Presentation to climate.com

advertisement
Weather and Climate Prediction
Cliff Mass
University of Washington
Outline
• Evolution of numerical weather prediction
• Application to regional climate and seasonal
forecasting
• Some ideas for climate.com
The Resolution Revolution in
Numerical Modeling
NGM,
80 km,
1995
1995
2007-2008
4-km UW
MM5
System
2013 WRF Model at 1.3 km
Last Week
WRF, 1.3 km
1.33 km resolution temperature
But just as important as the
computer revolution has been the
weather data revolution, with
satellites giving us three dimensional
data over the entire planet
Example: The Pacific Data Void No
Longer Exists
Cloud Track Winds
Better than Star
Trek!
NOAA Polar
Orbiter Weather
Satellite
Satellite Sensors Provide Thousands of High
Quality Vertical Soundings Daily over the Pacific
Cosmic GPS Satellites Provide More
Soundings!
Impacts
• The addition of massive amounts of
new observations is causing a steady
improvement in weather prediction
• We are now starting to see frequent
examples of forecast skill past one
week:
• Hurricane Sandy is only one example
Superstorm Sandy: well predicted
over a week ahead of time
ECMWF Forecast of Sea Level Pressure
Observed
180 hr (7.5 days)
Skill Improvements (ECMWF)
Major improvements, mainly due to satellite data and
improved models
A Fundamental Problem
• The way we have been forecasting
has been essentially flawed.
• The atmosphere is a chaotic
system, in which small differences
in the initialization…well within
observational error… can have
large impacts on the forecasts,
particularly for longer forecasts.
• Not unlike a pinball game….
A Fundamental Problem
• Similarly, uncertainty in our model physics
(e.g., clouds and precipitation processes) also
produces uncertainty in forecasts.
• Thus, all forecasts have some uncertainty.
• The uncertainty generally increases in time.
This is Ridiculous!
Forecast Probabilistically
• We should be using probabilities for all our
forecasts or at least providing the range of
possibilities.
• There is an approach to handling this issue that
is being explored by the forecasting
community…ensemble forecasts
Ensemble Prediction
• Instead of making one forecast…make
many…each with a slightly different
initialization or different model physics.
• Possible to do this now with the vastly greater
computation resources that are available.
Ensemble Prediction
•Can use ensembles to give the
probabilities that some weather
feature will occur.
• Ensemble mean is more accurate
than any individual member.
•Can also predict forecast skill!
•When forecasts are similar, forecast
skill is generally higher.
•When forecasts differ greatly,
forecast skill is less.
The Transition
Numerical Weather Prediction is
progressively transitioning to ensemble
prediction and ensemble data
assimilation
The Data Assimilation Revolution
• The combing of observations and model output
to provide a three-dimensional description of
the atmosphere is called data assimilation.
• Until recently the leading technology was
4DVAR, 4D Variational Data Assimilation. NWS
has lagged in using this.
• Ensemble-based data assimilation has many
advantages and is increasingly being used.
• Future convergence between ensemble
prediction and data assimilation is probable.
The Technology of Regional NWP
Can Be Used for Seasonal or Climate
Prediction
Regional Dynamical Downscaling
• For regional numerical weather prediction we
can embed high resolution models within a
coarse resolution global forecasts.
• Can do the same thing for climate/seasonal
prediction by simply replacing global weather
forecasting models with global climate models
(GCMs) or seasonal global prediction models
(e.g., NOAA’s Climate Forecast System-CFS)
• Just need the computer resources.
UW Regional Dynamical
Downscaling
• Have completed a number of 100-year
regional climate simulations using the WRF
model at 12-km grid spacing.
• Driven by a half-dozen different climate
models and emission scenarios.
Change in Winter Surface Air Temperatures (F)
Change in Snowpack from 1990 to 2090
-40%
0%
+40%
Climate Simulations
• Will be running with many more climate
model driven simulations.
• Now evaluating the use for monthly and
seasonal prediction at high resolution using
output from the NOAA CFS model, a coupled
atmosphere/ocean modeling system.
• Is there useful predictive skill at 1-9 months
for mean quantities?
Some Ideas
Providing Useful Climate Information
Based on Historical Records
• One of the greatest deficiencies of the climate
community and the NOAA/NWS.
• There is a huge amount historical climate data
available (station data, reanalysis information)
but it is difficult or impossible for folks to get
the actionable information they need.
Some Climate Questions
• When is the best time for wedding in Seattle?
• What is the windiest time of the day in Tucson?
• When are the high temperatures in Rome
between 60 and 70?
• I want to take a vacation the second week in
March. Where will temperatures between 70
and 80 with less than a 30% chance of rain,
within a 7 hr flight?
• What is the climatological last day of freezing
temperatures at my house?
Climate Information Today
• Pre-generated tables and graphics.
National Climatic Data Center (NCDC)
God Help You if You are a Layman Looking
for Climate Information at the NCDC Site
Low Hanging Fruit
• Secure U.S. and International
Climate/Historical Weather Data (available
from NCAR, NCDC, and others for minimal
costs). I assume climate.com already has it.
• Put into a relational data base.
• Build an interface/inquiry engine using natural
language queries if possible.
Climate Apps: The Surface Has Been
Barely Scratched
GardenKeeper
GardenKeep
er
• Using calibrated radar-based
precipitation data, tells you when
watering is necessary at your
location (considering water
demands of your plants and
evapotranspiration based on recent
weather)
• Warns when freezing conditions are
imminent during the winter.
• Tells you when you can plant seeds
and young plants in the spring
Custom Automated Pinpoint
Forecasts
The Idea
• The owner of a vineyard wants accurate
forecasts that considers the microclimate of
his property.
• An owner of a private airport wants forecasts
tailored exactly to his airfield.
• The harbormaster of a yacht club wants
accurate forecasts at his location.
There is a way.
• They contact climate.com for pinpoint forecasting
service.
• Working with the client, weather instrumentation
is installed at the exact locations of interest, with
the data retrieved via wifi, cell phone, or wired
connection.
• As soon as several weeks of data are available,
statistical postprocessing is applied to operational
models (e.g., GFS, NAM) to provide an optimal
forecast at the observation location.
There is a way
• The longer the observations are in place the
better the statistical postprocessing.
• Could use linear regression, extended
logistical regression, or other approaches.
• Forecast biases could be radically reduced at
such sites.
• Could use ensembles or analog methods to
give probabilistic predictions.
The End
Download