Session 2 Statistical Modelling of Rain 2023-2024 Anglais
Session 2 Statistical Modelling of Rain 2023-2024 Anglais
Session 2 Statistical Modelling of Rain 2023-2024 Anglais
Outline
I. Statistical analysis
I.1 IDF Curves
I.2 Gumbel fitting
III. Spatialization
I- Statistical characterization of rainfall event :
Fixed origin
i(mm/h) i(mm/h)
40 40
30 30
20 20
Moving window
10 10 (d=15 mn here)
Averaging with origin kept constant Moving window for the averaging
whatever the duration d for a given duration d
d=15 mn
d=35 mn
I- Statistical characterization of rainfall event :
𝑗 𝑑𝑢𝑟𝑎𝑡𝑖𝑜𝑛 𝑑𝑗 𝑗
n pluviometric events : 𝐼𝑖 = max(𝐼𝑒𝑣𝑒𝑛𝑡 𝑖 ) (1 < 𝑖 < 𝑛) [1] 𝐼1
𝑗
𝑗
𝐼2
n values of 𝐼𝑖 following def [1] 𝑗
𝐼3
• One keep only p values (p<n).
• Intensities classified in decreasing order : (… )
𝑗
𝐼𝑖
rank i (1<i<p) = rank of the jeme maximum for this duration dj
(… )
𝑗
𝑗
𝐼1,𝑝 Random vector whose values that follow Gumbel law 𝐼𝑝−1
𝑗
𝐼𝑝
𝑗
𝐹′ = 𝑝𝑟𝑜𝑏 𝐼 < 𝐼𝑖
𝑗 𝑗 𝑗 𝑗
𝑗 𝐼1 > 𝐼2 > ⋯ > 𝐼𝑝−1 > 𝐼𝑝
𝐹 = 𝑝𝑟𝑜𝑏(𝐼 > 𝐼𝑖 ) = 1 − 𝐹′
𝑖−𝑎 (Empirical frequency)
approximated by 𝐹 =
𝑝+1−𝑎
a=0.5 (usual choice in France)
I- Statistical characterization of rainfall event :
𝑇𝑖 = 1Τ𝐹𝑖 (years)
By construction
Intensities associated
to the same return period
𝑇𝑖
for the different duration dj
𝐼(𝑑𝑗 , 𝑇𝑖 = 𝑇)
I- Statistical characterization of rainfall event :
4,50
T=50 ans
4,00
T=20 ans
3,50
T=10ans
3,00
T=5 ans
2,50
T=10 ans : i(t)=43/(t+10)
2,00
1,50
1,00
0,50
0,00
0 50 100 150
I- Statistical characterization of rainfall event :
• For duration less than 1h, the same formulation can be used but the b coefficient is usually
different than the one for duration >1h.
• Singularity when d→0. Usually several fit on different ranges of duration to hampers
uncertainties at small duration and long duration
Other formulations :
𝐼 𝑑, 𝑇 = 𝑎(𝑇)Τ(𝑑 + 𝑏 𝑇 ) Talbot
(…)
I- Statistical characterization of rainfall event :
Gumbel law
𝑥 − 𝑥0
′
𝐹 =𝑒 −𝑒 −𝑢 𝑢 Reduced variable 𝑢=
𝑔
Methology
Methods of moment :
6
𝑔= 𝜎 = 0.78𝜎
𝜋
With 𝑝
1
µ= 𝑥𝑖
𝑝
𝑖=1
𝑝
2
1 2
𝜎 = 𝑥𝑖 − µ
𝑝
𝑖=1
I- Statistical characterization of rainfall event :
Based on the IDF curves. Rainfall event shape = completly advanced pattern
1 𝑡
From the definition of the average on time t 𝐼 𝑡, 𝑇 = න 𝑖 𝜏 𝑑𝜏
𝑡 0
𝑑
So : 𝑖 𝜏 = 𝐼(𝑡, 𝑇) × 𝑡
𝑑𝑡
𝐼 𝑡, 𝑇 = 𝑎𝑡 𝑏
𝑑
𝑖 𝑡 = 𝑎𝑡 𝑏+1 = 𝑎(𝑏 + 1)𝑡 𝑏
𝑑𝑡
