Biogenic volatile organic compounds (BVOC) participate in reactions that can lead to secondarily formed ozone and particulate matter (PM) impacting air quality and climate. BVOC emissions are important inputs to chemical transport models applied on local to global scales but considerable uncertainty remains in the representation of canopy parameterizations and emission algorithms from different vegetation species. The Biogenic Emission Inventory System (BEIS) has been used to support both scientific and regulatory model assessments for ozone and PM. Here we describe a new version of BEIS which includes updated input vegetation data and canopy model formulation for estimating leaf temperature and vegetation data on estimated BVOC. The Biogenic Emission Landuse Database (BELD) was revised to incorporate land use data from the Moderate Resolution Imaging Spectroradiometer (MODIS) land product and 2006 National Land Cover Database (NLCD) land coverage. Vegetation species data are based on the US Forest Service (USFS) Forest Inventory and Analysis (FIA) version 5.1 for 2002–2013 and US Department of Agriculture (USDA) 2007 census of agriculture data. This update results in generally higher BVOC emissions throughout California compared with the previous version of BEIS. Baseline and updated BVOC emission estimates are used in Community Multiscale Air Quality (CMAQ) Model simulations with 4 km grid resolution and evaluated with measurements of isoprene and monoterpenes taken during multiple field campaigns in northern California. The updated canopy model coupled with improved land use and vegetation representation resulted in better agreement between CMAQ isoprene and monoterpene estimates compared with these observations.
Volatile organic compounds (VOCs) are known to contribute to ozone (O
Isoprene, a highly reactive BVOC, contributes to O
BVOC emissions are highly variable among different types of vegetation,
therefore the representation of vegetative coverage is critically important
for accurate spatial distribution of emissions. Northern California has a
large gradient in high-isoprene-emitting vegetation extending from the
Sacramento valley eastward toward the Sierra Nevada (Dreyfus et al.,
2002; Karl et al., 2013; Misztal et al., 2014). Many counties in California
have been designated as “nonattainment” areas for both the 8 h O
In this paper, BVOC emissions estimated with the existing, version 3.14 (Schwede et al., 2005), and updated version of BEIS, version 3.61, are input to the Community Multiscale Air Quality (CMAQ) photochemical transport model (Hutzell et al., 2012; Byun and Schere, 2006; Foley et al., 2010) and estimated BVOC ambient concentrations are compared to surface observations at these field campaigns in central and northern California. Canopy coverage and vegetation species data have been updated with the United States Forest Service Forest Inventory and Analysis (FIA) version 5.1 database and 2006 United States Geological Survey National Land Cover Database (NLCD) using more spatially explicit techniques for tree species allocation. BEIS 3.61 has been updated with new a canopy model of leaf temperature for emissions' estimation. Canopy leaf temperature estimates are also compared with infrared skin temperature measurements over a grass canopy made at Duke Forest. BVOC estimates from the Model of Emissions of Gases and Aerosols from Nature (MEGAN) (Guenther et al., 2012) are also input to CMAQ and model predictions are compared with field study measurements to provide additional context for BEIS updates.
BEIS 3.14 used the BELD 3 land use data set based on combined US county-level USDA-USFS Forest Inventory and Analysis (FIA) vegetation speciation circa 1992 information with the 1992 USGS land cover information (Kinnee et al., 1997). A new land cover data set (BELD 4) integrating multiple data sources has been generated at 1 km resolution covering North America. Land use categories are based on the 2001 to 2011 National Land Cover Data set (NLCD), 2002 and 2007 USDA census of agriculture county-level cropping data, and Moderate Resolution Imaging Spectroradiometer (MODIS) satellite products where more detailed data were unavailable.
