Can AI be used for disaster management and to limit the negative effects of natural disasters like Pakistan’s urban flooding of 2022?
In light of the unfortunate earthquake in Morocco, we’ve begun mulling over Pakistan’s disaster management and readiness, and it does not look good!
As the global climate crisis worsens, Pakistan is considered to be the fifth most climate-vulnerable country in the world. This is partially because of our geographical location but mostly because we just aren’t prepared!
Several countries have successfully tried and now readily use Artificial Intelligence tools to predict and minimize damages.
If properly utilized, using AI for disaster management could be a game-changer for Pakistan’s efforts in addressing the destructive effects of climate change.
Examples of Use Abroad
There are examples of several private open-source AI programs that have all been used from the beginning to the end of the process to mitigate the effects of natural disasters.
Prediction and Prevention
Machine Learning (ML) tools can help predict patterns of disasters for prediction and prevention. By ensuring greater planning time, they have served to accelerate and strengthen strategic planning and decision-making.
Google’s Hydronet Tool
Certain Machine Learning approaches to big data can trace patterns of old floods and predict behaviors for enhancing preparedness.
Google’s Flood Forecasting and ‘Hydronet’ solution is one such tool! It was used to detect vital vulnerable spots to prepare for floods in India and Bangladesh.
The Stanford Earthquake Detecting System (STEDS) is a similar ML model. It can detect earthquakes that go unnoticed by traditional models.
Its early prediction patterns in line with real-time updates allow for the triangulation of relief efforts to areas that most need them.
AI tools can also help target long-term sources of climate destruction! Rainforest Connection, a non-profit tech startup, utilizes such technology to detect sounds of illegal logging and wildfires in real-time.
Similarly, The UNOSAT S-1 Flood AI pipeline was used in Africa and Asia to predict flooded regions and train for evacuation and rescue planning.
AI tools using satellite imagery allow disaster response teams to prioritize areas needing immediate attention. By planning rescue efforts accordingly, teams can save lives and resources more efficiently!
Here are some standout tools to help with relief:
IBM’s PAIRS Geoscope
PAIRS stands for Physical Analytics Integrated Data Repository & Services. It scans areas to ascertain the extent of damage and also aids in recovery planning through future risk assessment in the affected area.
A US Department of Defense project that was created as an open-source program with multiple high-profile collaborators, including Carnegie Mellon University’s Software Engineering Unit and Microsoft.
xView2 has already been used for ground missions in Turkey following the recent earthquake. It can:
- Map out detailed areas affected by the disaster
- Locate areas of damage that relief teams were unaware of
- Scope out temporary shelter sites
- Offer plans for long-term reconstruction
Potential for Use in Pakistan
The potential disaster management uses of AI in Pakistan are plenty. Even some of the solutions mentioned could lead to massive demand and success!
However, since we can’t accurately predict several types of impending disasters, we’ll focus on the one that plagues Pakistan the most in recent years: flooding.
The 2022 floods caused over 1730 deaths and affected over 33 million people, half of them children. Livestock losses racked up to over 13 million. In Sindh alone, crop losses due to floods equaled PKR 364 billion. More than 2.1 million homes were lost, and over 7.5 million people were in need of humanitarian assistance.
Pakistan’s Susceptibility to Flooding
The World Bank rates Pakistan as 18th of 198 countries on the 2019 Inform Risk Index. This is due to both its susceptibility to natural disasters but also its social vulnerability and lack of protection mechanisms.
Pakistan also high exposure to all flooding types, including riverine, flash flooding and coastal flooding. It is also susceptible to tropical cyclones and their associated hazards. Furthermore, flooding, at 42%, was the majority constituent of the average annual natural hazard Occurrences from 1980-2o20.
Current sea erosion and Karachi’s low elevation in comparison to the sea level also place it at a high degree of risk of flooding. Climate change has insured rising temperatures which have caused many of the glaciers in the northern states to melt and increased chances of flooding in neighboring areas.
Additionally, consistent cycles of strong monsoon rains have only added to the threat. In line with that, it is important to have a clear and comprehensive risk prevention policy in place!
AI Solutions For Flooding in Pakistan
Perhaps the most exciting and relevant use of these tools is in addressing flooding within Pakistan.
Currently, the post-disaster side takes priority. AI tools can shift the mindset to prevention. ML models can analyze patterns of storms and flash flooding to find the most affected areas and put evacuation measures into place.
AI tools can create detailed flood maps that help authorities identify the extent of flooding. These maps use satellite imagery and aerial photographs to provide short-term relief planning alongside long-term planning for damage control to infrastructure and agriculture.
Specific machine-learning models can also address human-led activities that have adversely affected the Pakistani ecosystem. They can address deforestation caused by illegal logging, identify non-seasonal fishing trawlers, and even create long-term plans for curbing greenhouse gas emissions!
Need for Awareness
With the 2022 floods and the disproportionate experience of climate change facing Pakistan, authorities have begun to look towards non-traditional modes of disaster management.
Although some research institutions have suggested the usage for AI through discussion briefs, governmental interest has remained minimal.
The workshop held in Peshawar on 11 September 2023 has been the most recent push for using AI for disaster management. Held in conjunction by NDMA, PDMA and the University of Engineering Technology in Peshawar, the workshop brought together expert opinion on global collaboration for climate change mitigation and disaster management.
Dr. Fazal Ahmad Khan, GIK Swabi Rector, urged for exploring AI-based solutions and for creating policy that created awareness and training for such technology within local universities.
Challenges of Using AI for Disaster Management
No matter their growing progress, it remains important to understand the limitations of AI tools. Significant human and economic capital may be wasted if such limitations are not accounted for.
A Matter of Accuracy
AI and Machine Learning tools are only as good as the data they are trained on. If utilized without proper training, these tools will not provide accurate results and may even worsen prevention and mitigation efforts. This could lead to an overestimation of danger and waste economic resources. It could also lead to an underestimation of damage leading to worsening impacts of the disaster.
It is also essential to dedicate significant resources in creating localized datasets and detailed maps of the diverse natural climate and topography of Pakistan. Only then can we better and more accurately utilize machine learning to predict effective and efficient ways of disaster management.
Similarly, certain AI tools used for disaster management use satellite imaging. Uncertain weather conditions can distort these images and make them unusable. For example, cloudy conditions in Turkey ensured that all images of the disaster zone were unreliable. XView2 provided accurate analysis 3 days after the earthquake.
Satellite imaging is also not 100% accurate even in sunny conditions. A bird-eye view may not always provide the specific degree of damage as required. It is therefore important to recognize the limitations of AI tools for disaster management, while acknowledging their substantive assistance. A joint effort using AI with traditional methods of disaster management will ensure greater success.
A Lack of Trust
AI systems rely on vast amounts of data, which may include sensitive data about affected communities and even individuals. Concerns about misuse and mishandling are justified and require clear policies and data protection measures.
Another contending factor is the rescue personnel’s lack of familiarity with such disaster management tools. A fear of replacement and a lack of technological understanding can cause resentment and a refusal to accept the technology.
Creating awareness among traditional sources of disaster management is the only way to ensure a successful use of AI for disaster management.
Looking forward to speaking on a panel at @opensv about the challenges of climate change and AI tools developed for disaster management readiness, forecasting tools, and building resilience. pic.twitter.com/oAwJRcHJ2y
— Shabaz Patel (@shabazbpatel) November 6, 2022
The worst thing we can do is to wait for a reason to begin developing and using these tools! However, examples such as this OSV talk last year show that we are prepared to have conversations and get the ball rolling on the matter.