# Serge Guillas's research while affiliated with University College London and other places

**What is this page?**

This page lists the scientific contributions of an author, who either does not have a ResearchGate profile, or has not yet added these contributions to their profile.

It was automatically created by ResearchGate to create a record of this author's body of work. We create such pages to advance our goal of creating and maintaining the most comprehensive scientific repository possible. In doing so, we process publicly available (personal) data relating to the author as a member of the scientific community.

If you're a ResearchGate member, you can follow this page to keep up with this author's work.

If you are this author, and you don't want us to display this page anymore, please let us know.

It was automatically created by ResearchGate to create a record of this author's body of work. We create such pages to advance our goal of creating and maintaining the most comprehensive scientific repository possible. In doing so, we process publicly available (personal) data relating to the author as a member of the scientific community.

If you're a ResearchGate member, you can follow this page to keep up with this author's work.

If you are this author, and you don't want us to display this page anymore, please let us know.

## Publications (53)

Multilevel Monte Carlo is a key tool for approximating integrals involving expensive scientific models. The idea is to use approximations of the integrand to construct an estimator with improved accuracy over classical Monte Carlo. We propose to further enhance multilevel Monte Carlo through Bayesian surrogate models of the integrand, focusing on G...

Deep Gaussian processes (DGPs) provide a rich class of models that can better represent functions with varying regimes or sharp changes, compared to conventional GPs. In this work, we propose a novel inference method for DGPs for computer model emulation. By stochastically imputing the latent layers, our approach transforms a DGP into a linked GP:...

We develop a new method to analyze the total electron content (TEC) depression in the ionosphere after a tsunami occurrence. We employ Gaussian process regression to accurately estimate the TEC disturbance every 30 s using satellite observations from the global navigation satellite system (GNSS) network, even over regions without measurements. We f...

Markov Chain Monte Carlo (MCMC) is one of the most powerful methods to sample from a given probability distribution, of which the Metropolis Adjusted Langevin Algorithm (MALA) is a variant wherein the gradient of the distribution is used towards faster convergence. However, being set up in the Euclidean framework, MALA might perform poorly in highe...

The potential of a full-margin rupture along the Cascadia subduction zone poses a significant threat over a populous region of North America. Previous probabilistic tsunami hazard assessment studies produced hazard curves based on simulated predictions of tsunami waves, either at low resolution or at high resolution for a local area or under limite...

Instrumental global temperature records are derived from the network of in situ measurements of land and sea surface temperatures. This observational evidence is seen as being fundamental to climate science. Therefore, the accuracy of these measurements is of prime importance for the analysis of temperature variability. There are spatial gaps in th...

The state-of-the-art linked Gaussian process offers a way to build analytical emulators for systems of computer models. We generalize the closed form expressions for the linked Gaussian process under the squared exponential kernel to a class of Mat\'ern kernels, that are essential in advanced applications. An iterative procedure to construct linked...

We develop a new method to analyze the total electron content (TEC) depression in the ionosphere after a tsunami occurrence. We employ Gaussian process regression to accurately estimate the TEC disturbance every 30 s using satellite observations from the GNSS network, even over regions without measurements. We face multiple challenges. First, the i...

We propose a novel deep Gaussian process (DGP) inference method for computer model emulation using stochastic imputation. By stochastically imputing the latent layers, the approach transforms the DGP into the linked GP, a state-of-the-art surrogate model formed by linking a system of feed-forward coupled GPs. This transformation renders a simple wh...

In this paper, statistical emulation is shown to be an essential tool for the end-to-end physical and numerical modelling of local tsunami impact, i.e. from the earthquake source to tsunami velocities and heights. In order to surmount the prohibitive computational cost of running a large number of simulations, the emulator, constructed using 300 tr...

This paper presents the first end-to-end example of a risk model for loss of assets in households due to possible future tsunamis. There is a significant need for Government to assess the generic risk to buildings, and the concrete impact on the full range of assets of households, including the ones that are key to livelihoods such as agricultural...

Tsunamis are unpredictable and infrequent but potentially large impact natural disasters. To prepare, mitigate and prevent losses from tsunamis, probabilistic hazard and risk analysis methods have been developed and have proved useful. However, large gaps and uncertainties still exist and many steps in the assessment methods lack information, theor...

Investigating uncertainties in computer simulations can be prohibitive in terms of computational costs, since the simulator needs to be run over a large number of input values. Building an emulator, i.e. a statistical surrogate model of the simulator, using a small design of experiments, greatly alleviates the computational burden to carry out such...

