From: "Lemoine, Frank G. (GSFC-6980)" <frank.g.lemoine@nasa.gov>
To: "ids.analysis.forum@ids-doris.org" <ids.analysis.forum@ids-doris.org>
Date: Mon, 11 Oct 2010 11:07:55 -0500
Thread-Topic: TR : Jason-1 EoL: the price of change. (Discussions continued)
Thread-Index: AQHLaV7CpyQrHvwm1UOK5zOCb9gOPA==
Message-ID: <B2C0817592DAF84B8C114091347F7DE91ABF9B3E1A@NDMSSCC02.ndc.nasa.gov>
Accept-Language: fr-FR, en-US
Content-Language: fr-FR
acceptlanguage: fr-FR, en-US
Content-Type: text/plain; charset="Windows-1252"
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Subject: [ids.analysis.forum] TR : Jason-1 EoL: the price of change.
 (Discussions continued)

 Dear IDS Analysis Forum-

I forward to you additional information about the Jason-1 End of Life Discussions
from Gerald Dibarboure of CLS. Please see the extended email message below.

Best regards,
Frank Lemoine
NASA Goddard Space Flight Center

********************************************************
________________________________________
De : Gerald Dibarboure [gerald.dibarboure@cls.fr]
Date d'envoi : lundi 11 octobre 2010 10:18
À : ostst@list.jpl.nasa.gov
Objet : Jason-1 EoL: the price of change

Dear OSTST members,
The Jason-1 Extension of Life (EoL) discussion is taking place in a different mailing list. But since the OSTST list is in cc of the material sent by NOAA/LSA, I guess I must forward the email I sent to the EoL mailing list one week ago.
It would be an understatement to say that the approach used, the simulations & tests performed, and the conclusions are not in agreement with the material from Remko and Walter. But again, this mailing list is not the place to discuss this, nor to talk about MSS accuracy or outliers, so I won't comment any further. If you want the short version, feel free to read the email below, if you want the long story, you can check out the posters.
This material expands on discussions from the EoL mailing-list and on preliminary assessments made last spring before the REVEX meeting and the JSG meeting. To that extent, I tried to make it as self-sufficient as possible but feel free to point out missing bits of information. I will then try to complement it with former emails and documents.
I’m looking forward to seeing you in Lisbon.
Best regards,
Gerald

=========================================
Message sent to the EoL mailing list on 05/10/2010
=========================================


Dear EoL fellows,
As per Josh and Rosemary’s request, I have uploaded a draft of two upcoming OSTST posters to keep the EoL discussion going.
 ftp://ftp.cls.fr/pub/oceano/jt/PosterOSTST2010_DibarboureMssError.pptx
 ftp://ftp.cls.fr/pub/oceano/jt/PosterOSTST2010_DibarboureEolSampling.pptx
Since the July JSG discussions, we have run some tests and simulations to quantify various statements from last spring. The results support the current recommendation to stay on this orbit as long as possible. If Jason-1 must move, then it is either onto an 11-day repetitive orbit (if EoL is more than one year) or onto the 12+341/419 geodetic orbit (if EoL is less than one year). Both are ~ 50km below the T/P orbit.
The EoL phase essentially asks two questions:
1.       Does leaving the historical track add a significant error?
2.       Does the loss of satellite coordination induce a significant sampling degradation?
To make it short, the answer we found is yes and yes.  Leaving the current orbit comes with a non-trivial price. More details below.
All the best,
 Gerald Dibarboure


Additional SSHA error from the use of a gridded MSS
Regarding the first question, we have tried to estimate the additional error stemming the use of gridded MSS (as opposed to repeat track analysis).
SSH anomalies (SSHA or SLA) are obtained through the difference of SSH(x,y,t) and <SSH>(x,y) the temporal reference. Therefore any error on <SSH> translates into an error on SSHA. Furthermore, any discrepancy between SSH and <SSH> (for instance standard used, corrections, different cross-calibrations…) adds an error on the Sea Level Anomaly as well.
The first advantage of the Mean Profile (a.k.a along-track <SSH>) is that it has the ability to cancel out the static part of any geographically correlated errors on a given sensor. Conversely, the best gridded MSS in the world still has to build a coherent 2D field from heterogeneous altimetry errors. As optimal as it might be, the gridded <SSH> is not coherent with SSH, which induces more SLA errors. For the sake of illustration, a two year average of Jason/ENVISAT differences exhibits 2 to 3 cm of errors with 300km to 2000km wavelengths (reduced to 1 – 2cm after DUACS cross-calibration).
Repeat track analysis is an old process, which has been improved over the years. For instance mean profiles are not just time averages, various pre-processing are used to ensure that they are not corrupted by instantaneous ocean variability, that they are not affected by interannual discrepancies (e.g. T/P 1992-2005 vs GFO 2000-2008) and so forth. Knowing that gridded MSS are built from the same data plus SSH data from geodetic phases/missions, the minimum MSS error is the Mean Profile error: it is impossible to have a gridded MSS more coherent with instantaneous SSH than a profile specifically computed to ensure this coherency. But why would a gridded MSS be less accurate than repeat track profiles? There are essentially 4 types of errors, none of which is easy to pinpoint and to separate from the others.
1.       To ensure a global coherency between all datasets, the gridding process is basically averaging all sensor-specific errors and especially geographically correlated ones. Consequently, not only is the gridded MSS unable to cancel the static error from the mission processed (see above), but it has ingested and “averaged” the static errors from other missions. The gridding process is generally tuned to try and minimize this “mix of static mission-specific errors”.

