HomeMy WebLinkAboutDAQ-2024-010756FY23 Science for Solutions
Contract #230135
Final Report
Prepared for:
Christopher Pennell
Utah Department of Environmental Quality
195 N. 1950 W.
Salt Lake City, UT 84116
Prepared by:
Ramboll
7250 Redwood Blvd., Suite 105
Novato, California 94945
May 2024
Improved Vegetation Data for the
Biogenic Emission Inventory of the
Wasatch Front
Final Report
Ramboll
7250 Redwood Boulevard
Suite 105
Novato, CA 94945
USA
T +1 415 899 0700
https://ramboll.com
Improved Vegetation Data for the Biogenic Emission
Inventory of the Wasatch Front
Final Report
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
i
Contents
List of Acronyms and Abbreviations iv
Executive Summary 5
Introduction 6
1.1 Background 6
1.2 Overview of Approach 8
1.3 Overview of MEGAN Model 8
1.4 Overview of MEGAN Model Input Data 8
1.5 Overview of Report 8
Leaf Area and Vegetation Cover Distributions for the Wasatch
Front 10
2.1 Quantify LAI and VCF Data 10
2.2 Calibration and Assessment of LAI Data 11
Growth Form Fraction Distribution for the Wasatch Front and
Northern Utah 14
3.1 Ultra-High Resolution (0.5 m) Urban Growth Form Cover Fraction
Distributions 15
3.2 High resolution (10 meter) growth form distributions for urban and
surrounding areas. 19
3.3 Comparison with alternative vegetation cover. 22
Tree Speciation 25
Compile MEGAN Inputs and Assess Emissions 29
Conclusions 33
6.1 Summary of findings 34
References 35
Appendices
Appendix 1. Using SNAP to estimate LAI and VCF using 10-m resolution ESA Sentinel2-MSI data
Appendix 2. Site descriptions
Appendix 3. Methods to estimate LAI from field measurements
Appendix 4. Generating MEGAN growth form fraction distributions using high resolution imagery
(e.g., NAIP)
Appendix 5. Estimating tree cover fraction and compiling random tree locations using i-Tree
Appendix 6. Virtual urban tree survey
Appendix 7. Virtual Urban Tree Survey (VUTS) Wasatch Front Urban Tree Key
Table of Figures
Figure 1-1. A map depicting the Wasatch Front ozone nonattainment area
and its surrounding region. 7
Figure 1-2. Illustration of the three types of landcover data inputs required
for the MEGAN model. Leaf Area Index (LAI) is the total
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
ii
amount of BVOC emission foliar source in a 10 m x 10 m area.
Growth form fractions divide the LAI into four categories that
can be accurately mapped at 10-m spatial resolution. Growth
form speciation quantifies the plant species composition to
assign species-specific emission factors. 7
Figure 1-3. Map of regions with updated MEGAN landcover inputs.
Magenta, green and blue rectangles show regions with updated
growth form fractions, LAI and emission vegetation types,
respectively. The ten new urban emission type regions are
shown in the blue rectangle breakout box on the right. 9
Figure 2-1. Scatterplot of satellite-based and field-based LAI averaged for
117 plots. Solid line indicates 1:1 agreement. Landcover
categories include forest/woods (filled blue circles), grass and
shrub (open black diamonds), and built-up (red open
triangles). 12
Figure 3-1. Inner left blue rectangle (and blue rectangle on the right)
shows urban target region including the extent of the ten
Wasatch Front zones that have been designated as ten urban
MEGAN Emission Vegetation Types developed for this project
(each with a specified tree species composition). Outer left
rectangle shows the rural 3-degree x 3-degree rural region
characterized using iTree. 15
Figure 3-2. ArcGIS base map imagery showing Salt Lake City Hall and
Courthouse and surrounding parks. 16
Figure 3-3. 0.5-m spatial resolution classification of the landcover
distribution surrounding the Salt Lake City Hall and Courthouse
developed for this project. Segmented NAIP natural color
image (top) and classified cover (bottom) showing distributions
of buildings and pavement (red), groves and individual trees
(green), dry grass (brown) and green grass (yellow). Black
areas are associated with shadows. 17
Figure 3-4. Salt Lake City Tree cover distribution (%). 18
Figure 3-5. SLC shrub cover distribution (%). 18
Figure 3-6. SLC grass cover distribution (%). 19
Figure 3-7. 10-m spatial resolution landcover distribution surrounding the
Salt Lake City Hall and Courthouse from the ESA Global
WorldCover data. Landcover categories include built-up (red),
trees (green), grass (yellow) and barren (grey). 20
Figure 3-8. MEGAN Utah grass cover distribution (%). 21
Figure 3-9. Utah shrub cover distribution (%). 21
Figure 3-10. Utah tree cover distribution (%). 22
Figure 3-11. Difference between CGL and SNAP Sentinel 2 LAI (m2 m-2)
estimates. 22
Figure 4-1. MEGAN tree composition used in MEGAN 2.1, 3.1 and 3.2 for
US urban landcover classes that have not been surveyed. 26
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
iii
Figure 4-2. The ratio of new (this study) to old (default MEGAN 3.2) values
for LAI, tree cover, fraction of isoprene emitters and the overall
impact. 28
Figure 5-1. LAI comparison between the updated data developed for this
project and the 2019 MEGANv3.2 vegetation cover database.
Black boundaries delineate the northern and southern regions
of the Wasatch Front. 31
Figure 5-2. July 2017 average isoprene emissions (kg/day) for the Wasatch
Front region simulated using MEGAN3.2 with current and
updated vegetation cover developed in this project. 31
Figure 5-3. July 2017 average terpene emissions (kg/day) for the Wasatch
Front region simulated using MEGAN3.2 with current and
updated vegetation cover developed in this project. 32
Table of Tables
Table 2-1. Comparison of 300 m CGL and 10 meter Sentinel SNAP LAI
estimates for urban zones in Wasatch Front region. 13
Table 3-1. Comparison of tree cover percentage estimates for Wasatch
Front urban zones using i-Tree (i-Tree random point surveys
for 2022), NLCD (NLCD 30 m landcover data for 2016), CGL
(ESA CGL 100 m landcover data for 2015), WC (ESA
Worldcover 10 m landcover data for 2021) and the 0.5-m
landcover developed for this study using 2021 NAIP data as
described in Section 3.1. 24
Table 4-1. Percentage of trees classified as isoprene emitters for ten urban
zones in the Wasatch Front region. Estimates are based on Salt
Lake City tree inventory (limited to trees on public land), the
Virtual Urban Tree Survey (VUTS) approach outlined in
Appendix 6, and data from the 2023 urban field study. The
MEGAN EVT code and corresponding values from the MEGAN
model are provided. 28
Table 5-1. Comparison of default MEGAN 3.2 inputs with those developed
for this study. 29
Table 5-2. Total episode average BVOC emissions (tons/day) for July 2017
estimated in the Wasatch Front region using MEGAN3.2 with
the current and newly updated vegetation cover developed in
this project. 30
Table 6-1. Ratio of LAI, tree fraction, and isoprene emitter fraction
estimated in the current study compared to values in existing
MEGAN input files. Their combined impact on isoprene emission
is estimated as the product of these three values. 33
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
iv
List of Acronyms and Abbreviations
BEIS Biogenic Emission Inventory System
BVOC Biogenic Volatile Organic Compound
CGL Copernicus Global Land
EPA Environmental Protection Agency
ESA European Space Agency
EVT Emission Vegetation Type
GF Growth Form
LAI Leaf Area Index
LAIv Leaf Area Index for the vegetated fraction
MEGAN Model of Emissions of Gases and Aerosols from Nature
NAAQS National Ambient Air Quality Standard
NAIP National Agriculture Imagery Program
NLCD National Land Cover Dataset
NOx Nitrogen Oxides
PAR Photosynthetically Active Radiation
QA/QC Quality Assurance/Quality Control
S2/MSI Sentinel-2 satellite Multispectral Instrument
SL2P Simplified Level 2 Product Prototype Processor
SNAP Sentinel Application Platform
UDAQ Utah Department of Air Quality
VCF Vegetation Cover Fraction
VOC Volatile Organic Compound
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
5
Executive Summary
Isoprene and other biogenic volatile organic compound (BVOC) emissions play a significant role in
atmospheric chemistry in the Wasatch Front region, where they contribute substantially to total VOC
emissions (UDAQ, 2023). Urban areas are the most challenging for BVOC emissions estimation, due to
heterogeneity and a lack of vegetation information, and yet they continue to be the least studied. Two
widely used biogenic emission models, the Model of Emissions of Gases and Aerosols from Nature
(MEGAN) and the Biogenic Emission Inventory System (BEIS), have used generic national average
vegetation characteristic data (e.g., leaf area, tree cover fraction and species composition) to
represent urban landscapes that vary over an order of magnitude which results in uncertainty in the
Wasatch Front region BVOC estimates. Increasingly higher resolution remote sensing data products
have substantially improved the potential for characterizing the landcover inputs required for biogenic
emission models.
This project aimed to address this challenge by improving BVOC emission estimates for urban areas
within the Wasatch Front region. To achieve this, we utilized a novel approach known as the Virtual
Urban Tree Survey, in conjunction with the analysis of high-resolution (<10 meter) satellite imagery
using machine learning, object-based classifications that are calibrated and assessed by field
observations to characterize urban tree cover. Specifically, the project focused on enhancing the
MEGAN model landcover inputs for the region. This included improving parameters such as time-
varying Leaf Area Index (LAI), growth form fractions (tree, shrub, crops, herbaceous plants), and tree
species composition.
The project's findings offer several key benefits. Firstly, they provide an improved MEGAN emission
model tailored to the Wasatch Front region, resulting in more accurate BVOC emission estimates. This
enhanced emission inventory is invaluable to the Utah Department of Air Quality (UDAQ) for informing
air quality simulations and developing regulatory strategies to improve and maintain clean air in the
region.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
6
Introduction
1.1 Background
The Wasatch Front experiences exceedances of the National Ambient Air Quality Standard (NAAQS) for
ozone during summer months when biogenic emissions are prevalent. The Northern Wasatch Front
and Southern Wasatch Front areas are treated separately for nonattainment under the 2015 8-hour
ozone standard1. The Northern Wasatch Front is designated as moderate nonattainment, while the
Southern Wasatch Front is designated as marginal nonattainment. The Northern Wasatch Front
includes all or part of Salt Lake, Davis, Weber, and Tooele counties. The Southern Wasatch Front
includes parts of Utah county. Figure 1-1 displays the Wasatch Front nonattainment area and its
surrounding region.
The Wasatch Front was required to attain the standard by August 3, 2021. The Southern Wasatch
Front nonattainment area attained the standard, prompting the Utah Division of Air Quality (UDAQ) to
initiate the redesignation process to attainment. However, the Northern Wasatch Front nonattainment
area failed to attain the standard by that date and was subsequently bumped up to moderate
classification on November 7, 2022 2. As a result of this designation, the UDAQ was required to
develop a moderate SIP (UDAQ, 2023). Recent monitoring data suggests that the Northern Wasatch
Front may not attain the standard by the moderate attainment date, potentially facing reclassification
as a serious nonattainment area in 2025. Serious nonattainment areas are subject to stricter
emissions reporting/permitting levels and additional control measure requirements.
Accurate quantification and simulation of emissions, particularly of biogenic volatile organic
compounds (BVOCs), are crucial for understanding and managing air quality. BVOCs, emitted from
various natural sources like vegetation, play a significant role in atmospheric chemistry, especially in
the formation of ozone and aerosols. Their emissions are highly variable in both space and time,
posing challenges for estimation. BVOC emissions emitted into urban atmospheres interact with
anthropogenic emissions like nitrogen oxides (NOx) to accelerate ozone formation. Understanding the
contribution of BVOC emissions to ozone formation helps in developing effective strategies to mitigate
air pollution and protect public health, particularly in urban areas.
This project aimed to improve the biogenic emissions inventory by developing a high-resolution urban
landcover database for the Wasatch Front. The MEGAN model relies on vegetation cover inputs such
as Leaf Area Index (LAI), growth form fractions, and growth form speciation, as depicted in Figure 1-
2. While a global database of these inputs is available through the MEGAN website
(https://bai.ess.uci.edu/), the existing data may be suitable for global and regional modeling in rural
areas but lacks precision and certainty in urban areas and other heterogeneous landscapes.
1 EPA Green Book 8-Hour Ozone (2015) Area Information https://www.epa.gov/green-book/green-book-8-hour-
ozone-2015-area-information
2 https://www.epa.gov/air-trends/air-quality-design-values, accessed on 19 January 2024
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
7
Figure 1-1. A map depicting the Wasatch Front ozone nonattainment area and its
surrounding region.
