Entry Name: "Chengying-Hu-MC2"
VAST
Challenge 2019
Mini-Challenge 2
Team
Members:
Yueqi Hu, Chengying, hyq825@gmail.com, PRIMARY
Qi Ma, Chengying, heymarch@qq.com
Yang Chen, Chengying, chen1984yang@gmail.com
Haidong Chen, Chengying, chenhd925@gmail.com
Weiqing Jin, Chengying, jinmmd@gmail.com
Fenjin Ye, Chengying, yefenjin@csu.edu.cn
Student Team:
NO
Tools
Used:
Tableau
Js
Python
Approximately how
many hours were spent working on this submission in total?
We spent
600 hours on this submission.
May we post your
submission in the Visual Analytics Benchmark Repository after VAST Challenge
2019 is complete?
YES
Video
Questions
Your task, as
supported by visual analytics that you apply, is to help St. Himark's emergency
management team combine data from the government-operated stationary monitors
with data from citizen-operated mobile sensors to help them better understand
conditions in the city and identify likely locations that will require further
monitoring, cleanup, or even evacuation. Will data from citizen scientists
clarify the situation or make it more uncertain? Use visual analytics to
develop responses to the questions below. Novel visualizations of uncertainty are
especially interesting for this mini-challenge.
1 – Visualize radiation measurements over time from both
static and mobile sensors to identify areas where radiation over background is
detected. Characterize changes over time. Limit your response to 6 images and
500 words.
The normalized readings from all sensors are grouped into
16 geographical grids. Normalization details will be presented under question
2-a. Figure 1 shows the average radiation readings of each grid and within
every two hours.
Fig 1. The average radiation readings of each grid and
within every two hours.
From the overview. we can make following assumptions:
1: on 8th began at noon, the
radiation around unclear power plant begin to increase and back to normal at
6pm. During the next two days, the readings stayed higher during work hours
near the nuclear power plant.
2: start from 8th 4pm,
the readings at the gate of Jade Bridge increased during rush hours and back to
normal at other time of the day.
3: start from 2pm on the 8th, the
readings at East Parten increased and last till the end.
4: started from 8th’s
night, radiation increase at Broadview last until next morning.
5: Began at 4pm on 9th, the
readings at the gateway to Wilson Bridge increased dramatically and last until
10am the next day.
6: Around the same time, the readings at OldTown and
increase slightly and last until the end.
We then integrate this view with more details to verify
the assumptions above, as shown in figure 2, (a) Each row shows the normalized
readings of sensors from one grid, which is pinned on the map on left. (b)
radiation readings are scattered on the map to see their geo-relation. The size
shows radiation values using size. (c) each row shows readings from a sensor,
which is used to evaluate the reliability of related sensors. View a, b, and c
using the same color coding to discriminate sensors. (d) is the overview
discussed above.
Figure 3 shows the details around the radiation
increasing near East Parton at 1pm. From the grid detail view we can see two
sensors at this area increased while others stay the same. The sensor view
shows us that these two sensor function normally. We than compare the locations
of high readings and low readings. It shows the high readings are near the
highway 19. Therefore, we can conclude that radiation contaminated objects near
the highway.
Fig 3. The details around the radiation increasing near
East Parton at 1pm.
From the overview
Fig 4. The radiation readings at the Wilson Bridge.
Figure 4 break down the radiation readings at
the Wilson Bridge. From grid detail view, we can see that there are two
locations with high radiation. The one on the left starts earlier and stay the
same reading. And the readings at the bridge gate
This tool also reveal more trivial changes in each area.
For example, we find many short peaks in radiation
reading that last 1 to 5 minutes.
2 – Use visual analytics to represent and analyze uncertainty
in the measurement of radiation across the city.
a. Compare uncertainty of the static sensors to the mobile sensors.
What anomalies can you see? Are there sensors that are too uncertain to trust?
Figure 1 shows the distribution of the readings from each
sensor during the first day. The hour of the day is color coded using
red-white-blue. It shows the uncertainty of the all sensor without being
effected by the events afterwards.
Fig 1. The distribution of the readings from each sensor
during the first day.
It’s easy to see the static sensors are more stable
comparing to the mobile ones, as they show similar value distribution, small
variance and without outliers.
