Kashmir-born young scientist, Dr Jawad Fayaz, who did his PhD from the University of California, Irvine, USA and worked as a postdoctoral research scientist at University College London, UK, currently teaches at Teesside University, School of Computing, Engineering, and Digital Technologies, UK. His research revolves around interdisciplinary goals of improving community resiliency and sustainability, especially against extreme events like earthquakes using Artificial Intelligence and Machine Learning. He spoke to Kashmir Life’s Science Talks
KASHMIR LIFE (KL): Do you think that earthquakes can be predicted?
Dr JAWAD FAYAZ (DJF): Though I am not an expert in geological seismology and my expertise lies in the realm of earthquake engineering, an immediate response to this question would be a big No, especially in the timeframes that the general population thinks (weeks, months, or years in advance). One of the world’s renowned structural and earthquake engineers, once said, “There are only two types of people who can predict earthquakes: a con man and a madman”.
Discussions surrounding earthquake predictions generally encompass two distinct timeframes: long-term (days to years), and short-term (seconds to minutes). There has been no evidence that a long-term prediction is possible, we can only give general opinions based on the strain developed in the plates about the possibilities. There is no definite scientific method to assert these long-term timelines.
The second timeframe spans from the tectonic rupture and fault’s initial activity to the waves’ propagation to distant areas. As an example, when an earthquake hits northern Pakistan, the interval it takes for waves to reach Srinagar can serve as a predictive time period for assessing earthquake intensity. The prevailing challenge pertains to how one can predict an earthquake in mere seconds and institute precautionary measures within that brief interval. The answer, regrettably, is in fact that there is limited scope for action on an individual basis.
If we look through the delve into this uncertainty it is quite easy to understand that the primary cause of fatalities during earthquakes is not the tremor itself, but the compromised infrastructure and systems that fail during the event. Gas leaks, high-speed trains, and electricity disruptions, for instance, are significant contributors.
Consequently, our focus during these short timelines is not necessarily at an individual level but rather at a community level. We are exploring solutions such as AI-triggered alarms and intelligent grid systems, which could deactivate power or mitigate potential hazards like gas leaks. Our goal is to intervene just before the earthquake reaches a specific location, halting contributing factors and minimising damage. While significant strides may not be achievable within this tight timeframe, there is room for limited early warnings and potential preventive actions. In this way, we aim to make a tangible impact on reducing the impact of disasters.
Apart from this, there have been some notions in the general public that after the Turkey earthquake, there is much more probability of having an earthquake in north India, but it is all a hoax and there is no proof of it. As I mentioned earlier, based on the history of the fault movement, seismologists and geologists assess the potential strain in the fault and based on that postulate that a ‘Big One’ is due as we hear all the time in California, regarding the San Andreas Fault (SAF). But again, there is no conclusive methodology to estimate when it may happen or even get a rough timeline. It might as well take a hundred years, five hundred years, or maybe just one year. But yes, in reality, the Eurasian and Indian plates do in fact collide due to which even the tip of Mount Everest grows higher every year and some researchers from IITs also claim that there is a probability of a ‘Big One’ in the Himalayan region but again there is no proper consensus.
KL: What were the learning curves of your journey?
DJF: I belong to Lal Bazaar, Srinagar, and completed my higher secondary at Burn Hall School, Srinagar. After that, I went to RV College of Engineering, Bangalore to pursue my bachelor’s in civil engineering and subsequently got accepted by the University of California Irvine where I conducted my master’s in Structural Engineering. While working on my master’s thesis, I was offered a PhD position which I accepted and officially completed in June 2021.
During this time, I developed an acute interest in Statistics, AI, Machine Learning and Deep Learning, which I mainly self-taught along with 2-3 programming languages. These tools further led me to the successful culmination of my PhD journey.
