Abstract
It is so far from a simple problem to identify the true relation between the actual measurements on the ground or sea and the remote sensing MSS data that some statistical model like the canonical correlation model or the simple regression model must be assumed for the approximate description of the phenomena. Thus the correlation and regression analysis is very widely used for the remote sensing study like the water characteristics estimation, but what statistical models should be is hardly discussed, while so many misuses of the statistical methods are done with the inference based on the wrong model. The main purpose of this paper is to explain how to build a statistical regression model for the remote sensing study.
In this paper the view that the observation for ground (or sea) truth data is done to calibrate the measurement by the remote sensing, is presented, that is, the remote sensing data are considered as the substiture characteristic of the truth data. Then the following 2 points become very important in theuse of the regression model;
1) Not regression of ground truth on remote sensing data, but regression of remote sensing data on ground truth.
2) Meaninglessness of the correlation coefficient of the bivariate data obtained by some experiments.
Some example of the real data analysis is also given to clarify the idea.