Figure 1-2. Illustration of the three types of landcover data inputs required for the
MEGAN model. Leaf Area Index (LAI) is the total amount of BVOC emission foliar source in a
10 m x 10 m area. Growth form fractions divide the LAI into four categories that can be
accurately mapped at 10-m spatial resolution. Growth form speciation quantifies the plant
species composition to assign species-specific emission factors.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
8
1.2 Overview of Approach
The overall approach was to utilize high-resolution satellite imagery capable of characterizing the
individual trees in an urban landscape that are missing from other landcover datasets. Currently, BEIS
and MEGAN both use a single tree species composition profile for all urban landscapes in the US
except for some cities in Texas and California where we have applied this approach to update the
MEGAN landcover. These high-resolution data were calibrated with field measurements and used to
compile more accurate landcover input datasets for BVOC emission modeling in the Wasatch Front
region. The technical approach is outlined below:
1. Quantify high resolution (8- or 10-day, 10 m) calibrated LAI data for region (approx. 6600
km2) along the Wasatch Front and use them as the basis for ~1 km resolution input data for
the MEGAN model.
2. Characterize high-resolution (10 m) distributions of growth form fractions (tree, shrubs, crops,
herbaceous) in the Wasatch Front urban areas and use this to develop ~1 km resolution input
data for MEGAN.
3. Compile urban tree species composition data for the major cities of Wasatch Front and use this
to generate 1 km resolution inputs for MEGAN.
1.3 Overview of MEGAN Model
The project used the MEGAN model, a widely used biogenic emissions model designed to provide
inputs of all-important biogenic VOC on the required temporal and spatial scales for regional air quality
and global earth system models. This model comprehensively accounts for all BVOC emissions,
whether originating from natural ecosystems or managed landscapes such as urban areas. The current
version, MEGAN3.2, is updated from MEGAN3.1 and incorporates emission factors and vegetation
characteristics data from literature and global land cover products. The MEGAN model has been
extensively used in numerous biogenic emission modeling studies across the western US.
1.4 Overview of MEGAN Model Input Data
The MEGAN biogenic emissions model includes a global 30-second (i.e., 1/120 of a degree in latitude
and longitude, approximately 1 km) landcover database that users can use to generate landcover
inputs for their region of interest. For this project, we utilized MEGAN3.2 inputs, including emission
factors, landcover details (growth forms, ecotypes, and LAI), weather parameters (PAR, temperature,
wind speed, humidity, and soil moisture), and atmospheric composition (CO2, ozone W126 index).
The specific files are archived at DOI: 10.5281/zenodo.10939297. Furthermore, we made several
refinements to improve biogenic emissions for the Wasatch Front. These include updates to LAI, urban
tree species composition, and other vegetation characteristics. These updates were based on field
measurements conducted as part of this study and the analysis of fine-resolution satellite imagery.
1.5 Overview of Report
Northern Utah MEGAN landcover inputs were updated over the various regions shown in Figure 1-3. As
described in Section 2, Leaf Area Index (LAI) was characterized for an approximately 30,000 km2
region surrounding the Wasatch Front. Section 3 describes the approach used to characterize growth
form fractions across an approximately 86,000 km2 region (39 to 42N, 11 to 114W) of Northern Utah
using 10-m Worldcover data and iTree virtual surveys. In addition, the growth form distribution of an
approximately 2,000 km2 urban portion of the Wasatch Front was characterized using 0.5-m imagery
and an ArcGIS machine learning, segmented object approach. The estimation of tree speciation of ten
urban zones, together covering ~1200 km2, is described in Section 4.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
9
Figure 1-3. Map of regions with updated MEGAN landcover inputs. Magenta, green and
blue rectangles show regions with updated growth form fractions, LAI and emission
vegetation types, respectively. The ten new urban emission type regions are shown in the
blue rectangle breakout box on the right.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
10
Leaf Area and Vegetation Cover Distributions for the
Wasatch Front
To estimate the biogenic emissions from vegetation, we need to quantify the amount of emitting
vegetation. Plant leaves are the dominant emission source, so the amount of foliage is an important
variable for driving biogenic emission models. Two variables are used to quantify the amount and
density of vegetation: Leaf Area Index (LAI) and Vegetation Cover Fraction (VCF). LAI is defined as
one half (i.e., just one side of a leaf) of the total leaf area per unit ground surface area and has units
of m2 m-2. The LAI data used for MEGAN is green LAI which can be quantitatively defined as green leaf
elements that have a leaf chlorophyll content > 15 µg m-2. VCF is the fraction of total area that is
covered by green canopy elements as seen from the nadir (looking directly down at the canopy). VCF
is the sum of all vegetation cover and is a constraint on the sum of individual vegetation cover types
(growth forms) which include tree, shrub, grass/herbaceous and crop in the MEGAN model. The ratio
of the LAI to the VCF is used to calculate the MEGAN model input, LAIv, which represents the density
of foliage in the vegetation covered surfaces of a plant landscape. Some biogenic emissions, such as
isoprene, are dependent on solar radiation levels so an estimate of foliar density is required to
determine the amount of self-shading of leaves in a plant canopy. Section 2.1 describes the approach
for estimating LAI data for MEGAN input from high resolution satellite images and Section 2.2
describes the assessment of the new LAI data.
2.1 Quantify LAI and VCF Data
LAI and VCF input for MEGAN are commonly obtained from satellite imagery, although ground surveys
can also be utilized. LAI estimates based on satellite images are indirect measurements that rely on
retrieval methods, typically based on empirical data and radiative transfer models. For this project, we
used the Simplified Level 2 Product Prototype Processor (SL2P, Weiss and Baret 2016) to estimate LAI
from the Sentinel-2 satellite Multispectral Instrument (S2/MSI) 10 m resolution data.
Sentinel is a satellite system comprising two satellites: Sentinel 2A, launched in 2015, and Sentinel
2B, launched in 2017. Together, they provide approximately a 5-day temporal resolution for
monitoring vegetation and soil changes. The Satellite LAI Product (SL2P) utilizes a neural network
approach trained with a globally representative dataset derived from the PROSAILH canopy radiative
transfer model (Weiss et al., 2000). The S2/MSI data processing involves radiometric and geometric
correction using ground control points and a digital elevation model to mitigate parallax error. This
error occurs when the satellite's position deviates from directly overhead the observed area. The
broad swath width (290 km) of each S2/MSI tile spans across entire urban regions while the high
resolution (10 m) can accurately capture tree cover in an urban setting (Wong et al. 2019). European
Space Agency's (ESA) Sen2Cor processor was employed for cloud and shadow removal and for
converting Sentinel-2 Level 1C (MSIL1C) data into atmospherically corrected top-of-canopy
reflectance data. We selected relatively cloud free images for processing and visually inspected each
image to assess quality and geolocation errors. The S2/MSI satellite data was accessed from the
Copernicus Open Access Hub (https://scihub.copernicus.eu/).
We generated a 10-meter resolution LAI dataset for three specific time periods requested by the
UDAQ for the Wasatch Front region. Time interpolation between available images was employed to
estimate LAI at each location for 8-day periods:
• June & July 2017
• May, June, July, August, & Sept 2022
• May, June, July, August, & Sept 2023
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
11
A tutorial describing the approach is given in Appendix 1 and provides additional details of the
methods used. Appendix 1A describes how to access the data and process individual tiles using ESA’s
Sentinel Application Platform (SNAP) toolbox. Appendix 1B describes how to process individual tiles to
generate MEGAN LAI inputs of 8-day or 10-day average data using ArcGIS and integrate SNAP-derived
LAI with global LAI data.
2.2 Calibration and Assessment of LAI Data
The SL2P of the SNAP toolbox is a widely used tool for calculating “effective LAI” from 10 m Sentinel-2
data. While it demonstrates good performance over certain landscapes, it may encounter challenges,
especially over heterogeneous canopies such as urban forests. Brown et al. (2021) noted that SL2P
LAI estimates can exhibit comparatively poor performance in such environments. The effective LAI
measured by satellite imagery would be the same as actual LAI if the foliage in vegetation canopies
were randomly distributed. This is not the case for most vegetation and so the difference, defined by a
canopy clumping index, needs to be determined. Brown et al. (2021) conducted a comprehensive
assessment of SL2P LAI estimates across 19 sites in the US, encompassing various landcover types.
They used the SL2P processor to directly estimate effective LAI and compared this to in situ
measurements of effective LAI and true LAI. In comparison to in situ measurements of LAI, the
validation statistics for the SL2P LAI indicated that the values were highly correlated, with an average
r2 of 0.84, but with a significant mean bias of -0.68 and slope of 0.54. Validation statistics for
individual ecosystem types varied considerably. This demonstrates the need for field observations to
determine if calibration factors are needed to account for underestimations that occur for
heterogeneous canopies.
The high resolution (30 m) and detailed (847 vegetation types across the US) geospatial LANDFIRE
landcover map (see landfire.gov) contains 121 landcover types for the Wasatch Front regions but most
are minor with just 18 (38) of them together covering 80% (95%) of the Wasatch Front region. Our
field measurements targeted the 7 primary landcover types that dominate the total Wasatch Front
urban area. Additionally, we included measurements for 10 other landcover types, comprising various
forest, woodland, shrubland, grassland and agricultural areas.
Quantitative estimates of one-sided true green leaf area index were collected in June 27- July 1, 2023
at 19 representative field sites located within seven of the ten zones shown in Figure 1-3: North (e.g.,
Bountiful), Capital (e.g, Salt Lake City) Central West (e.g., West Valley), Central East (e.g., Millcreek),
South West (e.g., Herriman), South East (e.g., Draper) and South (e.g., Lehi). The study locations
and the surrounding landcover and LAI distributions are shown in Appendix 2. All field sites were in
built-up (e.g., roadway corridors, suburban, urban) locations.
Both the overstory (tree canopy) and understory (grass, forbs and shrub ground cover) LAI were
estimated. The total LAI at each point was calculated as the sum of understory and overstory LAI. The
overstory LAI was measured using digital hemispherical images and a canopy model and the
understory LAI was measured using optical sensing. LAI was measured at 585 points within 117 plots
across the 19 field sites. Each plot covered an area of about 20 m x 20 m (400 m2). The approaches
used to estimate LAI at the field sites are described in more detail in Appendix 3.
The LAI at the individual points within each plot were averaged to estimate the plot average LAI. Each
plot was associated with one of three LANDFIRE broad dominant cover types based on the growth
form and the degree of urbanization: forest/woods, grass/shrub, or built-up (pavement, structures,
etc.). Figure 2-1 shows that the field observations and satellite estimates are positively correlated for
each cover type. The 10 m satellite data was resampled and averaged to 30 m resolution to minimize
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
12
any geolocation errors. The linear best fit r2 values ranged from 0.82 for built up to 0.90 for
forest/woods. The slope was ~0.84 for both forest and ground cover and 0.95 for built-up. The
average LAI value for the 117 plots was 2.37 m2/m2 for the field measurements and 2.33 m2/m2 for
the satellite data indicating good agreement.
Figure 2-1. Scatterplot of satellite-based and field-based LAI averaged for 117 plots.
Solid line indicates 1:1 agreement. Landcover categories include forest/woods (filled blue
circles), grass and shrub (open black diamonds), and built-up (red open triangles).
Figure 2-1 shows that the Sentinel data underestimated the field LAI about half of the time, and never
overestimated, when the Sentinel values were < 1.2 m2/m2. This underestimate, of about a factor of
two on average, has little impact on the total LAI estimated for the Wasatch Front, especially forest
and woodland LAI which tends to have LAI values greater than 1.2. Another issue with the Sentinel-2
LAI data is that the roofs of some buildings were assigned high (>5) LAI values when they should
have a value of zero. These high LAI values are associated with bright white roofs and were not found
for other structures or built areas (e.g., parking lots, roads, off-white, grey or colored roofs).
However, it appears that this occurs relatively infrequently in the Wasatch Front, in comparison with
other urban areas we have examined. As these two issues (underestimated LAI for low LAI locations
and overestimated LAI for buildings with bright white roofs) are offsetting and relatively minor (each
impacting total Wasatch Front LAI by < 5%) we have not applied any adjustments to the raw Sentinel
LAI data other than to set negative LAI values to zero.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
13
Table 2-1. Comparison of 300 m CGL and 10 meter Sentinel SNAP LAI estimates for urban
zones in Wasatch Front region.
Zone Area SNAP LAI CGL LAI LAI
Difference
(km2) m2/m2 M2/m2 %
Far North (e.g., Ogden) 196 0.716 0.861 16.8
Central West (e.g., W.
