Short-term instead of long-term rainfall time series in rainwater harvesting simulation in houses: An assessment using Bayesian Network

Abstract

Rainwater harvesting simulations usually take long-term rainfall data as input to size system components. However, long-term time series data are not available in many cities around the world. This article aims to investigate the relation between the rainwater harvesting simulation results and the rainfall characteristics of various time series lengths. For 13 cities, 30-year time series were adopted as long-term time series to investigate the influence of time series length on the simulation outcomes. Short-term time series were extracted from those long-term time series and used as input in several computer simulations applied in a generic residential model. Comparison of short-term and long-term time series results allowed to obtain a representative short-term time series that leads to results similar to those obtained using long-term time series. A statistical analysis associated the simulation results of the representative time series with their rainfall characteristics using three indicators: average annual rainfall, seasonality index, and average number of dry days per year. The average number of dry days per year showed to be the most important short-term time series indicator to be considered. Additionally, a Bayesian Network was proposed to serve as a support-decision making tool for future studies. This network uses rainfall indicators and rainwater demand to predict if a given short-term time series leads to results similar to those obtained using a long-term time series. The validation of the Bayesian Network showed the tool is promising and useful to assess the representativeness of short-term time series used in rainwater harvesting simulations in houses.