Entry Name: “CerebriAI-Wu-MC1”
VAST Challenge 2019
Mini-Challenge 1
Team
Members:
Wenjie Wu, Cerebri AI, Inc, wenjie.jessie.wu@gmail.com PRIMARY
Zheng Zhou, Cerebri AI, Inc, zhengzhou.purdue@gmail.com
Yingjie Chen, Purdue University, victorchen@purdue.edu
Zhenyu Qian, Purdue University, qianz@purdue.edu
Student Team
No
Tools Used:
D3.js, React.js, Django, MySQL
Approximately how many
hours were spent working on this submission in total?
80
May we post your submission in the Visual Analytics Benchmark
Repository after VAST Challenge 2019 is complete?
Yes
Video
https://va.tech.purdue.edu/vast2019/mc1/HeatMosaic.mp4
URL
Questions
1. Emergency responders
will base their initial response on the earthquake shake map. Use visual
analytics to determine how their response should change based on damage reports
from citizens on the ground. How would you prioritize neighborhoods for
response? Which parts of the city are hardest hit? Limit your response to 1000
words and 10 images.
If based on the earthquake shake map, the first
responders would put higher priority on areas that are close to the epicenter. However, the provided earthquake shake map did not come with a timeline,
so it failed to provide the changing patterns of three earthquakes and update
the damages. Figure 1
clearly show the patterns and changes of the reported damage metrics from Safe
Town (Area 4) in the three phases of the earthquake – fore-quake,
main shock, and after-quake.
·
The first earthquake,
a.k.a fore-quake, from April 6th 14:30 to
April 6th 21:00, was light and it caused not much damage.
·
The second earthquake, main-quake,
was the most serious one and it caused a number of
damages and presumable casualties in most areas, happened between April 8th
8:30 to April 8th 17:00.
·
The third one was an after-quake
also caused some damages and medical needs, happened approximate between April
9th 15:00 to 19:00, made condition worse in some area.
Figure 1
Report category heatmap from Safe Town (4) describing three earthquakes: fore-quake,
main-quake, and after-quake
During the fore-quake period (April 6th,
14:30 ~ April 6th 21:00), from Figure 2 we can see that:
·
Area 3, 8, 9, 10, and 19
needed more urgent response for building damage.
·
Area 3, 5, 6, 9, and 11 needed
higher priority of medical response.
·
Area 3, 9, and 11 needed
higher priority power damage attention.
·
The road damage was reported
high in area 3, 8, 9, 10 and highest in 13.
·
Sawyer and water leakage
happened in area 3, 18, and 19.
Overall area 3 was the highest priority for first
response during fore-quake period.
Figure 2 Reports aggregation for 19 locations during 1st earthquake (fore-quake) between 13:30 and 19:30 on April 6th
From Figure 3 (main-quake)
we find that area 3, and 4 experienced strongest shake, followed by 7, 12, 14, and 18. The overall shake
intensity appeared high in most areas and the reports were much more than from fore-quake.
Figure 3 Map view of the shake intensity
during the main quake
Figure 4 Reports aggregation for 19 locations during 2nd earthquake (main-quake) between 08:00 and 14:00 on April 8th
It’s notable that power outage occurred in area 3, 8, 9, 10, and 11
because there has been no incoming report over a time span of ~12 hours for
area 3, 8, and 9, ~20 hours for area 10, and ~2 hours for area 11, respectively
(Figure 4).
Figure 5 Power outages in area 10, 11, 3, and 9 after 2nd earthquake (main-quake)
Meanwhile after these blank-report period we found a spike on the report
values. We infer that such spike was due to the reports that people sent during
these periods finally reached the server thanks power got restored (Figure 5).
For responding, the highest priority would be for area 10, 11, 3, then 8
and 9 as their power outage (10, 11, 3) started about 40
minutes (09:05 4/8) after the main-quake (~08:35 4/8) and the outage in area 8
and 9 began in ~4 hours (~12:30 4/8) after the main-quake. Except power outage,
area 3 also had high severity reports on almost all other issues when it
started. Besides, area 1, 7, 14, 17, 18 reported power issues but no outage
(secondary priority).
Building-wise, all major areas reported damages. Area 9 and 8 was the
most severe while several others (1, 3, 10, 11, 14, 18) also needed attention.