Prior to my graduation, in March 2021, I was offered a post-doctoral research scientist position at University College London, England. After working there for around a year, I received a compelling invitation from Teesside University to contribute to their computing and engineering department. Unlike other universities, Teesside University has a joint department for computing and engineering which encourages more interdisciplinary research and teaching. Given the multi-disciplinary nature of my research, this opportunity greatly intrigued me and I joined Teesside University as an Assistant Professor (Lecturer) and also remained associated with University College London as an Honorary Lecturer. Currently, I am actively engaged in the ongoing work and endeavours of both universities, contributing to the fields of computing, engineering, and beyond.
KL: What were the main takeaways from your PhD?
DJF: My PhD was a bit broad and unconventional as it involved different topics and frameworks to ameliorate bridge infrastructural systems. So, through machine learning and statistical methods, I was working on different frameworks to improve the analysis and designing techniques of earthquake-resistant bridge structures. The analyses included vulnerability assessment of structures under different seismic environments like crustal and subduction; testing different methods of using synthetic and simulated earthquakes; developing deep learning models to predict structural responses during an earthquake. The overarching focus of my research has been on improving community resilience and making our structures and infrastructural systems safer through the lens of artificial intelligence and statistics. By the end of my PhD, the results of my work had been published in around nine to ten prestigious Q1 journal papers.
Engineering PhDs are generally different that other fields. Typically, in a PhD, researchers focus on identifying and analysing a problem and possible solutions. However, in engineering, 95 per cent of the work is concentrated on providing novel solutions to known problems. Hence the main takeaways are solutions rather than a simple message.
For example, if a common person is asked to design and analyse an earthquake-resistant bridge structure, and asked how they would test if the designed bridge is “good enough”. The most common answer is to take the last 100 or so years of earthquake data and make sure that the bridge can withstand all of them. This is the classic approach. But this has two major problems: firstly, the earthquakes never repeat themselves not in the sense that a magnitude 7.6 cannot happen again, of course, it can, but any two same earthquakes do not have exactly the same characteristics in their waves. Just like identical twins are not the same in their behaviour, no two earthquakes generate exactly the same waves. So, an earthquake in future will not be the same as a previously observed one.
Secondly, taking the worst earthquake in history may not happen again in the lifetime of the bridge which is generally designed for 75-100 years. So, taking the worst-case scenario can make the bridge structure highly expensive. Hence, we try to take a probabilistic approach in such cases. Thereby, the focus of my PhD was on developing better methodologies using AI for a proper probabilistic scheme of analysing the structures for better design. Also, it should be noted that due to the points I described above there is nothing called an “earthquake-proof” structure, as no matter how you build your structure, due to the unpredictable nature of future earthquake waves there is always uncertainty.
KL: What was your postdoc all about?
DJF: During my PhD, I got more inclined to the deep learning domain of AI and started exploring it. During my postdoc, I worked on a major European project that aimed at developing earthquake early warning and decision support systems. I particularly explored the domain of deep learning and build data-driven systems for early warning and seismic hazard modelling.
One of my works specifically attracted quite a bit of attention. In this paper, we proposed a highly accurate deep-learning framework that can use the initial few seconds of the seismic shaking at any given site to predict the characteristics of the waves that have still not arrived at the site. The earthquake waves are mainly composed of two types: primary waves (p-waves) and shear waves (s-waves). The p-waves are faster (with a speed of around 6 km/s) and are not so damaging; while as s-waves are slower (with speeds of around 1.5 km/s) and are generally more detrimental. The early warning systems take advantage of this time delay between the non-damaging and damaging parts of the waves at a site to provide information. So, our goal in this early warning system was to take the initial three seconds of the p-waves, and through our AI model, we can predict the intensity and amplitude of the overall seismic waves that are going to occur during the earthquake event and know what accelerations and forces will be inflicted on the structures and hence estimate the damage. On this basis, we provide a decision support system to trigger alarms and take possible precautions.
KL: Has this device system been implemented at some level?
DJF: It was recently published in January 2023. I started this study in 2021 but by the time it was reviewed and published, it consumed a lot of time to be wholly applicable. The algorithm is now slowly drawing attention and making an impact within the research community. I was recently invited to Chile, which is one of the most seismically active regions in the world, to discuss the framework. Chile experiences an earthquake of magnitude 7.5 every 10 or 20 years and hence the infrastructure is designed accordingly. Apart from that I am also collaborating with Japanese researchers as Japan is also prone to earthquakes and we are exploring the options.