The potential of a full-margin rupture along the Cascadia subduction zone poses a significant threat over a populous region of North America. Traditional probabilistic tsunami hazard assessments produce hazard maps based on simulated prediction of tsunami waves either under limited ranges of scenarios or at low resolution, due to cost. We use the G...

Tsunamis are unpredictable events and catastrophic in their potential for destruction of human lives and economy. The unpredictability of their occurrence poses a challenge to the tsunami community, as it is difficult to obtain from the tsunamigenic records estimates of recurrence rates and severity. Accurate and efficient mathematical/computationa...

Instrumental temperature records are derived from the network of in situ measurements of land and sea surface temperatures. This observational evidence is seen as fundamental to climate science. Therefore, the accuracy of these measurements is of prime importance for the analysis of temperature variability. There are spatial gaps in the distributio...

The software package Volna-OP2 is a robust and efficient code capable of simulating the complete life cycle of a tsunami whilst harnessing the latest High Performance Computing (HPC) architectures. In this paper, a comprehensive error analysis and scalability study of the GPU version of the code is presented. A novel decomposition of the numerical...

Pre-print available at: https://www.essoar.org/doi/abs/10.1002/essoar.10502534.2

Pre-print available at:
https://www.essoar.org/doi/abs/10.1002/essoar.10502534.1

The software package Volna-OP2 is a robust and efficient code capable of simulating the complete life cycle of a tsunami whilst harnessing the latest High Performance Computing (HPC) architectures. In this paper, a comprehensive error analysis and scalability study of the GPU version of the code is presented. A novel decomposition of the numerical...

We generalize the state-of-the-art linked emulator for a system of two computer models under the squared exponential kernel to an integrated emulator for any feed-forward system of multiple computer models, under a variety of kernels (exponential, squared exponential, and two key Matérn kernels) that are essential in advanced applications. The inte...

A natural, if idealised, picture of the role of risk assessments in planning sees decision-makers drawing on the risk projections provided by natural and social scientific models and fashioning policies or plans that maximise expected benefit relative to this information. In this paper we draw on our study of the use tsunami science in development...

Storm surges cause coastal inundations due to the setup of the water surface resulting from atmospheric pressure, surface winds and breaking waves. The latter is particularly difficult to be accounted for. For instance, it was observed that during Typhoon Haiyan (2013, Philippines), a stretch of coral reef near the coast, which was expected to prot...

The computational burden of running a complex computer model can make optimization impractical. Gaussian Processes (GPs) are statistical surrogates (also known as emulators) that alleviate this issue since they cheaply replace the computer model. As a result, the exploration vs. exploitation trade-off strategy can be accelerated by building a GP su...

This paper provides a timely review of progress and ongoing research needs in tsunami hazard and risk science since the most recent major event, the Tohoku tsunami in 2011. The tsunami community has made significant progress in understanding tsunami hazard from seismic sources. However, this is only part of the inputs needed to effectively manage t...

The Indus Canyon in the northwestern Indian Ocean has been reported to be the site of numerous submarine mass failures in the past. This study is the first to investigate potential tsunami hazards associated with such mass failures in this region. We employed statistical emulation, i.e. surrogate modelling, to efficiently quantify uncertainties ass...

The rarity of tsunamis impels the scientific community to rely on numerical simulation for planning and risk assessment purposes because of the low availability of actual data from historic events. Numerical models, also called simulators, typically produce time series of outputs. Due to the large computational cost of such simulators, statistical...

In this paper, we present the VOLNA-OP2 tsunami model and implementation; a finite-volume non-linear shallow-water equation (NSWE) solver built on the OP2 domain-specific language (DSL) for unstructured mesh computations. VOLNA-OP2 is unique among tsunami solvers in its support for several high-performance computing platforms: central processing un...

Bayesian calibration of computer models tunes unknown input parameters by comparing outputs with observations. For model outputs that are distributed over space, this becomes computationally expensive because of the output size. To overcome this challenge, we employ a basis representation of the model outputs and observations: we match these decomp...

The submarine canyon of the Indus River in the North Indian Ocean spreads for a length larger than 185 km and constitutes one of the most prominent channel-levee systems in the region. Past bathymetric surveys have identified plethora of evidence that exhibit submarine slumping on the slopes of the Indus Canyon. In this study, we use recent high-re...

In this paper, we present the VOLNA-OP2 tsunami model and implementation; a finite volume non-linear shallow water equations (NSWE) solver built on the OP2 domain specific language for unstructured mesh computations. VOLNA-OP2 is unique among tsunami solvers in its support for several high performance computing platforms: CPUs, the Intel Xeon Phi,...