Valley) 203 0.480 0.635 24.5
Capital (e.g., Salt Lake City) 158 0.427 0.521 18.0
Far South (e.g., Provo) 107 0.767 0.756 -1.40
South (e.g., Lehi) 139 1.01 0.946 -7.11
Central East (e.g., Millcreek) 136 0.749 0.891 15.9
Mid North (e.g., Layton) 151 0.581 0.804 27.8
South West (e.g., Herriman) 111 0.656 0.807 18.7
South East (e.g., Draper) 97 0.676 0.797 15.1
North (e.g., Bountiful) 109 0.842 0.93 9.48
Total area 1407 0.691 0.795 13.1
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
14
Growth Form Fraction Distribution for the Wasatch Front
and Northern Utah
The initial step in characterizing the enormous diversity in the BVOC emission rates across various
plant species is to divide the total vegetation within a specific location into the four MEGAN growth
forms. These growth forms are used due to their distinct emission characteristics and their
identification feasibility through satellite and aerial imagery. The four MEGAN growth forms consist of
trees, woody ground cover (encompassing shrubs and woody forbs), herbaceous ground cover
(including grass and herbaceous forbs), and crops. While trees are the dominant BVOC emission
source in many regions, other vegetation types may also significantly contribute to BVOC emissions.
The MEGAN model incorporates geospatially resolved distributions of the four growth form fractions at
a spatial resolution of 30 second (approximately 1 km2) covering the entire globe, excluding Antarctica
and the Arctic. As illustrated in Figure 3-1, our study focuses on Northwestern Utah, spanning from 39
to 42 degrees North latitude and 111 to 114 degrees West longitude, encompassing an area of
approximately 85,600 km2. Additionally, within this region, there is an urban target zone along the
Wasatch Front, covering an area of approximately 1,400 km², as shown in Figure 3-1. The figure
further delineates 10 subzones within the urban area (e.g., Far North, Capital, Southwest, etc.)
designated as urban MEGAN Emission Vegetation Type (EVT) zones in the global MEGAN EVT
database, as discussed in Section 4. The growth forms within the urban zone were characterized at a
spatial resolution of 0.5 meters, as outlined in Section 3.1, while the broader areas were characterized
at a spatial resolution of 10 meters. The new landcover data for Northwestern Utah were aggregated
to 30 seconds (approximately 1 km2) and integrated into the existing MEGAN3 global growth form
cover fraction input files, referred to as the MEGAN Growth Form Database version 7 (MGFDv7). The
study’s growth form distributions are illustrated, compared with alternative datasets, and evaluated
using iTree virtual observations of growth form fractions in this section.
The fraction of the Earth’s surface covered by four general vegetation growth forms (e.g., trees,
shrubs, herbaceous plants, crops) serves as essential input to the MEGAN BVOC emission model.
Previous efforts have generated these fractions using global satellite imagery with a spatial resolution
of 1000 m (Guenther et al. 2006) globally and up to 30 meters in specific regions like the contiguous
U.S. (Guenther et al. 2012). Outside of specific urban areas, the MGFDv7 data are based on the 100-
meter resolution Copernicus Global Land (CGL) Service landcover products version 3 dataset, derived
from the PROBA-V satellite imagery (see land.copernicus.eu/global/products/lc). The CGL dataset
categorizes landcover into 10 types (bare, crops, grass, moss, shrub, tree, snow, urban, permanent
water, seasonal water), which are then mapped to the MEGAN four growth forms. This mapping is
straightforward for most landcover types, matching directly with their respective growth forms (e.g.,
crops = crops, grass = grass, tree = tree, shrub = shrub, bare = bare). However, for the urban
landcover type, a constant growth form distribution must be applied unless alternative data is
generated, as done for Utah in this project. The CGL dataset is available as an integrated global
dataset for 2015, with annual updates thereafter, covering a global scale from 78.25N to 60S at a
resolution of approximately 100 meters at the equator. Its accuracy, when compared to 28,000
independent validation points, is stated to be 80% (Buchhorn et al. 2020). While the CGL tree cover
fraction represents a significant improvement over previous MEGAN landcover data, it has been noted
to underestimate tree cover in heterogeneous landscapes, including urban areas and arid woodlands.
To address this limitation, higher resolution (30-meter) National Land Cover Dataset (NLCD) tree
cover data, available only for the contiguous US, was integrated with the CGL data. The NLCD values
were used for urban areas and arid woodlands to enhance the accuracy of the MGFDv7 database.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
15
Figure 3-1. Inner left blue rectangle (and blue rectangle on the right) shows urban target
region including the extent of the ten Wasatch Front zones that have been designated as
ten urban MEGAN Emission Vegetation Types developed for this project (each with a
specified tree species composition). Outer left rectangle shows the rural 3-degree x 3-
degree rural region characterized using iTree.
3.1 Ultra-High Resolution (0.5 m) Urban Growth Form Cover Fraction
Distributions
Mature trees typically have crown areas of several meters or more, allowing for their identification and
crown area quantification using imagery with a spatial resolution of 1 meter or less (Figure 3-2 and 3-
3). Conversely, individual ground cover plants such as small shrubs and herbaceous species are often
smaller than 50 cm but tend to occur in larger clusters that can be accurately identified and classified
at 50 cm or lower resolution. This clustering phenomenon is particularly prominent in urban areas
characterized by grass lawns, shrub hedges, or gardens. It is worth noting that the NLCD tree cover
product primarily focuses on taller trees exceeding 5 meters in height, thereby omitting some of the
shorter tree species commonly found in urban landscapes.
The false color imagery, incorporating the near infrared band as depicted in Figure 3-2, can distinguish
vegetation by the cooler temperatures along with differences in visual light wavelengths. The
segmentation of the imagery, conducted prior to classification using a machine learning algorithm,
allows for object-based growth form classifications, as illustrated in Figure 3-3. The segmentation
approach enhances classification accuracy compared to traditional methods, where each pixel is
individually classified. By grouping pixels into meaningful objects based on spectral and spatial
characteristics, the object-based classification method improves the accuracy and efficiency of growth
form identification, thereby providing more reliable results for urban vegetation mapping and analysis.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
16
Figure 3-2. ArcGIS base map imagery showing Salt Lake City Hall and Courthouse and
surrounding parks.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
17
Figure 3-3. 0.5-m spatial resolution classification of the landcover distribution
surrounding the Salt Lake City Hall and Courthouse developed for this project. Segmented
NAIP natural color image (top) and classified cover (bottom) showing distributions of
buildings and pavement (red), groves and individual trees (green), dry grass (brown) and
green grass (yellow). Black areas are associated with shadows.
This project utilized imagery from the National Agriculture Imagery Program (NAIP), which acquires
aerial imagery via aircraft across the contiguous U.S. The specific methodologies used to quantify
urban growth form (GF) cover at a resolution of 0.5 meters using NAIP imagery and ArcGIS
segmentation and machine learning tools are outlined in detail in Appendix 4. NAIP imagery for the
year 2021 was accessed from the ArcGIS Portal Living Atlas (https://livingatlas.arcgis.com/). NAIP
imagery can be utilized to generate urban GF cover datasets throughout the contiguous US for specific
years, starting in 2003, although earlier NAIP imagery typically has lower resolutions (e.g., 1 to 2 m).
The ArcGIS “USA NAIP Imagery: Color Infrared product” utilized to develop GF cover distributions
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
18
consists of three bands: near infrared, visible green and blue bands. The ArcGIS Pro “Segment Mean
Shift” geoprocessing tool was used to segment each NAIP image into discrete objects, such as
individual features like buildings, trees, lawns, and lakes. Subsequently, an ESRI Classifier Description
(ECD) file was created, encompassing 12 landcover categories, including various types of trees,
ground covers (e.g., urban trees, native trees, shrub, green grass, brown grass, bare soil, herbaceous
crops), and non-vegetated landscapes (e.g., water, asphalt, colored rooftops). Appendix 4 describes
the creation process of the ECD files. Training samples were then generated according to the
procedures described in Appendix 4 and used to classify the objects using the ArcGIS “Random Tree”
machine learning geoprocessing tool. The images were subsequently classified using the ArcGIS
“ClassifyRaster” geoprocessing tool, assigning specific landcover types (e.g., oak tree) to each object
class. Finally, the ArcGIS “RECLASS” tool was used to assign tree, shrub, and herbaceous cover
fraction estimates to each location.
An example of the growth form distribution maps, encompassing the Salt Lake City area, is shown in
Figures 3-4 to 3-6. In these maps, it is evident that tree cover predominantly prevails in residential
areas. Shrubs are notably prominent around the airport vicinity and grass cover is prominently
observed in parks and golf courses.
Figure 3-4. Salt Lake City Tree cover distribution (%).
Figure 3-5. SLC shrub cover distribution (%).
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
19
Figure 3-6. SLC grass cover distribution (%).
3.2 High resolution (10 meter) growth form distributions for urban and
surrounding areas.
The approach described in Section 3.1 can generate accurate GF distributions over urban or similar
regions with comparable landcover characteristics. However, its applicability tends to be limited to
urban areas of approximately 150 km2 or less due to computational constraints. For instance,
characterizing a larger region would require significant computational resources, considering that a
100 km2 region comprises approximately 60 billion locations at a spatial resolution of 0.5 m x 0.5 m.
In the case of the 1400 km² urban Wasatch Front region studied in this project, it was subdivided into
ten regions, and the 0.5 m landcover characterization approach (as described in Section 3.1) was
applied separately to each of these divisions. Extending this methodology to the 85,600 km2 region of
Northwestern Utah, encompassing the urban Wasatch Front and its surrounding areas, would require
replicating the landcover classification approach approximately 570 times (85600 km2/ 150 km2)
which is impractical. To address the challenge of improving landcover characterization for larger
regions surrounding urban areas like the Wasatch Front, and to streamline the process of
characterizing landcover in other urban areas, an alternative approach was developed. This alternative
method involves leveraging the 10-meter Worldcover data, combined with the USDA i-Tree tool
(https://www.itreetools.org/) approach for quantifying tree, shrub, herbaceous, and crop cover within
a defined zone. The ESA Worldcover data, available globally for the years 2020 and 2021 through esa-
worldcover.org, offers comprehensive landcover information at a broader scale. The i-Tree approach
utilizes high-resolution Google Earth imagery and a random point generator to characterize vegetation
cover, as detailed in Appendix 5.
The effectiveness of the global 10-meter Worldcover data in representing landcover, including urban
areas, is demonstrated in Figure 3-7, which can be compared to the NAIP 0.5-meter product (Figure
3-3) and the original landcover image (Figure 3-2) of the SLC City Hall and Courthouse. It is evident
from the comparison that the Worldcover data generally captures the urban landcover accurately.
However, it should be noted that the Worldcover data does not provide fraction information regarding
growth forms but rather classifies each 10 m x 10 m location as a single landcover type. This
classification approach tends to classify locations with any trees present as solely trees, often leading
to an overestimation of actual tree cover (e.g., a location with 50% tree cover being classified entirely
as trees). To address this limitation, we constrained the Worldcover landcover distribution data using
i-Tree estimates of landcover continuous fractions. By doing so, we were able to accurately estimate
MEGAN growth form distributions at a 10-meter resolution. For this project, this was achieved by
dividing the study area into ten urban zones (as shown in Figure 3-1) and six rural zones, which were
used to characterize the forests and desert landscapes surrounding the urban areas in Utah.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
20
Figure 3-7. 10-m spatial resolution landcover distribution surrounding the Salt Lake City
Hall and Courthouse from the ESA Global WorldCover data. Landcover categories include
built-up (red), trees (green), grass (yellow) and barren (grey).
The resulting 10-meter Growth Form distribution data for the entire 3-degree x 3-degree area of
Northwestern Utah (as depicted in Figure 3-1) was integrated into the global MEGAN landcover
dataset (MGFDv7), which utilizes 100-meter CGL landcover data. The distribution of grass (Figure 3-
8), shrubs (Figure 3-9), and trees (Figure 3-10) across the region is illustrated below. The delineation
of zone borders indicates that the estimated rural grass and shrub cover derived from this project is
notably lower than the corresponding global CGL estimates. However, the tree cover estimates
demonstrate similarity between the two datasets. This discrepancy may stem from a tendency of the
CGL approach to overestimate grass and shrub cover, particularly in desert landscapes. The CGL
estimates for grass and shrub distributions appear reasonable for savanna, shrublands and grasslands
with mean annual precipitation of >30 inches, like Eastern Texas or Southern California chaparral. But
for more arid (<10 inches per year) desert landscapes in Utah, they may not be accurate. Further
investigation is required to determine if this is the case. Also, we found that the global CGL dataset
tends to show higher LAI in forested areas compared to SNAP Sentinel 2 data, but slightly lower in
urban areas (see Figure 3-11).
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
21
Figure 3-8. MEGAN Utah grass cover distribution (%).
Figure 3-9. Utah shrub cover distribution (%).
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
22
Figure 3-10. Utah tree cover distribution (%).
Figure 3-11. Difference between CGL and SNAP Sentinel 2 LAI (m2 m-2) estimates.
