Importantly, the amount of explained variability can be greatly influenced by the choice of the spatial and temporal scales. Furthermore, monthly rainfall comprises influential variability from seasonal, inter-annual and inter-decadal time scales. Eventually, there is need to first deseasonalize the monthly series before doing the EOF analysis. 

You may find the following papers relevant for your interest.

Article Influence of Spatial and Temporal Scales on Statistical Anal...

Article Geospatial Trends and Decadal Anomalies in Extreme Rainfall ...

Hi Charles

Thank you very much for your answer. I am very new in this field. I agree with you. If I am interested to look only in inter annual variability then what to do? I mean how to remove seasonal and Inter decadal variability? I have used cdo for doing the eof analysis since it is easy to use. I have attached my datafile here containing monthly rainfall for 30 years during 1951-1980 (12*30=360 months). It is only over Indian landmass. I am interested to see dominant mode of rainfall in the past 30 years.

Will you be so kind to check and comment on that. I need some more help from you regarding the software you are using and the steps for  doing so.

It may happen that my queries are silly due to that I don't have experience. But I want to learn the topic.

I expect a positive reply from you.

You can still consider the use of monthly series. However, the steps to follow to remove seasonal component of monthly series before applying EOF are in one of the two papers for which I provided the links in my first comment. Some of the vital references for you to learn and apply EOF are also cited in the above papers.

EOF can be done in:

• PYTHON,

              https://github.com/ajdawson/eofs

• R

              https://www.rdocumentation.org/packages/spacetime/versions/1.1-5/topics/EOF

• MATLAB

https://nl.mathworks.com/matlabcentral/fileexchange/54416-empirical-orthogonal-function--eof--analysis

• etc etc

Good luck!

Try to take a loook on the temporal variability of your result or your 1st mode. Try correlate with dominant phenomena. When it highly correlate then that mode represent that phenomena. As simply as that

Dear Edvin Aldrian

Thank you very much for your suggestion. I agree your point. As I have seen in the case of SST over pacific ocean there is a well known strong variability due to ENSO, therefore people are doing correlation with 1st mode time series with ENSO index. But in my case the parameter is all India rainfall. Therefore I am not able to understand which dominant phenomena I need to correlate. It is a monthly data series for 30 years. Therefore which kind of variability (Intra-seasonal, Inter-sesonal, Inter-annual, Intra-annual etc) I need to look for? By the by I am getting confused with the terminology used in the literature. For example what is the meaning of seasonal variability? what will be its period? is it 1 cycle/year? if it is why not known as annual variability?

Because I have seen in a Matlab link (http://in.mathworks.com/matlabcentral/fileexchange/61345-eof) having example on SST data, they are declaring the 1 cycle/year as seasonal variability.

I am expecting a clarification from you.

Thanks in advance

Regards

Suman

Hello

If you are computing EOF on monthly data, it is not a big surprise that the first mode explains a lot of variance and is associated with a monopolar variation across your network. It simply reflects the fact that all Indian landmass (or almost except extreme SE and perhaps Kashmir) shares the same annual cycle and that the temporal variations associated with this clock overwhelms all other type of variations. If you want to concentrate on interannual variations, you need to remove the annual cycle. You can either remove the climatological monthly means and perform the EOF on the anomalies or sums the seasonal rainfall for JJAS for example to consider the main rainy season across India and then compute EOF on the seasonal amounts, which, by definition, filter our the mean annual cycle.

Vincent

Suman Maity Sir, how many modes have you got for the EOF because am doing the same analysis on the same timescale...?

According to literature, first dominant mode mostly directs towards SST and gph anomalies in the same exact time period of EOF analysis