For mobile sensors, sensors 23, 29 has similar
distribution to the static ones. They can be considered more reliable than the
others. For other sensors, there are several aspects that contribute to their
uncertainty.
l Bias and
variance: The static sensors ensure us that the background radiation is similar
and stable across the city. Therefore, the variant of the median and width of
value distributions of each mobile sensor shows each mobile sensor has
different bias and variance, which should be considered individually in
normalization.
l Sensitivity:
Seven sensors (8, 9, 10, 11, 18, 19, 21) contain gaps in the reading, which
shows those sensors cannot measure the radiation with high accuracy.
l Outliers:
Ten Sensor (2, 4, 7, 10, 13, 16, 25, 33, 47, 48, 50) have readings way bigger
than others, which may due to occasional malfunctions of the sensors, which
will be excluded before normalization. Sensor 12 has readings lower than the
background, should be also considered as outliers.
l Location:
sensor 5, 16, 37 have two peaks, the readings from the working hours are much
smaller. Figure 2 shows that during the working hours those sensors are parked
close to each other at downtown area. Maybe they parked in the basement where radiation
tend to be lower than outside. We will also exclude these reading when doing
normalization. Sensor 28 gives higher readings when it parks.
Fig 2. The sensors are parked close to each other at
downtown area during the working hours.
l Device
characteristic: we also examming the uncertainty using scattered plot. And find
sensor 1, 2 and 27 have an interference
(Figure 3). Those sensors are considered unreliable.
Fig 3. The readings of sensor 1, 2 and
27 changes anomaly without outside impacts.
b. Which
regions of the city have greater uncertainty of radiation measurement? Use
visual analytics to explain your rationale.
The radiation of an area is more certain if
1) it has a static sensor which gives stable and
consistent radiation readings +0.3.
2) the roadmap in the region has a good coverage so that
the mobile sensors can have a good spatial coverage + 0.3.
3) mobile sensors have good coverage over time +0.3.
There are also other trivial aspects considered, which will discussed in the
following table:
The
known unreliable sensors learn from above are excluded in this question.
Spatial
coverage:
Figure 1 shows the number of radiation readings at
different place of the city. The static sensor are labelled in orange. It shows
regions of Pepper Mill, Oak Willow, and Palace Hills and west part of Northwest
has low coverage.
Fig 1. The location of all
readings reveal the spatial coverage of sensors.
Temporary coverage:
Figure 2 shows areas are only covered partials of
time, or only by very few sensors, including the bridges and banks (see
Figure 2). The top two regions have
reading. But each of them has only one which seem unreliable.
Fig 2. Some areas are only covered partials of time, or
only by very few sensors.
Furthermore, the resident neighborhoods don’t have
sensors during the daytime, such as Broadview and Northwest (see
Figure 3). The color strip under the reading shows the hour of the
day. The red and blue parts are the night and the grey parts are working hours.
Fig 3. The resident
neighborhood may not have sensors during the daytime.
To concluding the observation above, we create the follow
table to rank the uncertain of each neighborhood.
|
Region |
Score |
Static sensor coverage |
Mobile sensor Time coverage |
Mobile sensor Spatial coverage |
comments |
|
7 |
0 |
0 |
0 |
0 |
|
|
17 |
0.1 |
0 |
0.1 |
0 |
|
|
10, 11, 12, |
0.2~0.3 |
0 |
0.1 |
0.1 |
|
|
2 |
0.4 |
0 |
0.2 |
0.2 |
|
|
1 |
0.5 |
0.2 |
0.2 |
0.1 |
|
|
13 |
0.4 |
0.2 |
0.1 |
0.1 |
|
|
4 |
0.5 |
0.3 |
0.2 |
0.1 |
Mobile sensor 9 and 13 are sent to this area to capture the radiation. |
|
5 |
0.5 |
0.2 |
0.2 |
0.1 |
|
|
14,19 |
0.6 |
0.0 |
0.3 |
0.3 |
|
|
15,16 |
0.7 |
0.1 |
0.3 |
0.3 |
Close to static sensor outside the region. |
|
6 |
0.85 |
0.3 |
0.3 |
0.3 |
Cars parked in this area tend to have extremely low readings |
|
18 |
0.6 |
0.1 |
0.3 |
0.2 |
Static sensor on every road leading to this region |
|
9 |
0.9 |
0.3 |
0.3 |
0.3 |
|
For the gateways, C is
the more certain, as a static sensor is placed at its gateway, followed by A
and B for the same reason. F is more certain because there are more than 10
mobile sensors stuck there for a long time. All other outbound ways are high uncertain.
c．What effects do you see in the sensor
readings after the earthquake and other major events? What effect do these
events have on uncertainty?
