cuahsi cyberseminar april 01, 11 · 2017-04-28 · climate forecasts and water management :...

Climate Forecasts and Water Management :

Opportunities and Challenges

Sankar Arumugam Department of Civil & Environmental Engineering

North Carolina State University, Raleigh

CUAHSI Cyberseminar – April 01, 11

Presentation Outline

• Forecast Producers and Consumers

– Perspectives on Risk: Supply and Demand

• Dynamic Risk Management Framework

– Water Contracts with reliability of supply

• Insights from Retrospective Analyses

• Opportunities and Challenges

– Uncertainty Reduction

– Scope for Nutrient Allocation

Large Scale Hydroclimatology & Water Management

Hydroclimate Dynamics

Diagnosis

Physical Significance

Assimilation

Climatic Indices Land Surface Indices

Hydrologic Fluxes Estimation

Modeling

Forecasting

Water Management

System Design

Impact/Assessment

Allocation/Operation

Understanding &

Monitoring of Large

Scale Hydroclimatic

Systems

General Circulation Model

“Downscaling”

Regional Climate Model

Hydrologic Model

Exogenous Climate Predictors

Statistical Model

Forecasts of Reservoir Inflows and Flood Flows

Forecasting Seasonal Streamflow volumes – Two Approaches






– Restriction Levels if forecasts go wrong

– Allocation Model conditioned on current storages

– Tariff, Compensation and Penalties



Climate induced Risk on Water

• Forecast Producers

– Climate Scientists and Hydrologists

– Express seasonal streamflow uncertainty as

terciles or as ensembles

• Forecast Consumers

– Water Managers and Reservoir Operators

– Often risk averse; No reward for using forecasts

– Difficulty interpreting/relating forecasts to releases

– Often manages the system based on rule curves

(based on long-term operation)

– Need not quantify conditional risk

Attributing Climate Risk on Water Management

• Needs to be prognostic

• Relate water supply risk to demand risk

– Assigning reliability of supply for yield types

– Estimate reliability as part of allocation process

• Include compensations and penalties

– if forecasts go ‘wrong’ (false alarms or missed

targets)

• Tariff if specified yield is delivered

– Could act as an insurance scheme

• Limited Skill or Skill only in few seasons

– Include probabilistic constraint on storages






– Restriction Levels if forecasts go wrong

– Allocation Model conditioned on current storages

– Tariff, Compensation and Penalties


• Challenges and Way forward

Dynamic Water Allocation Model - Formulation

• Reservoir Inflow Forecasts Ensembles

• Water Contracts Specification

• Water Allocation Model for Bulk Sector

contracts

– Simulation – Optimization Model

Sankarasubramanian et al., WRR, 2009.

Water Contracts Specification

Specification Notation Illustration

Duration T 1 Year or 3 months

Release Ri, ith

use

5000 mm3 over T

Monthly Release Rti = bti bti = 1/12 if uniform

supply

Tariff (Value) i $1000

Contract

Reliability

(1-pfi) 90% for municipal use

What if Forecasts go Wrong?

Specification Notation Illustration

Restriction Wi* 500 hm3 - Industry

Restriction level j = 1, nl 1st– Lawn, Golf

2nd – Industry & above

Restriction

Fraction

aij 1st– 0 hm3; 2nd – 500

hm3

Compensation gij 1st– $100; 2nd – $300

Contract Failure vi $ 2500 if Wi > 500 hm3

Water Allocation Model • Modify the Allocation Rule – Maximize the

annual/seasonal net value from releases

conditioned on the forecasts

O =

– Ri – Release (Yield) for use ‘i’

– i - Unit value of water for delivery – Tariff function

– N – Number of uses (Contracts)

– Wil – Restriction at level ‘l’ with compensation gil

– i- Contract penalty if wi > wi*

*

1 1 1 1

( ) ( )l

i

nn n n

i i il il i i

i i j i

E R W W W g

Objective Function:

Maximize the net value from contracts and surplus water provision

Subject to

•P(Wi Wi*) pfi - Contract Level Constraint

For Industry, at a reliability (1-pfi) = 90%, P(Wi 500) 0.10

•P(STST*) ps - End of the Year Storage Constraint

For ST* = 500 hm3 and (1-ps) = 75%, P(STST*) 0.25

•P(RLj) prj - Reservoir Level Constraint

For Restriction Level 1, P(RL1) < 0.25; pr1 – Restriction level 1 prob.

Similarly for level 2, P(RL2) < 0.10; pr2 – Restriction level 2 prob.