Fractional tree canopy coverage is based on the 30 m resolution 2001 NLCD
canopy coverage (
Vegetation speciation is based on multiple data sources. Tree species are based on 2002–2013 Forest Inventory and Analysis (FIA) version 5.1 and crop species information is based on 2002 and 2007 USDA census of agriculture data. The FIA includes approximately 250 000 representative plots of species fraction data that are within approximately 75 km of one another in areas identified as forest by the NLCD tree canopy coverage. USDA census of agriculture data is available on a state and county level only and has been used to refine the agricultural classes to the NLCD agricultural land use categories.
FIA version 5.1 location data have been degraded to enhance landowner privacy
in accordance with the Food Security Act of 1985 (O'Connell et al., 2012).
The provided locations are accurate within approximately 1.6 km with most
plots being within 0.8 km of the reported coordinates and have accurate
state and county identification codes (O'Connell et al., 2012). BELD 3 FIA
vegetation specie fractions were aggregated to county level based on national
aboveground biomass estimates for deciduous, pine, juniper, fir, and hemlock
species. In the BELD 4 data set, FIA plot-level forest biomass
(kg ha
The fractional species composition of the NLCD canopy coverage was then
calculated and the FIA 5.1 species were aggregated to the BELD 4 species
(Table S1 and Fig. S1 in the Supplement). The NLCD land cover defines trees
as greater than 5 m tall, forest refers to greater than 20 % canopy
coverage, with deciduous forests having more than 75 % foliage shed in
winter and evergreen forests having more than 75 % of foliage retained in
winter (
MEGAN and BEIS are both used to support regional- to continental-scale O
BEIS version 3.61 estimates emissions for 33 volatile organic compounds, carbon monoxide, and nitric oxide. Table 1 shows the complete list of compounds estimated by BEIS with mapping to contemporary gas-phase chemical mechanisms SAPRC07T and CB6. BEIS estimates isoprene, 14 unique monoterpene compounds, and total sesquiterpenes. In addition, emissions are estimated for 16 other volatile organic compounds and an aggregate group of other unspeciated VOC. All biogenic VOC emissions are a function of leaf temperature while only isoprene, methanol, and MBO are a function of both leaf temperature and photosynthetically activated radiation (PAR). All species emissions have small indirect impacts from PAR via the canopy module.
Species emissions estimated by BEIS and mapping to the SAPRC07T and CB6r2 gas-phase chemical mechanism lumped species.
Inputs to BEIS include normalized emissions for each vegetation species,
gridded vegetation species, temperature, and PAR. Temperature and PAR can be
provided from prognostic meteorological models, such as WRF or other sources,
such as satellite products (Pinker and Laszlo, 1992; Pinker et al., 2002) or
ambient measurements. The BELD 4 database contains vegetation specie
information for 275 different vegetation categories (Table S1). Table 2 shows
emission rates for each emitted compound by aggregated vegetation type to
illustrate variability in emissions. The variability in BEIS emission rates
is greater than MEGAN 2.1 (Guether et al., 2012) due to the more detailed
representation of vegetation species. These vegetation types include 20 MODIS
and 21 NLCD land cover categories, and 20 different types of both irrigated
and non-irrigated crops (40 total). The remaining categories include tree
species, much of which are broadleaf (e.g., oak) and needle leaf (e.g., fir)
species. A gridded file indicating leaf-on dates based on the 2009 modeled meteorologic
bioseasons file, is also provided as input to BEIS. In the future, leaf-out
and leaf-fall dates will be matched with LAI data. Plant genus type LAIs for
summer and winter are estimated following Kinnee et al. (1997). However, it
is unlikely the current simple leaf-on parameterization will impact typical
regulatory assessments since elevated O
Emissions (
For various sensitivity studies presented here, BEIS 3.14 is applied with
BELD 3 vegetation data, WRF temperature, and both WRF and satellite-derived
estimates of PAR. BEIS 3.61 is applied similarly but with BELD 3 and BELD 4
vegetation data to isolate the impact of the updates to the canopy model.
