Entry Name: "CSU-Jiang-MC2"

VAST Challenge 2019
Mini-Challenge 2

Team Members:
Haojin Jiang, Central South University, 1316918731@qq.com PRIMARY

Kui Yang, Central South University, kuiyang@csu.edu.cn

Yaqi Qin, Central South University, 614067608@qq.com

Yifei Yang, Central South University, 3901160307@csu.edu.cn

Ying Zhao, Central South University, zhaoying@csu.edu.cn SUPERVISOR

Fangfang Zhou, Central South University, zff@csu.edu.cn SUPERVISOR

Student Team:

YES

Tools Used:

D3.js

Python

Excel

Tableau

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

150 hours

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

YES

Video

index.files\video.wmv

Three high resolution figures to demonstrate how to read our visualization design.

index.files\Case1.jpg

index.files\Case2.jpg

index.files\Case3.jpg

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.

To begin with, Table.1 lists out major events that had impacts on radiation measurements.

Table.1 An overview of major events.

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.

To guide the analysis, we summarize three typical time-varying patterns of sensor radiation readings (Table.2) which can be found in the data.

Table.2 Descriptions of three typical time-varying patterns of sensor radiation readings.

Our analysis of sensors and areas with radiation levels over background proceeds with two aspects: static sensors (SSs) and mobile sensors (MSs), and each aspect is illustrated from two perspectives: all sensors and individual sensor.

1.1 Analysis of all SSs

Two lastly elevated radiation levels (P3 patterns) are found in Fig.1(a).

The first P3 (Fig.1(a-1)) possibly caused by contaminated cars staying close (E9). Its high radiation levels mainly distributed in Broadview and the entrance of Jade Bridge (Fig.1(b)).

The second P3 (Fig.1(a-2)) may be caused by the large-scale radioactive contaminations after the third earthquake (E4). Its high radiation levels distributed in nearly the entire city especially Old Town (Fig.1(c)).

There are no P1 and P2 patterns when analyzing all sensors as a whole, because short-term fluctuations of individual sensor are diluted after averaging

Fig.1 Analysis of all SSs.

1.2 Analysis of all MSs

Two lastly elevated radiation levels (P3 patterns) are found in Fig.2(a).

The first P3 (Fig.2(a-1)) may be triggered by contaminated cars and areas. Its high radiation levels mainly distributed near Always Safe plant, East Parton and the entrance of Jade Bridge. (Fig.2(b)).

The second P3 (Fig.2(a-2)) may be caused by contaminated cars parking nearby. Its high radiation levels distributed around the nuclear plant, East Parton and Scenic Vista. (Fig.2(c)).

Fig.2 Analysis of all MSs.

1.3 Analysis of individual SS

All the P1-P3 patterns can be frequently found in individual SS. Table.1 provides a number of examples. Here we take SS-11 as an example (Fig.3).

Several P1 patterns with sky-high impulse values over 500cpm occurred in SS-1 readings (Fig.3(c)). They were possibly caused by connection glitches. We processed them as noise.

A P2 pattern occurred at April 8 22:07 and lasted for one minute (Fig.3(d)), indicating one contaminated car passing by SS-11.

A P3 pattern lasted from April 8 22:03 to April 9 6:26 (Fig.3(e)), indicating contaminated cars staying nearby SS-11.

Fig.3 Radiation readings of SS-11 for two days.

1.4 Analysis of individual MS

All the P1-P3 patterns can be frequently found in individual MS. Table.1 provides a number of examples. Here we take MS-44 as an example (Fig.4).

Many P1 patterns occurred in MS-44 readings (Fig.4(b)). They were possibly caused by connection glitches.

A P2 occurred at April 9 6:32 and lasted for two minutes (Fig.4(d)), indicating one contaminated car passing by.

A P3 pattern lasted from13:30 to 16:45 on April 8 (Fig.4(c)). It could be the care taking MS-44 sensor parking near a contaminated location.

Fig.4 Radiation readings of MS-44 for two days.

