Entry Name: PKU-Wei-MC2

VAST Challenge 2019
Mini-Challenge 2

Team Members:

Datong Wei, Peking University, weidt@pku.edu.cn PRIMARY

Hanning Shao, Peking University, 1700012930@pku.edu.cn

Zijing Tan, Northwestern University, zijingtan2021@u.northwestern.edu

Chenlu Li, Shanghai Jiao Tong University, chenluuli@gmail.com

Zhixian Lin, Peking University, zhixian.lin@pku.edu.cn

Xiaoju Dong, Shanghai Jiao Tong University, xjdong@sjtu.edu.cn

Xiaoru Yuan, Peking University, xiaoru.yuan@pku.edu.cn ADVISOR

Student Team: YES

Tools Used:

RadiationMonitor: An Interactive System for Visualizing and Exploring Spatial-Temporal Data, developed by PKU Visualization and Visual Analytics Group

Approximately how many hours were spent working on this submission in total?

200 hours

May we post your submission in the Visual Analytics Benchmark Repository after VAST Challenge 2019 is complete?

YES

Video

vis.pku.edu.cn/vastchallenge2019/MC2RadiationMonitor.wmv

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.

At around 8:00 April 8th, the striped pattern of data from all mobile sensors disappeared. Thus, this time is identified as the possible time of the earthquake. Comparing the data on the map before and after this timestamp, we identify the area southwest to the nuclear plant to be a location where radiation over background was detected. (see Fig. 1-1 )

Fig. 1-1 Overview of the whole map during the first period (before 4-8 8:00 am) and the remaining period.

Sensors that once detected high radiation can also be identified from the pixel graph (see Fig. 1-2). We match those high readings to the location of the sensors on the map during the time interval and find three other locations of interest.

Fig.1-4 Time pixel graph. From this graph, we could identify the important time point: 8:00 am and the high radiation levels detected from sensors.

The Nuclear Plant and Southwest to the Plant

At 10:30 on April 8th, mobile sensors 13, 22, 37, 41 (in the following text, static sensors are written as S1, S4, etc.; mobile sensors are written as #1 to #50) detected a sudden increase in radiation level from 30 cpm to 55 cpm in an area southwest to Always Safe nuclear plant.

Fig. 1-2 All data in the area throughout the whole period is presented

Other than this overall incrementation in the area, some sensors also detected transient but high radiation level from 13:00 to 16:00 on the same day. #9 arrived at the nuclear plant at 13:30 and detected 450 cpm radiation. 30 times higher than the background level (15 cpm). After the high reading returned to background level, #9 left the nuclear plant and exited from Jade Bridge.

Fig. 1-3 #9 data presented in pink

Wilson Forest Highway

From 18:40 on April 9th to 8:40 on April 10th, several sensors detected high radiation level of around 1500 cpm, about 50 times of normal radiation level, at Wilson Forest Highway (southeast corner of the island).

Fig. 1-5 Identify high radiation level at Wilson Forest Highway

Before the high radiation was detected, the last time data appeared in the area is 6:57 on April 9th. Thus, we concluded that high radiation emerged in the area between 6:57 and 18:40 on April 9th. Starting from 6:21 on April 10th, the radiation decreased in a step-down pattern until 8:40 April 10th, when the measurement returned to around 35 cpm. At this time, the cars exited from WF highway and no more data was obtained from the region.

Jade bridge

Jade Bridge at the north shore of the island is also a location where high radiation was detected during some periods. From 16:40 to 22:40 on April 8th, #10 detected radiation level at over 1400 cpm while parking on Jade Bridge (maybe due to blockage). During the same period, S12, which is right next to where #10 was parking, also detected an increase in radiation from 15 cpm to 30 cpm. After the high measurement decreased back to 10 cpm, #10 exited the map, and the reading of S12 returned back to normal.

Fig. 1-6 Identify high radiation level at Jade Bridge.

Then, on the next day (9th), S12 showed an increment in measurement from 15 cpm to 23 cpm at 15:00. This increase is echoed by a similar pattern from S9 (southwest to S12 in neighborhood 3) at the same time. #10 drove through Jade Bridge again 9 hours later, but anomaly was detected by neither #10 nor S12.

