Session 4: 15:00 - 16:30

Chair: Ellie Fisher

Theme: Atmospheric data and modelling

Improving predictions of high-impact weather in East Africa

Eliza Karlowska

Mesoscale convective systems are the largest of convective storms, spanning a few hundred kilometres and often lasting several hours. They lead to extreme rainfall, flash flooding and hail, especially in the tropics where warm and moist air is conducive to the development of these storms. The Lake Victoria region is inhabited by over 40 million people and is a major source of food, water and economic activity in East Africa. Over 80% of extreme rainfall around Lake Victoria is produced by mesoscale convective systems. Here, we analyse conditions prior to the development of these storms and link the changes in the local environment to the large-scale climate variability in the tropics. Our analysis is used to inform a machine learning model that predicts the probability of a given storm type occurring in order to improve predictions of high-impact weather over the lake.

Global Carbon Models and the Palm Oil Conundrum: Improving Emissions Estimates for South East Asia

Ida Bagus Mandhara Brasika

Palm oil cultivation significantly contributes to land-use change and carbon emissions in Equatorial Asia, accounting for approximately 23% of deforestation in Indonesia between 2001 and 2016. Despite this, most Dynamic Global Vegetation Models (DGVMs) lack a specific Plant Functional Type (PFT) for palm oil, categorizing it under generic cropland, which introduces uncertainties in estimating carbon emissions from land-use change (ELUC). To address this, we incorporated a dedicated palm oil PFT into the DGVM JULES-ES and conducted simulations using MapBiomas Indonesia 2.0 and the Annual Oil Palm Dataset (AOPD). Results show that including palm oil increases ELUC estimates to 0.068 ± 0.011 PgC/year compared to 0.053 ± 0.011 PgC/year without it. Representing palm oil as generic cropland produces slightly higher estimates (0.070 ± 0.012 PgC/year), but explicit representation is vital for accuracy. We estimate palm oil accounts for 22.6% of total ELUC, or 0.36 PgC over the past two decades in Southeast Asia. These findings highlight the importance of including palm oil PFTs in DGVMs to improve regional and global carbon budget estimations.

Responsible computing: minimising the environmental impacts of research

Sadie Bartholomew

Atmospheric and aligned research often has a significant computational component, or is even fully computational (as opposed to lab-based). But use of computers, from local workstations and personal laptops to supercomputer clusters, requires energy and therefore from our research we can have a sizable contribution to greenhouse gas emissions, especially when the complexity and scale of geoscientific study means we tend to repeatedly run sophisticated models and churn through huge amounts of data. We can end up exacerbating the climate crisis we might be studying!

This talk promotes the responsibility to remain vigilant of the environmental impact of our research from our use of software (code), hardware (computers) and data (e.g. data centers or storage); outlines some ways to quantify the impact; and suggests basic approaches to consider to reduce the impact from computational processes that are essential to our research.

Parametric uncertainty in climate feedback representation 

Saloua Peatier

The uncertainty associated with climate model calibration is rarely represented in climate projections. Climate models produce high dimensional output across space, time and variable dimensions. Model calibration is often ad-hoc, and is subject to both data, time and computational constraints. Perturbed parameter ensembles (PPE) are ensembles of simulations for which the physical parameters have been perturbed according to a defined sampling. Using a PPE of the ARPEGE-Climat 6.3 model, the atmospheric component of the CNRM-CM6-1 model, we argue that there is a potential for comparably performing parameter configurations with a diverse range of  future climate evolution. 

In the ARPEGE-Climat PPE, 30 parameters of the model have been simultaneously perturbed, coming from the parametrizations of turbulence, convection, cloud microphysics and radiative properties. We propose a quasi-automatic optimization method for selecting a subset of optimal calibrations with a diversity of feedback parameter values. Using a statistical emulation/optimization framework, 15 parameter configurations are optimized, exhibiting climatological skill scores comparable to those of the CFMIP6 multi-model ensemble, while covering an estimated equilibrium climate sensitivity (ECS) interval of [4.1 − 6.1]. These results illustrate the impact of model tuning on climate sensitivity. This method provides a pathway for rapidly identifying high and low sensitivity model variants, which could allow for better quantification of uncertainty in model projections.

Quantifying the indirect atmospheric and climate effects of wetland restoration as nature-based climate solution

Becky Wagner

Wetland restoration has been proposed as a promising nature-based solution to climate change. However, wetlands are the largest natural source of methane (CH4), which is a potent greenhouse gas and precursor to air pollution. CH4 emissions are calculated for future wetland restoration scenarios under different climatic future. These emissions are then input into the UK Earth System Model (UKESM) to evaluate the impact of wetland restoration on atmospheric composition, radiative forcing and air quality. 

Climate trends in initialised hindcasts

Rhidian Thomas

I will present a method to quantify the range of multidecadal trends in an ensemble of seasonal re-forecasts (hindcasts). These models are routinely used for prediction on seasonal to decadal timescales, but remain underutilised for understanding longer term climate change and variability. Each run is launched with initial conditions that reflect the observed state of the atmosphere and ocean at that time. By combining knowledge of the observed state with high spatial resolution and large ensemble sizes, seasonal hindcasts offer the potential to bridge the gap between observed trends and those in freely evolving climate models.

Panel Discussion

An opportunity to ask questions of all the presenters, find out more and explore other ideas.