Skip to Content

The ASIMUD project

Introduction

To be usefully applicable land-use change models need extensive calibration. Current calibration methods, however, do not take into account uncertainties in reference land-use data and uncertainties in the parametrisation of land-use change models. As a result, uncertainty in land-use change predictions are mostly unknown.

Probabilistic framework

In the ASIMUD project a particle filter data-assimilation algorithm will be used in a probabilistic framework in order to quantify and reduce uncertainties in land-use simulations. Its advantage is that no assumptions are made on the probability distribution of the model states. The particle filter calculates state predictions and their confidence intervals. This requires that the uncertainties of the model input variables and parameters are known. To our knowledge, the particle filter has not been used in land-use change modelling before.

Calibration framework using remote sensing

The probabilistic framework will be applied to a calibration procedure developed in the Belspo STEREO II MAMUD project. This procedure uses remote sensing data in the historic calibration process instead of land-use map time series, which are often lacking or show poor consistency. Spatial metrics derived from the land-use simulation and remote sensing observations are compared and used to tune model parameters.

Uncertainty in remote sensing data analysis

Remote sensing data analysis involves uncertainty caused by limitations of the data and the image interpretation methods used. Since uncertainties propagate through the processing chain, they will affect land-use maps inferred from remote sensing images and the derived land-use patterns, quantified by means of spatial metrics. An important part of the project is to characterize error and uncertainty in the different steps of the land-use interpretation process, using ground-truth data and process-related uncertainty models based on classification approaches.

Uncertainty in land-use change modelling

Spatially-dynamic modeling of land-use change involves uncertainty caused by attribute errors, positional errors, logical inconsistencies, incompleteness and temporal errors in the model and in the reference land-use maps used for initiation and calibration. Uncertainties in the reference land-use maps can be important, but are difficult to quantify objectively. Therefore, the only uncertainties that will be considered here are uncertainties in input parameters.

Uncertainty propagation in the modelling and remote sensing chain