A methodology for probabilistic real-time forecasting – an urban case study

Abstract

The phenomenon of urban flooding due to rainfall exceeding the design capacity of drainage systems is a global problem and can have significant economic and social consequences. The complex nature of quantitative precipitation forecasts (QPFs) from numerical weather prediction (NWP) models has facilitated a need to model and manage uncertainty. This paper presents a probabilistic approach for modelling uncertainty from single-valued QPFs at different forecast lead times. The uncertainty models in the form of probability distributions of rainfall forecasts combined with a sewer model is an important advancement in real-time forecasting at the urban scale. The methodological approach utilized in this paper involves a retrospective comparison between historical forecasted rainfall from a NWP model and observed rainfall from rain gauges from which conditional probability distributions of rainfall forecasts are derived. Two different sampling methods, respectively, a direct rainfall quantile approach and the Latin hypercube sampling based method were used to determine the uncertainty in forecasted variables (water level, volume) for a test urban area, the city of Aarhus. The results show the potential for applying probabilistic rainfall forecasts and their subsequent use in urban drainage forecasting for estimation of prediction uncertainty.

Real-time urban flood forecasting and modelling – a state of the art

“…The quality of the weather forecast has a major influence on the quality of a flood forecast system (Renée et al 2013). It is generally accepted that a rain gauge network alone is not sufficient for a proper weather forecast.…”

A bootstrap regional model for assessing the long-term impacts of climate change on river discharge

“…Different methods are used to assess the uncertainty of model parameters, especially due to insufficient data, such as the Monte Carlo method (Blasone et al, 2008;Nakhaei and Etemad-Shahidi, 2012), the Latin hypercube sampling (LHS) method (McKay et al, 1979;René et al, 2013), the generalised likelihood uncertainty estimation (GLUE) (Beven and Binley, 1992;Blasone et al, 2008) and the bootstrap method (Trichakis et al, 2011;Burn and Taleghani ,2013). In this study, the distribution-independent bootstrap method, first proposed by Efron (1979), was used to generate sufficient data via resampling to further evaluate parameter uncertainty of the RFIM.…”

Evaluation of an Application of Probabilistic Quantitative Precipitation Forecasts for Flood Forecasting

“…Individual member QPFs are commonly used to create probabilistic QPF (PQPF) [31,32], which provides a measure of uncertainty in the rainfall forecasts. Although one could simply assign probabilities at grid points based on the number of members showing precipitation above a threshold, often these forecasts are calibrated by comparing the model forecasts to observations for a period of time to account for systematic biases in the ensembles.…”