While there is evidence for an acceleration in global mean sea-level (MSL) since the 1960s, its d... more While there is evidence for an acceleration in global mean sea-level (MSL) since the 1960s, its detection at local levels has been hampered by the considerable in uence of natural variability on the rate of MSL change. Here we report an MSL acceleration in tide gauge records along the U.S. Southeast and Gulf coasts that has led to rates (>10 mmyr-1 since 2000) that are unprecedented in at least 120 years. We show that this acceleration is primarily induced by an ocean dynamic signal exceeding the externally forced response from historical climate model simulations. However, when the simulated forced response is removed from observations, the residuals are neither historically unprecedented nor inconsistent with unforced variability in simulations. Furthermore, a large fraction of the residuals can be explained by wind-driven Rossby waves in the tropical North Atlantic. This indicates that the acceleration represents the compounding effects of external forcing and internal climate variability.
A model system consisting of the Princeton ocean model forced by forecast surface fluxes of momen... more A model system consisting of the Princeton ocean model forced by forecast surface fluxes of momentum and heat from the regional atmospheric Eta model is at the heart of the East Coast Ocean Forecast System. Existing near-real-time data sets, including coastal water level gauge data and satellite-derived sea surface temperature and altimetry data, are being used operationally for model evaluation purposes and ultimately for assimilation into the ocean model. The first twelve months of comparisons between 24-hour forecasted and observed subtidal coastal water levels indicate a meridional average correlation coefficient of 0.65, an rms difference of l0 cm, and shows that the forecasts represent over 60% of the observed subtidal variability. A number of sensitivity experiments are underway and a series of enhancements are soon to be implemented, including modification of the surface heat and momentum fluxes; the inclusion of atmospheric pressure loading, riverine fresh water and surface fresh water (evaporation and precipitation) fluxes, and tidal forcing; and accounting for the effects of thermal expansion and contraction. In order to evaluate and improve the basic ocean model and system, the implementation of data assimilation is currently being withheld, however data assimilation methodologies have been developed and the sea surface temperature and altimeter data currently available in near-real-time will be used for these purposes.
Recent studies found that on long time scales there are often unexplained opposite trends in sea ... more Recent studies found that on long time scales there are often unexplained opposite trends in sea level variability between the upper and lower Chesapeake Bay (CB). Therefore, daily sea level and temperature records were analyzed in two locations, Norfolk in the southern CB and Baltimore in the northern CB; surface currents from Coastal Ocean Dynamics Application Radar (CODAR) near the mouth of CB were also analyzed to examine connections between the CB and the Atlantic Ocean. The observations in the bay were compared with daily Atlantic Meridional Overturning Circulation (AMOC) observations during 2005-2021. Empirical Mode Decomposition (EMD) analysis was used to show that variations of sea level and temperature in the upper and lower CB are positively correlated with each other for short time scales of months to few years, but anticorrelated on low frequency modes representing decadal variability and long-term nonlinear trends. The long-term CB modes seem to be linked with AMOC variability through variations in the Gulf Stream and the wind-driven Ekman transports over the North Atlantic Ocean. AMOC variability correlates more strongly with variability in the southern CB near the mouth of the bay, where surface currents indicate potential links with AMOC variability. For example, when AMOC and the Gulf Stream were especially weak during 2009-2010, sea level in the southern bay was abnormally high, temperatures were colder than normal and outflow through the mouth of CB was especially high. Sea level in the upper bay responded to this change only 1-2 years later, which partly explains phase differences within the bay. A persistent trend of 0.22 cm/s per year of increased outflow from the CB, may be a sign of a climate-related trend associated with combination of weakening AMOC and increased precipitation and river discharge into the CB.
