Each signal undergoes a validation process that incorporates a range of sanity checks and scientific validation
After reading this article, you will learn:
Sanity checks
Sanity checks ensure data from regional, hydrological, and global models combined with weather stations, remote sensing, and reanalysis products span a range which is physically possible. Such checks include:
1. Are there negative values in our data?This step checks the lower end of values in our datasets. Models can have very small negative numbers in some metrics, so this check is in place to ensure such instances are not carried through to our signals.
2. Are the maximum values sensible?This step checks the upper end of values in our datasets. For example, any winds greater than 150 m/s are deemed unphysical, and only data with wind values less than this maximum are carried through to our signals. Similarly, for our riverine signal, we take a global perspective to check that the global maximum discharge rates are in the Amazon basin.
3. Do the return levels increase with increasing return period?By definition, the return level calculated for each return period should increase with increasing return period - a 100-year event should be more severe than a 2-year event.
4. Does each data point (geographically and in time) make sense in the context of its neighboring points?Few of our signals are discrete phenomena; for example, one would not expect to have a grid location with a very high wind speed value surrounded by points with very small values. Where such examples are found, the neighboring points are used to smooth over the high value point.
Scientific validation
Scientific validation subsequently assesses these data against published scientific literature, where it is available. The atmospheric metrics (temperature, precipitation, and wind) are validated against the IPCC Atlas to ensure results from our Multiple Futures Model agree with the full CMIP6 model suite.
The scientific validation process is comparatively easier for some signals than for others. For example, there is a wealth of published information on regional temperature trends, whereas similar information for riverine flooding is not as readily available. In the latter example, alternative comparisons are made with historical data that are not already used to generate the metrics. For example, independent discharge data and flood maps from the UK and the US are compared with our metrics in the historical period (1980-2020) for scientific validation of our riverine signal.