NESDIS AIRS Data Missing - Does it impact the models?

[weatherdude]

NESDIS reported a hardware failure last Wednesday that took out MODIS and AIRS data. MODIS data came back up on Saturday the 17th, and AIRS data is still completely out and not getting into the models. 

 

Does anyone know how important this data is to the global models? 

 

I am aware that most items have "redundancies", but during the late October outage, those redundencies did not stop some strong lowering of model performance in the second week of the model forecast with the data outage at that time.

 

Thanks!

[dtk]

NESDIS reported a hardware failure last Wednesday that took out MODIS and AIRS data. MODIS data came back up on Saturday the 17th, and AIRS data is still completely out and not getting into the models. 

 

Does anyone know how important this data is to the global models? 

 

I am aware that most items have "redundancies", but during the late October outage, those redundencies did not stop some strong lowering of model performance in the second week of the model forecast with the data outage at that time.

 

Thanks!

Yes, it impacts the global models, but the loss likely isn't critical considering that we now have other hyperspectral IR sounders to assimilation (IASI on the European satellites and CrIS on NPP).  It's high up on the impact list, see here:

 

However, if AIRS is missing, some of the other bars will increase (in terms of impact/error reduction) to fill the void.

[ohleary]

Yes, it impacts the global models, but the loss likely isn't critical considering that we now have other hyperspectral IR sounders to assimilation (IASI on the European satellites and CrIS on NPP).  It's high up on the impact list, see here:

 

However, if AIRS is missing, some of the other bars will increase (in terms of impact/error reduction) to fill the void.

 

Where did you get this?  We for years have been trying to quantify the impact of a loss of a certain (usually satellite) data type in the models and EMC doesn't usually give us clear answers. 

[weatherdude]

http://gmao.gsfc.nasa.gov/forecasts/systems/fp/obs_impact/plots/summary_all+month+global+impact_per_anl.png

[dtk]

Where did you get this?  We for years have been trying to quantify the impact of a loss of a certain (usually satellite) data type in the models and EMC doesn't usually give us clear answers. 

To be clear, this plot doesn't tell you what happens if a satellite type is lost....it only tells you the estimated impact, for this metric, for this period, in the context of the observing system utilized.  If you redo the same period with a different observing system (i.e. no AIRS), the relative impacts will change since the level of observing system redundancy changes.  EMC has an ensemble based estimate for this type of thing but it is only a prototype and is not ready for real-time monitoring like the systems at ECMWF, UKMO, NASA, etc.

[ohleary]

To be clear, this plot doesn't tell you what happens if a satellite type is lost....it only tells you the estimated impact, for this metric, for this period, in the context of the observing system utilized.  If you redo the same period with a different observing system (i.e. no AIRS), the relative impacts will change since the level of observing system redundancy changes.  EMC has an ensemble based estimate for this type of thing but it is only a prototype and is not ready for real-time monitoring like the systems at ECMWF, UKMO, NASA, etc.

 

Yeah, this isn't even specifically tied to the GFS from what I see, so it's not really applicable.  Funny, this exact question was raised at the ops meeting this morning...what is the impact to the GFS from the lack of AIRS this last week?  It comes up all the time during a prolonged satellite data outage, and EMC can't usually quantify it...seems quite hard to quantify off the cuff.  Looking forward to EMC's prototype.

[dtk]

Yeah, this isn't even specifically tied to the GFS from what I see, so it's not really applicable.  Funny, this exact question was raised at the ops meeting this morning...what is the impact to the GFS from the lack of AIRS this last week?  It comes up all the time during a prolonged satellite data outage, and EMC can't usually quantify it...seems quite hard to quantify off the cuff.  Looking forward to EMC's prototype.

Well, it is a time mean plot for the NASA GEOS-5 model, which happens to use the NCEP DA system, so it is relevant.  However, as I said before, this type of plot still does not tell you what happens given a loss of data.  It only tells the impact within the context of all of the observations that are assimilated for that model, period, etc.  The only way to actually quantify addition/loss of data is to do full-blown, cycled Observing System Experiments (i.e. data addition or denial experiments).

[ohleary]

Well, it is a time mean plot for the NASA GEOS-5 model, which happens to use the NCEP DA system, so it is relevant.  However, as I said before, this type of plot still does not tell you what happens given a loss of data.  It only tells the impact within the context of all of the observations that are assimilated for that model, period, etc.  The only way to actually quantify addition/loss of data is to do full-blown, cycled Observing System Experiments (i.e. data addition or denial experiments).

 

Okay it's relevant because it's the NCEP GSI, but they might not be ingesting/assimilating the same satellite types/sensors/counts as the GFS does right?

 

So what it shows is for that time period and model, the relative importance of AIRS data with respect to the other types.  Is that accurate?  Would it be nonsensical to pull that graphic up when asked "how important is AIRS data to the GFS?"

[dtk]

Okay it's relevant because it's the NCEP GSI, but they might not be ingesting/assimilating the same satellite types/sensors/counts as the GFS does right?

They are pretty much using the same sensors and channels selections as NCEP.  The biggest difference is that they are using their own model (which has its own error characteristics) and I'm not sure their pseudo-operational system using hybrid DA (they may still be running 3DVAR).  Hopefully, NCEP/EMC will have their ensemble-based FSO monitoring tool up and running in short order.  I'm pretty sure it is a funded project now.

 

So what it shows is for that time period and model, the relative importance of AIRS data with respect to the other types.  Is that accurate?  Would it be nonsensical to pull that graphic up when asked "how important is AIRS data to the GFS?"

That is accurate.  It is not nonsensical at all, so long as you consider/mentioned the aforementioned caveats.  I always like to make sure that people realize what this doesn't show as it is a common misconception of this particular utility (i.e., what would happen if the observations were removed).  The other issue as this type of tool only considers the impact on short term forecasts (say 24-h), since there are all sorts of assumptions that are made including linearity.  It is not really a useful diagnostic (yet) for diagnosing the impact of observations on 5-7 day forecasts, for example.  Of course, errors early in the forecast are related to errors later in the forecast, but once non-linearity takes over (say beyond 24-48h), it becomes complicated.