The Copernicus Relays of Luxembourg and Belgium (i.e. LIST, Skywin, ISSeP, Spacebel and VITO) organised the 3rd edition of the AI4Copernicus event, focussing this time on “Earth Observation & Artificial Intelligence solutions for climate change challenges”.
We demonstrate that rainfall space-time maps will provide much more detailed inputs at the correct location and timing so that simulations become more realistic in particular when we do uncertainty propagation where we need to sample from realistic input space-time distributions and consider the correct scale (or space-time support).
We present and illustrate a proposed work flow to build and calibrate a space-time geostatistical model of rain, using radar imagery as a covariate in regression-kriging based simulation. Then, a work flow for run-off and sewer system modelling is presented as well. These work flows are repeated as many times as the Monte Carlo simulation design requires to obtain an ensemble of rainfall input maps and time series of quantity variables and quality variables are produced to analyze how input uncertainty propagates to output uncertainty.
We built an R-version of the EmiStat model which we call EmiStatR. EmiStatR can easily do uncertainty propagation analysis because it can make use of extensive functionalities of R. We propose an uncertainty framework which we apply to the EmistatR in a case study to analyse uncertainty propagation associated to rainfall and impervious areas.