Conference PaperPDF Available

Abstract and Figures

Recorded presentation: https://youtu.be/LDDONUuKaI4 Visualizing results is more important than ever in scientific dissemination. The analysis and communication of complex phenomena such as flash floods requires new approaches. The target is using a state-of-the-art model with a fast and robust predictive capability, which has been tested in small, steep rivers affected by recent flash floods, and visualizing this model in a relatable manner. To do so, this research project aims to achieve optimized simulations that could be carried out in the prototype of a serious gaming engine. The incentive for such optimized simulations is that complex hydraulic modelling is data-and computationally costly. This clashes with the need for low complexity solutions in the real-time based scenarios that an immersive experience and on-site decision-making requires.
Content may be subject to copyright.
ISBN: 978-90-824846-6-3 | PAGE: 177
OPTIMIZATION AND VISUALIZATION OF NUMERICAL MODELS OF FLASH FLOODS IN
STEEP NORWEGIAN RIVERS
Adina
Moraru
*1
,
Andre
w
P
erkis
1
,
Od
dbjørn
Bruland
1
,
Nils
R
u
¨
ther
1
1Norwegian University of Science and Technology (NTNU), Trondheim, Norway
*Correspondance : adina.moraru@ntnu.no
ABSTRACT
Visualizing results is more important than ever in scientific dissemination. The analysis and communication of com-
plex phenomena such as flash floods requires new approaches. The target is using a state-of-the-art model with a
fast and robust predictive capability, which has been tested in small, steep rivers affected by recent flash floods, and
visualizing this model in a relatable manner. To do so, this research project aims to achieve optimized simulations
that could be carried out in the prototype of a serious gaming engine. The incentive for such optimized simulations is
that complex hydraulic modelling is data- and computationally costly. This clashes with the need for low complexity
solutions in the real-time based scenarios that an immersive experience and on-site decision-making requires.
Keywords:
flash floods, steep rivers, hydraulic modelling, optimization techniques, visualization of numerical models.
1
RESEARCH MOTIVATION
Visualizing the dynamics during a flash flood based on hydraulic modelling, as well as enabling more efficient
numerical simulations, would enhance flood risk assessment and its communication. Available models for floods
in steep Norwegian rivers do not convey the estimated risk based on a user-friendly and three-dimensional real
world. Moreover, 3D visualizations that could be used for educational or training purposes are often not based
on precise hydraulic data.
The target is to implement the optimized numerical models into the prototype for a serious gaming flood platform
(e.g. Virtual Reality or Augmented Reality). Thus, the flood scenario will be complimented by gamification,
storytelling and immersive narrative techniques, which will provide a better user experience and improved risk
perception. The current paper provides an overview of ongoing research, where several methods have been
implemented to address the need for faster hydrodynamic models in two steep Norwegian rivers.
2
METHODOLOGY
The
f
ollo
wing
modelling
techniques
w
ere
implemented
in
tw
o
Norw
egian
r
iv
ers,
namely
Byr
te
a
˚
i
in
T
okk
e-
and
Storelva in Utvik municipalities (south and west of Norway, respectively):
Parallel CPU-based computing of 2D models;
GPU-based computing of 2D models;
Machine Learning (ML) for 1D emulation models.
Riv
er
Byr
te
a
˚
i
(in
T
okk
e),
which
w
as
flooded
in
2009,
w
as
used
to
identify
cr
itical
areas
based
on
geomor
phic
indexes and parallel CPU-based 2D numerical modelling. The river’s geometry was modified and its slope was
virtualized into constant values along the modelled reach to identify the influence of other parameters in the river
response. The results of such study and its applications are described in detail in (Moraru et al., 2019, 2020).
Such critical areas have the potential of being further combined into a locally refined, yet faster-solving, model.
The flash flood that affected Storelva river (in Utvik) in 2017 (Bruland, 2020) has been modelled in 2D (i.e. HEC-
RAS, Iber; 1(a)) and also visualized in 3D (i.e. Blender, Unity, Autodesk 3ds Max). The focus of the 2D models
was to understand the dynamics during the flood event in comparison to on-site documentation of the flood, as
well as searching for an optimal numerical method for small and steep rivers. The HEC-RAS model was carried
out on multiple parallel CPU threads, while the Iber numerical model was carried out on both CPU (single and
multi-thread), as well as on GPU (1(b)). Furthermore, the obtained 2D Iber model was used as input data for a
ML emulation model, and their performance was subsequently compared by Son (2020). Section 3 highlights
especially the outcome of using Iber CPU- and GPU-based computing in Storelva in Utvik.
3
PRELIMINARY RESULTS: CPU VS GPU FOR THE 2017 FLASH FLOOD
GPU computing, as well as the use of ML, are state-of-the-art optimization techniques (Moraru et al., 2020). The
computing tool used for numerical modelling will determine greatly the performance of the model, nevertheless.
For instance, preliminary results of a comparison of the computing performance of different CPUs and GPUs
in a small model of the 2017 flash flood (1(a)) shows that GPU-based computing is generally quicker than CPU-
based computing and that the dispersion of the data is larger in the CPU tests than in the GPU tests (1(b)).
(a)
Utvik flash flood modelled on GPU
80 1 Thread 80
70 3 Threads 70
4 Threads
60 8 Threads 60
GPU - Threads
50 50
40 40
30 30
20 20
10 10
0 0
CPU1 CPU2 CPU3 CPU4 CPU5 GPU-a GPU-b GPU-c GPU-d GPU-e GPU-f GPU-g
Processing unit
(b)
Preliminary CPU and GPU comparison for Utvik study case
Figure 1: (a) Study case, (b) average computing time per step for different CPUs (sing le- and multi-thread) and GPUs.
