Remote Sensing Methods for Urban Forest Planning

Remote Sensing Methods for Urban Forest Planning

Cities account for less than 3 percent of the global land surface but are home to

over 55 percent of the global population. The physical changes associated with

urbanization have altered the form, structure, and composition of the earth’s land

surface, affecting its thermal properties, energy balance, and biogeochemical cycles---

influencing the function of ecosystems and climate. Cities themselves are physically

diverse environments, demonstrating intraurban and interurban variability in terms of

their material composition and configuration, thermal characteristics, and ecological

capability, across a broad range of spatial, temporal, and geographic scales. (Upmanis

(McMichael et al., 2006; United Nations Development Programme, 2006; Padowski &

Gorelick, 2014; Liu et al., 2017). The design and implementation of effective

adaptation and mitigation strategies, not only require a complete knowledge and

understanding of the unique structure and function of constituent subcomponents, but

of their interrelationships. This involves the widespread and comprehensive collection

of data of the urban landscape, at a granular level (Masser, 2001; Künzer & Dech,

2013; Tang & Li, 2014; Luderwitz et al. 2015). Many of the negative consequences of

urbanization are both local and global in effect; however, our understanding of the

processes, impacts, and their linkages across time and space is incomplete.

There is scientific consensus that the modifications to land use and land cover

associated with urbanization, have caused changes to the local microclimate of many

(Gilbert, 1989; Heidt and Neef, 2008). These climatic differences are the result of

complex, diverse, and often collective interactions between the natural and built

environment.

Analytical methods for studying the role of forests in the urban environment

include applications remote sensing data and technology, and geospatial analysis.

Remote sensing is the acquisition of information about an object or phenomena that is

done without making physical contact with it. Remote sensors collect data of the

physical characteristics of an area or object by detecting its reflected or emitted

electromagnetic radiation at a distance from the targeted area. Remote sensing

represents the only practical means to acquire large areas of data of the earth’s surface

platforms: 1) ground based, 2) airborne, and 3) satellite. The platform chosen for a

particular sensor is determined by several factors including: the distance of the sensor

Remote sensing-based solutions can be used to promote more sustainable

urban forestry management practices that lead to more resilient forests in the future.

Remote sensing data and technology can provide critical information for urban

forestry management, such as monitoring tree growth, identifying areas of forest

stress, and tracking forest cover changes. This approach can help urban foresters to

tree height, allowing for more accurate estimates of forest biomass and carbon storage

(Beland et al., 2018). This information can be used to identify areas where tree growth

is lagging and prioritize interventions to improve tree health.

Another application of remote sensing in urban forestry management is

identifying areas of forest stress. Remote sensing data can be used to detect changes in

tree health, such as leaf chlorophyll content, water content, and photosynthesis rates

(Zhao et al., 2016). This information can be used to identify areas of forest stress, such

as those affected by drought or insect infestations. Early detection of forest stress can

help urban foresters to take proactive measures to mitigate the effects of stress on

forest health.

Remote sensing-based solutions can also be used to track forest cover changes

and identify areas of forest loss. Satellite imagery can provide information about

changes in forest cover and fragmentation over time. This information can be used to

identify areas where forest loss is occurring and prioritize conservation efforts (J.

Zhang et al., 2016). Additionally, remote sensing can be used to identify areas with

potential for reforestation or afforestation, promoting sustainable urban forestry

practices.

The most challenging aspect for most urban remote sensing applications,

relates to the spectral heterogeneity and morphological complexity of the urban fabric

resolution imagery (e.g., from manned or unmanned aircraft; or from satellites such as

IKONOS, QuickBird, Worldview) provides new opportunities to map detailed urban

areas at a fine scale by providing more detailed observations of the Earth (Cao et al.,

2010; Labchoon et al., 2017). The diversity of scale and shape of the structural

elements of the urban environment contribute to its morphological complexity. A

sensor system with a very high spatial resolution is necessary to capture the

morphological properties of urban features more accurately. An increase in spatial

resolution significantly expands the range of the detected spectrum and the observable

fundamentally composed of groups of pixels with similar spectral signatures (Voogt &

Oke, 2003; Zhou et al., 2019).

these ecosystems more effectively. Remote sensingbased solutions offer a cost-

effective and efficient means of gathering detailed spatial and temporal information

over large areas that would otherwise be difficult or impossible to obtain using

ground-based field survey methods. In the context of this research, remote sensing

data and technology are used to provide critical information for the effective

environmental planning, managing, and monitoring of urban forests. An overall goal

of this research is to improve on the knowledge and understanding of the intricate

dynamics of urban forest ecosystems and develop new tools to assist with their

management.

