Xunqiang Yin

The Sunway TaihuLight supercomputer: system and applications

The Sunway TaihuLight supercomputer is the world’s first system with a peak performance greater than 100 PFlops. In this paper, we provide a detailed introduction to the TaihuLight system. In contrast with other existing heterogeneous supercomputers, which include both CPU processors and PCIe-connected many-core accelerators (NVIDIA GPU or Intel Xeon Phi), the computing power of TaihuLight is provided by a homegrown many-core SW26010 CPU that includes both the management processing elements (MPEs) and computing processing elements (CPEs) in one chip. With 260 processing elements in one CPU, a single SW26010 provides a peak performance of over three TFlops. To alleviate the memory bandwidth bottleneck in most applications, each CPE comes with a scratch pad memory, which serves as a user-controlled cache. To support the parallelization of programs on the new many-core architecture, in addition to the basic C/C++ and Fortran compilers, the system provides a customized Sunway OpenACC tool that supports the OpenACC 2.0 syntax. This paper also reports our preliminary efforts on developing and optimizing applications on the TaihuLight system, focusing on key application domains, such as earth system modeling, ocean surface wave modeling, atomistic simulation, and phase-field simulation.

The prediction on the 2015/16 El Niño event from the perspective of FIO-ESM

Recently atmospheric and oceanic observations indicate the tropical Pacific is at the El Niño condition. However, it’s not clear whether this El Niño event of this year is comparable to the very strong one of 1997/98 which brought huge influence on the whole world. In this study, based on the Ensemble Adjusted Kalman Filter (EAKF) assimilation scheme and First Institute of Oceanography-Earth System Model (FIO-ESM), the assimilation system is setup, which can provide reasonable initial conditions for prediction. And the hindcast results suggest the skill of El Niño-Southern Oscillation (ENSO) prediction is comparable to other dynamical coupled models. Then the prediction for 2015/16 El Niño by using FIO-ESM is started from 1 November 2015. The ensemble results indicate that the 2015/16 El Niño will continue to be strong. By the end of 2015, the strongest strength is very like more than 2.0°C and the ensemble mean strength is 2.34°C, which indicates 2015/16 El Niño event will be very strong but slightly less than that of 1997/98 El Niño event (2.40°C) calculated relative a climatology based on the years 1992–2014. The prediction results also suggest 2015/16 El Niño event will be a transition to ENSO-neutral level in the early spring (FMA) 2016, and then may transfer to La Niña in summer 2016.

The improvement of the one-dimensional Mellor-Yamada and K-profile parameterization turbulence schemes with the non-breaking surface wave-induced vertical mixing

Both the level 2.5 Mellor-Yamada turbulence closure scheme (MY) and K-profile parameterization (KPP) are popularly used by the ocean modeling community. The MY and the KPP are improved through including the non-breaking surface wave-induced vertical mixing (Bv), and the improved schemes were tested by using continuous data at the Papa ocean weather station (OWS) during 1961–1965. The numerical results showed that the Bv can make the temperature simulations fit much better with the continuous data from Papa Station. The two improved schemes overcame the shortcomings of predicting too shallow upper mixed layer depth and consequently overheated sea surface temperature during summertime, which are in fact common problems for all turbulence closure models. Statistical analysis showed that the Bv effectively reduced the mean absolute error and root mean square error of the upper layer temperature and increased the correlation coefficient between simulation and the observation. Furthermore, the performance of vertical mixing induced by shear instability and the Bv is also compared. Both the temperature structure and its seasonal cycle significantly improved by including the Bv, regardless of whether shear instability was included or not, especially for the KPP mixing scheme, which suggested that Bv played a dominant role in the upper ocean where the mean current was relatively weak, such as at Papa Station. These results may provide a clue to improve ocean circulation models.

