Final Report Summary - OCEANDATAMODELS (Statistical modelling and estimation for spatiotemporal data with oceanographic applications)
The volume and variety of data collected from our oceans is rapidly increasing, from instruments such as surface drifters, sub-surface floats, underwater gliders, satellite altimeters, and acoustic sensors. There are now billions of observations, and the analysis of this data is key to our understanding of climate change and the environment at large. This project was concerned with developing new statistical methods for the analysis of such datasets. We sought to build models that could exploit the volume of data to estimate new features of the ocean, at previously unresolved resolutions.
A particular focus was on data from the Global Drifter Program, a vast and unique collections of observations encompassing most of the globe’s ocean surface. In total over 20,000 drifters have been deployed since 1979, resulting in over 100 million observations of measurements including surface velocity, sea surface temperature, and salinity. The challenge with such datasets are that the drifters are moving in time and space, so rather than observing fixed-point measurements, we instead observe “trajectories” where the observations from each instrument are moving across and between the oceans. This challenge required new models and new statistical methodology for their implementation.
The main contribution of this project was the delivery of a fully-integrated statistical model (from design to implementation stage) that can be used to better understand the trajectories from drifters and related instruments. Our model uses stochastic processes to describe the variability and uncertainty that is inherent to drifter trajectories, in addition to capturing swirling rotational oscillations caused by the Coriolis effect - the effect due to the Earth’s rotation on its own axis - as well as from tides and local features known as eddies. Details of our model have been published to both the statistics [1] and physical oceanography [2] communities. Furthermore, the results from our global analysis have been disseminated at leading international conferences such as Ocean Sciences, the American Geophysical Union Fall meeting, the European Geophysical Union General Assembly, the Joint Statistical Meetings, and the Royal Statistical Society Annual Conference.
The implementation of our model to large-scale data required several new methodological developments in statistics and spatiotemporal data analysis. These include a computationally-efficient procedure for estimating parameters from the model [3], a method for simulating synthetic data from the model [4], and methods to check whether the assumptions of the model hold for a particular dataset, in this case local stationarity [5]. We also contributed to a new interpolated product from the Global Drifter Program dataset [6], which was then used in subsequent analyses to maximise the resolution of our output.
A particular focus was on data from the Global Drifter Program, a vast and unique collections of observations encompassing most of the globe’s ocean surface. In total over 20,000 drifters have been deployed since 1979, resulting in over 100 million observations of measurements including surface velocity, sea surface temperature, and salinity. The challenge with such datasets are that the drifters are moving in time and space, so rather than observing fixed-point measurements, we instead observe “trajectories” where the observations from each instrument are moving across and between the oceans. This challenge required new models and new statistical methodology for their implementation.
The main contribution of this project was the delivery of a fully-integrated statistical model (from design to implementation stage) that can be used to better understand the trajectories from drifters and related instruments. Our model uses stochastic processes to describe the variability and uncertainty that is inherent to drifter trajectories, in addition to capturing swirling rotational oscillations caused by the Coriolis effect - the effect due to the Earth’s rotation on its own axis - as well as from tides and local features known as eddies. Details of our model have been published to both the statistics [1] and physical oceanography [2] communities. Furthermore, the results from our global analysis have been disseminated at leading international conferences such as Ocean Sciences, the American Geophysical Union Fall meeting, the European Geophysical Union General Assembly, the Joint Statistical Meetings, and the Royal Statistical Society Annual Conference.
The implementation of our model to large-scale data required several new methodological developments in statistics and spatiotemporal data analysis. These include a computationally-efficient procedure for estimating parameters from the model [3], a method for simulating synthetic data from the model [4], and methods to check whether the assumptions of the model hold for a particular dataset, in this case local stationarity [5]. We also contributed to a new interpolated product from the Global Drifter Program dataset [6], which was then used in subsequent analyses to maximise the resolution of our output.
