Developing Forecasting Tools

1
Developing Forecasting Tools

• OVERVIEW
• Acquiring historical data
• Sources
• Data issues
• Understanding the processes
• Climatology
• Weather typing
• Scatter plots
• Case studies

• Creating forecasting tools
• Range of tools
• Strengths/weaknesses
• How to develop a regression equation

2
Air Quality Data Sources

• Air Quality Data
• In-house (local data)
• EPAs AIRS (regional data) www.epa.gov/airs
• AIRNow (regional images) www.epa.gov/airnow
• Special Studies
• TexAQS 2000
• Southern Oxidant Study
• Gulf of Mexico AQ Study
• NARSTO-Northeast
• Others
• Contacts
• NOAA
• DOE

Source www.cgenv.com/Narsto/
3
Meteorological Data Sources

• Historical meteorological data
• Soundings and surface observations
• Regional climate centers wrcc.sage.dri.edu/rcc.ht
  ml
• University web sites weather.uwyo.edu/upperair/sou
  nding.html
• Data and weather maps
• Climate Prediction Center
• National Climatic Data Center lwf.ncdc.noaa.gov/o
  a/ncdc.html
• NOAAs ARL www.arl.noaa.gov/ready/arlplota.html
• Trajectories
• NOAA ARL www.arl.noaa.gov/ready/hysplit4.html

72558 OAX Omaha Observations at 12Z 27 Jan
2002 ---------------------------------------------
-------------------------------- PRES HGHT
TEMP DWPT RELH MIXR DRCT SKNT THTA
THTE THTV hPa m C C
g/kg deg knot K K K
-------------------------------------------------
---------------------------- 1000.0 56

966.0 350 5.2 -5.8 45 2.58
190 10 281.1 288.6 281.6 963.0 375
7.4 -3.6 46 3.06 193 12 283.6
292.5 284.1 954.0 452 8.2 -3.8 42
3.04 204 20 285.2 294.1 285.7 935.8
610 8.3 -4.3 41 2.98 225 35
286.9 295.7 287.4 925.0 705 8.4 -4.6
39 2.95 230 33 287.9 296.7 288.4
902.3 914 15.0 -8.0 20 2.33 240
26 296.8 304.1 297.2
4
Data Issues Sample Size

• At least 3 to 5 years of data
• Be aware of changes in emissions (fuel changes,
  new sources)

Number of days exceeding the federal 8-hr ozone
standard of 0.085 ppm in the Sacramento
Metropolitan Air Quality Management District.
The solid line indicates a five-year moving
average.
5
Data Issues Monitoring Network

• Does the monitoring network capture the peak
  concentrations?
• What types of monitors exist?
• Street, neighborhood, urban
• Background, downwind, rural

6
Data Issues Quality Assurance

• Understanding the quality of the data
• Range checks (maximum and minimum)
• Nearby site comparisons
• Pollutant comparisons (NO NO2 NOx)
• If data types are merged into a database, perform
  spot checks comparing original data to database

7
Data Issues Quality Assurance
Main and Roberts (2000)
Example of problems encountered with AIRS data.
The top figure shows ozone concentrations at a
site that normally experiences 10 times these
concentrations in July. Inspection of the data
showed that either the AIRS units code or decimal
point indicator code was in error. The bottom
plot shows anomalously high ozone concentrations
on the morning of the 11th the high ozone value
appears erroneous. (Level 0, AIRS data)
8
Data Issues Quality Assurance
The surface winds at the circled point are from
the southeast, whereas all other sites show
northerly winds on October 19, 1999, at 0900 LT.
(Dye et al., 2000)
Example of problems encountered with AIRS data.
The figure shows that there were an abnormally
large number of zero ozone concentrations at a
site during a few years possibly indicating
monitor or reporting problems. (Main and
Roberts, 2000).
9
Data Issues Miscellaneous

• Expect several different formats (significant
  effort is spent assembling a data set)
• Seek to standardize units (ex., ppm or µg/m3, m/s
  or mph)
• Carefully examine time standards and conventions
• Time zones (UTC, LST, LDT)
• Validation times for model data
• Time stamp (begin hour, end hour, middle)
• AQ data are generally in local standard or
  daylight begin hour
• Meteorological data are mixed, but UTC is common
• Matching data (00Z Day 2 Afternoon Day 1 USA)
• Continually examine data quality

