Ensemble prediction systems typically show positive spread-error correlation, but they are subject to forecast bias and dispersion errors, and are therefore uncalibrated. This work proposes the use of ensemble model output statistics (EMOS), an easy-to-implement postprocessing technique that addresses both forecast bias and underdispersion and takes into account the spread-skill relationship. The technique is based on multiple linear regression and is akin to the superensemble approach that has traditionally been used for deterministic-style forecasts. The EMOS technique yields probabilistic forecasts that take the form of Gaussian predictive probability density functions (PDFs) for continuous weather variables and can be applied to gridded model output. The EMOS predictive mean is a bias-corrected weighted average of the ensemble member forecasts, with coefficients that can be interpreted in terms of the relative contributions of the member models to the ensemble, and provides a highly competitive deterministic-style forecast. The EMOS predictive variance is a linear function of the ensemble variance. For fitting the EMOS coefficients, the method of minimum continuous ranked probability score (CRPS) estimation is introduced. This technique finds the coefficient values that optimize the CRPS for the training data. The EMOS technique was applied to 48-h forecasts of sea level pressure and surface temperature over the North American Pacific Northwest in spring 2000, using the University of Washington mesoscale ensemble. When compared to the bias-corrected ensemble, deterministic-style EMOS forecasts of sea level pressure had root-mean-square error 9% less and mean absolute error 7% less. The EMOS predictive PDFs were sharp, and much better calibrated than the raw ensemble or the bias-corrected ensemble.

Long-standing questions in marine viral ecology are centered on understanding how viral assemblages change along gradients in space and time. However, investigating these fundamental ecological questions has been challenging due to incomplete representation of naturally occurring viral diversity in single gene-or morphology-based studies and an inability to identify up to 90% of reads in viral metagenomes (viromes). Although protein clustering techniques provide a significant advance by helping organize this unknown metagenomic sequence space, they typically use only ∼75% of the data and rely on assembly methods not yet tuned for naturally occurring sequence variation. Here, we introduce an annotation-and assembly-free strategy for comparative metagenomics that combines shared k-mer and social network analyses (regression modeling). This robust statistical framework enables visualization of complex sample networks and determination of ecological factors driving community structure. Application to 32 viromes from the Pacific Ocean Virome dataset identified clusters of samples broadly delineated by photic zone and revealed that geographic region, depth, and proximity to shore were significant predictors of community structure. Within subsets of this dataset, depth, season, and oxygen concentration were significant drivers of viral community structure at a single open ocean station, whereas variability along onshore-offshore transects was driven by oxygen concentration in an area with an oxygen minimum zone and not depth or proximity to shore, as might be expected. Together these results demonstrate that this highly scalable approach using complete metagenomic network-based comparisons can both test and generate hypotheses for ecological investigation of viral and microbial communities in nature.virus | microbial ecology | Bayesian network M icroorganisms drive global biogeochemical cycles (1), with abundances and taxonomic composition tuned to spatiotemporally varying environmental conditions (2-5). Viruses then modulate these biogeochemical processes through mortality, horizontal gene transfer, and metabolic reprogramming (6). However, our understanding of how viral communities change in response to biological, physical, and chemical factors and host availability has been limited by technical challenges.Most viruses in the ocean lack both cultivated representatives [85% of 1,100 sequenced phage genomes derive from only 3 of 45 bacterial phyla (7)] and a universally conserved marker gene (8); thus, metagenomics is commonly applied to characterize the ecology and evolution of viral assemblages. Problematically, however, our ability to investigate these assemblages via metagenomics remains limited by the lack of known viruses and viral proteins in biological sequence databases. The first viral metagenome (virome) used thousands of Sanger reads and found that 65% of sequences were unknown [i.e., no database match for reads >600 bp (9)]. Adoption of next-generation sequencing (NGS) technologies then generated hundreds...

The focus of this paper is an approach to the modeling of longitudinal social network or relational data. Such data arise from measurements on pairs of objects or actors made at regular temporal intervals, resulting in a social network for each point in time. In this article we represent the network and temporal dependencies with a random effects model, resulting in a stochastic process defined by a set of stationary covariance matrices. Our approach builds upon the social relations models of Warner, Kenny and Stoto [Journal of Personality and Social Psychology 37 (1979) 1742-1757] and Gill and Swartz [Canad. J. Statist. 29 (2001) 321-331] and allows for an intraand inter-temporal representation of network structures. We apply the methodology to two longitudinal data sets: international trade (continuous response) and militarized interstate disputes (binary response).

A food web consists of nodes, each consisting of one or more species. The role of each node as predator or prey determines the trophic relations that weave the web. Much effort in trophic food web research is given to understand the connectivity structure, or the nature and degree of dependence among nodes. Social network analysis (SNA) techniques—quantitative methods commonly used in the social sciences to understand network relational structure—have been used for this purpose, although postanalysis effort or biological theory is still required to determine what natural factors contribute to the feeding behavior. Thus, a conventional SNA alone provides limited insight into trophic structure. Here we show that by using novel statistical modeling methodologies to express network links as the random response of within- and internode characteristics (predictors), we gain a much deeper understanding of food web structure and its contributing factors through a unified statistical SNA. We do so for eight empirical food webs: Phylogeny is shown to have nontrivial influence on trophic relations in many webs, and for each web trophic clustering based on feeding activity and on feeding preference can differ substantially. These and other conclusions about network features are purely empirical, based entirely on observed network attributes while accounting for biological information built directly into the model. Thus, statistical SNA techniques, through statistical inference for feeding activity and preference, provide an alternative perspective of trophic clustering to yield comprehensive insight into food web structure.