air quality forecasting in canada: current status ... · pdf fileair quality forecasting in...

Air Quality Forecasting in Canada: Current Status, Performance, and

Future Developments

WWOSC/IWAQFR 2014, MontréalDidier Davignon & colleagues,Meteorological Service of Canada,Environment Canada2014-08-19

– 31 octobre 2014

Summary

• Canadian context• RAQDPS overview

– GEM-MACH– Emissions & boundary conditions– Operational implementation– Statistical model– Products– Observation database & model evaluation– Objective analysis– Forest fire modelling

• Possible future directions

– 31 octobre 2014

Canadian context

• Air quality (AQ) involves many jurisdictions• Environment Canada (EC) assuming leadership on AQ

science– Developed expertise on AQ modelling– Building on existing weather forecast systems– Provide a nation-wide unified set of products

• AQ forecast as an health tool– Communicate to the public through the Air Quality Health Index– Issue warnings in collaboration with provinces– Program developed with Health Canada

– 31 octobre 2014

Who’s involved in Canadian AQ forecast

• Surface observations: EC S&T, provinces & large cities• Collecting real-time observations: EC Met Service (MSC)• Developing weather & AQ models: EC S&T• Designing & operating AQ forecast systems: EC MSC• Issuing forecast & warnings: MSC forecasting offices• Publishing public forecast & warnings: EC

communications, provinces• Developing Air Quality Health Index (AQHI): Health

Canada & health partners• Public health actions/decisions: public health partners

– 31 octobre 2014

RAQDPS: components

RAQDPS = Regional Air QualityDeterministic Prediction System

– 31 octobre 2014

RAQDPS: GEM-MACH

– 31 octobre 2014

GEM-MACH v1

• Was developed as a component of the Canadian weather model GEM

• Semi-lagrangian advection, semi-implicit formulation• Rotated lat-long grids, global or regional• Cascading possible• Verified world-class performance at global & regional scales

• Implementation of chemistry Air quality fully online, potential for 2-way coupling Chemistry solved by column Oxidant (ADOM-II) and aerosol chemistry, aqueous &

heterogeneous chemistry Vertical diffusion, wet & dry deposition Sectional representation of PM size distribution with 8 chemical PM

components

– 31 octobre 2014

GEM-MACH, operational version (v1.5.1)

GEM-MACH10 Grid

GEM-LAM10 Grid• Regional system with a domain covering North America

Subdomain of the regional weather model Inherits assimilated weather 10 km horizontal grid spacing 80 vertical levels with lid at 0.1 hPa runs twice daily (00z, 12z) 5 minutes time step for weather

(and tracer advection) 15 minutes time step for solving chemistry One-way coupling (meteorology affects chemistry) 2-bin sectional representation of PM size distribution

(i.e., 0-2.5 μm and 2.5-10 μm) with 8 chemical PM components

– 31 octobre 2014

RAQDPS: Emissions & input

– 31 octobre 2014

Emissions

• Processing emission inventories with SMOKE– Canada 2006, to be updated to 2010– USA EPA 2005 v4.2 projected to 2012– Processing area sources, point sources, mobile sources

▪ More than 10 000 major points, processed individually in the model– Provide hourly speciated profiles for a typical weak, each month

• Biogenic emissions– Four emission factors: NO, isoprene, monoterpenes & other

VOCs – Using BEIS system with BELD3 vegetation database (231

categories), + Canadian National Forest Inventory– Adjust emissions rates online according to meteorology

▪ Solar radiation, cloud cover, 10m temperature

– 31 octobre 2014

Boundary conditions

• Using previous 12h forecast as initial AQ conditions• Using the operational weather analysis as initial weather

conditions• Weather piloting from the operational weather runs

(which is on a larger domain)• AQ piloting: using a climatology at the boundaries

– Vary according to month of the year

– 31 octobre 2014

RAQDPS: statistical model

– 31 octobre 2014

Statistical model: UMOS-AQ

• Post-processing applied to GEM-MACH raw model output

• Corrects model bias (at point locations) through multi-variate linear regression approach

– Applied to meteorological variables since 2000– Adapted for air quality variables (O3, NO2, PM2.5) in 2010– Predictors:

▪ Meteorological and chemical variables from GEM-MACH▪ Persistence (observations at 00Z or 12Z, depending on model run)

– Equations are recalculated four times a month– Has two seasons (summer/winter) with a transitional period of

approximately six weeks

– 31 octobre 2014

Statistical model: UMOS-AQ

• Strengths:– Consistently removes model bias for improved day-to-day

predictability– Provides point forecast

• Weakness:– Method is dependent upon having a large statistical database of

past events, so performance during rarely occurring, bad air quality episodes can be poor

– Using linear techniques only ▪ Non-linear approaches are currently under investigation

– Equations must be recalculated for every model change ▪ Difficult to use in model development

– –

GEM-MACH/UMOS PerformanceGEM-MACHUMOS-AQ

BIA

SR

MS

E

O3

PM2.5

NO2

– 31 octobre 2014

GEM-MACH/UMOS PerformanceGEM-MACHUMOS

BIA

SR

MS

E

O3

PM2.5

NO2

– 31 octobre 2014

RAQDPS: products

– 31 octobre 2014

Products: Air Quality Health Index

Used for public forecast- Multi-pollutant index- Triggers warnings

– 31 octobre 2014

Products: forecasters tools

• Air quality forecast is prepared by site. • Forecasters examine time series of

– Recent observations– Hourly forecasted for the 3 AQHI pollutant (O3, PM2.5, NO2), with

a 3h running average▪ From UMOS-AQ

– Resulting AQHI

• Additionnal products are made available to forecasters– Internal website with all monitoring sites observations &

forecasts ▪ Plus output from parallel runs (when available)

– Allow investigation of special situations

– 31 octobre 2014

AQHI Forecaster Resource Site

– 31 octobre 2014

Other products

• Maps of model output for ozone, PM2.5 at the surface, 50m and 500m are published on our public website.