10
Understanding the Processes

• Some Basic Analyses
• Climatology of the air quality
• Day-of-week distribution
• Diurnal pattern
• Number of exceedances (by day, site, month,
  year)
• Maximum concentrations (duration and location)
• Weather typing
• Scatter plots
• Case studies

11
Climatology
Columbus 8-hr Ozone Exceedances by Day of Week
(1996-2000)
Days
12
Weather Typing

• Weather Typing Relating generalized synoptic
  weather patterns to air quality conditions
• Types of weather patterns
• 500-mb pattern
• Surface pressure pattern
• Determine patterns on all days, not just high
  days
• Pay special attention to ramp-up and cleanout
  days these are usually the most difficult to
  predict

13
Weather Typing Surface Patterns
MacDonald et al., 1998 Comrie and Yarnal, 1992
14
Weather Typing Upper-air Patterns

• Cairo, Egypt PM10 forecasting program
• Weather processes are the same
• Reviewed twice daily (00Z and 12Z) 500-mb weather
  maps from 1999
• Classified the synoptic weather into nine
  categories

15
Weather Typing Upper-air Patterns

• Result several different types of large-scale
  weather patterns can produce PM episodes in Cairo

16
Weather Typing

• Relate weather events to air quality episodes
• Examine relationship among 500-mb heights,
  temperature, ozone

17
Scatter Plots

• Qualitatively and quantitatively relates
  predictor variables with air quality
• Positive and negative relationships
• R2 is the variance explained

• R2 1.0 perfect
• R2 0.67 moderate correlation
• R2 0.21 low correlation

18
Case Studies

• Provide insight into important processes
• How to conduct a case study
• Select a few air quality episodes include
  ramp-up and cleanout days
• Review locations, magnitude, and duration of poor
  air quality
• Review surface and upper-air weather patterns
• Create back trajectories using NOAAs ARL Hysplit
  trajectory model to estimate transport
• Estimate upwind air quality to help quantify
  local versus background contributions (ex.,
  AIRNow maps)
• Review the twice daily temperature soundings to
  help estimate vertical mixing

19
Case Study - Egypt

• Select episodes to study include ramp-up and
  cleanout days

October 23 high PM10
Time series plot of PM10 concentrations in Cairo
in October 1999. The October 20-25, 1999,
episode is outlined (Dye et al., 2000).
20
Case Study - Egypt

• Review surface and upper-air weather patterns

500-mb heights (dm) (solid) and 850-mb
temperatures (oC) (dashed) (left), and sea level
pressure (mb) and surface wind flags (right) on
October 23, 1999, (high PM day) at 1400 LT
21
Case Study - Egypt
Review surface and upper-air weather patterns
500-mb heights (dm) (solid) and 850-mb
temperatures (oC) (dashed) (left), and sea level
pressure (mb) and surface wind flags (right) on
October 25, 1999, (cleanout day) at 1400 LT
22
Case Study - Egypt

• Review trajectories to estimate transport

12-hr trajectories ending in Cairo at 1400 LT
each day and 24-hr resultant winds for October
20 to 25, 1999
23
Case Study - Egypt

• Review upper-air soundings to estimate vertical
  mixing

0200 LT temperature soundings taken at Helwan
from October 20 to 25, 1999
24
Case Study Results

• Episode Characteristics
• Weak upper-level troughs passed over and north of
  Egypt
• The troughs lowered surface pressures to the
  north, thereby weakening the northerly winds, but
  winds still remained light to moderate from the
  north
• Aloft temperatures warmed, the capping inversion
  lowered, and vertical mixing was reduced
• Despite the moderately strong northerly winds,
  the reduced vertical mixing allowed for high PM10
  concentrations
• The episode ended when a strong trough followed
  by a ridge passed over Egypt, increasing the
  northerly winds and weakening the inversion

25
Forecasting Tools

• Forecasting tools can be
• Qualitative hot temperatures ? high ozone
• Quantitative wind speeds PM2.5
• Tools
• Help or aid in decision making processes
• Help forecasters to reach a consensus
• Tools should be used in conjunction with human
  experience

