I plan to use sequential Gaussian simulation (sGs) approaches in study. However, when I apply the inverse normal score transformation (NST) to the data after the calculation, there are some errors (especially in negative results). Therefore, I simulated the data with Box-Cox transformation instead of NST and obtained more appropriate results.
Can the Box-Cox transformation be used instead of normal score transformation in sequential Gaussian simulation (sGs)? Can I also find examples of such transformations previously used (to reference in the article)?
Thank you very much in advance
In my experience I would say that as long as the transformation is giving you the statistical distribution required by the sequential gaussian simulation, then it should work fine. Is sequential gaussian simulation sensitive to the statistical distribution of the data? Is the box-cox transform doing a good job of producing a normal distribution, if that is what is required? Is there a re-scaling that could be done to eliminate the negative values before the NST transform to eliminate the errors? That may be another alternative.
Dear Katona
First of all, thank you for your reply. After the Box-Cox transformation I tested with the normality tests (Shapiro-Wilk Test) and obtained the appropriate results for normal distribution. I cannot eliminate the negative results after NST because the number of data is quite high. As I understand it, do you suggest that Box Cox Transformation can be applied in sGs after fulfilling the normality requirement?
When I did training in statistical analysis we were taught to check a number of transforms and select the one that produced the "best" normal distribution. As long as you are able to back-transform it should not matter which transformation is used in this situation.
I'm surprised you are getting negative results on the back-transformation (if you don't have negative values in the input data). You may be having a issue with with you simulation process or algorithm if this is the case - perhaps check this first.
Dear Murphy
Thank you for your advice. But I did the necessary checks and did not encounter any algorithm problems. In particular, there are problems in maximum values and outliers in the data set. Even though I haven't found the source of this problem in NST, I haven't encountered these problems with Box-Cox.
Cafer - does the NST that you are using a quantile (data space) to quantile (gaussian space) transform - if coded correctly this should not produce negative values - perhaps have a look at the NSCORE program in GsLIB suite for programs http://www.statios.com/help/nscore.html. If you have a lot of same values you can add a little noise to these to break the ties.
Mr. Murphy
That could be a good idea. I'il take your advice. Thank you so much for your help.
- Badges
- Science topic
- Similar topics
- Geoscience
- Cartography
- Mapping