Research

We study different aspects of the predictability of weather and climate as complex physical systems.

Model Uncertainty

Optimal representing model uncertainty is a key challenge in climate modelling. Explicit representation of sub-gridscale uncertainties has been shown to provide a wealth of benefits to forecasts; reducing model biases, improving skill and fundamentally giving better estimates of forecast uncertainty along with the forecast. We study the impact of different methods of model uncertainty representation, particularly stochastic physics, and are in the process of developing new methods.

Representing land surface uncertainty in monthly to seasonal forecasts

In collaboration with resesarchers at ECMWF and Reading University, we are applying different methods of uncertainty representation to the ECMWF seasonal forecast system (System 4). Methods include stochastic tendencies, static and stochastically perturbed parameters.

People involved:

Dave MacLeod, Tim Palmer, Antje Weiseheimer

Relevant publication:

MacLeod et al (2016) QJRMS

Investigating impacts of stochastic physics on model climate

We are examining the impact of including ECMWF stochastic physics schemes on aspects of model climate including tropical variability, ENSO, the MJO and extratropical weather regimes.

People involved:

Hannah Christensen, Peter Watson, Aneesh Subramanian, Kristian Strømmen, Antje Weisheimer, Tim Palmer

Relevant publication:

Weisheimer et al., 2014, Phil. Trans. R. Soc. A

Constraining stochastic parameterisations

We work on constraining the structure and parameters of stochastic schemes using weather forecasts, coarse-graining using cloud-resolving models and simple dynamical systems such as Lorenz '96.

People involved:

Hannah Christensen, Peter Watson, Aneesh Subramanian, Antje Weisheimer, Tim Palmer, Andrew Dawson

Developing new stochastic parametrisation schemes for the atmosphere

We work on developing new representations of model uncertainty for atmospheric parametrisation schemes, including perturbed parameter and stochastic approaches

People involved:

Hannah Christensen, Aneesh Subramanian, Peter Watson, Tim Palmer

Comparing different representations of model uncertainty

Our work compares the performance of different representations of model uncertainty in weather, seasonal and climate forecasts using simple dynamical systems and comprehensive global circulation models

People involved:

Hannah Christensen, Aneesh Subramanian, Peter Watson, Tim Palmer

Stochastic Multiscale Modeling for Atmospheric Convection

In collaboration with researchers at ECMWF and SUNY Stony Brook, we are developing a stochastic super-parameterisation scheme for atmospheric convection in the ECMWF IFS for medium and long range prediction.

People involved:

Aneesh Subramanian, Antje Weisheimer, Tim Palmer

Stochastic Dynamical Cores for GCMs

We are developing and testing a stochastic dynamical core for GCMs.

People involved:

Aneesh Subramanian, Stephan Juricke, Antje Weisheimer, Tim Palmer

Representing ocean model uncertainty on timescales from seasons to decades

We are developing and testing stochastic ocean parametrizations to incorporate measures of subgrid scale variability and uncertainty in global ocean models, for a varity of timescales and in coupled as well as uncoupled systems.

People involved:

Stephan Juricke, Tim Palmer, Antje Weisheimer, Laure Zanna

Representing sea ice model uncertainty on timescales from seasons to decades

We are developing and testing stochastic sea ice parametrizations to incorporate measures of subgrid scale variability and uncertainty in global sea ice-ocean models, for a varity of timescales and in coupled as well as uncoupled systems.

People involved:

Stephan Juricke, Tim Palmer

Inexact computing

We study the use of inexact hardware in numerical weather and climate models. Inexact hardware is promising a reduction of computational cost and power consumption of supercomputers and could be a shortcut to higher resolution forecasts with higher forecast accuracy. Please find a presentation that provides a summary of our work on inexact computing here.

An imprecise land surface model

Using the ECMWF land surface model HTESSEL, we are exploring the impact of reduced precision arithmetic on simulation of land surface processes.

People involved:

Dave MacLeod , Andrew Dawson, Tim Palmer

Inexact computing on multiple spatial scales

Inexact techniques might make our models more efficient, but not if they are applied too strongly or in the wrong places. We emulate these techniques in multiscale models to investigate whether using different, optimal levels of 'inexactness' on different spatial scales can improve overall forecast accuracy. We are also developing and testing a multi-scale atmospheric model with reduced precision superparameterisation and double/single precision for atmospheric dynamics for modelling atmospheric convection.

People involved:

Tobias Thornes , Peter Dueben, Aneesh Subramanian, Andrew Dawson, Tim Palmer

Relevant publication:

Dueben and Palmer, 2014, Monthly Weather Review

Field-Programmable Gate Arrays (FPGAs) in Earth System modelling.

