Entry Name: "NENU-Xu-MC2"
VAST Challenge 2019
Mini-Challenge 2
Team Members:
Shaobin Xu, Northeast Normal University,
xusb531@nenu.edu.cn PRIMARY
Yiming Lin, Northeast Normal University, linym762@nenu.edu.cn
Dezhan Qu, Northeast Normal University, qudz862@nenu.edu.cn
Ke Ren, Northeast
Normal University, renk205@nenu.edu.cn
Huijie Zhang,
Northeast Normal University, zhanghj167@nenu.edu.cn (Advisor)
Student Team: YES
Tools Used:
A system
developed using node.js, vue.js, d3.js and Echarts
Approximately how
many hours were spent working on this submission in total?
500
hours
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.
We have used GIF files in our submission for better tracking
mobile sensors in animated map. Please wait patiently.
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.
Time division
Fig.
1: Radiation measurements over time.
A heat map (Fig. 1) is employed to visualize
radiation
measurements over
time from both static and mobile sensors, in which each grid is
colored according to the level of average measurements in an hour. Two visual
enhancements are implemented to make the view expressively.
Layout: Instead
of simply arranging the sensors in order of name on the vertical axis, we use MDS to project the 59 sensors to 1D, based on
similarity in radiation measurements over time. Arranging similar sensors
together helps to discover local features.
Statistics: We add a bar chart attached to the heat map, which can provide
an overview of the mean value for each row or column. Time steps or sensors
with high radiation measurements can be captured clearly.
The following meaningful
periods can be clearly found.
1. Before the earthquake (Fig. 1, purple box): All of the sensors exhibit low measurements.
2. Early stage after the
earthquake (Fig 1, yellow box): A small number of sensors detect higher radiation means
initially. Some of them return to normal after a few hours, while others
maintained high measurements. Some sensors fail to obtain their measurements,
which may be affected by the earthquake.
3.
Later period after the earthquake (Fig 1, red box): Radiation measurements peak, accompanied by a large
number of sensors leaving the area.
Discovery of contaminated
areas
We use an animated map to track changes in the sensors
over periods, by brushing in the heat map. The borders of the static sensors are purple to make them
easier to distinguish.
1. Before the earthquake (Fig.
2): Sensors distributed in
various areas show low measurements overall, although some of them occasionally
have higher monitoring values. There are no abnormal regions during this time period.
Fig. 2: Animated map before the earthquake.
2. Early stage after the
earthquake (Fig. 3): Sensors in the vicinity of Always Safe Nuclear plant have
clearly monitored the radiation. Higher radiation measurements are detected in both East Parton and the Jude Bridge than
before, which may be related to the spread of contamination or the
movement of contaminated cars.
Fig. 3 Animated
map at early stage of the earthquake
3. Later period after the
earthquake (Fig. 4): The radiation in the above areas still exists, at the same
time continuously high measurements are detected in the southeastern regions
(Scenic Vista and Wilson Forest). There is no longer a mobile sensor entering
Chapparal and Terrapin Springs at this period. Considering that these two
regions are adjacent to the radiation region, radiation may also occur. Sensors
in the central area have become more unstable, but in view of the overall situation, radiation has not spread
there.
Fig. 4: Animated map at later period
after the earthquake. Some cars leave in three directions
Exploration of changes
over time
Although the animated map plays a great role in detecting
radiation areas, it is not good at summarizing changes over long periods of
time. To give a clear display of the changes over time for each region, we
design a view to aggregate time, which consists of two parts. Grids are employed by us
to achieve more granular analysis.
Background: We use the max function as our aggregation criterion,
which means the background of each cell is colored by the max
measurements of all the 120 time steps. The irradiated regions can
then be quickly identified.
Uncertainty Glyph (Fig. 5): The outer ring is divided into 120 sectors, which are
colored according to the measured means. Changes over time of each gird can be
clearly visualized. This view also contains a lot of additional information
that will be described in question 2.
Fig. 5: Uncertainty glyph
of mobile 10th.
