Addressing the Stagnation of U.S. Operational Numerical Weather Prediction

U.S.  operational numerical weather prediction has stagnated.

• For decades, the skill of the U.S. global model has lagged those of leading international weather prediction centers
• U.S. operational numerical weather prediction (NWP) has fallen behind the state-of-the-science in other areas as well, even though the U.S. is the leader in meteorological research.
• Regional prediction, particularly ensemble-based probabilistic forecasting, is lagging far behind where it should be.
• And as easily demonstrated, U.S. operational weather prediction is not catching up.

But why the stagnation?

It is not for lack of resources.  It is not for the lack of sufficient personnel.  It is not because NOAA/National Weather Service meteorologists or leadership lack interest, knowledge, or motivation.  It is not because the research community does not want to assist.

It is because the structure of research, development, and operations in the U.S. is essentially broken and dysfunctional.  Divided and inefficient, with poor organization.  Lack of strategic planning. And because the operational and research communities are not working together effectively.

The cost of inferior weather prediction to the nation is immense.  Property and lives lost unnecessarily due to poorly predicted storms.  Economic loss and inefficiency resulting from weather prediction errors that could be substantially reduced.

The bottom line of this blog:

U.S. weather prediction will never become state-of-the-art until the U.S. governmental weather prediction enterprise is totally reorganized.

And in this blog I will describe how it could be reorganized to address the inherent problems, allowing U.S. operational prediction to be the best in the world within a few years.

The opportunity

Today there is a great opportunity to fix the problem.  The first thing needed to address a problem is to know that it exists.  The public knows, because there have been many stories in the national media on the trailing U.S. weather prediction enterprise.  Congress knows, and even passed a bill (the U.S. weather research and innovation act of 2017).  NOAA/NWS management knows and have started some well-motivated, but ineffective measures.   Private sector forecasting firms know and have pushed for reforms.  One U.S. National Academy of Sciences report after another has called for change.

Today we have new and highly able high-level managers in NOAA  (administrators Neal Jacobs and Admiral Tim Galludet) that understand the problems and would like to fix them.  We have an administration that wants "to make America great again."

 Incremental changes that leave current structures in place will fail.   After decades of talking about the problem, can we finally take the substantial measures that will make a real difference?

The warning signs

I have published at least a dozen blogs documenting U.S. inferiority in operational numerical weather prediction, the foundation of all weather forecasting in the U.S.  Others have described the problems as well.  But let me show you a few examples that demonstrate the problems.

Consider the 5-day forecast over the Northern Hemisphere for a typical level aloft (here, 500 hPa, about 18,000 ft, half the atmosphere is above and below this level).  Consider a measure of forecast error (root-mean-square error, RMSE) for the U.S. model (the GFS, red line) and the top world model (the European Center, ECMWF, black).  The  top plot below shows the forecast errors from 1996 to today, while the bottom one displays the difference between U.S. and European Center errors.

Note that the U.S. model (red line) has a higher error that the ECWMF at all times.  And the U.S. is not catching up.  In fact, our standing compared to ECWMF has worsened in the past few years.

 Another sign of NOAA/NWS problems is inferior statistical post-processing of model forecasts, a step in which several model forecasts can be combined, biases removed, and the other improvements.  The U.S. private sector uses much more sophisticated approaches (which they secured though cooperation with the university research sector, including the National Center for Atmospheric Research, NCAR). 

Here is an example of forecast accuracy from Chicago from the forecastadvisor website (virtually every other city shows a similar situation).  Major private sector firms (e.g., the Weather Channel), using some of the same model inputs, have much better forecasts that the NWS.

One national academy report after another (e.g., this one) has recommended the U.S needs a convection-allowing, high-resolution ensemble system that is large enough and designed sufficiently well to provide useful forecast uncertainty information.   The NWS has lagged in this area and has only created a kluged a small ensemble (the Storm Scale Ensemble of Opportunity, SSEO, now HREF).  Not good enough.

During the past several years, the NWS Short-Range Ensemble Forecast System (SREF) has stagnated in resolution and design.  Same with the Global Ensemble (GEFS).  The seasonal forecasting system (CFSv2) has remained basically unchanged for years, with no plans for upgrading for many years.