Recent scientific research indicates that the Rockall Bank Slide Complex in the NE Atlantic Ocean has formed as the result of repetitive slope failures that can be distinguished in at least three major phases. These sliding episodes took place during and before the Last Glacial Maximum. This work attempts the modelling of each sliding episode with...

Numerical inversions for earthquake source parameters from tsunami wave data usually incorporate subjective elements to stabilize the search. In addition, noisy and possibly insufficient data result in instability and non-uniqueness in most deterministic inversions, which are barely acknowledged. Here, we employ the satellite altimetry data for the...

Instrumental records showing increases in surface temperature are some of the robust and iconic evidence of climate change. But how much should we trust regional temperature estimates interpolated from sparse observations? Here we quantify the uncertainty in the instrumental record by applying multiresolution lattice kriging, a recently developed i...

Statistical methods constitute a useful approach to understand and quantify the uncertainty that governs complex tsunami mechanisms. Numerical experiments may often have a high computational cost. This forms a limiting factor for performing uncertainty and sensitivity analyses, where numerous simulations are required. Statistical emulators, as surr...

Computer simulators can be computationally intensive to run over a large number of input values, as required for optimization and various uncertainty quantification tasks. The standard paradigm for the design and analysis of computer experiments is to employ Gaussian random fields to model computer simulators. Gaussian process models are trained on...

Usual inversion for earthquake source parameters from tsunami wave data incorporates subjective elements. Noisy and possibly insufficient data also results in instability and non-uniqueness in most deterministic inversions. Here we employ the satellite altimetry data for the 2004 Sumatra-Andaman tsunami event to invert the source parameters. Using...

In this paper, we introduce a new procedure for the estimation in the nonlinear functional regression model where the explanatory variable takes values in an abstract function space and the residual process is autocorrelated. Moreover, we consider the case where the response variable takes its values in . The procedure consists in a pre-whitening t...

The Bayesian computer model calibration method has proven to be effective in a wide range of applications. In this framework, input parameters are tuned by comparing model outputs to observations. However, this methodology becomes computationally expensive for large spatial model outputs. To overcome this challenge, we employ a truncated basis repr...

High accuracy complex computer models, also called simulators, require large resources in time and memory to produce realistic results. Statistical emulators are computationally cheap approximations of such simulators. They can be built to replace simulators for various purposes, such as the propagation of uncertainties from inputs to outputs or th...

Total column ozone variations estimated using ground-based stations
provide important independent source of information in addition to
satellite-based estimates. This estimation has been vigorously
challenged by data inhomogeneity in time and by the irregularity of
the spatial distribution of stations, as well as by interruptions in
observation rec...

Submarine scarps of large extent have been observed on the eastern margin of the Rockall Bank, c.400km NW of Ireland, in the NE Atlantic Ocean. The scarps provide evidence that major submarine mass movements have taken place in the area over the past 16 ka. The Rockall Bank Slide Complex (RBSC) has dimensions of 120 by 150km (width-to-length aspect...

Total column ozone variations estimated using ground-based stations provide important independent source of information in addition to satellite-based estimates. This estimation has been vigorously challenged by data inhomogeneity in time and by the irregularity of the spatial distribution of stations, as well as by interruptions in observa-5 tion...

Gaussian fields (GFs) are frequently used in spatial statistics for their
versatility. The associated computational cost can be a bottleneck, especially
in realistic applications. It has been shown that computational efficiency can
be gained by doing the computations using Gaussian Markov random fields (GMRFs)
as the GFs can be seen as weak solutio...

In this paper we carry out a Bayesian calibration for uncertainty analysis in Computational Fluid Dynamics modelling of urban flows. Taking the case of airflow in a regular street canyon, and choosing turbulent kinetic energy (TKE) as our quantity of interest, we calibrate 3-D CFD simulations against wind tunnel observations. We focus our calibrati...

In this paper, we forecast ground level ozone concentrations over the USA, using past spatially distributed measurements and the functional linear regression model. We employ bivariate splines defined over triangulations of the relevant region of the USA to implement this functional data approach in which random surfaces represent ozone concentrati...

CFD simulations of complex outdoor environments present a significant modelling challenge. Simulations of airflow within an idealized street canyon are performed here. We test the model sensitivity to the empirical constants contained within the κ-ε turbulence model and examine how a systematic variation of these values could produce improved predi...

CFD simulations of complex outdoor environments present a significant modelling challenge. Simulations of airflow within an idealized street canyon are per-formed here. We test the model sensitivity to the em-pirical constants contained within the k-ε turbulence model and examine how a systematic variation of these values could produce improved pre...