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?

b. Which regions of the city have greater uncertainty of radiation measurement? Use visual analytics to explain your rationale.

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?

Limit your responses to 12 images and 1000 words.

A.

In order to compare the uncertainty between static and mobile systems, we first compare the standard deviation of data from static and mobile sensors. From the pixel graph, in which the size of the pixels represent the standard deviation, it is clear that the data from mobile sensors had higher standard deviation than that from static sensors from the start. The uncertainty of data from all sensors increased over the period for all sensors. However, the standard deviation of static data remains relatively slow, while increase in that of mobile data is more obvious from the pixel graph. Conclusively, the static sensors have lower uncertainty compared to mobile sensors.

Other than an overall comparison, we also discover some abnormal patterns in data of mobile sensors (Fig. 2-1).

Striped pattern

Before 8:00 April 8th, almost all the mobile sensors showed a striped pattern in their reading. Most but not all of them are caused by the discrete property of integer reading. This pattern disappeared in all mobile sensors after 8:00 on April 8th, and the reading from the sensors turned continuous.

Constant-valued Reading

At the same time the striped pattern of all mobile data disappeared, data from #1, #23, #26, #35, #47 turned constant at different values. These sensors were possibly broken after 8:00 April 8th, and the data from these sensors after the turning point is not trustworthy.

Extreme Values

Another difference between the static system and mobile system is that data from mobile sensors includes more extreme values. For example, data from #3 included extreme value larger than 1000 cpm, while normal #3 reading was around 20 cpm. This kind of extreme values added to the uncertainty of mobile sensors, but the trustworthiness of the data is not affected.

#18

#18 disappeared a few times after reaching a specific location in St. Himark and then reappeared at the same location after some time. Shortly before the data disappeared and after it reappeared, the reading ranged from 0 cpm to around 60 cpm.

Fluctuation in measurement

Several mobile sensor had obvious fluctuation in their measurement over time even when they were not moving. #40 and #41 are the two mobile sensors with most fluctuation

#2 measurement continuously rising with constant gradient

Measurement from #2 was increasing with a constant gradient throughout the whole period. This pattern is different from the general pattern discovered from other sensors, where radiation level increases after several events. Thus, we categorize #2 as being unreliable.

Thus, based on the anomalies listed above, the following sensors are viewed as untrustworthy: #1, #2, #18, #23, #26, #40, #41, #47.

Fig. 2-1 Abnormal patterns of mobile sensors.

B.

We first examine locations on the map with high standard deviation. If the standard deviation of the location is high because of differences among measurements of several sensors, we mark this area as location with high uncertainty.

End of Jade Bridge

This is a location where lots of disappearance of mobile data happened when cars left the region. The main causes of this high uncertainty were the lack of data and the high variability of existing data read by sensors that entered and left St. Himark from Jade Bridge. Data was obtained from the area only when cars pass by, usually one reading from each car. This low density of data and the variation of measurements caused the high uncertainty.

Static Sensor 12

The radiation detected by #10 and S12during anomaly was greatly different. Even though both sensors noticed increase in radiation, #10 detected radiation as high as 1370 cpm, while S12 only gave 28 cpm. Thus, this discrepancy increased the uncertainty in the area around S12 during the period.

Furthermore, after #10 left the area, the uncertainty of data from both S12 and mobile sensors passing by increased

Palace Hill

The area included a great deal of data from #2. The data from #2 was increasing with an almost constant gradient with time. This trend of #2 data was different from all the other sensors, leading to the high uncertainty in this region.

Neighbourhood 19

Data in this neighbourhood can be categorized into two groups. The first group of sensors (#45, #46) detected a radiation level at 40 cpm, while the other group (#24, #27, #28) measured 25 cpm radiation. It is hard to know the actual radiation level of this area.

Fig. 2-2 The regions with greater uncertainty.

C.

We identified three events that have the most impact on the radiation measurements.

1. April 8th Morning, 07:30 - 08:30 Earthquake

After the earthquake in the morning, the stripe pattern in data of mobile sensors disappeared. Our hypothesis is that the accuracy of the system was increased during emergency situations, or the part of the system rounding measurements to integers was broken during the earthquake. The amount of mobile data also decreased due to leaving and broken sensors. This decrease in data led to some areas where radiation was measured only once in a long period.

