Jarno Vanhatalo

Approximate inference for disease mapping with sparse Gaussian processes

Gaussian process (GP) models are widely used in disease mapping as they provide a natural framework for modeling spatial correlations. Their challenges, however, lie in computational burden and memory requirements. In disease mapping models, the other difficulty is inference, which is analytically intractable due to the non-Gaussian observation model. In this paper, we address both these challenges. We show how to efficiently build fully and partially independent conditional (FIC/PIC) sparse approximations for the GP in two-dimensional surface, and how to conduct approximate inference using expectation propagation (EP) algorithm and Laplace approximation (LA). We also propose to combine FIC with a compactly supported covariance function to construct a computationally efficient additive model that can model long and short length-scale spatial correlations simultaneously. The benefit of these approximations is computational. The sparse GPs speed up the computations and reduce the memory requirements. The posterior inference via EP and Laplace approximation is much faster and is practically as accurate as via Markov chain Monte Carlo.

Toward integrative management advice of water quality, oil spills, and fishery in the Gulf of Finland: a Bayesian approach.

Understanding and managing ecosystems affected by several anthropogenic stressors require methods that enable analyzing the joint effects of different factors in one framework. Further, as scientific knowledge about natural systems is loaded with uncertainty, it is essential that analyses are based on a probabilistic approach. We describe in this article about building a Bayesian decision model, which includes three stressors present in the Gulf of Finland. The outcome of the integrative model is a set of probability distributions for future nutrient concentrations, herring stock biomass, and achieving the water quality targets set by HELCOM Baltic Sea Action Plan. These distributions can then be used to derive the probability of reaching the management targets for each alternative combination of management actions.

Preparing for the unprecedented - Towards quantitative oil risk assessment in the Arctic marine areas.

The probability of major oil accidents in Arctic seas is increasing alongside with increasing maritime traffic. Hence, there is a growing need to understand the risks posed by oil spills to these unique and sensitive areas. So far these risks have mainly been acknowledged in terms of qualitative descriptions. We introduce a probabilistic framework, based on a general food web approach, to analyze ecological impacts of oil spills. We argue that the food web approach based on key functional groups is more appropriate for providing holistic view of the involved risks than assessments based on single species. We discuss the issues characteristic to the Arctic that need a special attention in risk assessment, and provide examples how to proceed towards quantitative risk estimates. The conceptual model presented in the paper helps to identify the most important risk factors and can be used as a template for more detailed risk assessments.

Experiences in Bayesian Inference in Baltic Salmon Management

Abstract. We review a success story regarding Bayesian inference infisheries management in the Baltic Sea. The management of salmonfisheries is currently based on the results of a complex Bayesian pop-ulation dynamic model, and managers and stakeholders use the prob-abilities in their discussions. We also discuss the technical and humanchallenges in using Bayesian modeling to give practical advice to thepublic and to government officials and suggest future areas in which itcan be applied. In particular, large databases in fisheries science offerflexible ways to use hierarchical models to learn the population dynam-ics parameters for those by-catch species that do not have similar largestock-specific data sets like those that exist for many target species.This information is required if we are to understand the future ecosys-tem risks of fisheries. Key words and phrases: Bayesian inference, Baltic salmon, risk anal-ysis, fishery management, decision analysis.1. INTRODUCTIONWe introduce a case of fisheries managementwhere Bayesian inference has been extensively used.Fisheries management is a field of applied science,and one could easily argue that fisheries science is

Spatiotemporal modelling of crown‐of‐thorns starfish outbreaks on the Great Barrier Reef to inform control strategies

We thank all those who have contributed to the Australian Institute of Marine Science (AIMS) surveys through the decades. The program has been partially supported by the Australian Governments Marine and Tropical Sciences Research Facility and National Environmental Research Program (NERP). This work was undertaken for the Marine Biodiversity Hub, a collaborative partnership supported through funding from the Australian Governments National Environmental Research Program (NERP). NERP Marine Biodiversity Hub partners include the Institute for Marine and Antarctic Studies, University of Tasmania; CSIRO, Geoscience Australia, Australian Institute of Marine Science, Museum Victoria, Charles Darwin University and the University of Western Australia. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. We thank Keith Hayes and Dan Pagendam for their comments on the manuscript. J.V. was funded by the Academy of Finland (grant 266349), research funds of the University of Helsinki and the Finnish Science Foundation for Economics and Technology (KAUTE). G.R.H. and H.S. were supported by the NERP Marine Biodiversity Hub. The authors declare no conflict of interest.

