Work Order No. 582-17-74141-26 Contract # 582-15-50417

Project 2017-14 Task 5.2

Final Report

Fire Impact Modeling with CAMx

PREPARED UNDER A CONTRACT FROM THE TEXAS COMMISSION ON ENVIRONMENTAL QUALITY

The preparation of this report was financed through a contract from the State of Texas

through the Texas Commission on Environmental Quality. The content, findings, opinions and conclusions are the work of the author(s) and

do not necessarily represent findings, opinions or conclusions of the TCEQ.

Prepared for: Mark Estes

Texas Commission on Environmental Quality 121 Park 35 Circle MC 164

Austin, TX 78753

Prepared by: Jeremiah Johnson, Gary Wilson,

Michele Jimenez, Tejas Shah, Ross Beardsley, Kurt Richman

and Greg Yarwood Ramboll Environ

773 San Marin Drive, Suite 2115 Novato, California, 94998

www.environcorp.com P-415-899-0700 F-415-899-0707

July 2017

July 2017

i

CONTENTS

1.0 EXECUTIVE SUMMARY ................................................................................................... 1

1.1 Updates from 2016 NRT Ozone Modeling System .......................................................... 1

1.2 Fire Impacts and Model Evaluation ................................................................................. 1

1.3 Sensitivity Analysis Summary .......................................................................................... 2

1.4 Recommendations ........................................................................................................... 3

2.0 BACKGROUND ............................................................................................................... 4

3.0 FIRE IMPACT MODELING SYSTEM ................................................................................... 5

3.1 Modeling Cycle ................................................................................................................ 5

3.2 Modeling Domains .......................................................................................................... 5

3.3 Models, Configurations, and Data ................................................................................... 7

3.3.1 Meteorology ......................................................................................................... 9

3.3.2 CAMx Configuration .............................................................................................. 9 3.4 Sensitivity Tests ............................................................................................................. 11

4.0 MODEL EVALUATION ................................................................................................... 13

4.1 Analysis of Fire Impacts ................................................................................................. 13

4.1.1 Case Study: April 28, 2017 .................................................................................. 13 4.2 WRF Meteorological Model Performance Evaluation................................................... 19

4.3 Operational Evaluation .................................................................................................. 24

4.4 Model Performance Evaluation and Sensitivity Analysis .............................................. 24

4.4.1 Statistics .............................................................................................................. 24

4.4.2 1-Hour Ozone Statistical Evaluation ................................................................... 25

4.4.3 Daily Maximum 8-hour Ozone Statistics ............................................................. 29

4.4.4 MDA8 Ozone Local Increment ............................................................................ 32

4.4.5 No Mexico Anthropogenic Emissions Sensitivity ................................................ 36

4.4.6 POA SSA Sensitivity Simulations ......................................................................... 37 4.5 Overall Assessment ....................................................................................................... 40

4.5.1 Main Findings ...................................................................................................... 40

4.5.2 Sensitivity Analysis Summary .............................................................................. 40

5.0 RECOMMENDATIONS FOR IMPROVEMENTS TO MODELING SYSTEM IN 2018 ................ 42

5.1 CAMx Modeling Improvements for 2018 ...................................................................... 42

July 2017

ii

5.2 Website Improvements ................................................................................................. 43

6.0 REFERENCES ................................................................................................................ 44

TABLES Table 3-1. WRF v3.8 physics options used in the NRT ozone modeling system. ..................... 9

Table 3-2. CAMx v6.40 options used for the FIM system. ..................................................... 10

Table 3-3. CAMx sensitivity simulations utilized in the FIM system. The sensitivity tests in the last two rows were run for Apr 23-30, 2017. ..................................... 12

Table 4-1. Number of MDA8 ozone observations and model concentrations that exceeded 70 ppb threshold for April 26 – May 31, 2017 for Dallas, Houston, San Antonio and Tyler regions. .............................................................. 29

Table 4-2. Number of MDA8 ozone observations and model concentrations that exceeded 70 ppb threshold for April 26 – May 31, 2017 by day across all Texas regions. ................................................................................................... 30

FIGURES

Figure 3-1. WRF and CAMx 36/12/4 km modeling domains as used in the FIM system. .................................................................................................................... 6

Figure 3-2. CAMx Model Layer Structure. TCEQ figure from http://www.tceq.texas.gov/airquality/airmod/rider8/modeling/domain. .............................................................................................................................. 7

Figure 3-3. CAMx flow chart detailing input/output and processing streams for Fire Impact Modeling system. ................................................................................. 8

Figure 4-1. True-color image of fire activity across the Yucatan by MODIS aboard NASA’s Terra satellite on April 25, 2017. Each red hot spot is an area where the thermal bands on the instrument detected temperatures higher than normal. When accompanied by smoke, as in this image, such hot spots mark actively burning fire. Image and description provided by NASA’s MODIS website (https://modis.gsfc.nasa.gov/gallery/individual.php?db_date=2017-05-01). ................................................................................................................... 15

Figure 4-2. Hourly PM2.5 smoke tracer concentrations for April 28, 2017 18:00 CST for the base model. Obtained from the FIM website (www.texasaq.com). ............................................................................................. 15

July 2017

iii

Figure 4-3. Hourly PM2.5 smoke tracer (left), base model ozone (middle) and base – No Fires ozone differences (right) for April 28, 2017 18:00 CST. Images obtained from the FIM website (www.texasaq.com). ............................. 17

Figure 4-4. Observed (black dotted line), base model (blue) and No Fires sensitivity run differences from base model (base – No Fires) ozone time series for April 27-29, 2017 at the Harlingen Teege C1023 (top) and Karnack C85 (bottom) monitors. .................................................................... 18

Figure 4-5. Map of ds3505.0 meteorological monitoring stations used in the WRF meteorological model performance evaluation. .................................................. 19

Figure 4-6. Soccer plots for wind speed (top left), wind direction (top right), temperature (bottom left) and humidity (bottom right) for all Dallas ds3505.0 monitoring stations covering April 26 - May 31, 2017. ......................... 22

Figure 4-7. Soccer plots for wind speed (top left), wind direction (top right), temperature (bottom left) and humidity (bottom right) for all Houston ds3505.0 monitoring stations covering April 26 - May 31, 2017. ......................... 22

Figure 4-8. Soccer plots for wind speed (top left), wind direction (top right), temperature (bottom left) and humidity (bottom right) for all San Antonio ds3505.0 monitoring stations covering April 26 - May 31, 2017. ...................................................................................................................... 23

Figure 4-9. Soccer plots for wind speed (top left), wind direction (top right), temperature (bottom left) and humidity (bottom right) for all Tyler ds3505.0 monitoring stations covering April 26 - May 31, 2017. ......................... 23

Figure 4-10. Number of occurrences of MDA8 ozone concentrations above 70 ppb for the years 2005-2017 for the Dallas-Fort Worth, Houston-Galveston-Brazoria, Tyler-Longview-Marshall, and San Antonio metropolitan regions. Each data point represents the maximum concentration over all monitors in each region. ................................................... 24

Figure 4-11. Dallas model performance statistics for ozone (20 ppb cutoff) by area for the base and No Fires simulations for April 26 - May 31, 2017. ..................... 27

Figure 4-12. Houston model performance statistics for ozone (20 ppb cutoff) by area for the base and No Fires simulations for April 26 - May 31, 2017. ............. 27

Figure 4-13. San Antonio model performance statistics for ozone (20 ppb cutoff) by area for the base and No Fires simulations for April 26 - May 31, 2017. ............. 28

Figure 4-14. Tyler model performance statistics for ozone (20 ppb cutoff) by area for the base and No Fires simulations for April 26 - May 31, 2017. ..................... 28

Figure 4-15. Soccer plots for MDA8 ozone for all CAMS sites in Texas (top), Dallas (middle) and Houston (bottom) covering April 26 - May 31, 2017. ...................... 31

July 2017

iv

Figure 4-16. Map of Dallas-Fort Worth CAMS monitoring locations. The 10 potential background sites have green markers and are labeled. ........................ 32

Figure 4-17. Map of Houston CAMS monitoring locations. The 16 potential background sites have green markers and are labeled. ....................................... 33

Figure 4-18. Quantile-quantile plots for Dallas MDA8 ozone local increment for base (left) and No Fires (right). ............................................................................. 34

Figure 4-19. Quantile-quantile plots for Houston MDA8 ozone local increment for base (left) and No Fires (right). ............................................................................. 34

Figure 4-20. Quantile-quantile plots for May 2016 NRT (left) and Apr 26-May 31, 2017 FIM (right) MDA8 ozone local increment for Dallas (top) and Houston (bottom). ................................................................................................. 35