3.3 Comparison with alternative vegetation cover.
The i-Tree measurement results, detailed in Table 3-1, reveal a range of tree cover percentage across
different zones, varying from 9% in the Central West (e.g., West Valley) zone to 20.5% in the Central
East (e.g., Millcreek) zone. These values are significantly higher than the tree cover data previously
used for MEGAN modeling including The Wasatch Front urban tree cover in the NLCD (1.29%) and
CGL (5.25%) datasets that underestimate tree cover by 92% and 33%, respectively, indicating that
those datasets are missing a substantial fraction of tree cover in these heterogeneous urban
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
23
landscapes. The ESA Worldcover data has the correct tree distribution but tends to overestimate tree
cover by approximately 50%. This overestimation occurs because it classifies each 10 m x 10 m
location as tree cover, even if the location is only partly covered by trees. Major cities within the ten
zones consistently exhibit approximately 20% higher tree cover compared to the overall zone average.
This observation may indicate that urbanization has contributed to an overall increase in tree cover
within the region.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
24
Table 3-1. Comparison of tree cover percentage estimates for Wasatch Front urban zones
using i-Tree (i-Tree random point surveys for 2022), NLCD (NLCD 30 m landcover data for
2016), CGL (ESA CGL 100 m landcover data for 2015), WC (ESA Worldcover 10 m landcover
data for 2021) and the 0.5-m landcover developed for this study using 2021 NAIP data as
described in Section 3.1.
Zone iTree This study WC CGL NLCD
% % % % %
Year 2022 2021 2021 2015 2016
Spatial resolution (meters) N/A 0.5 10 100 30
Far North (e.g., Ogden) 18.7 7.4 24 6.2 1.7
Central West (e.g., W. Valley) 9.0 13.8 14.5 3.8 0.3
Capital (e.g., Salt Lake City) 15.4 11.6 14.9 2.9 1.5
Far South (e.g., Provo) 17.4 15.2 25.2 4.3 0.5
South (e.g., Lehi) 14.7 12.1 29.6 8 0.6
Central East (e.g., Millcreek) 20.5 23 36.9 5.4 3.6
Mid North (e.g., Layton) 14.4 7.2 22.8 4.7 0.7
South West (e.g., Herriman) 14.0 15.9 20.3 4.4 0.2
South East (e.g., Draper) 19.3 17.3 30.5 6 2.3
North (e.g., Bountiful) 14.8 14.9 31.2 6.7 2.3
All regions 15.5 13.3 24.1 5.3 1.3
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
25
Tree Speciation
The emission factors of certain BVOC, such as isoprene, vary by more than order of magnitude
between tree species. For instance, species like Oaks and Sweetgum are known to have high isoprene
emission rates, whereas others like Elm, Juniper and Pecan emit negligible amounts of isoprene. This
wide variation underscores the importance of quantifying tree species composition within each
landscape to accurately represent the emission potential of the ecosystem.
Virtual Urban Tree Surveys (VUTS) were conducted as part of this project to estimate tree speciation
for municipalities in the Wasatch Front. To validate the effectiveness of the VUTS approach, we
compared it with two other methodologies outlined by Shah et al. (2021): urban Forest Inventory
Assessment (uFIA) data collected by the US Forest Service and Municipal Street Tree Inventories
(MSTI) compiled for city governments. For this comparison, both uFIA and MSTI data were available
for Houston, while only uFIA data were available for San Antonio, and only MSTI data were available
for Austin. The comparison revealed that the VUTS approach generally aligns with the findings of the
other two methodologies. Each of these approaches are described briefly below:
• Virtual Urban Tree Surveys (VUTS): The VUTS species composition data was based on several
hundred randomly selected trees within each city urban area. The resulting data is
representative of each entire city but represents relatively few sampling points. The
procedures for selecting the trees are described in Appendix 5 and the approach for identifying
them using high resolution aerial imagery and, where available, Google Street View images is
described in Appendix 6 using the Texas urban tree key in Appendix 7.
• Urban Forest Inventory Assessment (uFIA): The uFIA approach involves ground surveys of 1/6
acre plots that are randomly selected within each major land use type in the city. For instance,
in Houston, one plot was selected per approximately 3 square miles, resulting in about 200
plots covering the city’s 640 square miles. Each tree within these plots is identified, and the
Diameter at Breast Height (DBH) of each tree is measured. With an average of around 8 trees
per plot in Houston, this method samples approximately 1,600 trees. Crown cover area was
estimated from these data using the equations described by Geron et al. (1994) and the
species composition of the landscape was estimated based on the relative crown cover area.
• Municipal Street Tree Inventories (MSTI): MSTI inventories are conducted by cities to quantify
trees on city-owned property, particularly along city streets. Data from MSTI inventories were
available for all of Houston (representing approximately 200,000 trees throughout the city)
and a portion of downtown Austin (approximately 7,300 trees). It is important to note that the
number of trees in the MSTI data is significantly higher than that sampled for uFIA, MSTI data
only represent city streets, which account for approximately 15% of the total urban area.
Therefore, MSTI data may differ significantly from the city's average tree species composition,
as it does not capture trees from other land use types within the city.
MEGAN categorized urban areas in the US into four classes based on the LANDFIRE landcover
database. These classes include low intensity (areas with few built structures), medium intensity, high
intensity (densely built-up areas) and roads. The tree composition assigned to each urban category is
derived from trees observed in FIA plots, which serve as a sampling frame for tree measurements.
These FIA plots are selected within the LANDFIRE landcover types corresponding to each urban class
and are averaged over the entire US. Figure 4-1 illustrates that the assigned tree composition is
nearly identical for all four urban types. However, it is essential to note that this composition is based
on a US average and is generally reflective of forests across the country.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
26
Figure 4-1. MEGAN tree composition used in MEGAN 2.1, 3.1 and 3.2 for US urban
landcover classes that have not been surveyed.
The survey, conducted using iTree, analyzed nearly 10,000 trees to assess growth form fractions and
identify urban trees at the genus level using the VUTS approach. Out of the 30 observed genera, just
four—Acer (Maple, 17.2%), Fraxinus (Ash, 12.9%), Ulmus (Elm, 12.5%), and Populus (Aspen and
Poplars, 11.2%)—accounted for over half of all trees surveyed. An additional six genera contributed to
32% of the total, each representing 4 to 7% of the surveyed trees, including Picea (Spruce), Malus
(Apple), Platanus (Sycamore), Gleditsia (Honey Locust), and Tilia (Basswood). The remaining 15%
comprised various genera such as Pyrus, Elaeagnus, Juniperus, and others.
The locations of 214 of the 917 VUTS sampled trees were surveyed in the field. Six of these locations
no longer had a tree. This likely occurred because the tree had been removed in the time between the
field survey and when the imagery used for VUTS was obtained. The field survey identified both the
tree nearest to the target location as well as the two nearest trees (if they were within ~5 m). This
second dataset both reduces geolocation errors (which could lead to the VUTS tree and field tree being
different) and resulted in a larger field tree dataset. We calculated tree genera statistics using both
approaches. The single tree approach (208 trees surveyed) results found that five genera (Acer,
Fraxinus, Ulmus, Gleditsia, and Pinus) comprised between 8 and 13%, four genera (Picea, Pyrus,
Quercus, Populus) were between 5 and 7% three genera (Platanus, Tilia, Prunus) were between 2 and
4% and 14 genera were between 0.4 and 2% (Ailanthus, Elaeagnus, Malus, Juniperus, Catalpa,
Koelreuteria, Morus, Robinia, Salix, Aesculus, Ligustrum, Thuja, and Zelkova). The multiple tree
approach (529 trees surveyed) indicated that three genera (Fraxinus, Acer, and Ulmus) were 8 to
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
27
11% each, five genera (Picea, Pinus, Gleditsia, Populus, and Pyrus) were 7 to 8 % each, seven genera
(Prunus, Quercus, Tilia, Malus, Juniperus, Platanus, Ailanthus) were 2 to 5% each, seven genera
(Robinia, Elaeagnus, Morus, , Catalpa, Koelreuteria, Salix, and Betula) were 0.7 to 1.5% and ten
genera (Cercis, Aesculus, Liriodendron, Rhus, Thuja, Zelkova, Ginko, Juglans, Ligustrum, and Taxus)
were each between 0.1 and 0.5%.
An additional available tree inventory dataset is the Salt Lake street and parks tree inventory
(https://www.arcgis.com/home/item.html?id=5005408e8d2742eb830b9cee87833d4a)
containing the location and species of the 85,000 trees growing on city land in Salt Lake City.
The four tree inventories (VUTS, Field-one, field-multiple, SLC city street and park tree inventory) all
revealed that Acer, Fraxinus and Ulmus as the three most dominant genera, collectively constituting
30 to 42% of the total trees surveyed. The subsequent seven genera accounted for 42 to 43% of the
total trees in the field and VUTS inventories. However, the SLC tree inventory recorded only 36% for
the next seven genera, possibly because the 85,000 trees captured more rare species trees.
Alternatively, this disparity could be due to SLC’s greater diversity of trees. In the top ten genera,
which collectively represent approximately 80% of all trees, the field-multiple approach included Picea,
Pinus, Gleditsia, Populus, Pyrus, Prunus, and Quercus alongside the top three Acer, Fraxinus and
Ulmus. The Field-1 approach shared the same genera in the top ten, except that Platanus was ranked
#10 and Prunus was #12. The VUTS approach had seven of the same top ten genera, with Malus,
Platanus and Tilia included, while Pyrus, Quercus and Prunus were not. Similarly, the SLC inventory
comprised seven of the same top ten genera, with Tilia, Platanus and Malus included, and Populus,
Picea and Quercus (all isoprene emitters) omitted. It's worth noting that the SLC inventory solely
pertains to Salt Lake City, whereas the others cover a much larger region, thus differences in species
distribution are to be expected.
Table 4-1 presents the fraction of isoprene emitters estimated by the field (average of the single and
multiple tree approach), VUTS and SLC inventories for individual zones, alongside the value utilized in
MEGAN, which represents the weighted average of the various approaches. Across all approaches, the
isoprene emitter fraction was relatively low for Salt Lake City, with estimates of 13.9% (Field), 15.1
(SLC street tree inventory), and 18.2% (VUTS). Notably, VUTS yielded higher estimates than the field
in four of the seven zones with field data, while it was lower in three. The average isoprene emitting
fraction for the seven zones was 26% for VUTS and 25.8% for the field. However, there was
considerable variability among individual zones, with differences ranging from 54% higher to 35%
lower as shown in Figure 4-2. The 24% decrease in isoprene emission due to a lower fraction of
isoprene emitters found by this study was more than offset by the higher tree cover fraction (Figure 4-
2).
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
28
Figure 4-2. The ratio of new (this study) to old (default MEGAN 3.2) values for LAI, tree
cover, fraction of isoprene emitters and the overall impact.
Table 4-1. Percentage of trees classified as isoprene emitters for ten urban zones in the
Wasatch Front region. Estimates are based on Salt Lake City tree inventory (limited to trees
on public land), the Virtual Urban Tree Survey (VUTS) approach outlined in Appendix 6, and
data from the 2023 urban field study. The MEGAN EVT code and corresponding values from
the MEGAN model are provided.
Zone MEGAN EVT Area SLC
inventory VUTS Field MEGAN
code (km2) % % % %
Far North (e.g., Ogden) 2406 196 - 25.7 - 25.7
Central West (e.g., W.
Valley) 2407 203 - 26.9 - 26.9
Capital (e.g., Salt Lake City) 2408 158 15.1 18.2 13.9 14.9
Far South (e.g., Provo) 2409 107 - 18.9 - 18.9
South (e.g., Lehi) 2410 139 - 35.0 37.5 36.4
Central East (e.g., Millcreek) 2411 136 - 27.3 20.0 26.0
Mid North (e.g., Layton) 2412 151 - 30.0 23.8 27.3
South West (e.g., Herriman) 2413 111 - 25.8 16.7 17.6
South East (e.g., Draper) 2414 97 - 27.1 41.5 40.9
North (e.g., Bountiful) 2415 109 - 18.7 26.9 22.8
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
29
Compile MEGAN Inputs and Assess Emissions
We processed updated urban landcover data as outlined in Sections 2 through 4, converting it into the
format required by the MEGAN model. Subsequently, we conducted an evaluation to assess the
model’s sensitivity to updated urban landcover input data. This evaluation involved comparing BVOC
emissions estimates for the Wasatch Front region generated with both the current MEGANv3.2
vegetation cover data and the new updated vegetation cover data developed in this project. The two
input data sets are described in Table 5-1.
Table 5-1. Comparison of default MEGAN 3.2 inputs with those developed for this study.