By-Catch of Grey Seals (Halichoerus grypus) in Baltic Fisheries—A Bayesian Analysis of Interview Survey

Baltic seals are recovering after a population decline. The increasing seal stocks cause notable damage to fisheries in the Baltic Sea, with an unknown number of seals drowning in fishing gear every year. Thus, sustainable seal management requires updated knowledge of the by-catch of seals—the number of specimens that die in fishing gear. We analyse the by-catch of grey seals (Halichoerus grypus) in Finland, Sweden, and Estonia in 2012. We collect data with interviews (35 in Finland, 54 in Sweden, and 72 in Estonia) and analyse them with a hierarchical Bayesian model. The model accounts for variability in seal abundance, seal mortality and fishing effort in different sub-areas of the Baltic Sea and allows us to predict the by-catch in areas where interview data was not available. We provide a detailed description of the survey design and interview methods, and discuss different factors affecting fishermens motivation to report by-catch and how this may affect the results. Our analysis shows that the total yearly by-catch by trap and gill nets in Finland, Sweden and Estonia is, with 90% probability, more than 1240 but less than 2860; and the posterior median and mean of the total by-catch are 1550 and 1880 seals, respectively. Trap nets make about 88% of the total by-catch. However, results also indicate that in one of the sub-areas of this study, fishermen may have underreported their by-catch. Taking the possible underreporting into account the posterior mean of the total by-catch is between 2180 and 2380. The by-catch in our study area is likely to represent at least 90% of the total yearly grey seal by-catch in the Baltic Sea.

Scenario Based Risk Management for Arctic Shipping and Operations

Arctic oil and gas explorations and Arctic shipping must ensure the safety and protection of this sensitive environment in spite of the challenging operational conditions. However, current regulations and assessment methods do not predict the associated risk level reliably. In other words, ships transiting ice-covered waters are not designed according to physical measures, such as accurate limit states under ice loading, but according to economic and empirical design measures. Similarly, offshore installations should be designed according to the accurate limit states, but the actual ice loads are uncertain so this is not possible at present. Risk-based design methodologies using first principal methods offer a way to advance safe operations and transport of natural resources within and out of the Arctic Sea. This paper introduces a holistic treatment of the design relevant features and their identification to improve safe Arctic operations and transport. The focus is on design relevant Arctic aspects related to extreme and accidental ice events. The approach includes estimating ice loads, including extreme load events, assessing structural consequences of the loading events, assessing associated potential environmental consequences, and establishing a risk based design framework for managing risks. Copyright

Hydrographic responses to regional covariates across the Kara Sea

The Kara Sea is a shelf sea in the Arctic Ocean which has a strong spatiotemporal hydrographic variation driven by river discharge, air pressure, and sea ice. There is a lack of information about the effects of environmental variables on surface hydrography in different regions of the Kara Sea. We use a hierarchical spatially varying coefficient model to study the variation of sea surface temperature (SST) and salinity (SSS) in the Kara Sea between years 1980 and 2000. The model allows us to study the effects of climatic (Arctic oscillation index (AO)) and seasonal (river discharge and ice concentration) environmental covariates on hydrography. The hydrographic responses to covariates vary considerably between different regions of the Kara Sea. River discharge decreases SSS in the shallow shelf area and has a neutral effect in the northern Kara Sea. The responses of SST and SSS to AO show the effects of different wind and air pressure conditions on water circulation and hence on hydrography. Ice concentration has a constant effect across the Kara Sea. We estimated the average SST and SSS in the Kara Sea in 1980–2000. The average August SST over the Kara Sea in 1995–2000 was higher than the respective average in 1980–1984 with 99.9% probability and August SSS decreased with 77% probability between these time periods. We found a support that the winter season AO has an impact on the summer season hydrography, and temporal trends may be related to the varying level of winter season AO index.

Laplace approximation and natural gradient for Gaussian process regression with heteroscedastic student-t model

We propose the Laplace method to derive approximate inference for Gaussian process (GP) regression in the location and scale parameters of the student-