<html><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; ">Tim,<div><br><div><div>So you give us three options to choose from for the location of obs error variance in the obs sequence file.<div>1. The original way to put it at the end of all available copies.</div><div>2. New method to put it right after prior/posterior mean/std but before any ensemble prior/posterior mean/std if any.</div><div>3. New new method to move it all the way front right after the observation.</div><div><br></div><div>I guess this issue came up while you were working on the obs sequence files for my ensemble forecasting data. (Thanks!)</div><div>Given that we will potentially have more interest in expanding our ensemble data assimilation to ensemble forecasting and in obs-space verification for the real-obs cases, the current way (i.e., option 1) doesn't look ideal anyway.</div><div>If it is time we can make a change, it looks to me more natural to put the observation error right after the observation regardless of our filter design. And I think it's a good thing to update our hard-wired scripts in a more generalized way. </div><div>Although option 2 may be a safe choice at present as we don't have many users who worked with ensemble observations yet, I would be happy to modify my scripts (if any) for option 3. </div><div><br></div><div>Soyoung</div><div><br></div><div><div><br><div><div>On Nov 5, 2010, at 4:30 PM, Tim Hoar wrote:</div><br class="Apple-interchange-newline"><blockquote type="cite"><div style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; "><div>I've gotten a request asking to change the storage order of the 'copies' in</div><div>the netcdf format version of the DART obs_sequence files. The copy order is</div><div>described with strings in the CopyMetaData variable. My strategy has always been</div><div>to never assume anything about the storage order - this won't affect me or the matlab</div><div>scripts (see the get_copy_index(), get_qc_index() matlab functions).</div><div><br></div><div>For people who like to hardwire magic numbers into scripts (and who doesn't!)</div><div>this makes a potentially profound change. The NCO operators are also geared</div><div>to using magic numbers (as far as I can tell), so the time to speak up is NOW.</div><div><br></div><div>Presently, the obs error variance is always 'last'. If </div><div>filter_nml:num_output_obs_members > 0, these ensemble members get preserved</div><div>right after the 'mandatory' copies:</div><div><br></div><div><div>NCEP BUFR observation (for example) </div><div>prior ensemble mean </div><div>posterior ensemble mean </div><div>prior ensemble spread </div><div>posterior ensemble spread </div><div>prior ensemble member 1 </div><div>posterior ensemble member 1 </div><div>prior ensemble member 2 </div><div>posterior ensemble member 2 </div><div>prior ensemble member 3 </div><div>posterior ensemble member 3 </div><div>...</div></div><div><div>observation error variance</div></div><div><then the QC values follow></div><div><br></div><div>The proposed change is to make it such that the copy number of the observation_error_variance is invariant to the filter_nml:num_output_obs_members setting:</div><div><br></div><div><div><div>NCEP BUFR observation (for example) </div><div>prior ensemble mean </div><div>posterior ensemble mean </div><div>prior ensemble spread </div><div>posterior ensemble spread </div><div>observation error variance </div><div>prior ensemble member 1 </div><div>posterior ensemble member 1 </div><div>prior ensemble member 2 </div><div>posterior ensemble member 2 </div><div>prior ensemble member 3 </div><div>posterior ensemble member 3 </div><div>...</div></div><div><div><then the QC values follow></div></div></div><div><br></div><div>If you're still reading, keep the following in mind: The observation error</div><div>variance is probably ALWAYS present, even if you convert obs_seq.out files to</div><div>netCDF. In that case (I believe, but am unwilling to check at Friday at 4:30)</div><div>that the storage order is:</div><div><br></div><div><div>NCEP BUFR observation (for example) </div><div>observation error variance </div></div><div><then the QC values follow></div><div><br></div><div>So in some ways, I'd like to move the observation error variance to copy 2</div><div>for complete consistency. If the prior/posterior means/spread exist, they</div><div>get appended after that. This would cause the world as we know it to</div><div>cease to exist. I know of lots of "other people's scripts" that REQUIRE the prior_ensemble mean copy 2.</div><div><br></div><div>As you may have surmised, I have no opinion on the matter ... I'm happy</div><div>to follow the group decision. This is a democracy. I'd like to be a benevolent</div><div>dictator, but those pesky minions keep proving me wrong.</div><div><br><div> <div style="font-family: Helvetica; ">Tim Hoar, Associate Scientist</div><div style="font-family: Helvetica; ">National Center for Atmospheric Research</div><div style="font-family: Helvetica; "><a href="mailto:thoar@ucar.edu">thoar@ucar.edu</a></div><div style="font-family: Helvetica; ">303 497 1708</div> </div> <br></div></div>_______________________________________________<br>Dart-users mailing list<br><a href="mailto:Dart-users@mailman.ucar.edu">Dart-users@mailman.ucar.edu</a><br>http://mailman.ucar.edu/mailman/listinfo/dart-users<br></blockquote></div><br></div></div></div></div></div></body></html>