An optimization framework for adjoint-based climate simulations

Yuan, Shijin; Wen, Shicheng; Li, Hongyu; Zhang, Xinfeng; Qin, Liu

doi:10.1177/1094342013495096

Cited by 2 publications

(6 citation statements)

References 12 publications

(14 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…And this framework was applied to ZebiakCane (ZC) adjoint model and achieved great performance improvement. However, ZC model just includes a relatively simple data flow logic so the classical data stack strategy, First In Last Out can perform well in [5] and yet may rarely be suited to other complex models. Therefore, More sophisticated data flow models are needed.…”

Section: Introductionmentioning

confidence: 98%

“…Therefore, it is of urgency to accelerate the numerical weather prediction system simulation. In the work of Yuan et al [5] , a performance optimization framework for adjoint model was proposed, which included I/O optimization, adaptive strategy optimization, and parallel optimization. And this framework was applied to ZebiakCane (ZC) adjoint model and achieved great performance improvement.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

ACStack: Adaptive Composite Stack for Adjoint Models in Code Optimization

Zhou²,

Yuan

et al. 2013

2013 IEEE 10th International Conference on High Performance Computing and Communications &Amp; 2013 IEEE International Conferen

Self Cite

View full text Add to dashboard Cite

Adjoint model is one of the key technologies of numerical weather forecasting. In adjoint model, the observed information can be inputted into the model continually to improve the quality of the initial data and thus to improve the quality of forecast. However, adjoint models usually adopt the Calculate-All strategy for implementation, which is of low computational efficiency in complex numerical simulations. To improve the efficiency, we propose an Adaptive Storage Strategy in this paper. An Adaptive Composite Stack is also designed to manage the data flows. To further demonstrate the applicability of our method, the Fifth-Generation NCAR/Penn State Mesoscale Model (MM5) is chosen as the prediction model. Three cases of MM5 with different resolutions are studied and the results are presented. Experimental results show that the strategy can greatly improve the computing efficiency of adjoint models.

show abstract

Section: Introductionmentioning

confidence: 98%

Section: Introductionmentioning

confidence: 99%

ACStack: Adaptive Composite Stack for Adjoint Models in Code Optimization

Zhou²,

Yuan

et al. 2013

2013 IEEE 10th International Conference on High Performance Computing and Communications &Amp; 2013 IEEE International Conferen

Self Cite

View full text Add to dashboard Cite

show abstract

“…It is worth mentioning that this situation will likely change in the future, but not any time soon. In addition, heavy I/O involved in reading back the forward wavefield can significantly slow down the simulation 65 . For nondissipative acoustic or elastic media, a standard solution for large 3D problems is to perform three simulations per source 60;43 .…”

Section: Introductionmentioning

confidence: 99%

“…Changing the sign of one of the terms in the backward integrations to stabilize the calculations is also classically done in so-called Back-and-Forth Nudging algorithms 6;7 . An alternative solution consists of resorting to so-called 'partial checkpointing' or 'optimal checkpointing', i.e., using partial storage to disk, but, realizing that storage size limitations or slowdown related to disk storage are important issues 65 , defining an optimized sequence in which the forward and adjoint time steps are performed, essentially trading storage requirements for longer computation times. This elegant idea was introduced by 45 …”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Anelastic sensitivity kernels with parsimonious storage for adjoint tomography and full waveform inversion

Komatitsch¹,

Xie

Bozdag

et al. 2016

Geophys. J. Int.

View full text Add to dashboard Cite

We introduce a technique to compute exact anelastic sensitivity kernels in the time domain using parsimonious disk storage. The method is based on a reordering of the time loop of time-domain forward/adjoint wave propagation solvers combined with the use of a memory buffer. It avoids instabilities that occur when time-reversing dissipative wave propagation simulations. The total number of required time steps is unchanged compared to usual acoustic or elastic approaches. The cost is reduced by a factor of 4/3 compared to the case in which anelasticity is partially accounted for by accommodating the effects of physical dispersion. We validate our technique by performing a test in which we compare the K α sensitivity kernel to the exact kernel obtained by saving the entire forward calculation. This benchmark confirms that our approach is also exact. We illustrate the importance of including full attenuation in the calculation of sensitivity kernels by showing significant differences with physical-dispersion-only kernels. *** This manuscript is now published as a paper in the Geophysical JournalInternational, 2016. *** I IntroductionEfficient numerical methods for simulating the propagation of acoustic, elastic, or anelastic waves in the time domain are widely available, for instance based on finite-difference methods see e.g., 63 for a review, spectral-element methods e.g., 32;61;30;31 , or standard finiteelement methods e.g., 28 . Nowadays, these techniques are heavily used for imaging based on full waveform inversion (FWI) or adjoint tomography e.g., 59;44;60;63;20;41 . FWI involves fitting band-pass filtered versions of observed seismograms by minimizing least-squared differences between observed and synthetic seismograms. Adjoint tomography generalizes FWI by considering arbitrary measures of misfit, e.g., cross-correlation traveltimes, multi-taper phase and amplitude anomalies, or instantaneous phase measurements.In the context of imaging, it is useful to resort to the concept of sensitivity kernels e.g., 56;57;58;59;60;39;20;21 . Let s denote the forward displacement wavefield and s † the adjoint wavefield. In an isotropic Earth model, the kernels K κ and K µ represent Fréchet derivatives with respect to relative bulk and shear moduli perturbations, respectively. These kernels are given by e.g.,whereanddenote the traceless strain deviator and its adjoint, respectively, x is the position vector, and κ and µ are the bulk and shear moduli, respectively. Their expression remains valid for elastic perturbations superimposed on an anelastic Earth model 39 if the regular and adjoint wavefields are computed in that anelastic reference model. In practical applications, it is often useful to define compressional and shear wavespeed sensitivity kernels, namely 59;60andIn order to perform the convolution involved in the calculation of the kernels (1) and (2), simultaneous access to the forward wavefield s at time t and the adjoint wavefield s † at time T − t is required (or conversely, sinceconvolving two funct...

show abstract

An optimization framework for adjoint-based climate simulations

Cited by 2 publications

References 12 publications

ACStack: Adaptive Composite Stack for Adjoint Models in Code Optimization

ACStack: Adaptive Composite Stack for Adjoint Models in Code Optimization

Anelastic sensitivity kernels with parsimonious storage for adjoint tomography and full waveform inversion

Contact Info

Product

Resources

About