東華大學圖書館 |

Data-Driven Approaches for Predictability and Uncertainty Quantification of Sea Level Variability /

Record Type:	Electronic resources : Monograph/item
Title/Author:	Data-Driven Approaches for Predictability and Uncertainty Quantification of Sea Level Variability // Andrew Edward Brettin.
Author:	Brettin, Andrew Edward,
Description:	1 electronic resource (144 pages)
Notes:	Source: Dissertations Abstracts International, Volume: 86-12, Section: B.
Contained By:	Dissertations Abstracts International86-12B.
Subject:	Physical oceanography. -
Online resource:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31931990
ISBN:	9798286426379

Data-Driven Approaches for Predictability and Uncertainty Quantification of Sea Level Variability /
Brettin, Andrew Edward,

Data-Driven Approaches for Predictability and Uncertainty Quantification of Sea Level Variability /Andrew Edward Brettin. - 1 electronic resource (144 pages)

Source: Dissertations Abstracts International, Volume: 86-12, Section: B.

Sea level is modulated by a variety of processes on a continuum of timescales ranging from seconds to centuries. On multidecadal to centennial timescales, global mean sea level rise is a prominent impact of anthropogenic climate change, with implications for the significant global population living in vulnerable coastal regions. Beyond long-term trends, sea level variability on shorter timescales can also be substantial. Such short-term variability can give rise to sea level extremes and exacerbate the socioeconomic impacts of sea level rise. Understanding the conditions that yield predictable variability and constraining uncertainties in sea level forecasts can help mitigate potential adverse impacts of sea level extremes. This dissertation seeks to (1) build understanding of the mechanisms contributing to predictable sea level variability, and (2) develop constrained uncertainty estimates for sea level based on observed conditions, using data-driven techniques.Chapter 2 explores sources of state-dependent predictability for dynamic sea level on daily-to-seasonal timescales. In line with previous studies which have demonstrated that certain initial states can extend predictability horizons, this study seeks to identify initial conditions that improve forecast skill for dynamic sea level. Here, sources of dynamic sea level predictability are identified using neural networks trained on CESM2 large ensemble data to forecast dynamic sea level. The forecasts yield not only a point estimate for sea level, but also a standard deviation to quantify forecast uncertainty based on the initial conditions. The forecasted uncertainties identify sources of state-dependent predictability at most locations and forecast leads, exhibit skillful deterministic predictions, and skillfully forecast exceedance probabilities relative to local linear baselines. At Guam, propagating Rossby waves are identified as a potential source of dynamic sea level predictability. In the western Indian Ocean, persistence of thermosteric sea level anomalies may be a source of predictability on subseasonal timescales, whereas on seasonal timescales, El Nino emerges as a dominant driver of predictability.In Chapter 3, the predictability of basin-scale sea level variations is examined using a suite of different data-driven dynamical propagators. Statistical methods such as Linear Inverse Models have been successful for developing data-driven forecasts, but they often rely on assumptions that can be hard to satisfy given the nonlinear dynamics of the climate system. This study uses convolutional autoencoders with a dynamical propagator in the latent space to generate forecasts of sea surface height anomalies. Learning a nonlinear dimensionality reduction and the prediction timestepping together results in a propagator that generates better predictions for daily- and monthly-averaged sea surface heights in the North Pacific and North Atlantic than if the dimensionality reduction and dynamics are learned separately. The reconstruction skill of the model highlights regions in which better representation of the state results in improved predictions, such as in the tropics for North Pacific daily SSH predictions and in the Caribbean Current for the North Atlantic.In Chapter 4, we develop a method for constraining estimates for sea level nontidal residuals using neural networks trained to predict conditional quantiles of a probability distribution. The quantile regression neural network includes two novel yet simple modifications to the loss function to enforce accuracy in quantile predictions and to mitigate the risk of predicting invalid distributions. The quantile regression neural network can predict conditional distributions on datasets in which dependencies are nonlinear or conditional distributions are non-Gaussian- situations where other methods such as linear quantile regression and Gaussian maximum likelihood networks have deficiencies. Performance of the quantile neural network at predicting nontidal residuals at US East Coast tide gauges indicates that conditional distributions of nontidal residuals may be relatively well-described as Gaussian, though functional dependencies may be nonlinear. On Tropical Rainfall Measuring Mission (TRMM) precipitation observations, the quantile regression neural network strongly outperforms linear quantile regression and Gaussian maximum likelihood neural networks, indicating skill in capturing nonlinear dependencies and non-Gaussian distributions. The performance of the quantile neural network on a diverse set of datasets indicates that it is a flexible yet general approach which can constrain aleatoric uncertainties for geophysical quantities which may have non-linear or non-Gaussian conditional dependencies.