Figure 4-21. Hourly ozone difference concentrations for base – No Mexico Anthro for April 28, 2017 16:00 CST. ................................................................................. 36

Figure 4-22. Observed (black dotted line), base model (blue) and base – No Mexico Anthro (green) ozone time series for April 26 – May 31, 2017 at the San Antonio Northwest C23 monitor. ................................................................... 37

Figure 4-23. Hourly ozone difference concentrations for POA SSA=0.7 - base (left) and POA SSA=0.9 - base (right) for April 29, 2017 11:00 CST. .............................. 39

Figure 4-24. Observed (black dotted line), base model (blue) and POA sensitivity run differences from base model (SSA=0.7 in green; SSA=0.9 in red) ozone time series for April 26-29, 2017 at the Fort Worth Northwest C13 monitor. .......................................................................................................... 39

Figure 5-1. Map of crude oil production and location of oil and gas fields in Mexico and Gulf of Mexico. .................................................................................. 42

July 2017

v

LIST OF ACRONYMS AND ABBREVIATIONS

4HMDA8 4th highest 8-hour daily maximum ozone BC Black Carbon BCs Boundary Conditions CAMS Continuous Air Monitoring Station CAMx Comprehensive Air quality Model with extensions CMAQ Community Multiscale Air Quality Model CO Carbon Monoxide CST Central Standard Time DFW Dallas-Fort Worth Area EBI Euler Backward Iterative method EC Elemental Carbon EDGAR Emissions Database for Global Atmospheric Research EPA Environmental Protection Agency EPS3 Emissions Processing System version 3 FIM Fire Impact Modeling FINN Fire INventory from NCAR GEOS Goddard Earth Observing System GDAS GFS Data Assimilation System GFS Global Forecasting System HGB Houston-Galveston-Brazoria Area HMS Hazard Mapping System HTAP Hemispheric Transport of Air Pollution ICs Initial Conditions IGBP International Geosphere Biosphere Programme ISD Integrated Surface Data LAI Leaf Area Index LI Local Increment LSM Land Surface Model mb millibars MDA1 daily maximum 1-hour average MDA8 daily maximum 8-hour average MEGAN Model of Emissions of Gases and Aerosols from Nature MODIS Moderate Resolution Imaging Spectroradiometer MOPITT Measurements Of Pollution In The Troposphere MOZART Model for OZone and Related chemical Tracers m meter MM5 Fifth-Generation Penn State/NCAR Mesoscale Model MPI Message Passing Interface MSKF Multi-Scale Kain Fritsch NAM North American Model NASA National Aeronautics and Space Administration NCAR National Center for Atmospheric Research

July 2017

vi

NCDC National Climatic Data Center NCEP National Centers for Environment Prediction NCL NCAR Command Language NDAS NAM Data Assimilation System NMB Normalized Mean Bias NME Normalized Mean Error NAAQS National Ambient Air Quality Standard NO Nitric Oxide NOAA National Oceanic and Atmospheric Administration NOx Oxides of Nitrogen NRT Near Real-Time OC Organic Carbon OMP Open Multi-Processing PBL Planetary Boundary Layer PFT Plant Functional Type PM Particulate Matter PM2.5 Particulate Matter < 2.5 ug m-3 POA Primary Organic Aerosol ppb parts per billion PPM Piecewise Parabolic Method Q-Q Quantile-Quantile RAMS Regional Atmospheric Modeling System RE Ramboll Environ RMSE Root Mean Squared Error RPO Regional Planning Organization RRTMG Rapid Radiative Transfer Model for GCM applications SIP State Implementation Plan (for the ozone NAAQS) SSA Single Scattering Albedo TCEQ Texas Commission on Environmental Quality TOMS Tropospheric Ozone Mapping Spectrometer TUV Tropospheric UltraViolet radiative transfer model ug micrograms US United States WRF Weather Research and Forecast model WSM6 WRF Single-Moment 6-Class Microphysics Scheme YSU Yonsei University WRF planetary boundary layer parameterization

July 2017

1

1.0 EXECUTIVE SUMMARY Ramboll Environ (RE) is assisting the TCEQ by deploying a fire impact modeling (FIM) system that models the ozone and smoke impact of fires detected by satellite observations such as agricultural burning in Mexico and Central America. We tested the FIM system using May 2017 as the test period and delivered model results via a website. The FIM system was developed from a near-real time (NRT) ozone modeling system by adding NRT estimates of fire emissions and tracers of smoke transport. This FIM system tracks both the smoke and ozone impacts of biomass burning events to other areas within the modeling domain.

The FIM system uses the Weather, Research and Forecasting (WRF) meteorological model, CAMx air quality model and emissions data provided by the TCEQ. Model configurations are similar to those used for the TCEQ’s State Implementation Plan (SIP) modeling. This report describes the implementation of the 2016 FIM system that modeled April 26 – May 31, 2017 and our evaluation of system performance. We provide specific recommendations for future modeling system improvements.

Updates from 2016 NRT Ozone Modeling System 1.1In developing the FIM system, Ramboll Environ made the following updates from the 2016 NRT ozone modeling system:

1. Expanded southwards the 36 km domains for both the WRF and Comprehensive Air Quality Model with Extensions (CAMx) domains to encompass all of Mexico and the Yucatan peninsula;

2. Added anthropogenic emissions in the expanded portion of the CAMx 36 km domain from the EDGAR-HTPA version 2 global 0.1 degree inventory;

3. Added NRT estimates of fire emissions throughout the CAMx domain from the Fire INventory of NCAR (FINN);

4. Added a primary particulate matter (PM) smoke tracer to CAMx; 5. Quantified ozone impacts of fires from the difference between two model runs, with and

without fire emissions; 6. Ran FIM system modeling cycles in retrospective mode in order to incorporate weather

analyses (instead of forecasts) and NRT FINN fire emissions for the entire cycle; 7. Updated the website interface to present new information generated by the FIM system

and more easily select any date modeled

In addition, we expanded the 4 km WRF and CAMx domains to match TCEQ’s domains and updated to the latest versions of both models.

Fire Impacts and Model Evaluation 1.2

We performed quantitative and qualitative evaluations of WRF meteorological and CAMx ozone performance. The objective was to evaluate the No Fires simulation with respect to the base model to determine whether ozone performance is substantially impacted or improved by including fire emissions.

July 2017

2

First, we examined the impact of fire emissions on model ozone concentrations. The largest impacts occurred when a dense smoke plume from agricultural burning in Mexcio/Central America drifted over East Texas April 28-29, 2017. The aerosols within the smoke plume reduced sunlight and thereby decreased ozone concentrations within the plume relative to No Fires simulation. Much of East Texas was covered by this smoke plume and ozone impacts within the plume ranged from about 2-4 ppb.

Evaluation of WRF meteorological and CAMx ozone statistical metrics revealed that the FIM system agreed as well or better with observations than the 2016 NRT system.

Next, we examined MDA8 ozone local increment (LI) as a way to measure the model’s ability to estimate the range of ozone observations in a given metropolitan area. We found very little difference for the LI between the base and No Fires runs. Performance for the LI was found to be quite good in Dallas but we found consistent LI underestimates in Houston, although the May 2017 FIM system performed better than the May 2016 NRT system. The FIM system is expected to perform better because it has the benefit of relying on weather analyses rather than forecasts. However, we cannot conclude that this is the reason for the better performance because we did not test this hypothesis explicitly.

Sensitivity Analysis Summary 1.3We designed two sets of sensitivity simulations. The first was designed to quantify ozone impacts of Mexico anthropogenic emissions by taking the difference between the base case and a simulation with Mexican anthropogenic emissions set to zero. We found that for areas away from the immediate vicinity of the U.S./Mexico border, maximum ozone impacts across Texas were around 5-10 ppb, with the peak impacts occurring April 28-29. For this modeling episode, it appears that days with the largest Mexico emissions impacts seem to occur when observed ozone is low to moderate. Modeling an entire ozone season would provide a more complete understanding of Mexico emissions contributions to ozone concentrations in Texas.

The second set of sensitivity simulations investigated the ozone sensitivity to the optical properties of organic aerosols found in smoke plumes. We found that by reducing the single scattering albedo from 0.8 to 0.7, we obtained about a 1 ppb decrease within the smoke plume in East Texas in late April. Similarly, we found about a 1 ppb increase over the same area when increasing the single scattering albedo to 0.9. While the ozone sensitivity to the SSA optical parameter was found to be relatively small, there are other sources of uncertainty regarding the treatment of fires in the model that could be explored in future applications.

The fire impact modeling system can help TCEQ refine its photochemical grid modeling system used for State Implementation Plan (SIP) modeling. The system demonstrates usefulness by identifying potential days when exceptional events may be responsible for ozone exceedances. The FIM system could be extended by adding simulations for additional exceptional events.

