Bayesreef: A Bayesian inference framework for modelling reef growth in response to environmental change and biological dynamics
Introduction
Developing data-driven models of reef evolution is challenging because the complexity of the process exceeds the amount of available data necessary. Reef evolution is determined by the interaction between environmental factors such as water chemistry, light availability, sedimentation and hydrodynamic energy (Veron, 2008). The data for understanding and modelling reef evolution can be extracted from the geological record of reef-drilled cores, which is sparse, and also expensive to obtain; hence, limited work has been done in this area.
pyReef-Core (Salles et al., 2018) is an example of a carbonate stratigraphic forward model (SFM) for reef evolution that captures a number of important ecological dynamics in coral reef systems. pyReef-Core is the first modelling attempt to constrain hydrodynamic energy and sediment input exposure thresholds for coralgal assemblages on a geological timescale. It is a one-dimensional (1-D) model that simulates the vertical (and not lateral, hence 1-D) coralgal growth patterns observed in a reef drill core. pyReef-Core has several parameters representing external environmental factors which impact reef development. Examples of these factors include sea-level changes and the relationship between sediment input and depth. It also has parameters describing the response of coralgal assemblage growth to these environmental factors, such as water flow and parameters for internal population dynamics such as the Malthusian parameter. Fig. 2 shows the workflow of pyReef-Core, which shows vertical accumulation contributed by different coralgal assemblages over a timeframe.
Given limited and sparse data, we need to quantify uncertainty from different factors in the estimation of unknown parameters in stratigraphic forward models such as pyReef-Core. Moreover, geophysical and stratigraphic forward models rarely have a unique solution (Isakov, 1993; Charvin et al., 2009a) which is also known as non-uniqueness (Burgess and Prince, 2015). For example, different combinations of a range of environmental parameters such as water flow, temperature and population dynamics of the coral assemblages in pyReef-Core may give rise to the same simulated reef-core stratigraphy. Stratigraphic forward models produce a set of solutions that represent multiple and competing hypotheses regarding geological system evolution (Cross et al., 1999; Heller et al., 1993). However, the explicit temporal and depth structure simulated by pyReef-Core presents an opportunity to restrict the number of possible solutions and thus reduce uncertainty.
Bayesian inference provides a rigorous methodology for estimation and uncertainty quantification of unknown parameters in a given model by incorporating information from multiple sources (Robert and Rousseau, 2009). The information from prior research, expert opinion, and knowledge regarding the nature of specific physical processes can be incorporated via a set of prior beliefs, also known as priors. Moreover, information from observed data is used to update these prior beliefs via the likelihood function. In the case of environmental modelling, Bayesian inference for uncertainty quantification has been deployed for a number of problems (O'Hagan, 2012; Hassan et al., 2009; Raje and Krishnan, 2012; Refsgaard et al., 2007).
We present a novel probabilistic framework for the estimation and uncertainty quantification of environmental processes and factors which impact the depth and temporal distribution of communities of corals and coralline algae (coralgal assemblages) found in fossil reef drill cores. The framework is called Bayesreef which employs Bayesian inference using Markov Chain Monte Carlo (MCMC) sampling to estimate the unknown parameters of pyReef-Core Although we chose pyReef-Core to demonstrate the idea, the framework is general, and can be adapted, for other stratigraphic forward models. The goal of Bayesreef is to provide estimation and uncertainty quantification for complex processes with sparse data which presents four significant contributions to the literature.
First, we transform a selected deterministic stratigraphic forward model (pyReef-Core) into a probabilistic framework known as Bayesreef, where the parameters are sampled via MCMC and represented using a probability distribution. Bayesreef employs a multinomial likelihood using the data from drilled reef-core featuring the coralgal assemblages over a timeframe. The expert opinion and the results of previous studies are incorporated into the prior, while knowledge of the physical processes from the pyReef-Core is connected to the observed assemblage in the reef-core via the multinomial likelihood.
Second, we use Bayesreef to constrain the number of solutions that represent the unique palaeo-environmental history of the reef core by incorporating two different types of data representation, where one features the temporal structure and the other the depth structure of the reef-core. By depth structure, we refer to the thickness and type of coralgal assemblage at varying depths. By temporal structure, we refer to the thickness and type of sediment and/or coralgal material laid down at varying points in time.
Third, we demonstrate the effectiveness of Bayesreef with two sampling methods, that includes single-chain MCMC and parallel tempering MCMC, for synthetic and real reef-core drilled from a selected location in the Great Barrier Reef.
Fourth, we make the methodology and its implementation available for other researchers as a software tool ,1 which can be used to make inference regarding the factors affecting reef evolution.
The rest of the paper is outlined as follows: Section 2 provides background and related work, while Section 3 presents the methodology and techniques used including the multinomial likelihood function. Section 4 presents experiments and results. Section 5 provides discussion, and Section 6 concludes the paper with a discussion of future research.
