on the use of data assimilation in the assessment of the...
TRANSCRIPT
On the use of data assimilation in the
assessment of the cost/benefit of
meterological observations and observing
systems.
Lars Peter Riishojgaard
Director, Joint Center for Satellite Data Assimilation
Chair, OPAG-IOS, WMO Commission for Basic Systems
6th WMO Data Assimilation Symposium
Overview
• Money and weather forecasting – Enabling capabilities
• Data and weather forecasting – Data requirements
– Cost of meteorological observations
• Data impacts and the WMO RRR
• Can we put a monetary value on individual observing systems and/or individual observations? (and should we?)
– Role of Data Assimilation community and WMO RRR
• Final remarks
WMO Data Assimilation Symposium,
College Park Oct 7-11 2013 2
WMO Data Assimilation Symposium, College
Park Oct 7-11 2013
Weather Prediction and the US
Economy; A Macroscopic View
• Department of Commerce: “20% of overall US economy is weather sensitive”: ~$3 trillion/year
– Impact to air and surface transportation, agriculture, construction, energy production and distribution, etc.
• Assume that half of this is “forecast sensitive”: $1.5 trillion/year
• Assume that the potential savings due to weather forecasting amount to 5% of the “forecast sensitive total”: ~$75B/year
3
(discussed during CBS TECO in Windhoek,
2010)
WMO Data Assimilation Symposium, College
Park Oct 7-11 2013
… a Macroscopic View … (II)
• “Perfect forecast” is an NWP run with useful skill at two
weeks!
• 0 h useful forecast range => $0 in savings
• 336 h useful forecast (two weeks maximum predictability)
range => $75B in savings
• Assume now that the savings are distributed linearly over the
achieved forecast range for the global NWP system: – $75B/336h ~ $223B/hr
• This implies that the value to the United States economy of
weather observations, dissemination, forecast products and
services is >$220M per hour of forecast range per year !
4
(discussed during TECO in Windhoek, 2010)
The global picture
• The amount of $75B/year is one estimate of the
magnitude of the total potential socioeconomic benefit
of weather prediction activities to the US economy
• Scaling exercise, using World Bank (2011) numbers:
• Annual GDP of United States: ~$15T
• Annual GDP of all nations combined: ~$70T
– Assuming on average (i) equal sensitivity to weather, and (ii)
equal potential benefits from ability to predict across all
nations, we get an estimated
$75B *($15T/$70T) = $350B as the total global potential
benefit of weather prediction activities (indicating a likely
range of $100B to $1T)
WMO Data Assimilation Symposium,
College Park Oct 7-11 2013 5
Weather Prediction Enabling Capabilities
1. Observing Systems (GOS0
2. Dissemination Systems
3. Numerical Weather Prediction
– Science (modeling, data assimilation)
– High-end computing
WMO Data Assimilation Symposium,
College Park Oct 7-11 2013 6
• 1, 2 and 3 are of a foundational nature
• Among the foundational capabilities, 1 represents
the single largest expenditure
WMO Data Assimilation Symposium, College
Park Oct 7-11 2013
NWP requirements for upper-
air data coverage
7
Hence the need for a global
observing system, irrespective
of target location of forecast!
Estimating the total cost of running
all components of the GOS
• Informal exercise launched by the WMO Commission
for Basic Systems in 2012
• Approach was (perhaps too?) simple:
– Survey a small, but representative number of WMO member
states about their total annual investment in running,
maintaining and updating the GOS
– Use Cost-to-GDP ratios to extrapolate that to the rest of the
WMO members
– Add estimates for external (non-NMHS owned) contributions:
• Satellite systems
• Aircraft observations
• Third party networks
• Marine observations
• …
WMO Data Assimilation Symposium,
College Park Oct 7-11 2013 8
NMHS input to survey (anonymous)
Country GOS cost
(K USD, 2011)
Expenditure
on GOS, GDP
fraction
Comment
1
8330 1.3 x 10-5 Excludes radar data
2 39096 1.1 x 10-5
3 7793 1.6 x 10-3 (LDC)
4 14400 2.3 x 10-4 (Amortization
unclear)
… … … …
WMO Data Assimilation Symposium,
College Park Oct 7-11 2013 9
What do we think we know about the global
cost of acquiring the observations?
