Novel dynamical systems concepts for climate change prediction

What is it about?

This paper approached the problem of predicting climate under climate change in the annual to multi-decadal time scales. To that end, the authors introduced novel dynamical systems techniques and, using a conceptual but climate-like model, they investigated how climate projections can be affected by several sources of model-related uncertainty. Specifically, the authors combined the climate modelling idea of micro initial condition uncertainty (so-called micro-ICU) with the mathematical concept of pullback attractor to define a new pair of objects in the phase space, which they named as the "evolution set" and its associated "evolution distribution". These objects, although intuitive, represent the evolution of climate constrained to today’s best knowledge available. The authors explored how this evolution pair can be affected by two important sources of uncertainty: macro uncertainty in initial conditions, and uncertainty in anthropogenic forcing scenario. Additionally, the authors discussed the implications to the design and interpretation of ensembles in climate models.

Featured Image

Photo by NOAA on Unsplash

Photo by NOAA on Unsplash

Why is it important?

Predicting climate under climate change requires predicting the future under conditions never experienced before. The key tool used for this task are mathematical models of the planet Earth, which ranges from simple energy-balance models to the highly complex Earth System Models (ESMs). Essentially, ESMs are non-autonomous dynamical systems which possesses 4 key characteristics: they are 1) very complex (consisting of several highly complicated equations defined in an intricate spatial domain), 2) multi-component (e.g. they have an ocean, atmosphere, land, etc.), 3) multiscale (meaning that the modelled processes occur in all sorts of temporal and spatial scales), and 4) chaotic – the latter meaning that any computation within the model is highly sensitive to the finest details, making prediction a highly non-trivial task. A key question is therefore: how to predict the climate given all the above? Answering this question requires one to understand how a non-autonomous and chaotic dynamical system with these characteristics evolves over a finite number of years and decades. KEY TAKEWAY MESSAGE: In the temporal scales of interest for climate prediction, the projected climate is usually constrained within a object that is different from the system's attractor; the authors call this object the “evolution set”. Predicting climate therefore involves computing evolution sets and its associated distributions, rather than the system’s attractor. Furthermore, this evolution pair (set and distribution) is sensitive to various sources of uncertainty in initial condition, which has a practical consequence for initialising models (whether from spinup or from observations).

Perspectives