English

ISBN: 9798286426379Subjects--Topical Terms:

3168433
Physical oceanography.
Subjects--Index Terms:

Machine learning

Data-Driven Approaches for Predictability and Uncertainty Quantification of Sea Level Variability /
LDR:06313nmm a22004453i 4500 001 2437401
005 20251119142808.5
006 m o d
007 cr|nu||||||||
008 261011s2025 miu||||||m |||||||eng d
020 $a 9798286426379
035 $a (MiAaPQD)AAI31931990
035 $a AAI31931990
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Brettin, Andrew Edward, $e author. $3 3828361
245 1 0 $a Data-Driven Approaches for Predictability and Uncertainty Quantification of Sea Level Variability / $c Andrew Edward Brettin.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2025
300 $a 1 electronic resource (144 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 86-12, Section: B.
500 $a Advisors: Zanna, Laure Committee members: Barnes, Elizabeth A.; Shamekh, Sara; Fernandes-Granda, Carlos; Smith, Shafer.
502 $b Ph.D. $c New York University $d 2025.
520 $a Sea level is modulated by a variety of processes on a continuum of timescales ranging from seconds to centuries. On multidecadal to centennial timescales, global mean sea level rise is a prominent impact of anthropogenic climate change, with implications for the significant global population living in vulnerable coastal regions. Beyond long-term trends, sea level variability on shorter timescales can also be substantial. Such short-term variability can give rise to sea level extremes and exacerbate the socioeconomic impacts of sea level rise. Understanding the conditions that yield predictable variability and constraining uncertainties in sea level forecasts can help mitigate potential adverse impacts of sea level extremes. This dissertation seeks to (1) build understanding of the mechanisms contributing to predictable sea level variability, and (2) develop constrained uncertainty estimates for sea level based on observed conditions, using data-driven techniques.Chapter 2 explores sources of state-dependent predictability for dynamic sea level on daily-to-seasonal timescales. In line with previous studies which have demonstrated that certain initial states can extend predictability horizons, this study seeks to identify initial conditions that improve forecast skill for dynamic sea level. Here, sources of dynamic sea level predictability are identified using neural networks trained on CESM2 large ensemble data to forecast dynamic sea level. The forecasts yield not only a point estimate for sea level, but also a standard deviation to quantify forecast uncertainty based on the initial conditions. The forecasted uncertainties identify sources of state-dependent predictability at most locations and forecast leads, exhibit skillful deterministic predictions, and skillfully forecast exceedance probabilities relative to local linear baselines. At Guam, propagating Rossby waves are identified as a potential source of dynamic sea level predictability. In the western Indian Ocean, persistence of thermosteric sea level anomalies may be a source of predictability on subseasonal timescales, whereas on seasonal timescales, El Nino emerges as a dominant driver of predictability.In Chapter 3, the predictability of basin-scale sea level variations is examined using a suite of different data-driven dynamical propagators. Statistical methods such as Linear Inverse Models have been successful for developing data-driven forecasts, but they often rely on assumptions that can be hard to satisfy given the nonlinear dynamics of the climate system. This study uses convolutional autoencoders with a dynamical propagator in the latent space to generate forecasts of sea surface height anomalies. Learning a nonlinear dimensionality reduction and the prediction timestepping together results in a propagator that generates better predictions for daily- and monthly-averaged sea surface heights in the North Pacific and North Atlantic than if the dimensionality reduction and dynamics are learned separately. The reconstruction skill of the model highlights regions in which better representation of the state results in improved predictions, such as in the tropics for North Pacific daily SSH predictions and in the Caribbean Current for the North Atlantic.In Chapter 4, we develop a method for constraining estimates for sea level nontidal residuals using neural networks trained to predict conditional quantiles of a probability distribution. The quantile regression neural network includes two novel yet simple modifications to the loss function to enforce accuracy in quantile predictions and to mitigate the risk of predicting invalid distributions. The quantile regression neural network can predict conditional distributions on datasets in which dependencies are nonlinear or conditional distributions are non-Gaussian- situations where other methods such as linear quantile regression and Gaussian maximum likelihood networks have deficiencies. Performance of the quantile neural network at predicting nontidal residuals at US East Coast tide gauges indicates that conditional distributions of nontidal residuals may be relatively well-described as Gaussian, though functional dependencies may be nonlinear. On Tropical Rainfall Measuring Mission (TRMM) precipitation observations, the quantile regression neural network strongly outperforms linear quantile regression and Gaussian maximum likelihood neural networks, indicating skill in capturing nonlinear dependencies and non-Gaussian distributions. The performance of the quantile neural network on a diverse set of datasets indicates that it is a flexible yet general approach which can constrain aleatoric uncertainties for geophysical quantities which may have non-linear or non-Gaussian conditional dependencies.
546 $a English
590 $a School code: 0146
650 4 $a Physical oceanography. $3 3168433
650 4 $a Statistics. $3 517247
650 4 $a Computer engineering. $3 621879
653 $a Machine learning
653 $a Sea level variability
653 $a Uncertainty quantification
653 $a CESM2
653 $a Timestepping
690 $a 0415
690 $a 0463
690 $a 0800
690 $a 0464
710 2 $a New York University. $b Mathematics. $e degree granting institution. $3 3765635
720 1 $a Zanna, Laure $e degree supervisor.
773 0 $t Dissertations Abstracts International $g 86-12B.
790 $a 0146
791 $a Ph.D.
792 $a 2025
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31931990