• Ratios are too disparate to be used for scaling, but
the running costs of the conventional parts of the
GOS appear to be in the single-digit $B-range
• “Non-NMHS”-provided data (satellite, third-party
networks, marine, aircraft observations) add an
estimated $5B total
• Total costs of all meteorological observations
add up to an estimated $5-10B per year
• That is a lot of money, and decision-makers are
on the hunt for justification and objective metrics!
WMO Data Assimilation Symposium,
College Park Oct 7-11 2013 10
What do NWP and data assimilation have to do with this?
Objective, quantitative metrics:
NWP poses a well-defined prediction problem with a “right” answer
(and an infinity of wrong ones)
Well-defined measures for quality of output
Well-established methodologies for assigning merit (or blame) to individual observing systems
WMO Data Assimilation Symposium, College Park Oct 7-11 2013 11
WMO Commission for Basic System, Rolling Review of Requirements
Under its Commission for Basic Systems WMO maintains a standing Expert Team (ET-EGOS) which is responsible for documenting
1. Observational data requirements for all 12 WMO application areas (database)
2. Capability of all relevant observing system (databases)
3. Statements of Guidance, or gap analyses, matching 1. against 2 (a brief narrative for each application area)
4. …
ET-EGOS is helped in its work by numerous other ET’s and
Rapporteurs and by the WMO NWP Impact Workshops taking place once every four years
WMO Data Assimilation Symposium, College Park Oct 7-11 2013 12
WMO Workshops on the Impact of Various
Observing Systems on NWP
Five Workshops so far (the last two in close collaboration with THORPEX):
• 1st - Geneva, 1997
• 2nd – Toulouse, 2000
• 3rd – Alpbach, 2004
• 4th – Geneva, 2008
- Workshop Report available on
http://www.wmo.int/pages/prog/www/OSY/Reports/NWP-4_Geneva2008_index.html
• 5th – Sedona AZ, May 2012
http://www.wmo.int/pages/prog/www/OSY/Meetings/NWP5_Sedona2012/Final_Report.pdf
Workshops aim to bring together scientists from all major NWP centers to
assess the contribution to forecast skill of various elements of the global
observing system; guidance to participants provided well in advance of
Workshop itself.
13 WMO Data Assimilation Symposium,
College Park Oct 7-11 2013
OSE and FSO (I)
• OSEs (Observing System Experiments) are based on data denial (or addition)
• Impact focuses on the medium to long range
• Results show the impact of withdrawing (or adding) certain data
• OSE results are absolute; e.g. “observing system X extends the useful forecast range by N hours in the NH”
WMO Data Assimilation Symposium, College Park Oct 7-11 2013
14
Jung et al., WMO Impact Workshop in Sedona, May 2012
OSE and FSO (II)
• FSO (Forecast Sensitivity to Observations) are based on the adjoint of the model/analysis system or an ensemble approach
• Approach focuses exclusively on the short (quasi-linear) range
• Results show the impact of observations in the presence of all other observations
• FSO measures of impact are relative (e.g. often expressed in percentages that add up 100, even for poor forecasts or poor system performance)
WMO Data Assimilation Symposium, College Park Oct 7-11 2013
15
Gelaro et al, Fifth WMO Impact Workshop, Sedona 2012
4th WMO Impact Workshop, Alpbach 2008; impact
summary slide
Overall impact
(“marginal skill”) on
global NWP
16 WMO Data Assimilation Symposium,
College Park Oct 7-11 2013
Some Preliminary Conclusions from the Fifth
WMO Impact Workshop in Sedona, May 2012
• Modern, 4-dimensional data assimilation methods (4D-VAR,
ENKF) have led to greatly improved use of data, especially of
– Asynoptic data (e.g. aircraft, satellite observations)
– Observations with complex relationships between measured
and model variables (satellite radiances, GPSRO, radar,…)
• Broad consensus about highest-ranking contributors to forecast
skill, but not necessarily about their ranking order:
– AMSU-A (microwave temperature sounder)
