The krypton voltage at 4 is low, less than 0.1. It is also quite humid: H2O > 20 g/m^3
kh2o goes as -log(V), so the variance goes up as the voltage reaches the noise floor.
The covariance with w (w'kh2o') still looks generally good.
http://www.eol.ucar.edu/isf/projects/CWEX11/isfs/qcdata/plots/20110630/krypton4_20110630.png
http://www.eol.ucar.edu/isf/projects/CWEX11/isfs/qcdata/plots/20110630/h2o_20110630.png
It's probably wise to also clean the krypton at 2.