IS428 AY2019-20T1 Assign Nurul Khairina Binte Abdul Kadir TaskFindings Q2

VAST 2019 MC1: Crowdsourcing for Situational Awareness

Introduction

Data Analysis and Transformation

Interactive Visualization

Task Findings

References

Question 1

Question 2

Question 3

Question: Use visual analytics to show uncertainty in the data. Compare the reliability of neighborhood reports. Which neighborhoods are providing reliable reports? Provide a rationale for your response. Limit your response to 1000 words and 10 images.

Insights

In order to assess the reliability of the reports, we will consider the following factors:

• A surge of reports were made before the earthquake
• Delay in reports due to power outage and reports made after the earthquake
• Comparison of maximum impact scores over time (Pre, Major and Post earthquake)
• Variation in reported impact levels by neighbourhoods in the same region. The region is based on the estimated distance to the epicentre of the earthquake (Close, Moderate, Far and Very Far). The variation within the neighbourhood can also be observed.
• Relationship between shake intensity and other damage types to analyze damage uncertainty

This is the timeframe that we will be using for this analysis.

A. Pre-Earthquake – 6 April to 8 April

B. Major Earthquake – 8 April. Note this refers to the day of the earthquake

C. Post-Earthquake – 9 April to 11 April

Summary	Detailed Description
Surge of reports made before the earthquake on 6 April, 4 PM	We will analyze the data for the Pre-Earthquake Period (before 8 April). The filter is used to focus on this data. There is a spike in the number of reports on 6 April, 4 PM. We can see that Old Town, Scenic Vista and Broadview made a high number of reports at this timing across all damage types. Note that impact scores that are Null are not filtered out in this analysis. With reference to the shake map on 6 April, the shaking intensity was ‘Not Felt’. We will focus on this timing 6 April, 4 PM to 6 PM for this analysis. Since the shaking is Not Felt, we should expect less damage during this time period. We can expect some reports coming from the 3 neighbourhoods closest to the epicenter which is Old Town, Safe Town, and Cheddar Ford but the impact level shouldn’t be high. The average shake intensity for Safe Town was 2. The damage type filter for the mode chart is set to Shake Intensity. Safe Town was the only town with a mode of Weak from 4 PM to 6PM. At 7 PM, the mode changed back to Not Felt Since it is close to the source of the earthquake, it is acceptable and not an anomaly. The citizens might have overestimated the intensity of the shaking. Another possibility of why there is a surge of reports could be due to maintenance work. For example, we will focus on Roads and Bridges damages. There is maintenance work going on at Cheddarford which will result in delays. There is also more than 1 mode in the chart for Cheddarford and this is on 6 April, 6 PM which is 2 hours after the spike in reports.
Delay in reports due to power outage and reports made after the earthquake	We can observe a surge in reports for Old Town, Scenic Vista, Broadview and Chapparal on 9 April. All of them show a similar pattern of a gap of not having any reports made before the surge. Chapparal had 1375 reports and Old Town had 4490 reports once power was restored. However, only a small number of reports were made by Wilson Forest once power was restored and it was only for 4 hours since there was another gap after 7 PM on 9 April. In order to confirm that a power outage is the reason behind the surge of reports, we can analyze the Mode of Power Damages By Neighbourhood chart. For instance, on 8 April at 8 AM, the mode of impact score for Power was ‘Violent’ for Chapparal. 1 hour later, there were no reports made by Chapparal. Since it was so violent, the power could have been cut off. In fact, there were no reports made till 9 April at 4 AM and we can infer that the power was restored at that time. The power outage can affect the accuracy of the true extent of the damage and affect the emergency response plan. Once the power is restored, the backlog starts to flow which results in a spike in the number of reports that were originally made on 8 April and not 9 April. Lastly, reports were still being made after the earthquake. There is a spike in reports on 9 April, 3 PM. There is a possibility that there was another small earthquake that happened after the main one on 8 April. The source of the 2nd earthquake could have been in the same area since the same neighborhoods reported a higher average impact score as compared to the rest. Since more than 3 neighbourhoods reported shaking after the earthquake, it should be regarded seriously.
Comparison of maximum impact scores over time (Pre, Major and Post earthquake)	We are assuming that there is no 2nd earthquake on 9 April in this analysis. Hypothesis: The maximum impact score before the earthquake (pre-earthquake) and after the earthquake (post-earthquake) should NOT be higher than the maximum impact score on the day of the earthquake (major earthquake). These are the results when we compare between Major Earthquake and Post Earthquake: For Medical damages, Cheddarford has a maximum impact score of 7 for Major Earthquake (1 report) and 9 for Post Earthquake (2 reports). For Shake Intensity, Chapparal has a maximum impact score of 4 for Major Earthquake (1 report) and 5 for Post Earthquake (1 report). This report is inaccurate and 1 person could have reported it wrongly. Through this analysis, we can also conclude that using the maximum as a metric is not advisable since outliers will be considered as the maximum even if they are inaccurate.
Variation in reported impact levels by neighbourhoods in the same region	This is the time frame used for the following analysis: 8 April 8 AM to 9 AM. The reliability check preset is set to compare the impact scores 0 to 3 (in green) and 8 to 10 (in red) which are the 2 extreme ends. On 8 April from 8 AM to 9 AM, we can see that Old Town made the largest number of reports with impact score (8 to 10). This is extremely large as compared to other neighbourhoods that are close to the epicenter. As for neighbourhoods that are Far and Very Far, we expect to see a higher proportion of impact scores of 0 to 3 compared to 8 to 10 since they are further away and they are less likely to have a high shake intensity score. However, we can observe that Palace Hills have the largest number of reports with impact scores 8 to 10 as compared to the other neighbourhoods in the ‘Very Far’ region. It is important to ‘zoom out’ and look at the big picture before making conclusions. Use the Select Distance to Epicenter filter and set it to Very Far to focus on these neighbourhoods. The main issue here is the impact scores. 57.93% of reports in Palace Hills have an impact score of 6 to 7. This is unusual since Palace Hills is very far from the earthquake. Therefore, the reliability of the reports made by Palace Hills is questionable since the impact scores are inflated. A possible reason for this is that the tremors from the earthquake were spread to the West side of the city but we also have to consider that the shake map on 8 April revealed that the shaking was Not Felt at Palace Hills. Light shaking is expected from Northwest but for Downtown and Southwest the mode for shake intensity damages would be Not Felt like Palace Hills. We can check to see if the reports reflect the same observations we pointed out. We will assume that the shake map on 8 April is based on the data for the entire day in the analysis below. Palace Hills - We can see a variation in impact levels reported. A key observation made is that 37.93% of the reports made on 8 April have impact scores of 0 to 3 whereas 36.45% of reports made have impact scores of 6 to 7. If we were to use mode for our analysis, we can conclude that the findings are aligned to what the shake map tells us since the mode will be 'Not Felt' or 'Weak'. However, since the difference between these 2 impact levels is very small, we can't fully trust the information provided by this neighbourhood. Also, we can see how 13.22% of the reports made have an impact score of 8 to 10 for shake intensity. Downtown and Southwest - Both neighbourhoods have a mode shake intensity of 0 to 3 (Not Felt to Weak). It matches what the shake map tells us. Therefore, we can conclude that the shaking wasn't strong and Palace Hills shouldn't be affected greatly. Next, we can see that there were still reports with a shake intensity of 8 to 10 (Severe to Extreme) made by these 2 neighbourhoods and these are false reports. As for Downtown, 7.65% of reports have an impact score of 8 to 10 and for Southwest the proportion is much smaller at 2.81%. Therefore, we can conclude that the reports made by Downtown and Southwest are accurate since the observations from the shake map matches what we see in the data and the proportion of reports with impact scores of 8 to 10 is rather small. Overall, Palace Hills made the highest number of reports compared to other neighbourhoods and a possible reason for this could be the size of the neighbourhood. Palace Hills is bigger in size as compared to Downtown but not Southwest.
Damage Uncertainty	The analysis below is for 8 April (entire day). Firstly, we will focus cities that are located near to the epicenter and they are considered as some of the hardest-hit areas. For Old Town, we can observe that the shaking intensity corresponds to the other damage types. As for Pepper Mill, the shaking is less intense and the other damage types show that they are less affected by the earthquake. So far the relationship between shake intensity and other damage types reveals that the reports made were accurate. It would be strange if the impact on the other damage types like Housing / Roads is higher than Old Town, especially if there is no maintenance work going on in that area. Lastly, we can look at Wilson Forest and the lack of reports is not sufficient for proper analysis to be done. Secondly, we will focus on cities with a distance of Moderate from the epicenter. For Scenic Vista, there are a high number of reports with a shake intensity of 2 but Power, Roads and Bridges and Sewer and Water seem to be greatly affected since they have an impact score of 7.