– AIRS/IASI (hyper-spectral infrared sounders)
– Radiosondes
– Aircraft observations
– Atmospheric motion vectors (feature tracking satellite winds)
WMO Data Assimilation Symposium,
College Park Oct 7-11 2013 17
Some Preliminary Conclusions (II)
• Radio occultation data (GPSRO) also have substantial impact
but data volumes are currently declining as COSMIC is
approaching the end of its lifetime
• There is now no single, dominating satellite sensor; several
sensors contribute to forecast skill in roughly equal measure
• The relative impacts of specific observation types depends on
which other observations are used and how
– If certain data are withheld, other datatypes can in some
contexts compensate for the lost skill
• However, the continued value of in situ data, and in particular of
wind measurements, was clearly demonstrated
• Regional data assimilation systems making progress in the use
of radar and satellite observations
– Radiance assimilation still problematic
WMO Data Assimilation Symposium,
College Park Oct 7-11 2013 18
Forecast impact experiment from Dec. 2010 to Jan. 2011
Impact Impact / Obs. number
WMO Workshop on the Impact of Various Observing Systems on NWP Sedona – 22-25 May 2012
Could we use this type of FSO information
to rank observing systems by impact per
dollar?
1
9
Of course we can! Simply divide the
impact by the cost of running the
system and come up with a third
“impact per dollar” bar chart!
Obs impact and value, versus cost
• Armed with the following three pieces of information
– Cost of acquiring the observations (difficult but not impossible to
acquire this information)
– Overall economic benefits of products derived from
observations (can be WAG’ed or properly analyzed by trained
economists)
– Individual contribution of observing systems to NWP skill (done
through RRR, previous slide),
we could in fact take the reasoning on the previous
slide one step further and provide cost/benefit
numbers for individual components of the GOS; it is
not an easy exercise, but probably doable
WMO Data Assimilation Symposium,
College Park Oct 7-11 2013 20
Should we use DA methodology for
Cost Benefit analysis for observations?
• Caveat!
– The respective contributions of the components of the GOS vary
from center to center, even at comparable levels of skill
– The contributions vary depending on which other GOS
components are used in the experiments
– OSE results are expensive to acquire, and often inconclusive
– FSO impacts are relative; even if the overall forecast
performance is poor, some observing systems may stand out
due to their large share in a modest improvement
– The sum of the contributions always add up to 100, down until
the very last GOS component standing
• When are we done cutting?
WMO Data Assimilation Symposium,
College Park Oct 7-11 2013 21
Final remarks
• The economic impact of weather is relatively well understood
• In contrast, the economic impact of weather prediction is generally not well studied and documented
• The cost/benefit of meteorological observations are a subject of intense interest among program managers and decision makers
– The costs are incurred (and known) mostly at the regional levels, but the benefits are realized and assessed globally
– Tendency to focus too much on NWP diagnostics due to their compelling nature
WMO Data Assimilation Symposium,
College Park Oct 7-11 2013 22
Final remarks (II)
• The NWP and data assimilation community is being dragged into this whether we like it or not
– Victim of our own success
– Does our community need to collectively (e.g. through WMO/CAS, THORPEX) adopt a position on this issue?
• We need to remind people that while NWP is important and can provide very compelling metrics, it is only one of many application areas supported by the GOS
• New, widely adopted community standard metrics needed?
– Not much room for further improvement of the 500 hPa AC scores
WMO Data Assimilation Symposium,
College Park Oct 7-11 2013 23