Bayesian model comparison with un-normalised likelihoods

Models for which the likelihood function can be evaluated only up to a parameter-dependent unknown normalizing constant, such as Markov random field models, are used widely in computer science, statistical physics, spatial statistics, and network analysis. However, Bayesian analysis of these models using standard Monte Carlo methods is not possible due to the intractability of their likelihood functions. Several methods that permit exact, or close to exact, simulation from the posterior distribution have recently been developed. However, estimating the evidence and Bayes’ factors for these models remains challenging in general. This paper describes new random weight importance sampling and sequential Monte Carlo methods for estimating BFs that use simulation to circumvent the evaluation of the intractable likelihood, and compares them to existing methods. In some cases we observe an advantage in the use of biased weight estimates. An initial investigation into the theoretical and empirical properties of this class of methods is presented. Some support for the use of biased estimates is presented, but we advocate caution in the use of such estimates.

Bayesian analysis of uncertainty in the GlobCover 2009 land cover product at climate model grid scale

Land cover data derived from satellites are commonly used to prescribe inputs to models of the land surface. Since such data inevitably contains errors, quantifying how uncertainties in the data affect a model’s output is important. To do so, a spatial distribution of possible land cover values is required to propagate through the model’s simulation. However, at large scales, such as those required for climate models, such spatial modelling can be difficult. Also, computer models often require land cover proportions at sites larger than the original map scale as inputs, and it is the uncertainty in these proportions that this article discusses. This paper describes a Monte Carlo sampling scheme that generates realisations of land cover proportions from the posterior distribution as implied by a Bayesian analysis that combines spatial information in the land cover map and its associated confusion matrix. The technique is computationally simple and has been applied previously to the Land Cover Map 2000 for the region of England and Wales. This article demonstrates the ability of the technique to scale up to large (global) satellite derived land cover maps and reports its application to the GlobCover 2009 data product. The results show that, in general, the GlobCover data possesses only small biases, with the largest belonging to non–vegetated surfaces. In vegetated surfaces, the most prominent area of uncertainty is Southern Africa, which represents a complex heterogeneous landscape. It is also clear from this study that greater resources need to be devoted to the construction of comprehensive confusion matrices.

Statistical inference of OH concentrations and air mass dilution rates from successive observations of non-methane hydrocarbons in single air masses

Bayesian inference has been used to determine rigorous estimates of hydroxyl radical concentrations () and air mass dilution rates (K) averaged following air masses between linked observations of nonmethane hydrocarbons (NMHCs) spanning the North Atlantic during the Intercontinental Transport and Chemical Transformation (ITCT)-Lagrangian-2K4 experiment. The Bayesian technique obtains a refined (posterior) distribution of a parameter given data related to the parameter through a model and prior beliefs about the parameter distribution. Here, the model describes hydrocarbon loss through OH reaction and mixing with a background concentration at rate K. The Lagrangian experiment provides direct observations of hydrocarbons at two time points, removing assumptions regarding composition or sources upstream of a single observation. The estimates are sharpened by using many hydrocarbons with different reactivities and accounting for their variability and measurement uncertainty. A novel technique is used to construct prior background distributions of many species, described by variation of a single parameter . This exploits the high correlation of species, related by the first principal component of many NMHC samples. The Bayesian method obtains posterior estimates of , K and following each air mass. Median values are typically between 0.5 and 2.0 × 106 molecules cm−3, but are elevated to between 2.5 and 3.5 × 106 molecules cm−3, in low-level pollution. A comparison of estimates from absolute NMHC concentrations and NMHC ratios assuming zero background (the “photochemical clock” method) shows similar distributions but reveals systematic high bias in the estimates from ratios. Estimates of K are ∼0.1 day−1 but show more sensitivity to the prior distribution assumed.

A simple model of convection with memory

There are at least three distinct time scales that are relevant for the evolution of atmospheric convection. These are the time scale of the forcing mechanism, the time scale governing the response to a steady forcing, and the time scale of the response to variations in the forcing. The last of these, tmem, is associated with convective life cycles, which provide an element of memory in the system. A highly simplified model of convection is introduced, which allows for investigation of the character of convection as a function of the three time scales. For short tmem, the convective response is strongly tied to the forcing as in conventional equilibrium parameterization. For long tmem, the convection responds only to the slowly evolving component of forcing, and any fluctuations in the forcing are essentially suppressed. At intermediate tmem, convection becomes less predictable: conventional equilibrium closure breaks down and current levels of convection modify the subsequent response.

