Quantifying and attributing time step sensitivities in present-day climate simulations conducted with EAMv1

This study assesses the relative importance of time integration error in present-day climate simulations conducted with the atmosphere component of the Energy Exascale Earth System Model version 1 (EAMv1) at 1∘ horizontal resolution. We show that a factor-of-6 reduction of time step size in all major parts of the model leads to significant changes in the long-term mean climate. Examples of changes in 10-year mean zonal averages include the following: up to 0.5 K of warming in the lower troposphere and cooling in the tropical and subtropical upper troposphere, 1 %–10 % decreases in relative humidity throughout the troposphere, and 10 %–20 % decreases in cloud fraction in the upper troposphere and decreases exceeding 20 % in the subtropical lower troposphere. In terms of the 10-year mean geographical distribution, systematic decreases of 20 %–50 % are seen in total cloud cover and cloud radiative effects in the subtropics. These changes imply that the reduction of temporal truncation errors leads to a notable although unsurprising degradation of agreement between the simulated and observed present-day climate; to regain optimal climate fidelity in the absence of those truncation errors, the model would require retuning. A coarse-grained attribution of the time step sensitivities is carried out by shortening time steps used in various components of EAM or by revising the numerical coupling between some processes. Our analysis leads to the finding that the marked decreases in the subtropical low-cloud fraction and total cloud radiative effect are caused not by the step size used for the collectively subcycled turbulence, shallow convection, and stratiform cloud macrophysics and microphysics parameterizations but rather by the step sizes used outside those subcycles. Further analysis suggests that the coupling frequency between the subcycles and the rest of EAM significantly affects the subtropical marine stratocumulus decks, while deep convection has significant impacts on trade cumulus. The step size of the cloud macrophysics and microphysics subcycle itself appears to have a primary impact on cloud fraction in the upper troposphere and also in the midlatitude near-surface layers. Impacts of step sizes used by the dynamical core and the radiation parameterization appear to be relatively small. These results provide useful clues for future studies aiming at understanding and addressing the root causes of sensitivities to time step sizes and process coupling frequencies in EAM. While this study focuses on EAMv1 and the conclusions are likely model-specific, the presented experimentation strategy has general value for weather and climate model development, as the methodology can help researchers identify and understand sources of time integration error in sophisticated multi-component models.

QMC integration errors and quasi-asymptotics

A crude Monte Carlo (MC) method allows to calculate integrals over a d-dimensional cube. As the number N of integration nodes becomes large, the rate of probable error of the MC method decreases as {O(1/\sqrt{N})}. The use of quasi-random points instead of random points in the MC algorithm converts it to the quasi-Monte Carlo (QMC) method. The asymptotic error estimate of QMC integration of d-dimensional functions contains a multiplier {1/N}. However, the multiplier {(\ln N)^{d}} is also a part of the error estimate, which makes it virtually useless. We have proved that, in the general case, the QMC error estimate is not limited to the factor {1/N}. However, our numerical experiments show that using quasi-random points of Sobol sequences with {N=2^{m}} with natural m makes the integration error approximately proportional to {1/N}. In our numerical experiments, {d\leq 15}, and we used {N\leq 2^{40}} points generated by the SOBOLSEQ16384 code published in 2011. In this code, {d\leq 2^{14}} and {N\leq 2^{63}}.

Monte Carlo integration with a growing number of control variates

It is well known that Monte Carlo integration with variance reduction by means of control variates can be implemented by the ordinary least squares estimator for the intercept in a multiple linear regression model. A central limit theorem is established for the integration error if the number of control variates tends to infinity. The integration error is scaled by the standard deviation of the error term in the regression model. If the linear span of the control variates is dense in a function space that contains the integrand, the integration error tends to zero at a rate which is faster than the square root of the number of Monte Carlo replicates. Depending on the situation, increasing the number of control variates may or may not be computationally more efficient than increasing the Monte Carlo sample size.

Sources of path integration error in young and aging humans

Path integration is a vital function in navigation: it enables the continuous tracking of one’s position in space by integrating self-motion cues. Path integration abilities vary across individuals but tend to deteriorate in old age. The specific causes of path integration errors, however, remain poorly characterized. Here, we combined tests of path integration performance with a novel analysis based on the Langevin diffusion equation, which allowed us to decompose errors into distinct causes that can corrupt path integration computations. Across age groups, the dominant errors were due to noise and a bias in speed estimation. Noise-driven errors accumulated with travel distance not elapsed time, suggesting that the dominant noise originates in the velocity input rather than within the integrator. Age-related declines were traced primarily to a growth in this unbiased noise. Together, these findings shed light on the contributors to path integration error and the mechanisms underlying age-related navigational deficits.

Long Time Prediction of Uncertain Systems Using Singular Perturbation

This paper considers the problem of long time prediction of uncertain dynamic systems. Spectral methods such as polynomial chaos expansion (PCE) provides a suitable alternative for classical Monte Carlo method with lower computational load. However, polynomial chaos expansion has a major drawback of long time integration error. In this paper, we will apply singular perturbation (SP) method for reducing long time integration error. Using SP the accuracy of long time predictions are improved with comparable computational load. We will apply SP to illustrative exemplify problems to show effectiveness of our approach. Moreover, we will illustrate application of SP together with a model order reduction tool to reduce long time integration error as applied to distributed parameter.