Interpolation issue
Hi,
As I commented in issues #179 , we discovered an issue with @mmenegoz when we were calculating correlation based on a combination of two dataset ERA40 and ERAinterim.
Here is the concerned plot:
~cprodhom/postscript/decadal/correlation_skill_1y_tas.eps
After long investigation, we discovered that there is a big issue with the interpolation in Load. We called load two times with and without specifying the output grid. And the plot below show the difference for the three loaded observational products:
/home/Earth/cprodhom/postscript/decadal/test-reso/era40_obs_diff_nogrid_grid_2000grid.eps
/home/Earth/cprodhom/postscript/decadal/test-reso/erainterim_obs_diff_nogrid_grid_2000grid.eps
/home/Earth/cprodhom/postscript/decadal/test-reso/ncep-reanalysis_obs_diff_nogrid_grid_2000grid.eps
You will see that in all these plot appears very strange lines, which strong numerical errors in the interpolation. This has been hidden because people were always calculating anomalies and correlation with one single product. This appears very clearly as soon as you mix two products but is affecting (slightly?) all the results.
The script used to produce this plot is here:
/home/Earth/cprodhom/analyse/decadal_forecast/decadal_test_reso.R