To achieve these goals, this three-part investigation introduces new approaches

1. Apply new approach for mapping and evaluating the effects of UTC shade on

surface temperature at multiple local scales. Show that tree shade metrics

2. Use maximum likelihood HLM estimation to evaluate the contextual and

climatic effects of neighborhood-level land use factors on local-scale average

surface temperature of sidewalks within individual nested census block groups

Level 2 variables that best explain variance in mean sidewalk surface

temperature.

3. Review the current state of practice and knowledge related to the use of

Unmanned Aircraft Systems (UASs) in urban forestry. Ascertain the efficacy of

using UASs for common urban forestry tasks, such as species inventory, tree

health monitoring, and outbreak identification. Determine whether UAS-based

approaches might be both technically feasible and cost-competitive with field-

based methods.

Remote sensing data and technology are used to provide critical information

for planning, managing, and monitoring urban-forest environmental systems. Each

Chapter Summaries

Given the significant impact of urban tree canopy (UTC) on local climates,

there is an urgent need for metrics that encompass the structural and functional

qualities of UTC, not just canopy coverage. This study introduces a method for

detailed quantification and mapping of tree shade across urban areas, illustrating its

effectiveness in identifying microclimates and predicting urban heat on various spatial

scales. Our aim was to analyze spatial and temporal patterns of UTC and its shade,

determining the best metric for predicting urban heat by studying its correlation with

land surface temperature in Denver, Colorado, at two scales: a coarse

assessment using Landsat 8 data with 30-meter resolution and a fine-scale analysis

management.

Research

Questions

Intellectual

Merit

underscores the necessity for refined urban planning strategies, particularly in the

context of the Urban Heat Island (UHI) effect. This study examines the significant

influence of material composition and spatial configuration on urban climate

dynamics, focusing on the role of impervious surfaces and urban greenery in

modifying urban thermal environments. It delves into the nuanced role of urban trees

and other vegetation in regulating climate, emphasizing their strategic use to lessen

solar loads on buildings and improve comfort in public spaces. By modifying land

cover and land use, urban planners can influence microclimates within cities,

mitigating the UHI effect through various scales of intervention—from the types of

surfaces in urban designs to the density and arrangement of green spaces.

complex interactions between microclimate and the green and grey components of

urban infrastructure. Sidewalks within census block groups were selected to serve as a

of urban fabric have resulted in a diversity of methods for measuring thermal effects

both within and between cities. For example, the effects of land cover on LST can be

There are still knowledge-gaps regarding the specific heat reduction

performance of different types of vegetation interventions and interactions between

green and grey infrastructure in various cities at different spatial scales (i.e. local

versus city scale). The preponderance of past research shows that vegetation cooling

effects is predominantly driven by shading (Hoelscher et al., 2016; Speak et al., 2020;

that shading accounted for 60 % of the total cooling effect of urban vegetation and

Cameron et al. (2014) found that shading contributed up to 80 %. McPherson et al.

(1988), showed dense shade on all surfaces of homes reduced peak cooling loads 31-

49% depending on the climate. This is supported by the work of Akbari et al. (1997)

illustrating the impacts of tree shading on residential energy consumption, which

demonstrates that trees can reduce surface temperatures (within the shaded area) by as

much as 20–45 °F (11–25 °C). Research from Brown and Gillespie (1995)

Most previous research on the cooling effects of urban tree canopy (UTC)

shade were mainly conducted at a small spatial scale and focused on a few buildings

spaces in Tel-Aviv, Israel. Yu et al. (2020), used geospatial and empirical modeling to

simulate and measure the cooling effect of UTC shade, and evaluate its ability to

explain the variation in land surface temperature (LST), for New York City, NY and