The only exception to this dismal situation is the excellent work done by the NOAA ESRL rapid refresh team  for frequently updated, short-term forecasts (RAP, HRRR).

The NWS is developing a new global modeling system (FV-3), but there is no evidence that it will significantly improve forecasts without better data assimilation and physics (see figure below for comparison of the 5-day forecast).

I could list a dozen other examples of ways U.S. NWP has fallen behind, and virtually everyone who knows the situation would acknowledge that U.S. numerical weather prediction is woefully lagging where it should be.

Poor Coordination and Lack of Cooperation, Coupled with Insufficient Planning and Organization

So how does a nation with leading weather scientists, leadership in computer science and information technology, and substantial vulnerabilities to storms and other weather features (e.g., hurricanes, windstorms, severe thunderstorms, wildfires) end up with second rate forecasting capabilities?

Poor organization, lack of cooperation, acceptance of inferiority, among others.

You can think of U.S. numerical prediction as an old tree that has been left without pruning, trimming, or planning:  its eventually becomes very overgrown and unhealthy over time.

A visual metaphor of U.S. operational NWP

Consider the poor organizational structure of U.S operational numerical weather prediction.  Keep in mind that the National Oceanographic and Atmospheric Administration (NOAA) is part of the Dept. of Commerce, and that the National Weather Service is part of NOAA.

Operational weather modeling is the responsibility of the Environmental Modeling Center (EMC) of the NWS National Centers for Environmental Prediction (NCEP).  But they don't run the models--they let another group in NCEP do it (NCEP Central Operations, NCO).   Now, although EMC is responsible for the weather modeling, they don't develop most of their own models.   That is done by a number of other centers, some in the NWS and some not.  GFDL in Princeton develops some (hurricane models, new global model), NOAA ESRL (not in the NWS) develops others (e.g., rapid refresh models).  The statistical postprocessing is developed outside of NCEP in MDL (Meteorological Development Lab).  Decision making for new models (and the financial management) is done by different folks outside of NCEP  (Office of Science Technology Integration, OSTI).  No one individual or group is responsible for U.S. operational weather research, operations, and research.  It would be hard to make up a more ineffective approach to managing such a complex task.

Even more inefficient, consider that other agencies (like NASA, the Navy, and the Air Force) are all running their own (and different) numerical weather prediction systems, with independent development groups.

Not bad enough?   The research community through the National Center for Atmospheric Research (NCAR) has developed a separate suite of weather and environmental modeling systems, from the global to local scales. 

Worse?   A NOAA Office of Water Prediction and a National Water Center was set up in Alabama to take on the nation's water prediction (and hydrological forecasts), as if that should be separated from the rest of water prediction (this boondoggle was the result of a Senator wanting some pork for his district).

Vast sums are spent on uncoordinated weather forecasting research by the National Science Foundation, NOAA, DOD, NASA and others, much of redundant or never destined to assist operational capabilities.

And did I mention that there is no concrete strategic plan in NOAA/NWS for the development of U.S. environmental prediction?   Just vague platitudes.   This is in stark contrast to the detailed planning by NOAA/NWS's big competitors (e.g., the European Center, the UKMET office, etc), which as extensive, detailed, and coherent planning processes.

The bottom line:  we have a disastrously inefficient system of developing, running and post-processing numerical weather prediction guidance for the U.S., with agencies doing their own things, disorganization growing, and U.S. capabilities stagnating.

Most decision makers are thinking about protecting their own turf and resources, and few high-level NOAA administrators have known enough to see the problem or were willing to take the heat for the necessary changes.  Numerical weather prediction is technically very challenging and few in Congress has sufficient background to revamp the organization of U.S. weather forecasting.

The latter point is worth repeating:  numerical weather prediction is perhaps the most complex activity of our species, involving billion dollar satellites, global data collection, running complex models that encompass molecular to global scales, and that use the largest computers on the planet.  You can't do this in a haphazard, uncoordinated away as done in the U.S. and expect to be state-of-the science.