Correlating Bayesian date estimates with climatic events and domestication using a bovine case study

The tribe Bovini contains a number of commercially and culturally important species, such as cattle. Understanding their evolutionary time scale is important for distinguishing between post-glacial and domestication-associated population expansions, but estimates of bovine divergence times have been hindered by a lack of reliable calibration points. We present a Bayesian phylogenetic analysis of 481 mitochondrial D-loop sequences, including 228 radiocarbon-dated ancient DNA sequences, using a multi-demographic coalescent model. By employing the radiocarbon dates as internal calibrations, we co-estimate the bovine phylogeny and divergence times in a relaxed-clock framework. The analysis yields evidence for significant population expansions in both taurine and zebu cattle, European aurochs and yak clades. The divergence age estimates support domestication-associated expansion times (less than 12 kyr) for the major haplogroups of cattle. We compare the molecular and palaeontological estimates for the Bison-Bos divergence.

Spatial analysis of the error in a model of soil nitrogen

Models of the dynamics of nitrogen in soil (soil-N) can be used to aid the fertilizer management of a crop. The predictions of soil-N models can be validated by comparison with observed data. Validation generally involves calculating non-spatial statistics of the observations and predictions, such as their means, their mean squared-difference, and their correlation. However, when the model predictions are spatially distributed across a landscape the model requires validation with spatial statistics. There are three reasons for this: (i) the model may be more or less successful at reproducing the variance of the observations at different spatial scales; (ii) the correlation of the predictions with the observations may be different at different spatial scales; (iii) the spatial pattern of model error may be informative. In this study we used a model, parameterized with spatially variable input information about the soil, to predict the mineral-N content of soil in an arable field, and compared the results with observed data. We validated the performance of the N model spatially with a linear mixed model of the observations and model predictions, estimated by residual maximum likelihood. This novel approach allowed us to describe the joint variation of the observations and predictions as: (i) independent random variation that occurred at a fine spatial scale; (ii) correlated random variation that occurred at a coarse spatial scale; (iii) systematic variation associated with a spatial trend. The linear mixed model revealed that, in general, the performance of the N model changed depending on the spatial scale of interest. At the scales associated with random variation, the N model underestimated the variance of the observations, and the predictions were correlated poorly with the observations. At the scale of the trend, the predictions and observations shared a common surface. The spatial pattern of the error of the N model suggested that the observations were affected by the local soil condition, but this was not accounted for by the N model. In summary, the N model would be well-suited to field-scale management of soil nitrogen, but suited poorly to management at finer spatial scales. This information was not apparent with a non-spatial validation. (c),2007 Elsevier B.V. All rights reserved.

Calibration and sensitivity analysis of a river water quality model under unsteady flow conditions

Water quality models generally require a relatively large number of parameters to define their functional relationships, and since prior information on parameter values is limited, these are commonly defined by fitting the model to observed data. In this paper, the identifiability of water quality parameters and the associated uncertainty in model simulations are investigated. A modification to the water quality model `Quality Simulation Along River Systems' is presented in which an improved flow component is used within the existing water quality model framework. The performance of the model is evaluated in an application to the Bedford Ouse river, UK, using a Monte-Carlo analysis toolbox. The essential framework of the model proved to be sound, and calibration and validation performance was generally good. However some supposedly important water quality parameters associated with algal activity were found to be completely insensitive, and hence non-identifiable, within the model structure, while others (nitrification and sedimentation) had optimum values at or close to zero, indicating that those processes were not detectable from the data set examined. (C) 2003 Elsevier Science B.V. All rights reserved.

Performance and prediction: Bayesian modelling of fallible choice in chess

Evaluating agents in decision-making applications requires assessing their skill and predicting their behaviour. Both are well developed in Poker-like situations, but less so in more complex game and model domains. This paper addresses both tasks by using Bayesian inference in a benchmark space of reference agents. The concepts are explained and demonstrated using the game of chess but the model applies generically to any domain with quantifiable options and fallible choice. Demonstration applications address questions frequently asked by the chess community regarding the stability of the rating scale, the comparison of players of different eras and/or leagues, and controversial incidents possibly involving fraud. The last include alleged under-performance, fabrication of tournament results, and clandestine use of computer advice during competition. Beyond the model world of games, the aim is to improve fallible human performance in complex, high-value tasks.