## Citations

... The latter modeling toolbox is developed by ONERA, ISAE-Supaero, ICA (CNRS), NASA Glenn and the University of Michigan [26]. Our modeling software is free and open-source and has been used regularly in the aircraft industry, for example with a deep learning model [27][28][29][30] or with a deep gaussian process [31,32]. ...

... Future work worthy of investigation include DGP emulator-based sensitivity analysis, Bayesian optimization, and calibration, taking advantage of the DGP emulators' analytically tractable mean and variance implemented in SI. Coupling SI with sequential design (Beck and Guillas 2016;Salmanidou, Beck, and Guillas 2021) to further reinforce the predictive performance of DGP emulators with reduced computational costs is another promising research direction. Applications of sequential designs to FB-based DGP emulation are explored by Sauer, Gramacy, and Higdon (2022). ...

... It is algorithmically effective and straightforward for DGP surrogate modeling with different hierarchical structures. Unlike other studies that treat DGPs simply as compositions of GPs, we see DGPs through the lenses of linked GPs (Kyzyurova, Berger, and Wolpert 2018;Ming and Guillas 2021) that enjoy a simple and fast inference procedure. By exploiting the idea that a linked GP can be viewed as a DGP with its hidden layers exposed, our approach is to convert DGPs to linked GPs by stochastically imputing the hidden layers of DGPs. ...

... Most of the novel techniques in the field of PTHA are based on the notion of reducing the number of required computational runs with the aid of Gaussian process emulators, which are capable of maintaining good output accuracy and uncertainty quantification. The investigations of Gopinathan et al. [20] and Salmanidou et al. [21] are good examples of this approach, where the former delivered millions of output predictions based on 300 numerically simulated earthquake-tsunami scenarios, and the latter produced 2000 output predictions at each prescribed location, examining 60 full-fledged simulations. ...

... The victims are trapped inside buildings and unable to evacuate themselves during the earthquake. The next cause of death was that the victims were carried away by the current of the tsunami waves, which had high speed [13]. Hence they had no chance to escape. ...

... Furthermore, due to the low frequency, scarcity of data, and complexity of the phenomenon, tsunami hazard and risk assessment are affected by a strong component of uncertainty, that also influences people's perception and risk communication (see e.g., Behrens et al., 2021;Lorito et al., 2022;Rafliana et al., 2022). ...

... Although, in some cases, tsunamis reach the coast very fast, to apply our method there must be a minimum window of almost 10 minutes between generation and arrival. However, this is perfectly valid for 375 tsunami hazard assessment over populous regions with larger arrival times, as for example tsunami hazard assessment in the city of Victoria, British Columbia, from a tsunami generated in the Cascadia subduction zone (Salmanidou et al., 2021). Our implementation on the 2011 Tohoku Earthquake in Japan demonstrates that our method works well there. ...

... Geological activities leading to tsunamis are unpredictable. 39 Field observation data on tsunami propagation and interaction with structures are scarce. 40 The tide is a periodic fluctuation phenomenon caused by the gravitational pull of the moon, the sun, and other celestial bodies on the ocean surface, and the tidal bore is at the front of the tidal wave. ...

... For example, a scientist might be uncertain about the value of certain model parameters, and might therefore wish to estimate the expected value of some quantity of interest involving the model with respect to distributions on these parameters. An example which illustrates this problem (and which will be revisited in Section 5) is the modelling of landslidegenerated tsunamis, where the evolution of the wave through space and time is described through a complex system of differential equations (Behrens and Dias, 2015;Giles et al., 2020;Marras and Mandli, 2021;Reguly et al., 2018); see Figure 1 for an illustration. In this context, designers of tsunami resistant buildings, prevention structures or early warning systems might be interested in estimating the total wave energy or momentum flux of the tsunami at a fixed location. ...

Reference: Multilevel Bayesian Quadrature

... To account for that, the nascent leading approach has been to run the computer model over a small design of experiments and use this to create a surrogate, also called an emulator. Such an approach requires care in the design and the process of emulation (e.g. the parameterisation of the source), but has already shown success in various regions, such as Cascadia, NE Atlantic and the Western Indian OceanSalmanidou et al. [2017],Guillas et al. [2018], Salmanidou et al. [2019b,Gopinathan et al. [2020b]. An important source of uncertainties is the bathymetry in the coastal areas that have not been thoroughly surveyed; for Indonesia and other countries with extremely long and complex coastlines, it can be a challenge. ...