The theme of the 2022 Open Access week was Open for Climate Justice. PhD researchers Fiona Spuler and Jakob Wessel are working on the biases in climate models that lead to difficulties in relating the findings of global climate models to locally changing conditions and extreme events. Although these biases in no way invalidate the overall findings of climate change research, predictions of climate change, they may affect the impacts felt by local communities directly and hamper activities such as the implementation of flood defence measures.
In this blog post, Fiona and Jakob describe ibicus – a python open-source software package for the statistical bias correction of climate models. The package was developed in partnership with ECMWF and as part of the ECMWF Summer of Weather Code.
Although climate models are continuously improving in their representation of the atmosphere, land, and oceans — not least due to the contributions of various researchers here at the University of Reading — model biases persist. A model bias is defined as a systematic difference between the distribution of a simulated climate statistic compared to the observed climate statistic during the same time period. These biases, or unrealistic representations, in the climate model are, amongst other things, due to the fact that climate models have limited spatial resolution and there are some processes that occur at smaller spatial scales than the model can explicitly capture.
These biases don’t affect the broad findings of climate science regarding the large-scale impacts of anthropogenic climate change. However, they do become an issue when trying to relate the findings of global climate models to locally changing conditions and extreme events: even though slightly misrepresented maximum precipitation events might not matter too much for the overall picture of climate change, they will matter a lot to the community that lives in this region and is asking what type of flood defences to put in place.
An option to deal with these biases is statistical bias correction, which essentially means applying a correction function to the distribution of a meteorological variable such as precipitation or temperature. Although these empirical methods can reduce some biases, they cannot correct fundamental misspecifications of the climate model and are prone to misuse. Nevertheless, bias correction has become common practise and is applied prior to most climate impact studies. Bias correction, if applied, should therefore at least be evaluated with care. This is where ibicus comes in by enabling users compare and evaluate a range of different methodologies in a transparent and easy-to-access way. The package is published open-source and comes with extensive tutorials and documentation, therefore making it as easy as possible to use the best bias correction method for the location and problem at hand.
There already exist a variety of channels through which climate research and models are made publicly available – such as the IPCC, or the Copernicus Climate Data Store. However, this does not mean that communities – especially those at the frontlines of the climate crisis – have access to and can easily contribute to research on how climate change is going to impact their lives and livelihoods or develop plans on how they can reduce the impacts. This is due to various reasons – one of them being the unavailability of open-source software to relate the large-scale processes modelled on supercomputers around the world to local realities and impacts. Although ibicus in no way addresses this larger issue, it is with this bigger picture in mind that we developed it and will continue to work on it during our PhDs.
Fiona Spuler is a PhD student at the University of Reading at the Meteorology department in collaboration with ECMWF, working on hybrid statistical-dynamical models to improve seasonal forecasts. Jakob Wessel is a PhD student at the University of Exeter in collaboration with the Met office, working on statistical post-processing of weather forecasts with a particular focus on compound extremes.