Urban tree canopy (UTC) cover is an important metric for estimating overall

urban forest benefits, including cooling effects (Nowak & Greenfield, 2012). UTC

However, coverage metrics only capture the 2-dimensional cover of tree crowns,

which cannot accurately represent the distribution of projected shade, as it lacks the

UTC coverage metrics are usually obtained with remote sensing, including

imagery generated from LiDAR data and/or high-resolution digital aerial photographs

or satellite images. The accuracy of UTC simulation and quantification has steadily

improved with the use of 3D

LiDAR data and the application of object-oriented classification techniques advanced

by the University of Vermont Spatial Analysis Laboratory. The advantage of LiDAR is

spectrum provides information on reflected longwave radiation needed to compute

land surface temperature. The signal is inversely proportional to the evapotranspiration

of UTC and can provide an indirect estimate of the surface energy balance and plant

water stress (Delogu et al., 2018). TIR data have been used extensively to document

the effects of vegetation cover and structure on surface temperature and emissivity

over an urban area (Weng et al., 2004; W. Zhou et al., 2011; Coutts et al., 2016; X. Yu

et al., 2017; W. Li et al., 2017). The most widely accepted range for TIR remote

sensing in vegetation studies is 3–14 μm, divided into the MWIR (3–5 μm) and the

LWIR (8–14 μm). The MWIR receives both emitted radiation from Earth and reflected

radiation from the Sun, while the LWIR is dominated by the emitted component

(Jensen & Hardin, 2005). Urban green spaces and trees are known to interact with

solar IR radiation in a way that is dramatically different to the way they deal with

types, i.e., asphalt, porphyry, and grass.

Aircraft Systems (UASs) remote sensing techniques, emerges as a promising solution,

offering a way to bypass the limitations of traditional remote sensing methods by

providing high-resolution, cost-effective data collection (Colomina & Molina, 2014;

Banu et al., 2016). Stability, safety, control, reliability, and autonomy of UAS,

combined with the specificity of the sensors, are the factors that allow successful

implementation of this new platform for remote sensing of urban forests.

Despite recent innovations, the use of UAS and very high resolution TIR imagery to

quantify the thermal effect of shades projected by urban trees remains an

underexamined application.

This research demonstrates a methodology for the comprehensive

quantification and spatial mapping of tree shade across an entire urban area,

showcasing its utility in delineating urban microclimates and predicting urban heat

across a continuum of spatial scales. The objective was to quantify and map the spatial

and temporal patterns of urban tree canopy and urban tree canopy shade, and to

evaluate which coverage metric is the better predictor of urban heat by examining its

resolution of the UAS-collected imagery, both spatially and spectrally, allows

researchers to precisely measure the cooling effects of UTC shade on the surface

temperature of different urban land cover, specifically, turf grass and impervious

(1) What is the relationship between remotely sensed measures of land surface

(3) How are these relationships influenced by the specific spatial scale used for

METHODS

Analytical Approach

The fundamental modeling approach is to establish a statistical relationship

between microclimatic variation in land surface temper temperature and the

heterogeneous urban environment characterized by various 2D and 3D representations

of different types of land cover, and the shade cast by buildings and trees for a

specified location, date, and time. The methodology for both the coarse-scale and fine-

scale analyses are comprised of the same three primary steps including: 1) the creation

The LST data for the coarse-scale analysis study area was obtained from

Landsat 8 Thermal Infrared Sensors (TIRS) Analysis Ready Data (ARD), with a

weather research objectives and the limited temporal availability of Landsat 8 thermal

imagery. Landsat 8 data has been widely used across the world in urban surface

temperature research (Park et al., 2021). The ARD images are post-processed to the

highest scientific standards required for direct use in land cover change assessment.

