Inflow Forecasting forHydro Catchments

Ross Woods and Alistair McKerchar

NIWA Christchurch

Inflows

• Water flowing into hydro storages

• Usually measured by monitoring thelevels and outflows from hydro storages

Inflow = Outflow + (Rate of increase of storage)

• Variable at many timescales, differentfrom place to place

• Why make forecasts? How?

Nearly 80 years of levels for Lake Wakatipu

Data courtesy Contact Energy Ltd

309

310

312

313

level m

Jan-1924 Jan-27 Jan-30 Jan-33 Jan-36 Jan-39 Jan-42 Jan-45 Jan-48 Jan-51

309

310

312

313

level m

Jan-57 Jan-60 Jan-63 Jan-66 Jan-69 Jan-72 Jan-75 Jan-78 Jan-81 Jan-84

309

310

312

313

level m

Jan-87 Jan-90 Jan-93 Jan-96 Jan-99 Jan-02site 89135 L Wakatipu at Unadjusted Level

Main hydropower storagesMain hydropower storagesLake Storage Capacities

0

400

800

1200

1600

2000

Tau

po

Rot

oaira

Moa

wha

ngo

Wai

kare

moa

na

Cob

b

Col

erid

ge

Tek

apo

Puk

aki

Oha

u

Ben

mor

e

Haw

ea

Wan

aka

Wak

atip

u

Te

Ana

u

Man

apou

ri

Mah

iner

angi

Lake

Cap

acit

y (G

Wh

)

Forecasting for Three Timescales

• Weather systems: 0-3 days from now

• Seasonal forecasting: 1 week to severalmonths

• Interannual variability: Decadal-scale

Timescale: 0-3 days

• Why? How much electricity can we make?Flushing flows; High flow management

• What we need to know– What water is already in the catchment (river flows,

soil water, groundwater)?– How will weather system affect the rainfall, snowfall

and evaporative demand?

• With this information, we can use a model tocalculate snow melt, runoff from land into rivers,flow along rivers to lakes

• This can be done in more or less detail …

TOPNET - 1• Snowpack, canopy and root-zone

submodels over a shallow water table

Sat. zone

Rain, Evap

Subsurfaceflow

Recharge

ThroughfallSurfaceflow Sub-basin

outflow RiverNetwork

Othersub-basins:each one is

unique

Root zone

Canopy

Topo. controls

Snowpack

Melt

TOPNET - 2

• Sub-basin outflows are connected to rivernetwork routing

• Model gives results for every sub-basin andevery river reach, every hour

Balclutha (20,500 km2)Pomahaka R.

L. Roxburgh

L. L. HaweaHaweaL. Wanaka

L. WakatipuShotover Shotover R.R.

Alexandra

The Clutha River Catchment

The Catchment Model

RAMS Rainfall Forecasts overClutha Catchment

Balclutha

Alexandra

Real Time Forecasts for Alexandra - Nov. 1999using RAMS forecast on 20 km grid

0

200

400

600

800

1000

1200

10-Nov-99 12-Nov-99 14-Nov-99 16-Nov-99 18-Nov-99 20-Nov-99

Ale

xand

ra In

flow

(um

/h)

0

200

400

600

800

1000

1200

10-Nov-99 12-Nov-99 14-Nov-99 16-Nov-99 18-Nov-99 20-Nov-99

Ale

xand

ra In

flow

(um

/h)

Diamonds mark forecast time at 0800 each day

0

200

400

600

800

1000

1200

10-Nov-99 12-Nov-99 14-Nov-99 16-Nov-99 18-Nov-99 20-Nov-99

Ale

xand

ra In

flow

(um

/h)

Diamonds mark forecast time at 0800 each day

0

200

400

600

800

1000

1200

10-Nov-99 12-Nov-99 14-Nov-99 16-Nov-99 18-Nov-99 20-Nov-99

Ale

xand

ra In

flow

(um

/h)

Diamonds mark forecast time at 0800 each day

0

200

400

600

800

1000

1200

10-Nov-99 12-Nov-99 14-Nov-99 16-Nov-99 18-Nov-99 20-Nov-99

Ale

xand

ra In

flow

(um

/h)

Diamonds mark forecast time at 0800 each day

0

200

400

600

800

1000

1200

10-Nov-99 12-Nov-99 14-Nov-99 16-Nov-99 18-Nov-99 20-Nov-99

Ale

xand

ra In

flow

(um

/h)

Diamonds mark forecast time at 0800 each day

0

200

400

600

800

1000

1200

10-Nov-99 12-Nov-99 14-Nov-99 16-Nov-99 18-Nov-99 20-Nov-99

Ale

xand

ra In

flow

(um

/h)

Diamonds mark forecast time at 0800 each day

Successfulflood forecast

on a largeevent in a

slowly-respondingcatchment

Next Generation of Forecasts

• Operational Forecasting for catchments,with linked weather and hydrology models

• Operational Forecasting for Regions, withlinked weather and hydrology models

Timescale: 1 week – 6 months

• Why? Is there a hydro drought coming? Whendo we need pre-empt with thermal?

• What we need to know for inflow forecasts:– How much rainfall/snowfall?

– What temperatures?

