Reducing storage of global wind ensembles with stochastic generators

Abstract

Wind has the potential to make a significant contribution to future energy resources. Locating the sources of this renewable energy on a global scale is however extremely challenging, given the difficulty to store very large data sets generated by modern computer models. We propose a statistical model that aims at reproducing the data-generating mechanism of an ensemble of runs via a Stochastic Generator (SG) of global annual wind data. We introduce an evolutionary spectrum approach with spatially varying parameters based on large-scale geographical descriptors such as altitude to better account for different regimes across the Earth's orography. We consider a multi-step conditional likelihood approach to estimate the parameters that explicitly accounts for nonstationary features while also balancing memory storage and distributed computation. We apply the proposed model to more than 18 million points of yearly global wind speed. The proposed SG requires orders of magnitude less storage for generating surrogate ensemble members from wind than does creating additional wind fields from the climate model, even if an effective lossy data compression algorithm is applied to the simulation output.

Principles for statistical inference on big spatio-temporal data from climate models

“…where H r (c, L m , t k ) is the spectral process assumed to be independent across wavenumbers, c. Hence, for the storage and computation of f (c), model (4) requires O(N ) entries and O(N logN ) operations, respectively, as opposed to the covariance function, which requires O(N 2 ) entries and O(N 3 ) operations, respectively. We can also generalize (4) to allow for changing behavior across domains such as land and ocean (Castruccio and Guinness, 2017) or altitude (Jeong et al, 2018). Besides diminishing the storage and computational burdens, this approach allows the inference to be performed in parallel across m, i.e., across latitude, following Principle 2.…”

“…Here, we show only the temporal dependence; we defer the longitudinal and latitudinal step to the next section. We apply Principle 2 to model (3) below; the diagnostics can be found for annual wind speed in Jeong et al (2018) and for temperature in Castruccio and Genton (2016). If the vector of all values of r at time t k is denoted by r (t k ), then…”

“…In this work, we focus on the very end of the spectrum in terms of data size, i.e., on data sets generated from climate model ensembles, which are typically between 100 million to 10 billion points, and we provide three general principles that enable us to perform inference for nontrivial models within a reasonable time. These principles have emerged from a series of recent works on inference from extremely big spatio-temporal data (Castruccio and Stein, 2013;Castruccio and Genton, 2014;Castruccio, 2016;Castruccio and Genton, 2016;Castruccio and Guinness, 2017;Jeong et al, 2018) and ascribe to the general philosophy of the methods previously described, i.e., reduction of the information by exploiting the structure of a particular problem. These principles advocate designing a statistical model that fully leverages on parallel and high performance computing.…”

Visualizing Spatiotemporal Models with Virtual Reality: from Fully Immersive Environments to Applications in Stereoscopic View

“…3(a) for a snapshot). A similar app, called 'global surface wind' displays the annual winds between 1920 and 2100 across the globe according to a climate model and another statistical model (full details on the model are provided by Jeong et al (2018)). This app uses a similar visualization scheme to that in the temperature case, but it lacks the three-dimensional component as the model focuses on surface wind speed.…”

Comment

“…In particular, the multi-step spectrum model was designed to consider nonstationary covariance models across longitudes and to allow analysis of very large datasets by evaluating the likelihood with parallel and distributed computing. Castruccio and Guinness (2017) and Jeong et al (2018) showed how these models can be coupled with a land/ocean indicator and mountain ranges in the evolutionary spectrum. Since data from orbiting satellites have a particular observational location structure, the implementation of the above spectrum procedure on a non-gridded structure may be problematic.…”