Regridding and interpolation of climate data for impacts modelling – some cautionary notes

Users of climate data must often confront the problem that information is not available at the precise spatial locations of interest; or the related problem that multiple sources of information provide data at different collections of locations. An example of the first situation is the use of weather station data to calibrate a hydrological or land surface model requiring inputs on a regular grid; an example of the second is the use of information from an ensemble of climate models to sample structural uncertainty, but where each model produces output on its own grid. Dealing with this spatial misalignment is a common first step in any analysis, and is usually done by some form of interpolation. In this poster, we use standard approaches to convert regional climate model (RCM) outputs from the EuroCORDEX ensemble to the common grid used in the UK national Climate Projections (UKCP). We find that although these standard approaches perform acceptably in some situations, in others they can induce surprisingly large biases and inconsistencies in the statistical properties of the resulting fields – particularly those relating to variability and extremes. For example, although the resolutions of the UKCP grid and the EuroCORDEX RCMs are all similar, it is not hard to find locations where the maximum daily precipitation within a month is systematically underestimated by 5-10% in the regridded data; and where the maximum daily precipitation over a 20-year period is systematically underestimated by 25%. These effects could have major implications for impacts studies carried out using interpolated or regridded data, if they are not recognised and dealt with appropriately. We offer some suggestions, varying in ease of implementation, for dealing with the problem.