Key Considerations and Highlights

In conclusion, there is a variation in impact levels reported within the neighbourhood and neighbourhoods in the same region at a specific timing. If St. Himark intends to use crowdsourced data for decision-making in future disasters, these are some of the key considerations that have to be taken into account:

1. Power outages can result in a delay in reports and this suggests that these cities won't be able to get the assistance they require, even if the damages are severe if the analysis was done based on the crowdsourced reports. The mobile application is only useful in providing the team with valuable data if citizens can report their damages easily.
2. The difference in reported impact levels could be based on what they perceive the impact to be. The team can include a basic description on what exactly does 'Weak' or 'Severe' means in order to increase the accuracy of reports made.
3. The size of the neighborhood may have an influence on the number of reports made. A neighbourhood that is larger in size is most likely to have a high number of reports as compared to a really small one.
4. Upon doing further research, I discovered that earthquakes come in clusters. In any earthquake cluster, the largest one is called the mainshock; anything before it is a foreshock, and anything after it is an aftershock. Therefore, the surge in reports before and after the earthquake on 8 April could also be due to foreshock and aftershock. Other possible reasons for the surge in reports on 6 April at 4 PM and 9 April at 3 PM could also be due to maintenance work going on.
5. The shake maps were used by the team to deploy their emergency response and the highest magnitude is 4. However, the crowdsourced data shows that the shake intensity is higher than 4.