Keywords
rainwater tanks, upscaling, surrogate modelling
Start Date
1-7-2012 12:00 AM
Abstract
Mathematical modelling of decentralised water supply options requires a finertemporal resolution than models of centralised regional systems. To avoid having to runa coarse scale model with the timestep needed for the fine scale model, the latter can beupscaled to a larger timestep. One way of doing this is to use a surrogate model.This paper presents a nearest-neighbour surrogate model for upscaling a detailed clusterscale urban water model from a sub-daily to a monthly time scale. The approach consistsof running a detailed cluster model with a fine temporal resolution using a historicalrainfall timeseries, adjusting the daily rainfall values by clipping peaks and assigning rainfalloccurring near the end of a month to the following month, and then aggregating theoutput and corrected input timeseries to monthly timesteps. This set of monthly valuesis then used as a surrogate model in subsequent simulations, using nearest-neighboursampling to select an appropriate output value for a given monthly rainfall.It was found that the surrogate model performs well in emulating the detailed model at amonthly timestep, producing a good model fit and succeeding in reproducing autocorrelationwhile running faster than the detailed model by several orders of magnitude.
A Nearest-Neighbour Surrogate Model for the Simulation of Rainwater Tanks
Mathematical modelling of decentralised water supply options requires a finertemporal resolution than models of centralised regional systems. To avoid having to runa coarse scale model with the timestep needed for the fine scale model, the latter can beupscaled to a larger timestep. One way of doing this is to use a surrogate model.This paper presents a nearest-neighbour surrogate model for upscaling a detailed clusterscale urban water model from a sub-daily to a monthly time scale. The approach consistsof running a detailed cluster model with a fine temporal resolution using a historicalrainfall timeseries, adjusting the daily rainfall values by clipping peaks and assigning rainfalloccurring near the end of a month to the following month, and then aggregating theoutput and corrected input timeseries to monthly timesteps. This set of monthly valuesis then used as a surrogate model in subsequent simulations, using nearest-neighboursampling to select an appropriate output value for a given monthly rainfall.It was found that the surrogate model performs well in emulating the detailed model at amonthly timestep, producing a good model fit and succeeding in reproducing autocorrelationwhile running faster than the detailed model by several orders of magnitude.