The physical properties of the atmosphere in the new Hadley Centre Global Environmental Model (HadGEM1). Part II: Aspects of variability and regional climate

The performance of the atmospheric component of the new Hadley Centre Global Environmental Model (HadGEM1) is assessed in terms of its ability to represent a selection of key aspects of variability in the Tropics and extratropics. These include midlatitude storm tracks and blocking activity, synoptic variability over Europe, and the North Atlantic Oscillation together with tropical convection, the Madden-Julian oscillation, and the Asian summer monsoon. Comparisons with the previous model, the Third Hadley Centre Coupled Ocean-Atmosphere GCM (HadCM3), demonstrate that there has been a considerable increase in the transient eddy kinetic energy (EKE), bringing HadGEM1 into closer agreement with current reanalyses. This increase in EKE results from the increased horizontal resolution and, in combination with the improved physical parameterizations, leads to improvements in the representation of Northern Hemisphere storm tracks and blocking. The simulation of synoptic weather regimes over Europe is also greatly improved compared to HadCM3, again due to both increased resolution and other model developments. The variability of convection in the equatorial region is generally stronger and closer to observations than in HadCM3. There is, however, still limited convective variance coincident with several of the observed equatorial wave modes. Simulation of the Madden-Julian oscillation is improved in HadGEM1: both the activity and interannual variability are increased and the eastward propagation, although slower than observed, is much better simulated. While some aspects of the climatology of the Asian summer monsoon are improved in HadGEM1, the upper-level winds are too weak and the simulation of precipitation deteriorates. The dominant modes of monsoon interannual variability are similar in the two models, although in HadCM3 this is linked to SST forcing, while in HadGEM1 internal variability dominates. Overall, analysis of the phenomena considered here indicates that HadGEM1 performs well and, in many important respects, improves upon HadCM3. Together with the improved representation of the mean climate, this improvement in the simulation of atmospheric variability suggests that HadGEM1 provides a sound basis for future studies of climate and climate change.

The retrieval of ice water content from radar reflectivity factor and temperature and its use in evaluating a mesoscale model

Ice clouds are an important yet largely unvalidated component of weather forecasting and climate models, but radar offers the potential to provide the necessary data to evaluate them. First in this paper, coordinated aircraft in situ measurements and scans by a 3-GHz radar are presented, demonstrating that, for stratiform midlatitude ice clouds, radar reflectivity in the Rayleigh-scattering regime may be reliably calculated from aircraft size spectra if the "Brown and Francis" mass-size relationship is used. The comparisons spanned radar reflectivity values from -15 to +20 dBZ, ice water contents (IWCs) from 0.01 to 0.4 g m(-3), and median volumetric diameters between 0.2 and 3 mm. In mixed-phase conditions the agreement is much poorer because of the higher-density ice particles present. A large midlatitude aircraft dataset is then used to derive expressions that relate radar reflectivity and temperature to ice water content and visible extinction coefficient. The analysis is an advance over previous work in several ways: the retrievals vary smoothly with both input parameters, different relationships are derived for the common radar frequencies of 3, 35, and 94 GHz, and the problem of retrieving the long-term mean and the horizontal variance of ice cloud parameters is considered separately. It is shown that the dependence on temperature arises because of the temperature dependence of the number concentration "intercept parameter" rather than mean particle size. A comparison is presented of ice water content derived from scanning 3-GHz radar with the values held in the Met Office mesoscale forecast model, for eight precipitating cases spanning 39 h over Southern England. It is found that the model predicted mean I WC to within 10% of the observations at temperatures between -30 degrees and - 10 degrees C but tended to underestimate it by around a factor of 2 at colder temperatures.

Evaluation of the sea ice simulation in a new coupled atmosphere-ocean climate model (HadGEM1)

[ 1] A rapid increase in the variety, quality, and quantity of observations in polar regions is leading to a significant improvement in the understanding of sea ice dynamic and thermodynamic processes and their representation in global climate models. We assess the simulation of sea ice in the new Hadley Centre Global Environmental Model (HadGEM1) against the latest available observations. The HadGEM1 sea ice component uses elastic-viscous-plastic dynamics, multiple ice thickness categories, and zero-layer thermodynamics. The model evaluation is focused on the mean state of the key variables of ice concentration, thickness, velocity, and albedo. The model shows good agreement with observational data sets. The variability of the ice forced by the North Atlantic Oscillation is also found to agree with observations.

