Year: 2023
Primary Supervisor: Dr John Marsham <J.Marsham@leeds.ac.uk>
Institution: University of Leeds (Met Office joint chair) (School of Earth and Environment)
Academic Supervisors: Dr Christopher Taylor, Dr Cristina Charlton-Perez, Dr Massimo Bollasina (University of Edinburgh)
Research Themes: Atmosphere and Climate, Climate, Earth Observation, water, weather
Project Partners: Met Office
Research Keywords: Atmosphere, Climate, Earth Observation, Water
Supervisors: Dr John Marsham (Met Office), Dr Christopher Taylor (CEH), Dr Cristina Charlton-Perez (Met Office), Massimo Bollasina (University of Edinburgh)
This project will use Earth Observation to understand how the surface state controls the predictability of convection in Africa, and the implications for forecasting.
Scientific Background and Motivation
The need for improved risk-based precipitation forecasts for Africa is recognised nationally and internationally. Numerical Weather Prediction (NWP) skill for rainfall is, however, far lower in tropical/sub-tropical Africa than in the mid-latitudes. Earth Observation (EO) has provided critical evidence of how there are strong couplings between the land surface and rainfall in the Sahel, but much less is known about the rest of Africa, where the nature of the land surface and the convection both differ. This PhD will use EO to improve understanding of land-atmosphere couplings in Africa, to support improvements in risk-based forecasts from the next generation of weather models.
EO has also shown that standard weather and climate models that use parameterised convection fail to capture observed land-atmosphere feedbacks in Africa, but this is improved in high resolution convection-permitting models. The Met Office has started to provide convection-permitting simulations over Africa, and Leeds has a strong partnership with the Met Office in this area. Permitting convection improves skill, but research shows that ensemble forecasts, that are needed to quantify uncertainty, exhibit too little spread, limiting their value. Addressing such issues in ensemble spread is not only important for prediction, but also raises fundamental scientific questions around the processes controlling the predictability of convection in Africa.
In-situ observations of rainfall and land-surface properties are rare over Africa. There are now though a selection of remote sensing observations from polar orbiting satellites that provide high quality data for rainfall, soil moisture and vegetation. In addition, more frequent infrared and visible data from geostationary satellites have been shown to provide effective proxy measurements, especially when appropriately combined with higher-quality datasets.
Figure: An example of the role of soil moisture variations controlling convective rainfall from Taylor et al, 2022. Left: A mesoscale convective system (MCS) shown by IMERG rainfall accumulation (coloured, mm). Right: Preceding land surface temperature anomalies (coloured, K). The MCS was observed to follow warm land surface temperatures located between cooler areas that had been wet by rainfall over preceding days. Although a single case, this preference of mature Sahelian MCSs for warmer soils is established statistically.
Aims and Objectives
This PhD will use EO understand how land surface state (soil moisture, vegetation and wetlands/inland water) controls the predictability of convection, and the implications for NWP ensembles. Do models respond appropriately to the land surface? Can improved initialisation of the land surface across the ensemble improve skill and ensemble spread? We hypothesise that: (i) errors in the land surface initialisation lead to identifiable errors in predictions, (ii) accounting for land surface uncertainty can improve ensemble spread, (iii) models fail to capture the full extent of land-surface controls on convection, and we therefore expect that improving the land surface will bring greater benefits as such errors are addressed. Initially focusing on day+1 and day+2 timescales, the PhD will use EO to: (1) Understand how the land-surface state controls predictability of convective rainfall, and how uncertainty in the land surface generates uncertainty in rainfall. (2) For the next generation of ensemble convection-permitting NWP determine how improvements to land-surface initialisation from better use of EO data can help generate cost-effective and reliable probabilistic predictions. Extensions could link this to most likely to nowcasting predictability, but potentially to seasonal.
Methodology
The starting point will be to use EO to evaluate the first convection-permitting ensembles run for Africa during GCRF African SWIFT, focusing on West Africa where land-atmosphere interaction is strong. We will bring together satellite retrievals of soil moisture, inland water/wetlands and rainfall, with simulations to quantify how the land surface controls storms in observations versus models. A next step will be to examine how improved land-surface initialisation from EO can be used to improve ensemble spread and reliability. Throughout there will be opportunities to develop alternative statistical and perhaps AI/ML approaches, both as a tool for understanding, and as an alternative route for predictions, potentially building on work from CEH (UK) and KIT (Germany).
Further Reading
Klein and Taylor, 2020, https://doi.org/10.1073/pnas.2007998117
Vogel et al., 2020, https://doi.org/10.1175/WAF-D-20-0082.1,
Cafaro et al, 2019, https://doi.org/10.1002/qj.3531,
Taylor et al., 2022, DOI 10.1088/1748-9326/ac536d
