NATS 101 Lecture 19 Weather Forecasting Keep Clickers Handy 108 h ensemble forecast valid 1200 utc 20100401

NATS 101

Lecture 19Weather Forecasting

Keep Clickers HandyKeep Clickers Handy

108 h ensemble forecast valid 1200 utc 20100401

http://www.weatheroffice.gc.ca/ensemble/index_e.html

Important Upcoming Deadlines

• Apr 5 Monday

Homework08

submissions deadline 6:00 PM• Apr 6 Tuesday

Quiz 3• Apr 12 Monday

Project Due at turnitin.com

submissions deadline 11:00 PM

Reasons to Forecast Weather

• Should I bring my umbrella to work today?• Should Miami be evacuated for a hurricane?• How much heating oil should a refinery process

for the upcoming winter?

• Will the average temperature change if CO2 levels double during the next 100 years?

• How much to charge for flood insurance?

These questions require weather-climate forecasts for today, a few days, months, years, decades

Forecasting Questions

• How are weather forecasts made?• How accurate are current weather forecasts?• How accurate can weather forecasts be?

How can we solve the problem?

Simple Approach vs. Complex Approach(a.k.a. Cheap vs. Costly)

Simple forecasting approaches should be used as a “sanity check” to see if complex

approaches are worth it.

Simple Approach #1Persistence Forecast

Persistence: Future atmospheric state is the same as the current state.

Good Example: Tropical rainforest during wet season when the ITCZ is around. It’s raining today, so predict rain for tomorrow.

HIGH: 83°FLOW: 70°F



TODAY THURSDAY FRIDAY

Simple Approach #2Trend forecast

Trend: Add past change to current condition to obtain forecast for future state

Good Example: Temperature in Tucson increasing at 3°F per hour in the morning on a clear, calm day. Use this to forecast temperatures later in afternoon because the surface heats at a steady rate due to solar heating.

9 AM 12 PM 3 PM

93°F 96°F 99°F

Simple Approach #3Climatology forecast

Climatology: Forecast future state as the average of past weather for a given period

Good example: Forecast about six inches of rain to occur during the monsoon in Tucson, the average for the 1971-2000 period.

Simple approach #4: Analog forecast

Analog: Find a previous atmospheric state that is like the current state and forecast the same evolution. This one does require some more skill because no two situations are EVER exactly alike…

Good example: If a surface low pressure forms in the eastern Gulf of Mexico with a deep upper-level trough to the west, a Nor’ester will roll up the Eastern seaboard—like the 1993 Superstorm

500-mb MAP: 1993 Superstorm SURFACE MAP: 1993 Superstorm

LOW LOW TRACKTRACK

One more simple approach—and perhaps the one meteorologists and climatologists try their hardest to beat but are always

asked about…

Simple Approach #5FORECAST FROM SOMEBODY WHO IS NOT A METEOROLOGIST IS BASED ON

WHATEVER THEY PLEASE!

QuickTime™ and a decompressor

are needed to see this picture.

The complicated way to make a forecast is to use a physical and mathematical model of the atmosphere, starting from an observed state at

an initial time.

This is called

Numerical Weather Prediction

(NWP)

Why do Numerical Weather Prediction?

NUMERICAL WEATHER PREDICTION IS ONLY USEFUL IF YOU CAN SHOW IT GIVES A BETTER FORECAST THAN ALL THE SIMPLE WAYS TO MAKE A FORECAST:

PERSISTENCE

TREND

CLIMATOLOGY

ANALOG

ACHY KNEES; THE OLD FARMER’S ALMANAC…etc.

Steps in Numerical Weather Prediction

1. ANALYSIS: Gather the data (from various sources)

2. PREDICTION: Run the NWP model

3. POST-PROCESSING: Display and use products

Courtesy ECMWF

Sparse data over oceans and Southern Hemisphere

Analysis Phase: Surface Data

ASOS: Automated Surface Observing System

Electronic sensors to measure all elements of weather:

TemperaturePressureMoisture Wind speed and directionVisibilityPrecipitation and precipitation type

Located at virtually every major airport.

Many observations you see on a surface map are taken from ASOS.

Analysis Phase: Surface Buoys

Drifter (red) and moored (blue) buoys

Analysis Phase: Weather Balloons

Courtesy ECMWF

Little data over oceans and Southern Hemisphere

Analysis Phase: Aircraft Reports

Analysis Phase: Satellites

Geostationary

Polar Orbit

Geostationary

Fixed over one location at all times over equator

Polar Orbiting

Orbit over the poles covering earth in swaths

The most important source of data for NWP models

Ahrens, Figs. 9.5 & 9.6

Courtesy ECMWF

Geostationary Data Coverage

Courtesy ECMWF

Polar Orbiter Data Coverage

So we get all that data, say about every six hours or so.