The below figure shows:
1. The western and central grids are not significantly
affected by nuclear radiation, but sensor measurements become more unstable
after the earthquake.
2. Eastern part of the Safe Town
area remained at a low radiation level before the earthquake, but high levels
of radiation are detected immediately after the earthquake (Fig. 6 purple box).
3. Western part of the Safe Town area and East Parton are affected by radiation diffusion at
the beginning of the earthquake. Grid 65 (in Western part of the Safe Town
area) has high measurements in a short time but recovers quickly. Nevertheless
grid 38 maintains consistently high measurements after the earthquake. (Fig. 6
green box)
4. Grid 6 (in Scenic Vista) and grid 14 (in WILSON FOREST)
show high radiation measurements in the later stage of the earthquake, and grid
14 has the highest value in all grids. (Fig. 6 red box)
5. Many grids in the south were well measured before and
during the earthquake, but no sensors entered these areas later in the
earthquake.
Fig. 6: Grid
Summarization View of the 5 days.
2 Use visual analytics to
represent and analyze uncertainty in the measurement of radiation across the
city.
Division of time and space
To measure uncertainty accurately,
we divide both the time range and the space region. For the time range, a time
interval is one hour. As for the space region, we introduce the grid to take
the place of the administrative district.
Subsequently, uncertainty can be measured on a smaller scale. (e.g., one
hour in a grid)
Description of uncertainty
We introduce the uncertainty of
sensors to determine whether a sensor has completed the task of measurement
accurately within all the five days. This kind of uncertainty is quantified for
two indicators: consistency and missing.
Consistency: Consistency is used to represent the difference in
multiple measurements of the same sensor in a grid over a time interval. We model the difference with Gaussian distribution.
Mean and standard deviation are extracted to
measure uncertainty. High standard deviation means low consistency. We will use
“standard deviation” more often in the answers for ease of understanding. Note
that if a mobile sensor passes two girds in one hour, it can generate two
Gaussian distribution, we average the mean and standard deviation of these two
distributions.
Missing: Missing values are used
extensively to measure sensor uncertainty. One sensor is usually considered to
have high uncertainty if it is often missing.
In this challenge, the missing is caused by two cases: sensor failure and car
with sensor driving out of the island. We consider both cases together because
they all mean that the sensors cannot perform measurement tasks on the island.
For the purpose of visualization, we use the number of measurements to
represent the missing, which means the total number of sensor readings in an
hour.
As for uncertainty of regions, we
measure it by similar two indicators. The difference is that when constructing the Gaussian
distribution, we use measurements from all sensors in the region in an hour.
Visualization of uncertainty
In order to better visualize
uncertainty with two indicators, two design goals are proposed: visualize all
the sensors or regions to compare their differences and visualize a single
sensor or region for detailed information. To fulfill the both two design
goals, we put forward an uncertainty graph as a portrait of a sensor or a region. For a
single sensor or region, the portrait shows the mean, standard deviation,
number of measurements, and detailed distribution of measurements for each time
step (Fig. 5). As for all the sensors or regions, we use a simplified version
of the portrait by removing the detailed distribution. Three views are
implemented via this design
1. Grid Summarization View: As mentioned in question 1,
background of each cell is colored by the max measurements of all the 120 time
steps. One uncertainty graph is added to each grid (Fig. 6).
2. Sensor Projection View: MDS is employed for 2D layout of sensors. We use an
improved distance function to calculate the similarity between sensors. We
calculate the Euclidean distance of the mean if two sensors have measurements
at a certain time step. But if one of the sensors is missing for an entire
hour, we add a penalty for the distance. Missing is paid more attention by this
distance function. (Fig. 7)
3. Detail Inspection
View: When a region or sensor is clicked, this view displays its information completely
(Fig 5).
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?
In the Sensor View, different sensor patterns can be clearly
found via the appearance and location of the graphs.
Fig. 7: Sensor view after MDS projection.