Fig. 2-3 The first major event we detected.

The uncertainty of both static and mobile sensors increased after the earthquake. This pattern is clear from the size of pixels of mobile sensor data and the size of circles representing static sensors in the map.

2. April 8th 16:15 - 17:00 Contaminated cars left the nuclear plant

During this time, S15 measured several spiky jumps in radiation at the entrance of the nuclear plant. This signaled the time when several contaminated cars left the nuclear plant. Shortly after anomalies were detected by S15, several other static sensor (S12, S13, S14) detected similar but less numbered spikes showing passage of contaminated cars. This spiky pattern corresponds to the time when #10 and S12 detected rise in radiation at Jade Bridge. Some of the contaminated cars drove to Jade Bridge after coming out of the nuclear plant and was blocked on the bridge. The accumulation of contaminated cars at the blockage caused the high reading detected by #10 and S12. The uncertainty of many sensors increased at this timestamp, due to the sudden change in radiation. This is also a timestamp after which the uncertainty of the data started to increase more rapidly.

Fig. 2-4 The second major event we detected.

Fig. 2-5 Sensor 9 detected some contaminated cars leaving the nuclear plant.

3. April 9th 14:30 Second Accident

This was the time when radiation measurement of S9 and S12 suddenly increases. The measurement of #20 in area Scenic Vista also increased at this timestamp. Furthermore, data from the southern part of the city almost disappeared after this accident. Possibilities are the damage of roads or the evacuation of the area.

Fig. 2-6 The third major event we detected.

Fig. 2-7 The data is missing after 14:30 April 9th.

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.

Limit your responses to 10 images and 1000 words

A．

Locations of Contamination

Fig. 3-1 The locations of Contamination

l Jade Bridge

On Jade Bridge, there are two locations where potential contamination took place.

The first place is the area near S12. From 16:40 to 22:04 April 8th, both S12 and #10, which was parking next to S12, detected sudden increase in radiation measurements. A great possibility is that contaminated cars arrived at the location and parked there for some time because of blockage, causing contamination in the area. Later, starting from 15:00 April 9th, the measurement of S12 increased from 15 cpm to 22 cpm and never dropped back. This is another proof that the area around S12 suffered from contamination.

The second location is the location where Jade Bridge exits St. Himark. After the earthquake, lots of cars drove away and back through Jade Bridge. We find out that when the cars reach the end of the Jade Bridge in the map from outside, the sensor measurements quickly dropped from high level to normal radiation level detected on the bridge. Also, when the cars exited the city through the bridge, radiation rose rapidly to high value before the sensors disappeared.

l Area Southwest to Nuclear Plant

Changes in radiation detected by mobile sensors while passing the area indicated the contamination in the area. Below are the data from three sensors chosen randomly from all the sensors which passed the area at different times after the earthquake.

Data from #15, #22, #39, all the jumps were in the area southwest to the nuclear plant:

Fig. 3-2 Data from #15, #22, #39, all the jumps were in the area southwest to the nuclear plant.

l Wilson Forest Highway

From 18:40 April 9th to 8:40 April 10th, several sensors detected high radiation of 1500 cpm on Wilson Forest Highway. When the cars with sensor left the specific location with high radiation measurement, the radiation at the position had not returned to the previous level. After this group of cars left, there was no further data from the location in the remaining time. Thus, it is not clear whether the radiation at the location dropped back down to background level or not. With the high radiation detected, however, this location is a potential location of contamination.

Location of Contaminated cars

We found three locations where there was aggregation of contaminated cars

1. Jade Bridge

When #10 detected the high radiation at Jade Bridge, the measurement showed a step-up and step-down pattern. This pattern is likely caused by the gradual arrival of contaminated cars.

2. Wilson Forest Highway

The case here is similar to that at Jade Bridge

3. Entrance of Always Safe plant

When S15 detected the spiky increase in radiation on April 8th, the contaminated cars were exiting the nuclear plant.

