Considering the importance of this decision regarding the main US numerical weather prediction model, this blog will attempt to describe:
- What this means.
- Some issues regarding this decision.
- Some recomendations on the route the NWS must take if it wants to develop a world class global modeling system.
As I have discussed in several blogs (e.g., here), the current NWS Global Forecast System (GFS) is out-of-date in many ways. It was designed for low-resolution weather prediction and does not scale well on modern supercomputers (which can have tens or hundreds of thousands of processors). GFS physics (the descriptions of physical processes such as convection and clouds/precipitation) are ancient, representation the state of the science 20-30 years ago. The range and amount of data assimilated into GFS forecasts is less than leading centers (such as the European Center) and the quality control system lags. We can do much better.
72-h cumulative precipitation from the GFS forecast initialized at 1800 UTC 30 July 2016. The "popcorn" look to the precipitation in mountains is not realistic and reflects inferior, 20-year old precipitation physics in the model.
The NWS GFS global model is now third or fourth globally, a particular embarrassment since the range and quality of US. weather research is by far the greatest in the world. The general U.S. population finally understood the situation when important forecasts by the GFS (e.g., Hurricane Sandy) were clearly inferior to those of the European Center and UKMET office. So does NWS/ NOAA management and Congress, which provided funds for both a new operational computer and model replacement.
Replacing the core
The first step in moving to a new modeling system is to replace its guts: the dynamical core. But what is this? A reasonable analog is the chassis of a a car, including the frame and engine (see below). No body, no seats, no electronics, no AC or ventilation, no steering column, no gas tank. Although it is perhaps 25% of what makes up a modern car, the chassis does serve as the frame on which everything else is mounted and connected. To build the best car, you want a good chassis.
Numerical prediction dynamical cores are like that. They provide the framework of the model: its grid structure, the central atmospheric equations (conservation of mass, momentum, energy, and water substance), how the model can provide more detail in local areas (e.g., nesting or variable resolution grid elements).
The dynamical core does not take encompass key physical processes, such as boundary layer drag, condensation in clouds, radiation from the sun/clouds/surface, and much more. And the core does not deal with the critical element of data assimilation, taking all of the sources of weather information to create a physically consistent initialization, or starting place, for the forecast.
So a key point is this: although the National Weather Service picked a dynamical core, MOST of the work to develop a state of science global model for the US is still ahead.
The Decision is Made: GFDL's FV-3
The announcement last week was that the National Weather Service picked the FV-3. In some sense, it was the safe choice, but I believe it wasn't the best choice. Let me be clear: FV-3 is far, far better than the current GFS model and the handiwork of a very talented numerical modeler, S.J. Lin. But I believe it was not the best choice for important reasons. The alternative (MPAS) uses a grid structure that is far better at high resolution (like 1-4 km grid spacing), which is where global modeling is going in the next ten years. Thus, picking FV-3 was like buying a house that is good for two, when you are planning on a large family. FV-3 is far more conventional that MPAS in structure (e.g., it has grids, not the hexagonal elements of MPAS) and its compromises make it faster at the same grid spacing.
The NWS might have been better off with a house more suitable for the future
But just as important, picking MPAS would have brought the research and NOAA/NWS efforts together, and I believe the lack of nation cooperation is one of the key reasons US operational NWP is lagging.
But the decision, good or bad, has been made, and as noted above, the dynamical core is perhaps 25% of a modern modeling system. As Caesar would say, the die has been cast. And if future testing supports the superiority of MPAS, its dynamical core could be switched for FV-3.
So What Should Be Done Now?
Get a plan
The National Weather Service needs a plan to complete the remaining 75% of the new global modeling system and to ensure that it is state-of-the-art and STAYS state-of-the-art.
The trouble? They don't have a plan at this point and need to put it together immediately. Actually, they need TWO plans: a broad strategic plan on where they are going during the next ten years, and a shorter term implementation plan for the next few years.
Planning has been a singular weakness of the National Weather Service weather forecast modeling efforts, with a tendency to do too little of it, and what they have done has been in-house with inadequate vision.
It is clear that NOAA/NWS needs to open up the planning process to include the research community, the private sector, and others. They have done a bit of this with the NCAR UCACN and UMAC committees, but they need to take the planning process to a completely different level in detail, vision, and inclusion. Perhaps, a standing committee that mest several times a year. Let's call it the NOAA Global Model Advisory Committee (NGMAC), with modeling experts from NOAA (ESRL and GFDL), academia, NCAR, the private sector, and other Federal agencies. The committee would recommend a broad strategic plan to NOAA/NWS managemen,t and subcommittees (with additional experts) could build the implementation plans.