Events:
The first earth quake took place on 8th at 8am.
Five sensors are broken down, as all their readings after stay same.
Other four sensor seems also broken down as them don’t
provide readings any more. After put them on the map, two of them actually left
the city.
On
the static sensors, temporary peaks, as automatically detect and labelled using
orange bubble appears only after the earth quake. That may because contaminated cars are
running on the streets now.
We also created an animated
map study and movement of mobile sensor right after earth quake. When a leak occurs, sensors 9, 13 and 32
come to rescue. Then when 9 and 32 went home to rest, 10 came up, and finally
32 came back to participate in the rescue. The sensor does not move passively
because of the nuclear leak, but also moves actively because of the human
nature.
There are other changes in uncertainty that happens
generally and cannot be associated with any single event. However, we can
capture these changes by comparing the first and last day.
The following figure counts the number of readings of
each sensor on the first day and on the last day. The static sensors are in
orange.
It shows that, other than sensor failure discussed above,
the remaining mobile sensor are more likely to absent due to two reasons.
First, after earth quake, more mobile sensors left the city during working
hours. Second, even available, the mobile sensors are more like to miss
samples, such as sensor 32 which opened whole day but only get 75% valid
readings.
Noise
The follow figure is created
using the same method of figure 1 except that the reading from 10th are added for comparison.
The events have similar
impact on static sensor but impacts on the mobile sensor are different.
Meanwhile, it
is easy to see both static and mobile sensor give more outliers after these event. But the mobile sensors are more like to
create readings lower than the baseline.
The coolant leak happened at the noon on the 8th. After that, Mobile sensor 9 moving to the plant and its reading was high. The sensors nearby reading also raised.
Leaving the city and come back:
3 – Given the uncertainty you observed in question 2, are the
radiation measurements reliable enough to locate areas of concern?
a. Highlight
potential locations of contamination, including the locations of contaminated
cars. Should St.Himark officials be worried about contaminated cars moving
around the city?
Through the observation of static sensors, almost all of
the radiation increased dramatically at 3pm on 9th. The biggest change can be
found at static sensor 9 and 12. As shown in the Figure 1.
Fig 1. The change of
static sensors.
Therefore, we can make assumption that the major
radiation leak took place at that time. We also find static sensor 12, which
located at the gate of Jade Bridge, shows a consistent high radiation readings.
By looking at the trajectories of mobile sensor of the same time, we find there
are lots of cars wanted to leave the city though the bridge but no one success.
Therefore, the high radiation may contributed by the contaminated cars also
stuck there.
So, we can conclude that the coolant leak happened before
4:40 p.m. on the 8th, so we need to observe in detail the changes of sensor
readings near the nuclear power plant before that. The nearest static sensor to
the nuclear power plant is s15, we studied in detail the s15 readings before
this time. After 4 p.m. on the 8th, the s15 readings showed a large value. At
the same time, we observed that the mobile sensor m9 was also nearby, so we
combined the data of the two. Looking at it, we find that over-standard
readings occur in the same place for the same duration. After half an hour,
these readings become normal again, and large-scale nuclear leaks usually last
for a long time, so we conclude that this situation is due to the nuclear
contaminated employee cars passed by this place after work.
(PS: The animation (.gif) needs to be opened using Google Chrome.)
Now, we can make sure that the coolant leak happed time
before 4 p.m. on the 8th. So we going on
studying the sensor readings nearby, and we found that 4 p.m. on the 8th is not
the first time s15 has a short time high readings, the same situation is
happened at 10:28 a.m. on the 8th before. And the
first time of contaminated car leaving the plant is 10:28 a.m. on the 8th.
In efforts to find the trajectory of the contaminated
cars, we conclude that the detection range of the sensors for the contaminated
cars is actually very limited. If it exceeds a certain range, it is basically
undetectable, which brings us great difficulty in locating the contaminated
cars.
