The GCRF-AFRICAP team boasts ten postdoctoral research associates including climate expert Dr Sarah Chapman at the University of Leeds. Sarah presented on her work at the programme’s inaugural knowledge-sharing seminar on 30 January 2019. We wanted to know more so sent along our non-expert interviewer for some climate modelling 101…
Thanks for taking the time to talk more about your work, Sarah. I’ll jump straight in: climate change and resilience are a big component of the GCRF-AFRICAP project. Which aspect of this does your research examine?
We are using climate models to understand future climate in the partner countries (Malawi, South Africa, Tanzania and Zambia) and what this means for agriculture. We want to know what the most important aspects of climate are for agriculture in these countries, how climate will change in the future, and why.
Sounds like a big job! How will you go about this?
We are going to answer these questions using climate models – these are computer programmes that use equations to represent the processes driving our climate system. At first, we’ll be using two regional climate models. Regional models model a small region at a high resolution, rather than modelling the entire world like a global climate model. These regional models get their boundary conditions (the climate at the edge of the region being modelled) from a global model, also termed a ‘driving model’. They also inherit biases from the driving model, though as the domain of the regional models we are using is quite large, they are also free to develop their own circulation (1, 2). What I mean by this is that the climate in a regional model is determined by the boundary conditions from the driving model and the internal dynamics of the regional model. With a large domain, the boundary conditions are less important relative to the regional model dynamics and the regional model climate can differ more from the driving model. In some cases, this means the climate change signal can differ in sign between the driving model and the regional model. This is sometimes the case with precipitation, particularly where land-atmosphere feedbacks are important and are captured by the regional model but not the global model.
Got it – global models drive the regional ones, which in turn can give us more detail about smaller areas. Can you tell us more about these regional models? What are their strengths and weaknesses?
The two regional models we are using are called the CP4-Africa and P25 models (3). Both models are driven by the Unified Model (which incidentally is also used by GCRF-AFRICAP partner the UK Met Office for weather forecasting). CP4-Africa has a 4.5 x 4.5km resolution and is a convective permitting model, which means it can explicitly resolve convective clouds. P25 has a 26 x 39km resolution and a parametrization scheme for convection, which means parameters are used to estimate the value of sub-grid scale processes, rather than explicitly calculating what is happening at a sub-grid scale. This works well for estimating area-average rainfall but doesn’t work as well for extremes or looking at the distribution of rainfall (4). Because CP4-Africa can explicitly resolve convection, we expect it to do better with rainfall than P25, particularly for extreme rainfall.
Interesting stuff and shows how different models come into their own at different scales. And with tools like this in place, I guess it’s all systems go?
Not quite – before we can use these models to look at future climate, we need to know how well they represent current climate. To do this we need to compare the model outputs with actual observations. Validating climate models over East Africa is difficult due to the sparse network of rain gauges and weather stations. Instead of using weather stations, we used satellite products, which provide gridded estimates of rainfall over a large portion of the world. We need to be careful when using satellite data to validate climate models however, as the satellite-based estimates of rainfall aren’t perfect. There are many products available, and they all differ from each other, particularly when it comes to extremes (5,6).
I see. These products sound like a good alternative in the absence of the other observation tools you mention. How do they work?
Satellite rainfall products estimate rainfall from measurements of infrared and passive microwave (some products only use infrared). Infrared measures cloud top temperature and assumes colder clouds mean a higher cloud top and more rainfall. Passive microwave measures scattering from ice particles in clouds. Satellite-based rainfall products tend to under-estimate rainfall over mountains, particularly products that use infrared measurements only. In East Africa, there is also a tendency to under-estimate heavy precipitation. This is region-dependent, and in Asia the tendency is towards over-estimating heavy precipitation. We used the TRMM 3B42 daily product, which performs well in evaluations compared to other products, though it still has all these problems (7,8). This makes validating the climate model difficult, as we don’t know for sure what the actual rainfall is.
So even this is not an exact science…How did CP4-Africa and P25 perform?
For most of southern Africa the models were doing quite well for several metrics relevant to agriculture; onset of the rainy season, and indices of extremes such as consecutive wet and dry days and intense precipitation. CP4-Africa was doing much better when it came to extremes but wasn’t improving much on onset of the rainy season. While the models were doing well across most of the study area, they weren’t getting the boundary of the unimodal and bimodal rainfall areas in Tanzania quite right.
Good to hear that CP4-Africa is improving, but there are obviously still gaps in both models. Do you have any way to address this?
Yes – we decided to use the CORDEX ensemble of regional climate models. CORDEX is a coordinated regional downscaling experiment. Essentially, several of the global climate models that participated in the CMIP5 experiments (Coupled Model Intercomparison Project) were downscaled using several different regional models. Different global models will have different biases – some may represent parts of Africa better than the Unified Model, which drives CP4-Africa and P25, some may represent parts less effectively. Using this will allow us to look at a broad range of possible futures for East Africa, and the implications of different futures for agriculture.
Exciting stuff, which feeds into GCRF-AFRICAP’s wider work on building a picture of present-day agriculture systems in the focal countries and evaluating the pathways countries should follow to achieve climate-smart development.
Thanks, Sarah, for taking the time to explain more about your work to this layperson. Any of our readers who would like further information on our climate modelling work, can get in touch through our contact form.