Good day, Dr. Soumya Kanti Maiti. I'm Arjun. I've had a look through your background and your professional record. I'm interested in hearing more about your work and how you see yourself fitting in here with our faculty. Ready to start?

Please start by introducing your professional journey in academia.

Well, presently I'm working as an associate professor at Institute of Geophysics Police Academy of Science in Poland, where I am attached with the EU Horizon and DTGO project. And the work here is basically to develop ground motion prediction model for anthropogenic seismicity and to predict future earthquakes. Prior to that I have worked at.

Eunice, that is ulsan National Institute of science and Technology.

In South Korea, where I have got the brain Pull project and my work is primarily focused on machine learning model to develop ground motion prediction equations and the probabilistic seismic hazard assessment of Korea.

Before that also I have a postdoctoral experience in Israel where my primary focus was to estimate the side effect studies for probabilistic seismic hazard assessment. I have also work experience in as a project scientist in India where I am attached to develop probabilistic seismic hazard assessment of 40 cities in India.

My PhD on probabilistic seismic hazard assessment and I have published more than 14 research journal papers in high impacted.

Yeah, journals. My work, uh, research experience mainly focused on probabilistic seismic hazard assessment, ground motion prediction models, site characterization, seismic microgenition, and risk analysis.

Thank you.

Would you like?

Yes.

Professor, could you elaborate on your most significant research contribution to the field of earthquake engineering or structural engineering? How has it impacted the academic or practical aspects of this discipline?

Well, my I have done many. My recent one is to develop the ground motion prediction equations for anthropogenic seismicity. Previously I was developing ground motion prediction equation for natural seismicity. As you know for natural seismicity we have various ground motion prediction models like NGA, W 2 and others. But for anthropogenic seismicity, the ground motion prediction model is not present quietly hardly 3:00 to 4:00. So I am now developing the ground motion prediction.

Equations for mining induced seismicity in Poland. Prior to that I have also done the effect of side effects and other factors on probability seismic hazard assessment in Indian perspective. I have developed the ground motion prediction equations for Himalayan region. For West Bengal like a continental stable continental region, I have developed ground motion prediction equations for India. Also I have developed the probabilistic seismic.

For 40 cities in India which will be used for further studies. This was a project given by Ministry of Earth Sciences. I have contributed to them.

That's impressive, Professor. Could you explain the methodological approach you used in developing the ground motion prediction equations, particularly for the Himalayan region and the mining-induced seismicity in Poland? Did you encounter any significant challenges during your work, and if so, how did you address them?

Yes. So first you are talking about two different aspect. 1 is the like the Himalayan region, 1 is the mining, both are completely different. So I will be coming 1 by 1. The for the Himalayan region I have done in my PSD. So during the Himalayan region, not only for the Himalaya, I have also done for lower Himalaya. So the main challenges I faced was not the data was not available, earthquake data was not available. Also the major earthquakes in Himalaya were.

During the 1950s where data was not available, so to incorporate the challenges, I have done the stochastic simulations to generate synthetic ground earthquakes accelerograms. So that was the challenge and there I have developed that. So this was the main challenge and I have eliminated it by stochastic ground motion simulation. Now coming to the mining induced seismicity, here there is no.

Lack of data because it is maintained well, but the main problem is the near.

Effects side effects that is like the for mining induced the earthquakes are very small like 1 to 4. But for national earthquake the earthquakes are higher. So it is a not a problem. But for here we have to incorporate the near field effect in the ground motion prediction equations otherwise it will be completely it will be out of focus. Like previously people uses the national seismicity for mining induced but that I have incorporated by giving my own.

And incorporating near field effects.

That's an excellent clarification, Professor. It demonstrates your ability to adapt methodologies to unique regional and seismic characteristics. Now, could you briefly share how your ground motion prediction equations for Himalayan earthquakes have influenced seismic risk mitigation strategies or engineering practices in that region?

Well previously they used to use the ground motion prediction equation way back, but now they are incorporating new grammar motion prediction equations. They are like site amplification factor was not incorporated previously. But for my ground motion prediction equation I've given the site.

Amplification factor incorporated also for various periods like for various frequencies. Previously it was like the PGA and .2 and one second was used but I have given for like range of equations which could be used by the structural engineers for their development of buildings and all those stuffs.

