EXPLORING MACHINE LEARNING AND COMPUTATIONAL TECHNIQUES TO INFORM RAPID ADAPTIVE CLIMATE CHANGE (RACC)

The aim of the project is to explore the modelling of smaller datasets, using machine learning, to produce useful statistical forecasts and subsequently produce effective recommendations about climate and other timeseries datasets where the dataset is lacking richness or quality, for example where the climate is changing more rapidly than represented in the historical data. Building on the ideas in (Kenny, 2016), (Kenny, 2020) and (Ismail et al., 2015) using a fractal-based system, after (Kunze et al., 2015) , (Garralda-Guillem et al., 2020) to derive the mathematical attractor for a specific dataset and use that to "measure" the distance between the model and the subsequently observed data. Based on research by (Sévellec & Drijfhout, 2018) using an iterative approach, one could develop a series of models which are not based on big data models but would in the case of climate data enable predictions to be made over the 5-to-10-year period, but more importantly to keep this cycle iterating so that the model can be more responsive and adaptive to changes in the climate.

The goal of the PhD is to use and develop machine learning techniques which enable the identification and utilisation of patterns within climate data given the relatively rapid changes currently being experienced, and the context of the previous dependents on historical data to develop models. Historical data is no longer as useful as it used to be because of the rapidly changing climate. Although we can still learn from GCM models and longer-term hindcast modelling. There is an opportunity for a statistical-based approach to be developed.

This PhD studentship is part of a wider project to model the climate in a way which allows iterative model development. Data could the iterated comparatively between models and a meaningful comparison made from which forecasts can be drawn. The purpose of this is to produce an iterative model which continues to inform the changing climate by means of the difference between the current model and the observed data within the specified period. By doing this we anticipate achieving a model which can produce forecasts without relying on large datasets collected over a longer period. The innovative Rapidly Adaptive Climate Change (RACC) model is expected to be particularly useful given the increasing rapidity of climate change. As an initial step towards this goal, we intend to build a model which integrates the atmospheric data with the hydrospheric data with the intent of allowing the model to derive its own relationship between the heat cycle within the hydrosphere[1] and the effect on the global mean atmospheric temperature. The RACC approach of building models at different layers preserves the abstraction needed to keep individual datasets distinct whilst at the same time relating them in a way which could be brought together to produce forecasts.

Skills required

• A good understanding of computational and algorithmic techniques, related to machine learning.
• Strong coding skills - preferably including Java and Python
• A sound foundation in using statistics in code.

Background reading

Garralda-Guillem, A.I., Kunze, H., Torre, D.L. & Galán, M.R. (2020) 'Using the generalized collage theorem for estimating unknown parameters in perturbed mixed variational equations', Communications in Nonlinear Science and Numerical Simulation, vol. 91, p. 105433, Available: https://doi.org/10.1016/j.cnsns.2020.105433.

Ismail, D.K.B., Lazure, P. & Puillat, I. (2015) 'Advanced Spectral Analysis and Cross Correlation Based on the Empirical Mode Decomposition: Application to the Environmental Time Series', IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, vol. 12, no. 9, pp. 1968-1971.

Kenny, I. (2016) PhD thesis The Open University., Available: http://oro.open.ac.uk/47999/.

Kenny, I. (2020) 'Hydrographical Flow Modelling of the River Severn Using Particle Swarm Optimization', The Computer Journal, vol. 63, no. 11, Nov., pp. 1713-1736.

Kunze, H., Torre, D.L., Levere, K. & Galán, M.R. (2015) 'Inverse Problems via the (Generalized Collage Theorem) for Vector-Valued Lax-Milgram-Based Variational Problems', Mathematical Problems in Engineering, Available: https://doi.org/10.1155/2015/764643.

Sévellec, F. & Drijfhout, S.S. (2018) 'A novel probabilistic forecast system predicting anomalously warm 2018-2022 reinforcing the long-term global warming trend ', Nature Communications, vol. 9, August.