• No deterministic answers: uncertainty is central.Aim to reduce the uncertainty

• This can be done in more or less detail …

MaxMin

75%25%

Median

Hermitage monthly rainfall (1931-1996)

Monthly rainfall (m

m)

0

500

1000

1500

JAN

FEB

MAR

APR

MAY

JUN

JUL

AUG

SEP

OCT

NOV

DEC

MaxMin

75%25%

Median

Lake Wanaka monthly inflows 1931-1995

Monthly inflow

(m3/s)

0

200

400

600

800

JAN

FEB

MAR

APR

MAY

JUNE

JULY

AUG

SEPT

OCT

NOV

DEC

Hermitage monthlyHermitage monthlyrainfallrainfall

Lake WanakaLake Wanakamonthly meanmonthly meaninflowsinflows

SeasonalPatterns

The Past as a guide to the Future1. It is common to forecast the performance of energy systems

over the next season by starting from the current lakestorage, and then using weekly inflow sequences fromeach of the last 70 years of inflows. This lets us sample thevariability of inflows at this time of year.

BUT, the unspoken assumption is that each of the last 70years is equally likely in the next 3 months. Once we are ina strong El Nino or La Nina, some years are much morelikely than others.

2. Some energy system models use auto-regressive models toforecast streamflows: on average these might be useful, butextreme seasons have a different persistence structure

3. Use “ENSO”-streamflow or rainfall-streamflow forecasting

Lake Taupo Natural Spring Inflows 1935-1992

Spring SOI

Sp

ring

inflo

w (m

3/s

)

50

100

150

200

250

-30 -20 -10 0 10 20 30

Non-Outlier MaxNon-Outlier Min

75%25%

Median

Lake Taupo natural spring inflows 1935-1992

Spring SOI

Sp

ring

inflo

w (m

3/s

)

50

100

150

200

250

< -5 (-5,5] > 5

Clutha lakes summer inflow vs spring SOI

Spring SOI

Su

mm

er in

flow

(m3

/s)

0

200

400

600

800

1000

1200

-30 -20 -10 0 10 20 30

MaxMin

75%25%

Median

Clutha lake summer inflows vs spring SOI

Spring SOI

Inflo

w (m

3/s

)

0

200

400

600

800

1000

1200

5

NIWA National Climate Centre andNational Centre for Water Resources• Forecasts each month of the streamflows for the

coming 3 months www.niwa.co.nz/ncc• Based on: climate forecasts, current status of river

flow and soil moisture, expert knowledge andstatistical analysis of past data.

• Forecasts developed as the probability of “Abovenormal”, “Normal” and “Below normal” flows. Inthe long run each of these occurs 33% of the time.So random guesses have 33% hit rate. We havehad 40-45% hit rate over the last 3 years

• www.niwa.co.nz/ncwr - water quality, quantity,lakes, groundwater, …

Next Steps

• Develop new methods for forecasting climate1. Statistical weather generators2. Regional climate models

• Both of these are ways to generate manysequences of daily rain and temperature for thecoming season (conditioned on the current ocean-atmosphere status – ENSO etc)

• Either way, we need to convert these climateforecasts to inflows: can use the same model as forthe 0-3 day forecasting.

Topnet• Example of Regional Modelling in Northland

RiverNetwork

Sat. zone

Rain, Evap

Subsurfaceflow

Recharge

ThroughfallSurfaceflow Sub-basin

outflow

Othersub-basins:each one is

unique

Root zone

Canopy

Topo. controls

Snowpack

Melt

Timescale: Interannual• Why? Investment in new generation; Assessing the

expected performance of existing system• Fact: Inflows can be very different from one decade

to another• Up to 15% changes in South Island• Inflow differences caused mainly by rainfall

variability. This is climate variability (as opposed toclimate change: e.g. rise in temperature & rain inwest and south of South Island by 2050)

• Rainfall differences are caused by slow changes inocean-atmosphere circulation, called InterdecadalPacific Oscillation (IPO, PDO)

• 1945-1977, 1978-1999?, 1999?-

South Island West Coast Rainfall

Clutha River at Balclutha 1947/48 to 1998/99Clutha River at Balclutha 1947/48 to 1998/99(data courtesy Contact Energy Ltd)(data courtesy Contact Energy Ltd)

300

500

700

900

1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000

An

nu

al m

ean

flo

w (

m3 /

s)

Mean = 536 m 3/s

Mean = 612 m 3/s

Lake Te Anau inflow, 1931/32 to 2001/02Lake Te Anau inflow, 1931/32 to 2001/02(data courtesy FRST, Meridian Energy Ltd & The Market Place)

100

200

300

400

500

Jan-30 Jan-40 Jan-50 Jan-60 Jan-70 Jan-80 Jan-90 Jan-00

An

nu

al

me

an

in

flo

w (

m3/s

) 1977/78-1998/99

Mean =305 m3/s

1946/47-1976/77

Mean = 269 m3/s

1931/32-1945/46

Mean = 277 m3/s

Summary

• 0-3 days: weather system forecasting and a modelof catchment processes

• 1 week – 6 months: Effects such as ENSO can shiftthe distribution of inflows far from the “typical”distribution

• Interdecadal: Differences between decades of order15% - be careful which years of data you use!

• For these inflow forecasts to be useful, we needbetter links with energy system modellers, esp foradvice on how best to communicate inflowforecasts