Now what?

Objective Analysis

Data must be interpolated to some kind of grid so we can run the numerical weather prediction model—this is called the initial analysis.

For a regional model these are equally spaced points.

Grid spacing = 35 km

Now the “fun” begins—actually running the model to

make a prediction!

But how do NWP models work?

Not a simple answer!!

Structure of atmospheric models

Dynamical Core

Mathematical expressions ofConservation of motion (i.e. Newton’s 2nd law F = ma)Conservation of massConservation of energyConservation of water

These must be discretized to solve on a grid at given time interval, starting from the initial conditions (analysis).

Parameterizations

One dimensional column models which represent processes that cannot be resolved on the grid.

Called the model “physics”—but it is essentially engineering code.

Equations represented in dynamic coreMUST SOLVE AT EVERY GRID POINT!

MASS CONSERVATION

ENERGY CONSERVATION

CONSERVATION OF MOTION

CONSERVATION OF MOISTURE

Why is just doing this REALLY, REALLY HARD?

Have discretized the equations, so they can be solved on a grid.Equations are non-linear.

We haven’t even accounted for parameterizations yet!

(Pielke 2002)



Physical Processes in Models

Stuff falls between the cracks

Dynamic core

Discretizeddynamical equations

Precipitation processes

Radiation

Land surface energy balance

Boundary layer

Turbulent diffusion

Boundary conditions

Model Grids

“A Lot Happens Inside a Grid Box”(Tom Hamill, CDC/NOAA)

Approximate Size of One Grid Box for NCEP Global Ensemble Model

Note Variability in Elevation, Ground

Cover, Land Use

Source: www.aaccessmaps.co

Rocky Mountains

Denver~50 km Developing

Thunderstorm Cell

13 km Model

TerrainBig mountain ranges,

like the Sierra Nevada, are resolved.

But isolated peaks, like the Catalinas, are not

evident!

100 m contour

By now, it may be be evident that…

To run a numerical weather prediction model you need a big

HUGE number cruncher!

ENIAC One of the first computers

It wasn’t until the development of computers in the 1940s and 1950s that NWP could be even attempted.

Even at that, the very first NWP models were pretty basic (simple dynamical core, no parameterizations)

Hardware unstable: vacuum tubes in the giant computers often blew.

BEFORE THIS TIME, THE METEOROLOGISTS MADE FORECASTS JUST BY READING MAPS AND EXPERIENCE!

NWP’s First Baby Steps: Mid-Twentieth Century

Modern NWP

NCAR SUPERCOMPUTER(Millions of $$)

LINUX PC CLUSTER(Tens of thousands of $$)

Today, NWP models are typically run on supercomputers or networked clusters of PCs.

A Linux PC cluster within the UA Atmospheric Sciences Dept. is used to generate forecasts during the monsoon season and for significant weather events during the winter.

AND YOU NEED TO HAVE EXCELLENT TECH SUPPORT!!

Post-Processing Phase

• Computer then draws maps of projected state to help humans interpret weather forecast

• Observations, analyses and forecasts are disseminated to private and public agencies, such as the local NWS Forecast Office and UA

• Forecasters use the computer maps, along with knowledge of local weather phenomena and model performance, to issue regional forecasts

• News media broadcast these forecasts to public

Weather vs. Climate Forecasts

Weather Forecast

Run NWP model for a period up to two weeks (synoptic timescale)

Objective: Forecast relatively precise weather conditions at a specific time and place

Example: NWP model suggests it will likely rain tomorrow afternoon because mid-latitude cyclone will occur over the U.S.

Climate Forecast

Run NWP model for a period longer than two weeks.

Objective: Forecast probability of deviation from average conditions, or climatology.

Example: In the fall before an El Niño winter, a NWP model forced with warm sea surface temperatures in eastern tropical Pacific projects a circulation pattern favorable for above-average winter precipitation in Arizona.

NOT DESIGNED TO PREDICT EXACT WEATHER FOR SPECIFIC PLACES/TIMES MONTHS IN ADVANCE.

Suite of Official NCEP Forecasts

CPC Predictions Page

WEATHERWEATHERFORECASTSFORECASTS

CLIMATE CLIMATE FORECASTSFORECASTS

Climate Change Projections

http://www.cpc.ncep.noaa.gov/products/predictions/

COMMON PUBLIC MISPERCEPTION

Because the short-term weather forecast is sometimes wrong, why should we even trust climate forecasts, like seasonal forecasts or global warming

projections? LOGICAL FALLACY: The purpose of the climate

forecast is confused with that of the weather forecast.

A COMMON ARGUMENT MADE BY THE UNINFORMED

DON’T FALL VICTIM TO IT!!