July 2017

3

Recommendations 1.4We recommend using the lessons learned in this study to improve 2018 ozone modeling. We provide the following recommendations to improve the usefulness of the modeling system by incorporating:

• Stratospheric ozone intrusions by calculating the difference from a simulation that places a 100 ppb cap on ozone boundary conditions above 100 mb

• Ozone transport from Mexico • Better representation of fires in the model following literature review and/or updates to

the NRT FINN fire emissions if available

In addition to the suggestions above to improve the modeling system, we provide the following recommendations to make the modeling website better:

• Improve display of MDA1 and MDA8 ozone plots by including observation overlays • Add capability to download gridded model/observations as text files for user-defined

date range and geographical area • Add new page for cumulative Q-Q plots for ozone, similar to the zoom-able statistics

charts that include pick lists for region and site

July 2017

4

2.0 BACKGROUND The purpose of this study is to implement and refine the photochemical grid model system used by the TCEQ for the State Implementation Plan (SIP) modeling by adding a new fire impact modeling tool. We evaluate ozone impacts from fires and evaluate model performance statistics to measure the impacts of different model configurations and identify areas for improvement.

The FIM system was adapted from the NRT ozone modeling system. Lessons learned from the sensitivity simulations run in previous NRT projects have aided us in our design of the 2017 base model run configuration, in terms of performance and reliability.

We presented a complete overview of the 2013 project in Johnson et al. (2013). We found that the ozone model performed well when high ozone was observed. A general lack of cloud cover and stagnant conditions from WRF meteorology led to ozone over-predictions when observed ozone was low to moderate. Johnson et al. (2015) provides a summary of the 2014 project. The 2014 modeling improved overall ozone bias and error relative to the 2013 modeling, despite much lower observed ozone overall. In the 2015 project, we found improvement in reducing persistent positive ozone bias and discovered that choice of analysis fields in WRF has a substantial impact on ozone (Johnson et al., 2016a). Finally, the 2016 project found further improvement in reducing positive ozone bias, though background ozone was still consistently overestimated in Houston and, to a lesser extent, Dallas. We constructed an ensemble model by averaging together results from five simulations to investigate whether a forecast ensemble could demonstrate greater skill than the individual simulations that go into the ensemble. Lack of sufficient ensemble member diversity hampered our ability to produce a useful ensemble prediction (Johnson et al., 2016b).

This report describes the various components of the development of the FIM system, and presents an evaluation of fire impacts and other model results.

First, we detail our modeling cycle in Section 3.1, including information about run schedule and data sources used. We then specify our WRF and CAMx configurations in Section 3.3, and describe our sensitivity tests and why they were selected in Section 3.4. Next, we present qualitative and quantitative analyses of model results in Section 4, including relative performance of the base case versus sensitivity tests. Finally in Section 5, we discuss various recommendations as improvements to the 2018 ozone modeling system.

July 2017

5

3.0 FIRE IMPACT MODELING SYSTEM This section describes the components of the FIM system. We detail our modeling cycle, including information about run schedule and data sources used. We then describe our WRF and CAMx configurations, CAMx sensitivity tests and finally, features of the FIM website.

Modeling Cycle 3.1We utilize the modeling system as developed for the 2013-2016 NRT projects, with updates to model fires. Ramboll Environ runs the FIM system for 48 simulation hours (2 full days from midnight to midnight in CST). The term initialization is used because the meteorological simulation is started from initial conditions at this time. Ramboll Environ uses 0.25 degree Global Forecasting System (GFS) data assimilation system (GDAS) analysis data (Ek et al., 2014) as initial conditions for the WRF meteorological model. This GDAS data is also used for boundary conditions and data assimilation. Because the FIM system runs a modeling cycle with at least a 1 day lag, observations and analyses are available to the WRF model for the entire modeling cycle and therefore no GFS forecast data needs to be used. We are not able to utilize the NAM (North American Mesoscale) data assimilation system (NDAS) analysis data because it does not cover our expanded 36 km domain (shown in Figure 3-1) used in the FIM system.

Model images were uploaded to the FIM website as model results were processed. Images for the entire modeling period (April 17 – May 31, 2017) were generated for:

o Hourly ozone, NO, NOx, CO concentrations o Daily maximum 1-hour and 8-hour average ozone concentrations o Hourly 2-m temperature, PBL height, wind speed, wind vectors, incoming solar radiation

Users can select images for any modeling cycle for the base case and all sensitivity simulations.

As developed for the 2016 NRT project, Ramboll Environ delivered zoom-able, interactive statistics charts to the site which use observations from the TCEQ CAMS monitors and monitors in Arkansas, Oklahoma and Louisiana. Normalized Mean Bias (NMB), Normalized Mean Error (NME) and correlation coefficient (r) statistics are available for ozone, NO, NOx, CO, 2-meter temperature, wind speed, wind direction and solar radiation.

Modeling Domains 3.2Figure 3-1 presents the 36/12/4 km WRF and CAMx modeling domains used for the FIM system. We expanded the 36 km WRF and CAMx modeling domains such that the 36 km CAMx domain now includes all of Mexico, the Yucatan Peninsula, Belize and much of Guatemala. The 12/4 km domains are the TCEQ SIP domains, which have been used for other modeling efforts performed by Ramboll Environ. The vertical layer mapping table from lowest 38 layers (43 total) in WRF to 28 layers in CAMx, is presented in Figure 3-2. As with the modeling domains, this layer mapping is from the TCEQ SIP modeling and other recent modeling work performed by Ramboll Environ.

July 2017

6

WRF Domain Range (km) Number of Cells Cell Size (km)

Easting Northing Easting Northing Easting Northing

North America Domain

(-2916,2916) (-3024,3024) 163 169 36 36

South US Domain (-1188,900) (-1800,-144) 175 139 12 12

Texas Domain (-396,468) (-1620,-468) 217 289 4 4

CAMx Domain Range (km) Number of Cells Cell Size (km)

Easting Northing Easting Northing Easting Northing

RPO 36km Domain (-2736,2592) (-2808,1944) 148 132 36 36

Texas 12km Domain (-984,804) (-1632,-312) 149 110 12 12

Texas 4km Domain (-328,436) (-1516,-644) 191 218 4 4

Figure 3-1. WRF and CAMx 36/12/4 km modeling domains as used in the FIM system.

July 2017

7

Figure 3-2. CAMx Model Layer Structure. TCEQ figure from http://www.tceq.texas.gov/airquality/airmod/rider8/modeling/domain.

Models, Configurations, and Data 3.3We present a general overview of the input/output and processing streams for the FIM system in Figure 3-3. A description of the inputs used and configuration of the WRF and CAMx models follows.

July 2017

8

Figure 3-3. CAMx flow chart detailing input/output and processing streams for Fire Impact Modeling system.

July 2017

9

3.3.1 Meteorology We are utilizing WRF v3.9 (released April 2017; Skamarock et al., 2008) for the FIM system, the latest version of the model available at the start of the project. We provide the WRF physics options in Table 3-1. This configuration is similar to that used for the TCEQ SIP modeling. We are using MPI (Message Passing Interface) for our WRF simulations, utilizing 28 cores. Previous experience with WRF guided us to this configuration, as performance gains from either increasing the number of cores or using a hybrid MPI/OMP (Open Multi-Processing) approach were found to be minimal for WRF, in contrast to CAMx.

Table 3-1. WRF v3.8 physics options used in the NRT ozone modeling system. WRF Physics Option Option Selected Notes

Microphysics WRF Single-Moment 6-class (WSM6)

A scheme with ice, snow and graupel processes suitable for high-resolution simulations.

Longwave Radiation RRTMG Rapid Radiative Transfer Model. An accurate scheme using look-up tables for efficiency. Accounts for multiple bands, and microphysics species.

Shortwave Radiation RRTMG Rapid Radiative Transfer Model. An accurate scheme using look-up tables for efficiency. Accounts for multiple bands, and microphysics species.

Surface Layer Physics MM5 similarity Based on Monin-Obukhov with Carslon-Boland viscous sub-layer and standard similarity functions from look-up tables

LSM Noah NCEP/NCAR land surface model with soil temperature and moisture in four layers, fractional snow cover and frozen soil physics.

PBL scheme Yonsei University (YSU)

Non-local-K scheme with explicit entrainment layer and parabolic K profile in unstable mixed layer

Cumulus parameterization

MSKF WRF Multi-Scale Kain-Fritsch (MSKF) cumulus parameterization includes feedback of subgrid cloud information to the radiation schemes.

3.3.2 CAMx Configuration Ramboll Environ selected CAMx version 6.40 (released December 2016; Ramboll Environ 2016) for the FIM system, the latest version of the model at the start of the project. While some significant new features have been added since version 6.30, we don’t expect any substantial impacts with the configuration used for FIM modeling.

