HAL Id: hal-02599965
https://hal.inrae.fr/hal-02599965
Submitted on 16 May 2020
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Sensitivity analysis and metamodeling to help sizing vegetative filter strips in a watershed
Claire Lauvernet, Dominikus Noll, R. Muñoz Carpena, Nadia Carluer
To cite this version:
Claire Lauvernet, Dominikus Noll, R. Muñoz Carpena, Nadia Carluer. Sensitivity analysis and meta-modeling to help sizing vegetative filter strips in a watershed. EGU 2014, Apr 2014, Vienna, Austria. pp.1, 2014. �hal-02599965�
www.irstea.fr
■ The virtual scenarios are performed from the whole combination of :
■ for two areas in France, considering they will more or less cover the diversity of rainfall situations encountered in Europe.
for one given catchment in Europe, it may be realistic to assimilate the rainfall characteristics to the “North” or “South” ones.
■ For each area :
for each scenario, we get Runoff Delivery Ratio of a VFS in function of its length : RDR* = Runout/(Runin + Rain)
Scenario : 4794
11 lengths of VFS (5-25) : 52800 simulations, in 6 days Soil type : A/B/C/D = 1200
VFS Soil type : CLO/SAL/SCL/SIL = 1200
■ The user can choose in the plots catalog in function of his knowledge of the field : water table in summer & winter, soils types, contributive area length
Optimal VFS length for each scenario for several types of storms
Conclusion : A complete tool but quite difficult to apply for non-modelers
Simulations on many
combinations to simulate virtual scenarios and meta-modeling
Blue = winter
Yellow = summer soil humidity condition II
Green = summer soil humidity condition III
Synthesis of the 52000 simulations
VFS length (m)
RDR*
Contributive area length (m)
op
tim
al
VFS l
en
gth
(m)
Optimal VFS
length for
70%
efficiency
A metamodel/emulator/surrogate of the whole toolchain would allow to
■ evaluate an output of the toolchain at any point of the domain
■ evaluate sensitivity indices larger GSA at smaller numerical cost To perform a metamodel, one need to answer simultaneously to :
Which sampling strategy ?
What kind of metamodel ?
How to validate the metamodel?
Sensitivity analysis and metamodeling to help sizing
vegetative filter strips in a watershed.
Lauvernet C.1, Noll D.1, Muñoz-Carpena R. 2, Carluer N. 1
1 IRSTEA : National Research Institute of Science and Technology for Environment and Agriculture, Irstea centre de Lyon-Villeurbanne, France.
2 Agricultural and Biological Engineering. University of Florida. 287 Frazier Rogers Hall, PO Box 110570 Gainesville, FL 32611-0570 (EEUU).
Contact : claire.lauvernet@irstea.fr Eu ropean Geophysi cal Un io n, V ie nna , 201 4 Conclusion :
■Irstea developed a toolchain which allows to perform the whole modeling
■an application of this toolchain on a large number of scenarios allows giving some operationnal answer to the user
with a graphical tool which helps chosing the optimal VFS size in a catalog of situations. However, it is still limited to only 2 climatic regions.
the current metamodel is not satisfying because of a lack of sampling and/or dynamics
Perspectives :
■Perform the same method on many different climatic regions
■Develop a metamodel to perfom sensitivity analysis, and to be used by non modelers though being robust.
■A larger sampling should allow giving keys for VFS sizing for non-modelers
Vegetative Filter Strip in the
Dombes region (France)
■ Contaminants in surface water are partly due to pesticide applications.
■ Vegetative filter strips (VFS) are a common tool to reduce non point source pollution of water by reducing surface runoff.
■ VFS need to be adapted to the agro-pedo-climatic conditions, both in terms of position and size, in order to be efficient.
TOPPS-PROWADIS project* involves European experts and stakeholders to develop and recommend Best Management Practices (BMPs) to reduce pesticide transfer by drift or runoff in several European countries.
Irstea developed a guide to design VFS by simulating their efficiency to limit transfers (of water).
This study aims at developing a tool which can be used by operators
Objectives
Contributive Area
Buffer Zone
1 2
■ Hypothesis : buffer zone efficiency = ability to retain surface runoff (Possibility of evolution if pesticides and sediment data available)
■ Definition of agro-pedo-climatical scenarii, adapted to the studied area 2 steps :
1. Definition of the surface runoff entering the buffer zone scenario 2. Definition of buffer zone characteristics
Description of the tools’chain
1 2
1/ Contributive area definition
(surface, length, slope)
HydroDem
3/ Hydrograph definition
Curve Number method
Initial soil status choice, depending on the season
*http://abe.ufl.edu/carpena/vfsmod/index.shtml
**Munoz-Carpena, Lauvernet, Carluer, submitted
Input of
Simulations on virtual catchments
Contributive area characteristics VFS characteristics
X
North oceanic temperated climate, Amiens (N)
South mediterranean climate, Roujan (S)
1 winter scenario (W): intermediate intensity (I), quite long duration : 12h00
2 summer scenarios (S): short with high intensity (Peak), 2h more moderated (Intermediate), 6h
Initial humid. cond. Crop : wheat,corn
Hydr. Soil Type : ABCD Slope : 2%,5%,10% Length : 50-300m
Soil type :
SIL, SAL, CLO, SCL
Water table depth :
1 m/2.5 m.
efficiency of 70%
RD R* Sensitivity analysisMetamodeling
Here, the sampling is not optimal
Parametric regression Nonparametric regression Gaussian process regression Learning methods
Kriging Neural Network
SVM
Random Forests Linear
Polynomial
Spline
Generalized Additive Model
Need for another database for an independant validation
GAM and Random Forest are promising BUT: - no metamodel could be determined from these simulations
- need for more simulations on a predefined optimal sampling
take into account the flow dynamics & surface/subsurface interactions ?
need for a mecanistic dynamic model anyway? Part of explained variance per method : only 50% in the best cases
VFS Optimal
length
2/Precipitation event definition
HyétoHydro
Intensity – Duration – Seasonal frequency
(french data base)
Hyetogramm duration and form choice
models
m
odels
VFSMOD*
simulating water, suspended matters (and pesticides) transfer
inside a vegetative filter strip,
adapted to water table presence ** VFS length (m)
Test on different methods