NWP model types to generate weather and climate forecasts

General Circulation Model

Vs.

Limited Area Model

General Circulation Model (GCM)

NWP model run over the entire globe

Utility:

Forecast the evolution of large-scale features, like ridges and troughs.

Use to generate long-range weather forecasts (beyond three days), climate forecasts and climate change projections.

Disadvantage:

Can’t get the local details right because of course resolution and model physics.

NCEP Global Forecast System (GFS) Model

Grid spacing = ~100 km

Limited Area Model (LAM)

NWP model run over a specific region

Utility:

Very good for short-term weather forecasting (up to 3 days)

Provides high enough spatial resolution for a detailed local forecast (like thunderstorms in AZ).

May also be useful for climate forecasting.

Disadvantage:

Dependent on a larger-scale model (GCM) for information on its lateral boundaries.

Weather Research and Forecasting (WRF) Model

Forecast Surface TemperatureGCM vs. LAM

General Circulation Model Limited Area Model

Different Models, Different Forecasts!

Why different?

Due to all of the various components of the specific modeling systemAnalysis schemeModel dynamical core + parameterizations

So should we just let the computer do all the job of forecasting?

NO! The meteorologist DOES add value and can play a VERY important role in improving forecasts

IF he/she keeps to appropriate applications!

Value Added of the Meteorologist

(Agudo and Burt)

Knowledge of local weather and climate

Experience

Can correct for model biases

Knows how the model works and realizes it isn’t just a black box!

MOST IMPORTANT:

ISSUE WATCHES AND WARNINGS WHEN SEVERE WEATHER THREATENS PUBLIC SAFETY.

So why do forecasts go wrong?

Think about ALL the possible caveats we’ve already discussed:

Model sensitivity

Inadequate data to specify the initial state (analysis)

Unresolved scaled scales and physical processes

Still is a lot about processes in weather and climate we don’t understand

An inexperienced meteorologist

EVEN IF WE COULD “FIX” ALL OF THE ABOVE, IT WOULD STILL BE IMPOSSIBLE TO MAKE SKILLFUL AND ACCURATE WEATHER FORECASTS USING A NUMERICAL MODEL BEYOND ABOUT TWO WEEKS.

Chaos: A Fixed Limit to Weather Forecasting—Independent of the specific model

Chaos: System exhibits erratic behavior in that small errors in the specification of the initial state lead to unpredictable changes sometime in the future.

In NWP, there will ALWAYS uncertainty in the specification of the initial state—no way around it!

Bottom line: After about two weeks, can’t rely on NWP to provide an accurate and skillful weather forecast.

Sometimes called the “butterfly effect” Dr. Ed LorenzProfessor, MIT

First one to describe chaos

Beyond the two week limit, any forecast with a NWP model is a climate forecast because it has lost the

sensitivity to the initial state.

Why is there STILL is value in the climate forecast?

These can project the probability of departure from average conditions due to factors that vary on a long-time scale

Examples of long term forcing: ocean temperatures, soil moisture, increase in CO2

CPC Winter Climate Forecast vs. Obs.Temperature forecast Precipitation forecast

Observed temperature anomalies

Why was this 2010 forecast a

bust in Arizona?









Observed precipitation anomalies

2009-10 Winter Forecast Reflected El Niño



Key Concepts

• Forecasts are needed by many users• There are several types of forecasts

Simple, no-cost approaches

Numerical Weather Prediction (NWP)

-Analysis Phase

-Prediction Phase

-Post-Processing Phase

Key Concepts

• NCEP issues forecasts out to a season.• Human forecasters can improve NWP forecasts• NWP forecast go awry for several reasons:

measurement and analysis errorsinsufficient model resolutionincomplete understanding of physicschaotic behavior and predictability

• Chaos always limits forecast skill

Assignment for Next Thursday Thunderstorms

• Reading - Ahrens3rd: pg 253-267, 267-273, 273-285

4th: pg 257-271, 271-276, 277-290

5th: pg 263-276, 277-283, 284-296

• Homework09 – D2L (Monday Apr 19)3rd-Pg 286: 10.1, 3, 4, 5, 6, 7, ??, 15

4th-Pg 290: 10.1, 3, 4, 5, 6, 7, 9, 15

5th-Pg 296: 10.1, 3, 4, 5, 9, 10, 11, 16

Assignment for Next Tuesday

• Exam #3 -

D2L• Total Time Allowed to Finish Exam -

75 minutes• Time Interval to Start and Complete Exam -

7:00 am -12:00 pm• Individual Exam -

Not to be taken with others

Documents

NATS 101 Lecture 19 Weather Forecasting Keep Clickers Handy 108 h ensemble forecast valid 1200 utc 20100401