A highly effective global surface wave numerical simulation with ultra-high resolution

Surface wave is the most energetic form of motions in the ocean and is crucially important to navigation safety and climate change. High-resolution global wave model plays a key role in accurate surface wave forecasting. However, operational forecasting systems are still not in high-resolution due to entailed high demand for large computation, as well as low parallel efficiency barrier. Here breakthroughs encompassing the design and application of irregular quasi-rectangular domain decomposition, master-slave cooperative computing workflow and pipelining scheme were applied to a global wave model, which has been used in several operational forecasting systems and earth system models. Our realistic surface wave simulations on Sunway TaihuLight Supercomputer demonstrated that our model had outstanding scalability and achieved 45.43 PFlops in ultra-high resolution of (1/100)°, using full-scale supercomputer with 10,649,600 cores. That provides a highly effective solution for accurate surface wave forecasting and climate change prediction.

Analytical estimation of mixing coefficient induced by surface wave-generated turbulence based on the equilibrium solution of the second-order turbulence closure model

Based on the equations of motion and the assumption that ocean turbulence is of isotropy or quasi-isotropy, we derived the closure equations of the second-order moments and the variation equations for characteristic quantities, which describe the mechanisms of advection transport and shear instability by the sum of wave-like and eddy-like motions and circulation. Given that ocean turbulence generated by wave breaking is dominant at the ocean surface, we presented the boundary conditions of the turbulence kinetic energy and its dissipation rate, which are determined by energy loss from wave breaking and entrainment depth respectively. According to the equilibrium solution of the variation equations and available data of the dissipation rate, we obtained an analytical estimation of the characteristic quantities of surface-wave-generated turbulence in the upper ocean and its related mixing coefficient. The derived kinetic dissipation rate was validated by field measurements qualitatively and quantitatively, and the mixing coefficient had fairly good consistency with previous results based on the Prandtl mixing length theory.

Evaluation on data assimilation of a global high resolution wave-tide-circulation coupled model using the tropical Pacific TAO buoy observations

In order to evaluate the assimilation results from a global high resolution ocean model, the buoy observations from tropical atmosphere ocean (TAO) during August 2014 to July 2015 are employed. The horizontal resolution of wave-tide-circulation coupled ocean model developed by The First Institute of Oceanography (FIOCOM model) is 0.1°×0.1°, and ensemble adjustment Kalman filter is used to assimilate the sea surface temperature (SST), sea level anomaly (SLA) and Argo temperature/salinity profiles. The simulation results with and without data assimilation are examined. First, the overall statistic errors of model results are analyzed. The scatter diagrams of model simulations versus observations and corresponding error probability density distribution show that the errors of all the observed variables, including the temperature, isotherm depth of 20°C (D20), salinity and two horizontal component of velocity are reduced to some extent with a maximum improvement of 54% after assimilation. Second, time-averaged variables are used to investigate the horizontal and vertical structures of the model results. Owing to the data assimilation, the biases of the time-averaged distribution are reduced more than 70% for the temperature and D20 especially in the eastern Pacific. The obvious improvement of D20 which represents the upper mixed layer depth indicates that the structure of the temperature after the data assimilation becomes more close to the reality and the vertical structure of the upper ocean becomes more reasonable. At last, the physical processes of time series are compared with observations. The time evolution processes of all variables after the data assimilation are more consistent with the observations. The temperature bias and RMSE of D20 are reduced by 76% and 56% respectively with the data assimilation. More events during this period are also reproduced after the data assimilation. Under the condition of strong 2014/2016 El Niño, the Equatorial Undercurrent (EUC) from the TAO is gradually increased during August to November in 2014, and followed by a decreasing process. Since the improvement of the structure in the upper ocean, these events of the EUC can be clearly found in the assimilation results. In conclusion, the data assimilation in this global high resolution model has successfully reduced the model biases and improved the structures of the upper ocean, and the physical processes in reality can be well produced.