Description Default This study
LAI spatial resolution 333 m 10 m
LAI Year 2019 2017, 2022 and 2023
LAI Source ESA CGL This study based on SNAP Sentinel 2
Growth Form spatial resolution 100 m 0.5 m and 10 m
Growth Form Year 2020 2020
Growth Form Source ESA CGL This study based on NAIP, Worldcover, iTree
Tree species Composition Source rural FIA This study based on observations
We utilized the latest version of the MEGAN model (MEGAN3.2) to assess emissions sensitivity to
variations in vegetation cover. The comparison focused on BVOC simulated during the July 2017
summer month, within the UDAQ 1.33 km domain used in UDAQ’s ozone modeling for air quality
planning.
In these sensitivity simulations, we disabled the effects of drought, ozone stress, CO2 concentrations,
and bidirectional exchange LAI response, while activating the emission responses to windstorms and
extreme temperatures. Among these factors, drought and ozone exert significant influences, but they
are typically associated with large uncertainties.
Our analysis indicated a slight decrease (1%) in total BVOC emissions for the 1.33 km domain,
focusing on the Wasatch Front region, when using updated vegetation cover compared to estimates
derived from the current vegetation cover (see Table 5-2). This decline in BVOC emissions primarily
stems from lower LAI values observed in the S2/MSI satellite data across the Wasatch Front region
compared to MEGANv3.2. It should be noted that the S2/MSI LAI data in Figure 5-1 are for 2017 while
the MEGANv3.2 default data are for 2019. The decrease in emissions due to LAI is partially offset by
the increase in emissions due to vegetation cover updates.
Regarding individual BVOC species, we observed that isoprene emissions increased by approximately
3.6% when using the updated vegetation cover data for the Wasatch Front region. On the other hand,
volatile terpenes showed a negligible change of 0.1%. These changes were more pronounced for
isoprene emissions compared to terpene emissions with the updated vegetation cover data.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
30
Figure 5-2 illustrates the comparison of isoprene emissions for the Wasatch Front region between the
current and updated vegetation cover developed in this project. Generally, MEGAN3.2 estimates lower
isoprene across the entire Wasatch Front region with the updated vegetation cover input compared to
the current input. However, certain areas along the Wasatch Front show higher isoprene emissions.
Figure 5-3 illustrates the comparison of terpene emissions for the Wasatch Front region. It's worth
noting that the difference plot does not align precisely with the isoprene emissions difference plot
footprints.
Table 5-2. Total episode average BVOC emissions (tons/day) for July 2017 estimated in the
Wasatch Front region using MEGAN3.2 with the current and newly updated vegetation cover
developed in this project.
VOC type Current vegetation cover
(tons/day)
Updated vegetation cover
(tons/day) % Difference
ISOP 187.8 194.6 3.6%
TERP 93.6 93.7 0.1%
VOC 465.2 460.5 -1.0%
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
31
MEGAN3.2 2019 LAI Updated 2017 LAI Differences
Figure 5-1. LAI comparison between the updated data developed for this project and the
2019 MEGANv3.2 vegetation cover database. Black boundaries delineate the northern and
southern regions of the Wasatch Front.
Figure 5-2. July 2017 average isoprene emissions (kg/day) for the Wasatch Front region
simulated using MEGAN3.2 with current and updated vegetation cover developed in this
project.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
32
Figure 5-3. July 2017 average terpene emissions (kg/day) for the Wasatch Front region
simulated using MEGAN3.2 with current and updated vegetation cover developed in this
project.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
33
Conclusions
Biogenic emissions play a significant role in shaping air quality in Utah's urban areas and contribute
substantially to total VOC emissions in certain locations. However, these biogenic emissions remain
relatively uncertain in urban areas due to heterogeneity of the urban landscape and a lack of studies
on urban vegetation. The overall goal of this project was to improve estimates of BVOC emissions,
with a specific focus on the Wasatch Front area, particularly emissions from urban areas.
To accomplish this, we employed a novel urban tree characterization approach, called the Virtual
Urban Tree Survey, and used high spatial resolution (60 cm to 10 m) aerial and satellite imagery to
develop fine spatial resolution (~1 km) time-varying LAI, total vegetation cover, and the relative
abundance of high BVOC-emitting trees such as oaks, elms, ash, maples, pines, and juniper, as well
as other vegetation cover types present in urban areas of the Wasatch Front. Furthermore, we
assessed the sensitivity of the urban biogenic emissions to the updates developed in this project. The
results from this study were subsequently integrated into MEGAN3.2, accessible at DOI:
10.5281/zenodo.10939297. The differences between the standard MEGAN3.2 inputs and those
generated for this study are summarized in Table 6-1. Overall, these results indicate a small (15%)
increase in LAI, a 24% decrease in tree speciation, and a doubling in the tree fraction. It should be
noted that updating BVOC emission factors was not part of the scope of this project. Many of the
urban trees in the Wasatch Front have few reported data and the emission factors are uncertain. Even
greater uncertainties are associated with the BVOC emission factors of the shrub and grass species
that dominate the emissions of the shrub and grasslands surrounding the Wasatch Front.
Table 6-1. Ratio of LAI, tree fraction, and isoprene emitter fraction estimated in the current
study compared to values in existing MEGAN input files. Their combined impact on isoprene
emission is estimated as the product of these three values.
Zone
MEGAN
EVT Area LAI Tree
Fraction
Isoprene
Emitters Combined
code (km2) ratio ratio ratio ratio
Far North (e.g., Ogden) 2406 196 1.17 1.14 0.77 1.02
Central West (e.g., W. Valley) 2407 203 1.25 2.71 0.82 2.75
Capital (e.g., Salt Lake City) 2408 158 1.18 2.01 0.44 1.05
Far South (e.g., Provo) 2409 107 0.99 2.68 0.57 1.51
South (e.g., Lehi) 2410 139 0.93 1.37 1.08 1.38
Central East (e.g., Millcreek) 2411 136 1.16 1.93 0.79 1.76
Mid North (e.g., Layton) 2412 151 1.28 1.5 0.83 1.58
South West (e.g., Herriman) 2413 111 1.19 3.56 0.53 2.25
South East (e.g., Draper) 2414 97 1.15 2.13 1.24 3.03
North (e.g., Bountiful) 2415 109 1.09 1.7 0.61 1.14
Total 1407 1.15 2.02 0.76 1.72
The primary benefit of this project is more accurate BVOC emission estimates for the Wasatch Front
region. This improved emission inventory is available to the UDAQ for use in air quality simulations
that are crucial for scientific understanding and developing regulatory control strategies aimed at
improving and maintaining clean air in the region. Key findings from the project are summarized
below:
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
34
6.1 Summary of findings
• The vegetation landcover data developed by this project will reduce uncertainty in BVOC
estimates for the Wasatch Front region by improving LAI, tree cover fraction and tree species
composition.
• Our comparisons of alternative LAI data for the Wasatch Front area, derived from high-
resolution Sentinel-2 satellite imagery, found relatively small differences from the current
MEGAN LAI.
• Landcover datasets based on 30-meter (and coarser) resolution imagery tend to underestimate tree
cover in urban areas by approximately 50%. Sub-meter resolution imagery (e.g., NAIP) can
accurately capture tree cover associated with individual trees, making it suitable for quantifying
urban tree cover.
• Virtual urban tree surveys (VUTS) emerge as a cost-effective and reliable approach for quantifying
urban tree species composition for BVOC emission modeling.
• Spruce, aspen/poplars, and oaks trees are the primary sources of isoprene emission in the urban
forests of the Wasatch Front region. Sycamore, Black locust, willows and redbud trees make minor
contributions to the total isoprene emissions.
• The estimation of tree species composition shows good general agreement, particularly for the three
main genera (maples, ash, and elms).
• MEGAN3.2 should be used for estimating emissions of isoprene, monoterpene, sesquiterpenes and
other biogenic emissions in the Wasatch Front region. The primary reasons for choosing MEGAN
are: (1) BEIS data structures can’t characterize urban vegetation, and (2) MEGAN simplifies the
process of updating and improving landcover and emissions data. The current MEGAN inputs could
be improved including shrub cover fraction and emission factors for the grass and shrublands
surrounding the Wasatch Front.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
35
References
Arietta, A.Z., Estimation of forest canopy structure and understory light using spherical panorama
images from smartphone photography, Forestry: An International Journal of Forest Research,
Volume 95, Issue 1, January 2022, Pages 38–48, https://doi.org/10.1093/forestry/cpab034.
Brown, Luke A., Richard Fernandes, Najib Djamai, Courtney Meier, Nadine Gobron, Harry Morris,
Francis Canisius, Gabriele Bai, Christophe Lerebourg, Christian Lanconelli, Marco Clerici,
Jadunandan Dash, Validation of baseline and modified Sentinel-2 Level 2 Prototype Processor
leaf area index retrievals over the United States, ISPRS Journal of Photogrammetry and
Remote Sensing, Volume 175, 2021, Pages 71-87, ISSN 0924-2716,
https://doi.org/10.1016/j.isprsjprs.2021.02.020.
Buchhorn, M., Lesiv, M., Tsendbazar, N. E., Herold, M., Bertels, L., & Smets, B. (2020). Copernicus
global land cover layers—collection 2. Remote Sensing, 12(6), 1044.
Fang, H., Baret, F., Plummer, S., & Schaepman‐Strub, G. (2019). An overview of global leaf area
index (LAI): Methods, products, validation, and applications. Reviews of Geophysics, 57,
https://doi.org/10.1029/ 2018RG000608.
Geron, C. D., Guenther, A. B., & Pierce, T. E. (1994). An improved model for estimating emissions of
volatile organic compounds from forests in the eastern United States. Journal of Geophysical
Research: Atmospheres, 99(D6), 12773-12791.
Guenther, A., Karl, T., Harley, P., Wiedinmyer, C., Palmer, P. I., & Geron, C. (2006). Estimates of
global terrestrial isoprene emissions using MEGAN (Model of Emissions of Gases and Aerosols
from Nature). Atmospheric Chemistry and Physics, 6(11), 3181-3210.
Guenther, A. B., Jiang, X., Heald, C. L., Sakulyanontvittaya, T., Duhl, T. A., Emmons, L. K., & Wang,
X. (2012). The Model of Emissions of Gases and Aerosols from Nature version 2.1 (MEGAN2.
1): an extended and updated framework for modeling biogenic emissions. Geoscientific Model
Development, 5(6), 1471-1492.
Ramboll (2021). Biogenic Emission Inventory Improvements for the Waco Area.
Ramboll (2023). Leaf Area Data Improvements for the Biogenic Emission Inventory of the East Texas
Council of Governments
Shah et al. (2021) Texas Urban Vegetation BVOC Emission Source Inventory, Final Report, AQRP
Project No. 20-007.
Trezza, Ricardo & Allen, Richard & Tasumi, Masahiro. (2013). Estimation of Actual Evapotranspiration
along the Middle Rio Grande of New Mexico Using MODIS and Landsat Imagery with the
METRIC Model. Remote Sensing. 5. 5397-5423.
UDAQ, 2023. State Implementation Plan, 2015 Ozone NAAQS Northern Wasatch Front Moderate
Nonattainment Area, Section IX Part D.11. https://documents.deq.utah.gov/air-
quality/planning/DAQ-2023-011344.pdf.
Weiss M. and Baret. F. (2016). S2ToolBox level 2 products. Version 1.1. Available online at
step.esa.int/docs/extra/ATBD_S2ToolBox_L2B_V1.1.pdf
Wong, M. M. F., Fung, J. C. H., & Yeung, P. P. S. (2019). High-resolution calculation of the urban
vegetation fraction in the Pearl River Delta from the Sentinel-2 NDVI for urban climate model
parameterization. Geoscience Letters, 6, 1-10.
Yan, K.; Park, T.; Yan, G.; Liu, Z.; Yang, B.; Chen, C.; Nemani, R.R.; Knyazikhin, Y.; Myneni, R.B.
Evaluation of MODIS LAI/FPAR Product Collection 6. Part 2: Validation and Intercomparison.
Remote Sens. 2016, 8, 460. https://doi.org/10.3390/rs8060460
Yarwood, G., Wilson, G., Shepard, S., & Guenther, A. (2002). User’s guide to the global biosphere
emissions and interactions system (GloBEIS) version 3. ENVIRON International Corporation,
773.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix 1: Using SNAP to estimate LAI and VCF using 10-
m resolution ESA Sentinel2-MSI data
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-1
Appendix 1. Using SNAP to estimate LAI and VCF using 10-m
resolution ESA Sentinel2-MSI data
This section provides a tutorial to process Sentinel2-MSI data to develop LAI inputs for the MEGAN
model. Appendix 1A describes how to access S2/MSI data and process individual tiles using the SNAP
tool. Appendix 1B describes how to process individual tiles to generate MEGAN LAI inputs of 8-day or
10-day average data using ArcGIS.
