Estimating Inhomogeneous Spatio-Temporal Background Intensity Functions using Graphical Dirichlet Processes
Isaías Bañales, Tomoaki Nishikawa, Yoshihiro Ito, Manuel J. Aguilar-Velázquez · 2025
Read Estimating Inhomogeneous Spatio-Temporal Background Intensity Functions using Graphical Dirichlet Processes by Isaías Bañales, Tomoaki Nishikawa, Yoshihiro Ito, Manuel J. Aguilar-Velázquez (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-04974.txt.
Introduction
Isaías Bañales and co-authors build a new way to model background quake rates that shift in both place and time. Dense sensor nets now record more small quakes, while strong quakes or slow slip events can also change local rates.
Their model treats background events as a Poisson process whose strength can vary across space and over time. It uses graphical Dirichlet processes, a Bayesian non-parametric tool, to track that change with honest doubt bounds. They apply the method to quakes in southern Mexico, where rates are far from flat.
Catalog record
| CREATOR | Isaías Bañales, Tomoaki Nishikawa, Yoshihiro Ito, Manuel J. Aguilar-Velázquez |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 34 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-04974
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Estimating Inhomogeneous Spatio-Temporal Background Intensity Functions using Graphical Dirichlet Processes Isaías Bañales, Tomoaki Nishikawa, Yoshihiro Ito, Manuel J. Aguilar-Velázquez Abstract An enhancement in seismic measuring instrumentation has been proven to have implications in the quantity of observed earthquakes, since denser networks usually allow recording more events. However, phenomena such as strong earthquakes or even aseismic transients, as slow slip earthquakes, may alter the occurrence of earthquakes. In the field of seismology, it is a standard practice to model background seismicity as a Poisson process. Based on this idea, this work proposes a model that can incorporate the evolving spatial intensity of Poisson processes over time (i.e., we include temporal changes in the background seismicity when modeling). In recent years, novel methodologies have been developed for quantifying the uncertainty in the estimation of the background seismicity in homogeneous cases using Bayesian non-parametric techniques. This work proposes a novel methodology based on graphical Dirichlet processes for incorporating spatial and temporal inhomogeneities in background seismicity. The proposed model in this work is applied to study the seismicity in the southern Mexico, using recorded data from 2000 to 2015. keywords ETAS, Dirichlet Process, Bayesian Nonparametrics, Background seismicity † † equal-contributors: These authors contributed equally to this work. † † equal-contributors: These authors contributed equally to this work. † † equal-contributors: These authors contributed equally to this work. 1 Introduction Since its introduction, the Epidemic Type Aftershock Sequence (ETAS) model ( 38 ) has proved to be one of the most important tools for assessing stochastic seismicity, in which earthquake activity is modeled using Hawkes’ process ( 22 ) . In this context, the work by ( 38 ) assumed that the mainshocks (or background earthquakes) originate from a Poisson process. Recently, due to the importance of having uncertainty quantification of the estimators, such authors as 46 and 34 have worked with Bayesian non-parametric approaches to model a temporally homogeneous background intensity function \mu(x,y) in equation ( 1 ) using Dirichlet processes mixtures (DPM) ( 1 ) and Gaussian processes, respectively. Since the homogeneity of the ETAS model can be affected by tectonic or pore pressure changes ( 18 ; 19 ; 31 ) , such phenomena may have implications for the occurrence of tremors. However, these changes can be produced by strong earthquakes 7 , and can also lead to an increase in the seismicity rate ( 36 ) . Furthermore, the presence of slow slip events—slow fault slip phenomena that mainly occur along the plate interface—can alter seismic activity ( 30 ; 14 ; 37 ) , as well as periods of microearthquake quiescence ( 32 ) . Moreover, not only variations in seismicity can affect ETAS model estimates, but also factors such as short-term aftershock incompleteness ( 20 ; 2 ) and changes in the cut-off magnitude ( 21 ) . These temporary changes in activity emphasize the need for non-homogeneous models over time for the background activity function, such as the one introduced in this paper. Consequently, this work contribution lies on the following key points: • We propose a nonparametric Bayesian approach to estimate the background seismic function based on Graphical Dirichlet Processes (GDP) ( 5 ) , which allows for inhomogeneities in both space and time. • We focus on inferring the background intensity adding inhomogeneities in time, i.e. \mu(x,y,t) , following a GDP as proposed by 5 . In the context of tectonic activity, this variation of the background seismic activity model holds significance, as it may provide valuable insights into the actual tectonic physics that govern earthquake occurrence. • We only study the background intensity function because the 47 database is used. Its main feature lies in the fact that aftershocks, foreshocks, and earthquake swarms were already removed using the methodology proposed by 15 , leaving only mainshocks (background events) available (i.e., we will only consider the Poissonian intensity). As discussed in detail in Section 2 , the usage of hidden variables in our methodology can be extended and incorporated in the context of the ETAS model, enhancing the methodology by 46 . In the frequentist context, 28 have used piecewise constant functions and kernel density estimators to model an inhomogeneous space-time background activity function in the context of the Hawkes process. They assumed that \mu(x,y,t)=\gamma f(x,y)v(t), where \gamma\in\mathbb{R} , f is a probability density function, and v is a real-valued function. Nevertheless, the assumption that space changes can be separated from changes in time is generally not realistic for earthquake activity. In the non-parametric Bayesian framework, the estimation of \mu(x,y,t) can be done using the Generalized Spatial Dirichlet Process introduced by 11 . Their approach allows modeling different probability density functions across space and discretized time, where the weights and point masses in the stick-breaking representation ( 48 ) are assumed to vary smoothly over space and time. However, their formulation requires modeling the weights through an auxiliary thresholding scheme based on a collection of independent stationary Gaussian random fields. Since we keep the mass points fixed in our work, we have chosen to use the representation proposed by 5 , which allows the weights to vary through a hierarchical model with beta distributions. It is also worth mentioning the 12 work, which deals with the spatio-temporal problem that occurs when point data are not directly observed and only aggregated counts are available. In this context, 24 proposed a non-parametric Bayesian approach to estimate the time-homogeneous spatial Poisson-process intensity function using Dirichlet processes, work that has been extended to the non-homogeneous case by 56 using dependent Dirichlet processes (DDP) ( 29 ) . If the reader is interested in DDP and their stick-breaking representation, we suggest consulting the work 41 . The main difference between the methodology by 56 and the one proposed in our work lies on the DDP structure; while Kottas keeps the weights of the mixing elements fixed over time during the Dirichlet process (while varying their locations), we keep the locations fixed and move the weights, which will be discussed in detail in section 2 . The idea of keeping the mixture locations constant over time is reasonable, given that regions where tectonic plates interact require very long periods to undergo significant changes. Therefore, the main seismogenic regions will remain stable in their locations. When a change in seismicity occurs, for example, due to a strong earthquake or a slow slip event (SSE), the weight associated with the mixture component in these regions will deviate from zero, but only during the periods near the occurrence of the event. This, combined with the fact that the number of parameters to be estimated using the GDP is smaller, makes the approach presented in this work particularly attractive in the context of seismology. 2 Methodology As is pointed out by 13 , the usual approach to model the mainshocks (i.e., earthquakes that are not produced by other earthquakes, as the aftershocks), is to use an inhomogeneous Poisson Process in space and constant over time with intensity \mu(x,y) , and the rate of aftershocks is modeled using a triggering function \nu depending on the magnitude ( M ) which decays according to the spatial and temporal distance between earthquakes. Then, the intensity function in the ETAS