Modelling volcanic source deformation using a particle filter-based inversion
Detecting volcanic unrest is of utmost importance and requires active monitoring of several key indicators including surface deformation. In recent years, increasing satellite capabilities and ground data acquisition have made deformation data more temporally and spatially dense. However, current me...
| Κύριος συγγραφέας: | |
|---|---|
| Άλλοι συγγραφείς: | |
| Μορφή: | Final Year Project (FYP) |
| Γλώσσα: | English |
| Έκδοση: |
2019
|
| Θέματα: | |
| Διαθέσιμο Online: | http://hdl.handle.net/10356/77180 |
| _version_ | 1826114952154316800 |
|---|---|
| author | Tay, Cheryl Wen Jing |
| author2 | Benoit Taisne |
| author_facet | Benoit Taisne Tay, Cheryl Wen Jing |
| author_sort | Tay, Cheryl Wen Jing |
| collection | NTU |
| description | Detecting volcanic unrest is of utmost importance and requires active monitoring of several key indicators including surface deformation. In recent years, increasing satellite capabilities and ground data acquisition have made deformation data more temporally and spatially dense. However, current methods used to model volcanic source deformation either use limited data such as in static inversions or assume that the likelihood of models is Gaussian-distributed such as in the Kalman filter. Hence, I used an alternative method, the particle filter, which is potentially more robust than current methods as it assimilates newly received data into past data and does not require Gaussian assumptions that usually do not hold in reality. The aim of this study was to adapt the particle filter to model volcanic source deformation as it has never been applied to this field before. I trialled variants of the particle filter on synthetic data sets of differing complexities, and on real data from the 2014 dike propagation event at Bárðarbunga volcano which has good ground truth. The results show that a particle filter using a simulated annealing-based resampling with at least 750 particles was able to accurately model the evolution of a single volcanic source over time. Based on parameter misfits, data misfits and consistency, this was the most effective variant when compared to the other tested variants which used probabilistic or Metropolis-Hastings-based resampling instead. I also found that an additional optimization step improved the accuracy of modelling when there was more than one volcanic deformation source involving a greater number of model parameters. The tested particle filter variants were unable to accurately model a simulated case with multiple volcanic deformation sources and the Bárðarbunga case study. However, as this study provides an exploration into the use of particle filter in modelling volcanic source deformation, further validation of the particle filter’s performance is possible with an improved modelling approach. This could involve ensuring that the inaccuracies of models from previous times of analysis do not affect the models in future times of analysis, and using more constraints to facilitate the convergence of models towards accurate solutions. |
| first_indexed | 2024-10-01T03:47:35Z |
| format | Final Year Project (FYP) |
| id | ntu-10356/77180 |
| institution | Nanyang Technological University |
| language | English |
| last_indexed | 2024-10-01T03:47:35Z |
| publishDate | 2019 |
| record_format | dspace |
| spelling | ntu-10356/771802023-02-28T16:47:11Z Modelling volcanic source deformation using a particle filter-based inversion Tay, Cheryl Wen Jing Benoit Taisne Asian School of the Environment University of Leeds Andrew Hooper DRNTU::Science::Geology Detecting volcanic unrest is of utmost importance and requires active monitoring of several key indicators including surface deformation. In recent years, increasing satellite capabilities and ground data acquisition have made deformation data more temporally and spatially dense. However, current methods used to model volcanic source deformation either use limited data such as in static inversions or assume that the likelihood of models is Gaussian-distributed such as in the Kalman filter. Hence, I used an alternative method, the particle filter, which is potentially more robust than current methods as it assimilates newly received data into past data and does not require Gaussian assumptions that usually do not hold in reality. The aim of this study was to adapt the particle filter to model volcanic source deformation as it has never been applied to this field before. I trialled variants of the particle filter on synthetic data sets of differing complexities, and on real data from the 2014 dike propagation event at Bárðarbunga volcano which has good ground truth. The results show that a particle filter using a simulated annealing-based resampling with at least 750 particles was able to accurately model the evolution of a single volcanic source over time. Based on parameter misfits, data misfits and consistency, this was the most effective variant when compared to the other tested variants which used probabilistic or Metropolis-Hastings-based resampling instead. I also found that an additional optimization step improved the accuracy of modelling when there was more than one volcanic deformation source involving a greater number of model parameters. The tested particle filter variants were unable to accurately model a simulated case with multiple volcanic deformation sources and the Bárðarbunga case study. However, as this study provides an exploration into the use of particle filter in modelling volcanic source deformation, further validation of the particle filter’s performance is possible with an improved modelling approach. This could involve ensuring that the inaccuracies of models from previous times of analysis do not affect the models in future times of analysis, and using more constraints to facilitate the convergence of models towards accurate solutions. Bachelor of Science in Environmental Earth Systems Science 2019-05-15T05:17:48Z 2019-05-15T05:17:48Z 2019 Final Year Project (FYP) http://hdl.handle.net/10356/77180 en Nanyang Technological University 130 p. application/pdf |
| spellingShingle | DRNTU::Science::Geology Tay, Cheryl Wen Jing Modelling volcanic source deformation using a particle filter-based inversion |
| title | Modelling volcanic source deformation using a particle filter-based inversion |
| title_full | Modelling volcanic source deformation using a particle filter-based inversion |
| title_fullStr | Modelling volcanic source deformation using a particle filter-based inversion |
| title_full_unstemmed | Modelling volcanic source deformation using a particle filter-based inversion |
| title_short | Modelling volcanic source deformation using a particle filter-based inversion |
| title_sort | modelling volcanic source deformation using a particle filter based inversion |
| topic | DRNTU::Science::Geology |
| url | http://hdl.handle.net/10356/77180 |
| work_keys_str_mv | AT taycherylwenjing modellingvolcanicsourcedeformationusingaparticlefilterbasedinversion |