In order to do this, we need to analyze which parts of
the city have a marked increase in reading after 10:28 a.m. on the 8th. We
divided the city into several areas. By visualizing all sensor readings in each
area, we found that there are two sensor reading modes that we should focus
between 10:28 a.m. on the 8th and the morning of
the next day. First, the short time high readings, which results from the
passage of contaminated cars through the area. Second, the long time high
readings, which results from contaminated cars were parked nearby (night more
likely). Like this, we can find a lot of high readings at the gate of jade
bridge, it’s combine of the two reading modes. That means some contaminated
cars (or one) try to leave the city by Jade bridge, but found the bridge is
stacked. Then the contaminated car came back.
The picture below show that there is a contaminated car
park near m42 in East Parton at 4:34 p.m. on the 8th and
leave at 9:45 a.m. on the 9th.
This is the reading mode for a contaminated car pass an
unmoving sensor.
By the way, some short high reading do not really reflect
the passage of contaminated cars, it just because an
uncertain mobile sensor come to this area, like the picture show below. In this
situation we need to use the original reading of this sensor to check whether
it’s just come to this area and it’s uncertain.
By finding these two modes after 10:28 a.m. on the 8th. We can
find this picture below, which mean where the contaminated cars once parked.
The points on the map show where the contaminated cars were detected for a long
time.
And this picture below, which means where once
contaminated cars passage nearby. The points on the map show where the
contaminated cars were detected for a short time.
The points on the map show where the contaminated cars
were detected.
The first picture show where the contaminated cars were
parked, and we can make it sure.
For the reason of uncertain for some area, we can’t make
sure where the contaminated cars which appear in the second picture finally
went.
The second picture show that the movement of contaminated
cars is not limited, and some of the area have a lot of population, it’s very
dangerous for the citizens. And as we will discussed next part of the question,
some contaminated cars even leave the city freely. So, of course, the officers need
to worry about the contaminated cars.
b. Estimate
how many cars may have been contaminated when coolant leaked from the Always
Safe plant. Use visual analysis of radiation measurements to determine if any
have left the area.
As mentioned before, it is very difficult to detect the
trajectory of contaminated cars through limited sensors, so some special method
is needed to estimate the number of contaminated cars. By studying the movement
of the mobile sensors, we found that they perform normally work and rest. This
situation was not broken until 5 p.m. on the 8th. Many people evacuated
directly to Jade Bridge in the north of the city after work. Normally, they had
already returned home or were on their way home at this time. This feature is
also applicable to the staff of nuclear power plants. Therefore, after the car
was contaminated, the last normal off-duty time happened on the 8th evening, so
we observed the sensor readings near the nuclear power plant in detail in the
afternoon of 8th. During this period, the available effective sensors included
s12, s13, s14, s15 and m9. M9 arrived here at 1:24 p.m. and kept a very high
reading until about 4 p.m. when the readings began to decline. After half an
hour, it returned to normal. During this period, M9 did not move. Therefore, we
speculate that it is the time when employees leave work that the
nuclear-contaminated car exits. Therefore, we only need to count the number of
peak readings of s15 in this period to figure out the number of cars left with
nuclear contamination. A total of 16 nuclear-contaminated cars left the plant
during this period, as shown in the figure. The first one left at 10:28 a.m. on
the 8th. The other 15 left after 4 p.m. on the 8th. And we
can get the number verified by the nearby static sensors.
As we mentioned before, the first evacuation at Jade
Bridge did not success, but there are other two evacuation at Jade Bridge, and
we can make sure that there were contaminated car involved in these two
evacuation from the readings.
Other evacuation way is Wilson forest highway. And there
were contaminated car involved in these two evacuation from the readings.
(PS: The animation (.gif) needs to be opened using Google Chrome.)
c. Indicated
where you would deploy more sensors to improve radiation monitoring in the
city. Would you recommend more static sensors or more mobile sensors or both?
Use your visualization of radiation measurement uncertainty
to justify your recommendation.
Before discussing this problem, we must first compare the
difference between dynamic sensors and static sensors. Static sensors are
stable and reliable. Dynamic sensors are convenient and have a wider coverage.
However, the readings need to be corrected and processed before they can be
used.
In our analysis process, for a wide range of radiation
detection, it is basically reliable, but in the southwest region, such as 10,
11, 12, 13 sensors are really too few, and there are fewer vehicles, it is
recommended to be close to the main road in these areas. One or two static
sensors are added to the upper one, and some dynamic sensors can be added to
the vehicles of the residents here.