Section snippets
Coral reef evolution
The ability of corals to vigorously grow and build reef structures is dependent upon favourable environmental conditions (Done, 2011). Three related environmental factors examined in pyReef-Core are vital in influencing coral reef evolution on multi-decadal to centennial timescales. They are water depth (accommodation), hydrodynamic energy, and autochthonous (reef-derived) sediment input.
The accommodation is the vertical space in the water column above the substrate within which corals can grow
Methodology: Bayesreef
While deterministic models of long-term interactions of organisms in a marine ecosystem exist (Clavera-Gispert et al., 2017), and software tools for coral reef evolution are available (Salles et al., 2018; Barrett and Webster, 2017), there are no probabilistic methods which combine these deterministic models with observed data. We note that such models have several or many non-unique solutions; however, even if we manage to constrain them to be unique, there remains uncertainty with the
Experimental design and results
We investigate the performance of Bayesreef using the multinomial likelihood function outlined in the previous section for the time and depth structure and their associated time-based and depth-based likelihoods as the basis of prediction and evaluation of results. We use two different MCMC sampling methods for synthetic reef-core and apply to a selected application from the Great Barrier Reef. The experiments follow the plan below.
- 1.
Single-chain MCMC sampling (Algorithm 2) for two free
Discussion
We applied Bayesreef to infer how coralgal community structure changes in relation to prevailing changes in accommodation, hydrodynamic energy and sediment input. The results show that Bayesreef provides a reliable prediction of the synthetic ground truth data in the experiments that consider the depth and time-based likelihood and different combination of selected free parameters. Moreover, the estimated parameters provide an accurate prediction of the reef-core when compared to the synthetic
Conclusion and future work
We presented Bayesreef, which is a comprehensive Bayesian framework that incorporates multiple sources of information, including forward models, priors and empirical data from geological reef-cores. The Bayesreef framework employ single-chain and parallel tempering MCMC sampling to address the challenges of multimodal posteriors distributions given synthetic reef-core and an application from the Great Barrier Reef. The framework addresses geophysical inverse problem posed by unobserved
Data and software
The source code that implements the framework of Bayesreef can be downloaded from the Github repository .2
Supplementary results for selected experiments are given here 3
Acknowledgements
We would like to acknowledge Australian Research Council (ARC - DP120101793 and FT140101266) and Sydney Research Excellence Initiative 2017, The University of Sydney for providing the grants for supporting this work. We also wish to acknowledge Sydney Informatics Hub for providing research engineering support for this project.
References (80)
- et al.
Impact of sea-level rise and coral mortality on the wave dynamics and wave forces on barrier reefs
Mar. Pollut. Bull.
(2014) - et al.
Holocene evolution of the great barrier reef: insights from 3d numerical modelling
Sediment. Geol.
(2012) - et al.
Reef sedimentary accretion model (reefsam): understanding coral reef evolution on holocene time scales using 3d stratigraphic forward modelling
Mar. Geol.
(2017) A history of Runge-Kutta methods
Appl. Numer. Math.
(1996)- et al.
Bayeslands: A Bayesian Inference Approach for Parameter Uncertainty Quantification in Badlands
(2019) - et al.
Langevin-gradient parallel tempering for bayesian neural learning
Neurocomputing
(2019) An abstract volterra equation with applications to linear viscoelasticity
J. Differ. Equ.
(1970)- et al.
“Holocene ”turn-on” and evolution of the southern great barrier reef: revisiting reef cores from the capricorn bunker group
Mar. Geol.
(2015) - et al.
Influence of hydrodynamic energy on holocene reef flat accretion, great barrier reef
Quat. Res.
(2016) - et al.
The evolution of the great barrier reef during the last interglacial period
Glob. Planet. Chang.
(2017)
Environmental impacts of dredging and other sediment disturbances on corals: a review
Mar. Pollut. Bull.
A bayesian belief network to assess rate of changes in coral reef ecosystems
Environ. Model. Softw
Markov chain Monte Carlo (mcmc) sampling methods to determine optimal models, model resolution and model choice for earth science problems
Mar. Pet. Geol.
Using Markov chain Monte Carlo to quantify parameter uncertainty and its effect on predictions of a groundwater flow model
Environ. Model. Softw
Relationship between species diversity and reef growth in the holocene at ishigaki island, pacific ocean
Sediment. Geol.
History of indo-pacific coral reef systems since the last glaciation: development patterns and controlling factors
Earth Sci. Rev.
Probabilistic uncertainty specification: overview, elaboration techniques and their application to a mechanistic model of carbon flux
Environ. Model. Softw.
“Uncertainty in the environmental modelling process – a framework and guidance
Environ. Model. Softw
Holocene reef growth over irregular pleistocene karst confirms major influence of hydrodynamic factors on holocene reef development
Quat. Sci. Rev.
Scleractinian assemblages under sediment input: their characteristics and relation to the nutrient input concept
Palaeogeogr. Palaeoclimatol. Palaeoecol.