Fields that are related to public health take time from being published to getting implemented, as even small mistakes can be catastrophic. The same thing applies to the high-risk medical field as well as structural engineering. Though early warning systems may not seem that risky, the worst thing imaginable is just a “false alarm”. However, in the context of earthquakes, false alarms can create a major phobia and panic in the public which may result in mental stress and hesitation to work.
Even though our framework has shown more than 90 per cent accuracy in the theoretical analysis, we are thoroughly trying to scrutinise it with the researchers of various countries involving Chile, Japan, Greece, and the Himalayan regions to test it and hopefully, it may be properly implemented in the near future.
KL: Is there any authentic software generated for it?
DJF: I have successfully developed user-friendly software, accessible freely on my website. Most of my published papers are accompanied by complementary software, available at no cost, and these tools have been globally downloaded and adopted. We hold a positive outlook that our novel early warning system will garner greater recognition and adoption in the near future.
Early warning systems have already gained traction in Western countries such as California and Japan. Even in India, researchers have crafted mobile applications for early warning, currently undergoing testing across northern states. Our mission extends to promoting this technology within India, aiming for practical application and performance assessment.
KL: What was the takeaway from the recent Turkey devastation?
DJF: Given my detachment from practical structural engineering for quite some time, I cannot say much about it. Undoubtedly, experts have already shared their insights on this matter. From my perspective based on the surveys, not only in Turkey but across most seismically vulnerable communities, the core issue transcends engineering and delves into the socio-political landscape. I may be oversimplifying, a substantial portion of destruction arises not solely from engineering deficiencies, but from the corruption and carelessness entwined with construction activities.
In Turkey, for instance, post-event reconnaissance revealed divergent outcomes for similar structures at the same locations. Many collapsed while others remained unscathed. The disparity often traces back to non-compliance with seismic regulations and codes during design and construction.
In 2019, during my PhD in California, we surveyed areas hit by successive earthquakes of magnitudes 6.4 and 7.1, alongside numerous aftershocks. Although such large back-to-back events are not usual for California, the structural damage was minimal. Losses primarily resulted from ruptured pipes, gas leaks, road deformations, etc., and structures mainly displaying minor cracks. Such occurrences underscore the efficacy of buildings constructed to code.
I do understand the economic and purchasing power of the local population plays a vital role in the decisions we make in developing countries, however, time and again we are reminded by such extreme events that neglecting community resilience and adopting lax construction guidelines can bear significant costs. The challenges extend beyond engineering prowess, emphasising the crucial interplay of societal and political factors.
KL: What is your outlook on Kashmir as it comes under seismic zone 4 and 5?
DJF: First and foremost, I advocate for a thorough education on the subject and adherence to established code books. We considerably know that in Kashmir, most of the infrastructure is built without the involvement of any proper engineer and hence are not engineering designs. While I do recognise and understand the financial constraints that individuals face in our society and hiring professionals can be daunting, I still implore people to prioritise this matter as the long-term benefits would significantly outweigh the short-term costs.
Evidence underscores that abiding by design principles, professional insights, and prior experiences, coupled with investments in early warning systems for prompt assistance, can significantly alleviate this issue. Even in a country like Chile which often faces high-magnitude 7 and 8, structures hardly experience any structural damage due to their practice of over-designing buildings for earthquake resilience. Following similar processes in Kashmir may relieve some stress from a structural perspective, but again the crux lies in economic constraints. Designing and constructing such structures is often much more expensive than our simple masonry structures. But we must remember that though financial crises may drive some to cut corners, the resultant devastation outweighs the short-term gains of leniency.
KL: In context to Kashmir’s traditional Dhajji Dewari system known to be earthquake resistant which has been modified to concrete, is there some kind of an attempt that will have a sort of fusion between the traditional and the modern system to add up to disaster management?