Static sensor
1. Static sensors (the inner ring is made into purple)
except the 15th are in a large cluster in common. No missing occurs
in all five days, but the uncertainty of these sensors increases after the
earthquake, which can be clearly found by the changes of the blue fold line of
the outer ring.
2. In contrast, the static 15th lose its
measurement function after a period of the earthquake. This sensor around the
nuclear power plant is difficult to be trusted for the analysis of nuclear
leak.
Mobile sensor
1.
The mobile sensors in the large cluster at the lower right are similar
to the first pattern of the static sensors, although the overall uncertainty is
larger. These sensors are rarely missing, and the uncertainty is very low
before the earthquake except mobile 19th and 20th. These
two sensor are considered to be not accurate enough to trust.
2.
Mobile sensors in the central and top have experienced two long-term
missing in later period after the
earthquake (Fig.7 yellow polygon). The upper left part is more uncertain after the earthquake (Fig.7 blue polygon). Among them, the 21st sensor has
a high uncertainty before the earthquake, we believe it is not trusted based on
the same reason as in 1.
3.
The sensor on the lower left is
similar to the static sensor 15th, and the measurement function is
lost after a certain time, but the missing phenomenon is more serious. This may
mean damage to the sensor or the departure of the vehicle with the sensor.
4.
An abnormal sensor numbered 18
can be found at the bottom right. Considering its missing and standard
deviation, this is an untrusted sensor.
5.
There is also a sensor that is
always trustworthy until their measurements no longer change after the time of
the earthquake. They have zero standard deviation, which can be found by
observing whether the outer circle is existing (Fig. 8). These sensors cannot trust
after the earthquake. They are the 1st, 26th, 35th,
23rd, 47th mobile sensors (Fig. 7 red ovals).
Fig.
8: A simplified diagram of mobile 26th. Its standard deviation turn
into 0 after the earthquake, which means the reading no longer changes.
b. Which regions of the city have greater uncertainty of radiation
measurement? Use visual analytics to explain your rationale.
In order to analyze the uncertainty of the regions by means of
the measurements of the sensor, we use a line chart to show the changes of
sensor readings in the selected region. Several patterns can be found.
1. A large number of missing:
Many regions of the city are not adequately measured throughout the 5 days.
2. Medium uncertainty with large number of measurements:
Many central area grids exhibit this pattern and do not have
high anomaly measurements. The accumulation of many sensor uncertainties leads
to this phenomenon.
Fig. 9: Sensor
measurements over time in grid 45.
3. Short time high uncertainty with medium number of measurements: Grid 38 (in East Parton) and grid 65 (in Safe Town) exhibit this pattern. For grid 38, high uncertainty may be
related to anomalous events (Fig. 10). The first peak is caused by the spread
of nuclear radiation, and the second one may be related to aftershocks.
Fig. 10: Uncertainty
Glyph of grid 38
For grid 65, between 16:00 and 23:00 on the 8th,
there is a great difference between the three sensors (static 12th,
mobile 10th, and mobile 8th) measurements in this region.
Mobile 10th has an abnormality affected by the environment at this
time, most likely because of a contaminated car (Fig. 11).
Fig. 11: Sensor measurements over time in grid 65.
The measurements of mobile 10th is significantly different from
other sensors
4. High uncertainty with small number of measurements: As
for grid 50 (in Wilson Forest) and grid 70 (in Safe Town), there is no car
passing through these two regions most of the time, but the measurements will
be high once they are measured after the earthquake (Fig. 12). This may be
related to that vehicles go through the bridges in the grid to leave the
island.
Fig. 12: Uncertainty
Glyph of grid 70.
5. Some grids contain only damaged sensors for a long time (Fig.
13).
Fig. 13: Some areas in
the south contain only damaged sensors after the earthquake.
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?
We
use the sensor matrix view mentioned in question 1 to visualize the sensor
readings over time. More detailed information
as well as the uncertainty can be found in other views. With these views, we
can summarize the effects of events on sensor readings and uncertainty.
1. Earthquake
caused growth in sensor measurements of some sensors. The uncertainty of both
mobile and static sensors is significantly increased, showing more missing and
larger standard deviations.
