No label defined (Q4359724)

Geological departments need to estimate mine surfaces according to the mineral samples. Because of earth movement, the mine surface usually splits into several segments. It is very important, for mine exploring and for taking precautionary actions from accidents, to know the split locations of the mine surface. From a statistical view point, this is an estimation problem of 2-dimensional (2-D) regression functions with discontinuities. We can find such problems in many other application fields such as meteorology, oceanography, image processing, etc. Here, we discuss estimation of bivariate jump regression functions. An a.s. consistent estimator of the jump location curve is suggested. This estimator is based on difference of two one-sided kernel smoothers. A rotation transformation is also used. We consider an ideal case that the jump location curve has an explicit function form first and then generalize it to a more general case where the explicit function form does not exist. Compared to some existing methods on this topic, mainly to the edge detection methods in image processing literature, our method uses less conditions on the design points and on the underlying regression function. So it is expected to find more applications.