The current NOAA/NWS response will not solve the problem

NOAA/NWS management know there is a big problem, and to be honest, some things have improved.  A few years ago, there was outright tension between model developers in NOAA ESRL (Earth System Research Lab in Boulder) and the NWS EMC folks.  That is much better now.  The NWS is putting more funds into extramural research in the academic community and is holding workshops and meetings to gain input.  Recently, the NWS begin developing a new global model (based on the GFDL FV-3 system) to replace the hoary old model (the GFS).

NOAA/NWS has not dealt with the central problem of poor organization, coordination and planning.

They have not created a structure to develop actionable strategic plans.  Different folks are responsible for model development and operations.  The  NOAA extramural funding has often not been well spent.  The University community is drifting away, with nearly all using the NCAR models.  There is  insufficient computer resources to support weather prediction and weather prediction research that the nation requires. And there is little evidence that the new global model will improve verification scores very much, since the real problems are in data assimilation, physics, ensembles, and post processing.

The National Center for Atmospheric Research Mesa Lab

The NWS has held workshops and organized all kinds gatherings to garner support, but these have led to long laundry lists, with little ability to lead to actionable, organized efforts.

And let me be clear that I am not singling out NOAA management for this problem.  It is difficult for a line manager in NOAA to visualize and change a complex organizational structure.   Important partners (like NCAR, DOD, NASA) have often shown little interest in working with NOAA, wanting to maintain their own modeling sandbox. Congress has not been sufficiently attentive to the dysfunctional structures they have created and maintained.

How to fix the problem of U.S. numerical weather and environmental prediction

The key is reorganization, coordination and planning.  The best possible circumstance would be to bring all environmental prediction activities of NOAA into one group-- research, development, operations--with one individual responsible for the whole operation.  We are talking about combining the NWP activities and responsibilities of NWS NCEP, NWS MDL, NOAA ESRL, NOAA GFLD, the National Water Center, and NWS NCO (and others) into one group.  Even better would be to create a truly national center, and bring in the NASA and DOD components--but that is a heavy lift.

But let's imagine something that is smaller lift, that might be a good first step.  Leave responsibility for running the weather/environmental prediction models in the National Weather Service, but move all model development and testing to a new integrated entity within NOAA:  The National Environmental Prediction Research and Development Center (NEPRDC).  A schematic of the organizational structure is shown below.  It would include a chief scientist to help organize and lead the scientific work, with a strong scientific advisory committee.

One organization in NOAA would be responsible for research, development, and testing of the nation's environmental models (weather, ocean, hydrological, coastal, etc.)  The center would have ample supercomputer resources for model testing (which does not exist now) and control of extramural funding, which would help support university (and other) researchers working in research that will directly address current and future modeling problems.

An ideal location for the center would be in Boulder, Colorado-- the center of U.S. weather research, with both NOAA ESRL and NCAR (the university community's center) located there today.   The Developmental Testbed Center (which can provide extramural support for U.S models and help evaluate the new ones, is already in Boulder). Some  NWS EMC personnel (who are now in DC) could work as a satellite center in Boulder.  Bringing NCAR  (also in Boulder) and NOAA together through such a center is critical, and Boulder is a far more central venue than DC for combining the nation's modeling efforts.

Improvements in environmental prediction will only come from sustained, coordinated hard work in physics, data assimilation, and other key areas.  This requires coherent planning and the coordination of the vast scientific and technical resources of the U.S.  The current system is incapable of such effort, the proposed one will be up to the task.

As a first step in fleshing out this proposal might be a national workshop on U.S. environmental prediction, bringing together the entire environmental prediction community, with a detailed white paper coming out of it.  And with so many changes in the bureaucratic structures, Congress will have to be involved.

But in the end, it is clear that the current structures, the result of legacy and administrative drift over decades, are failing.   Only major restructuring and reimagining of U.S. environmental prediction can result in the necessary changes.    The U.S. can easily regain leadership in weather prediction if we only have the will to acknowledge the current failed structures and replace them with something better. 

The U.S. has fallen behind in so many areas due to complacency, poor leadership, self-interest, and loss of energy.  Can we do better in weather and environmental prediction, rebuilding our capabilities when we still have the resources to be the best and give the American people state-of-science forecasts?   I hope so.