Build a National Team
The history of the past few decades is clear: when the National Weather Service develops models in house without the entrainment of the outside community, the models are inevitably not state of the science. But when they partner with others (such as ESRL working with NCAR on the High Resolution Rapid Refresh, HRRR), amazing things happen.
Specifically, NOAA must make the model development a cooperative national effort. Bill Kuo of UCAR , a wise and experienced player in US numerical weather prediction efforts, has suggested a wonderful plan. Center the development of the new national modeling effort in UCAR (the University Corporation for Atmospheric Research), with a center of gravity in Boulder, Colorado (the intellectual center of the US weather community). A team of ESRL researchers, NCAR and academic scientists, private sector modelers, GFDL and NOAA EMC staff) could build the system there. The UCAR Developmental Testbed Center (also in Boulder ) would provide extensive testing of new model components and could organize community support. ESRL, with considerable personnel in model development and basic physics, is found in Boulder.
EMC scientists in DC could work on the new system components, complete pre-operational testing, and take on key tasks such as improving quality control.
NOAA/NWS could continue and expand its research funding of advances that support the new system and hopefully NSF could help support associated basic research.
Think Unified and Probabilistic.
The future of numerical weather prediction is clearly moving towards unified modeling across all scales. That is the way the real atmosphere works. The days of one model for global forecasts and another at ultra-high resolution are numbered, if not over all ready. Leading groups, such as the UKMET Office, have already moved to a unified approach. NOAA must do the same.
Future prediction will be essentially probabilistic at all scales, with ensembles (running models many times with different initial states and physics) being the foundation. The NWS has generally not given enough emphasis and thought to their ensemble systems, with their size and resolution being insufficient. They have also held back from creating a high-resolution (2-4 km grid spacing) over the US, something recommended by endless National Academy reports and national workshops. Ensembles need more resources and thought.
US NWP has become risk adverse and conservative. It needs to be more forward-leaning and innovative.
The Bottom Line
The National Weather Service made a conservative choice for the core of its new global modeling system, but that is not the end of the world. The U.S. can still build a far better global prediction system IF NOAA/NWS will methodically plan the development of the model and entrain the large and strong US research community. Such strategic planning and inclusive development would be a very different approach than NOAA has done in the past. Leaders of NOAA (e.g., Kathryn Sullivan) and the Weather Service (Louis Uccellini, Director, and Bill Lapenta, had of NCEP) say they are ready to follow such an approach, and with their support, U.S. numerical weather prediction could move in a new direction. If U.S. numerical weather prediction stagnates because the NWS follows the path of the past several decades, you will know who to blame.
Cliff:
ReplyDeleteI've been reading your blog for almost a year now, and greatly appreciate the effort you've put in presenting timely comments and insight into weather (and other topics) in the Pacific Northwest.
Although I'm not a meteorologist, I am a scientist (Ph.D. in Astronomy), and have been following your discussion of how the NWS is going about updating both their hardware and software with some interest.
Your arguments for suggesting that the NWS choose the MPAS model over the FV-3 model are well presented and quite convincing, but I believe that you have omitted an additional, but very important aspect: the training of future NWS meteorologists.
In a nutshell, I suspect that students who are training to both use and improve existing models will be trained using the MPAS model, as it is the model which has been developed by the research community.
It makes little sense for a student, who has made a significant contribution to some aspect of the underlying physics by spending years living and breathing the MPAS model, to be hired by the NWS only to spend the first year(s) of their job learning the idiosyncrasies of a different model (and perhaps finding that the results of their earlier work may not be easily implementable). Note also that the rate at which the FV-3 model evolves will also be limited by the much smaller number of active researchers using this model (a point you also made in your most recent blog post).
Finally, it makes little financial sense for the national research funding agencies (NSF, NASA, NOAA) to fund meteorological research which utilizes the MPAS model while simultaneously funding the NWS to research and implement a competing model.
Again, thanks for your exemplary efforts in explaining what's keeping me from seeing the stars that I love in the Pacific Northwest.
Dave Lien
I'm surprised none of your compadres have commented on this very insightful post. Hoping TPTB do right for future generations.
ReplyDelete