2. The
earthquake caused some sensor failures, including sensors with loss of
measurement or with measurement no longer changing (Fig. 14).
Fig. 14: Animated map of sensor 6th and sensor 34th.
These two sensors fail at similar times.
Fig. 15: Some damaged
sensors. The readings of these sensors no longer change after 8:30. This may mean the time of the earthquake.
3. Some
vehicles enter and leave the island multiple times, which makes the two bridges
(Jade Bridge and Wilson Forest HWY) highly
uncertain.
4. Aftershocks
caused a lot of changes in the readings of some sensors (Fig. 16, Fig 10 red
ovals).
Fig. 16:
Measurements over time for static 9th. An obvious ladder can be found.
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?
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.
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.
In the first two questions, we explore nuclear radiation contamination from the perspectives of both the regions and the
sensors. After analysis, we can clearly see that there is a close relationship
between the two. Regional contamination affects the readings
of the sensors, and sensor anomalies can affect the assessment of regional
measurements and uncertainty. In this question, we will explore the relationship
between the two in more depth, which can provide a better assessment of contaminated
cars and locations of contamination.
a. locations of contamination
1.
Locations around Always Safe Nuclear plant: Grid 49 is a typical contaminated region. This area is
rarely measured, but different sensors have detected high radiation
measurements in this region at different periods.
Fig. 17: Sensor measurements over
time in grid 49.
2.
Locations which the
radiation is spread to: Two areas are identified as potentially contaminated areas: grid 38 (in East Parton) and grid 65 (in Safe Town). In
question 2.b, we have explored the uncertainty of these two grids. Gird 38 is a
contaminated region as the reason that the measured values of the sensors
passing through this area have increased after a period of the earthquake. Contamination has been spread here. In contrast, in the grid 65, only
one sensor (mobile 10th) has detected high pollution in this area
for a short time, which lead to high uncertainty. We prefer to trust other
sensors in this grid, especially the static sensor 12. Contamination has never spread to this area. Considering that the
mobile 10th is in the normal state for the rest of the time, we
think that some abnormality occurred during this time, for example, a contaminated car is near it. In total, locations on the south of the nuclear
power plant are more likely to become a potentially contaminated
area, especially the regions with great uncertainty after the earthquake.
Fig. 18: Uncertainty Glyph of mobile 10th. No
obvious abnormalities except for high values in a short time. There may be a
contaminated car near it at that time.
3.
Locations in the
south: Grid 6 is monitored
for contamination on the afternoon of the 9th. The
radiation measurement starts to increase significantly at 3.p.m., as shown by
the mobile 20th readings. The start time corresponds to the peak
time of the uncertainty in grid 38, which means possible anomalous events, such
as aftershocks. It is worth noting that other areas in the south have not been
well measured during this time. Some of them have no sensors passing through,
and some only have malfunction sensors. They are all potential locations of
contamination.
Fig. 19: Sensor
measurements over time in grid 6. The increase in the measured value of mobile 20th
coincides with the aftershock time we guessed.
4.
Locations around
bridges and the highway: A bridge and a highway have caught our attention because of
their uncertainty. Grid 14 (containing Wilson
Forest HWY) is monitored for very high radiation measurements on the
evening of the 9th. Despite its high standard deviation, we trust this result
because of the similar trends of all the sensors in the grid (Fig. 20).
However, for grid 70 (containing JADE BRIDGE), we don't think this is a
contaminated area based on the discussion above.
Fig. 20: Sensor
measurements over time in grid 14(containing Wilson Forest HWY).
The sensors that stay on the bridge are showing high
levels of radiation measurements.
b. contaminated cars
Cars with mobile sensors: In fact, there is no car in grid 49(containing Always
Safe Nuclear plant) in the period of the earthquake. Some cars entered this
area for a period of time after the earthquake and are considered by us to be
potentially contaminated vehicles. We track these cars on the map and find out
if they are contaminated. When they left the area, the readings immediately
return to normal low values (Fig. 21).