In the 3, 6, 14, 15, 18, 19 and other areas of the city
center, a large number of vehicles pass through every day, but because the
reliability of the dynamic sensor is too low, it is very difficult to find the
correct reading in it. Things, therefore, there is no need to add dynamic
sensors in this area. Due to the large number of vehicles, 4~8 static sensors
can be added near the traffic roads, which can also easily correct the dynamic
sensor readings when processing data.
In the urban export area, we found that only three
outlets A, B, and C have stable static sensors. Therefore, it is necessary to
add static sensors to other outlets in the city.
In the seaside area, during the analysis, we found that
in some areas of the city, there is very little data, but these areas do not
have too many vehicles passing through. Therefore, it is recommended to add
some dynamic sensors to the residential vehicles living on the edge of the
city.
Limit your responses to 10 images and 1000 words
4 –– Summarize the state of radiation measurements at the end
of the available period. Use your novel visualizations and analysis approaches
to suggest a course of action for the city. Use visual analytics to compare the
static sensor network to the mobile sensor network. What are the strengths and
weaknesses of each approach? How do they support each other? Limit your
response to 6 images and 800 words.
Strictly control the entry and exit of nuclear power
plants, test them one by one, and then identify the contaminated vehicles and
centrally manage them in a relatively safe place. Or in some traffic routes,
detect vehicle radiation and control suspicious vehicles. The recommended
monitoring area is shown below.
Comparison
between static and mobile sensors in following perspectives.
What are the strengths and weaknesses of each approach?
How do they support each other? Limit your response to 6 images and 800 words.
Overview, the static sensors are more reliable than the mobile ones. Its
only disadvantage is spatial coverage.
All conclusions here are already discussed in the questions above.
|
|
Static |
Mobile |
|
Spatial cover |
Very limited with only 13 sensors.
But they can be placed in are without a road network. As |
Have bigger spatial coverage. But still
cannot cover area without roads. |
|
Time coverage |
Always open and very stable sampling |
The total number of available
sensors is not stable, as the sensor may leave the city during the working
hours. However, it can monitor the nearby
changes consistently when it not moving. Therefore, it gives good time
coverage near workplace and resident neighborhood. |
|
Bias and variance |
All device has similar bias and
variance, which made it easy to normalize. |
Each device Various in bias and
variance. Furthermore, they are not stable after bigger readings. |
|
Vulnerability during radiation
leakage |
Impacted by outside events, but the
impact are similar on different sensors. |
The impacts on different devices are
different, which make it even harder to normalize. Also device fails in
different ways |
|
Other Uncertainty |
|
Uncertainty may introduced by malfunctions
of devices itself, which is very hard to predict. |
|
Other benefits |
|
Mobile sensor can be organized temporarily to
monitor emergency event. The mobile sensor can also be used to evaluate the
traffic trends in the city and detect traffic jams |
The bubble sizes there shows the totally length of time at each spot. We
can see that the downtown area has density coverage.
Sensor should work together to get better results in following ways:
The mobile sensors can collect radiation readings from placed without
static sensors nearby.
The mobile sensors can assigned to replace the static sensor temporarily
when the static sensor stops working. As the mobile is less reliable. We can
assign more than one sensors to get more reliable readings using cross
verification.
The static sensors give more reliable readings, which can be used to
evaluate the uncertainty of mobile sensors nearby.
Even though the mobile sensors have high uncertainty in area with busy
traffic we can always find several different sensors. We can easily cross
verify the readings though visualization as we did in 2B. Therefore, the static
sensor is more effective in the open area with no road and light traffic.
Dissipate residents near OldTown to other districts.
5 –– 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.
In our
analysis, there are three aspects that are most important, it will be the main
part of our real time analysis system.
1) Real-time sensor reading verification, this method is answered as we
answered in the first question, and can also be compared and verified by a
nearby static sensor.
2) Real-time sensor reading maps, it allows us to maintain an awareness of the
overall situation.
3）A short-term continuous fluctuation of a certain location or sensor, as
show below, recording these historical fluctuations, and project these into the
map which providing a trajectory analysis of nearly 1 to 2 hours, allowing us
to observe the trajectory of the mobile pollution source in real time, thereby
predicting its moving direction.
