change. Additionally, we are able to evaluate the uncertainty associated with a specific input source. Data uncertainty in terms of the represented spatial neighborhood substantially influences the modeling results. A further possible enhancement is the integration of qualitative expert knowledge. The investigation of uncertainty propagation to land-use change scenarios is a future challenge.