We received major medical reports from areas where hospitals are located
(1, 3, 6, 9, 11, 16). We also assume there’s a medical facility in area 5 due
to similar pattern, but the city’s introduction did not mention it. If no
medical facility, area 5 would need immediate medical response. Area 8
has the most medical reports, followed by 4, 14, 15, 18. These areas would also
need medical help as we can see a few serious reports. Medical team may
need to send out to these areas with higher priority on area 8.
Area 8, 9 has the most road and bridge damages, followed by 3, 14, and
17. Resources might need to repair these
road damages.
Area 8, 9, and 14 reported serious water and sewer damages which would
require prioritized response. Also 3, 16, 17 are the next ones need attention.
Overall area 8 would need highest priority of first response.
It’s also notable that the damage reports from area 1 were obviously
diverted, including building, medical, power, and water. Considering area 1 is
in a “not-felt” zone, as well as the doubtful uncertainties (see answer to
question 2), we do not suggest list area 1 as top priority even it reported
high values.
Figure 6 Reports aggregation for 19 locations
during 3rd earthquake between 14:00 and 20:00 on April 9th
From Figure 5 above we
can see after-quake caused area 3 and 8 lost power immediately. Area 3 experienced
the longest 21-hour power outage from 15:00 April 9th to 12:00 April 10th.
Area 8 and 14 also had a power outage. But at the middle there was a short
period that power resumes supply in area 14 (see Figure 6).
Figure 7 Power outage pattern in Easton (14)
after after-quake
From Figure 7 below we can also find area 14, then 4, then
12, then 17 experienced power outages starting ~02:00, ~7:00, ~10:00, and ~16:00
on April 10th, respectively.
Figure 8 Power outage in Easton (14), Safe
Town (4), Pepper Mill (12), and Oak Willow (17) in hours after the after-quake
For the after-quake, area 16 has many reports at
two modes. Many medium level reports as well as low severity reports. Area 13
appeared the same polarized pattern on power, road and bridge, and water and
sewing.
Most areas had less severity reports during after-quake.
But area 4’s condition got worse during after-quake. Therefore area 4 needed
more resources during the after-quake period (see Figure 8).
Area 8, followed by 3, 9, and 11 experienced
most severe water and sawing damage during after-quake, which would require
higher attention (Figure 6).
For road and bridge, area 3, 8, 9, 10 had more damage
during after-quake (Figure 6).
Building wise, area 3, 8, 9, 10 also had the most severe
reports (Figure 6).
2.
Use visual analytics
to show uncertainty in the
data. Compare the
reliability of neighborhood reports. Which neighborhoods are providing reliable reports?
Provide a rationale for your response. Limit your response to 1000 words and 10
images.
For visualizing the uncertainty, we created a strip
chart (Figure 9): the middle
line of the strip is the mean value of each categories every hour, and the strips
on the line are the reports. The height of the strips is twice of the standard
deviation of that time calculated hourly. The color of each strip is a gradient
heat color. Cool color means low severity, and warm color means high severity. The
strips are semi-transparent (opacity: 0.1), so that when there are a lot of reports
coming up in a short amount of time, the strips will overlay with each other
and create a more solid area. Therefore, when we look at the strip chart, the
wider the ribbon is, the larger the uncertainty is; solid area means large
amount of reports, and the color of the ribbon indicate the severity.
Figure 9 expample
of strip chart
Overall, area 2, 3, 6, 8, 9, and 15 have more
wide and solid ribbons during all the categories and all the range during the
peaceful time, which are recognized as high uncertainty. It seems that those
are areas with larger population and their data have more noises. Figure 10 gives
an example of large-population area 6 versus small-population area 13. Higher
population might lead to higher density of buildings and more old buildings. We
can observe that in area 3, 6, 8, and 15, the damages on the buildings during the
aftershock were more severe than the mainshock.
Figure 10 6-downtown as an example of higher
population and noises, and 13 – Cheddarford as an example of lower population
and less noises
Areas
13 and 16 has polarized pattern through every category (Figure 6), which caused higher uncertainty during shake time. It
may because the area cross two geological condition.
Area
1 is not reliable especially at the beginning time of the main earthquake. High
severity reports suddenly disappeared. Also, these reports do not consistent
with shake reports (Figure 11).