II- Synthetic hyetograph
𝑏
𝐼 𝑡, 𝑇 = 𝑎Τ 𝑡 + 𝑐
𝑏 𝑏−1 Τ 𝑡+𝑐 2𝑏
𝑖(𝑡 = 𝑎 × 𝑡 + 𝑐 −𝑎×𝑏× 𝑡+𝑐
imax (t, T )
t
II- Synthetic hyetograph
II1 – Keifer method (Chicago synthetic hyetograph)
𝐼 𝑛𝑑𝑡, 𝑇 − 𝐼( 𝑛 − 1 𝑑𝑡, 𝑇)
𝑖 𝑛𝑑𝑡 = × 𝑛𝑑𝑡 + 𝐼( 𝑛 − 1 𝑑𝑡, 𝑇)
𝑑𝑡
One can also work with the water depth during time step dt
ℎ 𝑛𝑑𝑡 = 𝑖 𝑛𝑑𝑡 × 𝑑𝑡 = 𝑛 × 𝑑𝑡 × 𝐼 𝑛𝑑𝑡, 𝑇 − 𝑛 − 1 × 𝑑𝑡 × 𝐼( 𝑛 − 1 𝑑𝑡, 𝑇)
(mm)
4 40 0,97 38,95 5,65 10,00
5 50 0,88 43,79 4,84 8,00
6 60 0,80 48,07 4,28
6,00
7 70 0,74 51,93 3,86
4,00
8 80 0,69 55,48 3,54
9 90 0,65 58,76 3,28 2,00
10 100 0,62 61,83 3,07 0,00
11 110 0,59 64,72 2,89
10
30
50
70
90
0
12 120 0,56 67,46 2,74
11
13
15
17
19
21
23
13 130 0,54 70,07 2,61
14 140 0,52 72,56 2,49
duree (mn)
15 150 0,50 74,95 2,39
16 160 0,48 77,25 2,30
17 170 0,47 79,46 2,21
18 180 0,45 81,60 2,14
19 190 0,44 83,67 2,07
20 200 0,43 85,67 2,01
21 210 0,42 87,62 1,95
22 220 0,41 89,52 1,90
23 230 0,40 91,36 1,85
24 240 0,39 93,16 1,80
a.t (1-a).t
t1 t=t2-t1
DT
t t2
1 𝑡2
𝐼 𝑡, 𝑇 = න 𝑖 𝜏 𝑑𝜏
𝑡 𝑡1
𝑑 𝑡2 𝑑
න 𝑖 𝜏 𝑑𝜏 = 𝐼(𝑡, 𝑇) × 𝑡
𝑑𝑡 𝑡1 𝑑𝑡
Leibnitz rule :
𝑑 𝑑 𝑑
𝑡 (𝑡) × 𝑖 𝑡2 (𝑡) − 𝑡 𝑡 × 𝑖 𝑡1 𝑡 = 𝐼(𝑡, 𝑇) × 𝑡
𝑑𝑡 2 𝑑𝑡 1 𝑑𝑡
𝑑
1 − 𝛼 × 𝑖 𝑡2 (𝑡) + 𝛼 × 𝑖 𝑡1 𝑡 = 𝑓(𝑡) with 𝑓 𝑡 = 𝐼(𝑡, 𝑇) × 𝑡
𝑑𝑡
II- Synthetic hyetograph
II1 – Keifer method (Chicago synthetic hyetograph)
Calculate : 𝑡1 𝑡 = 𝛼 𝐷𝑇 − 𝑡
𝑡2 𝑡 = 𝛼𝐷𝑡 + 1 − 𝛼 𝑡
𝑖 𝑡1 = 𝑖(𝑡2 ) = 𝑓(𝑡)
Repeat with different values of t to build the rain from a set of points
II- Synthetic hyetograph
II1 – Keifer method (Chicago synthetic hyetograph)
Synthetic hietograph
IDF curve
Imax(t,T)
t
DT
t
DT
By construction,
areas : = =
t DT So …
Keiffer synthetic hietrograph
II- Synthetic hyetograph
II1 – Keifer method (Chicago synthetic hyetograph)
Advanced
Centered
Discrete hietograph
Discrete hietograph
(…etc)
1 2 3
3 1 2
4 5
5 4
6
7 6
8 7 8
II- Synthetic hyetograph
II1 – Keifer method (Chicago synthetic hyetograph)
2 3
Non centered discrete
1 2 3
hietograph
4 5 4 1 5
6 6
7 7 8
8
{
{
b) Kiefer method (Chicago synthetic hyetograph)
II- II- Synthetic
Synthetic hyetograh
hyetograph
II1 – Keifer method (Chicago synthetic hyetograph)
Example (same IDF curve as the previous examples for the advanced rainfall
event)
II- Synthetic hyetograph
II1 – Keifer method (Chicago synthetic hyetograph)
18
16
14
12
10
Hk mm
0
10
20
30
40
50
60
70
80
90
100
110
120
130
140
150
160
170
180
190
200
210