Case study on the three-dimensional structure of meso-scale eddy in the South China Sea based on a high-resolution model

Meso-scale eddies are important features in the South China Sea (SCS). The eddies with diameters of 50–200 km can greatly impact the transport of heat, momentum, and tracers. A high-resolution wave-tide-circulation coupled model was developed to simulate the meso-scale eddy in the SCS in this study. The aim of this study is to examine the model ability to simulate the meso-scale eddy in the SCS without data assimilations The simulated Sea Surface Height (SSH) anomalies agree with the observed the AVISO SSH anomalies well. The simulated subsurface temperature profiles agree with the CTD observation data from the ROSE (Responses of Marine Hazards to climate change in the Western Pacific) project. The simulated upper-ocean currents also agree with the main circulation based on observations. A warm eddy is identified in winter in the northern SCS. The position and domain of the simulated eddy are confirmed by the observed sea surface height data from the AVISO. The result shows that the model has the ability to simulate the meso-scale eddy in the SCS without data assimilation. The three-dimensional structure of the meso-scale eddy in the SCS is analyzed using the model result. It is found that the eddy center is tilted vertically, which agrees with the observation. It is also found that the velocity center of the eddy does not coincide with the temperature center of the eddy. The result shows that the model has the ability to simulate the meso-scale eddy in the SCS without data assimilations. Further study on the forming mechanism and the three-dimensional structure of the meso-scale eddies will be carried out using the model result and cruise observation data in the near future.

A comparison of two global ocean-ice coupled models with different horizontal resolutions

A global eddy-permitting ocean-ice coupled model with a horizontal resolution of 0.25° by 0.25° is established on the basis of Modular Ocean Model version 4 (MOM4) and Sea Ice Simulator (SIS). Simulation results are compared with those of an intermediate resolution ocean-ice coupled model with a horizontal resolution of about 1° by 1°. The results show that the simulated ocean temperature, ocean current and sea ice concentration from the eddy-permitting model are better than those from the intermediate resolution model. However, both the two models have the common problem of ocean general circulation models (OGCMs) that the majority of the simulated summer sea surface temperature (SST) is too warm while the majority of the simulated subsurface summer temperature is too cold. Further numerical experiments show that this problem can be alleviated by incorporating the non-breaking surface wave-induced vertical mixing into the vertical mixing scheme for both eddy-permitting and intermediate resolution models.

ESTABLISHMENT OF THE OCEAN DYNAMIC SYSTEM WITH FOUR SUB-SYSTEMS AND THE DERIVATION OF THEIR GOVERNING EQUATION SETS

Based on their differences in physical characteristics and time-space scales, the ocean motions have been divided into four types in the present paper: turbulence, wave-like motion, eddy-like motion and circulation. Applying the three-fold Reynolds averages to the governing equations with Boussinesq approximation, with the averages defined on the former three sub-systems, we derive the governing equation sets of the four sub-systems and refer to their sum as “the ocean dynamic system”. In these equation sets, the interactions among different motions appear in two forms: the first one includes advection transport and shear instability generation of larger scale motions, and the second one is the mixing induced by smaller scale motions in the form of transport flux residue. The governing equation sets are the basis of analytical/numerical descriptions of various ocean processes.

Evaluation on surface current observing network of high frequency ground wave radars in the Gulf of Thailand

Due to the high cost of ocean observation system, the scientific design of observation network becomes much important. The current network of the high frequency radar system in the Gulf of Thailand has been studied using a three-dimensional coastal ocean model. At first, the observations from current radars have been assimilated into this coastal model and the forecast results have improved due to the data assimilation. But the results also show that further optimization of the observing network is necessary. And then, a series of experiments were carried out to assess the performance of the existing high frequency ground wave radar surface current observation system. The simulated surface current data in three regions were assimilated sequentially using an efficient ensemble Kalman filter data assimilation scheme. The experimental results showed that the coastal surface current observation system plays a positive role in improving the numerical simulation of the currents. Compared with the control experiment without assimilation, the simulation precision of surface and subsurface current had been improved after assimilated the surface currents observed at current networks. However, the improvement for three observing regions was quite different and current observing network in the Gulf of Thailand is not effective and a further optimization is required. Based on these evaluations, a manual scheme has been designed by discarding the redundant and inefficient locations and adding new stations where the performance after data assimilation is still low. For comparison, an objective scheme based on the idea of data assimilation has been obtained. Results show that all the two schemes of observing network perform better than the original network and optimal scheme-based data assimilation is much superior to the manual scheme that based on the evaluation of original observing network in the Gulf of Thailand. The distributions of the optimal network of radars could be a useful guidance for future design of observing system in this region.