Appendix 1A
Step 1. Download Sentinel 2 data
Go to https://scihub.copernicus.eu/dhus/#/home. Log in (sign up for account if you have not yet
done so). Start in “Navigate mode” and navigate to the desired region of the planet and zoom into
the appropriate scale. Switch to “Area mode” and select area of interest on map by holding down left
button and drag polygon over area. Go to advanced search by clicking on the three bars at top left.
Enter desired sensing period (start date on the left and end date on the right). Click on the box to the
left of “Mission: sentinel 2”. Do not put anything in “Satellite platform”. Choose S2MSI2A for product
type. Enter acceptable cloud cover, e.g. “[0 TO 10]”. Note that the “TO” must be capital letters. Click
on search (right side of top bar). Download images by placing cursor over the tab and then click on
arrow at far-right bottom. Note that you can have only three files downloading at a time. The time to
download a file depends on the size (many are just a fraction of a full tile), your internet speed and
probably how busy the server is. A 500 MB file can take from 10 seconds to 10 minutes to download.
The download is a zipped file containing all of the MSI sensor bands.
Step 2. Use SNAP to calculate LAI and VCF
Open SNAP app. If you don’t have it then it can be download from
http://step.esa.int/main/download/snap-download/.
Choose sentinel 2 toolbox. Click on the “open file” icon on the far left and locate the zipped file that
you downloaded. It will then show up in the product explorer window.
First you need to resample the bands so they are all at 10m: Go to Raster> Geometric operation >
Resampling to open the Resampling tool window. Click on “resampling parameters” and in the “define
size of resampled product” box under “By reference band from source product” select “B2”. This will
set the size to 10 m and it will indicate a target height and width of 10980. Click “Run” then “OK” and
then close the resampling window.
Next you will calculate LAI and veg cover fraction: Go to Optical> Thematic land processing>
Biophysical Processor (LAI, fAPAR…) to open the Biophysical processor tool window.
In the “Source Product” box, select the resampled file you just created (it will say “resampled” at the
end of the file name (you can widen the box to enable viewing the whole file name).
Save as BEAM-DIMAP format
Click on “Processing parameters” and deselect “FAPAR”, Cab, and CWC.
Then click on “Run”. It will take up to ~1 hour for a full tile (Windows system with 64GB RAM and 8
core processors) but most files are only a fraction of the whole tile and so take less time.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-2
Step 3: Process the 10-m data to generate inputs for MEGAN.
The LAI and Fcover files (and their associated “flag” files) are created in a folder that is named the
same as the original downloaded file but with “resampled_biophysical.data” attached. The files are
ENVI image files which can be uploaded into ArcGIS or other programs where they can be
interpolated/resampled to the appropriate time step (8-day for LCDB13, 8 or 10 day or monthly for
MEGAN 3.2) and spatial resolution (1/120 of a degree). The flag files can be used to mask out cloud
contamination.
Appendix 1B Processing Sentinel LAI and VCF data using ArcGIS.
As a first step, the LAI and fcover images (.img files) generated using the ESA SNAP tool should be
visually inspected to identify and remove high cloud cover images and the SNAP output should be
checked for corrupt images. For an example, see Figure A1-1 below.
Figure A1-1. Corrupted LAI image for January 27, 2019 in Houston tile RTN
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-3
ArcGIS python can be used to process the uncorrupted images. The processing includes 5 steps that
can be accomplished using the 5 ArcGIS python programs:
1. Merge partial scenes (see SentinelRTNStep1.py). For some locations, the 100 km x 100
km Sentinel tile is fully within the swath of both Sentinel sensors (2A and 2B) so that each
satellite pass provides a full image. An example is the area around San Antonio, Texas (ESA tile
T14 RNT). In that case step 1 is not required. For other locations, such as the Austin Texas
example shown in A1-2, some or all images includes a swath of missing data on the left (west)
or right (east). The missing data in each of the scenes can be replaced by the nearest (in time)
scene with data in the missing area. The missing data includes both “nan” shown in black in
A1-2 and constant values appearing as grey triangles along the border of the “nan” data (see
red oval in A1-2, bottom). The “black” areas in A1-2 can be identified by a value of zero in the
Sentinel Scene classification (SCL) provided at 20 meter resolution for each image. The “grey”
triangle missing data (A1-2 bottom) can be identified as locations where SCL is greater than
zero and Lai is greater than zero.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-4
Figure A1-2. Examples of partial LAI images for Austin TX region that have missing data
on the west side (top) or on the east side (bottom)
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-5
2. Identify missing data (see SentinelRTNStep2.py). Missing data in each image can be
replaced by the data from the nearest (in time) scene. Clouds are typically flagged by SNAP with
a value of “9” and “10” in the SCL image and so can be identified as missing data. Cloud
shadows and the borders of some clouds are assigned values of 2, 3, 7 and 8 in the SCL images
but in some cases these values are not clouds but are locations with low vegetation cover. They
can be at least partially identified by checking to see if a given location is assigned these values
in most scenes (in which case it is probably due to the landcover) or in only a few scenes (in
which case it is probably due to clouds).
3. Replace missing data and identify maximum vegetation cover fraction (see
SentinelRTNStep3.py). The missing vegetation cover data identified in step 2 is replaced by
the data from the nearest (in time) scene. The annual maximum vegetation cover is then
determined as the maximum of all cover fractions for a location.
4. Replace missing data and identify maximum vegetation cover fraction (see
SentinelRTNStep4.py) Interpolate LAI. The missing LAI data identified in step 2 is replaced
by the data from the nearest (in time) scene. The LAI range for any location is then set with a
minimum of zero (SNAP will give some negative numbers) and a maximum based on the
maximum allowed LAIv (SNAP gives some unrealistically high values of LAI). MEGAN2 and
MEGAN3 data preprocessors expect LAI data for 8-day time periods (46 in a year). This was
designed to align with time step of MODIS, the primary source of satellite LAI and cover data.
While Sentinel has the potential to have data coverage every 5 days (there are two satellites
that each have a return of 10 days) cloud cover and missing scenes typically reduces the
number of available images. For these cases, values for each 8-day period can be interpolated
using a linear interpolation of the available images to generate a time weighted average value
for each 8-day period.
5. Calculate LAI and LAIv: vegetation covered surfaces (see SentinelRTNStep5.py). 8-day
LAIv inputs required for MEGAN2.1 and MEGAN3.1 are calculated by dividing LAI by annual
maximum vegetation cover fraction. MEGAN3.2 uses just LAI and can be 8- or 10-day LAI.
6. Integrate Wasatch urban LAI data into MEGAN global LAI data. The 2021 version of the
MEGAN global LAI data uses the CGL Service 2019 LAI 1km version 2 products which are
provided as global, multi-band netCDF4 files with metadata according to the Climate and
Forecast (CF) conventions (v1.6). The CGL LAI data are available at a temporal resolution of
~10 days and spatial resolution of 1/112 degrees (i.e., ~1 km) for the whole world.
To merge SNAP-derived LAI with global data, the two datasets need to be aligned in temporal and
spatial resolution. The global 10-day LAI data were converted into 8-day to align with the Texas urban
LAI data. ArcGIS was used to remap the fine resolution LAI data from SNAP onto ~ 1km resolution
global LAI data. To achieve this, the global dataset in ncf format was first converted into raster
dataset for processing. ArcGIS “Zonal Statistics” was used to calculate average LAI data from the 10-
m raster data within each grid cell of the global raster data. This can then be converted to a lat/lon
grid with 1/112 degrees resolution (0.008-degree cell size) using ArcGIS Resample set to “bilinear
interpolation”. Multiple input raster datasets were merged into one file using ArcGIS “Mosaic” tool for
ease of processing. Once the SNAP-derived LAI data were merged and on the same grid as global
data, the urban LAI data were merged into global data using ArcGIS “Mosaic to New Raster tool” and
then converted into NCF format needed by MEGAN.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-6
Appendix 2: Site descriptions
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-7
Appendix 2. Site descriptions
LAI estimates were obtained at 18 representative urban and suburban field sites across the Wasatch
Front, extending as far south as Lehi and as far north as Farmington. Figure A2-1 shows the location
of these eighteen field sites, with seventeen are in the Northern Wasatch Front and one (Lehi) is in the
Southern Wasatch Front.
Figure A2-1. LAI field site locations.
Figures A2-2 to A2-5 show the locations of field plots within the context of regional landcover
distributions. There are a total of 117 plots. LAI was estimated at 585 individual points (5 points
within each plot). The points within a plot were 10 to 20 meters apart. The two figures for each study
region show the locations of the plots and study sites relative to 1) the ESA 10 meter resolution
WorldCover landcover data and 2) the Sentinel 2 SNAP SL2P 10 meter LAI data.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-8
Figure A2-2. Northern Wasatch Front LAI sampling locations north of Salt Lake City (e.g.,
Farmington, Bountiful, N. Salt Lake) are shown as compass points. Left Panel shows
landcover (green = forest, yellow=grass, purple=crop, grey =bare, red=built/urban,
blue=water). Right panel shows LAI distribution based on 10 meter Sentinel2 SNAP data
(white <0.5 m2/m2, yellow is 0.5 to 2 m2/m2, green is >2 m2/m2.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-9
Figure A2-3. Northern Wasatch Front LAI sampling locations in and near Salt Lake City
(e.g., Salt Lake City, South Salt Lake City, West Valley City) are shown as compass points.
Top Panel shows landcover (green = forest, yellow=grass, purple=crop, grey =bare,
red=built/urban, blue=water). Bottom panel shows LAI distribution based on 10 meter
Sentinel2 SNAP data (white <0.5 m2/m2, yellow is 0.5 to 2 m2/m2, green is >2 m2/m2.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-10
Figure A2-4. Northern Wasatch Front LAI sampling locations south of Salt Lake City (e.g.,
West Jordan, South Jordan, Sandy, Draper) are shown as compass points. Top Panel shows
landcover (green = forest, yellow=grass, purple=crop, grey =bare, red=built/urban,
blue=water). Bottom panel shows LAI distribution based on 10 meter Sentinel2 SNAP data
(white <0.5 m2/m2, yellow is 0.5 to 2 m2/m2, green is >2 m2/m2.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-11
Figure A2-5. Southern Wasatch Front LAI sampling locations in Lehi are shown as compass
points. Left Panel shows landcover (green = forest, yellow=grass, purple=crop, grey =bare,
red=built/urban, blue=water). Right panel shows LAI distribution based on 10 meter
Sentinel2 SNAP data (white <0.5 m2/m2, yellow is 0.5 to 2 m2/m2, green is >2 m2/m2.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-12
Appendix 3: Methods to estimate LAI from field
measurements
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-13
Appendix 3. Methods to estimate LAI from field measurements
One sided green leaf area was quantified at 600 points within 120 plots in 8 cities (Millcreek,
West Valley, Bountiful, Salt Lake City, Lehi, Draper and Herriman) in the Wasatch Front.
Both the overstory (tree canopy) and understory (grass, forbs and shrub ground cover) LAI
were estimated. Overstory and understory LAI were estimated using the image processing
approach described in Section B1. Understory LAI was also estimated using the optical
approach described in Section B2. The total LAI at a point was estimated as the sum of
understory and overstory LAI. The LAI at the individual points in a plot were averaged to
estimate the plot LAI. The landcover characteristics (general description, dominant growth
forms and plant species, degree of urbanization) of each plot were recorded. The plot level
was the primary unit for comparison with satellite imagery.
3A. Digital imagery LAI estimation
Overstory LAI was estimated using the hemispherical photography (HP) technique that has
been widely used to estimate canopy LAI. We used smartphone spherical panorama photos
(SPP) to obtain HP images based on the approach recently developed by Arietta (2022). The
SPP HP approach has several advantages compared to the traditional HP approach including
lower cost, less variability due to operators, and higher accuracy (Arietta 2022). There are
three main steps to estimate LAI with this approach: 1) obtain SPP image, 2) transform SPP
image into a binary upward (or downward) HP image, 3) calculate LAI using a canopy gap
model.
The SPP images were captured using the “create photosphere” feature of the google “street
view” app on Android smartphones. As described by Arietta (2022), taking a good spherical
panorama requires keeping the center of the smartphone at a single point while moving the
screen around in different directions. The SPP images were stored on the cameras and
downloaded daily to an archive. The date and location (latitude and longitude) were
recorded and embedded in the image. The 600 SPP images collected during the field study
were transferred to a computer for further processing.
