Transcript Sub-daily rainfall in Australia

School of Civil, Environmental and Mining Engineering
Wednesday, 4th April 2012
Changes to sub-daily
rainfall in Australia
Dr Seth Westra
Life Impact | The University of Adelaide
Slide 1
Slide 2
Slide 3
Slide 4
Presentation overview
• Part 1: The sub-daily rainfall dataset in Australia
• Part 2: The observed relationship between
temperature, humidity and rainfall intensity
• Part 3: Detection of trends in sub-daily rainfall
• Part 4: Towards a downscaling algorithm for sub-daily
rainfall
• Part 5: Evaluating regional climate model (WRF)
performance using the diurnal cycle of sub-daily
precipitation
Slide 5
Part 1: Australian rainfall record
• More than 19000 daily precipitation stations (read at
9am daily)
• More than 1500 pluviograph stations (6-minute
resolution)
Pluviograph (sub-daily)
Slide 7
Daily (only locations > 100 years)
Slide 8
Australian rainfall record – record length
Pluviograph
Daily
Slide 9
Part 2: Link between temperature and extreme
rainfall
Extreme rainfall will
scale at C-C rate of
~7%/C or “super C-C”
rate of ~15%/C
Slide 10
Methodology
• Reproduce this work using Australia-wide data:
– 137 long pluviograph records (average length 32
years, with average of 6% missing)
– Mean and maximum daily 2m air temperature
extracted for each wet day
– Data grouped into 15 bins by temperature – and
different percentile (e.g. 50, 99%ile) rainfall
extracted in each bin
– Where available, relative humidity also extracted
Methodology
Hardwick-Jones, R., Westra, S. & Sharma,
A., 2010, “Observed relationships between
extreme sub-daily precipitation, surface
temperature, and relative humidity”,
Geophysical Research Letters, 37, L22805
Slide 12
60-minute rainfall intensity against
average daily temperature
Blue = 99 percentile rainfall (representing behaviour of ‘extremes’)
Red = 50 percentile rainfall (representing behaviour of ‘average’
events)
60-minute rainfall intensity against
average daily temperature
10
10
ALICE SPRINGS AIRPORT 015590
2
10
Maximum daily 60-minute precipitation (mm)
Maximum daily 60-minute precipitation (mm)
10
1
0
10
15
20
25
Mean Daily Temperature (C)
30
10
10
DARWIN AIRPORT 014015
2
1
0
24
25
26
27
28
29
Mean Daily Temperature (C)
30
31
Blue = 99 percentile rainfall (representing behaviour of ‘extremes’)
Red = 50 percentile rainfall (representing behaviour of ‘average’
events)
Scaling of 99th percentile maximum daily 60-minute burst
with mean daily surface temperature


1
5
0
E
E
5
3
1

165  E
E
0
12
School of Civil, Environmental and Mining Engineering
Wednesday, 4th April 2012
NORTH

5
1 S
EST
CENTRAL- W

13% to 20%.C-1
SOUTH
EAST
30 S
7% to 13%.C-1
2% to 7%.C-1
-2% to 2%.C-1
-7% to -2%.C-1

