Discussion Urban terrain segmentation for the Marmara Region

Discussion: Urban terrain segmentation for the Marmara Region Speaker: Akarun Discussant: Lerner-Lam

Classic Geophysical Inverse Problem Forward Problem: Image = F { Classifications [segments] } Segments = geographic basis functions, change with time. Useful for hypothesis testing and examining prior assumptions about geographic variation.

Inverse Problem l Classifications [segments] = F-1 {Image} l l l F-1 = F† Inverse = generalized inverse (approximate) l Treated by Stochastic Techniques and Genetic or Evolutionary Algorithms, AI (training), matched filtering, wavelets, empirical orthogonal eigenfunctions analysis, etc.

Basic Issues l l l Uniqueness of derived classifications. Do different combinations of classifications fit the data equally well? Error model, including covariance. Uncertainties as input to further calculations. Resolving power of data. – Geographical resolution (function of instrument) – Ability to distinguish between classifications l Optimizing a penalty function, including prior knowledge

Other issues l Proxy set – What needs to be quantified? l Relate to multidisciplinary needs – What is measurable/detectable? l l Collateral data sets Groundtruth: minimum (necessary and sufficient) tests Baselining: time dimension Background and causal changes – Global experience database – EOF ( t): time variation of geographic basis functions