Jacob, J. C., Armstrong, E. M., Bourassa, M. A., Cram, T., Elya, J. L., Greguska, F. R., III, et al. (2018). OceanWorks: Enabling Interactive Oceanographic Analysis in the Cloud with Multivariate Data. In American Geophysical Union (Vol. Fall Meeting).
Abstract: NASA's Advanced Information System Technology (AIST) Program sponsors the OceanWorks project to establish an integrated data analytics center at the Physical Oceanography Distributed Active Archive Center (PO.DAAC). OceanWorks provides a series of interoperable capabilities that are essential for cloud-scale oceanographic research. These include big data analytics, data search with subsecond response, intelligent ranking of search results, subsetting based on data quality metrics, and rapid spatiotemporal matchup of satellite measurements with distributed in situ data. The software behind OceanWorks is being developed as an open source project in the Apache Incubator Science Data Analytics Platform (SDAP – http://sdap.apache.org). In this presentation we describe how OceanWorks enables efficient, scalable, interactive and interdisciplinary oceanographic analysis with multivariate data.
Interactivity is enabled by a number of SDAP features. First, SDAP provides Representational State Transfer (REST) interfaces to a number of built-in cloud analytics to compute time series, time-averaged maps, correlation maps, climatological maps, Hovmöller maps, and more. To access these, users simply navigate to a properly constructed parameterized URL in their web browser or issue web services calls in a variety of programming languages or in a Jupyter notebook. Alternatively, Python clients can make function calls via the NEXUS Command Line Interface (CLI). Authenticated users can even inject their own custom code via REST calls or the CLI.
To enable interdisciplinary science, OceanWorks provides access to a rich collection of multivariate satellite and in situ measurements of the oceans (e.g., sea surface temperature, height and salinity, chlorophyll and circulation) and other Earth science data (e.g., aerosol optical depth and wind speed), coupled with on-demand processing capabilities close to the data. We partition the data across space or time into tiles and store them into cloud-aware databases that are collocated with the computations. We will provide examples of scientific studies directly enabled by OceanWorks' multivariate data and cloud analytics.
|
Cintra, R., Campos Velho, H., & Cocke, S. (2016). Multilayer Perceptron on data assimilation system applied to FSU global model..
|
Cocke, S., Boisserie, M., & Shin, D. - W. (2013). A coupled soil moisture initialization scheme for the FSU/COAPS climate model. Inverse Problems in Science and Engineering, 21(3), 420–437.
|
Yu, P. (2006). Development of New Techniques for Assimilating Satellite Altimetry Data into Ocean Models. Ph.D. thesis, Florida State University, Tallahassee, FL.
Abstract: State of the art fully three-dimensional ocean models are very computationally expensive and their adjoints are even more resource intensive. However, many features of interest are approximated by the first baroclinic mode over much of the ocean, especially in the lower and mid latitude regions. Based on this dynamical feature, a new type of data assimilation scheme to assimilate sea surface height (SSH) data, a reduced-space adjoint technique, is developed and implemented with a three-dimensional model using vertical normal mode decomposition. The technique is tested with the Navy Coastal Ocean Model (NCOM) configured to simulate the Gulf of Mexico. The assimilation procedure works by minimizing the cost function, which generalizes the misfit between the observations and their counterpart model variables. The “forward” model is integrated for the period during which the data are assimilated. Vertical normal mode decomposition retrieves the first baroclinic mode, and the data misfit between the model outputs and observations is calculated. Adjoint equations based on a one-active-layer reduced gravity model, which approximates the first baroclinic mode, are integrated backward in time to get the gradient of the cost function with respect to the control variables (velocity and SSH of the first baroclinic mode). The gradient is input to an optimization algorithm (the limited memory Broyden-Fletcher-Goldfarb-Shanno (BFGS) method is used for the cases presented here) to determine the new first baroclinic mode velocity and SSH fields, which are used to update the forward model variables at the initial time. Two main issues in the area of ocean data assimilation are addressed: 1. How can information provided only at the sea surface be transferred dynamically into deep layers? 2. How can information provided only locally, in limited oceanic regions, be horizontally transferred to ocean areas far away from the data-dense regions, but dynamically connected to it? The first problem is solved by the use of vertical normal mode decomposition, through which the vertical dependence of model variables is obtained. Analyses show that the first baroclinic mode SSH represents the full SSH field very closely in the model test domain, with a correlation of 93% in one of the experiments. One common way to solve the second issue is to lengthen the assimilation window in order to allow the dynamic model to propagate information to the data-sparse regions. However, this dramatically increases the computational cost, since many oceanic features move very slowly. An alternative solution to this is developed using a mapping method based on complex empirical orthogonal functions (EOF), which utilizes data from a much longer period than the assimilation cycle and deals with the information in space and time simultaneously. This method is applied to map satellite altimeter data from the ground track observation locations and times onto a regular spatial and temporal grid. Three different experiments are designed for testing the assimilation technique: two experiments assimilate SSH data produced from a model run to evaluate the method, and in the last experiment the technique is applied to TOPEX/Poseidon and Jason-1 altimeter data. The assimilation procedure converges in all experiments and reduces the error in the model fields. Since the adjoint, or “backward”, model is two-dimensional, the method is much more computationally efficient than if it were to use a fully three-dimensional backward model.
|
Srinivasan, A., Chassignet, E. P., Bertino, L., Brankart, J. M., Brasseur, P., Chin, T. M., et al. (2011). A comparison of sequential assimilation schemes for ocean prediction with the HYbrid Coordinate Ocean Model (HYCOM): Twin experiments with static forecast error covariances. Ocean Modelling, 37(3-4), 85–111.
|
Yu, P., Morey, S. L., & O'Brien, J. J. (2009). A reduced-dynamics variational approach for the assimilation of altimeter data into eddy-resolving ocean models. Ocean Modelling, 27(3-4), 215–229.
|
Smedstad, O. M., Hurlburt, H. E., Metzger, E. J., Rhodes, R. C., Shriver, J. F., Wallcraft, A. J., et al. (2003). An operational Eddy resolving 1/16° global ocean nowcast/forecast system. Journal of Marine Systems, 40-41, 341–361.
|
Belyaev, K. P., Tanajura, C. A. S., & O'Brien, J. J. (2001). A data assimilation method used with an ocean circulation model and its application to the tropical Atlantic. Applied Mathematical Modelling, 25(8), 655–670.
|
Perrie, W., Zhang, W., Bourassa, M., Shen, H., & Vachon, P. W. (2008). Impact of Satellite Winds on Marine Wind Simulations. Wea. Forecasting, 23(2), 290–303.
|
Jardak, M., Navon, I. M., & Zupanski, M. (2009). Comparison of sequential data assimilation methods for the Kuramoto-Sivashinsky equation. Int. J. Numer. Meth. Fluids, .
|