Application of DSmT for Land Cover Change Prediction

This chapter presents an environmental application of DSmT for the land cover prediction. The spatial prediction of land cover at the field scale in winter is useful to reduce the bare soils in agricultural intensive regions. Fusion process with the Dempster-Shafer theory (DST) proved to have limitations with the increase of conflict between the sources of evidence that support land cover hypotheses. Several modifications may be used such as source weighting or the hedging methods, but with no benefit in the considered case studied since the conflict may not explain by itself all the bad decisions. Actually, sources of evidence may induce all together a wrong decision. Then, it is necessary to introduce paradoxical information. Nevertheless, sources of evidence that are in use, are defined according to hypothesis “covered soil” or “bare soil” in the frame of DST. We investigate several points of view to define the belief assignments of the hyper-power set of the DSmT from the initial power set of DST. So, smart belief assignments induce a better prediction of bare soils. Samuel Corgne is also affiliated with TAMCIC, CNRS FRE 2658, team TIME, GET/ENST Bretagne, France. 372 CHAPTER 17. APPLICATION OF DSMT FOR LAND COVER CHANGE PREDICTION 17.