Reservoir Simulation (for each ensemble ‘k’)

• Inflow Forecast : qtk; t=1…,T; k=1,…,N

• Continuity Equation: t=1,2, …, T

• SDt = -St | St < 0 (Account the Deficit)

• Rti=btiRi (Target Release for each user)

• Evaporation :

n

i

titttt REqSS1

1

2)2/)(( 11 tttt SSE






– Ceara, North East Brazil

– Angat, Philippines

– Falls Lake, North Carolina


TALE OF TWO RIVER BASINS Jaguaribe-Metropoilitan

HydroSystem, Ceara, Brazil

Angat Reservoir, Philippines

Angat Reservoir during the 1997-98 El Nino

Hydroclimatology of the Basins JMH, Ceara, NE Brazil

• Semi-Arid

• One Rainy season,

Rest Zero flows

Angat, Philippines

• Humid

• Two Rainy Seasons

(Jun-Sep) & (Oct-Dec) Seasonality of Oros Inflow

0

50

100

150

200

250

300

350

400

1 2 3 4 5 6 7 8 9 10 11 12

Month

Flo

w (

m^

3/s

)

Mean

Median

Quantile (75)

Quantile (25)

Quantile (90)

Quantile (10)

Seasonality of rain determined by N-S migration of the ITCZ

0

50

100

150

200

250

300

350

JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC

Month

Str

eam

flo

w (

in h

m3)

0

50

100

150

200

250

300

350

400

450

500

Rain

fall (

mm

)

Streamflow

Rainfall

JJAS – 30%

OND – 46%

3-months lag correlation

(Nino3.4,QJJAS) = -0.20

(Nino3.4,QOND) = -0.51

Predictors Correlation

Nino 3.4

Dipole

-0.31

-0.42

Water Management Context JMH, Ceara, NE Brazil

Angat, Philippines

Agriculture

Julyt Julyt+1 Jant+1 Dect+1

Seasonal Variation Rain/Flow

Municipal and Industrial

Priority Uses

• Municipal

• Industrial

• Agriculture

ANGAT H.E. PLANT

150

160

170

180

190

200

210

220


EL

EV

AT

ION

(m

)

1996 1997 1998 1999 2000 2001 UPPER LOWER

Priority Uses

• Municipal

• Industrial

• Agriculture

• Hydropower

Storage/Annual

Demand Ratio: 0.72

Storage/Annual

Demand Ratio: 5.1

Hydroclimatology of the Basins JMH, Ceara, NE Brazil Angat, Philippines

• ECHAM4.5 Precipitation

Forecasts Forced with

Persisted SST conditions

Seasonality of rain determined by N-S migration of the ITCZ

Predictors Correlation

Nino 3.4

Dipole

-0.31

-0.42

(PF,Qobs):0.51

ONDJF

Climate Information based Streamflow Forecasts JMH, Ceara, NE Brazil

• Predictors: Nino3.4 & Atlantic Dipole

• Semi-parametric Resampling algorithm

Angat, Philippines

• ECHAM 4.5 ONDJF Forecasts with persisted SSTs

• Parametric Regression

0

50

100

150

200

250

300

350

400

1968 1978 1988 1998

Year

Str

eam

flo

w (

MC

M)

ONDJF-obs

ONDJF-pred

(Qpred,Qobs):0.58

0

25

50

75

100

125

150

175

200

225

1970 1975 1980 1985 1990 1995

Year

Flo

w (

m3/s

)

obsFirst QuartileMedianThird QuartileMax

(Qobs,Qpred) = 0.75

De Souza et al., 2003

Sankarasubramanian et al., 2008

Retrospective Analyses on Water Allocation

JMH, Ceara, NE Brazil:1990 – 2000;Annual Demand

-20

0

20

40

60

80

100

120

1991 1992 1993 1994 1995 1996 1997 1998 1999 2000

Year

Fo

recas

t -

Zero

In

flo

w Y

ield

(h

m3)

0

10

20

30

40

50

60

70

Ob

se

rve

d F

low

s (

m3/s

)

Agriculture

Industrial

Flow

Forecast


Oros - System Losses

-100

-90

-80

-70

-60

-50

-40

-30

-20

-10

0

10

1991 1992 1993 1994 1995 1996 1997 1998 1999 2000

Year

Fo

reca

st

- Z

ero

(in

hm

3)

0

10

20

30

40

50

60

70

Ob

se

rve

d F

low

s (

in m

3/s

)

Evaporation+spill

Storage in July

Observed Flows

Angat, Philippines

The Angat Reservoir during the 1997-98 El Nino (September 12,

1998 – RWL = 158.15 m AMSL)

ANGAT H.E. PLANT

150

160

170

180

190

200

210

220


EL

EV

AT

ION

(m

)

1996 1997 1998 1999 2000 2001 UPPER LOWER

IRI Forecast for OND 1998

Retrospective Analyses on Water Allocation Angat, Philippines: 1987-2001 ONDJF Allocation