Landsat uniquely provides large-scale land surface temperature (LST) data with each

30-meter pixel representing an average of the local thermal environment. Very high-

resolution land cover and shade variables are generated at a granular scale of 0.5 to 1

The LST data for the fine-scale analysis was acquired with UAS on September

4, 2019, under cloud-free conditions. Specifically, a DJI Matrice 600 hexacopter

Micasense Altum sensor as payload, distinguished by six spectral bands (Blue, Green,

Red, Red Edge, Near Infrared, Thermal), was used for the collection of very high-

resolution thermal and multispectral imagery, at 41 centimeters (16.14 inches) and 2.6

centimeters (1.02 inches), respectively. The September 2019 mission had an earlier

start time of 9:30 AM. Ideally, UAS missions are flown between the hours of 11 AM

and 1 PM, when the sun was at its apex to optimize spectral reflectance, maximize

were used for subsequent radiometric correction of the multispectral images. Ground

control points (GCPs) were used to correct the absolute spatial accuracy of the

multispectral orthomosaics with a typical accuracy of 2 cm in the horizontal direction.

The very high-spatial resolution of 41 centimeters (16.14 inches) for the thermal band,

became the determining factor in the sample design for the fine-scale analysis.

Similarly, very highresolution land cover and shade variables are generated at a

granular scale of 0.5 to 1 meter, but instead of their values being aggregated to a 30-

Study Area and Sampling

Denver is situated at the eastern edge of the High Plains, adjacent to the Front

Range of

diurnal temperature variation throughout the year. Historically, in Denver’s prairie

ecosystem, trees were largely absent until planted for landscaping purposes during

urbanization. The Denver urban forest would not exist, nor would it persist without the

aid of deliberate human intervention, in the form of planting, maintenance, and

irrigation.

The study area for the coarse-scale assessment is a reduced 28610-hectare

(286.1 km2) portion of the total 40135-hectare (401.35 km2) area for the entire City

and County of Denver. We excluded Denver’s recently annexed airport zone,

consistent with previous UTC analyses conducted for the city of Denver (Noble et al.,

2019), given that this zone is almost treeless. The data sample for the coarse-scale

random and used to extract the spatially coincident information from the underlying

thematic data layers representing the city’s land cover and shade variables. The full

quadrant of the park was the designated study area (Figure 1). This zone is estimated

to house 260 trees representing 31 different species, consisting predominantly of

Green Ash (Fraxinus pennsylvanica), Blue Spruce

the LittleLeaf Linden (Tilia cordata).

The data sample for the fine-scale assessment was assembled using 1000

random point features. GIS was used to generate the random points to fall within the

specified extent of City Park study area, such that each point had minimum nearest-

neighbor distance of 10 meters and the underlying areas intersecting UTC crowns

were excluded as suitable locations for potential sample points. The thematic data

random points, which were used to extract the spatially coincident variable

information which lay directly below.

2D Land Cover Variable Creation

The surface cover for both the coarse-scale and fine-scale assessment was first

measured using pre-classified land cover data with a very high-resolution of 0.5 to 1

meter. For the CoarseScale assessment, these variables are subsequently aggregated as

areal metrics to the 30-meter resolution fishnet sample grid. This integrated dataset

effectively reflects the combined influence of land cover diversity and distribution

within the scope of a single sampled Landsat LST pixel. In contrast, the independent

predictors of surface temperature for the fine-scale assessment were coded as binary to

assessment. The areas of UTC Shade modeled for the 9:30 AM time-step were found

to be a near perfect match for the observed areal footprints of the shadows cast by City

Park trees in the natural color version of the UAS-collected thermal-multispectral

UTC Area (VIF = 5.56), but not for Turf Area (VIF = 2.92) and Impervious Surface

(VIF = 2.90). Preceding regression analysis, Spearman’s rank order correlation was

High Shade Area (Table 3). This collinearity between variable pairs prohibited the use

of model configurations that combined the use of both types of variables as predictors

within the same model. Ultimately, multimodal inference was used to determine

individual variable importance by comparing the predictive performance of different

of Low Shade and High Shade, also containing

Area (Model 3).

for impervious surface and turf from the second OLS regression equation. A VIF value

of 4 is typically considered a key threshold for collinearity.

