Fundamentals of Disaster Management Sytems: A Computer Science Perspective
Mehmet Aksit - University of Twente - 3-7-2023
Outline:
Disaster Management as it is today
Problem Statement
Process Automation for Disaster Management
Today:
1) Disaster Prediction
2) Post-Disaster:
- Identification
- Assess
- Understand
- Cope
- Strategy
- Recovery Procedures
Problem statement
He worked with the auhtorities and developed requirements for dealing with disaster and without looking at the trends, really focused on talking to authoraties' stakeholders.
They used different viewpoints to elicit requirements
They used quesionnaires with wanted-not wanted 5 likert scale in a large organization in Turkey. This study identified 65 requirements that are now prioritized. For prioritization, they identified which ones could be deffered to a later time without causing technical problems during disaster management (priorization in 3 groups or requirements). After that, they synthesized to identify technologies and skills to fulfill the requirements.
Obsserved problems: they analyzed the problems and saw that it would be difficulty to deliver Process Automation in the way they intended.
They generalized the problem of process automation as a problem of Demand and Supplies. There are several related works in aid optimization for disasters. But they have some limitations: generally offline optimizations, problem-specific (case by case), product specific (for specific systems), weak or absence of automated process support, accordingy, lcak of an online control system platform to manage disasters effectively and efficiently.
Solutions for the demand and supply problem
Example:
They have about 100 tables like the one in the picture below, with established rules. Then the cocneptual model on the right hand corner guides the automation of a process to deal with a particular disaster. They create tasks and group tasks that compose jobs, and this is assigned to different people.
If there isn't enough resource, then some strategies are developed (trade-off analysis, prioritization of jobs, etc.)
To sum up, for them, a disaster problem is a resource allocation problem.
Upcoming publication:
Disaster Prediction
Another direction of work they do is to predict disasters. They do that based on the concept of 'Event'.
I asked them if they do that by analyzing social media, but he said 'no'. They use other data sources.
Giancarlo has asked an interesting question about causality, in other words, how to identify that an event will happen because it is caused by a previous event occurence that has been already identified. Mehmet said that they treat this in their approach.
Event specification:
Upcoming publication:
Using Digital-Twins for disaster prediction and handling:
Conclusions
From a ML point of view, there is a lot to be developed such as:
Notes on the discussion subsequent to the talk:
There are many advanced simulation systems out there, and instead of producing new ones, we should use the existing systems that are already very sophisticated.
They keep collaborations with the Disaster Center at the Univ. of Sao Carlos, in Brazil, and they hope to talk to the Federal Disaster Management team as well. The main problem that he sees is that Brazil, like in Turkey and many other countries, there is a lot of data from satelites and other state of the art sensors, but there is no data on disaster management process. And the main problems for him are process problems. That is why they are working on process automation, and these partners have been very excited about this kind of work.
It is even relatively easy to find data sets on disasters, but not on disaster management processes. Therefore, it is very important to work on producing this data, as well as integrating such data from different data sources. Also during the pandemic he noticed that most of the work was on putting data together from different systems. That is why we need International Alliances and a lot of work on data integration.
Their system is responsible to make decisions to allocate tasks to teams in disaster management. The decisions are based on rules and ML algorithms that adapt the rules for making the predictions and recommendations.
How to cope with people's resistance in using such systems? Jan says that perhaps the operator should give the last word and not the system. But actually in many areas, we see systems that already take the decision on our behalf and they can do it better than us (e.g., piloting planes, flood gates in Rotterdam).
Mehmet says that something important is to categorize the type of disaster - meaningful categories! (see slide on ML opportunities above). Depending on the type of disaster, there can be more or less human intervention.
Giancarlo suggests that work should be done in: intentional (goal) modeling to understand what people's intention is in the middle of a disaster (e.g. Paris riots).
Mehmet says: another important issue is to deal with ethics and privacy in this domain.