0

100

200

300

400

500

600

1987 1989 1991 1993 1995 1997 1999 2001

Year

Oco

tber

All

ocati

on

(h

m3)

0

50

100

150

200

250

300

350

400

Ob

se

rve

d I

nfl

ow

(h

m3

)

Municipal

Irrigation

hydropower

M&I target

Irrigation target

Observed Flow

Performance - updated Forecasts

0

0.1

0.2

0.3

0.4

0.5

0.6

Target Month

Co

rre

lati

on

OctoberNovemberDecemberJanuaryFebruary

Oct Nov Dec Jan Feb

Sankarasubramanian et al., 2008, J.hydromet

Angat- Spill Volume (Max Hydro)

0

50

100

150

200

250

300

350

400

450

1987 1989 1991 1993 1995 1997 1999 2001

Years

Sp

ill

Vo

lum

e (

in h

m3

)

0

50

100

150

200

250

300

350

400

Ob

se

rve

d In

flow

ActualUpdated ForecastOctober ForecastObserved Inflow

Sankarasubramanian et al., 2009, JAMC

Retrospective Analyses on Water Allocation

Angat, Philippines: 1987-2001 ONDJF Allocation

0

200

400

600

800

1000

1200

1400

1987 1989 1991 1993 1995 1997 1999 2001

Year

Hyd

rop

ow

er

Ge

ne

rate

d (

in G

WH

)

0

50

100

150

200

250

300

350

400

Ob

se

rve

d In

flow

ActualUpdated ForecastOctober ForecastObserved


Meeting February Target Storage

0

100

200

300

400

500

600

700

800

Oct Nov Dec Jan FebForecast Issued

Fe

bru

ary

Sto

rag

e (

hm

3)

198919911997Target Storage


Utility of Forecasts in improving

hydropower generation

Maurer and Lettenmaier, 2004, Jclim.

Utility of Forecasts as storage/inflow

Maurer and Lettenmaier, 2004, Jclim.

Utility of Forecasts As Storage/Demand

0

1

10

0.0 1.0 2.0 3.0 4.0 5.0 6.0

Storage/Demand

% Im

pro

vem

en

t

Correlation = 1.0

Correlation = 0.75

Correlation = 0.5

Angat Oros


Insights from Retrospective Analyses

• For seasonal forecasts to be useful

– Initial storage should constrain the allocation

– If not, 100% reliability; Most systems belong to this

category, since reservoirs are designed to reduce

the variability and ensure reliability of supply.

• Use end of season target storage constraint

– If initial storage does not constrain allocation

– If skill is good only during a particular season

– To enforce restrictions for below normal conditions

– To reduce spillage and increase allocation

(primarily hydropower) for above normal conditions

• Perspectives from Forecasting

– Update the forecasts within the season (very

beneficial for hydropower systems)

– Multimodel climate forecasts are better, since it

reduces aggressiveness of individual models

• Forecasts are more useful than climatology

– Within year storage systems (typically humid

basins) than over year (arid basins) systems

– Reducing system losses (spill and evaporation)

– Systems with low storage/annual demand ratio

– Multiple uses constraining the allocation process

Insights from Retrospective Analyses


• Seasonal Climate Forecasts Applications

for Water Management

– Multimodel Combination

• Well calibrated forecasts improves confidence

– Nutrient Loadings Forecasts

• Scope for Extending this framework for Nutrient

Allocation

– Water and Energy Systems Management

• Covariability between precipitation and temperature

Why multimodel combination reduces uncertainty?

Weigel et al., QJRMS, 2008.

Mu

ltim

od

el c

om

bin

ati

on

-U

nc

ert

ain

ty R

ed

uc

tio

n

Mu

ltim

od

el c

om

bin

ati

on

-U

nc

ert

ain

ty R

ed

uc

tio

n

Select categorical climate forecasts/predictions, ,

m

i tQ , where m=1,2...,M

(M=8) denotes the model index including climatology, with i=1, 2..., N (N=3) representing

the categories in year ‘t’ with t=1,2…, T (T= 46 years).

Obtain the squared error, SEtm, between the ensemble mean and the observed

precipitation/temperature for each year for all GCMs.

Based on DJF Nino3.4 (X) as the predictor, compute the distance between the current

Nino3.4, Xt, and the rest of the Nino3.4, Xl, where l=1,2..., T-1 (leaving ‘t’ year out).