1.5 Analysis of special areas with radiation level over background

We also find several special areas with high radiation levels. For example, two locations were detected to be slightly high in radiation readings since April 6, possibly under current constructions for road resurfacing and waterpipe repair (see our major event analysis in 2.3). In addition, many MSs detected several high values when passing through Jade Bridge and Wilson Forest Highway on April 9 and April 10.

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.

2.1 Classification of uncertainty

We define four types of uncertainties observed in the data to guide our analysis (Table.3).

Table.3 Definitions of four types of uncertainties and corresponding measurement methods.

2.2 Uncertainty analysis of sensors and regions based on the classification

Our analysis proceeds with three aspects: SSs, MSs, and regional grids (RGs).

Table.4 Uncertainty types in terms of SSs, MSs, and RGs.

2.2.1 Uncertainty analysis of SSs

(1) U2 frequently happened to all SSs. The fluctuation degree of radiation readings of most SSs kept increasing after 8:00 on April 8, which could be caused by the second earthquake (E2). Fig.5(a) provides an example of SS-4.

(2) U4 occurred only once to SS-15 (Fig.5(b)).

Fig.5 Uncertainty Analysis of SSs.

2.2.2 Uncertainty analysis of MSs

(1) U1 was common in MSs. The radiation background levels of MSs were inconsistent even at the same time and place. Fig.6(a) demonstrates an example.

(2) For U2, similar with SSs, the fluctuations of readings of most MSs became fierce after the second earthquake (E2).

(3) U3 happened to most MSs at 8:36 on April 8 (E2 time). The decimal places of readings changed from 0 to utmost 15. Fig.6(b) provides examples of MS-4 and MS-33.

(4) U4 has two sub-types: timesteps skipping and continuous data missing. The former often happened to most MSs. The latter generally was found when the cars with MSs left the town especially during E10 (Fig.6(c)).

Fig.6 Uncertainty Analysis of MSs.

(5) Several special findings:

• Constant readings of five sensors (MS-1, 23, 26, 35, and 47) were found after 8:36 on April 8 due to E2. Fig.7(a) shows an example of MS-35.

• Decimal readings of three sensors (MS-1, 20, 27) were generated before April 8 8:36 (Fig.7(b)), which was different from most MSs.

• Readings of MS-2 kept increasing linearly(Fig.7(c)).

• Three sensors (MS-6, 18, and 29) lost readings for a period more than once at three fixed parking spots (Fig.7(d-e)). We infer the sensors could be turned off by cars’ owners.

• Three sensors (MS-20, 23, and 34) might have been shut down and never worked again by unknown reasons. We specially speculate that MS-34 was damaged by E2 because it broke down immediately when E2 took place.

All the above findings can cause the four uncertainties with severe levels. Therefore, the relevant readings cannot be trusted.

Fig.7 Several special findings of MSs.

2.2.3 Comparison analysis of uncertainty types and levels between SSs and MSs

In terms of uncertainty types, MSs had 4 types (U1, U2, U3 and U4) and SSs had two types (U2, U4).

In terms of uncertainty levels (U2 and U4), MSs were generally more severe than SSs:

(1) For U2, MSs commonly showed a lower credibility with a larger standard deviation and wider confidence interval than SSs (in Fig.8(a)).

(2) For U4, MSs are unstable in data uploading. Timestep skipping and continuous data missing can be frequently found in most MSs, but continuous data missing only occurred once to SS-15.

Such differences can be verified in the treemap clustering view of all sensors, as shown in Fig.8(b) and Fig.10(c).

In addition, compared with Fig.1 and Fig.2, we also find that the average background radiation level of MSs (25.7cpm) is higher than that of SSs (14.6 cpm).

Fig.8 Comparison of SSs and MSs.

2.2.4 Uncertainty analysis of regional grids

RGs had two types of uncertainties (U1 and U4).

(1) For U1, the regions with a high level of readings inconsistency before (April 7) and after (April 9) E2 are shown in Fig.9(a) and Fig.9(b) respectively.

(2) U4 occurred mainly to regions without coverage of sensors. Fig.9(c) depicts the regions covered by MSs for the 5 days.