45 S
-13% to -7%.C-1
Life Impact | The University of Adelaide
-17% to -13%.C-1
Regional scaling of 99th percentile
60-minute burst precipitation with surface temperature
Precipitation Depth (mm)
3
School of Civil, Environmental and Mining Engineering
10
East Region
North Region
SouthWednesday,
Region 4th April 2012
Central Region
Clausius-Clapeyron Relationship
2
10
10
10
1
0
0
5
10
Life15
Impact | The20
University of25
Adelaide
Temperature (C)
30
35
40
Regional scaling of 99th percentile daily precipitation
with surface temperature
Precipitation Depth (mm)
4
School of Civil, Environmental and Mining Engineering
10
East Region
North Region
SouthWednesday,
Region 4th April 2012
3
Central Region
10
Clausius-Clapeyron Relationship
10
10
10
2
1
0
0
5
10
Life15
Impact | The20
University of25
Adelaide
Temperature (C)
30
35
40
Slide 18
Slide 19
Does relative humidity stay constant with temp?
DARWIN AIRPORT 014015
PERTH AIRPORT 009021
100
100
95
Mean daily relative humidity (%)
Mean daily relative humidity (%)
90
80
70
60
50
40
90
85
80
75
70
65
60
30
20
5
10
15
20
25
Mean daily surface temperature (C )
Jun to Nov
Dec to May
55
Jun to Nov
Dec to May
30
50
22
35
ALICE SPRINGS AIRPORT 015590
32
100
90
90
Mean daily relative humidity (%)
Mean daily relative humidity (%)
26
28
30
Mean daily surface temperature (C )
SYDNEY (OBSERVATORY HILL) 066062
100
80
70
60
50
40
30
80
70
60
50
40
20
10
24
Jun to Nov
Dec to May
5
10
15
20
25
30
Mean daily surface temperature (C )
Jun to Nov
Dec to May
35
40
30
5
10
15
20
25
Mean daily surface temperature (C )
30
Summary of temperature scaling work
• Clear scaling of rainfall with temperature across
Australia
• Scaling depends on duration of storm burst, and
exceedance probability
• Scaling also depends on atmospheric temperature –
negative scaling with high temperatures!
– Likely to be due to access to atmospheric moisture
• BUT: Does a historical scaling relationship imply
similar future changes?
Slide 21
Part 2: Detection of trends in Australian rainfall
• We wish to detect whether there are trends or
other types of climatic non-stationarity in
extreme precipitation data
• Consider the following hypothetical example:
– ‘Extreme’ precipitation will scale at a rate of
7%/C in proportion to the water holding capacity of
the atmosphere
– Global warming trend has been ~0.74C over the 20th
century
– Therefore would need to be able to detect a ~5%
change
Motivation
• Assuming 50 years of data, such a trend would be
detected at the 5% significance level in only 8% of
samples (and a negative trend detected in 2% of
samples!)
What is a max-stable process?
• Formal definition: suppose for
, i = 1,..., n,
are independent realisations of a continuous process.
If the limit:
exists for all s with normalising constants an(s) and
bn(s), then
is a max-stable process.
• Spatial analogue of multivariate extreme value
models, which accounts for both data-level
dependence and parameter-level dependence.
– Distinct from ‘Spatial GEV’ models which only
account for parameter-level dependence.
Illustration of max-stable process
• The ‘storm profile’ model:
Benefits for trend detection
• Can improve the strength of the trend that can be
detected (given by value of parameter ‘β1’),
depending on the amount of spatial correlation.
Application to Australian rainfall data
• Of Australia’s ~1400 sub-daily records,
we selected the 35 most complete
stations with records from 1965-2005.
– Extracted annual maximum data for
6-minute through to 72 hour storm
bursts
• Also considered high quality daily data
from 1910 to 2005
Application to Australian rainfall
• Trends in annual
maximum 6-minute
rainfall
– Blue/red indicates
increasing/decreasing
trend
– Filled circles indicate
statistically significant
at the 5% level
Is there an increasing trend in
east-Australian precipitation?
Sensitivity to gauge changes
• Many sub-daily stations had at least one gauge
change over the record, usually from Dines
pluviograph to TBRG
• Tested sensitivity by extracting any ‘step change’ in
the year the gauge change occurred, and then refitting the trend.
• Did not make any significant difference to the
strength of the trends in the previous slide
Summary of trend detection work
• Max-stable processes provide an elegant way of
detecting non-stationarity in hydroclimatic data
– Enables substitution of ‘space-for-time’ while
accounting for spatial dependence
• In east-Australia an increasing trend in sub-daily
(particularly sub-hourly) precipitation data could be
detected, but not for daily data
• This would suggest that sub-daily precipitation is
increasing much more quickly than expected
• Also highlights that daily data cannot be used
for inference at shorter timescales
Westra, S. & Sisson, A., 2011, “Detection of non-stationarity in precipitation extremes using a max-stable
