Keywords
land use modelling, spatial metrics, stochastic model, remote sensing, calibration
Start Date
1-7-2012 12:00 AM
Abstract
A correct historic calibration of land-use models is important, because they are more and more used by decision makers. Existing calibration methods, however, do not sufficiently take into account uncertainties in input parameters. For that reason, uncertainties that propagate through simulations of future land use are mostly unknown. When uncertainties in model parameters can be estimated, Monte Carlo modelling can be used to approximate the uncertainty in simulated land-use patterns. This study shows that uncertainty information is not only indispensible for the interpretation of the output of land-use models, but that this can also be used to improve the calibration of landuse models by means of data assimilation. For this purpose the MOLAND model of Dublin has been integrated in a Python-based data-assimilation framework. Calibration of the land-use model is based on the comparison of spatial metrics derived from historic remote sensing images and land-use simulation results. Remote sensing derived probability density functions of class metrics are assimilated in the model at time steps for which they are available. The particle filter algorithm only continues successful realizations of the land-use model, thus reducing its uncertainty by removing unsuccesful particles from the ensemble. In this way, only parameters sets used in the simulation that match the patterns observed in the remote sensing imagery as quantified by the class metrics, are used in the simulation of future land use. It is expected that the automatic procedure results in an improved calibration of the land-use model. Furthermore, it provides data on the uncertainty of the results, which is important for drawing conclusions from simulations.
Uncertainty analysis and data-assimilation of remote sensing data for the calibration of cellular automata based land-use models
A correct historic calibration of land-use models is important, because they are more and more used by decision makers. Existing calibration methods, however, do not sufficiently take into account uncertainties in input parameters. For that reason, uncertainties that propagate through simulations of future land use are mostly unknown. When uncertainties in model parameters can be estimated, Monte Carlo modelling can be used to approximate the uncertainty in simulated land-use patterns. This study shows that uncertainty information is not only indispensible for the interpretation of the output of land-use models, but that this can also be used to improve the calibration of landuse models by means of data assimilation. For this purpose the MOLAND model of Dublin has been integrated in a Python-based data-assimilation framework. Calibration of the land-use model is based on the comparison of spatial metrics derived from historic remote sensing images and land-use simulation results. Remote sensing derived probability density functions of class metrics are assimilated in the model at time steps for which they are available. The particle filter algorithm only continues successful realizations of the land-use model, thus reducing its uncertainty by removing unsuccesful particles from the ensemble. In this way, only parameters sets used in the simulation that match the patterns observed in the remote sensing imagery as quantified by the class metrics, are used in the simulation of future land use. It is expected that the automatic procedure results in an improved calibration of the land-use model. Furthermore, it provides data on the uncertainty of the results, which is important for drawing conclusions from simulations.