Fig. 21: Animated map of sensor 9th. This vehicle is not contaminated though it has stayed
in a contaminated area.
Cars without mobile sensors: After the earthquake, the uncertainty of some areas in
the city has become greater, which is most likely due to the movement of
contaminated cars in the city. We evaluate this impact by exploring these
uncertain regions.
1.
Grid 65 has been
mentioned many times. The abnormality of the mobile 10th is probably
caused by the presence of contaminated cars
in the vicinity, considering that it is reliable for most of the rest of the
time.
2.
It is likely that there
are contaminated cars parked in grid 70 for the reason that every car passing
through this grid presents a large uncertainty with high measurement in a short
time (Fig 22).
Fig. 22:
Sensor measurements over time in grid 70. Short-lived high values occur when the sensor passes
through these areas
Cars leaving the area: We observe the sensors that have left the area. When
sensors leaving from the Jade Bridge return to the area, they have higher mean
values than before (Fig. 23). Cars with these sensors are contaminated at
varying levels, especially mobile 46th and mobile 36th.
These contaminated cars travel to and from the city and the outside world,
affecting the uncertainty of some areas (Fig. 24).
Fig.
23: Measurements over time for mobile 8th. When it
returned to the island from Jude Bridge, the readings became higher.
Fig.
24: Animated map of
sensors leaving the city from
Jude Bridge. These sensors have greater measurement and uncertainty after
returning
St.
Himark officials should be worried about contaminated cars because they are
moving around the city and affecting some areas and cars, although for most
areas this effect is not serious.
c. Extra sensors
Areas where sensors are
needed: We believe that areas
with high uncertainty, especially high missing,
need to be added more sensors (Fig. 25).
1. Most areas on the edge of the island are not well
measured.
2. In the south, many areas have no sensors or only damaged
sensors after the earthquake.
3. Nuclear power plants are rarely measured after a period of
the earthquake.
4. There must be some abnormal events in Jade Bridge and Wilson Forest HWY.
Fig.
25: Grid Summarization View. Many areas have high uncertainty and require more
sensors
Selection of sensors: Mobile sensors expand the range of measurements (Fig. 26),
while static sensors provide long-term accurate measurements (We discuss this
in question 4). We hope to add sensors through the following scheme:
Fig.
26: Mobile sensors trajectory. Mobile sensors cover most regions of the island
1
Few cars will pass
through the edge area of the island, static sensors need to be added if
necessary.
2
Most of the southern part
of the island needs to be more fully measured, and more mobile sensors can do
the task, if the owners of these cars work or live in the southern
regions.
3
More static sensors are
needed in the nuclear power plant area because few cars will enter after
nuclear radiation.
4
These two locations
require more static sensors because the mobile sensors are very unstable in
these areas. Static sensors can reduce this uncertainty and feed back reliable
results
5
In addition, placing
additional mobile sensors in taxis is a viable option, considering their more
frequent movements.
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.
1
The
animated map for the last few hours shows that sensors in the city are still
not stable enough. There are still some sensors with high radiation
measurements in the area (Fig.
27). Contaminated
cars should be effectively controlled and handled centrally.
Fig. 27: Measurements over time of all the sensors. There are still
some sensors that monitor contamination
2
For
nuclear power plants, contamination still exists in this area,
which is found by the readings of mobile 32nd. This area should be
regulated and prohibited from entering by unrelated vehicles (Fig. 17).
3
Grid 38 is
still suffering from the spread of contamination,
and there are still vehicles staying here for a long time (Fig. 28). The impact
of contamination in this area on people needs to be
effectively assessed to determine whether it is accessible.
Fig. 28: Sensor measurements over time in grid 38. There are still
cars in this contaminated area.
4
There is a
need to take measures to prevent contaminated vehicles from coming in and out
on Jade Bridge and Wilson Forest HWY.
5
More
sensors need to be placed to reduce uncertainty in some areas, especially in the
southern region. This will help determine the extent of contamination spread.