Fig.9 Uncertainty analysis of regional grids.

2.3 The influence analysis of major events on readings and uncertainty

Table.1 summarizes the major events that can be found in the data and background information of MC2 as well as the data in MC1 and MC3.

E1 (The first earthquake): No discernable changes of readings were caused by E1 due to its small earthquake magnitude.

E2 (The second earthquake): The earthquake magnitude of E2 was large, thereby causing the increase of the fluctuation degree of sensor readings (U2 mentioned in 2.2.2 and 2.2.1) and the changes of reading precision (U3 mentioned in 2.2.2) for many sensors especially mobile ones. It also caused more data missing (U4) as shown in Fig.10(a).

E3 (The coolant leak): The second earthquake led to the potential coolant leak, which took place during 8:35-10:28 on April 8. From then on, the P2 and P3 time-varying patterns occurred to many sensor readings due to movements of contaminated cars.

E4 (The third earthquake): The earthquake magnitude of E4 was close to E2, causing similar effects on radiation readings and uncertainties. Specifically, E4 caused the subtle increase of readings from all SSs and most MSs which maintained until the end (Fig.10(b)).

Fig.10 Analysis of the effects of E2 and E4.

E5 (Power maintenance): Before earthquakes, many skipped timesteps took place when the cars with MSs parked in Old Town or Southwest which were undergoing power maintenance (Location1 and Location2 in Fig.11(a,b and c)). We speculate that the power failure hindered data uploading in time.

E6 (Broken water main repairing): The broken pipe at the intersection of Blair and Quealy in East Parton may be a little radioactive (less than 20cpm), causing the small P2 patterns of readings of nearby MSs (Location3 in Fig.11(a,d)).

E7 (Streets resurfacing): The materials used in street constructions might be radioactive, triggering the rise (less than 20cpm) of readings of nearby MSs. Fig.11(a) gives the possible Location4 currently undergoing resurfacing projects. Many P2 patterns occurred when cars passed by Location4. And MS-28 once staying near Location4 detected subtle P3 patterns.

Fig.11 Analysis of the effects of E5, E6, and E7.

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

3.1 Analysis of potential contamination locations

Table.5 summarizes the potential contamination locations. Fig.12 marks out the locations.

Table.5 An overview of potential contaminated locations and possible causes.

Fig. 12 All the potential locations of contamination and examples of relevant patterns presented in the readings of nearby sensors.

L1 and L7 are needed extra explanation.

For L1, we find that the second earthquake (E2) led to the coolant leak (E3) and many cars of employees in Always Safe plant were contaminated inside the parking lot. The evidence can be found from Fig.13(a). The two cars carrying MS-9 and M3-13 stayed in the parking lot for four times in total. The radiation readings surged obviously in the last two times.

For L7, we find that the third earthquake (E4) may cause a large-scale radioactive contamination. Old Town was influenced significantly. The evidence can be found from Fig.13(b).

Fig.13 Analysis of potential contamination locations L1 and L7.

3.2 Estimation of the number and possible movements of contaminated cars

Contaminated cars had two types: ones drove away from AS and ones stayed inside AS. Three groups of the formers (total 19 contaminated cars) can be found based on the P2 patterns occurred to SS-15 which was located at the entrance of AS and MSs parking inside AS. But the latter were hard to quantify. Fig.14(a) demonstrates an example to explain how to detect the leaving from AS of contaminated cars.

(1) One contaminated car was detected by SS-15 to leave AS at 10:28 on April 8 and headed toward SS-14 (Fig.14(b-c)).

Fig.14 Analysis of a contaminated cars.

(2) 13 contaminated cars were detected by SS-15 to leave successively from 16:10 to 16:42 when at the same time the readings from MS-9 parking in AS-Parking-Lot steadily declined. Soon after, many SSs detected these cars passing (Fig.15).

Fig.15 Analysis of 13 contaminated cars.