As sensors always showed an increase in radiation when contaminated cars passed by, the city should worry about contaminated cars spreading contamination around the city.

B.

Estimate Counts of Contaminated cars

From the spikes in S15 on April 8th, we estimated that about 15 contaminated cars left the nuclear plant. In interpreting anomaly at Jade Bridge and Wilson Forest Highway, we interpret the step-wise increase and decrease of radiation as sign of contaminated cars coming and leaving. The contaminated cars likely left the city through Jade Bridge and Wilson Forest Highway. The contaminated cars exiting through highway was blocked at the blockage on the highway, after which, they likely exited through the highway. Several contaminated cars probably also exited from Jade Bridge as accumulation of radiation happened there and the end of the Jade Bridge in the map showed high radiation when cars passed by. From this count, about 12 contaminated cars left the city.

Fig. 3-3 The contaminated cars detected by Static Sensor 15.

Fig. 3-4 The contaminated cars detected by Mobile Sensor 9.

Fig. 3-5 S12 and #10 both detected contaminated cars on Jade Bridge.

Fig.3-6 Sensor 27 and sensor 29 stayed on WFHighway in different time period and they both detected some increases of radiation level on WFHighway. Each increase might indicate a contaminated car pass the highway.

C.

We recommend that the city deploy more sensors at the following locations:

Fig. 3-7 Areas needed to be deployed more sensors.

Fig. 3- 8: After 14:30 April 9th, there is almost no data from the southern part of the city.

l Jade Bridge and Wilson Forest Highway

High radiation measurement were found at the entrance through these roads to St. Himark. Static sensors can be placed here to monitor the contamination level.

l Entrances to St. Himark from Mainland

Place static sensors on main connections to mainland so that contamination level of cars passing by can be monitored to stop contaminated cars from spreading nuclear radiation to other areas.

l Areas with little data

Place static sensors in areas where the cars never go. More specifically, there is little or almost no data in Wilson Forest, Scenic Vista.Reports from these areas are too few. Static sensors should be placed to provide believable data to evaluate the radiation level in these areas.

l Northeast side of the nuclear plant

The area northeast to the nuclear plant along the shore lacks data as well. Placing static sensors in this area is important as the areas close to the nuclear plant should attract most attention.

l Southern Part of St. Himark

After 14:30 April 9th, there is almost no data from the southern part of the city. If this is because remaining cars in the region do not have sensors, add more mobile sensors in this area. If this is because the roads in this area are blocked, add static sensors to monitor the radiation level

We recommend static sensors as the uncertainty in static sensors are much lower than that of mobile sensors. Also, with data from April 6th to April 10th, it is already possible to locate areas where contamination took place. Thus, static sensors would be able to provide a more location-focused observation.

Fig.3-9 Size of rects denotes times the area had been measured by mobile sensors. It could be seen that the mobile sensors could cover the main road of the city before the earthquake.

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.

Radiation measurements at the end

Throughout the whole period, the background level of radiation gradually increased. Thus, at the end of the available period, the radiation level at all areas were higher than that at the start of the period. Below are some areas with more significant patterns.

Fig. 4-1 The overview of the radiation level across the city at the end of the available period.

l Nuclear Plant and Southwest to the Plant

The high level of radiation in this area persisted until the end of the available period. The radiation in areas directly south and west of the nuclear plant also increased throughout the whole period, yet with no abnormal increase compared to other areas. The area north and east to the nuclear plant lacked data, so the radiation measurements in those regions cannot be analyzed with current data. Thus, among all the areas with data around the nuclear plant, the area southwest to the nuclear is the only region with clear evidence of contamination.

Fig. 4-2 The high level of radiation of southwest to the nuclear plant persists

l Southern Part of St. Himark

From around 14:30 April 9th to the end of the available period, there was almost no data in the southern part of St. Himark (neighbourhood 8, 9, 10, 11, and 17). Two locations of importance in this area.

l Wilson Forest Highway

Several mobile sensors showed high radiation level at the southeast corner of the map near the end of the available period. However, most of the sensors showed a step-down pattern, and no more sensors pass the area for the last 14 hours. In order to be more certain, data for the last 14 hours at the area is needed.

l Scenic Vista

#20 detected radiation of 190 cpm from 15:00 April 9th to 13:04 April 10th before the signal suddenly disappeared. As only data at the location of anomaly is from #20, the radiation status from the end of #20 data to the end of the available period (11 hours) is unknown.

l Palace Hill and Friday Bridge

Measurements from the area near Friday Bridge in Palace Hill neighbourhood seemed higher than that in most other areas during the last 12 hours. However, scatterplot of this area shows that #2 is the main cause. As #2 is categorized as not trustworthy because of its abnormal trend, the high reading at Friday Bridge is not reliable.