Effects of sedimentation on settlement of acropora millepora
Coral Reefs
The great barrier reef: the chronological record from a new borehole
J. Sediment. Res.
Tracing the influence of land-use change on water quality and coral reefs using a bayesian model
Sci. Rep.
Non-unique stratal geometries: implications for sequence stratigraphic interpretations
Basin Res.
Postglacial growth history of a French polynesian barrier reef tract, tahiti, central pacific
Sedimentology
Coral reef response to quaternary sea-level and environmental changes: state of the science
Sedimentology
Multi-core Parallel Tempering Bayeslands for Basin and Landscape Evolution
Coral morphology, diversity and reef growth
Nature
A bayesian approach to inverse modelling of stratigraphy, part 1: Method
Basin Res.
A bayesian approach to inverse modelling of stratigraphy, part 1: Method
Basin Res.
Process-based forward numerical ecological modeling for carbonate sedimentary basins
Comput. Geosci.
Water flow modulates the response of coral reef communities to ocean acidification
Sci. Rep.
Construction and application of a stratigraphic inverse model
Growth fabrics and growth-rates of holocene reefs in the great barrier-reef
BMR J. Aust. Geol. Geophys.
Relationship between reef growth and sea-level in the great barrier reef
Evolution of the Great Barrier Reef over the Last 130 Ka; a Multifaceted Approach, Integrating Palaeo Ecological, Palaeo Environmental and Chronological Data from Cores
“Corals: Environmental Controls on Growth,” Encyclopedia of Modern Coral Reefs: Structure, Form and Process
Temperature and phytoplankton growth in the sea
Fish. Bull.
Mass-transfer limitation of nutrient uptake by a wave-dominated reef flat community
Limnol. Oceanogr.
Cited by (21)
DeepGR4J: A deep learning hybridization approach for conceptual rainfall-runoff modelling
2023, Environmental Modelling and SoftwareCyclone trajectory and intensity prediction with uncertainty quantification using variational recurrent neural networks
2023, Environmental Modelling and SoftwareBayesian neuroevolution using distributed swarm optimization and tempered MCMC[Formula presented]
2022, Applied Soft ComputingCitation Excerpt :We also note that gradient-descent suffers from local minima problems and is not applicable in models where gradient information is not readily available. This is often the case in Earth science applications such as reef modelling and landscape evolution models [52,21]. Additionally, there is a strong need for uncertainty quantification in models for decision making in medical science [53], such as in the case of detecting heart disease in patients.
Distributed Bayesian optimisation framework for deep neuroevolution
2022, NeurocomputingCitation Excerpt :We used simple to deep neural networks to demonstrate the challenge in using computationally expensive models. Our proposed Bayesian optimisation framework can be used for a wide range of models across different domains which includes other models, such as Earth and geoscientific models landscape [113] and modelling reef evolution [114]. We note that Langevin-gradients are limited to models where information about the gradients is available, but when it is not available such as in geoscientific models, PSO can be beneficial.
Precipitation reconstruction from climate-sensitive lithologies using Bayesian machine learning
2021, Environmental Modelling and SoftwareCitation Excerpt :As a result this distribution is usually approximated in two ways; 1.) sampling methods such as Markov chain Monte Carlo (MCMC) (Hastings, 1970; Geman and Geman, 1984; Gelfand and Smith, 1990; Chandra et al., 2019; Scalzo et al., 2019; Pall et al., 2020), and 2.) variational methods which provides more computationally efficient approximations (Jordan et al., 1999; Wainwright and Jordan, 2008).
Bayesian geological and geophysical data fusion for the construction and uncertainty quantification of 3D geological models
2021, Geoscience FrontiersCitation Excerpt :Bayesian inference provides a suitable framework for doing this by using Markov chain Monte Carlo (MCMC) sampling methods for estimation and uncertainty quantification of free parameters. Previous studies using Bayesian inference in the geosciences have primarily focused on (i) geophysical joint inversions (Bosch et al., 2006; Shen et al., 2013; Giraud et al., 2017), (ii) fluid flow through permeable reservoirs for groundwater, hydrocarbon or carbon dioxide storage applications (Ye et al., 2010; Refsgaard et al., 2012; Seifert et al., 2012; Oladyshkin et al., 2013), (iii) geomorphologic and climate evolution (Hapke and Plant, 2010; Chandra et al., 2019; Pall et al., 2020), and (iv) fusion of structural geology data with geophysical datasets (Jessell et al., 2010, 2014; Grose et al., 2018; Pakyuz-Charrier et al., 2018b; Wellmann et al., 2018). Despite a clear need for Bayesian fusion of solid Earth geological and geophysical datasets (Jessell et al., 2010, 2014, 2018), there is still relatively little work in developing Bayesian-inferred 3D geological models, particularly detailed models at the local and camp-scale in a mineral exploration context.