Emulation and interpretation of high-dimensional climate model outputs

Abstract

Running complex computer models can be expensive in computer time, while learning about the relationships between input and output variables can be difficult. An emulator is a fast approximation to a computationally expensive model that can be used as a surrogate for the model, to quantify uncertainty or to improve process understanding. Here, we examine emulators based on singular value decompositions and use them to emulate global climate and vegetation fields, examining how these fields are affected by changes in the Earth's orbit. The vegetation field may be emulated directly from the orbital variables, but an appealing alternative is to relate it to emulations of the climate fields, which involves high-dimensional input and output. The singular value decompositions radically reduce the dimensionality of the input and output spaces and are shown to clarify the relationships between them. The method could potentially be useful for any complex process with correlated, highdimensional inputs and/or outputs

Transient Earth system responses to cumulative carbon dioxide emissions: linearities, uncertainties, and probabilities in an observation-constrained model ensemble

“…Wigley and Raper, 2001;Knutti et al, 2002Knutti et al, , 2003Knutti et al, , 2005Schleussner et al, 2014;Bodman et al, 2013;Little et al, 2013;Harris et al, 2013;Holden et al, 2013;Bhat et al, 2012), Bern3D-LPJ features a dynamic threedimensional ocean with physically consistent formulations for the transport of heat, carbon, and other biogeochemical tracers, similar to work by Holden et al (2010) and Olson et al (2012), and includes a state-of-the-art dynamic global vegetation model, peat carbon, and anthropogenic land-use dynamics. The model is applied directly without using an emulator (Holden et al, 2010(Holden et al, , 2015Olson et al, 2012). Further, we note that no ocean carbonate chemistry or marine biology parameters were varied in this study.…”

Toward Data‐Driven Weather and Climate Forecasting: Approximating a Simple General Circulation Model With Deep Learning

“…They solve discretized physical equations of the physics of the atmosphere, in order to compute the evolution of atmospheric states over time. Recently, numerous applications using machine learning techniques in connection with GCMs and weather prediction models have been proposed, including learning relations between orbital parameters and climate fields from a climate model (Holden et al, 2015), learning from high-resolution simulations in order to improve predictions made with simpler models (Anderson & Lucas, 2018;Rasp et al, 2018), helping in decision making in extreme weather situations (McGovern et al, 2017), detecting extreme weather in climate data sets (Liu et al, 2016), and predicting the uncertainty of weather forecasts (Scher & Messori, 2018). All these proposed techniques are valuable techniques for climate science and meteorology.…”

Compression and Conditional Emulation of Climate Model Output

“…Recently, there has been interest in fitting statistical models to climate model output for the purpose of building computationally efficient model emulators (Castruccio et al, 2014(Castruccio et al, , 2013Williamson and Blaker, 2014;Holden et al, 2015;Tran et al, 2016) and for use as a tool for compressing the output. Previous work has viewed the parameters in the statistical model as the compressed object (Castruccio and Genton, 2016;Castruccio and Guinness, 2017), and thus if the model is well-specified, it can be used to generate emulated model runs whose mean and covariance structure matches that of the original dataset.…”