Note that the BEIS BVOC emission factors were the same in these BEIS 3.14
and 3.61 simulations. A gridded 0.5 by 0.5
BEIS 3.61 includes a two-layer canopy model. Layer structure varies with light intensity and solar zenith angle. Both layers of the canopy model include estimates of sunlit and shaded leaf area based on solar zenith angle and light intensity, direct and diffuse solar radiation, and leaf temperature. BEIS 3.14 previously used 2 m temperature to represent canopy temperature for emissions' estimation even though BVOC emission factors are typically based on leaf temperature (Niinemets et al., 2010). The canopy model has been updated to use land surface physics from the WRF model and air-surface exchange algorithms from the CMAQ model to approximate leaf temperature using an energy balance for the sunlit and shaded portion of each canopy layer. Emissions are estimated for sunlit and shaded fractions of the canopy and summed over the two layers for total canopy emissions.
A simple two-big-leaf (sun and shade) temperature model was developed based
on a radiation balance. The leaf radiation balance is solved for both the sun
(Eq. 1) and shaded (Eq. 2) leaf sides in each layer.
The infrared budget is parameterized as
The saturation vapor pressure of the leaf is defined as
where
The
The
Equation (5) is then further simplified:
Equations (1), (3), (5), (7), (8), and (9) are algebraically combined to estimate the
sunlit leaf temperature assuming that
Equations (2), (3), (5), (7), (8), and (9) are combined to estimate the shaded leaf
temperature:
The sunlit leaf area index, LAI
BVOC emission fluxes,
Chemical species are estimated using the Community Multi-scale Air Quality
Model (CMAQ) version 5.0.2 (
The model domain covers central and northern California with 4 km
Stationary point sources are based on 2009 specific emissions where
available and the 2008 National Emission Inventory (NEI) version 2
otherwise. Mobile emissions are interpolated between 2007 and 2011 estimates
provided by the California Air Resources Board (CARB) and allocated
spatially and temporally using the Spare Matrix Operator Kernel Emissions
(SMOKE) model (
Diurnal observed, and MEGAN 2.1 and BEIS 3.61 estimated leaf
temperatures
Between 15 June and 31 July 2009, the BEARPEX study was conducted to study
photochemical reactions and products in areas downwind of urban areas with
large biogenic influences. The study was located at a managed ponderosa pine
plantation in the foothills of the Sierra Nevada (38.90
During June 2010, the CARES study was conducted to study the formation of
organic aerosols and the subsequent impacts on climate. The study was
composed of two surface monitoring sites: T0 and T1. The T0 was located in
Sacramento, CA at the American River College campus (38.65
The sunlight leaf temperature in MEGAN 2.1 and the revised canopy model in
BEIS 3.61 were evaluated against observations taken in 2008 at the Blackwood
Division of the Duke Forest in Orange County, North Carolina, USA
(35.97
Total aboveground forest biomass (Mg ha
The canopy model updates for leaf temperature estimation are evaluated by comparing canopy model output with infrared skin temperature measurements of a grass canopy at the Duke Forest field site in central North Carolina (Fig. 1). BEIS 3.61 canopy model inputs are based on field measurements taken at this location coincident with the skin temperature data collection. The infrared skin temperature measurements do not represent a mean canopy leaf temperature but rather the temperature of the portion of the canopy exposed to the atmosphere. The infrared skin temperature measurement should be warmer than the mean leaf temperature during periods of solar irradiation and cooler during periods of radiative cooling due to the insulating effect of the unexposed portion of the canopy. Only the estimated exposed leaf temperature (Eq. 12) was used in the evaluation to account for this discrepancy between measurements and canopy model output. Figure 1 shows observed and predicted estimates of leaf temperature and difference between leaf and ambient temperature. The average temperature estimated by the BEIS 3.61 canopy model for the top of the canopy compares well with observations (mean bias of 0.3 K and mean error 1.2 K). Top of the canopy leaf temperature estimated by MEGAN 2.1 are comparable to BEIS 3.61 and the observations at the Duke Forest site.