220
230
240
n.dt (mn)
Example (same IDF curve as the previous examples for the advanced rainfall
event)
II- II- Synthetic
Synthetic hyetograh
hyetograph c) Desborde double triangle hyetograph
II2 – Double triangle rainfall event (Desbordes rainfall)
Parameters value: « Guide de Construction et d’utilisation des pluies de
projets », Service Techniques de l’Urbanisme, Paris, Hemain (1986)
Dp DT HT HM1
Example of parameters value. Toulouse-Blagnac (Region 2). Return period T=10 ans for the
duration of the intense period
Same procedure as IDF curves to get the values in the table above.
(which is an example)
For a given set of N events with the same duration DT,
- statistical analysis of the peak events on duration Dp from a
subsample of P events : water depth H(Dp,T) and associated return period
T
- for a given set of P’ subvent whose return period is T for the
peak intensity (so such as H roughly H(Dp,T) on Dp HM1
a) Calculation of the associated water depth on DT
b) Calculation of the return period of the associated
water depth
Dp
tp DT
II- II- Synthetic
Synthetic hyetograh
hyetograph
20,00
18,00
16,00
14,00
Hdesborde (mm)
12,00
10,00
8,00
6,00
4,00
2,00
0,00
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
n dt (mn)
II- Synthetic hyetograph
Chocat B., Thibault S., Bouyat M. ; "Etude comparative des résultats fournis par la méthode de Caquot et le modèle du
Chocat et al (1981): réservoir linéaire" ; TSM l'eau ; n°7 ; pp 417-424 ; juillet 1981.
1 - (0.1) 120 a 2
t
b +1
i(t2 ) = (0.25.K )b b
0.9 0.1 b
Qe
b (0.1) - 1
b
imax = (0.25.K ) 120 a 2 b
Qs K
0.9 0.1b
2.d=0.5 K t2=2.25 K t3=2.5.K r=(t3-t2)/t3=0.1
t
imax
t3 • K=time lag : K> 10 mn
t et K in mn
i(t2) i(t2) et imax in mm/h
t2 2.d
DT=2.t3
II- Synthetic hyetograph c) Desborde double triangle hyetograph
II2 – Double triangle rainfall event (Desbordes rainfall)
i(t2 ) = (d )b
1 - (0.1) 120 a 2
b +1
b
Canoe Software : symetrical rainfall event
0.9 0.1 b
• DT=10.d (arbitrary) (correspond to Chocat’s and al
imax = (d )
b (0.1) - 1
b
120 a 2b •
results)
t2=(DT-2d )/2
0.9 0.1b
imax
• a=a(T) et b=b(T) = coeff de Montana
t et K in mn
i(t2) et imax in mm/h
i(t2)
t2 2.d
DT
II- Synthetic hyetograph
𝑏(𝑇)
𝑖𝑚𝑎𝑥 = 2. 𝑎 𝑇 . 𝐷𝑇
𝑖𝑚𝑎𝑥
This kind of synthetic hietograph is suitable for catchment
with fast response (K below 30 mn).