Table 3-2 gives the CAMx configuration options that are currently in use. We utilize a hybrid MPI/OMP configuration for CAMx. We determined from model benchmarking that 14 MPI slice x 4 OMP thread setup was the optimum configuration for this application.

July 2017

10

Table 3-2. CAMx v6.40 options used for the FIM system. Science Options Configuration Comments

Model Code CAMx Version 6.40 Released December 2016 Time Zone Central Standard Time (CST) Vertical Layers 28 layers (model top approximately 100

mb) Lowest 21 CAMx layers match lowest 21 WRF layers

Chemistry Gas Phase Chemistry Aerosol Chemistry

CB6r4

None

CB6r4 combines a condensed set of reactions involving ocean-borne inorganic iodine from the CB6r2h full halogen mechanism with the temperature- and pressure-dependent organic nitrate branching ratio introduced in CB6r3. Primary PM smoke tracers from NRT FINN fire emissions are inert

Plume-in-Grid None Turned off; run-time consideration for FIM modeling

Photolysis Rate Adjustment

In-line TUV

Adjust photolysis rates for each grid cell to account for clouds and primary PM smoke tracers. Certain photolysis rates adjusted for temperature and pressure.

Meteorological Processor Subgrid Cloud Diagnosis

WRFCAMx CMAQ-based

Sub-grid clouds diagnosed from WRF grid-resolved thermodynamic properties.

Horizontal and Vertical Transport Eddy Diffusivity Scheme Diffusivity Lower Limit

K-Theory Kz_min = 0.1 m2/s

Vertical diffusivity (Kv) fields patched to enhance mixing:

1. over urban areas in lowest 100 m (OB70 or “Kv100” patch)

2. in areas where convection is present, by extending the daytime PBL Kv profile through capping cloud tops (cloud patch)

Dry Deposition

Wesely (1989)

Utilizes 11 landuse categories and does not use LAI

Numerical schemes Gas Phase Chemistry Solver Horizontal Advection Scheme

Euler Backward Iterative (EBI) Piecewise Parabolic Method (PPM) scheme

As used in TCEQ SIP modeling

July 2017

11

We are using the following CAMx inputs for the FIM system:

• Initial conditions and boundary conditions extracted from NRT MOZART-4/MOPITT chemical forecasts from NCAR (http://www.acom.ucar.edu/acresp/forecast). Chemical forecasts are run each day using MOZART-4, driven by GEOS-5 meteorology and including the standard (100 species) chemical mechanism (Emmons et al., 2010).

• Initial conditions were extracted only for the initialization of the April 17, 2017 modeling cycle; subsequent cycles restarted from previous cycle.

• For the boundary conditions, NRT MOZART-4 chemical forecasts were not available for May 17-18, so chemical forecasts for surrounding days (May 16 and 19) were used to extract boundary conditions for these missing days.

• 2017 day-of-week specific anthropogenic emissions inventory provided by the TCEQ • Month-specific elevated point source emissions provided by the TCEQ • 2010 EDGAR global 0.1 degree emissions based on EPA's HTAP emissions modeling

platform used outside TCEQ 36 km domain. • MEGAN v2.10 biogenic emissions using current WRF modeling cycle meteorology with:

o Isoprene emission factors incorporate aircraft data (EFvA2015) as in (Yu et al., 2015) o All non-isoprene species use EFvE2015 emission factors o PFTv2015 includes corn crop fix for plant functional type o Leaf Area Index (LAI) inputs include urban enhancements in Texas (Kota et al., 2015)

• WRFCAMx v4.6 using YSU Kv methodology • Kv landuse patch up to 100 m and Kv cloud patch applied • O3MAP: 2016 monthly averages from 1 degree TOMS satellite ozone column data • Photolysis rates files generated using O3MAP 2016 monthly averages • Land use / land cover inputs from TCEQ’s HGB SIP modeling database; MODIS IGBP

(International Geosphere Biosphere Programme) land use / land cover used outside TCEQ 36 km

Sensitivity Tests 3.4Table 3-3 describes the sensitivity simulations used in the 2017 FIM system. CAMx sensitivity simulations were run in parallel with the “base case” simulation. We selected the sensitivity tests to determine the impact of fires (No Fires and Primary Organic Aerosol Single Scattering Albedo [POA SSA] runs) and Mexico anthropogenic emissions (hereafter No Mexico Anthro) on Texas ozone.

The No Fires and No Mexico Anthro simulations were run for the entire modeling period of the Base simulation, Apr 17-May 31, 2017. The POA SSA (Primary Organic Aerosol Single Scattering Albedo) simulations were run for Apr 23-30, 2017 in order to capture the impacts from a smoke plume event that occurred Apr 28-29, 2017.

July 2017

12

Table 3-3. CAMx sensitivity simulations utilized in the FIM system. The sensitivity tests in the last two rows were run for Apr 23-30, 2017.

Sensitivity Simulation Description No Fires Run CAMx without ingesting NRT FINN fire emissions No Mexico Anthro Zero-out Mexico emissions from all anthropogenic sectors POA SSA=0.9 Change Single Scattering Albedo (SSA) for Primary Organic Aerosol (POA)

from 0.8 to 0.9 POA SSA=0.7 Change POA SSA from 0.8 to 0.7

We developed a No Fires sensitivity simulation that removes NRT FINN fire emissions. This run was designed so that its ozone concentrations could be subtracted from the base simulation to calculate the ozone impact from wildfire emissions. Next, we developed the No Mexico Anthro sensitivity which removed all anthropogenic emissions from Mexico. Biogenic and wildfire emissions within Mexico were included in this run. For this run, we applied a mask to all model grid cells covering Mexico and removed all anthropogenic emissions within those cells. We anticipate that the current FIM system can be extended in the future to model more exceptional events. One such exceptional event is transport of ozone across international borders as is modeled here. We wanted to quantify how much ozone could be attributed to Mexico and determine if any substantial impacts occurred when observed ozone was high during the entire April 26 – May 31, 2017 modeling period. Finally, we developed and tested two POA SSA sensitivity simulations covering April 23-30, 2017. This period was chosen in order to evaluate the ozone impacts relative to the base case simulation when a smoke plume from agricultural burning in Southern Mexico/Central America impacted much of East Texas on Apr 28, 2017. The base case simulation specified an SSA = 0.8 for POA, while the two POA SSA sensitivity tests used SSA = 0.7 and 0.9, respectively. This simulation was designed to test the sensitivity of ozone to the optical properties of the aerosols present in smoke plumes. Detailed information about this simulation and its results are provided in Section 4.4.6.

July 2017

13

4.0 MODEL EVALUATION This section presents quantitative and qualitative evaluations of WRF meteorological and CAMx ozone performance. The objective was to evaluate the No Fires simulation with respect to the base model to determine if ozone performance is substantially impacted or improved by including fire emissions in the model.

In the sections below, we first provide an analysis of fire impacts on ozone, focusing on a smoke plume event that occurred over East Texas April 28-29, 2017. Next, we provide results from the regional analysis of the base configuration, focusing on quantitative evaluation of the meteorological fields and ozone. Then we provide an evaluation of the daily maximum 8-hour average (MDA8) ozone local increment (LI) analysis in Dallas and Houston. Finally, we present the ozone impacts of the sensitivity simulations.

Analysis of Fire Impacts 4.14.1.1 Case Study: April 28, 2017 On April 28, 2017, a dense smoke plume over East Texas was observed. NOAA’s HMS (Hazard Mapping System) Analysis Team reported smoke in the Gulf and along the coast on April 28 and traced it to seasonal burning in Mexico/Central America1:

Friday, April 28, 2017 DESCRIPTIVE TEXT NARRATIVE FOR SMOKE/DUST OBSERVED IN SATELLITE IMAGERY THROUGH 1745Z April 28, 2017 SMOKE: Gulf of Mexico... This morning a broad area of light to moderately dense smoke from ongoing seasonal burning occurring over southeastern Mexico and Central America covered much of the Gulf of Mexico. The most dense part of the plume was over the western Gulf and stretched from the Bay of Campeche northward to the upper Texas/western Louisiana coast.

1 http://www.ssd.noaa.gov/PS/FIRE/DATA/SMOKE/2017D281844.html

July 2017

14

TCEQ’s Daily Air Quality Forecast also mentioned the presence of smoke and its origin in its 4/28 report:

Friday 04/28/2017 Smoke, mainly from agricultural burning in Mexico and Central America, has spread across much of the eastern two-thirds of the state this morning. Smoke may be heavy at times today, especially in South and Southeast Texas.