Figure 11 Palace hills has a pattern that a
group of people reports high damage during the main earthquake, but not in the
other two earthquakes
Due
to power outage, area 3, 9, 10, 11 were not reliable at the time (Figure 5). It is possible to derive status during the power
outage. But for area 3, it is impossible since the power outage was too long.
Areas 2, 5, 6, 15, 19 were complete and
reliable. The values were not polarized, aligning with normal distribution (Figure 12 shows area 5).
Figure 12 Location 5 Southwest as an example of
complete and reliable areas
Area 7 is not reliable since it has a small
number of reports. Area 10 also has much less reports than other areas, but
enough to present its conditions. Hence it is reliable (Figure 13).
Figure 13 Strip Chart of location 7 Wilson Forest and 10 Chapparal
Reporting period of area
12 is much longer (~6 hours) than other areas. When reports from other areas
already calmed down, reports still arise from area 12 (see Figure 14).
Figure 14 Pepper Mill (12) reporting damage
condition delayed
3.
How do conditions change over time? How does
uncertainty in change over time? Describe the key changes you see. Limit
your response to 500 words and 8 images.
Conditions
in the area changed drastically over the time of the three earthquakes. Fore-quake
was not severe, and it did not cause much damage. Main quake caused power
outage in area 3, 8, 9, 10, and 11. After-quake caused less damage, for
example, area 9 in Figure 15. But area 4 had more severe damage
during after-quake (see Figure 16).
Figure 15 Damage condition in Broadview (9)
less severe during after-quake than during main-quake
Figure 16 Damage level arisen in Safe Town
between main-quake and after-quake
Uncertainty reduced over time, for example, in area 1, as the polarized
pattern disappeared (see Figure 17).
Figure 17 Uncertainty changing over time in
reports from Palace Hill (1)
In other areas, generally in the time periods between earthquakes
(normal time), reports had a bigger variation range, but graduate decreased to
a narrow range (in earthquake), for example, from Figure 18 we can see that in normal time the downtown area
generates a lot of noisy reports which had large variance. During each earth
quake, the noise got decreased for a great deal and reports’ reliability increased.
Figure 18 Overview Downtown (6) reports
For areas 4, 12, 13, 14, 15, 18, and 19 which
has two or more poles on shaking reports, in the after-quake, they will switch
to single mode and be more trustable. (Figure 19 shows the uncertainty condition change in East Parton (18))
Figure 19 East Parton uncertainty change
For a prolonged power outage, although all reports were collected
at the end and can be used to derive the status during power outage, the
temporal details are still missing. For example, in area 3, we cannot see when
the report values decreased (the middle image in Figure 4).
4.
The data for this
challenge can be analyzed either as a static collection or as a dynamic stream
of data, as it would occur in a real emergency. Describe how you analyzed the data - as a
static collection or a stream. How do you think this choice affected your analysis? Limit your
response to 200 words and 3 images.
1. Static collection allows us the see the whole picture. But from
Dynamic stream, we cannot foresee data after current time. The data after could help to check the
reliability of current data. For example, in Area 1, at the beginning period of
mainshock between 8:30am to 10:30am 04/08, there were a lot of severe reports
happened. From a streaming report, we should send out first responder this area
1. But after a short while, these reports disappeared. We doubt these reports
were sent out in panic.
Figure 20 Overview Palace Hills reports
2. As static collection, we can see the max number of reports. We
can easily use this number as a base to decide the color scale of heatmap. For
stream analysis, we do not know the possible largest number. In our design, we
adopted a dynamic color.
3. For areas with power outage, in streaming mode, during the power
outage period, we may assume their severity would remain the same as when they
started to report. However, for some areas, e.g. area 3 and 8, when power came
back, the accumulated messages have higher values. It seems the severity of the
damage was increasing during the power outage period. Streaming mode may under
estimate the severity.
Figure 21 Overview Old Town (3) and Scenic
Vista (8)
4. Typically, after-quake should be less than the mainshock and
therefore less damage. Areas 15, 16, 17 etc. are in this pattern. However, for
certain areas, e.g. 4 and 8, the damage is aftershock is severe than main-quake. If we use the main-quake reports in streaming
mode as a reference to predict the aftershock damage, we may underestimate the
severity.
5.
Figure 22 Condition worsen in Safe Town
from main-quake to after-quake while condition better in Weston