Hello,
I see the problem. For the surface air temperature variable
data were available at all grid points. For SST there are some
grid points [ ~33% :-) ] that are all missing because they are
over land. The underlying fortran routine correctly set the internally
calculated lag-1 autocorrelation as missing (_FillValue). At this point
it should have returned with the equivalent sample size set to
_FillValue also. Unfortunaltely, it did not.
A fix will be applied in the near future.
Regards
Dennis Shea
>Hi. I had a question about equivalent sample size and ftest.
>I created variables sst1(lat,lon,time) and sst2(lat,lon,time),
>and applied a low-pass filter to them. Then I calculated their
>variances. For land pts. I use the FillValue of 1e+35.
>Now, I want to determine if their variances are significantly
>different using the ftest. I included my approach below.
>However, the equiv_sample_size command leads to this error:
>
> ==>SLATEC/NCL: DGAMMA:
> ANSWER LT HALF PRECISION BECAUSE X TOO NEAR NEGATIVE INTEGER: NERR=
[SNIP]
>
>Previously, I did the same thing for surface air temperature without any
errors.
>Any idea what is wrong here? Thanks.
>Mike
>
>**********************
>
> xEqv = equiv_
>sample_size (sst1, 0.1,0)
> yEqv = equiv_sample_size (sst2, 0.1,0)
> gwt = latGauWgt (128, "lat", "gaussian weights", "dimension_less ")
> xN = wgt_areaave (xEqv, gwt, 1., 0)
> yN = wgt_areaave (yEqv, gwt, 1., 0)
>
> alpha=new((/128,128/),float)
> alpha = 100.*(1. - ftest(var1,xN,var2,yN,0))
> alpha!0="lat"
> alpha!1="lon"
> alpha&lat=lat
> alpha&lon=lon
> alpha@_FillValue=1e+35
> alpha=mask(alpha,oro,0)
>
>
>_______________________________________________
>ncl-talk mailing list
>ncl-talk@ucar.edu
>http://mailman.ucar.edu/mailman/listinfo/ncl-talk
_______________________________________________
ncl-talk mailing list
ncl-talk@ucar.edu
http://mailman.ucar.edu/mailman/listinfo/ncl-talk
This archive was generated by hypermail 2b29 : Sun Aug 24 2003 - 20:46:34 MDT