The GIMP image editor (downloaded from www.gimp.org) was used to extract two binary
HP images (upward looking and downward looking) from each of the 600 SPP images to
generate a total of 1200 binary HP images. This was accomplished with the following steps:
1) import spherical panorama photo JPEG images obtained using smartphones
2) crop the upper (or lower) half of the rectangular photo sphere,
3) scale the image into a square (height is same size as width),
4) remap to a polar projection
5) extract blue component (optimal for green vegetation)
6) generate binary image
7) export as TIF image (suitable for import to GLA program)
The Gap Light Analyzer version 2.0 (GLA2.0, downloaded from
https://www.sfu.ca/rem/forestry/downloads/gap-light-analyzer.html) imaging processing
software was used to characterize forest canopy structure and gap light transmission indices
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-14
to quantify effective Plant Area Index (PAIeff) from the binary HP digital images generated
with the GIMP image processing. This was accomplished with the following steps:
1) import binary HP image TIF file
2) configure and register image (ensure that it is centered)
3) set threshold
4) select canopy structure
5) calculate LAI
6) record displayed LAI results for ring 4
One sided green LAI was calculated as described by Fang et al. (2019)
LAI = (1- a) PAIeff / ΩE
where “a” is the woody-to-total-plant-area ratio, used to represent the contribution of
woody material to the total canopy area, and ΩE is the element clumping index, which
quantifies the effect of foliage clumping within shoots. Values of “a” and ΩE for each
growth form (broadleaf trees, needleleaf trees, shrubs, grass) were based on literature
values summarized by Fang et al. (2019), Beland and Baldocchi 2021 and Brown et al.
2021.
3B. Optical sensing LAI estimation approach
Understory LAI was measured at each point using a Trimble GreenSeeker handheld
Normalized Difference Vegetation Index (NDVI) sensor. The sensor measures reflected light
at two wavelengths, Red light at 660 nm and Near Infrared (NIR) at 780 nm, and calculates
NDVI as
NDVI = (NIR – Red) / (NIR + Red)
NDVI has been widely used to estimate LAI from satellite optical measurements. The
GreenSeeker sensor must be positioned 2 to 4 feet from the foliage. This is easily
accomplished for understory grass, forbs and shrubs but is difficult to achieve for tall trees
as it requires access to a location several feet above the top of the tree canopy. For this
reason, the optical LAI measurement was conducted only for understory vegetation.
Literature reports of LAI to NDVI relationships (e.g., Trezza et al. 2013) were used develop
an equation to estimate one sided green LAI from NDVI as
LAI = 0.16e0.0424 NDVI
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-15
Appendix 4: Generating MEGAN growth form fraction
distributions using high resolution imagery (e.g., NAIP)
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-16
Appendix 4. Generating MEGAN growth form fraction distributions
using high resolution imagery (e.g., NAIP)
Summary
This appendix section provides a tutorial on ArcGIS procedures for calculating growth form fraction
distributions (e.g. tree cover, grass cover, shrub cover) using high resolution imagery than enables
mapping at the level of individual trees and incorporating the results into the MEGAN growth form
cover fraction landcover input files. The three main steps are:
• Step 1. Generate high resolution tree cover fraction maps for selected areas.
Approach: Use ArcGIS tools to process ultra-high resolution imagery (e.g. 60 cm NAIP) for
selected region.
Product: High resolution growth form fraction maps for selected areas.
• Step 2. Assess accuracy. Return to step 1 if necessary.
Product: Accuracy estimates.
• Step 3. After achieving desired accuracy, integrate growth form fraction data into MEGAN global
1 km landcover database.
Approach: Use ArcGIS tools to reproject, aggregate and integrate growth form data into 1 km
global grid.
Product: MEGAN global growth form cover fraction files (tree, shrub, crop, grass).
Step 1. Mapping individual urban trees using high resolution imagery and ArcGIS
There are four main tasks required for this approach to mapping trees and other land cover using high
(60 cm or less) spatial resolution using ArcGIS. This includes 1) acquire the imagery, 2) create an
object-based image (this groups neighboring pixels together based on their similarity to create objects
that are then used for the classification) the objects include trees, buildings, lawns, cars, etc 3)
compile a database of training samples to identify specific landcover types (e.g., oak trees, juniper
trees, water bodies, laws, buildings, roads, etc) within the area, 4) classify all of the objects in the
image.
1. Obtain the high-resolution imagery for the targeted region.
The example given here is for using NAIP imagery which is available for all the contiguous US with
new imagery available approximately every other year. The ArcGIS Portal Living Atlas contains NAIP
data in an integrated file. Use the “USA NAIP Imagery: Color Infrared”. This is a 3 band dataset that
includes two visible light (blue and green) and one near-infrared (NIR). It works well for identifying
vegetation in most landscapes by may not be optimal for all landscapes so you could also try “USA
NAIP: Natural Color” which is also three bands but has red visible light instead of infrared. CLIP the
national NAIP database to generate an image of the target region. With 64 Gb Ram, an intel i9 chip or
better, and an SSD drive you should be able to do an area of several thousand km2 (note that this is
~10 billion points for 60 cm data). Output the clipped image (Right click on image, select data, export
raster) to a TIF file that will be used for the processing described below. If a specific year and month
is desired (NAIP is available about every 2 years and can be taken in different months of the year
which might help with identifying certain vegetation types) the data can be accessed by right-clicking
on the base “USA NAIP Imagery: Color Infrared” file and selecting individual NAIP files (~ 7 km x ~7
km). NAIP files can also be downloaded from location such as https://datagateway.nrcs.usda.gov/.
You can mosaic the individual tiles into one file for processing using ArcGIS “Mosaic to new raster”.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-17
2. Use ACRGIS Geoprocessing tool “Segment Mean Shift” to segment and
create an image with discrete objects (e.g., individual features such as
buildings, trees, lawns, lakes).
Select the TIF file generated in step 1 and then go to Imagery> Classification tools > Segmentation
(or Geoprocessing tool “Segment Mean Shift”). Settings that generally work well for heterogeneous
urban landscapes are: spectral detail= 18, spatial detail=18, minimum segment size = 50
Other settings may be optimal for other specific landscapes.
An example of a small (800 m x 800 m) object based segmented image based on NAIP color infrared
(NIR blue and 2 visible bands) is shown in A4-1.
Figure A4-1. Object based segmented image based on 60 cm NAIP visible and IR
bands. The figure shows a golf course with trees (dark red), grass (pink),
buildings, roads and trails (grey/bluish/white) in this false color image.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-18
3. Use ArcGIS to create an ESRI Classifier Description (.ecd) file
ESRI Classifier Description files are used to classify images. To create one, first select the segmented
object image file generated in step 2 and then go to Imagery> Classification tools > Training Samples
Manager. Create a schema (or upload an existing one) that includes the targeted landcover types that
are present within the image. To add classes to a new schema, right click on “new schema”. This may
include several types of trees (e.g., live oak, juniper, pine, maple, oaks, other deciduous trees, etc),
several ground cover types (e.g., shrub, green grass, dry grass, crops, bare soil), and several non-
vegetated types (e.g., dark water, clear water, pavement, colored buildings, rock). Populate each
landcover class with training samples by first selecting the class and then using the “segment picker”
to select “objects” in the segmented object image that represents the landcover type (i.e, select live
oak and then click on a live oak tree in the image). Ground cover and non-vegetated types will also
need training samples. Breaking down total tree cover into individual species or subtypes requires
obtaining geolocated features (points or polygons) representing each tree type and uploading them to
ArcGIS. They can then be used within the training samples manager to identify trees representing the
targeted types. The features used for training specific tree types can be obtained from geolocated
municipal tree inventories (available for some cities such as Houston and Austin) or from surveys that
can be conducted in situ in the field or virtually using Google Earth and Google Street Map.
After obtaining sufficient training samples (usually dozens of each landcover type but could require
more) they can be used with Imagery> Classification tools > Classify with classifier= Random Trees
(default values). The “Random Tree” classifier works well for some urban areas but other settings may
work better for other landscapes. This task may require some iteration. Use ArcGIS “Tabulate Area”
to calculate the total area covered by each landcover type. After an initial classification is generated
and assessed to determine if landscapes are accurately assigned landcover types, this step can be
redone to generate additional training areas for any landcover types that are incorrectly classified.
4. Use ArcGIS “ClassifyRaster” Function to classify segmented object images (generated
by step 2) using ESRI Classifier description (.ecd) file (generated by step 3).
An ESRI Classifier description file and a segmented image are used to generate a classified image with
various types of trees and other landcover. The RECLASS function generates a numeric growth form
fraction file at 0.5 m resolution. An example of the resulting 0.5 m resolution landcover distribution
data is shown in Figur and can be compared with the natural color image shown in Figure A.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-19
Figure A4-2. Landcover map (60 cm resolution) of the golf course shown above in
false color showing trees (various types in different shades of green), grass
(yellow) and bare soil (brown), buildings and pavement (white, grey and red).
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-20
Figure A4-3. NAIP natural color image (60 cm resolution) of the golf course shown
in the previous figures.
5. Use ArcGIS “Reclassify” to set numeric values to cover types.
The landcover classes created in section 1.4 can be converted to numeric values using ArcGIS
“Reclassify”. For example, if landcover types 1,2, and 3 are trees, then a numeric tree cover map can
be created by assigning a value of 1 to each of those categories and a value of 0 to all others.
Indeterminate classes (e.g., shadows) may be assigned a value between 0 and 1.
Step 2. Assess accuracy
Trees and other growth form distributions can be assessed using geolocated observations of individual
trees and other vegetation types. These data can be generated in several ways including:
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-21
1) i-Tree: this program uses random sampling of hundreds of points on a google Earth image
where the user identifies objects at random locations within a specified zone which can be user
defined or geopolitical boundaries such as city or county boundaries.
2) Available geolocated tree inventories and other growth forms. These are becoming increasingly
available for city street trees. The inventory should include latitude, longitude, growth form
(and tree species) and can be uploaded into ArcGIS as a CSV file. The “Display XY data” (right
click on file in standalone tables) is used to generate a layer feature class of points. Then the
“extract values to points” tool is used to find the classification value for each point. The
attribute table of the resulting file can be opened and copied and pasted into an excel
spreadsheet.
3) Conducting field surveys. Additional geolocated tree inventories and growth form data can be
generated with field surveys.
4) Available tree and vegetation cover summaries. Published reports provide average vegetation
cover fraction for some cities and other areas. This includes urban FIA, city reports, and other
publications.
Use ArcGIS “Zonal Statistics as Table” to calculate vegetation cover fractions for defined areas (such as city
boundaries) that can be assessed with the cover fractions obtained from the above sources.
Step 3. Integrate data into MEGAN global 30 second (~1 km) grid growth form
database.
The MEGAN global growth form database uses the CGL Service landcover products version 3 dataset. The
CGL data are available as an integrated global dataset for 2015, with annual updates after that, on global
scale (78.25N to 60S) at a resolution of 0.0099206 (~100 m at the equator) and accuracy of 80% when
compared to 28000 independent validation points. These data were generated using imagery from the
PROBA-V satellite and released in September 2020 (Buchhorn et al. 2020;
https://land.copernicus.eu/global/products/lc). The MEGAN Growth Forms include tree, shrub, crop, and
herbaceous with the remaining area considered to be barren of vegetation. The CGL tree cover fraction can
be underestimated in heterogeneous landscapes, including urban areas and arid woodlands. The higher
resolution (30-m) NLCD tree cover data can capture some of these missing trees and has been integrated
into the CGL data (i.e., the higher NLCD values are used for urban areas and woodlands). However, ultra-
high resolution imagery (e.g., the 60 cm NAIP) indicates that both the CGL and NLCD also misses 10 to
>50% of trees in the heterogeneous urban landscape. As shown in A4-3, the NAIP imagery can capture the
tree cover contributed by isolated individual trees. Urban development can also lead to rapid changes in
time (e.g., deforestation for housing developments) which should be considered, for example, if comparing
2015 CGL data with 2018 NAIP data. There could also be an increase from tree planting efforts although
that would typically take longer.
MEGAN (30 second resolution = 0.0083 degree cell size) and CGL (3.57 second resolution = 0.000992063
degree cell size) data are on a lat/lon grid while NLCD (30 m resolution) and NAIP (60 cm resolution) are on
the Albers_Conical_Equal_Area grid. The 60-cm NAIP data can be set to the NLCD 30-m grid using ArcGIS
“Aggregate” (cell factor of 30). This can then be converted to a lat/lon grid with 1.2 second resolution
(0.0033 degree cell size) using ArcGIS Resample set to “nearest neighbor”. The data can then be set to the
MEGAN 30 second resolution (0.0083 degree cell size) using the ArcGIS Aggregate tool with a Cellfactor of
25. The domain (in ArcGIS environment) needs to be set so that it covers a latitude and longitude that
starts and ends on a 30 second grid value so that the grids will line up. This 30 second file can then be
integrated into the MEGAN global 30 second grid. This can be accomplished by integrating the entire image
or a subset such as within the boundaries of a city or for all of the urban locations within the image.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-22
Appendix 5: Estimating tree cover fraction and compiling
random tree locations using i-Tree
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-23
Appendix 5. Estimating tree cover fraction and compiling random
tree locations using i-Tree
This appendix describes how to use the i-Tree tool to generate a random tree location database and to
estimate area average tree cover fraction (and fractions of other cover types) for individual cities and
other regions. The results can be used to assess the tree cover estimates generated using the
approach described in Appendix 2 and to generate a database of random tree geolocations that can be
used for the Virtual Urban Tree Survey approach described in Appendix 6.
