<div dir="ltr">I didn't chose the most instructive colorbar to show the magnitude of the errors. Attached are plots with colorbars to indicate the magnitude of the errors. In particular, note that the errors due to single precision in the mass budget are on the order of 10% (saturated on the 1% colorbar). Note these are the errors for a particular time step based on monthly output. <div><br></div><div>The issue with budgets with single precision output is that the tendancy is always a small residual of larger terms. While of course the uncertainty of ECCO estimates is larger than even these 10% errors, I think this is not particularly relevant to this discussion. ECCO in the end is also a forward model and MITgcm developers have take great pains to insure that budgets close. I don't know why we'd sacrifice the ability to close budgets rigorously in order to save a factor of 2 in space by storing as single precision. Also, for the user it is a lot easier if the budgets close to machine precision rather than having a user wonder if an error is "acceptable" or not. </div><div><br></div><div>Martha</div><div><div><br></div><div><br></div></div></div><div class="gmail_extra"><br><div class="gmail_quote">On Sun, Nov 5, 2017 at 2:00 PM, Menemenlis, Dimitris (329C) <span dir="ltr"><<a href="mailto:Dimitris.Menemenlis@jpl.nasa.gov" target="_blank">Dimitris.Menemenlis@jpl.nasa.gov</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div style="word-wrap:break-word">
Hi Martha, single precision is 32-bit base-2, or approximately ~10^6 decimal.
<div>So no surprise that you get noise at the 10^-7 level.</div>
<div>What application requires that budgets be closed at the 10^-7 level?</div>
<div>From an engineering perspective, the ECCO budgets are so uncertain</div>
<div>that anything past the second decimal point is almost</div>
<div>certainly without much meaning.</div><span class="HOEnZb"><font color="#888888">
<div><br>
</div>
<div>Dimitris</div></font></span><span class="">
<div><br>
</div>
<div>
<div>
<blockquote type="cite">
<div>On Nov 4, 2017, at 6:39 PM, Martha Buckley <<a href="mailto:marthabuckley@gmail.com" target="_blank">marthabuckley@gmail.com</a>> wrote:</div>
<br class="m_-505667917707168754Apple-interchange-newline">
<div><span style="font-family:Helvetica;font-size:12px;font-style:normal;font-variant-caps:normal;font-weight:normal;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;float:none;display:inline!important">Hi
Ou,</span>
<div style="font-family:Helvetica;font-size:12px;font-style:normal;font-variant-caps:normal;font-weight:normal;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px">
It may be possible to close the budgets for the global ocean with single precision output, but this is not the case for regional budgets. For these we really do need to have double precision outputs. Attached are a couple of plots that show this. </div>
<div style="font-family:Helvetica;font-size:12px;font-style:normal;font-variant-caps:normal;font-weight:normal;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px">
<br>
</div>
<div style="font-family:Helvetica;font-size:12px;font-style:normal;font-variant-caps:normal;font-weight:normal;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px">
I am looking at mass and heat budgets over the top 1000m. What I did was take the r3/iter3 output, which has double precision, and after loading each file (e.g., UVELMASS, etc.), I converted to single precision (the back to double so the gcmfaces programs work).
When I use the double precision output, (tend-ocean convergence - forcing)./ tend < 1e-7 for both mass and heat (see budgetM1000m_errorDouble.<wbr>eps and budgetHC1000m_errorDouble.<wbr>eps). However, when the outputs are converted to single precision the errors are
much, much larger, (see budgetM1000m_errorSingle.<wbr>eps and budgetHC1000m_errorSingle.eps)<wbr>. </div>
<div style="font-family:Helvetica;font-size:12px;font-style:normal;font-variant-caps:normal;font-weight:normal;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px">
<br>
</div>
<div style="font-family:Helvetica;font-size:12px;font-style:normal;font-variant-caps:normal;font-weight:normal;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px">
One of the touted advantages of ECCO is the ability to do budgets, so we need outputs that allow us to do budget over any arbitrary regions, not just simple ones, like the global average. So can we get the new outputs in double precision please?! </div>
<div style="font-family:Helvetica;font-size:12px;font-style:normal;font-variant-caps:normal;font-weight:normal;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px">
thanks,</div>
<div style="font-family:Helvetica;font-size:12px;font-style:normal;font-variant-caps:normal;font-weight:normal;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px">
Martha</div>
</div>
</blockquote>
<br>
</div>
<br>
</div>
</span></div>
<br>______________________________<wbr>_________________<br>
ecco-support mailing list<br>
<a href="mailto:ecco-support@mit.edu">ecco-support@mit.edu</a><br>
<a href="http://mailman.mit.edu/mailman/listinfo/ecco-support" rel="noreferrer" target="_blank">http://mailman.mit.edu/<wbr>mailman/listinfo/ecco-support</a><br>
<br></blockquote></div><br><br clear="all"><div><br></div>-- <br><div class="gmail_signature" data-smartmail="gmail_signature">Martha W. Buckley<br><a href="mailto:marthabuckley@gmail.com" target="_blank">marthabuckley@gmail.com</a><br><a href="http://sites.google.com/site/marthabuckley/home" target="_blank">http://sites.google.com/site/marthabuckley/home</a><br></div>
</div>