patrickbdevaney/mia9 · Datasets at Hugging Face

text stringlengths 0 6.44k
both predetermined and random locations where people tend to
congregate (such as parks, bus stops, playgrounds, street intersections) and recorded the site-specific characteristics at the deployment site (such as full vs open canopy cover, or the material
on which the sensor was placed). The date and time when each
sensor was deployed and retrieved was entered on mobile forms
completed by participants. Participants were encouraged to
complete the mobile form at the sensor’s exact location because
the form uses a digital tool that picks up the GPS (latitude/
longitude) location of the submission site. In addition to this,
they documented the iButton location by providing both a
close-up picture of the sensor as well as a contextual picture of
the area. These photos were later used to ensure proper retrieval of the sensors and to validate the site-specific attributes
associated with each iButton deployment location, including approximate height of the sensor, which we estimate varied from 0.5
to 2 m. The form also provided a box for notes where participants
864 JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY VOLUME 62
Unauthenticated \| Downloaded 02/29/24 07:14 AM UTC
could explain their observations of the site in more detail, often
offering further insight on the location and its characteristics.
The range of temperatures recorded by iButtons (Thermocron
models DS1922L-F5#, DS1921H-F5#, and Hygrochon model
DS1923) was evaluated against a National Institute of Standards
and Technology (NIST)-calibrated thermometer in a laboratory
oven, recording at 1-min intervals, resulted in an average absolute error among sensors of 3.68F (28C). The sensor accuracy is
60.98–1.88F (0.58–1.08C). The hourly temperatures were downloaded using Thermodata Viewer software and aggregated, analyzed, and displayed using R Studio 4.1.1 software and ArcPro
2.8.0. Values above the temperature range of the sensor were removed from the set of observations as were anomalous values recorded from sensors that coincided with photodocumentation
indicating the sensor may have been exposed to direct sun
(artifact of improper placement). In the analysis, the first recording used was the recording closest to (either at or after) the listed
deployment time. If only a date was given, then the first day
used was the day after the listed deployment date, since the time
of actual deployment on the listed date was uncertain. In a few
instances, the deployment time was listed by participants as
“12:00 a.m.” [0000 local time (LT)], when the first recording was
after “01:00 a.m.” (0100 LT), and in these cases, it was assumed
that the participant meant “12:00 p.m.” (1200 LT); in these instances, the first recording accepted in the analysis was at 1200 LT.
For retrieval, the last recording used in the analysis was the recording closest to (either at or before) the listed retrieval time. If
only a date was given, then the last day used was the day prior to
the listed retrieval date, since the time of actual retrieval on the
listed date was uncertain. If no date or time was listed, the day
FIG. 1. Map of Miami–Dade County showing the locations of the MIA temperature measurement
and the iButton sensors. The tree canopy is overlaid for reference (Hochmair et al. 2022).
JULY 2023 CLEMENT ET AL. 865
Unauthenticated \| Downloaded 02/29/24 07:14 AM UTC
prior to the earliest date/time of retrieval of another sensor during
that period’s deployment was used. The data contain the description of each sensor deployed, and both their listed deployment/retrieval dates and the first and last dates used in the analysis.
Individual files based on unique identification (ID) numbers were used for analysis. For the hourly analysis, in any
iButton that measured both temperature and relative humidity, the heat index was calculated (https://www.wpc.ncep.
noaa.gov/html/heatindex_equation.shtml). Then, for all the
files, an hourly mean resample was used to take the mean of
the observations recorded during that hour. For the daily
analysis, a daily minimum, mean, and maximum resample of
the hourly resampled data was performed. Postprocessing was
then performed on the final hourly and daily resampled files
to only keep the variables that would be used in the analysis.
Hourly climatological data from Miami International Airport
(MIA) for the years 1950 to 2021 were downloaded in decadal
comma-separated values (CSV) files from NOAA’s Local
Climatological Data website (https://www.ncei.noaa.gov/products/
land-based-station/local-climatological-data). The data used here
are from the Miami Airport National Weather Service ASOS
station, though conclusions are the same if we use other ASOS
stations in the area. The downloaded files were then manually
processed to keep hourly or subhourly observations and remove any daily or monthly observations as well as any variables
excluding time, temperature, and relative humidity. These processed decadal files were then merged to a master file to group
all the years from 1950 to 2021. An hourly mean resample was
then performed on this master file, and then a daily minimum,
mean, and maximum resample of hourly values concluded the
postprocessing.
Scatterplots with a simple linear regression line were created
by first adding columns of hourly MIA observations matching
the time of sensor observations to each individual sensor file.
Then the updated individual files were combined based on classification (sunny or shaded). The combined sunny files were
then plotted as iButton variable versus MIA values, and using
simple linear regression, a line of best fit was drawn through
the points. The same was performed for shaded files. An analysis was then performed by taking an equation of the line of best
fit from the scatterplots for a specific variable and applying it to
the 1950–2021 MIA values. We then use this regression to reconstruct daily maximum heat index values based on the MIA
values from the past 70 years.
3. Results
The iButton sensors were deployed in a range of different conditions that can significantly impact temperature and humidity
observations including under open (sunny) and treed (shaded)
canopy (Fig. 1). Averaging over hyperlocal conditions, recorded
temperatures with iButton sensors matched daily fluctuations in
minimum temperature when compared with MIA data (Fig. 2).
Average maximum daily temperatures, however, well exceeded
MIA temperatures, and this variation was present in both sunny
and shaded locations (to varying degrees), and during each of
the times of the year sampled (Fig. 2).
While the conditions at the iButton sites had, on average,
higher maximum temperatures, there was a large range of values
across the sites. For scatterplots of all daily iButton Tmin and
Tmax, the variation in Tmin is smaller than Tmax, and does not
depend much on the absolute temperature (Figs. 3a,b). Tmax
also departs more from the MIA values at higher temperatures
for sunny locations than shaded locations, though both have a
warm offset at all temperatures. On average the maximum temperature from the iButtons was 78F (3.98C) warmer in sunny locations than the MIA values with a standard deviation of 68F
(3.38C), and 68F warmer in shaded locations with a standard deviation of 7.58F (4.28C).
A subset of the iButtons also recorded humidity, allowing us
to compute the heat index (see the methods section, section 2).
Figure 3c compares the heat index from iButtons versus MIA.
Here we see that both sunny and shaded locations have heat
indices well above the MIA value. On average the maximum heat index from the iButtons was 118F (6.18C) warmer

