Data Assimilation Education Forum
Part I: Overview of Data Assimilation
Identifying Current and Future Shortfalls in
Data Assimilation Education
January 21, 2008
NASA/GSFC/Global Modeling and
Thank you to…..
• Eugenia Kalnay, UMD
• Robert Miller, OSU
• Carl Wunsch, MIT
• Keith Haines, U. Reading
• Nancy Nichols, U. Reading
• Lack of qualified personnel in data assimilation
(state estimation, inverse methods) for large
• Lack of qualified personnel with interest and
experience in radiative transfer
• Lack of programming and computing skills for
high end computers
University Experience - 1
• UMD: a very successful data assimilation educational program -
• Interdisciplinary: Mathematicians, Physicists, Atmospheric and Oceanic scientists
• Started in 2001, 12 completed PhD’s, about 10 underway.
• 2 graduate courses on data assimilation taught on a regular basis
• Introductory course attracts some of the best students in Atmospheric Sciences and Applied
Math (~ 8-10 students/ year)
• ~1/2 take the follow-on, advanced course
• It has been essential to develop a collection of simple models and methods for data assimilation
that the students can learn from and work with.
• Developing the infrastructure (including computational resources) for
development and testing of methods is essential.
• Need access to an operational system
• 2007 JCSDA summer workshop - ~ 60 applicants - lecturers
from UMD, JCSDA, NCEP, GMAO, NRL, and other universities.
University Experience - 2
• OSU: Andrew Bennett’s summer school is now taught as a regular
course, but # of eligible students is pretty small
• Problem: understanding DA requires a level of mathematical
sophistication that most students simply do not have
• Solution: incorporate more mathematical concepts into graduate
courses from the beginning, including problems of high
• Problem: students are averse to this!
University Experience - 3
• MIT: “Inference from Data and Models” - ~50% of students not
meteorology or oceanography
• Much of what is now done in NWP centers and in large university
projects involves the application of these ideas to real systems to get
real results --- a numerical engineering problem
• A growing interest in the methods by the Engineering Schools
• Headed toward a situation in which data assimilation-like methods will
become part of the standard engineering curriculum
• The students involved have to learn enough about how we do things to
make it work, but none of them seems interested in extending the
• A growing tendency to run big numerical models as black boxes, and to
download vast data sets from the web that they then regard as 'truth'.
We need to produce a new generation that is adept in both using models
and data, has a realistic sense of what both are good for, and retains a
healthy skepticism about what was assumed and done.
U. Reading Experience
• Data assimilation program promoted through Mathematics and
• Strong offering of PhD projects - PhD students are supported by
grants from the UK Research Councils.
• NERC and EPSRC PhD CASE awards - in co-operation with industry and
scientific research agencies. The project is agreed between the
university and the industry and must be scientifically competitive to win
the award. The project has both industrial and academic supervisors
and the student is funded to spend some part of the time getting work
experience in the industry. This is attractive to good students and is an
excellent way to gain interest in the subject!
• At least 16 students funded by this arrangement working on topics in
data assimilation with support from the Met Office.
NERC: Natural Environment Research Council
EPSRC: Engineering and Physical Sciences Research Council
CASE: Co-operative Awards in Science and Engineering [Cooperating organization provides at least 1/3 of required funding]
U. Reading Experience (ctd)
• NERC funded Centre of Excellence in Data Assimilation (DARC) -
a distributed research centre specifically in Data Assimilation
with the Directorate centred at Reading:
･University of Reading
･University of Oxford
･University of Cambridge
･Rutherford Appleton Laboratory
･University of Leeds, and
• DARC Post-Doctoral Fellows at Reading help to supervise PhD
students, and also collaborate with the MetOffice and ECMWF
on projects on data assimilation.
• This creates a critical mass doing research in the area -
provides a strong research environment for students.
U. Reading Experience (ctd)
• One of the DARC’s main objectives is to provide training in data
assimilation - lecturing and tutorial teaching in summer schools,
NATO ASI meetings, and and similar training courses, by giving
seminars at other institutions, …..
• A web-page dedicated to providing tutorial examples and
computer programs that can be used freely by other groups for
training. This is a popular website and is used internationally. We
get around 100 hits per month on this site.
What we need
• Ph.D. level scientists with good grounding in data assimilation methods
and experience with large models and/or expertise relevant to satellite
data (radiative transfer)
• Scientists who can advance our science, not just apply existing systems
as a black box to a science problem
• Scientists who have some experience with (or exposure to) large,
complex systems and models and don’t require extensive OJT
• Scientists who can program in modern Fortran on high end computing
• Challenge: Need to entice students to be interested in
assimilation development, not just in using an existing system
• Challenge: Need to excite students to work in this discipline
• Need to have a partnership with Math depts to ensure that
students have a strong background in stats, analysis, and PDEs
• Need to reach out to engineering schools - appropriate
curriculum and new graduates
• Universities consortium ? - going it alone is not as effective
• Summer schools not effective by themselves
• Need to involve “industry” in partnership with the university -
don’t just leave it to the university - requires an investment
from the operational centers
• Partnership with operational centers is best way to provide
experience with systems of relevant complexity and with
relevant computational infrastructure