BELD 4 total forest biomass estimates were evaluated against the independent
estimates of Blackard et al. (2008). Blackard et al. (2008) created a
spatially explicit live forest biomass data set for the United States based
on FIA observations mapped to MODIS, 250 m aggregated NLCD, topographic,
and climatic data. Figure 2 shows the BELD 4 and
Blackard et al. (2008) estimates of forest biomass for this model
domain at 4 km resolution. The Blackard et al. (2008) 250 m grid
resolution data set was projected and aggregated to the CMAQ 4 km grid
resolution projection using rgdal and raster libraries in R (Bivand et al.,
2014). The BELD 4 estimates evaluated well against those of Blackard et al. (2008) with a Pearson's correlation coefficient of 0.872 (
BELD 3 spatial allocation of ponderosa pine (
There are currently no continental US or global databases to quantitatively evaluate the fractional tree species data coverage developed here. However, the species range maps of Critchfield and Little (1966) and Little (1971, 1976) can be used for a qualitative evaluation. The tree species that constituted the largest fraction of biomass observations in the FIA database generally fell within the tree species range maps (Fig. 3). Note that the maps represent a binary distribution of the tree species natural range and the BELD 4 estimates represent a gradient of species density. Species that did not constitute a large fraction in FIA observations typically had a much smaller estimated spatial range than indicated by the range maps. This could partially be due to the criteria, e.g., tree height greater than 5 m, etc., for trees carried over from the NLCD classification scheme or due to sparse sampling of these tree species in the FIA database due to the species scarcity. However, these species likely represent a small fraction of the forest coverage in the domain and a small fraction of the domain-wide BVOC emissions. Also, it is possible that tree coverage has changed in California since the 1970s when the trees were surveyed due to urban planning, plantations, fire, forest growth, and climate change. Future iterations of the BELD data set and the evaluation of the BELD data set can likely be improved by incorporating land cover data with more plant-species-specific information, such as the California Gap Analysis Project (Davis et al., 1998).
Biogenic VOC emissions estimated with BEIS using the new canopy model (BEIS 3.61) and updated vegetation data (BELD 4) are shown for the northern California region in Fig. 4. A similar figure of spatial biogenic emissions estimated with BEIS 3.14 and BELD 3 are shown in Fig. 5. In this model domain, isoprene emissions are highest in the foothills of the Sierra Nevada where high-isoprene-emitting vegetation (e.g., oak trees) are located. Monoterpene emissions are highest in the Sierra Nevada Mountains where high-emitting needleleaf trees are located. Sesquiterpene emissions are highest in the Sacramento and San Joaquin valleys where grasses are common. Most other biogenic VOC emissions show similar spatial patterns as isoprene or monoterpenes (Fig. 4).
The fractional coverage of oak (high-isoprene-emitting species) and needle leaf trees (high monoterpene emitting species) are shown using BELD 3 and BELD 4 in Fig. S2. The BELD 4 representation shows a higher intensity of fractional coverage in much of the Sierra Nevada as county-level information is allocated more spatially explicitly than in BELD 3. Smearing out vegetation coverage, as in BELD 3, will lead to lower emission estimates where narrow features, such as the band of oak trees in the western Sierra Nevada foothills, exist and overpredictions in areas that get allocated vegetation that does not exist in that area. Changes in oak and needle leaf fractional coverage between BELD 3 and BELD 4 are notable for both the Cool and Blodgett Forest sites meaning the observation data available at these locations are useful for evaluating the methodology used to generate BELD 4 (Fig. S2).
BEIS 3.61/BELD 4 estimated total emissions (tons) for the modeling period.
The updated leaf canopy module increases biogenic VOC emissions throughout California (Fig. 5). The changes to the vegetation input data show increases and decreases in isoprene and monoterpene emissions related to changing spatial allocation of high emitting vegetation species and changes to leaf area estimates. Sesquiterpene emissions generally decrease due to the changes in land use and vegetation for this region (Fig. 5). The new vegetation allocation approach employed here for BELD 4 provides more detailed sub-county-level representation of emitting species compared to BELD 3 and those changes are reflected in biogenic VOC emission differences.