Choose the number of neighbors, K, and compute Mean Square Error (MSE) over ‘K’

neighbors using: ( )

1

1 Km m

t, K j

j

SEK

Compute weights ( ,

m

t Kw ) for each model for each time step based on MSE over ‘K’

neighbors

M

m

m

Kt

m

Ktm

Ktw

1

,

,

,

/1

/1

Obtain categorical forecasts/predictions of multimodels, ,

MM

i tQ , based on individual models’

weights ( ,

m

t Kw ): , ,

1, ,

,

1

Mm m

t K i tMM mi t K M

m

t K

m

w Q

Q

w

Get the skill of the multimodel forecasts/predictions, , ,,MM MM

t K t KSE RPS (Rank Probability Score)

(Devineni et al., WRR, 2008)

Reliability of Multimodel Forecasts – BN Precipitation

(Devineni and Sankarasubramanian, Monthly Weather Review, 2010)

Why conditioned multimodel better?

(Devineni and Sankarasubramanian, Monthly Weather Review, 2010)

0

10

20

30

40

50

60

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1991 1993 1995 1997 1999 2001 2003 2005

Ob

serv

ed

Str

eam

flo

w (

cm

s)

Pro

b(S

T<

ST

* )

Year

Regression

Resampling

Multimodel

Climatology

Observed

0.33 percentile

0.66 percentile

Probability of Meeting the Target Storage

Golembesky et al., 2008

Invoking Restriction based on Target Storage Probabilities

0

20

40

60

80

100

120

140

0

10

20

30

40

50

60

70

80

1991 1992 1993 1994 1997 1998 2002 2005

Ob

serv

ed

S

tream

flo

w,

cfs

Restr

icti

on

(%

)

Year

Resampling

Multimodel

Regression

Observed

(a)

0

20

40

60

80

100

120

140

0

2000

4000

6000

8000

10000

12000

1991 1992 1993 1994 1997 1998 2002 2005

Ob

serv

ed

S

tream

flo

w,

cfs

Dif

fere

nce i

n S

ep

tem

ber

Sto

rag

e, acre

-feet

Year

Resampling

Multimodel

Regression

Observed

(b)

Golembesky et al., 2008

Water Quality Management

• Water Quality Trading

– Successful programs in Neuse and Tar basins

– Association - Municipal and Industrial discharges

– Target nutrient loadings – Point and nonpoint

– Trading – Point to point and point to nonpoint

– Point to nonpoint – Cost share program for BMP

• Opportunity to use Forecasts

– Pre-season estimates of loadings from runoff

– Optimal loadings between point and nonpoint

– Forecasts work better with contracts/trading

Initial Results on Nutrient Forecasts

Covariability in precipitation and temperature

• Positive (Negative) Correlation during winter

– Tennessee Valley, New England (Upper Mid West)

• Negative Correlation during summer

– Midwest, Midwest and part Southeast

Winter Summer


• Seasonal to Interannual Climate Forecasts

– Uncertainty Reduction and Reliable Forecasts

– Appropriate Policy Instruments – e.g., water

contracts with specified reliability, penalties and

tariffs.

– Climate Information based River Basin

Management

• Seasonal Nutrients Allocation and Water Quality Trading

• Combined Water and Energy Systems Management

Acknowledgements

• IRI, Columbia University, Palisades

– Upmanu Lall, Naresh Devineni

• FUNCEME, Ceara, North East Brazil

– Assis Filho De Souza

• PAGASA, Manila, Philippines

– Susan Espinueva

• NC State University

– Kurt Golembesky, Jeseung Oh

Related Publications • Sankarasubramanian, A., U.Lall, F.D.Souza Filho, A.Sharma (2009), Improved

Water Allocation utilizing Probabilistic Climate Forecasts: Short-term Water

Contracts in a Risk Management Framework, Water Resources Research.

• Sankarasubramanian, A., U. Lall, and S. Espunevea (2009), Utility of

Operational Streamflow Forecasts in Improving within-season Reservoir

Operation, Journal of Applied Climatology & Meteorology.

• Golembesky, K., and A. Sankarasubramanian, and N. Devineni (2009), Improved

Management of Falls Lake Reservoir during the Summer Season using Climate

Information based Monthly Streamflow Forecasts: Role of Restrictions in

Water supply and Water Quality Management, Journal of Water Resources

Planning and Management, 2009.

• Devineni, N., A. Sankarasubramanian, and S. Ghosh (2008), Multi-model

Ensembling of Probabilistic Streamflow Forecasts: Role of Predictor State

Space in Skill Evaluation, Water Resources Research, 44, W09404,

doi:10.1029/2006WR005855.

• Sankarasubramanian, A., U. Lall, and S. Espuneva (2008), Role of

Retrospective Forecasts of GCMs Forced with Persisted SST anomalies in

Operational Streamflow Forecasts Development, Journal of Hydrometeorology,

9(2), 212-227.

cuahsi cyberseminar april 01, 11 · 2017-04-28 · climate forecasts and water management :...

Documents