Comparing Probabilistic and Geometric Models On Lidar Data

A bottleneck in the use of Geographic Information Systems (GIS) is the cost of data acquisition. In our case, we are interested in producing GIS layers containing useful information for river flood impact assessment. Geometric models can be used to describe regions of the data which correspond to man-made constructions. Probabilistic models can be used to describe vegetation and other features. Our purpose is to compare geometric and probabilistic models on small regions of interest in lidar data, in order to choose which type of models renders a better description in each region. To do so, we use the Minimum Description Length principle of statistical inference, which states that best descriptions are those which better compress the data. By comparing computer programs that generate the data under different assumptions, we can decide which type of models conveys more useful information about each region of interest.