The most recent publicly available version of BEIS (version 3.14) and BELD 3 vegetation input were used to provide biogenic emissions for a 4 km CMAQ simulation covering northern and central California for the period of time coincident with the 2009 BEARPEX field study. Additional simulations were done to illustrate the impact of updating the leaf canopy module in BEIS 3.61 and also how updating vegetation input data has an effect on biogenic VOC model performance. Model runs were also done using satellite-derived PAR as input to BEIS in addition to WRF-estimated solar radiation. The MEGAN 2.1 model was also run using WRF and satellite estimates of PAR for the same domain and period.
Temperature and solar radiation used for the biogenic emissions models were compared to measurements at these field sites (Sacramento, Cool, and Blodgett Forest) to determine how meteorological inputs may bias model-estimated BVOC. WRF model evaluation against meteorological variables is summarized in Table 3. The WRF model does well at capturing daytime high temperatures at Blodgett Forest and slightly overestimates daily peak PAR. Daytime minimum temperatures at Blodgett Forest are largely overestimated by WRF (Fig. S3). Temperature maximums and minimums are well characterized at Sacramento and Cool (Figs. S4 and S5) and are similar at these sites during the 2009 and 2010 field study periods (Fig. S3). The satellite-estimated PAR underestimates the ground measurements at Blodgett Forest on certain days but does better at capturing daytime peaks than WRF. In general, meteorological model performance at Blodgett Forest and nearby areas in northern California (Fig. S6) should result in overestimated emissions of isoprene and monoterpenes due to model overestimates in PAR and nighttime ambient temperature. While mixing layer depth has been shown to be well represented by WRF for California using the configuration used here (Baker et al., 2013), mixing layer depth was not continuously measured at these field sites so it could not be directly evaluated, meaning that differences between modeled and actual surface layer mixing depth and also differences in local- to regional-scale transport could impact CMAQ estimates of biogenic VOC.
Baseline BEIS 3.14/BELD 3 emissions (tons; left column) and difference between canopy update and baseline BEIS 3.61/BELD 3 (center column) and between the canopy update and land use/vegetation species updates BEIS 3.61/BELD 4 (right column).
Field study measurements of isoprene and monoterpenes taken in 2010 at Sacramento and Cool, and 2009 at Blodgett Forest provide an opportunity to better understand if the changes to BEIS and BELD better reflect the biogenic VOC gradient seen over these sites. Figure 6 shows the observed distribution of isoprene concentrations at Sacramento and Cool from 2010, Blodgett Forest in 2009, and model estimates from 2009 for the baseline CMAQ/BEIS simulation (BEIS 3.14 and BELD 3), canopy model updates (BEIS 3.61), vegetation data updates (BELD 4), and using satellite PAR with all formulation and other input data updates. Measured isoprene concentrations are lowest in Sacramento and highest at Cool where a high density of oak trees exist. The baseline simulation predicts the highest isoprene at Blodgett Forest rather than Cool, but when canopy parameterization updates and vegetation data inputs are used the modeling system captures the gradient in concentration well across these three sites and also the distribution in observations at each site (Fig. 6).
Measured monoterpenes are highest at Blodgett Forest and lowest at Sacramento (Fig. 7). The baseline model captured this gradient but notably overestimated monoterpenes at Cool. When BELD 4 is used as input, the modeling system compares much closer to observations at Cool and begins to slightly underestimate at Blodgett Forest. The use of satellite PAR rather than solar radiation estimated by WRF does little to change model performance of isoprene. Monoterpenes are not directly sensitive to PAR input and change little due to indirect use of PAR in the canopy model.
Distribution of observed and modeled isoprene. Observations at Sacramento and Cool represent June 2010. Observations at Blodgett Forest match the modeled period.
