As we all know that use of scorr function in grads calculates the spatial correlation between two variables over an X-Y domain. I am using this function in a different ways:
In place of two variable, I am using 1) model and 2) observation for any particular variable. Then it will compute the spatial correlation over an X-Y domain of that particular variable. My question is:
1) Whether my method is correct?
2) How to interpret this correlation? I mean what I can say from this correlation value?
Any kind of help will be useful.
http://www.utdallas.edu/~dag054000/Taiwan_lectures/background%20readings/IEHG-2007.pdf
Hi Firoz
Thank you very much for your answer. For your kind information, my question is not regarding spatial auto-correlation. Please try to read my query twice and answer.
Hi Suman,
I also use scorr function. The aim of this technique is simply to get the accuracy/similarity of spatial patterns between two variables. For instance, you have two precipitation data from observation and simulation, and do the spatial correlation between them. If the scorr value is high and satisfies a certain significance level (if needed), you can say the model accurately simulates the precipitation pattern. However, don't forget to calculate the bias or RMSE to assess the accuracy in precipitation's amount.
Dear Muhammad
Thank you very much for your answer. I exactly agree what you have told and your methodology. I am doing the same thing. Only query I want to ask you about significance level. How I will check that? As I know this is done using t distribution. Since I have dataset at large no. of grid points and therefore how to find the degrees of freedom (dof)? for example dimension of the dataset is 333x222. How to find the dof for significance level for correlation? Please reply at the earliest.
You are right, we have to use an accurate dof. t test assumes the samples are independent while the spatial grid points are likely correlated each others. You can obtain an effective dof by removing autoregressive properties of both series using an algorithm in Livezey and Chen (1983) (Eq. 1 and 2). Although they used time series data for correlation analysis and your case is spatial correlation, I think the equation is still applicable since the aim is to remove the dependency of samples.
http://journals.ametsoc.org/doi/abs/10.1175/1520-0493(1983)111%3C0046%3ASFSAID%3E2.0.CO%3B2
Dear Muammad
Thank you very much for your answer. Now I have got the point and I am trying to to through the paper. Let's see whether I can solve the problem or not. If needed, I will write to you. Anyway again heartily many many thanks to you.
Dear Muhammad
I need your help regarding eof analysis for meteorological datasets. I have explored a lot but not confident enough. Do you have any document with meteorological examples and how to interpret those results? If you don't mind will you provide your mail-id so that I can contact you through emails. Please reply.
These documents may be useful:
http://journals.ametsoc.org/doi/abs/10.1175/MWR3150.1
http://www.met.reading.ac.uk/~han/Monitor/eofprimer.pdf
https://websites.pmc.ucsc.edu/~dmk/notes/EOFs/EOFs.html
I am not that expert. However if you want, you can email to m.rais.a[at]dc.tohoku.ac.jp.
- Badges
- Science topic