Precipitation elasticity of streamflow in
catchments across the world
Francis Chiew, Murray Peel, Tom McMahon, Lionel Siriwardena
(CSIRO, Australia & University of Melbourne, Australia)
[Harry Lins, Richard Vogel, Mike Bonell, Wolfgang Grabs et al.]
[WMO/UNESCO WCP-Water]
FRIEND 2006, Havana Cuba, 26 Nov - 1 Dec 2006
www.csiro.au
Presentation outline
•
Climatic variability and climate change
•
Hydrologic sensitivity to climate – modelling approaches and topdown data driven methods
•
Global monthly/annual precipitation, temperature and streamflow
dataset for over 500 catchments across the world
•
Precipitation elasticity of streamflow
•
Summary and conclusions
Climatic variability and climate change
Temperature has been rising
Hydroclimatic variability occurs
over different time scales
1910
1930
1950
1970
1990
2010 2030
What will the future climate look like?
Large uncertainties in climate change projections
% change in mean annual rainfall
for 2030 (relative to 1990)
-40% -20%
0 +20%
+40%
Modelling hydrologic sensitivity to climate
Rainfall
PET
Run model with
“future” rainfall input
Scale historical
rainfall by change
Calibrate hydrological model
Hydrological Runoff
Model
winter
Rainfall
summer
PET
+15%
0
-15%
Range of change
estimated by GCMs (2030
relative to 1990)
Hydrological Runoff
Model
Climate change impact on runoff
• choice of hydrological model
• calibration method and criteria used
• method used to obtain future
climate time series.
40
Monthly runoff (mm)
Results are dependent on
Present runoff
2030 runoff
30
20
10
0
J
F
M A M J
J
A
S
O N D
Estimating precipitation elasticity of streamflow
directly from historical time series data
• Precipitation elasticity of streamflow, εp, is defined as proportional change in
mean annual streamflow divided by proportional change in mean annual
precipitation.
• Sankarasubramaniam et al. nonparametric estimator
εP = median
Qt - Q
P
Pt - P
Q
• Other estimates of εP
εP = ρPQ
CVQ
CVP
[Qt - Q] = εP [Pt - P]
• The above methods estimate εP directly from a set of concurrent annual
catchment precipitation and streamflow data (i.e., they are model independent).
They are particularly useful for global studies, where it is difficult to define
hydrological models appropriate for large parts of the world, and to obtain the
data required to run such models.
εP values
– hydrological model versus nonparametric estimator
6
εP (Non-parametric)
5
4
3
2
1
εP (SIMHYD model)
0
0
1
2
3
4
5
εP (SIMHYD model)
6
<2
2 – 2.5
2.5 – 3
>3
Global hydroclimate data used in this study
• Concurrent annual (“water year”) precipitation, temperature and streamflow time
series for 521 catchments.
• Data length varies from 23 to 64 years and catchment areas from 100 to 76,000
km2 (10th to 90th percentile).
Development of global hydroclimate dataset (1)
• Streamflow data are drawn from global database of monthly streamflow data for
over 1,200 catchments described in Peel et al. The streamflow data are believed
to be unregulated over the period of record in the database.
• Source of precipitation and temperature data is the Global Historical Climatology
Network (GHCN), which contains precipitation and temperature data for over
20,000 and 7,000 stations respectively.
• To compile catchment-average monthly precipitation data, catchment boundaries
are constructed from the USGS HYDRO1k DEM. The catchment is used only if
the area defined by the boundary is within 5% of the published catchment area.
• For each catchment, precipitation stations within 200 km of the catchment
boundary (with more than 50% data over the period of streamflow data) that
contribute to the Thiessen weighting are used to estimate the catchment-average
monthly precipitation time series. Missing precipitation data are infilled with data
from another station with the highest correlation.
• For temperature, stations within 500 km of the catchment boundary are used.
Development of global hydroclimate dataset (2)
• Biggest sources of error in the data are:
- streamflow measurements (flow-stage rating); and
- likelihood of more precipitation stations located in lower parts of the catchment.
• Other main sources of error include:
- catchment may not be unregulated as believed and/or may have undergone
land use change;
- poor precipitation data and/or poor infilling of missing precipitation data; and
- insufficient precipitation stations to represent the catchment.