For the design purposes basically.

Understood, Professor. Incorporating the site amplification factors for various frequencies is indeed a significant advancement, especially for structural engineers. How have practitioners or policymakers, such as those in the Ministry of Earth Sciences, responded to the application of your methodologies and outcomes in practical scenarios?

My base is to do the research purpose and give the results. So I provide them with the results. But after that it's up to them whether they will implement it for real practices or not. That depends on like not only ministry, the government and all those purposes.

Well, that's out of my scope because my work was to give the results how they have incorporated in real buildings, that's ministry affairs and all that effect. So that is not my.

Can I tell one thing more?

Of course, Professor. Please go ahead.

So here it is more actually implemented.

That's truly fascinating, Professor. A dynamic, predictive model that adapts with new data and forecasts seismic events within such short timeframes is cutting-edge. Could you elaborate on the algorithm or approach that underpins this predictive capability? How does it handle the inherent uncertainties in seismic processes?

Well in seismic processes the main uncertainty is the epistemic uncertainty and also like for the ground motions so that with the mode datasets and everyday that epistemic uncertainty is gone. And for prediction I am working on various algorithms like Bayesian algorithm and all those and I take the short time period and I use a machine learning to predict it. The model like I take for two months or three months data backwards I do the.

Analysis and I predict what will be the activity rate in the near future and I validate it with one week prior datasets and then I give the results to them.

Thank you for outlining your approach, Professor. Utilizing machine learning alongside Bayesian algorithms to address epistemic uncertainty is a sophisticated method, particularly with short-term predictions. How has the validation of your model performed so far? Are the predicted activity rates aligning well with observed seismicity in these mining regions?

Uh, actually it's a new approach. So it is not, uh, I would not say I could accurately predict it that my accuracy is like around 70 to 72%. I know it's quite low, but as you know, earthquakes are unpredictable. You cannot predict like 100%. So presently my model accuracy is around 70%.

A 70% accuracy in seismic predictions, given the inherent unpredictability of earthquakes, is a commendable start, Professor. It's an evolving field, and developing a model with even this level of predictive capability is significant progress. Are there plans or areas you're exploring to further increase the accuracy or refine the model?

Well, now my future plan is to work on the earthquakes in high seismic regions like for the natural earthquake and I will see whether my model is also applicable for natural earthquakes or not. So if I get a chance to join there, then I would have the datasets and there and I will incorporate my model to predict short term. Now I like all the models previously using is like you could say probabilistic seismic hazard for 50 years.

But 50 years a long term. So my idea is to predict for the short term like 2-3 months if it is possible to have an earthquake of greater than 5 or not. So that is the model and I will be incorporating it if I get a chance on the natural earthquake and see how it comes. Because now I am only focusing on the mining based earthquake like the induced seismicity.

Your vision for applying this short-term predictive model to natural earthquakes is both ambitious and necessary, Professor. Bridging induced and natural seismicity with predictive algorithms could redefine seismic risk management. Could you discuss how you plan to adapt or modify your current models for the complexities and scale of natural seismicity? For instance, accounting for larger magnitudes, diverse tectonic settings, and potentially sparser data availability?

MMM Actually it will be quite easier I would say, because for large earthquake we could have a large data set. And as you know for if you have a large data set, my training model will have large data. So it will be, I think it will have a higher accuracy also there for the.

Natural earthquake we could incorporate a site amplification or the VS 30 as we call VS not VS 30. The 1D is here by velocity model in the uncertainty and I have done a brief research previously like incorporating all these aspects like sleep aid VS 30 defines the model validity mode. I'm talking in ground motion prediction equations perspective. But I am sure if I incorporate this perspective into machine learning model and incorporate all these data set which is.

Available right now, but for natural earthquake we have this data set by various analysis. So we could incorporate that and have a greater, I say predictability.

Thank you, Professor. Your approach to leveraging the abundance of data from natural seismicity, along with incorporating parameters like VS30 and ground motion amplification into predictive models, is well-grounded. It's clear that integrating these factors could enhance both accuracy and reliability. It seems you've laid a strong foundation to extend your work to broader seismic challenges.

Could you repeat it what you want to know?

