Global IDF curves created from local observations using machine learning

Intensity–duration–frequency (IDF) curves are representations of the probability that a given rainfall intensity will occur within a given period. At the global scale, however, only for a few locations sub-daily rain gauge data is available from which global IDF curves could be derived. This poses a major challenge for simulations of global pluvial flood hazard and risk which require information of intensity, duration, and probability as boundary conditions. Therefore, efficient yet accurate means for scaling the locally available data to the global extent need to be found.

Consequently, we use available quality-controlled sub-daily precipitation data from the GSDR data set to derive growth curve parameters at around 10,000 locations world-wide. After combining these scale and shape parameters with globally available data of main precipitation drivers, a regionalized machine learning model is first trained and tested and then applied to produce global IDF maps.

Finally, we evaluated these maps against an ensemble of openly available local IDF curves found in literature. By selecting locations spread across the globe, we try to ensure to include as much variability as possible in the evaluation. Additionally, the global IDF curves were benchmarked against available more bespoke IDF data in the USA and UK.

While such data-driven approaches clearly depend on the quality and quantity of available sub-daily rainfall observations, the method still shows to capabilities of current data-driven modelling approaches to scale local data to global data applicable in both flood risk research and practice.