Deep Learning for Geophysics: Current and Future Trends

Yu, Shi; Ma, Jianwei

doi:10.1029/2021rg000742

Cited by 284 publications

(74 citation statements)

References 165 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We use these labeling experiments and their resulting scores to quantitatively examine disagreement and its causes. Results from this study can inform future labeling work, which will continue to be necessary as ML use in the Earth and environmental science continues to rise (e.g., Karpante et al, 2018;Goldstein et al, 2019;Yu and Ma 2021;Razavi et al, 2021).…”

Section: Accepted Articlementioning

confidence: 92%

“…2019; Yu and Ma 2021;Razavi et al, 2021). Correctly labeled data are also critical for building trustworthy models, which is a current focus for environmental ML research (e.g., McGovern et al, 2021).…”

Section: Accepted Articlementioning

confidence: 99%

“…Labeling tasks will likely continue to be necessary in tandem with increased adoption of ML use for Earth and environmental science studies (e.g., Goldstein et al, 2019;Karpante et al, 2018;Razavi et al, 2021;Yu & Ma, 2021). Correctly labeled data are also critical for building trustworthy models, which is a current focus for environmental ML research (e.g., McGovern et al, 2020).…”

Section: Next Steps For Labeling Studies In Earth and Environmental S...mentioning

confidence: 99%

See 2 more Smart Citations

Labeling Poststorm Coastal Imagery for Machine Learning: Measurement of Interrater Agreement

Goldstein

Buscombe

Lazarus

et al. 2021

Earth and Space Science

View full text Add to dashboard Cite

show abstract

Section: Accepted Articlementioning

confidence: 92%

Section: Accepted Articlementioning

confidence: 99%

Section: Next Steps For Labeling Studies In Earth and Environmental S...mentioning

confidence: 99%

See 1 more Smart Citation

Labeling Poststorm Coastal Imagery for Machine Learning: Measurement of Interrater Agreement

Goldstein

Buscombe

Lazarus

et al. 2021

Earth and Space Science

View full text Add to dashboard Cite

show abstract

“…In contrast, as a type of data-driven approach, machine learning methods train a model with adjustable parameters to represent the nonlinear and complicated relations between a response variable and the predictors (e.g., Yu & Ma, 2021). Thus, given the same set of predictors, an improved prediction and representation of cross-stream mixing should be possible using machine learning methods instead of the dynamical theory SMLT.…”

Section: Smltmentioning

confidence: 99%

Seasonal Surface Eddy Mixing in the Kuroshio Extension: Estimation and Machine Learning Prediction

et al. 2022

View full text Add to dashboard Cite

Results of coarse‐resolution climate models are sensitive to the specification of ocean eddy mixing coefficients. Therefore, it is important to estimate, rationalize and predict eddy diffusivities. Here, we estimate the seasonal variability of surface eddy diffusivities in the Kuroshio Extension region using numerical particles advected by a submesoscale‐permitting model solution. We find that both the spatial structure and the domain‐averaged value of the particle‐based eddy diffusivities have a significant seasonal cycle. We also assess the predictability of cross‐stream mixing lengths in this region using the methods of machine learning, suppressed mixing length theory (SMLT), and multiple linear regression (LR). The predictors we choose are all variables from SMLT that represent eddy‐ and mean‐flow properties, and these predictors correlate well with the particle‐based cross‐stream mixing lengths. We demonstrate that, compared to SMLT and LR, machine learning methods, in particular the random forest (RF) and convolutional neural network (CNN), can better represent both the spatial structure and the domain‐averaged value of cross‐stream mixing lengths. The skill in predicting the mixing lengths with CNN has much less seasonal variability than that with RF. Our results indicate that the machine learning approach may be useful in future development of eddy parameterization schemes.

show abstract

“…Deep learning has been rapidly and widely applied to geophysical investigations (Bergen et al, 2019;Yu & Ma, 2021). In some studies of exploration geophysics, it has produced results comparable to and in some cases surpassing expert human performance (Bi et al, 2021;Liang et al, 2019;Reading et al, 2015).…”

mentioning

confidence: 99%

Deep Learning‐BasedH‐κMethod (HkNet) for Estimating Crustal Thickness andVp/VsRatio From Receiver Functions

Wang

Song

2022

JGR Solid Earth

View full text Add to dashboard Cite

A receiver function (RF) is the response of the Earth's structure below a seismometer to an incident teleseismic wave and consists of a series of P-to-S (Ps) or S-to-P (Sp) converted waves generated at structural interfaces (Langston, 1979), mainly those generated by the velocity discontinuities in the crust and upper mantle below the seismometer. Crustal thickness (H) and P-wave and S-wave velocity ratio (κ) are important parameters reflecting the crustal structure and internal material composition and can provide an important basis for regional tectonics and dynamics. Zhu and Kanamori (2000) noted that when the Moho converted phase Ps and crustal multiple waves are considered at the same time, H and κ can be estimated under the assumption of the average crustal P wave velocity. This is currently the most commonly used H-κ method. The H-κ algorithm stacks the amplitudes of RFs at predicted arrival times for Ps and crustal multiples (PpPs and PpSs + PsPs; here referred to as M1 and M2, respectively, for convenience, as in J. Li et al., 2019) for different values of H and κ in a grid search.The assumptions of the H-κ method are very simple, which leads to its limitations. It assumes that the crustal medium is isotropic, the Moho surface is horizontal, and the P-wave velocity of the crust is fixed. However, the above assumptions cannot be satisfied under complex geological settings, so the results of the H-κ method may be inaccurate. The Chinese continent is known to have highly heterogeneous crustal structure and extremely variable Moho depth, as revealed for example, by seismic tomography (

show abstract

Deep Learning for Geophysics: Current and Future Trends

Cited by 284 publications

References 165 publications

Labeling Poststorm Coastal Imagery for Machine Learning: Measurement of Interrater Agreement

Labeling Poststorm Coastal Imagery for Machine Learning: Measurement of Interrater Agreement

Seasonal Surface Eddy Mixing in the Kuroshio Extension: Estimation and Machine Learning Prediction

Deep Learning‐BasedH‐κMethod (HkNet) for Estimating Crustal Thickness andVp/VsRatio From Receiver Functions

Contact Info

Product

Resources

About