Spatialtemporal modelling of extreme rainfall

1
Spatial-temporal modelling of extreme rainfall
Climate Adaptation Flagship

• Mark Palmer
• TechFest 2009

2
Spatial-temporal modelling of extreme rainfall

• Acknowledgements
• Santosh Aryal (CLW),
• Bryson Bates (CMAR),
• Eddy Campbell, (CMIS),
• Yun Li (CMIS),
• Neil Viney (CLW),
• IOCI,
• BoM
• WA Water Corp.  
• Australian Greenhouse Office (Department  of
  Climate Change)
• Upper Parramatta River Catchment Trust
• Sydney Water 
• Sydney Metropolitan Catchment Management
  Authority 
• Hunter-Central Rivers Catchment Management
  Authority
• Southern Rivers Catchment Management Authority

3
Spatial-temporal modelling of extreme rainfall

• How does it fit in to Tech Fest 2009
• (relatively) large data sets
• Use of Bayesian Hierarchical modelling to
• Amalgamate data sets,
• Reduce dimensionality.

4
Spatial-temporal modelling of extreme rainfall
5
Spatial-temporal modelling of extreme rainfall

• Where is this changing, clearly not uniformly
• WA Wheatbelt, formerly most reliable wheat
  growing area in Australia
• The Eastern urban fringe, relies on dam water
• Kimberley rainfall is increasing
• What about changes in extreme rainfall, this may
  be more dramatic?

6
Spatial-temporal modelling of extreme rainfall

• Why are we doing this?
• With climate change, engineering specifications
  will need to change dams, culverts roads.
• Farming practices may need to change
• So
• Can we pickup changes?
• Can we relate changes to climate drivers?
• Can we link our analyses to climate models?

7
Spatial-temporal modelling of extreme rainfall

• What are we really interested in?
• Return Period/Levels, IDFs and DA curves
• Need to look at this from both a spatial and
  temporal aspect
• So, how are we going to proceed
• Data
• Characterize extremes at a station
• Allow characteristics of extremes to vary
  spatially, and be driven by covariates

8
Spatial-temporal modelling of extreme rainfall

• Return Period
• The return period T, for a given duration and
  intensity i(d), is the average time interval
  between exceedance of the value i(d)

• Intensity Duration Frequency Curves
• X-axis duration
• Y-axis intensity
• Each line corresponds to a fixed return period

• How have these been derived?
• Generally in an adhoc manner
• No measures of variability associated with them
• Generally not (very) specific to a location

9
Spatial-temporal modelling of extreme rainfall

• Data
• Daily data 1950-2003
• Pluvio data essentially continuous recording,
  but aggregated into sub to super daily rainfall
  records
• There is an issue with daily versus 24 hour data
• Data divided into summer and winter (region
  specific)
• Numbers of sites
• SW WA 1501 made up of (1227, 274)
• UPRCT 607 (346, 261)

10
Spatial-temporal modelling of extreme rainfall

• Bayesian Hierarchical Model
• Model rainfall at a station
• GEV distribution for extremes
• Spatial Modelling of GEV parameters, to borrow
  'strength
• Geostatistical approach based on GPs, such
  askriging for a single variable is well
  understood and developed, shall pursue an
  approach that induces GPs.
• Include covariates
• looking for trends in parameters of the GEV,
  covariates (time, drivers eg SOI, AAO, SST, heat
  content of oceans)
• Hierarchical framework
• Allows us to combine the above
• Bayesian
• Coherent approach to inferences
• Applications of the model
• Use it to derive return levels, IFD curves, areal
  statistics such as Depth Area curves

11
Spatial-temporal modelling of extreme rainfall

Di-graph representation of the Spatial-Temporal
model for extreme rainfall

• Issues
• Implemented correctly? (Gelman, simulation)
• Does the model fit?

12
Spatial-temporal modelling of extreme rainfall

• Examples UPRCT GEV surfaces

13
Spatial-temporal modelling of extreme rainfall

• Differences in 50 year Return Level Surfaces
  (2003 1953)

14
Spatial-temporal modelling of extreme rainfall

• IDF curves

Sample of IDF curves for the pluviograph station
566038, for a 50 year return period (Ocean heat
anomaly 0, ie effectively the historical long
term average), drawn from the MCMC procedure.
Estimated average IDF curves, for return periods
of (a) 5 years, (b) 20 years and (c) 50 years.
Each figure shows the IDF curves calculated using
an ocean heat anomaly of -2.5, 0.0 and 2.5
respectively.
15
Spatial-temporal modelling of extreme rainfall
Driven by covariates
16
Spatial-temporal modelling of extreme rainfall

• Perils of climate research
• Syd Levitus, John Antonov, Tim Boyer (March 2008)
• Improved estimates of upper-ocean warming and
  multi-decadal sea-level rise (Catia M.
  Domingues1, John A. Church1,2, Neil J. White1,2,
  Peter J. Gleckler3, Susan E. Wijffels1, Paul M.
  Barker1 Jeff R. Dunn1) CSIRO June 2008, Nature

17
Spatial-temporal modelling of extreme rainfall

• Whats achieved
• A spatial model for GEVs which characterizes
  extreme rainfall at gauged and ungauged locations
  over a range of durations
• Can be driven by covariates (either measured or
  derived from other models)
• Can derive objects of interest such as IDFs and
  measures of variability

18
Spatial-temporal modelling of extreme rainfall

• Whats next predictor selection
• Large numbers of covariates, (both 2D and 3D)
• L1, Lasso, RChip
• What about spatial correlation (2 and 3 D data)?
• What about time lags?

19
Thank you
CSIRO Mathematical and Information Sciences Mark
Palmer Phone [phone] Email
mark.palmer_at_csiro.au Web www.csiro.au/cmis
Climate Adaptation Flagship
Contact UsPhone 1300 363 400 or 61 3 9545
2176Email Enquiries_at_csiro.au Web www.csiro.au