Next, the model was assessed for spatial autocorrelation. A distance-based

spatial weights matrix was created in GeoDa (Anselin, 2003) and the model was

assessed for spatial dependency with respect to that matrix. The resulting diagnostics

included a Moran’s I value of

0.2439 (p=0), indicates the presence of a moderate level of positive spatial

autocorrelation. This value suggests that there is a tendency for similar values to be

located near each other more than would be expected by chance, but the clustering is

not extremely strong. Next, we tested for lag versus error term dependency in the

model using Lagrange Multiplier tests (Anselin & Rey, 1991). The spatial error model

while the spatial lag model assumes that there is systematic autocorrelation due to

spillovers in the dependent variable (Anselin, 1988). Lagrange Multiplier and Robust

the Lagrange tests on lag (Robust LM lag= 70.3, p=0; Robust LM error= 11. P=0),

indicating the need to model spatial lag.

There are several forms of spatial lag regression from which to choose. The

most common is the Simultaneous Autoregressive (SAR) model (Eq. 2), which

assumes the existence of spillover effects in the dependent variable such that the value

of yi influences yj which in turn influences yk and so on, with the result that a marginal

change in the observational unit can have a cascading impact on the outcome variable

in all other observational units.

spillovers in the dependent variable. In this case, we hypothesize that it is likely due to

misclassified instances of land cover type among the independent variables

representing different surface cover.

While SAR does not explicitly model the direct impact on the dependent

variable in observation i of independent variables for i’s neighbors, as the so-called

“Spatially Lagged X” (SLX) and “Spatial Durbin” models do, it does account for

neighboring independent variable impacts indirectly, as those variables influence the

dependent variable of the neighbors, which in turn influence the dependent in the

interpreted in the same way as those in ordinary least squares (OLS) regression.

Rather, a special table of spatial spillover impact effects is needed to interpret the

coefficients’ so-called direct, indirect, and total effects. The first of these represents the

marginal effect of each x on y within a given spatial observation, i; the second

OLS and SAR regressions were conducted in R 4.2.1 using the LAGSARLM

required in order to generate the “impacts” table. We conducted regressions where the

effects of both UTC and turf land cover variables were modeled (and omitted as

control variables) separately in two different equations. Additionally, while it was not

possible to include the spatially correlated UTC and tree shade variables in the same

models for the Coarse-Scale assessment, this was a possibility for the finescale

spatial lag of the dependent variable can help to stabilize the variance and reduce the

multicollinearity impact on coefficient estimates.

Nonetheless, including variables that are perfectly collinear would still be problematic.

Consequently, it is still important to examine the variables for collinearity and

RESULTS

positive relationship with the dependent variable for LST, Mean Surface Temperature

(Table 5). The remaining variables, UTC Area, Turf Grass Area, Mean Shade-Hours,

Low Shade Area (<=4 hours), and High Shade Area (>=5 hours), demonstrated a

negative relationship with Mean Surface Temperature (Table 5). All variables in all

models were significant at the 99% confidence level. Because data were

untransformed, each coefficient represents the marginal increase or decrease in

degrees Celsius of surface temperature associated with a one squaremeter increase in

the area of that variable.

Model 1 assessed the effects of the High- and Low-Shade Area variables by

including the two predictors in the same model, along with Turf Grass and Impervious

Area. Recall, that the collinearity between UTC Area and all UTC shade variable pairs

prohibited the use of model configurations that combined the use of both types of

variables as predictors within the same model. The coefficient of High Shade Area was

the largest in magnitude for Model 1, indicating that a single square meter of High

Model 1 (0.4707) was the largest, indicating that it explained the most variance in LST

with 47.1%.

Model 2 assessed the effects of the Mean Shade-Hours variable, in addition to

interpretable variable, since it represents a sum of areas averaged across the time they

are shaded. This contrasts with the other explanatory variables, which represent area

only. Still, the pseudo R2 value for Model 2 (0.4702) was the second largest, indicating

that it explained 47.0% of the variance in LST.