Figure 4-1 shows an image of fire activity across the Yucatan peninsula captured by MODIS on April 25, 2017. Actively burning fires are shown by red “hot spots” and accompanied by smoke. These fires continued burning over the next several days/weeks and were ingested by the NRT FINN emissions product and provided to the model.

Figure 4-2 shows CAMx PM2.5 from fire smoke tracers (> 10 ug m-3) originating from Mexico/Central America impacting East Texas and reaching into the Mississippi River Valley on April 28, 2017 at 18:00 CST. We note a second smoke plume associated with the West Mims Fire, located in the Okefenokee National Wildlife Refuge along the Florida/Georgia border.

July 2017

15

Figure 4-1. True-color image of fire activity across the Yucatan by MODIS aboard NASA’s Terra satellite on April 25, 2017. Each red hot spot is an area where the thermal bands on the instrument detected temperatures higher than normal. When accompanied by smoke, as in this image, such hot spots mark actively burning fire. Image and description provided by NASA’s MODIS website (https://modis.gsfc.nasa.gov/gallery/individual.php?db_date=2017-05-01).

Figure 4-2. Hourly PM2.5 smoke tracer concentrations for April 28, 2017 18:00 CST for the base model. Obtained from the FIM website (www.texasaq.com).

July 2017

16

Next, we examine the ozone impacts of this smoke plume on East Texas. Figure 4-3 presents the PM2.5 smoke tracers (left; same as in Figure 4-2), base model ozone (middle) and base – No Fires ozone differences (right) for April 28, 2017 18:00 CST. We note that within the smoke plume over East Texas/Louisiana and reaching into the Mississippi River Valley, ozone concentrations are decreased relative to the No Fires run. The model is adjusting photolysis rates to account for the smoke tracers, which results in slight ozone decreases in East Texas on this day. We also note similar behavior (ozone decreases within smoke plume) for the West Mims Fire, located in the Okefenokee National Wildlife Refuge along the Florida/Georgia border. Further away from the smoke plumes (e.g. Intermountain West, Canada, Upper Midwest and Northeast), we find ozone increases from fire emissions.

July 2017

17

Figure 4-3. Hourly PM2.5 smoke tracer (left), base model ozone (middle) and base – No Fires ozone differences (right) for April 28, 2017 18:00 CST. Images obtained from the FIM website (www.texasaq.com).

Finally, we present hourly ozone time series for the April 27-29, 2017 period at the Harlingen Teege C1023 (top) and Karnack C85 (bottom) monitors in Figure 4-4. The maximum ozone impact at the Harlingen Teege monitor is 3.6 ppb at occurs on April 29 at 8:00 CST. The maximum impact at the Karnack monitor is lower (2.3 ppb) and occurs the previous night, April 28 at 22:00 CST. This is consistent with the direction of the smoke plume, which first reached Northeast Texas in the afternoon/evening of April 28, then meandered to the west early on April 29. These ozone impacts from the fires provided motivation for POA SSA sensitivities presented in section 4.4.6.

July 2017

18

Figure 4-4. Observed (black dotted line), base model (blue) and No Fires sensitivity run differences from base model (base – No Fires) ozone time series for April 27-29, 2017 at the Harlingen Teege C1023 (top) and Karnack C85 (bottom) monitors.

July 2017

19

WRF Meteorological Model Performance Evaluation 4.2We evaluate WRF 2-m temperature, 2-m humidity and 10-m wind speed and direction using the Integrated Surface Data (ISD) data set ds3505.0, archived at the National Climatic Data Center (NCDC). Meteorological data from the TCEQ’s CAMS are not used because some locations are known to have nearby obstructions that bias wind measurements from certain sectors (Johnson et al., 2015) and a systematic evaluation of which CAMS could be used for meteorological model performance evaluation is not available to us. For this report, we examine performance at ds3505.0 monitoring locations in Dallas, Houston-Galveston-Brazoria (HGB), San Antonio and Tyler (Figure 4-5).

Figure 4-5. Map of ds3505.0 meteorological monitoring stations used in the WRF meteorological model performance evaluation.

Emery et al. (2001) derived and proposed a set of daily performance “benchmarks” for typical meteorological model performance. These standards were based upon the evaluation of about 30 MM5 and RAMS meteorological simulations of limited duration (multi-day episodes) in support of air quality modeling applications. These were primarily ozone model applications for cities in the eastern and Midwestern U.S. and Texas that were primarily simple (flat) terrain and simple (stationary high pressure causing stagnation) meteorological conditions. More recently

July 2017

20

these benchmarks have been used in meteorological modeling studies that include areas with complex terrain (McNally, 2009; ENVIRON and Alpine, 2012).

The purpose of the benchmarks is to help characterize how good or poor the results are relative to other model applications run for the U.S. In this section, the meteorological variables are compared to the benchmarks as an initial indication of the WRF performance. These benchmarks include model bias and error in temperature, wind direction and water vapor mixing ratio as well as the wind speed bias and Root Mean Squared Error (RMSE). The benchmarks for each parameter are as follows:

• Temperature Bias: less than or equal to ±0.5 °K; alternative of ≤±2.0 °K for complex conditions.

• Temperature Error: less than or equal to 2.0 °K; alternative of ≤2.5 °K for complex conditions.

• Mixing Ratio Bias: less than or equal to ±0.8 g/kg; alternative of ±1.0 g/kg for complex conditions.

• Mixing Ratio Error: less than or equal to 2.0 g/kg. • Wind Direction Bias: less than or equal to ±10 degrees. • Wind Direction Error: less than or equal to 30 degrees; alternative of 50 degrees

for complex conditions. • Wind Speed Bias: less than or equal to ±0.5 m/s; alternative of ±1.5 m/s for

complex conditions. • Wind Speed RMSE: less than or equal to 2 m/s; alternative of 2.5 m/s for

complex conditions. The equations for bias and error are given below, with the equation for the Root Mean Squared Error (RMSE) similar only being the square of the differences between the prediction and observation and a square root is taken of the entire quantity.

Bias = ( )∑=

−N

iii OP

N 1

1

Error = ∑=

−N

iii OP

N 1

1

The April 26-May 31, 2017 average statistics for wind speed, wind direction, temperature and humidity for all ds3505.0 stations within Dallas are displayed graphically in Figure 4-6 using the so-called “soccerplot” displays. Soccerplots graph monthly average bias versus monthly average error. For wind speed, error is replaced with RMSE. Along with the results are the simple (plotted in black) and complex benchmark results (red). It is desirable to have the symbols lay inside the benchmark outline (i.e., score a goal in the soccer plot analogy).

July 2017

21

We present soccerplot diagrams for Houston, San Antonio and Tyler in Figure 4-7, Figure 4-8 and Figure 4-9, respectively. We find generally good performance with points falling within or close to the simple benchmark goals. Tyler shows the worst performance across the four regions, with poor wind speed (just within the complex benchmarks for RMSE and bias) and wind direction (bias just outside benchmark for simple/complex benchmark) performance. Overall, performance looks similar to that observed for 2016 NRT modeling. This kind of assessment should prove to be more useful when multiple months can be compared over an entire ozone season.

July 2017

22

Figure 4-6. Soccer plots for wind speed (top left), wind direction (top right), temperature (bottom left) and humidity (bottom right) for all Dallas ds3505.0 monitoring stations covering April 26 - May 31, 2017.

Figure 4-7. Soccer plots for wind speed (top left), wind direction (top right), temperature (bottom left) and humidity (bottom right) for all Houston ds3505.0 monitoring stations covering April 26 - May 31, 2017.

July 2017

23

Figure 4-8. Soccer plots for wind speed (top left), wind direction (top right), temperature (bottom left) and humidity (bottom right) for all San Antonio ds3505.0 monitoring stations covering April 26 - May 31, 2017.

Figure 4-9. Soccer plots for wind speed (top left), wind direction (top right), temperature (bottom left) and humidity (bottom right) for all Tyler ds3505.0 monitoring stations covering April 26 - May 31, 2017.

July 2017

24

Operational Evaluation 4.3We present the number of occurrences of MDA8 ozone above 70 ppb for the DFW, HGB, Tyler-Longview-Marshall and San Antonio regions in Figure 4-10. The plot suggests that May 2017 had approximately as many high ozone days as May 2015. The exception is San Antonio, where May 2017 (3 occurrences) was closer to May 2016 (2) than May 2015 (10).

Figure 4-10. Number of occurrences of MDA8 ozone concentrations above 70 ppb for the years 2005-2017 for the Dallas-Fort Worth, Houston-Galveston-Brazoria, Tyler-Longview-Marshall, and San Antonio metropolitan regions. Each data point represents the maximum concentration over all monitors in each region.

