-------------
NOTE:
There's one relevant piece of publication missing, which is the Ph.D. thesis of M. Baptist, one of the co-authors of our work. As the thing is 12 MB large, and only a certain part of the work is relevant to the application, I thought it best not to send it to you in this email. It can be found however through http://www.library.tudelft.nl/dissertations/2940/f_203505_true_EN.html
And in principle it's okay that it's mirrored for the competition. Note however that only chapter 4 is relevant for the application.
-------------
(1)
Determining Equations for Vegetation Induced Resistance using Genetic Programming
Modelling Floodplain Biogeomorphology (Chapter 4)
(2)
Maarten Keijzer, mkeijzer@xs4all.nl
Martin Baptist, M.J.Baptist@citg.tudelft.nl
Vladan Babovic, babovic@wldelft.nl
Javier Uthurburu, j.uthurburu@hotmail.com
For Keijzer, Babovic & Uthurburu
WL | Delft Hydraulics
Rotterdamseweg 185
P.O. Box 177
2600 MH Delft
The Netherlands
For Baptist:
M.J. Baptist
Delft University of Technology
Faculty of Civil Engineering and Geosciences
Hydrology section
Stevinweg 1
2628 CN Delft
the Netherlands
(3)
Maarten Keijzer
(4)
Determining Equations for Vegetation Induced Resistance using Genetic Programming:
Inducing equations based on theory and data is a time-honoured technique in
science. This is usually done manually, based on theoretical understanding and
previously established equations. In this work, for a particular problem in
hydraulics, human induction of equations is compared with the use of genetic
programming. It will be shown that even with the use of synthetic data for
training, genetic programming was capable of identifying a relationship that
was more concise and more accurate than the relationship uncovered by
scientists. As such this is a human-competitive result. Furthermore it will be
shown that the genetic programming induced expression could be embedded in a
line of theoretical work, filling in a few gaps in an already established line
of reasoning. The resulting equation is the most accurate and elegant
formulation of vegetation induced resistance to date.
Modelling Floodplain Biogeomorphology:
Understanding the interactions between the ecosystem and the morphology of
river floodplains, i.e. floodplain biogeomorphology, is becoming increasingly
important in view of modern river management and climate change. There is a
need for predictive models for the natural response of river floodplains to
hydraulic measures and river rehabilitation, such as river widening,
construction of secondary channels and floodplain lowering. This thesis
investigates floodplain biogeomorphology from a modeller's perspective. It
addresses management concepts such as room-for-the-river and cyclic floodplain
rejuvenation, in which a symbiosis is sought for between flood management and
nature rehabilitation, and it shows how 1-D, 2-D and 3-D numerical models can
be used to support them. Its main focus is on one of the knowledge gaps
emerging from these model applications, viz. the effect of floodplain
vegetation on the bed shear stress and subsequent bedload sediment transport.
By means of a combination of theory development, flume experiments and field
work, this research has resulted into a number of practically applicable and
validated formulations for the hydraulic resistance and bed shear stress
reduction of vegetation. These can be applied in large-scale numerical
morphodynamic models to better design river measures in combination with nature
rehabilitation.
(5)
B, D, E, F, G
(6)
The equation found by Genetic Programming for resistance induced vegetation is
ultimately a refinement of the formula proposed by Antoine Chezy in 1776. Over
the years the problem for non-vegetated streams has been satisfactory solved,
for streams containing vegetation however, the formulation of resistance
(roughness) was still largely unsolved. In a recent literature survey, more
than 800 papers have been published in the last two decades on the subject of
vegetation roughness alone (G). These papers present experimental results and
equations aimed to fit the experiment results. Many of these equations are
attempts at curve-fitting, and don't have a solid theoretical backing. Both the
GECCO paper and the Ph.D. thesis named here present two equations that have
been build by one of the co-authors in order to create a physically sound
equation for resistance induced vegetation based on first principles. Alongside
it, GP was used to create an expression based on data generated by a
significantly more complex dynamical model. Compared with the expressions
based on first principles, which are state of the art in the field, the GP
induced expression was (a) more accurate, (b) amenable to analysis and (c)
significantly more simple. Comparing the GP equation with the equations in the
literature corroborates this. The GP induced equation is the most accurate
equation for resistance induced vegetation over a wide range of conditions
(B,E). This was tested using experimental results from 17 different studies.
The work of Kouwen is particularly relevant for appreciating the GP induced
expression. In 1969 Kouwen proposed a skeleton expression for vegetation
induced resistance in which two key components are still missing. This skeleton
is theoretically sound. The GP induced formulation is an instantiation of
Kouwen's framework, with two concrete expressions that fill in the gaps (F).
This was found while analysing the formula. No part of Kouwen's work was used
to steer the GP search, it was independently discovered by GP.
The equation was created on data generated by a dynamical model. After
selection and analysis, the GP-induced equation and the dynamical model were
tested on laboratory experiments gathered from 17 independent studies. In the
comparison, the GP-expression was not significantly worse than the more complex
dynamical model (E). It is important to note here that while the dynamical
model needs extensive hardware to run, the GP-expression is simple enough to be
used by an engineer in the field only using a pocket calculator.
Currently a publication is in review for the Journal of Hydraulic Research in
order to publish the formula (D).
(7) citations
Keijzer, M., Baptist, M., Babovic, V., Uthuruburu, J.,
Determining Equations for Vegetation Induced Bonabeau, E.W.; Cantu-Paz, E.; Dasgupta, D.; Deb, K.; Foster, J.A.;
de Jong, E.D.; Lipson, H.; Llora, X.; Mancoridis, S.; Pelikan, M.;
Raidl, G.R.; Soule, T.; Tyrrell, A.; Watson, J.-P.; Zitzler, E. (editors).
Proceedings of the Genetic and Evolutionary Computation Conference GECCO-2005.
New York, NY: ACM Press.
Baptist, M.
Modelling Floodplain Biogeomorphology
DUP Science, Delft, 2005.
avalaible through: http://www.library.tudelft.nl/dissertations/2940/f_203505_true_EN.html
Resistance using Genetic Programming.
In Beyer, H.-G.; O