process model”, Journal of Hydrology, 406
Part 4: Disaggregating from daily to sub-daily
rainfall under a future climate
• We have shown that the scaling of rainfall with
atmospheric temperature depends on storm burst
duration, exceedance probability, and moisture
availability
– How can this be used for estimating change in subdaily rainfall under a future climate?
• Various techniques are available for downscaling daily
rainfall under a future climate
– We have developed an algorithm to disaggregate
from daily to sub-daily rainfall under a future
climate.
Slide 32
Importance of seasonality on daily to sub-daily
scaling
• Scaling from daily to sub-daily rainfall strongly
depends on atmospheric temperature
Slide 33
Plotting against both temperature and day of year
• BUT – most of the annual variation can actually be
attributed to atmospheric temperature!
Slide 34
Influence of location – before and after
regressing against atmospheric temperature
Slide 35
Considering a broader range of
atmospheric variables...
Variable
Abbreviated name
500, 700 and 850hPa
temperature
Dew point temperature
Relative humidity
Pressure reduced to
mean sea level
850hPa wind strength
and direction
t500, t700, t850
Daily mean, maxima,
minima and/or diurnal
range
mean, maxima, minima,
range
mean
Td
RH
prmsl
maxima
mean and maxima
mean and minima
Degrees Celsius
Percentage (%)
Pa
wnd850_str,
wnd850_theta
mean
10m wind strength and
direction
wnd10m_str,
wnd10m_theta
mean
500 and 850hPa
geopotential height
z500, z850
mean
(derived from u and v
components of wind;
units of m/s)
(derived from u and v
components of wind;
units of m/s)
Geopotential meter
(gpm)
2m surface temperature tmp2m
Units
Degrees Celsius
Degrees Celsius
Algorithm
Assume we have future sequences of daily rainfall
available (e.g. from a statistical or dynamical downscaling
algorithm), as well as atmospheric covariates
1. Given a future daily rainfall amount and associated
atmospheric covariates (e.g. temperature, relative
humidity, geopotential height...)
2. Find days in the historical record which have a ‘similar’
atmospheric state and daily rainfall amount and also
the complete sub-daily rainfall sequence
3. Sample from one of those days
Slide 37
A disaggregation algorithm for downscaling sub-daily rainfall
Westra, S., Evans, J., Mehrotra, R. & Sharma, A., “Disaggregating from daily to sub-daily rainfall under a
future climate”, submitted to Journal of Climate
Slide 38
Summary of sub-daily disaggregation
• Disaggregation algorithm is a simple ‘analogues’
based approach for understanding sub-daily rainfall
behaviour under a future climate
• Requires daily downscaling information, but such
information is often readily available
• Shows substantial changes can be expected at hourly
or sub-hourly timescales.
Slide 39
Part 5: Diurnal
cycle of modelled
and observed
rainfall
• Good performance of a
dynamical model in
capturing the diurnal
cycle provides a
positive indication that
the processes of subdaily precipitation are
correctly represented.
Evans, J. & Westra, S., “Investigating the mechanisms of diurnal rainfall variability using a Regional Climate Slide 40
Model”, submitted to Journal of Climate
Diurnal cycle of different precipitation generating mechanisms
Slide 41
Conclusions and ongoing work
• Evaluated scaling relationships of sub-daily rainfall
and found strong dependence on temperature and
atmospheric moisture
• Trend detection work also shows increasing trends in
fine time-scale (particularly sub-hourly) rainfall
– Significant implications for urban flood risk and
risk of flash flooding
• Developed statistical disaggregation algorithm to
generate sub-daily rainfall sequences conditional to
daily rainfall, under a future climate.
• Also collaborating with dynamical climate modellers
to evaluate capacity of regional climate models to
Slide 42
simulate sub-daily precipitation
Slide 43
References
• Hardwick-Jones, R., Westra, S. & Sharma, A., 2010, “Observed
relationships between extreme sub-daily precipitation, surface
temperature, and relative humidity”, Geophysical Research
Letters, 37, L22805
• Westra, S. & Sisson, A., 2011, “Detection of non-stationarity in
precipitation extremes using a max-stable process model”, Journal
of Hydrology, 406
• Westra, S., Mehrotra, R., Sharma, A. & Srikanthan, S., 2012,
Continuous rainfall simulation: 1. A regionalised sub-daily
disaggregation approach, Water Resources Research, 48
(W01535).
• Westra, S., Evans, J., Mehrotra, R. & Sharma, A., “Disaggregating
from daily to sub-daily rainfall under a future climate”, submitted
to Journal of Climate
• Evans, J. & Westra, S., “Investigating the mechanisms of diurnal
rainfall variability using a Regional Climate Model”, submitted to
Slide 44
Journal of Climate