estCrossVal and estSurfaces failing at random points

I am encountering random points of failure with modelEstimation during the initial estCrossVal and estSurfaces processes. I am attempting to run short-term trends and am seeing either one process run to completion and the other fail, but during a re-run seeing the first process fail with no set pattern that I can define. For some of my problem sites I was able to run a long-term trend, along with bootbands, with no problem.

I've never encountered this problem before and am up to date with R and EGRET packages. The strange thing is that I was failing with many more sites until I upgraded R last week. This, again randomly, fixed some sites but not others. I'm wondering if there is a bug because I cannot find any rhyme or reason to where or why these sites are failing.

Also, despite estCrossVal and estSurfaces failing, the resampling process continues just fine and prints out estimations and likelihoods, however the AnnualResults table is empty and therefore leaves my work incomplete.

Has anyone come across this problem recently?
Perhaps an NWIS Modernization issue?

Start by reproducing modelEstimation with the affected short-term-trend sites, comparing estCrossVal and estSurfaces across reruns. Inspect how the resampling output reaches the AnnualResults table and record the R and EGRET versions plus site inputs. Done means the intermittent failure has a reproducible cause and the affected estimation produces a populated AnnualResults table.

Written by the indexing model from the issue text.