3. Conclusions 1. Introduction 2. Results Comparison of two stochastic methods for disintegrating daily precipitation to a sub-hourly series Erle Kristvik 1 , Ashenafi Seifu Gragne 1 , Tone Merete Muthanna 1 , and Komlan A. Kpogo-Nuwoklo 2 1 Norwegian University of Science and Technology, Department of Civil and Environmental Engineering, 2 Freie Universität Berlin, Institut für Meteorologie Statistical Meteorology Acknowledgements References The BINGO project has received funding from the European Union’s Horizon 2020 Research and Innovation program, under the Grant Agreement number 641739. Water managers in the city of Bergen, Norway, deal with high amounts of annual precipitation loads, causing frequent flooding of combined sewer overflows (CSOs) and consequently pollution to receiving water bodies. An adaptation of the current system is necessary in order to reduce the environmental damage today, and in a future climate. Although climate scenarios are becoming more and more available, projections of sub-hourly precipitation series for modelling urban hydrologic systems and impacts of climate change are scarce. In order to generate precipitation series of five-minute intervals and make practical use of the existing climate projections, we apply and compare a non-parametric K-nearest neighbor (KNN) and a parametric Poisson cluster model (PCM) weather generator: Non-parametric (KNN T) • Traditional K-Nearest neighbor approach • Data driven bootstrapping of time series based on multivariate nearest neighbor probability density estimation [1] • implemented by adapting the algorithm presented in [1] to sub-hourly disintegration of precipitation series Parametric (BLRPRx) • Randomized Bartlett-Lewis Rectangular Pulse model with exponentially distributed cell intensities • 6 parameter {λ, ι, α, φ, κ, ν} variant of the model as described in [2] • implemented using available R package ‘HyetosMinute’ [2] which includes both parameter estimation and disintegration schemes. • Statistics used for optimization (mean and variance) well preserved for both methods and durations (Fig1) • KNN T shows better skill in capturing important statistics, such as skewness, proportion dry and empirical cumulative distribution (Fig2, Fig3, and Fig4) • BLRPRx shows better skill in disintegrating 1h precipitation, while KNN is equally successfull for 5min and 1h durations (Fig3 and Fig4) • Successful disaggregation with the KNN method enables for temporal downscaling of higher resolution climate projections to be further used in climate change impact analyses and adaptation planning of urban drainage systems Fig1 Fig2 Fig3 Fig4 [1] Lall, U. & Sharma, A. A Nearest Neighbor Bootstrap For Resampling Hydrologic Time Series. Water Resour. Res. 32, 679–693 (1996). [2] Kossieris, P., Makropoulos, C., Onof, C. & Koutsoyiannis, D. A rainfall disaggregation scheme for sub-hourly time scales: Coupling a Bartlett-Lewis based model with adjusting procedures. J. Hydrol. 556, 980–992 (2018). Contact: [email protected]

VHULHV - BINGO · Microsoft PowerPoint - poster_egu.pptx Author: erlekr Created Date: 4/3/2018 2:01:37 PM

Download PDF Report

Upload
others
View
2
Download
0

Embed Size (px)

Citation preview

Page 1: VHULHV - BINGO · Microsoft PowerPoint - poster_egu.pptx Author: erlekr Created Date: 4/3/2018 2:01:37 PM

3. Conclusions1. Introduction 2. Results

Comparison of two stochastic methods for disintegrating daily precipitation to a sub-hourly series

Erle Kristvik1, Ashenafi Seifu Gragne1, Tone Merete Muthanna1, and Komlan A. Kpogo-Nuwoklo2

1Norwegian University of Science and Technology, Department of Civil and Environmental Engineering, 2Freie Universität Berlin, Institut für Meteorologie Statistical Meteorology

Acknowledgements

References

The BINGO project has received funding from the European Union’s Horizon 2020 Research and Innovation program, under the Grant Agreement number 641739.

Water managers in the city of Bergen,Norway, deal with high amounts of annualprecipitation loads, causing frequentflooding of combined sewer overflows(CSOs) and consequently pollution toreceiving water bodies. An adaptation of thecurrent system is necessary in order toreduce the environmental damage today,and in a future climate.

Although climate scenarios are becomingmore and more available, projections ofsub-hourly precipitation series for modellingurban hydrologic systems and impacts ofclimate change are scarce.

In order to generate precipitation series offive-minute intervals and make practical useof the existing climate projections, we applyand compare a non-parametric K-nearestneighbor (KNN) and a parametric Poissoncluster model (PCM) weather generator:

Non-parametric (KNN T)• Traditional K-Nearest neighbor approach• Data driven bootstrapping of time series

based on multivariate nearest neighbor probability density estimation [1]

• implemented by adapting the algorithm presented in [1] to sub-hourly disintegration of precipitation series

Parametric (BLRPRx)• Randomized Bartlett-Lewis Rectangular

Pulse model with exponentially distributed cell intensities

• 6 parameter {λ, ι, α, φ, κ, ν} variant of the model as described in [2]

• implemented using available R package ‘HyetosMinute’ [2] which includes both parameter estimation and disintegration schemes.

• Statistics used for optimization (mean and variance) well preserved for both methods and durations (Fig1)

• KNN T shows better skill in capturing important statistics, such as skewness, proportion dry and empirical cumulative distribution (Fig2, Fig3, and Fig4)

• BLRPRx shows better skill in disintegrating 1h precipitation, while KNN is equally successfull for 5min and 1h durations (Fig3 and Fig4)

• Successful disaggregation with the KNN method enables for temporal downscaling of higher resolution climate projections to be further used in climate change impact analyses and adaptation planning of urban drainage systems

Fig1 Fig2

Fig3 Fig4[1] Lall, U. & Sharma, A. A Nearest NeighborBootstrap For Resampling Hydrologic Time Series. Water Resour. Res. 32, 679–693 (1996).

[2] Kossieris, P., Makropoulos, C., Onof, C. & Koutsoyiannis, D. A rainfall disaggregation scheme for sub-hourly time scales: Coupling a Bartlett-Lewis based model with adjusting procedures. J. Hydrol. 556, 980–992 (2018).

Contact: [email protected]