Professor, I was acknowledging the depth of your approach and wanted to know more about how you plan to integrate parameters like VS30 and ground motion amplification with machine learning models. Specifically, could you elaborate on how these integrations might improve the accuracy and applicability of your predictions for natural seismicity?

Uh, the by prediction what you are meaning the predicting the PGA or PSA or predicting the occurrence of earthquakes, which one you are mentioning?

Like for if for prediction of the PGA or Yeah in using machine learning I have generated like machine learning based ground motion model. So all the ground motion previous model uses like.

Thank

Various factors but for all the regions all the factors are not available. So now machine learning has gained a lot of experience to estimate the PGA and PSA at various sites. So one could be incorporated apart from traditional probability seismic hazard assessment, we could incorporate machine learning based ground motion prediction equations where the predictor variables will be. Magnitude distance VS 30 VS 30 is basically gives for the site amplification.

And estimate the PGA and PSA values.

Approaches like Bayesian optimization or other approaches I have to see whatever is available. Now I am using the Bayesian optimization so there must be other optimizations technique for natural hazard and I will implement it and see how the occurrence rate increases. So by this I am also incorporating the peak ground accelerations and pseudospectral acceleration values using machine learning approaches. I will predict them and also I will predict the short term occurrence rate of the earthquakes.

Also for the occurrence rate we would we will use some.

Thank you for the detailed explanation, Professor. Integrating machine learning for estimating PGA and PSA with predictor variables like magnitude, distance, and VS30 is a powerful approach, especially when traditional models lack complete data. Combining this with Bayesian optimization for short-term occurrence rates creates a comprehensive framework for both seismic hazard assessment and prediction. It is truly a forward-thinking methodology.

Thank you.

You're welcome, Professor.

Would you want to know anything more?

What?

I’d like to understand how you ensure the robustness and validation of your machine learning models, specifically when using them for seismic predictions or hazard assessments. How do you ensure the models generalize well across different seismic regions?

For validation of the hazard model like I have a data sets for distance VS 30 and PGA and PSA with all the I will create a database and generally I train my model on 80% of the data set. I train it and I try to predict and test on the later 80 like 20% of the data set. So by that I validate my model like on the test and train data set then I.

Like R, square, RMSC and ma values and I check whether my R-squared value like I mentioned like I am getting now 70% so I want to increase it to more 80 to 8190% so that I will validate by that thing.

Thank you, Professor. Your approach of training on 80% of the dataset and testing on 20%, alongside using metrics such as R-squared, RMSE, and MAE, is a solid validation strategy. Your focus on continually improving model accuracy demonstrates a strong commitment to robust and reliable seismic predictions.

Thank you.

Or time.

You're welcome, Professor. If there's anything else you'd like to share, please let me know. If not, I appreciate your time and insights.

Well, I forgot to mention about one of my achievements like I have received a brain pull fellowships where the scientists are invited in South Korea to do the research. So I have been there for around 2 years as a Brainpool fellow and, and I have collaborated with them and presently also I collaborate with many other people in South Korea, Israel. So my main approach is also I have a collaboration for the data and all those things.

So I could bring it to VIT if I get a chance to join there.

Also, I have been working presently in the EU Horizon project in the DTGO project to develop like I told Grammar prediction equation for answering seismicity. It is under EU Horizon and DTGO project just mentioned. Okay.

Thank you for highlighting that, Professor. Receiving the Brain Pool Fellowship and fostering international collaborations with researchers from South Korea and Israel is a significant achievement and speaks to your global research impact. Your ability to bring these collaborations and datasets would certainly add immense value to an institution like VIT. This demonstrates not only your individual expertise but also your capability to enrich the academic and research environment around you.

That's excellent, Professor. Being part of the EU Horizon project and contributing towards the DTGO project for developing ground motion prediction equations for anthropogenic seismicity further underscores your active involvement in cutting-edge research. It also highlights your ability to contribute to significant international projects, bringing valuable expertise and exposure to the field. Thank you for sharing this important aspect of your current work.

Thank you for your time.

You're most welcome, Professor. It was a pleasure to hear about your groundbreaking work and achievements. Thank you for sharing your time and insights. Would you like to conclude the interview at this point?

Yes, I've told whatever is necessary.