Mechanisms of ocean heat uptake in a coupled climate model and the implications for tracer based predictions of ocean heat uptake

The distribution of tracers in the ocean is often taken as an indication of the ventilation pathways for oceanic water masses. It has been suggested that under anthropogenic forcing heat will be taken up into the interior of the ocean along isopycnal ventilation pathways. This notion is investigated by examining distributions of potential temperature and a passive anomaly temperature tracer in a coupled climate experiment where CO2 is increased at a rate of 2% per year. We show that interior temperature changes cannot be explained solely by passive tracer transport along isopycnals. Heat uptake is strongly affected by changes in circulation and has a substantial diapycnal component.

Representing El Niño in coupled ocean–atmosphere GCMs: the dominant role of the atmospheric component

A systematic modular approach to investigate the respective roles of the ocean and atmosphere in setting El Niño characteristics in coupled general circulation models is presented. Several state-of-the-art coupled models sharing either the same atmosphere or the same ocean are compared. Major results include 1) the dominant role of the atmosphere model in setting El Niño characteristics (periodicity and base amplitude) and errors (regularity) and 2) the considerable improvement of simulated El Niño power spectra—toward lower frequency—when the atmosphere resolution is significantly increased. Likely reasons for such behavior are briefly discussed. It is argued that this new modular strategy represents a generic approach to identifying the source of both coupled mechanisms and model error and will provide a methodology for guiding model improvement.

Effectiveness of landmark analysis for establishing locality in p2p networks

Locality to other nodes on a peer-to-peer overlay network can be established by means of a set of landmarks shared among the participating nodes. Each node independently collects a set of latency measures to landmark nodes, which are used as a multi-dimensional feature vector. Each peer node uses the feature vector to generate a unique scalar index which is correlated to its topological locality. A popular dimensionality reduction technique is the space filling Hilbert’s curve, as it possesses good locality preserving properties. However, there exists little comparison between Hilbert’s curve and other techniques for dimensionality reduction. This work carries out a quantitative analysis of their properties. Linear and non-linear techniques for scaling the landmark vectors to a single dimension are investigated. Hilbert’s curve, Sammon’s mapping and Principal Component Analysis have been used to generate a 1d space with locality preserving properties. This work provides empirical evidence to support the use of Hilbert’s curve in the context of locality preservation when generating peer identifiers by means of landmark vector analysis. A comparative analysis is carried out with an artificial 2d network model and with a realistic network topology model with a typical power-law distribution of node connectivity in the Internet. Nearest neighbour analysis confirms Hilbert’s curve to be very effective in both artificial and realistic network topologies. Nevertheless, the results in the realistic network model show that there is scope for improvements and better techniques to preserve locality information are required.

Constraining a North Atlantic Ocean general circulation model with chlorofluorocarbon observations

Measurements of anthropogenic tracers such as chlorofluorocarbons and tritium must be quantitatively combined with ocean general circulation models as a component of systematic model development. The authors have developed and tested an inverse method, using a Green's function, to constrain general circulation models with transient tracer data. Using this method chlorofluorocarbon-11 and -12 (CFC-11 and -12) observations are combined with a North Atlantic configuration of the Miami Isopycnic Coordinate Ocean Model with 4/3 degrees resolution. Systematic differences can be seen between the observed CFC concentrations and prior CFC fields simulated by the model. These differences are reduced by the inversion, which determines the optimal gas transfer across the air-sea interface, accounting for uncertainties in the tracer observations. After including the effects of unresolved variability in the CFC fields, the model is found to be inconsistent with the observations because the model/data misfit slightly exceeds the error estimates. By excluding observations in waters ventilated north of the Greenland-Scotland ridge (sigma (0) < 27.82 kg m(-3); shallower than about 2000 m), the fit is improved, indicating that the Nordic overflows are poorly represented in the model. Some systematic differences in the model/data residuals remain and are related, in part, to excessively deep model ventilation near Rockall and deficient ventilation in the main thermocline of the eastern subtropical gyre. Nevertheless, there do not appear to be gross errors in the basin-scale model circulation. Analysis of the CFC inventory using the constrained model suggests that the North Atlantic Ocean shallower than about 2000 m was near 20% saturated in the mid-1990s. Overall, this basin is a sink to 22% of the total atmosphere-to-ocean CFC-11 flux-twice the global average value. The average water mass formation rates over the CFC transient are 7.0 and 6.0 Sv (Sv = 10(6) m(3) s(-1)) for subtropical mode water and subpolar mode water, respectively.