Model Performance Evaluation and Sensitivity Analysis 4.4

4.4.1 Statistics The CAMx FIM website has been set up to compute model performance statistics for each CAMx run when observed data are available. Statistical metrics are computed for individual CAMS monitoring sites, major urban areas and the entire CAMS network.

The statistical metrics computed for CAMS monitoring locations are:

a) Normalized Mean Bias (NMB)

o NMB = � (𝑃𝑖−𝑂𝑖)

𝑁𝑖=1

� 𝑂𝑖𝑁𝑖=1

July 2017

25

where Pi and Oi are the predicted and observed values (Oi,Pi) in a data pair and N is the number of observed/modeled data pairs.

b) Normalized Mean Error (NME)

o NME = ∑ |𝑃𝑖−𝑂𝑖|𝑁𝑖=1∑ 𝑂𝑖𝑁𝑖=1

c) Correlation coefficient (r)

r =

Statistical metrics were computed for:

o Hourly ozone, NO, NOx and CO o Hourly temperature, wind speed, wind direction, total solar radiation

A 20 ppb threshold was applied to observed ozone concentrations; thresholds of 1 mph and 10 Watts/m2 were employed for wind speed and solar radiation, respectively. We note that because wind direction is an angular measurement, we replace NMB and NME with mean bias (MB) and mean error (ME), respectively.

We evaluated model performance for the base and No Fires simulations for the following modeling period, April 26 through May 31, 2017. This evaluation includes 2-meter temperature, 10-m wind speed and wind direction, solar radiation, and ozone for both simulations.

4.4.2 1-Hour Ozone Statistical Evaluation We present 1-hour ozone statistics across all Dallas CAMS sites in Figure 4-11. Similar plots for Houston, San Antonio and Tyler are found in Figure 4-12, Figure 4-13 and Figure 4-14, respectively. Similar figures are available on the FIM website with interactive selection of date, region/site and model simulation. We observe the poorest performance (high positive bias) when ozone is relatively low (1-hour peak ozone below about 50 ppb). In general, we note that the No Fires sensitivity simulation is nearly identical to the base simulation. The most substantial departure between the two runs occurs on April 28, when a smoke plume from agricultural burning in Southern Mexico and Central America spread over much of East Texas. On this day, the No Fires run shows substantial ozone overestimates across each of the four regions. The presence of the smoke plume on this day served to reduce the amount of

−

−

−

∑∑

∑∑

∑∑∑

=

=

=

=

=

==

N

i

N

ii

i

N

i

N

ii

i

N

i

N

ii

N

ii

ii

N

OO

N

PP

N

OPOP

1

2

12

1

2

12

1

11

July 2017

26

incoming solar radiation, thereby reducing ozone production via photolysis within the plume which lowered the large positive biases are reduced somewhat.

July 2017

27

Figure 4-11. Dallas model performance statistics for ozone (20 ppb cutoff) by area for the base and No Fires simulations for April 26 - May 31, 2017.

Figure 4-12. Houston model performance statistics for ozone (20 ppb cutoff) by area for the base and No Fires simulations for April 26 - May 31, 2017.

July 2017

28

Figure 4-13. San Antonio model performance statistics for ozone (20 ppb cutoff) by area for the base and No Fires simulations for April 26 - May 31, 2017.

Figure 4-14. Tyler model performance statistics for ozone (20 ppb cutoff) by area for the base and No Fires simulations for April 26 - May 31, 2017.

July 2017

29

4.4.3 Daily Maximum 8-hour Ozone Statistics We present the number of MDA8 ozone observations exceeding 70 ppb for the Dallas, Houston, San Antonio and Tyler regions in Table 4-1. Houston had the most days exceeding 70 ppb (41), followed by Dallas (17), San Antonio (5) and Tyler (none). The table suggests that the base and No Fires simulations underestimated MDA8 ozone (“false negatives”) in all regions. Both models predicted only two (out of 17 observed) occurrences of MDA8 ozone above 70 ppb in Dallas. Of the 41 observed occurrences of MDA8 ozone above 70 ppb in Houston, the base simulation predicted 17 occurrences and the No Fires simulation predicted 18 occurrences. Neither model predicted any occurrences above the threshold (of the 5 observed) for San Antonio.

Table 4-2 (daily results) shows that a substantial number of false negatives (42 of 50 or 84%) occur over just 2 days (May 6-7) for the base simulation. There was only one day (May 14) where the models predicted more occurrences (both runs: 5) than were observed (11).

Table 4-1. Number of MDA8 ozone observations and model concentrations that exceeded 70 ppb threshold for April 26 – May 31, 2017 for Dallas, Houston, San Antonio and Tyler regions.

Dallas Houston San Antonio Tyler Obs 17 41 5 0

Base 2 17 0 0

No Fires 2 18 0 0

July 2017

30

Table 4-2. Number of MDA8 ozone observations and model concentrations that exceeded 70 ppb threshold for April 26 – May 31, 2017 by day across all Texas regions. 5/2 5/5 5/6 5/7 5/12 5/13 5/14 5/15 5/20 5/23 5/30 5/31 Total Obs 3 1 33 11 2 5 5 2 2 1 4 1 70 Base 3 0 2 0 0 1 11 2 0 0 1 0 20 No Fires 3 0 3 0 0 1 11 2 0 0 1 0 21 Next, we present soccer plots for MDA8 ozone (Figure 4-15) for all Texas CAMS monitors (top), Dallas monitors (middle) and Houston monitors (bottom). Overall, we observe good performance – the base model NMB is < +2% and NME is < 16% across all Texas sites. The tight clustering of symbols for the two runs agrees with the analysis of statistics charts.

The Dallas plot (middle plot in Figure 4-15) shows even better performance than all Texas sites, with NMB of -0.4% and NME of 13.2% (No Fires NMB: -0.5%; NME: 13.3%). The Houston plot (bottom plot in Figure 4-15) demonstrates good overall performance (base NMB: +7.8%; NME: 19.8%) with very small differences from the No Fires simulation (NMB: +7.7%; NME: 20.0%). We note that these MDA8 ozone statistics for Houston outperform any single month (May-September) from the TCEQ NRT 2016 project.

July 2017

31

Figure 4-15. Soccer plots for MDA8 ozone for all CAMS sites in Texas (top), Dallas (middle) and Houston (bottom) covering April 26 - May 31, 2017.

July 2017

32

4.4.4 MDA8 Ozone Local Increment In order to determine how well the base and No Fires sensitivity runs estimate the range of ozone observations in a given metropolitan area, we calculate the MDA8 ozone local increment (LI) for observations and model simulations. This approach allows us to classify days into low, medium or high ozone production based on the observed LI, which we can then use in our model comparison. This approach may be a more appropriate way to evaluate ozone model performance in Houston, where very sharp observed ozone gradients may be impossible for the model to replicate.

Figure 4-16 displays a map of all Dallas-Fort Worth CAMS monitoring stations. A similar map for Houston monitoring stations is provided in Figure 4-17. We classify the monitors with green pushpins as potential background sites (meaning that when they are upwind of the urban area they are indicative of background) and calculate the median MDA8 ozone concentration across these monitors for each day. Then we find the difference between this value and the maximum MDA8 ozone concentration across all monitors in the same region. We refer to this difference as the MDA8 ozone LI.

Figure 4-16. Map of Dallas-Fort Worth CAMS monitoring locations. The 10 potential background sites have green markers and are labeled.

July 2017

33

Figure 4-17. Map of Houston CAMS monitoring locations. The 16 potential background sites have green markers and are labeled.

In Figure 4-18Figure 4-18, we present quantile-quantile (Q-Q) plots for Dallas MDA8 ozone local increment for the base (left panel) and No Fires (right panel). For each plot, the x-axis shows the observed LI and the y-axis shows each model run’s LI. We note very good agreement between the base model and observations. As documented previously, we note similar performance between base and No Fires runs.

We present similar Q-Q plots for Houston in Figure 4-19. In contrast to the Dallas plots, we see persistent underestimation throughout the entire range of observed LI. Again, we find very similar performance between the two runs.

Finally, we present Q-Q plots for Dallas (top) and Houston (bottom) MDA8 ozone local increment in Figure 4-20. The left column shows results from the May 2016 NRT modeling system’s base run, while the right column shows results from the Apr 26-May 31, 2017 FIM system’s base run. We note that this is not a direct comparison since we are showing results for two different modeling episodes. However, we do find improved performance over the May 2016 NRT base simulation, especially for Dallas. This kind of comparison may be more appropriate when we can compare results for complete ozone seasons.

July 2017

34

Figure 4-18. Quantile-quantile plots for Dallas MDA8 ozone local increment for base (left) and No Fires (right).

Figure 4-19. Quantile-quantile plots for Houston MDA8 ozone local increment for base (left) and No Fires (right).

