In climate models, the initial data set is very important. There are a few reasons of the growth such as the system you are working, precision is very low for floating point arithmetic. Take care when  values with which you are playing e.g. multiplying a very large nos. with a very small value, then the error growth becomes high. The earth dia is too high and snow depth at certain area is a very low some places. or considering different values of temperatures sometime. You can read my paper also which is explaining many other reasons like working with different  compilers also.

Thank Goel for your response.

In my view, there had been three kinds of error growth mechanism already known to the scientific world.

one is the error growth mechanism due to incomplete observation data. The error comes from initial data belong to this kind.

the second is the error growth mechanism due to incorrect scheme of the control equation groups expressing the physical processes in the atmosphere.

the third is the error growth mechanism due to the nonlinear item in the equation group ,which lead to "chao" phenomena in the numerical study, which is found by Lorenz.

What I mentioned here is the fouth:

This kind of error growth mechanism is caused by the strong pre-condition called uniform continuity for all the variables such as geopothential height fields, temperature fields, wind fields, .......etc.

I have found out this kind of error growth mechanism is popular in our climate models which modellers ignore. I have developed an index which can identify this kind of errors. I also got a funding from China National Natrual Science Fondation. But recently, some leading scientists didn't know this,and refuse to give positive comments on my new discovery when I send a new application for fundings in China. I asked this questions in order to find some people to collaborate with me in order to  continue my study by the funding from your countries.

I love all of your questions, and want to add a very important factor that I do not think has been included in many climate models so far, especially regarding predictions of monsoon flooding and droughts--atmospheric dust, especially the Pakistan-Arabia Dust Cloud.  Not including its effect can potentially cause a huge error in any climate model. 

I have found a new error growth machanism, and I have proposaled a method to improve it,but it is very strange that I can't apply for some reserch projects to make further study. Some mainstream scientist in China told me that" some people want to control this funding on doing meaningless ,although they have no progress with their ideas which is not in the direction of your field"