A continuous data assimilation scheme and a multilayer, primitive equation, numerical model are d... more A continuous data assimilation scheme and a multilayer, primitive equation, numerical model are described. The model is an eddy-resolving, coastal ocean model that has been extended to include the Gulf Stream region. It has complete thermohaline dynamics, a bottom-following, sigma, vertical coordinate system, and a coastal-following, curvilinear orthogonal, horizontal coordinate system. Calculated model fields are used to provide a model climatology and correlations between subsurface temperature and salinity anomalies and surface elevation anomalies. An optimal interpolation method, the surface to subsurface correlations, and estimated model and data errors are the basis of the assimilation technique. Altimetry anomaly data extracted from the model calculations according to the Geosat orbital schedule are used to test the assimilation scheme and to provide nowcasts and forecasts. Sensitivity studies are performed to test the effects of various parameters of the scheme. It is found that the scheme is less efficient in the shallow continental shelf area than in the deeper regions of the model. The results show significant nowcast skill, with area-averaged rms error for surface elevation and subsurface properties of about 40-50% of the corresponding error of the unassimilated case. Good forecast skill, better than persistence, is demonstrated for 10-20 days; there is little skill after 30-40 days. Increasing the density of the satellite altimetry data (especially by decreasing the separation distance between tracks) should decrease the nowcast rms error to about 15% and improve the forecast. 1. variables used in initialization and assimilation. Primitive equation models may fall short of providing a realistic climatology. In the North Atlantic, Gulf Stream separation has been a problem [e.g., Thompson and Schmitz, 1989]; models tend to separate north of Cape Hatteras. However, in the long run it is probable that good nowcasts and forecasts will partially depend on the capabilities of the numerical model to generate realistic climatologies and realistic statistics at scales comparable to the Rossby radius. Data assimilation schemes vary. Robinson et al. [1988, 1989] primarily use satellite sea surface temperature (SST) fields to locate the Gulf Stream and mesoscale eddies. They
While there is evidence for an acceleration in global mean sea level (MSL) since the 1960s, its d... more While there is evidence for an acceleration in global mean sea level (MSL) since the 1960s, its detection at local levels has been hampered by the considerable influence of natural variability on the rate of MSL change. Here we report a MSL acceleration in tide gauge records along the U.S. Southeast and Gulf coasts that has led to rates (>10 mm yr −1 since 2010) that are unprecedented in at least 120 years. We show that this acceleration is primarily induced by an ocean dynamic signal exceeding the externally forced response from historical climate model simulations. However, when the simulated forced response is removed from observations, the residuals are neither historically unprecedented nor inconsistent with internal variability in simulations. A large fraction of the residuals is consistent with wind driven Rossby waves in the tropical North Atlantic. This indicates that this ongoing acceleration represents the compounding effects of external forcing and internal climate variability.
The study addresses two important issues associated with sea level along the coasts of Thailand: ... more The study addresses two important issues associated with sea level along the coasts of Thailand: first, the fast sea level rise and its spatial variation, and second, the monsoonal-driven seasonal variations in sea level. Tide gauge data that are more extensive than in past studies were obtained from several different local and global sources, and relative sea level rise (RSLR) rates were obtained from two different methods, linear regressions and non-linear Empirical Mode Decomposition/Hilbert-Huang Transform (EMD/HHT) analysis. The results show extremely large spatial variations in RSLR, with rates varying from~1 mm y −1 to~20 mm y −1 ; the maximum RSLR is found in the upper Gulf of Thailand (GOT) near Bangkok, where local land subsidence due to groundwater extraction dominates the trend. Furthermore, there are indications that RSLR rates increased significantly in all locations after the 2004 Sumatra-Andaman Earthquake and the Indian Ocean tsunami that followed, so that recent RSLR rates seem to have less spatial differences than in the past, but with high rates of~20-30 mm y −1 almost everywhere. The seasonal sea level cycle was found to be very different between stations in the GOT, which have minimum sea level in June-July, and stations in the Andaman Sea, which have minimum sea level in February. The seasonal sea-level variations in the GOT are driven mostly by large-scale wind-driven set-up/ set-down processes associated with the seasonal monsoon and have amplitudes about ten times larger than either typical steric changes at those latitudes or astronomical annual tides.