2.       The gridding process is not able to resolve all scales in an isotropic way so it has to perform some smoothing to make up for signals which cannot be resolved away from known tracks (and to some extent from geodetic data). Comparing the CNES/CLS and DNSC/DTU MSS highlights that the smoothing is not the same. Furthermore, if one computes SSH anomalies at high resolution (from 0.5Hz to 20Hz), one can see quite a lot of content which is not resolved similarly by MPs and MSS.

3.       The omission error (i.e the error due to the lack of actual measurements) is often assumed to be the dominant error away from known tracks, as only geodetic data are available in crossover diamonds of repetitive tracks.

4.       Which leads to the difficult use of geodetic datasets in MSS gridding. Contrary to repeat Mean Profiles, geodetic data contain the full extent of oceanic variability, and not just the mean sea surface height content one would like to absorb in a gridded MSS. Although it is possible to minimize the MSS corruption by oceanic structures from geodetic data, residual errors are far from negligible.

To assess these errors we have used various tests: comparing gridded MSS from different groups, or comparing different MSS computed similarly but from different data  (e.g. with and w/o certain GFO or T/P tracks), looking at what happens on SSHA variance during the T/P drift phase and on the tandem track, looking at residuals between a MSS and the input data it ingested, looking at differences between MSS and EGM08 and so forth. The difficulty in this exercise is that it is often necessary to pass-band filter data so highlight differences out of a large signal, and that most results observe the sum of multiple error terms so one cannot just do a simple sum/RSS. I won’t go into details about how we obtained each result but feel free to check out the poster, and feel free to ask for more if comments and legends are not clear enough.
To make it short, the most optimistic error estimate is to assume we only have omission errors and to gloss over the rest. In this scenario we get an error range of 1cm (100 to 500km) + 2.5cm (shorter scales). This estimate is obtained only using a fully coherent CNES/CLS dataset and processing. A more realistic estimate is to look at the differences between CNES/CLS and DNSC/DTU (after interannual discrepancies are removed) which exhibit coherent discrepancies up to 5cm.
An average error of 3 cm is arguably a safe estimate, neither too optimistic and occulting non-trivial error terms, nor assuming that the full content of MSS differences is necessarily impacting Jason EoL SSHA products. The 3cm estimate is also quite consistent with the formal error delivered with the CNES/CLS MSS (the formal error is a theoretical information computed by the OI from input data distribution and error prescribed). To this static difference, one should add the inability of the gridded MSS to cancel out static errors on SSH from each sensor as mentioned above (add a best case 1 to 2 cm error after cross-calibration).
Note that these figures are by no means an estimate of the absolute precision of each surface: we expect the CNES/CLS MSS to exhibit more coherency on various metrics simply because altimetry data and pre-processing are essentially the same, whereas MSS from DNSC/DTU are derived from different data, standards and pre-processing (likely as good but different). Also note that since the time I have assembled this material, we have run tests on the most recent DNSC/DTU 2010 MSS which exhibits improvements upon the DNSC 2008, but no ground shaking change for these EoL estimates.

To anyone wondering the impact of such an additional error, the poster contains a map and a histogram of the ocean variability presently observed from altimetry. As an illustration, assuming an average error level of 3cm, at least 50% of the ocean would be affected by a systematic error higher than 50% of the local signal variability, and only 20% of the oceans would have an error inferior to 25% of the local variability. Modern MSS are unarguably much better than pre-Jason2 surfaces, but they are still not on par with modern repetitive altimetry processing.

