Introduction
Aquatic ecosystems play an important role in many aspects of our lives, including food security and tourism, and perhaps most importantly, they play a key role in regulating our climate. However, the high complexity of aquatic ecosystems means that creating simulations needed to make inferences about the impacts of anthropogenic threats on such systems is difficult. Furthermore, high uncertainties in underlying processes mean capturing such uncertainties in simulations is paramount for informing policy and creating aquatic digital twins.
We will improve how aquatic ecosystems are simulated by building a next-generation aquatic modelling framework that is highly composable and includes probabilistic components such that aquatic ecosystem models generated using our framework can represent uncertainties.
Explaining the science
Aquatic ecosystems are historically simulated using sets of ordinary differential equations (ODEs), with components of these systems representing organisms and their physiology, such as size, growth rate, or response to temperature. Parameter values used in such systems of ODE's are generally chosen based on laboratory experiments or through inverse modelling by fitting ODE's to observational data. Here, we will improve such models by implementing stochastic differential equations (SDE's), with noise terms representing uncertainties in underlying parameterizations. Since parameters often co-vary, for example, larger cell size also constitutes a higher nutrient requirement, components of the SDE will also be implemented to co-vary.
We will implement these systems of SDE's using Julia, exploiting the highly composable and interpretable nature this language offers without sacrificing the computational efficiency needed to derive large systems of SDE's on a global scale.
Project aims
Aim 1: Developing Agate.jl
Our first aim is to implement a framework to implement composable and efficient aquatic ecosystems that can capture uncertainty. This will be achieved through the development of Aquatic Gcm-Agnostic Tunable Ecosystems in Julia (Agate.jl).
The first phase of this package will be to port the global marine plankton model MITgcm-DARWIN from FORTRAN into Julia, and add stochastic components. However, implementing MITgcm-DARWIN will be only the first step; we will build Agate.jl such that it can support any system of aquatic ODE's such as simulations of fish, particles and macromolecular models.
Aim 2: Identify key processes driving uncertainties
While it is generally understood that predictions of aquatic ecosystems are uncertain, the scope and drivers of such uncertainties are poorly understood. Using Agate.jl, we will identify the key processes leading to model uncertainty and how such processes can be improved. Since simulations of aquatic ecosystems are limited by computational constraints, identifying and focusing on such key processes will allow for more efficient and realistic implementations of aquatic ecosystems while improving our fundamental understanding of ecological processes and systems.
Organisers
Ruth Drysdale
Programme Manager, Artificial Intelligence (AI)