|
Hi Yogesh, The general ideas are here: http://ferret.pmel.noaa.gov/Ferret/faq/correlations-and-variances which shows first a simple correlation between two time series, and then a correlation between two X-Y-T gridded variables. To compute a correlation between a single-point observed time series and the gridded data, we want to replicate the data at the point location across the XY grid, then compute the correlation. If your point data variable is called point_var and the variable on the XY grid is xy_var, then you can set it up this way: yes? let point_on_xy = point_var[x=@ave,y=@ave] + 0* xy_var The purpose of taking the average in X and Y is to remove the grid dependence in the XY direction. If the observation is on a 1-point grid in x and y we want to remove that. An averaging operation removes the axis over which you define the average (but in this case will do no computation). If the time axes of the observation and the gridded variable did not agree, you could define a regridding in time as well. Adding 0* xy_var makes use of the idea of "conformability" to promote the data from the observation to the grid points in XY. See what you have: yes? shade/L=1 point_on_xy ! will show a constant value yes? shade/y=0 point_on_xy ! constant in X, shows the T variation of the observation Now you can continue with the example in the FAQ and call the "variance.jnl" script. Define the variables in whichever order makes sense to you. yes? LET p = point_on_xy yes? LET q = xy_var yes? GO variance For your second question, you would use the ideas from the second example in the FAQ, but you will need to regrid the data of one model onto the grid of the other. Ansley On 3/16/2016 2:18 AM, 'Yogesh Tiwari'
via _OAR PMEL Ferret Users wrote:
|