Comparison of mobile sensors
network and
static sensors network
A total of
50 mobile sensors and 9 static sensors are included in this area. We compare
their differences from the following perspectives.
Radiation measurement
Static
sensors maintain relatively low measurements for most of the time. In contrast, there are two peaks in the mobile
sensor. Mobile sensors make it easier to find anomalous areas, which also means
more susceptible to impact.
Uncertainty
We discussed this in detail in question 2. In total, Mobile sensors are more prone to missing and damaged than static
sensors.
In some cases, more mobile sensors will not
reduce uncertainty but will increase it (Fig. 9). The area of the city center
is measured by multiple sensors, each of which is relatively stable, but with
large differences between each other. This leads to a large standard deviation
in the grid. We prefer to believe in the measurements of static sensors rather
than to average measurements of mobile sensors.
Coverage
The distribution of the nine static sensors is
not uniform, and most of them are distributed in the north of the island,
especially around Always Safe Nuclear plant. In contrast, the mobile sensors
trajectories cover most of the city, in addition to marginal areas such as
Wilson Forest. However, cars with mobile sensors are stationary
for most of the time, especially at night and during working hours. (Fig. 29).
Fig. 29: Animated map of all the sensors. At night, they are as stationary as static sensors.
At these times, mobile sensor networks can be
viewed as larger and more inaccurate static sensor networks, with a similar
uneven distribution. This means that although most areas are covered
superficially, some areas are not well measured during these two periods, which
is especially noticeable in the southern part of the island. In addition, after
the earthquake, it may be because of the traffic control, the number of
vehicles going to the southern region is significantly reduced, though these
areas are precisely the regions that need to be measured (Fig. 30).
In general, static sensor measurements fluctuate within a small range, have higher stability, are less susceptible to unrelated factors, and are more difficult to produce missing. Smaller coverage is a major weakness of static sensors. For mobile sensors, they have measurements with greater floating range and uncertainty, and are more prone to missing and damaged. However, with the regular movement of vehicles in the city, they have a larger coverage range and are more likely to detect the generation and spread of radiation. Of course, as we discussed above, the coverage is time-varying rather than long-lasting, which means that anomalies cannot be immediately discovered if the car with the sensor does not happen to pass there.
Fig.
30: Mobile sensors’
trajectories in the first two days and the last two days.
There are very few cars passing through the southern region after the
earthquake.
The synergy between mobile sensor networks and
static sensor networks provides cities with long-term monitoring of critical
areas and intermittent measurements of the entire area. Mobile sensors can
provide effective measurements in areas without static sensors, while static
sensors can provide continuous and accurate monitoring when there are no mobile
sensors in the area, and can also help reduce the uncertainty of mobile sensors
in the area when they coexist. To some extent, these sensors effectively
monitor the generation and spread of radiation and reduce its uncertainty.
Regrettably, there are still some areas that have not been effectively
measured, especially after the earthquake. Both dynamic sensor networks and
static sensor networks need to be extended. In order to better cover more
areas, it would be a better choice to install the sensor on a taxi, considering
its
more frequent movements.
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.
Based on the data of MC2, containing radiation
measurements from mobile and static radiation sensors, we analyze the data as a
static collection and focuse on summarizing the temporal behaviors of all the
sensors among the total period, finding the anomalies, and evaluating the
uncertainty of sensors and regions. For the target of providing suggestions for
the subsequent actions of the city based on our comprehensive analysis, we
design and develop a visual analytics system with six coordinated views (Fig.
31). An uncertainty glyph which can describe
the corresponding mean, the standard deviation, and the monitoring frequency in
each hour is designed for giving a comprehensive portrait of a sensor or a
gird. The overall views clearly summarize the situation within five days from
time and space perspectives, while detail views provide explanations for our findings.
Comparing to monitoring anomalies in real time simply, treating data as a
static collection is more helpful in exploring events from the whole to the
detail and evaluating existing conditions. In fact, the idea of real-time
monitoring is applied to our animated maps, and our system can be extended to include the ability to process
dynamic streaming data if necessary.
Fig.
31: System overview.