Model 3 assessed the effects of the UTC Area variable, in addition to Turf

Grass and Impervious Area. Like shade, the effect of UTC Area offset that of

Impervious Area. By comparison an equivalent unit of UTC Area offsets 60% of

Impervious. A comparison of Model 3 with Model 1 using AIC, indicate the model

with High Shade Area and Low Shade Area variables but no UTC Area variable

(Model 1), outperformed the model with UTC Area but no shade area variables

(Model 3) in Denver (AIC= 137707 for 1, and 137772 for 3). The lower AIC score for

Model 1 suggests that even though the model’s AIC is penalized for having one more

variable than Model 3 (i.e. the former has variables for High Shade Area and Low

Shade Area while the latter only has UTC Area), the increase in model fit more than

makes up for the loss in model parsimony (Akaike, 1978, 1979; Burnham &

Anderson, 2003).

SAR 1

Near Distance 0.133 0.038 3.534 0.00041 **

UTC Shade -5.454 0.238 -22.920 0.00000 ***

UTC 1.736 0.405 4.290 0.00002 **

SAR 2

Estimate Std. Error z value Pr(>|z|) Sig

0.00041 **

Near Distance 0.133 0.038 3.534

UTC Shade -5.454 0.238 -22.920 0.00000 ***

AIC: 5259, (AIC for lm:5438)

LM test for residual autocorrelation: 0.030, p-value: 0.8617

Table 8: Fine-Scale Assessment SAR Results of Direct, Indirect, & Total

Impacts

0.138

2.710

-9.320

-2.966

0.00056

0.00000

0.00000

0.00001

0.00170

0.00000

0.00000

0.00009

0.00075

0.00000

0.00000

0.00001

Unshaded

has become a key component of Denver planning policy. Recently, city crews have

(Weng et al., 2004; Chen et al., 2006; Yuan & Bauer, 2007; S. Li et al., 2010; W. Li et

al., 2017; Y. Liu et al., 2017), the impacts of tree shade on the urban thermal

This research supports the idea that a more nuanced approach to urban

greening is necessary. Simply planting more trees may not be sufficient for effective

heat mitigation in urban areas. Urban planners should focus on strategic shade

planning. It suggests that the two-dimensional metric of Urban Tree Canopy (UTC)

coverage is inadequate as it doesn't capture the threedimensional distribution of shade

and its dynamic nature with the sun's movement. There is a need for very high-

resolution (pixel area <= 1m) spatial data to accurately map, measure, and compare the

impact of UTC and UTC shade on urban LST, and cities could leverage modern

remote sensing techniques using LiDAR and thermal infrared imagery for this

purpose.

Understanding the specific heat reduction performance of different types of land cover

and shade variables can lead to more targeted and effective UHI mitigation strategies.

This research consists of two separate investigations conducted at different

spatial scales: 1) a Coarse-Scale assessment, and 2) a fine-scale assessment. In the

Coarse-Scale assessment, we introduce new methods for modeling and evaluating

hourly UTC shade at a local spatial scale of 30 meters, for an entire city, and compare

the heat predicting performance of the UTC shade metrics to that of UTC coverage.

This portion of the analysis relied solely on remote sensing data collected by

shade variables is quite limited, due to the high degree of spatial correlation between

UTC and its projected shade effects at a 30-meter scale. Further, models show that the

effect size (regression coefficients) of the High Shade Area variable was larger than

In the fine-scale assessment, the analysis methods were applied to a very high-

resolution dataset acquired with UAS and used to model and evaluate hourly UTC

shade at a microscale of

41 centimeters, to compare the heat predicting performance of the UTC shade metric

explanation of the variance in LST. The added capability of collecting hi-res

thermographic imagery significantly enhanced this phase of analysis, enabling

thermal impact of tree shade effects is supported by the statistical rigor of the results

from the SAR models implemented in this fine-scale phase of analysis. The results of

the SAR models confirmed the superior power and heat mitigating impact of tree

We believe this research demonstrates the value and feasibility of explicitly

pedestrian routes, outdoor gathering places, and safe routes to schools. That

information can in turn be used to target new green investments. The need for such

Nonetheless, this methodology does entail some limitations. One is that we

measure heat using surface, and not air temperature, which would require extensive in

situ monitoring data. Further, the surface temperature associated with UTC crowns is

representation of urban temperatures, in lieu of highresolution atmospheric

temperature data which does not yet exist. Ideally, such a study would be based on

atmospheric data, but conducting such an analysis would require a massive data

acquisition effort to measure atmospheric temperatures.