DJF: Taq and Dhajji-Dewari are traditional construction methods used to enhance the structural resilience of buildings. Taq refers to wooden or timber-framed construction, often featuring intricate interlocking joinery, while Dhajji-Dewari involves infilling a timber frame with various materials like bricks, rubble, or even earth.
This is a difficult question to answer since as a scientist I tend to base my opinions on what I read in the broad literature and peer-reviewed studies. Knowing whether the fusion between traditional and modern systems is better or not without any tests and research studies is not right as there can be several confounding variables. However, it is important to note that these techniques have historical significance and have demonstrated some earthquake resistance in the past. Incorporating traditional techniques into modern seismic design could be a fascinating area of research and innovation, potentially leading to hybrid construction methods, however, the question is who will fund this or take the risk.
For example, after completing my bachelor’s degree, I designed polymer reinforcing bars made up of waste plastic which was proved to be superior to steel in terms of strength and durability through several engineering testing. But after that, it could not get any further because nobody agreed to fund it. This is also because whenever there is any new material in the market, no matter how beneficial it is in theory, no one will prefer to take a risk by utilizing it in their house or buildings.
Conventional materials like steel, and concrete have sustained the test of time and are relatively simpler materials to understand and predict in terms of behaviour. Generally, no structural engineer or professional will suggest or agree to take a chance by using unconventional materials in construction as it involves public risk, laws, and other validation rules. But a thorough study of such systems can surely benefit in understanding the capabilities of traditional systems against the modern ones.
KL: Is there a possibility of biotic intervention in nature and variance of a proportion of the tectonic plates?
DJF: This is again a complex question as, generally, biotic intervention, which involves the influence of living organisms on natural processes, and the variance of tectonic plate proportions are two separate concepts. However, based on the examples you gave – use any bio device whether to fill in water, to level the fragile hillocks, or intervene in the nature of the land, all I can say is that such measures can only alter the upper crust of the earth to some extent. Earthquakes generally originate deep in the earth’s lithosphere due to the tectonic plates’ movements. Over geological time, these plates can change in size due to various processes like subduction (one plate sliding under another) or seafloor spreading. The proportions of tectonic plates can indeed vary over time, but this is not influenced by biotic factors. Instead, it is driven by tectonic and geodynamic processes occurring in the earth’s interior.
I do not think any research has been conducted in this realm. Furthermore, there is a basic issue that people take correlation as causation and mix up the effects. Also, in science, anecdotal evidence does not hold any significant weight as compared to scientific studies. For example, people often claim that certain animals understand earthquakes and indeed they might do. But just because you saw dogs barking or crows crowing before an earthquake does not necessarily mean the two are correlated. These indications are not scientifically accepted and are quite hard to investigate as we would have to monitor the animals for multiple earthquakes, and no one knows how much time between the earthquakes. The same is the case with biotic interventions as they may be anecdotal, but no one has tested these hypotheses.
KL: How are deep learning and machine learning going to help us in managing devastating catastrophes?
DJF: Machine learning, deep learning, and artificial intelligence have proven to be immensely successful tools even in infrastructure-related applications. In my own research, I have witnessed a plethora of ground-breaking advancements in this specific field. However, it is important to acknowledge that while these technologies hold great promise, they might not be a panacea for all challenges.
Back in 2019 or so, there was a statistic that by 2035, 65 per cent of the jobs that people will hold have not been even created yet. So technically, we do not even have a full understanding of what to expect and prepare for. This mirrors the situation with AI itself, a domain characterised by an exponential curve that is defying any predictions. AI’s pervasive role in numerous domains has been invaluable, yet it is prudent to approach it cautiously due to the potential for errors, given its reliance on the data. One has to be extremely cautious in scrutinising the data that is used to build and train these AI models. Specifically deep learning is notorious for its garbage-in garbage-out reputation. If trained properly, deep learning and machine learning algorithms have been extremely successful in surpassing all previous methods in many prediction and decision-making tasks related to catastrophe management. And there is no doubt that in the coming years, we will see many more comprehensive models that will help the public and communities in dealing with these natural hazards.
Umaima Reshi processed the interview