tp
DT
. From Chocat et al (1981) : K<10 mn. DT=8.9 C4.23.b . K and DT in mn
Constant rainfall
Rainfall with a chronogy : i(t). Triangle or double triangle
Fulfill the I.D.F statistic :
i(t) 1 𝑡2
𝐼 𝑡, 𝑇 = න 𝑖 𝜏 𝑑𝜏
𝑡 𝑡1
𝑖 𝑡, 𝑇
t
d
𝑖 𝑡, 𝑇 = 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡𝑒 = 𝐼(𝑡, 𝑇)
advanced
General case
II- Synthetic hyetograph
To summarize
Those synthetic rainfall events are punctual events i(x,y,t) associated to a water depth
H(x,y) over the duration t
For some applications, one are interested in average water depth over the surface
catchment A
1
A
LM ( A, d ) = H ( x, y, d )dA
A
This require spatialization methods
Constant rainfall
Rainfall with a chronogy : i(t). Triangle or double triangle
Fulfill the I.D.F statistic :
i(t) 1 𝑡2
𝐼 𝑡, 𝑇 = න 𝑖 𝜏 𝑑𝜏
𝑡 𝑡1
𝑖 𝑡, 𝑇
t
d
𝑖 𝑡, 𝑇 = 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡𝑒 = 𝐼(𝑡, 𝑇)
advanced
General case
III- Rainfall spatialization
III- Rainfall spatialization
III-1 Thiessen Polygon’s method (=Voronoi method)
1
LM ( A, d ) =
A A
H ( x, y, d )dA
1
LM ( A, d ) = Aj .H j = j w j .H j
A j
t
H j = i j ( x j , y j , )d
0
In the later case, spatialization often based on simplified assumption on the rainfall event
structure (epicenter), in particular, continuous decreasing of the rain intensity compared
to the distance with this epicenter
d
H e ( xe , ye ) = ie ( xe , ye , t )dt Corresponding water depth over duration d
0
d
H ( x, y, t ) = i( x, y, t )dt Water depth over duration d at an other localation (x,y)
0
1
LM ( A, d ) =
A A
H ( x, y, d )dA
1
d
=
A A 0
i ( x, y, t )dt dA
1
d
= a ( x, y )ie ( xe , ye , t )dt dA
Assumtion :
Spatial damping independant A A 0
on the time. Caution !!!!
d
1
=
A A a ( x, y )dA ie ( xe , ye , t )dt
0
LM ( A, d ) = a ( A) H e ( xe , ye , d )
<a>(A) dependance with the rainfall event duration but also the return period
Probabilistic approach (as for the I.D.F curves). Frequency analysis on the quantity L M(A,t) to get
<a>(A,t,T)
III- Rainfall spatialization
C) Spatial damping coefficient
III-2 Spatial damping coefficient
• Rainfall damping depending on the distance to the epicentre : a(r). Can be computed from
the knowledge of <a>
rI Intensité
Bassin I E Pluie à l’épicentre
B temps
Bassin II
As a conclusion
Simple definition of the spatial damping <a(A)> does not suit well the structure of real spatial
event, as on small time step, fluctutions from an event to another
Mathematical relations are empirical and assume a given « shape » of the spatial variation
(with a continuous decrease) around the center : radar measurements show that this simplified
view is not close to reality.
In urban hydrology :
- For surface catchment lower than 1000 ha : safer to consider a spatially constant
rainfall without damping (rainfall equal to the one considered at the « center »)
- For surface catchment greater than 3000 ha : more precise modelling of the spatial
repartition of rain. Synthetic hyetograph not advised. Better to work with real rainfall event