(3) Five contaminated cars left AS successively from 19:50 to 20:30 on April 9, which can be found by MS-13 parked in AS-Parking-Lot (Fig.16(a)). Four of the five cars passed MS-32 which parked near AS loop (Fig.16(b)) from 19:58 to 20:32. In addition, the radiation level of MS-13 (100cpm) is much larger than that of MS-9 (44cpm) which parked in AS-Parking-Lot on the previous day from 16:40 to 16:46 (Fig.16(c)). This indicates that some contaminated cars might drove back, which cannot be verified due to the relevant data missing of SS-15. According to the remained radiation level detected by MS-9 when it left AS-Parking-Lot at 16:46 on April 8, we roughly estimate there were still 1 to 2 cars staying in AS-Parking-Lot, but this cannot be verified by other information.

Fig.16 Analysis of 5 contaminated cars.

3.3 Estimation of contaminated cars which might leave the city

Our estimation depends on the P2 or P3 patterns observed near each bridge or highway heading out of the city.

(1) Through Friday Bridge and Magritte Bridge, no contaminated cars left. SS-1 which located in the crossroad leading to the two bridges did not detect any P2 or P3 pattern (Fig.17①).

(2) Through Jade Bridge, at least 7 contaminated cars left the city and 5 returned, indicating at least 2 cars left without coming back (Fig.17②).

(3) Through Wilson Forest Highway, four cars left the town, which can be found by the reading declines of MS-27 and MS-28 nearby the place (Fig.17③).

(4) Through Himark and 12th of July Bridge, movements of contaminated cars were uncertain due to the lack of readings from nearby sensors (Fig.17④).

Fig.17 Analysis of contaminated cars which might leave the city.

3.4 How to improve radiation monitoring in the city.

Our proposal mainly focuses on mitigating the uncertainties in regions and enforcing the monitor of contamination conditions. Fig.18 depicts the overall radiation and uncertainty levels of the city and recommends the places required new SSs (large pink icons) or MSs (large blue car icons).

(1) Regions with a greater inconsistency requires a SS. For example, Easton had experienced a high inconsistency after April 9 12:00 and thus a SS was needed to enforce monitoring (Fig.19①).

(2) Regions, which were covered with insufficient data and had a low population, require a SS. For example, considering Wilson Forest was a sparsely populated area where cars rarely passed by, a SS should be deployed here (Fig.19②).

(3) Regions, which were covered with insufficient data and has a large population, can be improved with some MSs. For example, Oak Willow was a center for entertainments and should have large traffic volume. So, the lack of data here could be compensated by setting MSs (Fig.19③).

(4) Set more SSs or MSs in the regions with high levels of radiation. For example, the high radiation level at L5 continued till April 10, but it was mainly measured by MSs which shortly stayed here. SS-11 once detected contaminated cars staying nearby on April 9, but few MSs around that time caused that we cannot detect the tracks of the contaminated cars (Fig.19④).

(5) Set more static sensors at some key sites, such as hospitals, schools, business areas, and the entrances of bridges and high ways. For example, due to the lack of SSs at the entrance of July Bridge, Himark Bridge, and Wilson Forest Highway, it was hard to determine whether contaminated cars had left or entered the town (Fig.19⑤).

Fig.18 Recommendations of regions required new SSs or MSs.

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.

4.1 Summary of the state of radiation measurements on April 10

We summarize the radiation measurements from two aspects: the conditions of all sensors and the radiation levels combined with uncertainty levels of the city.

(1) The credibility of all SSs and MSs became lower with large fluctuation degrees of readings, due to the disruption from the second earthquake. The overall state of MSs was worse than that of SSs. Among all MSs, four MSs were anomalous in precisions and variation trends, seven broke down after the second earthquake, 11 lost data due to the relevant cars left the town without coming back. Only 30 sensors were working but the trajectories mainly concentrated on the north of the city (Fig.19(c-d)).

(2) The radiation levels detected by most sensors scattering around the city had increased (Fig.19(a-b)). Old Town was possibly severely contaminated. And the detected radiation levels remained high around the AS parking lot and northeastern East Parton, indicating the radioactivity of leaking coolant was undiminished.