CPU- and GPU-based simulations need different computational effort based on the number of threads available
in the processing unit, i.e. the CPUs shown in 1(b) have from 1 to 8 threads, whereas the respective GPUs
have from 384 to 4352 threads. As the simulation will experience a speedup until all threads are fully used, the
speedup will be larger in GPU-based simulations than in CPU-based ones. In conclusion, this makes a personal
computer with a GPU suitable for fast hydrodynamic simulations of flash floods in steep Norwegian rivers.
Acknowledgements
The authors wish to thank an anonymous reviewer for the valuable feedback. This publication is part of the
World of Wild Waters (WoWW) project, which falls under the umbrella of Norwegian University of Science and
Technology (NTNU)’s Digital Transformation initiative.
References
Bruland O. (2020). How extreme can unit discharge become in steep Norwegian catchments? Hydrology Research 51(2), 290307.
doi:10.2166/nh.2020.055.
Mor
ar
u
A.,
R
u
¨
ther
N.,
Br
uland
O
.
(2020).
Current
trends
in the
optimization
of
hydr
aulic
flood
sim
ulations
in
ungauged
steep
r
iv
ers.
In
W. Uijttewaal et al. (Eds.), 10th International Conference on Fluvial Hydraulics (River Flow 2020), Delft, the Netherlands, pp. 12311238.
Taylor & Francis Group, ISBN 978-0-367-62773-7.
Moraru A., Usman K. R., Bruland O., Alfredsen K. (2019). River idealization for identification of critical locations in steep rivers using 2D
hydrodynamic modelling and GIS. In 22nd Northern Research Basins Workshop and Symposium, Yellowknife, Canada, pp. 144153.
doi:10.13140/RG.2.2.13276.64647.
Son S. (2020). Optimization of hydraulic flood simulations in steep rivers using GPU and Machine Learning. Msc thesis, Universitat
Politecnica de Catalunya - Norwegian University of Science and Technology.
ISBN: 978-90-824846-6-3 | PAGE: 178
Average time per step (s)
ResearchGate has not been able to resolve any citations for this publication.
Conference Paper
Full-text available
The numerical tools simulating flood events must be accurate, in order to provide useful data, and computationally efficient, to facilitate informed decision-making during a flood. One of the main limitations of modelling software at the present time is the CPU time needed to perform simulations over complex spatial domains. Recent studies show that the models lack the necessary implementations to take advantage of the parallelism available on current hardware, which hinders their use in interesting applications such as real-time flood forecasting, calibration and uncertainty analysis, or visualization and gamification of floods for improved risk communication. For instance, methods based on artificial intelligence have contributed highly to the state-of-the-art of flood simulations, providing better performance and cost-effective solutions than complex hydrodynamic fluid solvers. The current study provides an overview of strategies dedicated to the optimization of hydraulic flood simulations suitable for ungauged steep rivers, which are not extensively studied.
Article
Full-text available
This study presents results of observations and analysis of the flood event in Utvik on 24 July 2017. Observations during and after the event, hydraulic simulations and hydrological modelling along with meteorological observations, are used to estimate the peak discharge of the flood. Although both observations and hydraulic simulations of flood extremes are uncertain, even the most conservative assumptions lead to discharge estimates higher than 160 m3/s at culmination of the flood from the 25 km2-large catchment. The most extreme assumptions indicate it may have been up to 400 m3/s, but there is also strong evidence for the discharge at culmination being between 200 and 250 m3/s. Observations disclosed that the majority of water came from about 50% of the catchment area giving unit discharges up to 18 to 22 m3/s,km2. If the entire catchment contributed it would be from 9 to 11 m3/s,km2. This is significantly higher than previously documented unit discharges in Norway and in the range of the highest observed peak unit discharges in southern Europe. The precipitation causing this event is estimated to be three to five times higher than a 200-year precipitation taken from the intensity–duration–frequency curves for the studied region.
Conference Paper
Full-text available
Planform variation and hydrodynamics in steep rivers were characterized based on the combination of 2D hydrodynamic simulations and GIS tools. The idealization of topography (i.e. slope) and hydrology (i.e. discharge) permitted analysing their relevance and effect in river's response to identify critical locations in case of flooding. A total of 24 scenarios were idealized. Slope idealization was achieved by means of modifying the slope of the river bed while keeping the terrain constant outside of it. Slope ranged from 4% to 5.5%. Discharge idealization was implemented as constant steady flow from Q= 30 m3/s (>QMean) to Q= 105 m3/s (>Q200). Hydrodynamics at channel bends with a curvature of 35º to 95º along 1125 m of reach and their correlation with topography were analysed. Both river banks and channel's centre points were analysed; the left bank generally presented higher values and higher variability. Results show that discharge has a stronger influence in river's response than slope when s≤ 4.5% and Q≥ 60 m3/s. For slopes higher than 4.5%, discharge shows poorer correlation with hydraulic forces than slope gradient's influence. The same trend is observed for Q< 60 m3/s, where slope presents stronger correlations than discharge with river's response. However, the difference in river response decreases and shows a stabilizing tendency with increasing slope. Bend curvature influenced hydraulic response together with channel narrowing in subsequent cross-sections. Water surface elevation was usually lower at the convex side of bends, which presented higher hydraulic forces.
Optimization of hydraulic flood simulations in steep rivers using GPU and Machine Learning
  • S Son
Son S. (2020). Optimization of hydraulic flood simulations in steep rivers using GPU and Machine Learning. Msc thesis, Universitat Politecnica de Catalunya -Norwegian University of Science and Technology. ISBN: 978-90-824846-6-3 | PAGE: 178