New stochastic representations for the uncertainties in operational numerical weather prediction

Uncertainty in NWP models can be treated by randomly perturbing the parameterization calculations. However, the random component of a fully viable and properly defensible approach cannot be chosen arbitrarily but should be based on theory, observations and/or supporting simulations. For example, the uncertainties should depend in systematic ways on the current state of the atmosphere, and should scale appropriately and predictably with model resolution.

With a focus on atmospheric convection and related processes, this project will identify and quantify the physical deficiencies in the current methods for treating uncertainty in operational NWP and will develop new methods designed to eliminate or ameliorate those deficiencies.

Key questions to be addressed are:

1. What are the key dependencies exhibited by uncertainties as portrayed in the current ECMWF scheme, and how do these dependencies compare with the true uncertainties as estimated from ensembles of superparameterizations?
2. What physical relationships underly the true uncertainties, and how and why do the uncertainties vary around the globe?
3. Are the physical relationships better captured using the new perturbed-parameters scheme under development at ECMWF, which aims for better links of uncertainty to the prevailing physics? In making these assessments, we intend to focus particularly on atmospheric states that are close to transitions (e.g. from shallow to deep convection), which are important aspects of overall model uncertainty but which may not be well handled at present.
4. What physical ingredients are most important to include in new methods so that they can capture the most important dependencies on a firm physical basis?

Based on these results, we will identify and test new physically-motivated techniques and assess their behaviour in the ECMWF NWP model. The end goal of the project will be to have constructed proofs-of-concept for new methods that are capable of producing the right uncertainty for the right reasons, and which show promise for improving ensemble weather forecasting. In this way, the project will contribute to improved NWP ensemble methods and improved representations of uncertainty for future weather and climate models.

TRAINING:
We will expect and strongly encourage the student to take full advantage of the training opportunities available at the University of Reading, guiding them in an appropriate selection of the skills training courses. We will also advise the student on an appropriate selection of formally-taught modules at the Department of Meteorology, depending on their previous background and experience. The modules on numerical modelling will be particularly valuable for this project.

A range of other opportunities will also be available through our links with ECWMF. For example, ECMWF organize and host week-long training programmes in various aspects of numerical weather prediction, and these are naturally well-aligned with their modelling and forecasting practices. Details are available at https://www.ecmwf.int/en/learning/training Specifically we expect the training on “Parametrization of subgrid physical processes” and “Predictability and ensemble forecast systems” to be extremely valuable to support this project, and we intend that the student should attend one of these classes in year 1 and the other in year 2.

The supervisors will provide training in the use of the Integrated Forecasting System (IFS), the ensemble NWP system that is used for both operational and research purposes. By means of a “Special project” the student will be able to use the IFS on the ECMWF’s high-performance computing facilities. Some smaller tasks can alternatively be performed using OpenIFS (the portable, externally-available form of the IFS) on the Reading system. Training on the use of IFS/OpenIFS will be supplemented by attendance at the OpenIFS Users’ workshop. See https://www.ecmwf.int/en/about/media-centre/news/2017/openifs-users-explore-seasonal-predictability
for a description of the 2017 workshop. Supervisor Plant has successfully pioneered the use of the IFS at Reading (https://www.ecmwf.int/en/newsletter/152/news/openifs-used-university-reading-students) and is in discussion with ECMWF about hosting of the next OpenIFS workshop, which is provisionally scheduled for Reading University in June 2019. This would provide an excellent opportunity for the student, and would be perfect timing for this project.

To hear more about this project please follow the link: https://www.youtube.com/watch?v=U44-kRvfrLM&list=PLZWYaq_mWwsEM5dH1abHjYIgU2EVaegT9&index=3

To read more about this project please follow the link: http://www.met.reading.ac.uk/nercdtp/home/available/desc/entry2018/SC201834.pdf