July 2017

35

Figure 4-20. Quantile-quantile plots for May 2016 NRT (left) and Apr 26-May 31, 2017 FIM (right) MDA8 ozone local increment for Dallas (top) and Houston (bottom).

July 2017

36

4.4.5 No Mexico Anthropogenic Emissions Sensitivity In this section, we present results for the No Mexico Anthropogenic Emissions (hereafter No Mexico Anthro) sensitivity run. For this run, we applied a mask to all model grid cells covering Mexico and removed all anthropogenic emissions within those cells. We anticipate that the current FIM system can be extended in the future to model more exceptional events. One such exceptional event is transport of ozone across international borders as is modeled here. We wanted to quantify how much ozone could be attributed to Mexico and determine if any substantial impacts occurred when observed ozone was high during the entire April 26 – May 31, 2017 modeling period.

We present ozone differences from the base run (base – No Mexico Anthro) for April 28, 2017 16:00 CST in Figure 4-21. We note that much of Central Texas shows ozone impacts from 5-10 ppb along with 1-2 ppb impacts across large regions of the lower Midwest and Southeastern U.S. Maximum ozone impacts from this run were found on April 28 and 29, which coincides with the time of maximum ozone impacts from Mexico/Central America agricultural fires.

Figure 4-21. Hourly ozone difference concentrations for base – No Mexico Anthro for April 28, 2017 16:00 CST.

Next, we show hourly ozone time series for the April 26 – May 31, 2017 modeling period at San Antonio Northwest C23 monitor in Figure 4-22. Outside of the El Paso region and other CAMS monitoring locations located near the Mexico border, the largest ozone impacts from Mexico emissions are found in the San Antonio region. We find a maximum ozone impact of around 10 ppb on the afternoon of April 28. As noted above, this timing coincides with the smoke plume impact. April 29 shows about a 9 ppb ozone impact and several other days show impacts around or slightly exceeding 5 ppb. For this modeling episode, it appears that days with the

July 2017

37

largest Mexico emissions impacts seem to occur when observed ozone is low to moderate. We note that we are only examining one month of results. Over an entire ozone season, we may be more likely to find days where Mexico emissions contributions were substantial during high ozone periods.

Figure 4-22. Observed (black dotted line), base model (blue) and base – No Mexico Anthro (green) ozone time series for April 26 – May 31, 2017 at the San Antonio Northwest C23 monitor.

4.4.6 POA SSA Sensitivity Simulations In this section, we present the motivation for and results from two sensitivity simulations which changed the single scattering albedo (SSA) aerosol optical parameter for primary organic aerosol (POA). The first simulation reduced the POA SSA from the default value used in the base run, 0.8, to 0.7, while the second increased the parameter to 0.9. The goal of these simulations was to determine the sensitivity of ozone production to this important optical parameter.

SSA is the ratio of scattering efficiency to total extinction efficiency, which is defined as the sum of scattering and absorption efficiencies. SSA is unitless and a value of 1 implies that all particle extinction is due to scattering; conversely, SSA of zero implies that all extinction is due to absorption. SSA is an important optical property in climate applications, because it determines whether aerosols tend to have a warming (absorb solar radiation) or cooling (scatter solar radiation) effect.

July 2017

38

𝑆𝑆𝑆 =𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 𝑠𝑒𝑒𝑠𝑠𝑠𝑠𝑠𝑠𝑐

𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 𝑠𝑒𝑒𝑠𝑠𝑠𝑠𝑠𝑠𝑐 + 𝑠𝑎𝑠𝑎𝑠𝑎𝑠𝑠𝑎𝑠 𝑠𝑒𝑒𝑠𝑠𝑠𝑠𝑠𝑠𝑐

SSA of aerosols produced from biomass burning depends on the fuel type, burning conditions and oxidation level (Pokhrel et al., 2016). For the NRT FINN fire emissions used in this study, the fuel type is determined based on global datasets of land cover classification and vegetation type (Wiedinmyer et al., 2011). These factors are then used to derive aerosol speciation profiles for individual fires. Because the light-absorbing properties of elemental carbon (EC, assumed to equal black carbon, BC) and organic carbon (OC) are so different (EC SSA = 0.25; OC SSA = 0.8; Takemura et al., 2002), the ratio of EC/OC produced by the FINN fire emissions is critically important for characterizing the light absorbing properties of the smoke plume in the model. NRT FINN fire emissions used in this study have relatively low EC/OC ratios (from less than 0.1 to around 0.3). Therefore, the SSA of organic carbon specified in the model should have the most impact on light attenuation and therefore photochemical ozone production within the smoke plume. In CAMx, we track organic mass (OM) as primary organic aerosol (POA), which is different from the OC obtained from the FINN fire emissions. Therefore, we apply an OM/OC factor of 1.7 (Reff et al., 2007) to obtain POA in the model. We also extract other PM2.5 (Fine Other Primary or FPRM in CAMx) emissions from the FINN fire emissions by the following relation:

𝐹𝐹𝐹𝐹 = 𝐹𝐹2.5 − 1.7 ∗ 𝑂𝑂 − 𝐵𝑂

These other PM2.5 emissions account for less than 10% of the total fine (< 2.5 ug m-3) aerosol emissions in the FINN fire product.

We selected April 23-30, 2017 as the modeling period for the POA SSA sensitivity simulations to coincide with the smoke plume that covered much of East Texas April 28-29, 2017. Figure 4-23 shows ozone differences from the base run from the POA SSA=0.7 simulation (left) and POA SSA=0.9 simulation (right) for April 29, 2017 11:00 CST. In general, we find that the POA SSA=0.7 sensitivity results in ozone decreases within the smoke plume and the POA SSA=0.9 sensitivity shows ozone increases within the smoke plume. These results make sense because reducing SSA means that more solar radiation is absorbed within the smoke plume, which reduces the amount of radiation available for photochemical production of ozone. Conversely, increasing SSA will result in less solar radiation being absorbed by the smoke plume, which increases the photochemical production of ozone. In the area where fires are burning (near Mexico/Guatemala border), ozone impacts exceed 6 ppb. Over East Texas, we see broad regions of ozone impacts that are right around 1 ppb collocated with the position of the smoke plume at this time.

Examining the hourly ozone time series at the Fort Worth Northwest C13 monitor in Figure 4-24, we find that the POA SSA=0.7 sensitivity results in about a 1 ppb decrease in ozone when the smoke plume moves over the monitor on the morning of April 29, 2017. Conversely, the POA SSA=0.9 sensitivity shows about a 1 ppb increase in ozone during the same time period. We note that the two time series almost exactly mirror each other throughout the day of April 29. Although the timing of these impacts vary across different monitors in the state (depending

July 2017

39

on when the smoke plume passed over a given monitor), the magnitude of the ozone impacts (about 1 ppb) are remarkably similar across the state.

Figure 4-23. Hourly ozone difference concentrations for POA SSA=0.7 - base (left) and POA SSA=0.9 - base (right) for April 29, 2017 11:00 CST.

Figure 4-24. Observed (black dotted line), base model (blue) and POA sensitivity run differences from base model (SSA=0.7 in green; SSA=0.9 in red) ozone time series for April 26-29, 2017 at the Fort Worth Northwest C13 monitor.

July 2017

40

Overall Assessment 4.54.5.1 Main Findings We performed quantitative and qualitative evaluations of WRF meteorological and CAMx ozone performance. The objective was to evaluate the No Fires simulation with respect to the base model to determine if ozone performance is substantially impacted or improved by including fire emissions in the model.

First, we examined the impact of fire emissions on model ozone concentrations. The largest impacts occurred when a dense smoke plume from agricultural burning in Mexico/Central America drifted over East Texas April 28-29, 2017. The aerosols within the smoke plume reduced sunlight and thereby decreased ozone concentrations within the plume relative to No Fires simulation. Much of East Texas was covered by this smoke plume and ozone impacts within the plume ranged from about 2-4 ppb.

Evaluation of WRF meteorological and CAMx ozone statistical metrics revealed that the FIM system agreed as well or better with observations than the 2016 NRT system.

Next, we examined MDA8 ozone local increment as a way to measure the model’s ability to estimate the range of ozone observations in a given metropolitan area. While we found very little difference between the base and No Fires runs, performance was found to be quite good in Dallas. We found consistent LI underestimates in Houston, but performance is in better agreement with observations compared to a similar LI analysis performed for the May 2016 NRT ozone modeling system.

4.5.2 Sensitivity Analysis Summary We designed two sets of sensitivity simulations. The first was designed to quantify ozone impacts of Mexico anthropogenic emissions and found that for areas away from the immediate vicinity of the U.S./Mexico border, maximum ozone impacts across Texas were around 5-10 ppb, with the peak impacts occurring April 28-29. For this modeling episode, it appears that days with the largest Mexico emissions impacts seem to occur when observed ozone is low to moderate. We note that we are only examining one month of results. Over an entire ozone season, we may be more likely to find days where Mexico emissions contributions were substantial during high ozone periods.

