|
Hi Nitin,
Not meaning to give statistics advice (I’m the last one qualified for that!), but I’m not sure that those ferret scripts can correctly give you the 95% confidence interval (or indicate significance at the 5% level). They don’t account for things
like serial autocorrelation or help you adjust for a global significance level. For the former, you’d need to adjust the trend error estimate using the lag-1 correlation coefficient (e.g., see Santer
et al., 2000). For the latter, you need p values for each grid square, and then you can use the Wilks method I’ve previously emailed to list. There might
be ways to do these calculations in Pyferret, but I’m still in the habit of using IDL to get p-values and then saving to netcdf for plotting in ferret.
Anyway, once you’ve got a mask of which grid squares are significant, indicating this on a plot is quite straightforward using a mask:
———
let critical_p_value = xxxx
!Some value defined as per my other message
let signif = if (pval gt `critical_p_value`) then 1
!Mask of 1s and 0s for significance
fill/pal=white_centered/lev=$(my_levs) slope
!Plot the whole field
shade/over/lev=(0)/pal=grey/nolab signif
!Grey out non-significant grid squares (could use white if you weren’t using a white centered palette)
contour/over/line/lev=$(my_levs) slope
!Add contour lines so all the trend values can be seen, but only the significant trends are highlighted in color (my preferred way at least)
go land
———
Sorry it’s not a full answer, but hopefully helpful.
P
|