Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
GeoENV 2010 – Gent, 13-15 Sept. 2010
Improved mapping of daily precipitation over Quebec daily precipitation over Quebec
using the i i i hmoving-geostatistics approach
N. Jeannée (GEOVARIANCES) & D. Tapsoba (HYDROQUEBEC)
Contact: [email protected]: +33 (0)1 60 74 74 54 – Mob: +33 (0)6 84 04 35 41
Contents
Motivation
Input dataInput data
Methodology: the M-GS approachgy pp
Results
Conclusions and Perspectives
2/11
MotivationContext
− Daily precipitation: key parameter for predicting hydropower generationin Québec (Canada)
Several modeling issues: scarce monitoring network large spatial − Several modeling issues: scarce monitoring network, large spatial variability, local anisotropies, non stationarity, ...
− Development of the “Moving-Geostatistics” framework by Estimages & Geovariances dedicated to the local optimization of parameters involved Geovariances, dedicated to the local optimization of parameters involved in variogram-based models
Obj tiObjectives− Present the M-GS approach
i) Determination of local modeling parameters (ranges, sills, anisotropy directions…);
ii) Use of these parameters in subsequent estimation/simulation algorithms.
− Test its efficiency for the modeling of daily precipitationy g y p p
3/11
Input Data 6000 Data (mm)
22.40 1
Dataset from the « Réseau 5750
m)
Alti(m)
800
0.1No rainBasins
Météorologique Coopératif du Québec », a partnership initiave from the managers of
5250
5500
Y (km 800
700
600
500
400
300from the managers of meteorological networks.
0 500 1000
5000
300
200
100
0
Focus on daily precipitations for a specific day: January 18, 2009.
0.4 Nb Samples: 176Minimum: 0.0Maximum: 22.4Mean: 2.3
X (km)
Characteristics:
0.3
encies
Mean: 2.3Std. Dev.: 2.7
Characteristics:skewness, non stationarity, local anisotropies... 0.1
0.2
Frequ
0 10 20
Rainfall (mm)
0.0 ⇒ Which modeling framework?
4/11
Methodology
Moving-Geostatistics:Moving Geostatistics:− Methodology developed since 2007, dedicated to the local
optimization of variogram-based models.optimization of variogram based models.
− Numerous applications initially dedicated to the Oil & Gas Industry:
Gridding (reservoir geometry or properties)
Noise filtering (seismic processing)
Facies modeling (reservoir characterization)
5/11
Methodology
Common modelling issues addressed by M-GS:g y
« Structural » non stationarity(Bathymetry data)
Local accuracy
Small scale structures,strong anisotropy
Highlycontinuous
6/11
Methodology
Idea: estimate and use parameters mapsp p
Main challenge: determining local parameters− Several approaches:
Local cross-validation
Local variogram analysis
Mathematical morphology approach (Ray tracing) to convert images into structural characteristics
− Ability to integrate exogenous information
Then, use of the local parameters in estimation or simulation algorithmsg
7/11
Results
Application to daily precipitationApplication to daily precipitation− Comparison between two approaches:
U i l k i i ( ith li t d )Universal kriging (with linear trends)
Kriging with moving parameters (M-kriging)
− Elements of comparison:
Visual control of results
Use of a validation subset
8/11
ResultsM-kriging approach
Determination of local parameters:− Determination of local parameters:
Definition of an analysis grid, use of overlapping
M-parameter 1: main direction of anisotropy (local Cross-Validation)p py ( )
M-parameter 2: variogram sill (computation of local variances)
Other M-parameters: ranges (not shown)
6500 6500
6000
(km) Angle
90
6000
(km)
Sill
5500
Y (
80
70
60
50
40
30
5500
Y Sill
10 9 8 7 6 5
0 500 1000 1500
X (km)
5000 20
10
0 0 500 1000 1500
X (km)
5000 4 3 2 1 0
9/11
Results
Comparisonp− Visual control of daily precipitation maps
5750
6000
5750
6000
5250
5500
Y (km)
5250
5500
Y (km)
[mm]
10 98
0 500 1000
5000
5250
M-kriging
0 500 1000
5000
Univ. Kriging
8 7 6 5 4 3 2 1
0 500 1000
X (km)
0 500 1000
X (km)0
10/11
Results
Comparisonp− Visual control of daily precipitation maps
5750
6000
5750
6000
5250
5500
Y (km)
[mm]
5.04.50
5250
5500
Y (km)
0 500 1000
5000
5 50
Univ. Kriging StDev
4.03.53.02.52.01.51.00.5 0 500 1000
5000
M-kriging StDev
0 500 1000
X (km)0.0
X (km)
10/11
Results
Comparisonp− Visual control of daily precipitation maps
5750
6000
5750
6000
5250
5500
Y (km)
[mm]
5.04.50
5250
5500
Y (km)
0 500 1000
5000
5 50
Univ. Kriging StDev
4.03.53.02.52.01.51.00.5 0 500 1000
5000
M-kriging StDev
− Validation results (34 points - 20% subset)
0 500 1000
X (km)0.0
X (km)
Method rho RMSE
UK 0.63 1.75
M-GS 0.70 1.56
10/11
Conclusions and Perspectives
Conclusions− Accounting for local parameters: a promising and pragmatic
approach to improve the modeling of variables presenting structural characteristics which are locally varyingcharacteristics which are locally varying.
− Daily precipitation case: local anisotropies, non stationarity…
Perspectives− Several theoretical aspects still need to be addressed: model
authorization, local anamorphosis for simulations, distinguishingstructural variations vs. statistical fluctuations…
− Daily precipitation case: integration of radar data as an auxiliaryvariable, which might also be useful to guide the estimation of precipitation parametersprecipitation parameters.
11/11