• Criteria used to select catchments for this study:
- there must be less than 5% of “poor infilled data”;
- for catchments >5,000 km2, must have at least two precipitation stations
representing more than 70% of the Thiessen weighting;
- at least 20 years of concurrent annual precipitation and streamflow data;
- runoff coefficient must be less than 1.0
- εP must be greater than zero.
Estimates of εP
• 80% of εp estimates are between 1.0 and 3.0 (i.e., a 10% change in mean annual
precipitation results in a 10 to 30% change in mean annual streamflow).
• Higher εp values (> 2) are in Australia and southern and western Africa.
• Lower εp values (< 2) are in mid and high latitudes of the Northern Hemisphere.
Range of εP values in major climate zones
Climate zone
(Koppen)
εP
εP
Number of
catchments
Median
range
Tropical (A)
Very wet (Af, Am)
Moderately wet (Aw)
79
20
59
1.7
1.2
2.0
(0.8 - 3.1)
(0.8 - 1.9)
(0.9 - 3.3)
Arid (B)
Cold arid (BWk, BSk)
Warm arid (BWh, BSh)
45
32
13
1.8
1.6
2.0
(0.4 - 2.9)
(0.4 - 3.1)
(0.5 - 2.5)
Temperate (C)
Wet winter (Csa, Csb, Csc)
Wet summer (Cwa, Cwb, Cwc)
No seasonality (Cfa, Cfb, Cfc)
262
32
35
195
1.9
2.0
1.8
1.9
(0.9 - 3.1)
(0.9 - 3.4)
(0.8 - 2.8)
(1.0 - 3.1)
Cold (D)
135
1.1
(0.5 - 1.9)
• Lower εp values in cold climates.
• Lower εp values in very wet tropical climates.
• No clear distinguishment of εp values in other climates.
εP versus runoff coefficient
5
4
εP
3
2
1
0
0
0.2
0.4
0.6
0.8
1.0
Runoff coefficient
• Strong inverse relationship between εp and runoff coefficient.
• Precipitation-runoff process is nonlinear – same absolute change in runoff for a
given absolute change in precipitation would be reflected as higher εP in
catchments with low runoff coefficient.
• In most cases, upper limit of εP is the inverse of runoff coefficient.
εP versus mean annual streamflow and precipitation
5
4
εP
3
2
1
0
0
1000
2000
3000
4000
Mean annual streamflow (mm)
5
4
εP
3
2
1
0
0
1000
2000
3000
4000
Mean annual rainfall (mm)
5000
εP in Australian catchments
6
5
εP
4
3
2
1
0
0.0
6
0.2
0.4
0.6
Runoff coefficient
0.8
1.0
5
4
εP 3
2
1
0
• Clear spatial differences in εp values.
• Very strong correlation between εp and
runoff coefficient.
• Strong correlation between εp and mean
annual streamflow and rainfall.
0
1000
2000
3000
1000
2000
3000
Mean annual streamflow (mm)
6
5
εP
4
3
2
1
0
0
Mean annual rainfall (mm)
εP versus mean annual temperature
5
4
εP
3
2
1
0
-20
-10
0
10
20
30
Mean annual temperature (OC)
• Lower εp in cold climates.
• Low εP in cold climates because of high runoff coefficient (low energy available for ET).
• Where there is significant/permanent snowpack, annual streamflow is also dependent on
temperature which governs the amount of snowmelt and over-year snow storage, which
is not considered in the εP estimator used here.
Summary and Conclusions
•
The precipitation elasticity of streamflow, εP, is estimated for 521 catchments
across the world, using a nonparametric estimator.
•
The nonparametric estimator estimates εP directly from concurrent historical
annual precipitation and streamflow data. It is therefore particularly useful
for comparative global studies.
•
The key results are:
- εp generally range from 1.0 to 3.0 (i.e., a 10% change in mean annual
precipitation results in a 10 to 30% change in mean annual streamflow)
- there is a strong inverse relationship between εP and runoff coefficient
- εP is lower in colder climates.
•
Possible improvements to analyses/results:
- Better dataset?
- Consideration of combined precipitation/temperature or precipitation/PET
elasticity of streamflow.
Thank You
www.csiro.au