Purpose
The purpose of this page is to share discussion points or feedback
for approaches and interpretations.
Approaches
The relationship between redcedar and urban heat were investigated
with generalized mixed effect models with these data and analyses.
Findings
Overall, the probability of a tree having top dieback or being
classified as unhealthy increased with increases in temperatures above
average, but there was no evidence the probability of a tree being
classified as thinning is related to afternoon heat.
Evidence for increases in the probability of having top dieback with
increases in afternoon temperatures was found in Portland and King
County datasets, but not Tacoma.
- General results
- probability a tree was classified as unhealthy increased with urban
heat
- After noon temperature best fit
- More investigation into the importance of morning, afternoon, or
evening temperature would be valuable.
- No evidence for effect of co-factors
- Rural temps were higher than urban temps in king county
Questions of interest
- Initial Questions of Interest
- Does urban heat increase the probability of trees being classified
as unhealthy?
- which temperature time is most important (e.g. morning, afternoon,
evening, daily avg?)
- Does urban heat increase the probability of trees being classified
as dead?
- Are increases in urban heat associated with increases in percent
canopy dieback?
- Are percent poverty, HOLC rating, or environmental health disparity
ranks associated with increases in probability of trees being classified
as unhealthy?
- Do urban trees have higher proportions of unhealthy trees than rural
trees?
- Does tree size affect the tree’s vulnerability to urban heat?
Approach Considerations
- Aspects to consider
- Should we remove observations with very poor positional
accuracy?
- More investigation into the timing of heat?
- e.g. which temperatures (morning, afternoon, eve) are most important
for determining tree health?
- Check how dist from avg is calculated based on daily average
temperature
- Including dead trees?
- Current analysis does not include dead trees because
- there are a number of other factors that may have led to the dead of
those trees
- could not account for 1 inflation with glmmTMB package
- Beta logistic regression analysis does not include dead trees
because there are a number of other factors that may have led to the
dead of those trees
- Removing dead trees also removes the issue where we could not
account for 1 inflation with glmmTMB package
- Are there other random effects we should account for (e.g site type
or user?)
- Removing Hoyt Arboretum Trees?
- Caclulating Distance from mean after removing Hoyt
- should we remove observations with very poor positional
accuracy?
Eish, the portland dataset as is, only includes 15 trees with dead
tops.
Should we remove other tree health categories (e.g. other or
thinning) in analyses? for example, should we only include ‘dead top’
and ‘healthy’ trees when doing a dead top analysis?
Filtering trees to dead tops or healthy did not improve the
significance of the tacoma dead top model.
Why was there no evidence of a relationship in Tacoma?
- How to handle outliers?
- Accounted for currently
- removed outliers of trees with < -8 dist.from.mean.af
- Not accounted for currently
- check outliers for toxic release, whats up with those?
- check outliers with healthy trees with 100% dieback or with any %
dieback values
- Interactions?
- Heat and Pollution
- Does dieback increase with increases in heat
pollution?
- Heat and Poverty
- Are the impacts of urban heat on dieback worse for those in
poverty?
Analyses
of Portland data only indicate there is a relationship between tree
health and afternoon heat (DN_AF1) - non standardized. The probability
of a tree having top dieback increased with afternoon heat. However, no
evidence was found for a relationship between crown thinning and
afternoon heat. Perhaps there is a relationship with morning or
evening temperatures? However, it could also be that the observers
are not as confident in recognizing trees as thinning.
Data limitations
Preliminary analyses of data limited to those with HOLC
grades (n=395) may not have enough observations. For example, even
when distance from mean afternoon temp was included as a co-factor,
there was no evidence of a realtionship with heat.
Similarly, about 500 observations (oregon’s) were dropped for
analyses limited to observations with EHD data (n=787). Testing
hypotheses with EHD data may be best accomplished by including the full
dataset, rather than those shared only from King County or Tacoma.
Removal of Hoyt Trees Removal of Dead Trees
Data QA QC
Eight other tree species were added to the project, but the
iNaturalist community helped id the trees correctly.
How many identifiers were there? How many observers were there?
Additional Data
- Home Owners Loan Corporation Maps were downloaded from the Mapping
Inequality Project at the following links:
Data for the below metrics were included in the data, but not all
were investigated in the analyses
yet.
- Available data
- iNaturalist Data
- Tree health conditions
- Percent Dieback
- Number of additional unhealthy trees
- Tree size
- Site type (urban, suburban, rural)
- Location description (roadside, park or yard, forest edge, inside
forest)
- Summer access to water
- Urban heat
- HOLC Grade
- EHD Data (see V2.0
Report for more details)
- Indicators in Pollution Burden
- Environmental Exposures
- Diesel exhaust
- Ozone
- PM2.5
- Toxic Release
- Proximity to heavy traffic roadways
- Environmental Effects
- Lead risk from housing
- Proximity to hazardous waste treatment storage
- Proximity to superfund sites
- wastewater discharge
- Indicators of Population Characteristics
- Sensitive Populations
- Death from cardiovascular disease
- Low birth weight
- Socioeconomic Factors
- No high school diploma
- Unaffordable housing
- Transportation expense
- Limited English
- People living in poverty
- Race (people of color)
- Unemployment
References
- Voelkel, J., and Shandas, V. (2017). Towards Systematic Prediction
of Urban Heat Islands: Grounding Measurements, Assessing Modeling
Techniques. Climate 5, 41. doi: 10.3390/cli5020041.