both predetermined and random locations where people tend to

congregate (such as parks, bus stops, playgrounds, street intersections) and recorded the site-specific characteristics at the deployment site (such as full vs open canopy cover, or the material

on which the sensor was placed). The date and time when each

sensor was deployed and retrieved was entered on mobile forms

completed by participants. Participants were encouraged to

complete the mobile form at the sensor’s exact location because

the form uses a digital tool that picks up the GPS (latitude/

longitude) location of the submission site. In addition to this,

they documented the iButton location by providing both a

close-up picture of the sensor as well as a contextual picture of

the area. These photos were later used to ensure proper retrieval of the sensors and to validate the site-specific attributes

associated with each iButton deployment location, including approximate height of the sensor, which we estimate varied from 0.5

to 2 m. The form also provided a box for notes where participants

864 JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY VOLUME 62

Unauthenticated | Downloaded 02/29/24 07:14 AM UTC

could explain their observations of the site in more detail, often

offering further insight on the location and its characteristics.

The range of temperatures recorded by iButtons (Thermocron

models DS1922L-F5#, DS1921H-F5#, and Hygrochon model

DS1923) was evaluated against a National Institute of Standards

and Technology (NIST)-calibrated thermometer in a laboratory

oven, recording at 1-min intervals, resulted in an average absolute error among sensors of 3.68F (28C). The sensor accuracy is

60.98–1.88F (0.58–1.08C). The hourly temperatures were downloaded using Thermodata Viewer software and aggregated, analyzed, and displayed using R Studio 4.1.1 software and ArcPro

2.8.0. Values above the temperature range of the sensor were removed from the set of observations as were anomalous values recorded from sensors that coincided with photodocumentation

indicating the sensor may have been exposed to direct sun

(artifact of improper placement). In the analysis, the first recording used was the recording closest to (either at or after) the listed

deployment time. If only a date was given, then the first day

used was the day after the listed deployment date, since the time

of actual deployment on the listed date was uncertain. In a few

instances, the deployment time was listed by participants as

“12:00 a.m.” [0000 local time (LT)], when the first recording was

after “01:00 a.m.” (0100 LT), and in these cases, it was assumed

that the participant meant “12:00 p.m.” (1200 LT); in these instances, the first recording accepted in the analysis was at 1200 LT.