Field Programmable Gate Arrays promise a significant increase in computational performance for simulations in geophysical fluid dynamics compared with CPUs of similar power consumption and allow to adjust the precision of individual floating point numbers to specific application needs. Together with collaborators at Imperial College and Maxeler technologies, we have shown significant performance and energy efficiency gains by using FPGA's to implement low and medium complexity climate models.

People involved:

Stephen Jeffress, Peter Dueben, Tim Palmer

Relevant publication:

Dueben et al., 2015, JAMES

Pruned and bit-width truncated chips in Earth System modelling.

In close collaboration with hardware developers, we investigate the use of new hardware setups that allow to trade numerical precision against an increase in performance and a reduction in power consumption.

People involved:

Peter Dueben, Tim Palmer

Relevant publication:

Dueben et al., 2014, Phil. Trans. R. Soc. A

Interactions between rounding errors and model uncertainty.

We investigate the use of rounding errors to represent sub-grid-scale variability and compare rounding error patterns with random forcings of stochastic parametrisation schemes.

People involved:

Peter Dueben, Tim Palmer

Relevant publication:

Dueben and Dolaptchiev, 2015, Theor. Comput. Fluid Dyn.

Reduced numerical precision in atmosphere models.

We study the use of reduced precision hardware in three-dimensional models of the atmosphere. We started with tests in the spectral dynamical core IGCM and proceed now to the OpenIFS model. We also studied the use of single precision in the Integrated Forecast System of ECMWF. The use of single precision allows a significant cost reduction with almost no impact on model results.

People involved:

Peter Dueben, Tim Palmer

Relevant publication:

Dueben and Palmer, 2014, Monthly Weather Review

Reduced precision hardware in data-assimilation.

We study how rounding errors will influence ensemble data assimilation. We will also investigate if data storage can be improved using the optimal level of numerical precision.

People involved:

Sam Hatfield, Peter Dueben, Aneesh Subramanian, Tim Palmer

Predictability

We study the predictability of the climate system on multiple timescales. We have a particular focus on the seasonal scale, looking at the skill of retrospective hindcasts across multiple domains and their potential use for societally beneficial applications.

Impact of parameter uncertainty on soil moisture memory

Soil moisture memory is a key component of seasonal climate predicatability. We are investigating the impact of uncertainty in hydraulic parameters on the estimation of memory.

People involved:

Dave MacLeod, Antje Weisheimer

Relevant publication:

MacLeod et al (2016), HESSD (discussion paper)

The use of subseasonal-to-seasonal forecasts in wind energy applications

In collaboration with researchers at the Barcelona Supercomputing Center, we are interested in the best way to transform monthly to seasonal forecasts to useful predictions for the wind energy industry

People involved:

Dave MacLeod

Relevant publication:

MacLeod et al (2016), submitted to Weather, Climate & Society, available on request

Skill of NAO seasonal predictions in century-length seasonal hindcasts

Predictability of the North Atlantic Oscillation (NAO) is a key component in prediction of European seasonal climate. Using System 4, we have created a seasonal hindcast dataset from 1900 to present, and are using this to look at how skillful seasonal forecasts are, and the decadal variability in this skill.

People involved:

Nathalie Schaller, Dave MacLeod, Christopher O'Reilly, Tim Palmer, Antje Weisheimer

Relevant publication:

Schaller et al (2016), in preparation, available on request

The use of seasonal forecasts for climate services

We have a longstanding interest in the skill of seasonal climate forecasts and their application to provide societal benefit. Members of the group have been working with water companies, meteorological forecasting centers and humanitarian organisations to investigate the institutional barriers preventing forecast use. We also carry out extensive verification of forecasts; some recent areas of focus include heatwaves in Europe, monthly forecasting for UK water companies, and seasonal drought forecasting for the Middle East.

People involved:

Dave MacLeod, Ana Lopez, Tim Palmer, Antje Weisheimer

Relevant publication:

MacLeod et al, (2016) QJRMS

Forecast skill and model fidelity in tropical variability in 30 year seasonal hindcasts

We investigate the representation of tropical climate variability and the impact of different scales of stochastic forcing on these in 30 year seasonal hindcasts.

People involved:

Aneesh Subramanian, Tim Palmer, Antje Weisheimer

Impacts of stochastic parametrizations on sea ice and ocean predictability

We investigate the impact of stochastic schemes in sea ice and ocean models on potential predictability for seasonal and decadal forecasts.

People involved:

Stephan Juricke, Tim Palmer, Antje Weisheimer, Laure Zanna

FileSize
151201_Hannah_small.pdf5.27 MB
peter_dueben_homepage.pdf6.42 MB