– Raw model output to be used with care

• Hourly objective analysis for O3, PM2.5 (not public yet)• Internal website with AQHI including the impact of forest

fires• Animations of forest fire forecasted plumes over North-

America

– 31 octobre 2014

RAQDPS: verification

– 31 octobre 2014

VAQUM: Verification for Air Quality Models

• Designed a PostGIS database to store AQ observations and corresponding model outputs

– Can ingest both real time and QC’ed historical datasets– Allows to produce various statistics & categorical scores– About 1730 stations (265 CAN, 1465 USA)– Maximises the use of metadata– Collecting data since 2007

• Essential tool to assess the impact of model updates• Monitoring the performance of the operational system

– 31 octobre 2014

RAQDPS: objective analysis

– 31 octobre 2014

OA: Objective Analysis for Surface Pollutants

• Operational as of February 2013 • Blends model forecasts with surface observations from Canadian

regional networks and the U.S. EPA/AIRNow observation network– Using an optimal interpolation approach– Knowledge of the errors of model and observation data is applied to

weight each input accordingly• Products available hourly (2x = early and late analyses):

– O3 and PM2.5 (NO2 under development)– Possibility for AQHI maps derived from individual analyses for O3,

PM2.5, and NO2

• Analyses are not yet used to initialize GEM-MACH– Tests have been made, applying a correlation factor to spread

information at the surface into the vertical dimension– Results show an improvement in the short-term forecast

– 31 octobre 2014

GEM-MACH

Analysis Increments

Observations

Objective Analysis

Ozone

– 31 octobre 2014

RAQDPS: Forest Fires

– 31 octobre 2014

Forest Fire modelling

• Now possible thanks to:– Near real-time hotspot data available for all of North-America– Partners from the Canadian & US Forest Services provide fire

activity and characteristics in near real-time (NRCan)– BlueSky initiative provides tools to compute emissions from the

above data

• Large interest from emergency management partners• Mostly developed with the objective of improving the air

quality forecast– Run as a parallel product

– 31 octobre 2014

Sample Output (2013-07-02 run)PM2.5 animations : FireWork - Operational

– 31 octobre 2014

Possible Future Directions

• Change how we communicate AQHI?– Now publishing AQHI only for communities where monitoring is

available for all 3 pollutants▪ Forecast done only where AQHI can be observed

– Provide the analysed AQHI everywhere else?▪ Not all partners may want this

– Provide a forecasted AQHI across the domain?▪ Need a sound approach to 2D statistical correction

– 31 octobre 2014


• Benefits of activating feedbacks to meteorology?– Higher under special situations, e.g. forest fires– Need to better parametrise aerosol direct & indirect effects– SCI-PS129.02 - A study of feedbacks between weather and air-

pollution using GEM-MACH (last Monday, P Makar & W. Gong)

• Moving towards AQ forecast with Forest Fires?– How to deal with predictability of FF emissions– Should we keep contribution of FF to AQ distinct?– Need to secure hotspot remote sensing data provision– Implement more comprehensive plume rise algorithm (Freitas)– Challenges in adapting surface objective analysis / assimilation

systems to FF (defining model errors)

– 31 octobre 2014


• Benefits of higher resolution deterministic AQ systems?

– Have been testing 2.5km windows over Prairies & Southern Ontario

– Benefits of representing finer scale features important to AQ▪ Refined spatial allocation of emissions▪ Lake breeze▪ Urban Heat Island▪ Urban aerosols & feedbacks to radiation/convection/cloud

formation– Chemical parameterisations sensitive to resolution– See presentation by Craig Stroud, Wednesday 5:20pm: SCI-

PS225.03 - Development of an air quality prediction system for the 2015 PanAm Games in Toronto

– 31 octobre 2014


• Ensemble AQ systems?– Allowing probabilistic forecast (chances of exceeding AQ

thresholds)– Challenge of defining balanced perturbations

▪ Weather model processes▪ Emissions▪ Chemical processes

– Computationally expensive.▪ Look for multi-model, multi-agency systems?

– 31 octobre 2014


• AQ data assimilation?– Start with assimilation of surface OA– Presentation by A. Robichaud this morning: SCI-PS159.03 -

Assimilation of surface chemical species observations into the Canadian GEM-MACH model using optimal interpolation

– Move to En-Var assimilation of the whole atmosphere?

• Improve model skill for other pollutants?– Partners expressed interest in SO2, VOCs, PM1.0, HS, etc.– Challenge is validation with scarce observations– Designing intensive measurement campaigns for that purpose

– 31 octobre 2014


• Improve statistical models further?– Move to non-linear multivariate updatable models– Improve capacity to preserve exceptional events– Expand to additional predictors (model species, land-use, etc.)

• Take advantage of high resolution (sub-km) surface weather models

– Provide predictors to statistical model?– Use for downscaling?

▪ Should we provide different forecast for near-road, downtown, residential areas, etc?

– Design an AQ surface model for urban areas?

– 31 octobre 2014

Thank you!

Authors: D. Davignon1, D. Anselmo1, P.-A. Beaulieu1, S. Gilbert1, S. Gravel2, H. Landry1, P. Makar3, S. Menard1, M.D. Moran3, R. Pavlovic1 and C. Stroud 3

1Air Quality Modeling Applications Section, Environment Canada, Dorval, QC, Canada. 2Air Quality Research Division, Environment Canada, Dorval, QC, Canada. 3Air Quality Research Division, Environment Canada, Toronto, ON, Canada.

air quality forecasting in canada: current status ... · pdf fileair quality forecasting in...

Documents