Meteorology: Climate trends, coverage bias and rainfall (online)

Wednesday 26th May 2021

Location: Remotely by invite. See Royal Statistical Society main events page .

14:00-14:15 Login

14:15-14:30 Chair of the Leeds/Bradford Group

Introductions

14:30-15:10 Prof Kevin Cowtan, York Structural Biology Laboratory, University of York

This one weird trick from the 1970s could have avoided the false pause in global warming

Communicating complex statistical concepts to user communities can be challenging, when even supposedly simple metrics such as the mean of a dataset can show counterintuitive behaviours which lead to wrong conclusions. I will show how I attempted to communicate this issue to an audience of climate data users, along with some societal impacts of the misunderstanding.

15:10-15:50 Mr Donald Cummins, Department of Mathematics, University of Exeter

How reliable is detection and attribution of climate change trends?

Since the 1970s, climate scientists have developed statistical methods intended to formalize detection of changes in global climate, and attribute such changes to relevant causal factors, natural and anthropogenic. Detection and attribution of climate change trends is commonly performed using a variant of Klaus Hasselmann's "optimal fingerprinting" method, which involves a linear regression of historical climate observations on simulated output from numerical climate models. However, it is well known in time series analysis that regressions of non-stationary variables are in general inconsistent and liable to produce spurious results.

This study has shown, using an idealized linear-response-model framework, that if standard assumptions about the inputs to the numerical climate model hold, then the optimal fingerprinting estimator is consistent. In the case of global surface temperature, parameterizing abstract linear response models in terms of planetary energy balance and ocean heat storage provides this result with physical interpretability. Hypothesis tests have been conducted using global temperature output from 16 of the latest generation of numerical climate models, to assess whether the assumptions required for consistency hold in practice. Each of the 16 tests yielded strong evidence that the assumptions hold

15:50-16:30 Prof Douglas Parker, School of Earth and Environment & School of Mathematics, University of Leeds

Analysing and predicting African rainfall

Rainfall is probably the most important climatic variable for the African populations: reliable rain is vital for the success of crops, but rain events can be very intense, leading to flooding, high winds and lightning. Predicting rainfall can enable people to take action to improve their livelihoods and prevent or mitigate adverse impacts.

Unfortunately, the performance of global numerical models for African rainfall remains very poor, on timescales from days to decades, offering poor guidance to forecast users. Faced with such uncertainty from prediction systems, we need to represent rainfall differently. I will present some ideas from a wide range of prediction timescales.

“Nowcasting” (exploitation of near-real-time data) is a vital service for response to high-impact events. On longer timescales, understanding the relationship of complicated rainfall patterns with large-scale climatic drivers can help us to understand and predict seasonal rainfall and the nature of rainfall in Africa’s future climate.