Step 1. Initiate project
The i-Tree tool can be accessed at canopy.itreetools.org. Detailed tutorials are provided at
www.itreetools.org/support. The first step is to either start a new (i.e. for a new city) project
(“Project” > “File” > “New/Start/over”) or continue with an existing project by loading the project
from “project” > “file” > “open” and skipping down to step 2. A project area can be defined by
selecting an administrative region such as a city or county or by drawing an area - either within i-Tree
or created in another program (e.g., ArcGIS) and imported as a shapefile.
You will need to create a landcover scheme. A typical one would include the following seven classes:
Deciduous tree, Evergreen tree, Shrub, Grass and Bare Soil, Built (e.g., roads, buildings), Water,
Barren or import one that you have previously created and saved. Grass and Bare Soil may be
combined since what appears as bare soil in a winter image may appear as grass in a summer image.
Barren indicates a location containing bare rock, desert sand or other surface that appears unlikely to
have vegetation at any season.
Step 2. Identify landcover types and generate tree cover percentage and database
of randomly located trees
The i-Tree tool will randomly choose points within the selected domain and the operator will classify
the point as one of the landcover types in the landcover scheme. A typical minimum number of points
is 300. This can be tested by seeing if the percent tree cover values change if you add an additional
100 points. Save the project file often so you don’t lose your data and always save when you finish.
After completing all data points necessary and saving your project you can transfer the results (mean
and standard error for each cover class) to a spreadsheet. All of the data is stored in a project with a
name that you give it with a file type “.itrcnpy”. This file can be used to create a database of randomly
located trees that can be used for the Virtual Urban Tree Survey described in Appendix 4. This can be
accomplished by outputting all of the points into a spreadsheet and then removing all of the points
that are not trees. The file can then be used to create a KML file that can be imported into Google
Earth for the Virtual Urban Tree Survey.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-24
Appendix 6. Virtual urban tree survey
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-25
Appendix 6. Virtual urban tree survey
This appendix describes procedures for quantifying city average urban tree species composition that is
referred to here as the Virtual Urban Tree Survey (VUTS) approach. VUTS takes a dataset of trees that
are randomly distributed throughout a city, generated using the procedures described in Appendix 5,
and identifies them using a virtual tree identification key, described in Appendix 7.
Traditional approaches for characterizing tree species composition include random plots, tree
inventories, and tree community surveys. An example of the random plot method is the USFS Forest
Inventory Assessment (FIA) approach that uses ground surveys of 1/6 acre plots that are randomly
selected within each of the main vegetation types in a landscape that could be an urban area. FIA
selects one randomly located plot per ~3 square miles. Every tree in the plot is identified and the
diameter at breast height (DBH) of each tree is measured which enables an estimate of crown cover.
The FIA approach has been applied to most US forests including a few urban forests such as in
Houston, Austin and San Antonio. Tree inventories involve identifying every tree in an area and often
include DBH measurements. Urban tree inventories are becoming increasingly common, but they
typically include only trees on city property which generally is only ~15% of all trees in a city. Tree
community surveys identify the dominant trees in a landscape and so can provide a qualitative
estimate of the dominant trees in a city. All of these approaches require a survey team to travel to the
landscape being characterized. The VUTS approach was developed for this project as a low cost
alternative that does not require on site measurements. The VUTS approach was assessed by
comparison with random plot and tree inventory approaches.
VUTS is an extension of the i-Tree procedures (see Appendix 5) which are designed to quantify tree
cover and potentially broad tree type categories such as evergreen and deciduous trees. VUTS takes
this to the next level of BVOC emission types that can be identified using widely available Google
Earth imagery including aerial views and street views.
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-26
Appendix 7. Virtual Urban Tree Survey (VUTS) Wasatch
Front Urban Tree Key
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-27
Appendix 7. Virtual Urban Tree Survey (VUTS) Wasatch Front Urban
Tree Key
This appendix contains the taxonomic keys for identify trees using the VUTS tree identification
approach described in Appendix 4. A taxonomic key is a simple tool used to identify a specific object,
in this case to identify a tree. It begins by looking at large, important features that can divide the
possible answers into a few large groups, thus quickly ruling out most of them. The key shown in
Appendix 5A is designed for identifying trees at locations where a Google Street View image of the
tree is available. It considers a number of features that can be seen in a Google Street View image. If
a Google Street View image is not available for a specific location, then the aerial view key shown in
Appendix 5B can be used with a Google aerial image view. These are available for nearly all locations,
but image quality (resolution) can vary and the number of seasons with imagery can vary. This key
relies on features that can be seen from aerial images. The accuracy achieved with both keys can be
improved with availability of images from more than one season.
7A. Key for Wasatch Front trees for which Google Street View imagery is available
(i.e. the tree location is near a road with Google Street View imagery).
Utah Streetview Key – November 2022
1 leaves needle or scale like; evergreen
2 Leaves needle like; fruit a dry cone when mature
3 needles in clusters of 2 or more, leaves over 1 inch long…………………………………………………..……..…
Pinus
3 needles occur singly and < 1 inch
long……………………………………………………………………………..………....Picea
2 leaves scale like or awl-shaped
4 fruit berry like, tree like with wide
base………………………………………………………………….…………….Juniperus
4 fruit woody cone; often like a shrub with narrow base……………………………………………………………….Thuja
1 leaves broad and thin; usually deciduous
5 leaves opposite
6 leaves simple or with 3 leaflets
7 leave lobed, seed pods winged and wide………………………………………………………………………………………Acer
7 leaves not lobed, seed pods long and
narrow……………………………………………………………..…………..Catalpa
6 leaves compound with more than 5
leaflets………………………………………………………………………..….Fraxinus
5 leaves alternate
8 leaves simple
9 leaves lobed……….………………………….……………………………………………………………………………………….Group 1
9 leaves not lobed……………………………………………………………………………………………………………………..Group 2
8 leaves compound…………………………………………………………………………………………………………………….Group 3
Group 1
1 leave distinctly palmately lobed and veined; large 3 – 8 inches
wide……………………………………...Platanus
1 leaves not palmately lobed and mostly < 3 inches wide
2 leaf blades generally triangular in outline; small tree, fruit apple
like……………………………..…….Crataegus
2 leaf blades longer; often large trees; fruit an
acorn………………………………………………....………………Quercus
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-28
Group 2
1 leaf widest below the middle
2 leaf base not symmetrical; leaf tip gradually
pointed……………….………..…………………………..…..…..…..Celtis
2 leaf base symmetrical
3 bark light colored to white; lacking furrows
4 bark often very white; trunk with horizontal lenticles; small branches hanging
down……..………..Betula
4 bark light green; often large trees; branches horizontal to pointing upward.
……………..………….Populus
3 bark usually darker colored; often with lateral furrows
5 leaf blade heart shape
6 trees with light green bracts; medium size tree, tree often triangular shape- wider at
base and tapering to narrow
top….…………..………………………………………………………………………..………….Tilia
6 trees without bracts; showy flowers in spring; small
tree……....…………………...…………………………….Cercis
5 leaf bade more or less triangular; leaves often with some lobes; small
tree……….………………..Crataegus
1 leaf widest at the middle or above
7 leaf over 3 times longer than wide
8 leaves silvery color, short
petiole…..…………………………………………………………………………………....Elaeagnus
8 leaves not silvery; usually green or reddish
9 leaf lanceolate
.……………………………………………………………………….............……………………………………….Salix
9 leaf ovate to oblong
ovate…………………………………………………………………………………………………..…..Zelkova
7 leaf less than 3 times longer than wide
10 some leaves have lobes at base others without; leaves about as wide as
long………...…………....Morus
10 no leaves have lobes;
11 large trees, leaves with sharply serrated
margins…………………………………………………………………...Ulmus
11 moderate sized trees, leaves without sharply serrated margins
12 leaves often with a reddish tinge; bark of newer branches have horizontal
lines……………..…...Prunus
12 leaves usually greenish; bark does not have horizontal lines
13 Most branches are pointed straight up; fruit generally
elongated…….………..…………………………..Pyrus
13 Many branches are horizontal or not pointed straight up; fruit
round…..……………………………….Malus
Group 3
1 leaf odd pinnate (leaflet on end of leaf)
2 leaflets with large teeth or lobes at base; leaflets 1-4 inches long; flowers
yellow…………….Koelreuteria
2 leaflets entire, leaflets elliptical to oval; leaflets 1-2 inches long; flowers white or
pink………..….Robinia
1 leaf even pinnate
3 leaflets approximately 1 long; 15 to 30 leaflets per
leaf…………….……………………………..…………....Gleditsia
3 leaflets 2-6 inches long
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-29
4 approximately 10 leaflets per leaf; leaves approximately 10 inches
long………………………………...Pistacio
4 over 12 leaflets per leaf; leaves 12-30 inches
long...………………………….…………………..…………….Alianthus
Genera in Key
Acer Juniperus Prunus
Alianthus Koelreuteria Pyrus
Betula Malus Quercus
Catalpa Morus Robinia
Celtis Picea Salix
Cercis Pinus Thuja
Crataegus Pistacio Tilia
Elaeagnus Platanus Ulmus
Fraxinus Populus Zelkova
Gleditsia
Common Names
Apple – Malus Heaven Tree– Ailainthus Plum - Prunus
Arborvitae - Thuja Honeylocust – Gleditsia Pistacio - Pistacio
Ash - Fraxinus Juniper – Juniperus Redbud - Cercis
Birch – Betula Linden - Tilia Russian Olive – Elaeagnus
Catalpa – Catalpa Locust - Robinia Spruce - Picea
Cottonwood – Populus Maple – Acer Sycamore – Platanus
Elm - Ulmus Mulberry – Morus Willow – Salix
Goldenraintree - Koelreuteria Oak – Quercus Zelkova - Zelkova
Hackberry – Celtis Pear - Pyrus
Hawthorn - Crataegus Pine – Pinus
Notes: Pistachio used in this key is P. chinensis
7B. Key for Texas trees for which Google Street View imagery is not available (i.e.
the tree location is not near a road with Google Street View imagery).
Identification is based on Google Aerial View imagery.
Utah Aerial Tree Key 01/17/2023-Ver16
1 Trees evergreen - green or dark blue October – February - Group A
1 Tree deciduous - light green, colorful or without leaves October – February
2 Some colored leaves on 10/21- Group B
2 No leaves on 10/21 Group C
Group A Evergreem
1 tree color greenish in July – October
2 large tree, irregular shape, long needles in shadows………..………………………………....…….Pinus
2 smaller trees……………………………………………………..………………………..……..Juniperus
1 tree dark green or blue July -October; circular crown, shadow pyramid shape….………..…………Picea
Group B (Deciduous) Trees with some colored leaves in 10/21
1 Trees often large, irregular shape
Ramboll - Improved Vegetation Data for the Biogenic Emission Inventory of the Wasatch Front
Final Report
Appendix-30
2 Trees pale green to yellow -brown in 10/21………………………………..……………………Platanus
2 Trees yellow or orange in 10/21
3 trees orange in 10/21 and light green in 7/22…...………………………...…………...……………Acer
3 trees yellow to yellow green 10/21 and darker green in 7/22…………………………..………...Ulmus
1 Trees smaller, round shape…...……………………….…………………………..…………………Malus
Group C (Deciduous) Trees without leaves in 10/21
1 trees smaller, dark reddish/pinkish and blurry in 10/21……………………..…….………..……….Malus
1 trees often larger, lighter color in 10/21
2 trees with small branches that are whitish, tan, or pinkish making image fuzzy in 10/21
3 trees light white or tan colored or grey colored in
10/21…..............................................………Fraxinus
3 trees reddish or pinkish in 10/21, tree shape circular…………………………….………...………..Tilia
2 trees with few small branches making larger branches more visible and distinctive
4 some of the large and midsize branches are dark colored; August canopy not solid…..…….…Gledistia
4 most all of the large and midsize branches are light colored; August canopy solid
5 over 10% ground cover visible underneath canopy in October…………………………………….Acer
5 very little ground cover visible underneath canopy in October; yellow coloration…….……….Populus
Fraxinus = Fraxinus or Betula
Juniperus = Juniperus or Thuja
Malus = Malus or Pyrus or Prunus or Crataegus
Platanus = Platanus or Robinia
Ulmus = Ulmus or Catalpa or Salix
This key has approximately 82% of the genera that occur in the Salt Lake City Inventory