Sampling changes if we leave the tandem orbit
We have also run various sampling analyses to look at the sampling dynamics of EoL orbit candidates, and to quantify the differences between them.
Firstly, pure geometrical simulations and animations highlighted losses to the coherent sampling pattern of the tandem (e.g loss in ability to detect and to track sea state or mesoscale features in Near Real Time). One of the most spectacular changes is due to the combination of the Moiré effect (if Jason1 EoL and Jason2 overlap, it’s on thousands of kilometers) and the desynch process (no more 10-day cycle or 3-day sub-cycle). It is possible to observe a cyclic “pulse” in sampling quality in addition to systematic blind spots: sometimes both satellites are relatively well interleaved (at least as good as possible) and a few days later they are overlapping everywhere in the entire Pacific. The pulse cycle and impact is dictated by the first sub-cycles for sea state scales and larger sub-cycles or cycles for mesoscale. The consequence would be an unstable sampling capability in Near Real Time (good days, bad days, good weeks, 100% duplicate weeks…).
To assess the impact of this dynamics change, we have run OSSE-like simulations with DUACS software. In a nutshell, we took eddy resolving model outputs as an ocean “truth”. From this truth, we simulated along-track data from the tandem and from 11 EoL orbit options (7 repetitive with various cycle durations, and 4 geodetic, i.e. the best from each altitude family). From simulated satellite observations, we have run DUACS processes to reconstruct an observed state of the ocean in Near Real Time. Comparing the reconstructed fields to the ocean “truth” we get an estimate of the observing capability of the constellation. All results are then normalized to the error of the current tandem (so 100% means a Jason2 / EoL duo would be as good as what we have now).
To separate the impact of sampling changes from the impact of the MSS error, EoL data were first simulated with no additional MSS error (just standard repetitive altimetry errors), then with an optimistic +1cm EoL error, then with a pessimistic +3cm EoL error.
Even ignoring the MSS error, we observe that breaking the tandem adds 20 to 35% of additional reconstruction error due to measurement duplicates (Moiré+desynch). The best repetitive orbit is a 15-day orbit. The best geodetic is confirmed to be 12+341/419 (see previous emails and qualitative assessments) which has a pseudo-cycle of 17-days. The worst Moiré effect is on 11-day orbits (again, as expected from longitude/time plots sent in former emails).
To run a fair comparison, we have looked at the impact of the MSS error (both optimistic and pessimistic) on the current tandem track and on the 12+341/419 geodetic EoL (with the same results). If we add an optimistic 1cm error, the impact is unsurprisingly a measly +5% error in zones of intense mesoscale activity (15cm rms or more), and +12% wherever ocean variability is lower than 12cm (roughly 75% of the globe). If the MSS error is 3cm, the reconstruction error increases to +13% in zones of intense mesoscale variability and more than +50% wherever ocean variability is lower than 12cm! Moreover, the MSS-induced error stacks with the degradation observed from sampling alone.
If Jason-1 has to be moved to an EoL orbit, geodetic orbits are strictly better in terms of pure sampling capability (good blending with Jason2 observations). But even for strong currents, this gain is lost the MSS error if the error is larger than 1 or 2cm. To observe strong oceanic variability (15cm or more), both types of orbits are roughly equivalent. However, for the rest of the ocean, a repetitive orbit is strictly better, assuming that we can rebuild a decent Mean Profile after one year (i.e. cycles of 11 days only). Conversely, if the EoL phase lasts for less than one year, or if its duration is unknown, geodetic orbit 12+341/419 should be selected as the MSS error cannot not be removed anyway.
But most importantly all EoL options are largely inferior to the current tandem. The price of an EoL phase would be up to +30% observation error (with multi-satellite maps) from sampling alone, and up to +50% if the MSS error is a realistic 3cm. Furthermore, this sampling degradation is uneven in time and it goes through cyclic pulses of best/worst case phases particularly detrimental to near real time applications.


--
----------------------------------
Gerald Dibarboure
C.L.S, Space Oceanography Division
8-10 rue Hermès
31520 Ramonville St-Agne - France
Tel: +33 (0)5 61 39 47 83
----------------------------------
WARNING: please update email to
    gerald.dibarboure@cls.fr<mailto:gerald.dibarboure@cls.fr>
----------------------------------


                           Cliquez sur l'url suivante 
https://www.mailcontrol.com/sr/8jvPdODvdqHTndxI!oX7Uox5a7D76txBTqU9np8AwBcewYt70PTNHrlmcw5pzXf0xIHNdJ15nZDfDqWPgPvXQw==  
                    si ce message est indésirable (pourriel).