Model evaluation against field campaigns and network observations.
Distribution of observed and modeled monoterpenes. Observations at Sacramento and Cool represent June 2010. Observations at Blodgett Forest match the modeled period.
The MEGAN 2.1 model generally captures the gradient in observations between sites for isoprene and monoterpenes, but predicts much higher isoprene concentrations at each site compared to observations (see Fig. 6). This is consistent with other studies comparing MEGAN 2.1 isoprene flux with measurements in the Sierra Nevada of northern California (Misztal et al., 2014) and also with modeling systems using MEGAN 2.1 isoprene emissions compared with ambient isoprene concentrations in Texas (Kota et al., 2015) and southern Missouri (Carlton and Baker, 2011). The airborne flux measurements of Misztal et al. (2014) are lower than the MEGAN estimates for the northern California modeling domain evaluated here and the MEGAN canopy model behaved similarly to BEIS 3.61 (Fig. 1) indicating that the MEGAN overestimate in isoprene is likely due to the MEGAN 2.1 emission factors in the modeling domain. Using the MEGAN model estimates of monoterpenes resulted in overestimates at Cool and underestimates at Blodgett Forest. Estimates of isoprene using MEGAN improved when using satellite PAR as input rather than WRF solar radiation. This is consistent with similar evaluation in other parts of the United States (Carlton and Baker, 2011). The use of satellite PAR with MEGAN exacerbated monoterpene overestimates at Cool and increased model estimates at Blodgett Forest reducing the model underestimate. First-generation oxidation products of isoprene (methacrolein and methyl vinyl ketones) were also measured at Blodgett Forest in 2009. Model performance is similar to isoprene where BEIS estimates compare favorably with measurements, and MEGAN 2.1 emissions result in notable overestimates (Fig. S7) similar to previous studies (Kota et al., 2015). Methacrolein can further react in the atmosphere to form methacryloyl peroxynitrate (MPAN) which can form methacrylic acid epoxide (MAE) and subsequently secondary organic aerosol including aerosol methylglyceric acid, organic sulfates, and organic nitrates (Worton et al., 2013). CMAQ overestimates MPAN at Blodgett Forest using either biogenic emissions model, but overestimates are greater when using MEGAN. Model performance for isoprene propagates through secondary reactions and could lead to similar over- or underestimates of SOA.
The updated biomass and tree species vegetation characterization in BELD would benefit from additional evaluation for other parts of the conterminous United States. It is critically important to evaluate biogenic emissions models with field experiments designed for biogenic model evaluation or those that provide robust measurements of key biogenic VOC species, such as those used for this assessment. Future work is planned to evaluate BEIS against a larger field study in California designed for biogenic emissions model evaluation (2011 California Airborne BVOC Emission Research in Natural Ecosystem Transects; CABERNET) (Karl et al., 2013; Misztal et al., 2014) and also with a field study done in the southeast United States during the summer of 2013 (Southern Oxidant and Aerosol Study; SOAS). Evaluation of the model in urban areas would be useful although little field data exist for urban areas making this type of assessment difficult.
BEIS 3.61 code is available upon request prior to the public release of
CMAQ v5.1 and available now in SMOKE 3.6.5
(
WRF source code is accessible from
Additional model output, comparison with measurements, and a flowchart of land use data processing are provided in the Supplement.
The authors would like to acknowledge Lara Reynolds, Charles Chang, Allan Beidler, Chris Allen, James Beidler, and Chris Geron; Alex Guenther, Jeong-Hoo Park, and Allen H. Goldstein from the University of California, Berkeley; Berk Knighton and Cody Floerchinger from the University of Montana; Gunnar Shade and Chang Hyoun from Texas A&M University; Thomas Jobson from Washington State University; and David Simpson and Hannah Imhof from Chalmers University for a useful discussion on the canopy model. Although this work was reviewed by EPA and approved for publication, it may not necessarily reflect official Agency policy. Edited by: G. A. Folberth