For retrieval, the last recording used in the analysis was the recording closest to (either at or before) the listed retrieval time. If

only a date was given, then the last day used was the day prior to

the listed retrieval date, since the time of actual retrieval on the

listed date was uncertain. If no date or time was listed, the day

FIG. 1. Map of Miami–Dade County showing the locations of the MIA temperature measurement

and the iButton sensors. The tree canopy is overlaid for reference (Hochmair et al. 2022).

JULY 2023 CLEMENT ET AL. 865

Unauthenticated | Downloaded 02/29/24 07:14 AM UTC

prior to the earliest date/time of retrieval of another sensor during

that period’s deployment was used. The data contain the description of each sensor deployed, and both their listed deployment/retrieval dates and the first and last dates used in the analysis.

Individual files based on unique identification (ID) numbers were used for analysis. For the hourly analysis, in any

iButton that measured both temperature and relative humidity, the heat index was calculated (https://www.wpc.ncep.

noaa.gov/html/heatindex_equation.shtml). Then, for all the

files, an hourly mean resample was used to take the mean of

the observations recorded during that hour. For the daily

analysis, a daily minimum, mean, and maximum resample of

the hourly resampled data was performed. Postprocessing was

then performed on the final hourly and daily resampled files

to only keep the variables that would be used in the analysis.

Hourly climatological data from Miami International Airport

(MIA) for the years 1950 to 2021 were downloaded in decadal

comma-separated values (CSV) files from NOAA’s Local

Climatological Data website (https://www.ncei.noaa.gov/products/

land-based-station/local-climatological-data). The data used here

are from the Miami Airport National Weather Service ASOS

station, though conclusions are the same if we use other ASOS

stations in the area. The downloaded files were then manually

processed to keep hourly or subhourly observations and remove any daily or monthly observations as well as any variables

excluding time, temperature, and relative humidity. These processed decadal files were then merged to a master file to group

all the years from 1950 to 2021. An hourly mean resample was

then performed on this master file, and then a daily minimum,

mean, and maximum resample of hourly values concluded the

postprocessing.

Scatterplots with a simple linear regression line were created

by first adding columns of hourly MIA observations matching

the time of sensor observations to each individual sensor file.

Then the updated individual files were combined based on classification (sunny or shaded). The combined sunny files were

then plotted as iButton variable versus MIA values, and using

simple linear regression, a line of best fit was drawn through

the points. The same was performed for shaded files. An analysis was then performed by taking an equation of the line of best

fit from the scatterplots for a specific variable and applying it to

the 1950–2021 MIA values. We then use this regression to reconstruct daily maximum heat index values based on the MIA

values from the past 70 years.

3. Results

The iButton sensors were deployed in a range of different conditions that can significantly impact temperature and humidity

observations including under open (sunny) and treed (shaded)

canopy (Fig. 1). Averaging over hyperlocal conditions, recorded

temperatures with iButton sensors matched daily fluctuations in

minimum temperature when compared with MIA data (Fig. 2).

Average maximum daily temperatures, however, well exceeded

MIA temperatures, and this variation was present in both sunny

and shaded locations (to varying degrees), and during each of

the times of the year sampled (Fig. 2).

While the conditions at the iButton sites had, on average,

higher maximum temperatures, there was a large range of values

across the sites. For scatterplots of all daily iButton Tmin and

Tmax, the variation in Tmin is smaller than Tmax, and does not

depend much on the absolute temperature (Figs. 3a,b). Tmax

also departs more from the MIA values at higher temperatures

for sunny locations than shaded locations, though both have a

warm offset at all temperatures. On average the maximum temperature from the iButtons was 78F (3.98C) warmer in sunny locations than the MIA values with a standard deviation of 68F

(3.38C), and 68F warmer in shaded locations with a standard deviation of 7.58F (4.28C).

A subset of the iButtons also recorded humidity, allowing us

to compute the heat index (see the methods section, section 2).

Figure 3c compares the heat index from iButtons versus MIA.

Here we see that both sunny and shaded locations have heat

indices well above the MIA value. On average the maximum heat index from the iButtons was 118F (6.18C) warmer