The second set of sensitivity simulations investigated the ozone sensitivity to organic aerosol optical properties. We found that by reducing the single scattering albedo from 0.8 to 0.7, we obtained about a 1 ppb decrease within the smoke plume in East Texas in late April. Similarly, we found about a 1 ppb increase over the same area when increasing the single scattering albedo to 0.9. While the ozone sensitivity to the SSA optical parameter was found to be relatively small, there are other sources of uncertainty regarding the treatment of fires in the model that could be explored in future applications.

July 2017

41

The fire impact modeling system can help TCEQ refine its photochemical grid modeling system used for State Implementation Plan (SIP) modeling. The system demonstrates usefulness by identifying potential days when exceptional events may be responsible for ozone exceedances. The FIM system could be extended by adding simulations for additional exceptional events (discussed in Section 5).

July 2017

42

5.0 RECOMMENDATIONS FOR IMPROVEMENTS TO MODELING SYSTEM IN 2018 We discuss potential ozone modeling capabilities for the 2018 ozone season to model exceptional events other than biomass burning. We also make recommendations for a combination of CAMx modeling improvements as well as website improvements for the 2018 modeling system.

CAMx Modeling Improvements for 2018 5.1We provide the following recommendations to improve the usefulness of the modeling system by incorporating:

• Stratospheric ozone intrusions by calculating the difference from a simulation that places a 100 ppb cap on ozone boundary conditions above 100 mb

• Ozone transport from Mexico • Better representation of fires in the model following literature review and/or updates to

the NRT FINN fire emissions if available • Improving the anthropogenic emission inventory for Mexico including emissions

associated with oil and gas production in the Gulf of Mexico near the Yucatan peninsula (see Figure 5-1). Biomass burning plumes are likely to traverse this region and pick up additional emissions on the way to Texas.

Figure 5-1. Map of crude oil production and location of oil and gas fields in Mexico and Gulf of Mexico.

July 2017

43

Website Improvements 5.2In addition to the suggestions above to improve the modeling system, we provide the following recommendations to make the modeling website better:

• Improve display of MDA1 and MDA8 ozone plots by including observation overlays • Add capability to download gridded model/observations as text files for user-defined

date range and geographical area • Add new page for cumulative Q-Q plots for ozone, similar to the zoom-able statistics

charts that include pick lists for region and site

July 2017

44

6.0 REFERENCES Ek, M., Bill Lapenta, Geoff DiMego, Hendrik Tolman, John Derber, Yuejian Zhu, Vijay

Tallapragada, Mark Iredell, Shrinivas Moorthi, Suru Saha. 2014. The NOAA Operational Numerical Guidance System. 29th Session of the World Weather/Climate Research Programme (WWRP/WCRP) Working Group on Numerical Experimentation (WGNE-29) Bureau of Meteorology, Melbourne, Australia, 10-14 March 2014.

Emery, C., E. Tai, and G. Yarwood, 2001. “Enhanced Meteorological Modeling and Performance Evaluation for Two Texas Ozone Episodes.” Prepared for the Texas Natural Resource Conservation Commission, prepared by ENVIRON.

Emmons, L. K., Walters, S., Hess, P. G., Lamarque, J.-F., Pfister, G. G., Fillmore, D., Granier, C., Guenther, A., Kinnison, D., Laepple, T., Orlando, J., Tie, X., Tyndall, G., Wiedinmyer, C., Baughcum, S. L., and Kloster, S. 2010. Description and evaluation of the Model for Ozone and Related chemical Tracers, version 4 (MOZART-4), Geosci. Model Dev., 3, 43-67, doi:10.5194/gmd-3-43-2010.

ENVIRON and Alpine. 2012. Western Regional Air Partnership (WRAP) West-wide Jump-start Air Quality Modeling Study (WestJumpAQMS) – WRF Application/Evaluation. ENVIRON International Corporation, Novato, California. Alpine Geophysics, LLC. University of North Carolina. February 29. (http://www.wrapair2.org/pdf/WestJumpAQMS_2008_Annual_WRF_Final_Report_February29_2012.pdf).

Guenther, A., Karl, T., Harley, P.,Wiedinmyer, C., Palmer, Geron, C. 2006. Estimates of global terrestrial isoprene emissions using MEGAN (Model of Emissions of Gases and Aerosols from Nature), Atmos. Chem Phys., 6, 3181-3210.

Johnson, J., E. Tai, P. Karamchandani, G. Wilson, and G. Yarwood. 2013. TCEQ Ozone Forecasting System. Prepared for Mark Estes, TCEQ. November.

Johnson, J., G. Wilson, D.J. Rasmussen, G. Yarwood. 2015. “Daily Near Real-Time Ozone Modeling for Texas.” Prepared for Texas Commission on Environmental Quality, Austin, TX. January.

Johnson, J., G. Wilson, D.J. Rasmussen, G. Yarwood. 2016a. “Daily Near Real-Time Ozone Modeling for Texas.” Prepared for Texas Commission on Environmental Quality, Austin, TX. January.

Johnson, J., G. Wilson, A. Wentland, W. -C. Hsieh, and G. Yarwood. 2016b. “Daily Near Real-Time Ozone Modeling for Texas.” Prepared for Texas Commission on Environmental Quality (TCEQ), TX. December.

Kota, S. H., G. Schade, M. Estes, D. Boyer, Q. Ying. 2015. Evaluation of MEGAN predicted biogenic isoprene emissions at urban locations in Southeast Texas, Atmos. Environ., 110, 54-64. doi:10.1016/j.atmosenv.2015.03.027.

McNally, D. E., 2009. “12km MM5 Performance Goals.” Presentation to the Ad-Hoc Meteorology Group. 25-June. (http://www.epa.gov/scram001/adhoc/mcnally2009.pdf).

July 2017

45

Pokhrel, R., N. L. Wagner, J. M. Langride, D. A. Lack, T. Jayarathne, E. A. Stone, C. E. Stockwell, R. J. Yokelson, S. M. Murphy, 2016. Parameterization of single-scattering albedo (SSA) and absorption Ångström exponent (AAE) with EC / OC for aerosol emissions from biomass burning. Atmos. Chem. Phys., 16 (2016), pp. 9549-9561.

Ramboll Environ. 2016. User’s Guide to The Comprehensive Air quality Model with extensions version 6.40. ENVIRON International Corporation, Novato, CA. December. Available at www.camx.com.

Reff, A., B.J. Turpin, J.H. Offenberg, J. H., C. P Weisel, J. Zhang, M. Morandi, T. Stock, S. Colome and A. Winer, 2007. A functional group characterization of organic PM2.5 exposure: Results from the RIOPA study, Atmos. Environ., 41, 4585–4598, doi:10.1016/j.atmosenv.2007.03.054.

Skamarock, W.C., J.B. Klemp, J. Dudhia, D.O. Gill, D.M. Barker, M.G. Duda, X-Y Huang, W. Wang, J.G. Powers. 2008. “A Description of the Advanced Research WRF Version 3.” NCAR Technical Note, NCAR/TN-45+STR (June 2008). http://www.mmm.ucar.edu/wrf/users/.

Takemura, T., T. Nakajima, O. Dubovik, B.N. Holben, S. Kinne, 2002. Single scattering albedo and radiative forcing of various aerosol species with a global three dimensional model, J. Clim., 15 (2002), pp. 333-352.

Wiedinmyer, C., Akagi, S. K., Yokelson, R. J., Emmons, L. K., Al-Saadi, J. A., Orlando, J. J., and Soja, A. J.: The Fire INventory from NCAR (FINN): a high resolution global model to estimate the emissions from open burning, Geosci. Model Dev., 4, 625-641, doi:10.5194/gmd-4-625-2011, 2011.

Wesely, M. L.: Parameterization of surface resistances to gaseous dry deposition in regional-scale numerical models, Atmos. Environ., 23, 1293–1304, 1989.

Yantosca, B., Long, M., Payer, M., Cooper, M. 2013. GEOS-Chem v9-01-03 Online User’s Guide, http://acmg.seas.harvard.edu/geos/doc/man/.

Yu, H., A. Guenther, C. Warneke, J. de Gouw, S. Kemball-Cook, J. Jung, J. Johnson, Z. Liu, and G. Yarwood, 2015. “Improved Land Cover and Emission Factor Inputs for Estimating Biogenic Isoprene and Monoterpene Emissions for Texas Air Quality Simulations”. Prepared for Texas Air Quality Research Program. (AQRP Project 14-016, September 2015).