Matt Amos

Maths and Stats. Dept.

Lancaster University

I’m an environmental data-scientist working on merging physical and statistical models together to generate better predictions of our ever changing world. My particular focus at the moment is on not only making better predictions of environmental systems, but also ensuring that we are estimating the uncertainty properly.

Currently I work on the DSNE (data science for the natural environment) project, based in the maths and stats department at Lancaster University, but am actively looking for either academic or industry research jobs. As long as there are interesting and environmentally important problems to solve, I’m interested!

selected publications

AppliedML

Identifying latent climate signals using sparse hierarchical Gaussian processes

Amos*, M., Pinder*, T., and Young, P. J.

Accepted at NeurIPS Workshops Tackling Climate Change with Machine Learning, and Gaussian Processes, Spatiotemporal Modeling, and Decision-making Systems 2022

HTML PDF Code
AppliedML

A continuous vertically resolved ozone dataset from the fusion of chemistry climate models with observations using a Bayesian neural network

Amos, M., Sengupta, U., Young, P., and Hosking, J S

2023

Abs HTML PDF Code

Continuous historic datasets of vertically resolved stratospheric ozone, support the case for ozone recovery, are necessary for the running of offline models and increase understanding of the impacts of ozone on the wider atmospheric system. Vertically resolved ozone datasets are typically constructed from multiple satellite, sonde and ground-based measurements that do not provide continuous coverage. As a result, several methods have been used to infill these gaps, most commonly relying on regression against observed time series. However, these existing methods either provide low accuracy infilling especially over polar regions, unphysical extrapolation, or an incomplete estimation of uncertainty. To address these methodological shortcomings we used and further developed an infilling framework that fuses observations with output from an ensemble of chemistry-climate models within a Bayesian neural network. We used this deep learning framework to produce a continuous record of vertically resolved ozone with uncertainty estimates. Under rigorous testing the infilling framework extrapolated and interpolated skillfully and maintained realistic interannual variability due to the inclusion of physically and chemically realistic models. This framework and the ozone dataset it produced, enables a more thorough investigation of vertically resolved trends throughout the atmosphere.
ML

Ensembling geophysical models with Bayesian neural networks

Sengupta*, U., Amos*, M., Hosking, J. S., Rasmussen, C. and 2 more authors

Advances in Neural Information Processing Systems 2020

Abs HTML PDF Code

Ensembles of geophysical models improve projection accuracy and express uncertainties. We develop a novel data-driven ensembling strategy for combining geophysical models using Bayesian Neural Networks, which infers spatiotemporally varying model weights and bias while accounting for heteroscedastic uncertainties in the observations. This produces more accurate and uncertainty-aware projections without sacrificing interpretability. Applied to the prediction of total column ozone from an ensemble of 15 chemistry-climate models, we find that the Bayesian neural network ensemble (BayNNE) outperforms existing ensembling methods, achieving a 49.4% reduction in RMSE for temporal extrapolation, and a 67.4% reduction in RMSE for polar data voids, compared to a weighted mean. Uncertainty is also well-characterized, with 90.6% of the data points in our extrapolation validation dataset lying within 2 standard deviations and 98.5% within 3 standard deviations.
Thesis

Data-science techniques to improve the robustness, accuracy, and utility of chemistry-climate model ensembles

Amos, M.

2022

Abs HTML PDF

Ensembles of chemistry-climate models (CCMs) are fundamental for the exploration of the chemistry-climate system. A particular focus of chemistry-climate modelling is stratospheric ozone, whose concentrations have been decreased by anthropogenic releases of ozone depleting substances. In conjunction with observational data, CCM ensembles have been relied upon to simulate historic effects of ozone depletion and to project future ozone recovery. However, many widely used ensemble analysis methods are simplistic and are based upon incorrect assumptions about the design of the ensemble. Multi-model means used to construct future ozone projections do not account for variable model performance or similarity and therefore give biased and inaccurate projections. Similarly, simplistic linear regression methods used to infill historic ozone records underestimate interannual variability and are inaccurate in regions of sparse data coverage. Moreover, given advances in machine learning and data science and their increased use in environmental science, it is timely to apply more advanced tools to CCM ensembles. To address this methodological deficit, this thesis presents a set of novel tools to improve the predictions and projections from CCM ensembles of stratospheric ozone. A process-based weighted mean is developed which accounts for model performance and similarity in CCM ensembles. This improvement over pre-existing methods was used to generate accurate ozone hole recovery projections. This thesis also developed a Bayesian neural network (BNN) which fuses together CCMs with observational data to produce accurate and uncertainty-aware predictions. The BNN framework was used to produce historic continuous datasets of total ozone column and vertically resolved ozone, and represents a significant improvement in methods used to ensemble models. Though designed for CCM ensembles these flexible tools have the potential to be applied to other environmental modelling disciplines to improve the accuracy of projections, better understand uncertainty and to make better use of historic observations.