Predicting ENSO using Gaussian Density Neural Networks trained on distorted physics data
dc.rights.license | CC-BY-NC-ND | |
dc.contributor | H. A. Dijkstra | |
dc.contributor.advisor | Dijkstra, H.A. | |
dc.contributor.author | Goede, Ivo | |
dc.date.accessioned | 2022-04-13T00:00:59Z | |
dc.date.available | 2022-04-13T00:00:59Z | |
dc.date.issued | 2022 | |
dc.identifier.uri | https://studenttheses.uu.nl/handle/20.500.12932/41491 | |
dc.description.abstract | This research explores the performance of the Gaussian Density Neural Network (GDNN) framework’s performance predicting the El Nino Southern Oscillation (ENSO) when trained using distorted physics data produced by the Cane & Zebiak 1987 climate model (CZ87). Distorting the wavespeed and thermocline feedback in CZ87 shows a deterioration of the prediction skill of the GDNN at a 9 month lead time. Also shown is a notable capacity of the GDNN to account for differences in amplitude and period in the oscillation the target variable between test and distorted training datasets. Subsequent study may uncover new dynamical relationships at the core of the ENSO phenomenon. | |
dc.description.sponsorship | Utrecht University | |
dc.language.iso | EN | |
dc.subject | A project aiming to make insightful the physical relationships used by a certain artificial neural network trained to predict the El Niño climate phenomenon. To achieve this goal the neural network is trained on data from the Zebiak Cane 1987 climate model. The climate model data is purposely manipulated in order to generate a change in the behaviour of the predicting network thereby hopefully explaining the network's inner workings. | |
dc.title | Predicting ENSO using Gaussian Density Neural Networks trained on distorted physics data | |
dc.type.content | Master Thesis | |
dc.rights.accessrights | Open Access | |
dc.subject.keywords | ENSO; ML; Machine Learning; neural networks; pacific ocean; ZebiakCane1987 | |
dc.subject.courseuu | Climate Physics | |
dc.thesis.id | 3334 |