In this paper a sigma coordinate ocean model is modified to remove the commonly used Boussinesq a... more In this paper a sigma coordinate ocean model is modified to remove the commonly used Boussinesq approximation so that the effect of thermal expansion is exactly included in the basic equations in order to cope with the seasonal heating cycle and the detection of climate change through variation in sea level height. Tests are performed to evaluate the differences between Boussinesq and non-Boussinesq calculations under different heating and cooling conditions and different model domains. For an idealized case of a flat bottom, shallow ocean basin without wind forcing, simulations of a warm eddy show that the non-Boussinesq dynamics have only a minor effect on the baroclinic current field. However, vertically averaged velocities, though small compared with the baroclinic velocities, are cyclonic for the Boussinesq calculation and anticyclonic for the non-Boussinesq calculation. The results indicate that global or closed basin Boussinesq models should be able to simulate most of the observed steric sea level changes on seasonal or climate timescales, when corrected by a spatially uniform, time-dependent factor calculated from the volume-averaged density change. The seasonal variation of the globally averaged sea level calculated from climatological data is small, about 1 cm. Variations in steric sea level in regional models, both Boussinesq and non-Boussinesq, may differ from those of global models owing to the unknown transport across their boundaries associated with the local heating and cooling. A spatially uniform, time-dependent correction, similar to that associated with thermal expansion, is proposed to account for transport across open boundaries of regional models. Variations of sea level obtained from a Boussinesq model of the Atlantic Ocean approximate the seasonal signal due to the heating/cooling cycle of each hemisphere as observed by satellite altimeter data. Greatbatch [ 1994] has raised concern that since variations in sea level associated with expansion or contraction of the water column due to density. changes are missing from ocean models, they Copyright 1995 by the American Geophysical Union. Paper number 95JC02442. 0148-0227/95/95JC-02442505.00 may not correctly simulate seasonal [Pattullo et-al., 1955] and climatic [ Church et al., 1991 ] changes in sea level. For example, seasonal variations in sea level are observed in the global ocean with tide gauges [e.g., Tsirnplis and Woodworth, 1994] and satellite altimeters [ Sta•nrner and Wunsch, 1994] but may differ from ocean model simulations. Ocean models can simulate steric sea level changes associated with climatic changes in thermohaline structure as recently demonstrated by Ezer et al. [ 1995]. Coastal ocean models can also forecast variations in sea level; for example, an experimental, operational, coastal nowcast/forecast system for the U.S. east coast [Aikrnan et al., 1995] shows considerable skill in the prediction of short-term, wind-driven, sea level variations. However, the processes associated with long-term, seasonal, and interannual variations due to heating and cooling need further understanding before they can be accurately predicted by ocean models. Consider a closed basin. It is supposed that the non-Boussinesq sea level can be written as rl(x, y,t) = rl•(x, y,t) + tiE(t)+ rlas(X, y,t) (1) where r/• is the local sea level change due to the Boussinesq dynamics, r/E (t) is due to expansion or compression of the water column and is equal to the area average of-H6p/po, where H(x, y) is the bottom topography and •-ff(x, y, t) is the vertical average of the density deviation from a reference density Po In an application of (1), r/a s is unknown and neglected as a small error; it is, however, largely attributable to the so-called "Goldsbrough-$tommel gyres" [Greatbatch, 1994], a non-20,565 20,566 MELLOR AND EZER: THE BOUSSINESQ APPROXIMATION IN OCEAN MODELS Boussinesq vortex stretching effect due to density change. The Goldsbrough-Stommel gyre, first introduced for the case of forcing by mass flux due to evaporation and precipitation, is discussed also by Huang and Schmin [ 1993]. The hypothesis embodied in (1) is that r/oe is independent of the spatial variables, x and y. For (1) to be useful, r/a s should be small so that, for example, Boussinesq global models when compared with observations may be adjusted by a globally uniform, time-dependent r/E(t). In other words, the local elevation change induced purely by density change in the continuity equation is rapidly distributed over the entire domain with a timescale L/c, where L is the basin lateral scale and c = (gH) u2 is the barotropic wave speed. In the most simplified case of a motionless ocean with a uniform heating, r/= r/E. The main objective of this paper is to test a non-Boussinesq model and to compare it with its Boussinesq counterpart. It should be mentioned that here the term "non-Boussinesq" does not refer to the full non-Boussinesq dynamics, which might also include acoustic waves [ Veronis, 1973]. In the equation of state used here [Mellor, 1991 ] the density is calculated from the salinity, potential temperature, and pressure, p = p(S, T, •), but the pressure is calculated from the hydrostatic relation using an approximate, temporally constant density. Therefore sound waves are filtered out in both the Boussinesq and the non-Boussinesq models.