In summary, the overall state of radiation levels of the city was still in high risks. The number of sensors was inefficient and their readings cannot well measure the situation of the city.

Fig.19 the state of radiation measurements on April 10.

4.2 Suggestion on the course of action for the city in response to the crisis

(1) In response to the widespread contamination caused by E4, the city should enforce the radiation monitoring. The following areas needed special attentions.

-- Old Town was in a bad situation and the school here should be closed.

-- Scenic Vista was likely to be contaminated, but the drastic surge of readings detected by MS-20 was unusual and thus might be malfunctioning. Quick verifications of the real condition here was imperative.

-- Always Safe plant must be carefully examined. Because the leak of contamination was in a large scale, we highly suspect the damage caused by E2 was exacerbated by E4.

(2) In response to the coolant leak, L5 and L6 (in Table.5) needed a quick cleanup of the dropped coolant from contaminated cars. Those suspected to be temporal locations of contaminated cars should also be checked and cleaned if any coolant dropped after the cars left.

(3) In response to the damage of earthquakes, four neighborhoods (ID: 3, 4, 8, 9) were suspected to be in bad conditions and needed urgent rescue (Fig.20). There are the following three reasons.

--Seven broken MSs were mainly distributed in seven neighborhoods (ID: 3, 4, 8, 9, 13, 14, 18), which could be possibly under serious damages like building collapses.

--Most cars detected to left the city possibly for safe shelters after earthquakes were from neighborhood 3, 8 and 9.

--There were a great many old buildings in Old Town and Safe Town which were easily collapsed in the earthquake.

In summary, it is of high priority for rescue teams to ascertain the casualties in these four neighborhoods and brought medical resources due to lack of hospitals nearby.

Fig.20 Neighborhoods needed further monitoring, cleanup and rescue.

4.3 Comparison between the networks of SSs and MSs

We compare the two networks from the three aspects as follows:

(1) Sensitivity to the detection of radioactive objects and areas.

--The variation in static sensor readings was relatively moderate. The spikes in readings were generally lower than 20cpm when stationary contaminated cars or areas were detected. But the readings increased significantly (30-200cpm) as legible P2 patterns when contaminated cars passed by. Therefore, the static network did better at identifying the movements of polluted cars.

--Due to the inherent inconsistency and generally small spikes, the passing of radioactive objects was hard to discover for MSs. But the spikes were drastic to over 500-1,000cpm when contaminated cars staying nearby. So the mobile network was more sensitive to stationary radioactive objects.

(2) Capacity of resisting disturbances.

--In SSs, only SS-15 blacked out for a period. While in MSs, six broke down due to earthquakes and the in-time data uploading will be disrupted by power failures or earthquakes. Therefore, the static network was more resistant than the mobile network to external stimuli.

(3) Overall coverage of radiation detection.

--Due to limitation of sensor number, the static network failed to cover most central and southeastern areas of the city, which can be made up of by the patrols of MSs. MSs can cover all neighborhood except Wilson Forest.

Fig.21 Comparison between the networks of SSs and MSs.

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.

The data processing and interface design of our prototype system are compatible with real-time monitoring.

The data are currently processed by two time intervals: one hour and one minute, which respectively target at analysis of historical data with a long time span and fine-grained real-time data.

First, based on the analysis of readings, P2 patterns generally lasted for 1-3 minutes, so one-minute time interval can perfectly capture them.

Second, the statistical analysis on time series as well as the uncertainty calculation and interpolation of radiation measurements on map are in low time complexities, which can be completed within 1 minutes for this dataset.

Last, all the visualization views in our interface support a dynamic updating. Fig.22 is an animation demo of dynamic view updating of our prototype system. Moreover, our visualization design results in few visual clutter issues.

To sum up, our system can meet the need for real-time monitoring by using one-minute time slice to deal with the streaming input. In addition, we should also consider the situation with wider geographical areas and a great number of sensors, which poses challenges to the capability of real-time data processing and efficiency of interface updating.

Fig.22 An animated picture of dynamic updating of our prototype system. (This is a gif picture)