Transcript Streamflow Forecasting Project

Streamflow Forecasting
Project
By:
JD Emmert
Derek Rapp
Big Hole River
Melrose, Montana
OBJECTIVE
• The purpose of this project is to identify the
•
•
large scale climate signals that affect the
streamflow of the Big Hole River and to simulate
streamflow scenarios given a probabilistic
forecast of the climate signals.
Determine the 100-year flood for flood control
designs.
Develop models to estimate the monthly
streamflow using parametric and nonparametric
approaches.
High flows occur in the spring months and are
primarily driven by snowmelt. Therefore the
climate signal comes in the previous winter.
500
700
300
500
4000
4000 6000 8000
p(x)
0
500
1000
1500
1000
2000
200
p(x)
200
400
1000
6000
400
600
11
600
800
9
0.0000
p(x)
p(x)
10
0
8
0.0000
600
800
6
0.0000
3000
7
200
600
5
p(x)
p(x)
0 e+00
1000
400
3
p(x)
2000
4
0
200
0.00000
1000 2000 3000
0.00000
0
700
2
p(x)
0 e+00
p(x)
1
0.000
p(x)
100
0.0000
300
800 1000
0.000
100
0.000
p(x)
0.000
p(x)
Distribution of Monthly Streamflows
200
400
600
12
800
Climate Signal Indicators
• ENSO (El Nino Southern Oscillation)
• PDO (Pacific Decadal Oscillation)
• PNA (Pacific/North American Pattern Index)
These climate signals were scatterplotted against
the streamflows.
The climate signals were also correlated with the
streamflows.
0 20 40 60
-40
enso1
1000
2000
3000
4000
5000
4000
5000
4000
5000
0
-2 -1
pdo1
1
flow2
1000
2000
3000
50
-150
-50
pna1
150
flow2
1000
2000
3000
flow2
Flow vs. ENSO
-0.3013
Flow vs. PDO
-0.2873
Flow vs. PNA
-0.2699
4000
3000
2000
la nina
neutral
el nino
la nina
neutral
500000 1500000
el nino
• The means are shown to be different.
• The variances are shown to be similar
The t-test and F-test we performed agree with this.
Conditional and Unconditional Probabilities
• ENSO categories
– High ENSO > 7.0
– Low ENSO < -7.0
– Neutral ENSO is in between.
• Flow categories
– Broken into 33rd percentiles.
• Probabilistic Forecast
– P(high ENSO) = 0.2
– P(low ENSO) = 0.5
– P(neutral ENSO) = 0.3
Theorem of Total Probability
• P(high flow) = 0.4020
• P(low flow) = 0.2716
• P(nuetral flow) = 0.3265
We chose a higher probability for the low
ENSO and this results in a higher
probability for high flows.
4000
3000
1000
2000
streamflow (cfs)
5000
Streamflow Scenarios
Given the forecast for climate signals as shown earlier, this is where
the flow for the following spring would be expected to fall.
Histogram for Annual Maximum Flow
0.0000 0.0005 0.0010 0.0015 0.0020 0.0025 0.0030 0.0035
Density
Histogram of Q
0
200
400
600
800
Q
Gamma distribution is the best fit based on KS test
200 400 600 800
Log-normal
Log Pearson Type III
EV1
Locfit
0
Discharge (cms.)
Quantile function estimates for Big Hole River Near Melrose, MT
0.0
0.2
0.4
0.6
0.8
1.0
Cumulative Probability
200 400 600 800
Discharge (cms.)
Quantile function estimates for Big Hole River near Melrose, MT
Log-normal
Log Pearson Type III
EV1
Locfit
0.80
0.85
0.90
Cumulative Probability
0.95
1.00
100 year flood estimates for
different methods.
Gamma
579 cms
Log-normal
665 cms
Log-Pearson III
574 cms
Extreme Value I
587 cms
LOCFIT
769 cms
Models of Monthly Streamflow
• Parametric AR(1) model
• Lag-1 Nonparametric K-NN model
Models simulate the months of May and June
because they exhibit interesting distributions.
Key statistics are mean, standard deviation, skew,
maximum, minimum, and correlation.
0.5
-1000
9000
0
0.6
0.7
1000
11000
2
0.4
-2000
8000
1
0.3
-3000
7000
2
0.2
-4000
6000
1
1
2
1
2
1
2
1
2
3500
4000
4500
-0.5
0.0
0.5
1200 1400 1600 1800 2000 2200 2400
3000
7000
0.5
0.6
0.7
1400
8000
0.8
2
0.4
1200
7500
1
0.3
900 1000
6500
2
0.2
800
6000
1
1
2
1
2
1
2
1
2
1200
0.0
1400
3000
0.5
1600
3500
1800
1.0
2000
4000
2200
1.5
2400
4500
ANY QUESTIONS ?