Applicability of the Empirical Mode Decomposition for Power Traces of Large-Scale Applications
Lecture Notes in Computer Science, 2018
Current trends in HPC show that exascale systems will be power capped, prompting their users to d... more Current trends in HPC show that exascale systems will be power capped, prompting their users to determine the best combination of resources to satisfy a power budget. Hence, performance and energy models must interplay and aid users in this resource selection based on the desired application parameters. While existing performance models may predict application execution at a scale, current power models are inadequate for this propose due, in part, to the variability of instantaneous dynamic power and the need to handle large amount of power measurements at the runtime to populate the models. In this paper, the latter challenge is tackled by selecting certain power measurements and applying to them the empirical mode decomposition (EMD) technique, which itself already deals with instantaneous variability of power during the runtime. Specifically, it is proposed here to apply EMD to segments of a power trace to rapidly generate a quadratic model that describes overall time, power, and thus energy simultaneously. The proposed models have been applied to several realistic applications. The error across the proposed models and the measured energy consumption is within 5% for the smaller segments consisting of 2,000 trace samples and is about 2% for the segments of 6,000 samples.
International Journal of High Performance Computing Applications, Oct 31, 2017
Power draw is a complex physical response to the workload of a given application on the hardware,... more Power draw is a complex physical response to the workload of a given application on the hardware, which is difficult to model, in part, due to its variability. The empirical mode decomposition and Hilbert-Huang transform (EMD/HHT) is a method commonly applied to physical systems varying with time to analyze their complex behavior. In authors' work, the EMD/HHT is considered for the first time to study power usage of high-performance applications. Here, this method is applied to the power measurement sequences (called here power traces) collected on three different computing platforms featuring two generations of Intel Xeon Phi, which are an attractive solution under the power budget constraints. The high-performance applications explored in this work are codesign molecular synamics and general atomic and molecular electronic structure system-which exhibit different power draw characteristics-to showcase strengths and limitations of the EMD/HHT analysis. Specifically, EMD/HHT measures intensity of an execution, which shows the concentration of power draw with respect to execution time and provides insights into performance bottlenecks. This article compares intensity among executions, noting on a relationship between intensity and execution characteristics, such as computation amount and data movement. In general, this article concludes that the EMD/HHT method is a viable tool to compare application power usage and performance over the entire execution and that it has much potential in selecting most appropriate execution configurations.
For modern parallel applications, modeling their general execution characteristics, such as power... more For modern parallel applications, modeling their general execution characteristics, such as power and time, is difficult due to a great many factors affecting software-hardware interactions, which is also exacerbated by the dearth of measuring and monitoring tools for novel architectures, such as Intel Xeon Phi processors. To address this modeling challenge, the present work proposes to employ the Empirical Mode Decomposition (EMD) method to describe an execution as a series of modes culminating in a single residual trend, for which, in its turn, a model equation is obtained as a non-linear fit. As outcome, an overall energy consumption may be predicted using this model. A real-world quantum-chemistry application GAMESS and a molecular-dynamics proxy application CoMD were considered in the experiments. The results demonstrate that the energy modeled ranges within 10-30% of the measured energy, depending on the length of execution.
Uploads
Papers by Tal Ezer