26
Creating Forecasting Tools

• Tool development is a function of
• Amount and quality of data (AQ and
  meteorological)
• Resources for development (human, software,
  computing)
• Resources for operations (human, software,
  computing)
• Types of tools
• Persistence
• Climatology
• Criteria, Thresholds, Rules of thumb
• Regression equations
• Classification and Regression Trees (CART)
• Neural networks
• Fuzzy logic
• Numerical modeling
• Conceptual and experience

Fewer resources, lower accuracy
More resources, better accuracy
27
Example Tool Criteria

• Table of meteorological values associated with
  high ozone episodes
• Simple forecasting tool modest accuracy
• Compare forecasted value to criteria

• Strengths
• Easy to develop
• Quick to run
• Conceptual
• Weaknesses
• Qualitative
• Does not provide concentrations

28
Example Tool CART

• Classification and Regression Trees
• Software develops the decision tree with human
  guidance
• CART splits data sets into similar and dissimilar
  groups

Ozone (LowHigh)
29
Example Tool CART
CART classification PM10 in Santiago, Chile
Is forecasted temperature at 850 mb ? 10.5C ?

No
Yes
Node x Variable and criteria STD
Standard deviation Avg Average PM10
(ug/m3) N number of cases in node
Variables T850 - 12Z 850 MB temp DELTAP - the
pressure difference between the base and top of
the inversion MI0 - Synoptic weather potential
(scale from 1-low to 5-high). FAVGTMP - 24-hour
average temperature at La Paz FAVGRH - 24-hour
average relative humidity at La Paz.
Cassmassi, 1999
30
Example Tool Numerical Modeling

• Numerically model the processes
• Requirements
• Gridded emission inventory
• Meteorological model forecasts
• Other supporting data
• Land use
• Boundary conditions (air quality)
• Photochemical model
• Computer resources
• More sources of uncertainty

• Strengths
• Based on atmospheric and chemistry physics
• Provides forecast in areas without monitors
• Helps further understand processes
• Weaknesses
• High level of expertise and funding needed to
  develop, operate, and improve
• Requires substantial computer resources

31
Example Tool Ozone Modeling

• Research and operational modeling being conducted
  by a number of organizations Environment
  Canada, MCNC, NOAA, Ohio State, Sonoma
  Technology, Inc., SUNY-Albany, Washington
  University

32
Example Tool Ozone Modeling

• 1-hr ozone concentrations in central California
  on September 20, 2001

Forecast
Sonoma Technology, Inc.
33
Creating a Forecasting Tool Regression

• Several steps
• Determine what to predict
• Identify predictor variables
• Perform statistical analysis to select key
  predictor variables
• Develop regression equation
• Verify forecast

34
Creating a Forecasting Tool Regression

• Determine what to predict (predictand)
• 8-hr ozone concentration
• Daily peak (maximum) concentration
• At one site or all sites
• Recommend regional daily maximum
• Identify predictor variables
• Use conceptual understanding
• Create scatter plots of predictand with each
  predictor variable
• Check accessibility of variables selected
• Check physical relationship between each variable
  and ozone concentration

35
Creating a Forecasting Tool Regression

• Perform statistical analyses to identify key
  predictor variables

Cluster Analysis Partition data into similar
and dissimilar subsets
Correlation Analysis Evaluate association
among variables
Step-wise Regression Automate selection of
significant variables
Human Selection Identify a limited number of
variables based on above analyses
36
Creating a Forecasting Tool Regression

• Develop regression equation
• Use statistical software or MS Excel (LINEST
  function)
• Evaluate several equations to strike a balance
  between variable accessibility, equation
  accuracy, and physical meaning
• Example Columbus, next-day equation

O3 exp(2.421 0.024Tmax 0.003Trange -
0.006WS1to6 0.007V925 - 0.004RH -
0.002WS500)
Tmax maximum temperature in ºF Trange daily
temperature range WS1to6 average wind speed
from 12 a.m. to 9 a.m. in knots V925 V
component of the 925-mb wind at 00Z RH relative
humidity at the surface at 00Z WS500 wind speed
at 500 mb at 00Z
37
Summary

• Creating Forecasting Tools
• Data availability and quality
• Develop understanding
• Develop tools
• The more tools the better

• Next Steps
• Verification of forecasts
• Questions