Suggested Actions:

l Investigate what happened at the southeast corner of the island after 14:30 April. What is the cause of the disappearance of data from this area? What is the radiation level in the area now? Check if the government should evacuate the people living in the area.

l Investigate what caused the contamination from the nuclear plant to orient in the southwest direction.

l Calibrate the mobile sensors or make improvements on the design of the sensors. Currently, different mobile sensors detected different background levels at the start, ranging from 15 cpm to 40 cpm. The consistency in the background radiation detected by static sensors proved that there is little difference among background radiation at different locations at the start. Thus, recalibrating the mobile sensors may help to solve the problem. Also improvements on the design of the sensors can help lower the uncertainty of the mobile sensors, which would add to the reliability of data obtained.

l Place more sensors are suggested in question 3

l Trace down the contaminated cars so that they can be cleaned to avoid further contamination of the city

l Investigate to reason behind the lack of data from southern St. Himark after 14:30 April 9th

Co-working of the Static and Mobile Network

In a monitor system, below are the important properties:

l Completeness of Data after Events

The static system is far more consistent than the mobile one. Among all the 9 static sensors, only S15 lost part of its data. In the mobile system, however, 27 sensors experienced lack of data, and five sensors went broke, leaving only constant-valued reading. Thus, the static system is far more resilient towards events like the earthquake than the mobile system.

l Consistency in Reading from a Location

As the static sensors are placed at specific locations, the system is able to monitor changes in radiation at the specific point throughout a long period. This is an advantage while studying some locations of interest. While reading from mobile sensors is affected by both the time factor and the location factor, that from static sensors is only affected by the time factor and can be used to analyze time evolution of radiation.

l Coverage of Locations

An obvious advantage of the mobile system is its coverage of wide-spred locations. While 50 mobile sensor is able to reach every road of St Himark,far more static sensors are needed to reach the same spatial coverage. Yet admittedly, there are locations covered only shortly. The solution would be setting up static sensors at locations with abnormal patterns.

l Comparison among Sensors

Static sensors are all placed far away from each other to ensure spatial coverage. Thus with just the static system, it is hard to tell if a sensor is broken or not.

Static and mobile sensors are two totally different systems, and they have different strengths. The strength of one system may be the weakness of another. Thus, when the two systems monitor the condition together, their strengths can be combined.

By comparing data from the two systems, users are able to see that the differences in starting radiation level exists in the mobile system but the static system. Thus, the users know the discrepancies between starting radiation level of different mobile sensors is due to the errors in mobile sensors rather than the actual radiation level.

Also, as mentioned above, mobile sensors and static sensors can be applied at different stages of the exploration. When no clear pattern is known yet, mobile sensors help detect more areas. When some areas with anomalies are discovered, static sensors can be applied to the areas to monitor the area closely.

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.

We analyze the data as a static collection. Some statistical magnitudes, such as the means, standard deviations and frequency are introduced to distinguish different periods of time. For example, we color a sensor’s timeline to mark out whether it send a constant value, lost connection or worked normally, and a certain sensor’s value changing over time can be shown on the scatter. Besides, we analyze the data aggregated by selected areas to show if something abnormal happened in the place. We color the small squares on the map according to the means or other statistical magnitudes.

In fact, our system is able to run on the data stream as well. There will not be any difference drawing the map and the scatterplot. However, since we need statistical magnitudes to analyze a sensor’s working condition, the system has to wait for data to accumulate before having enough data to calculate those statistical magnitudes. As a result, the color of sensors’ timeline and the background colors of the scatter would be shown about 30 minutes later, as we draw the pattern based on 30-minute units.