Our Research in Focus 1 – Hydrological modelling in data-sparse areas

Hydrological modelling using ensemble satellite rainfall estimates in a sparsely gauged river basin: The need for whole-ensemble calibration

Chris Skinner, Tim Bellerby, Helen Greatrex, and David Grimes.

Published in Journal of Hydrology, Volume 522, March 2015, 110-122

Read here (paywall) or free, non-formatted version here

Summary

This research highlighted some of the issues with using information from satellites to estimate how much rain has fallen over an area. Many satellite rainfall estimation methods cannot directly observe the rainfall itself, instead inferring the amount from other information, in this case the temperature of clouds – clouds cool whilst raining so if they are below a certain temperature it is likely they are raining. This method was pioneered by University of Reading’s TAMSAT team.

A map showing satellite derived rainfall estimates for Africa
Example of rainfall estimates derived from satellite information using the TAMSAT cold cloud duration method. Image from the TAMSAT website.

The relationship between cloud temperature and rainfall is not perfect and changes with distance and time, causing the rainfall estimates to be wrong in places. As we increase the detail we want to use the chances of it being wrong increase so to overcome this we can use lots and lots of repeat estimates – each using different relationships between cloud temperature and rain in different places and at different times – to create a set of results, each unique but with an equal likelihood of being correct.

Map of Senegal Basin area from the paper. Black dots show the location of the sparse network of rain gauges available for the research.

This set of estimates can be used to run different types of models – for example, to run crop yield models to help us predict how well crops will have grown. For this research they were used to run a hydrological model to predict how much water will have flowed through a river basin. These models rely on us tuning values in them so that the outputs closely match observed values, a process called calibration, but when you have a set of different rainfall estimates for the same period which do you use to perform the calibration? The method that produced the best results was to use all of them together.

Why does it matter?

In the UK, our rainfall is monitored by a long-established and well-funded network of rain gauges and rain radar, providing us with a high detailed view of rainfall across the whole country in near real-time. This wealth of data helps us operate detailed forecasting systems to predict rainfall in the future and inform warning systems for drought and flooding. However, in many parts of the globe these networks to observe rainfall do not exist.

Across sub-Saharan Africa, including in the Senegal River Basin that provided the case study for this work, the lack of ground instrumentation is particularly acute. To fill the gaps in the data satellite information is used to estimate rainfall but this is not as accurate or as detailed as networks of ground instruments. These estimates are used to determine drought and flood warnings and by insurance companies that pay out to farmers who have lost crops, and consequently income, due to a lack of rain. It is important we understand fully what satellite estimates are telling us so we can make the correct decisions.

Why should you read it?

If you are interested in learning more about rainfall estimation is data-sparse regions and/or hydrological modelling using uncertain probabilistic input data.

New @EEIatHull Research – The impact of different rainfall products on landscape modelling simulations

Research led by Energy and Environment Institute Research Fellow, Chris Skinner, has recently been published in the journal Earth Surface Processes and Landforms. The research, featuring an international team from Hull, Bristol, and Zurich, showed how different methods of measuring rainfall can lead to different predictions of landscape change in computer models.

Rainfall is a slippery thing. For those of us in the UK, we are very much aware that it can be raining one minute and then sunny the next, or it can be raining over your house but not over your friend’s house just a few streets away. This makes it a difficult thing to measure accurately and consequently meteorologists use several methods to try and do so.

The simplest way is to use a rain gauge. There are many different designs but, essentially, they are mostly all glorified buckets that fill with water as it rains, although some do take different approaches. They generally give us a good idea of how much rain has fallen at that spot between the times readings are taken. By automating the readings we can get a good idea of how the rainfall rate has changed over time.

Rain gauges cannot tell us how much rain has fallen outside the bucket. This is generally ok if you are still close to the bucket but the further away you get, the more of a problem this becomes. By using lots of rain gauges we can get a better idea of how the rainfall is varying across an area and we can use geostatistics to try and fill the gaps. However, the results will be different depending on the geostatistical method you decide to use.

Weather radar on the other hand is able to tell us the relative intensity of rainfall over an area. The radar sends out signals that are bounced back to it by rain and depending on the timing and strength of that signal we can tell where it is raining, and the areas it is heaviest. It does not directly measure the rain though and it needs calibrating against a reference point. This calibration may be less accurate as you move away from the reference point or if conditions change over time.

The consequence of this is the availability of different methods to measure rainfall, the results of which we call products. Each product will be different in its estimation of where, when, and how much rain has fallen and with many computer models of rivers relying on a measurement of rainfall as an input, the choice of product can have a big influence on the results from the model.

Landscape evolution models (LEMs) are designed to model changes to the Earth’s surface, usually over large areas and long time periods (at least one hundred years). Some of these models have become sophisticated and fast enough they can be used to explore more local and shorter-term changes. They need to use a rainfall product to run yet only rarely does a product exist that has a record long enough to cover the time scales simulated. Instead, we can use weather generators that take the characteristics of rain as recorded by a product to create long records of rainfall that are possible and likely based on the data.

Graphical abstract from Skinner et al (2020). The chart shows the changing pattern of erosion and deposition from the channel head to the catchment outlet after 1500 years of computer model simulation. The different colours represent the results using the different rainfall products. The records that produced the most change were based on longer records that contain heavier rainfall events.

In newly published led by EEI Research Fellow, Chris Skinner, a weather generator was used to produce long rainfall records based on different rainfall products, as well as a combination of information taken from each product. These synthetic records of rainfall were used to run a landscape evolution model for periods of 50 and 1500 years, finding that the patterns of erosion and deposition varied along a river depending on which product was initially used.

Due to the relationship between river flows and the movement of sediment in rivers, something known as the geomorphic multiplier, a small increase in river flow can result in a large increase in the amount of material eroded and transported by the river. This makes models of erosion and deposition extremely sensitive to changes in rainfall and consequently, the initial choice of rainfall product used can have a big influence on the model results.

As these modelling approaches are increasingly used to help understand the impacts of climate change or to help predict flood risk, understanding how the choice of rainfall product can impact results is crucial and needs to be properly managed by modellers.

You can read the full article here.