Transcript Downscaling in sub-daily scale

Downscaling in sub-daily
scale – inventory of methods
Joanna Wibig
University of Lodz, POLAND
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
Outline:
•
•
•
•
•
Dynamical downscaling
Weather generators
Disaggregators
Evaluation procedures
Summary
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
dynamic downscaling
Regional climate models in relatively
high resolution (both in space and
time)
MOS technics: DC or bias corrections
Disaggregation, if necessary
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
DC + frequency adjustment procedure
There is an important limitation of
DC that a change in precipitation
frequency is generally not
considered and the future
frequency is assumed to be
identical to today’s.
Olsson J, Gidhagen L, Gamerith V, Gruber
G, Hoppe H, Kutschera P, 2012,
Sustainability, 2012, 4, 866-887;
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
Modeling of a diurnal cycle of
precipitation
An example of the estimated diurnal cycle of
precipitation amount from observation and
RCA3 simulations with 4 different
resolutions for the ‘Malexander’ station
Walther, A., et al., 2011
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
Comparison of different DC and bias
correction precedures
Räisänen, Räty, Clim.Dyn., September 2012 online first
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
Weather generators
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
RainSim
•
•
•
•
•
Burton, A., et al.., 2008: Rainsim: A spatialtemporal stochastic rainfall modelling system.
Environ. Mod. & Soft., 23, 1356-1369.
Rainfall only
Daily or hourly
Poisson cluster models: NSRP, GNSRP, BLRP
Single or multi-site locations
Models are calibrated separately within
different weather states
Name of software: RainSim V3
Developer: School of Civil Engineering and Geosciences,
Newcastle University, NE1 7RU, UK
Contact: Aidan Burton, School of Civil Engineering and
Geosciences, Newcastle University, NE1 7RU, UK,
[email protected]
Hardware: PC with windows 2000 or XP
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
WGEN (ClimGen)
Washington State University
http://www.bsyse.wsu.edu/CS_Suite
/ClimGen/index.html
Daily resolution
• Precipitation
• Maximum and Minimum temperature
• Solar radiation
• Maximum and Minimum relative humidity
• Maximum and Minimum dew point temperature
• Windspeed
• Vapor pressure deficit
• Reference evapotranspiration (Penman-Monteith,
Priestley-Taylor, Hargreaves).
1 to 1440 minute resolution
• Storm events (precipitation intervals)
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
• ClimGen, is a weather generator of a WGEN type
• ClimGen generates precipitation, daily maximum and
minimum temperature, solar radiation, air humidity, and
wind speed.
• ClimGen usesWeibull distribution to generate
precipitation amounts instead of the Gamma distribution
used by WGEN.
• In ClimGen, all generation parameters are calculated for
each site of interest
• ClimGen can be applied to any location with enough data
to parameterize the program.
• ClimGen uses quadratic spline functions chosen to ensure
that:
 The continuity of the daily average values across month
boundaries,
 The continuity of the first derivative across month
boundaries.
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
RMAWGEN
•
•
•
•
•
Cordano E., Eccel, E., 2011. RMAWGEN (R
Multi-site Auto-regressive Weather
GENerator): a package to generate daily time
series from monthly mean values.
http://CRAN.Rproject.org/package=RMAWGEN
Auto-regressive models
R-language
Daily resolution
Multi-site
Temperature, precipitation, wet, dry, hot
spells, others
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
GLIMCLIM
http://www.homepages.ucl.ac.uk/~ucakarc/work
/glimclim.html
• A GLM for daily observations of a climate variable is defined by
setting up a probability distribution for each of the daily values.
Each observation is regarded as a realization or a sample from its
own distribution.
• A typical assumption in GLMs is that all of the observations are
drawn from the same family of distributions, for example, normal,
Poisson, or gamma.
• A GLM is essentially a multiple regression model for the chosen
family of distributions; the regression-like approach enables the
individual distributions to change with time site and external
factors. Inference (judging if a possible factor has a genuine effect
on the studied phenomenon) is carried out using likelihood-based
methods. Such methods implicitly take into account the family of
distributions being used. It is therefore important to choose a
realistic distribution for a specific climate variable.
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
Disaggregators
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
K-nearest neighbours resampling
approach for disaggregation to
multisite hourly data
Mezghani, Hingrey, 2009, A combined downscaling-disaggregation weather
generator for stochastic generation of multisite hourly weather variables over
complex terrain: Development and multi-scale validation for the Upper Rhone
River basin, J.Hydrology,377:245-260
M. Sharif, D. H. Burn and K.M. Wey, 2007, Daily and Hourly Weather Data
Generation using a K-Nearest Neighbour Approach Challenges for Water
Resources Engineering in a Changing World, Winnipeg, Manitoba, August 22 –
24, 2007
End User Needs for Regional Climate Change Scenarios, 7-9 March 2012, Kiel
Disaggregation to finer scales
with stochastic methods
Onof, Arnbjerg-Nielsen, 2009 Atmospheric Research 92 350–363
Hingray, Ben Haha, 2005. Atmospheric Research 77, 152–175.
Ormsbee, L.E., 1989, J. Hydraul. Eng., ASCE 115 (4), 507– 525.
End User Needs for Regional Climate Change Scenarios, 7-9 March 2012, Kiel
Disaggregation by adjusting
The proportional adjusting procedure:
k


~
~
X s  X s  Z /  X j 
j 1


The linear adjusting procedure:
k


~
~
X s  X s  s  Z   X j 
j 1


The power adjusting procedure:


~
~
X s  X s  Z /  X j 
j 1


k
Koutsojannis & Onof, 2001, J.
Hydrology: 246:109-122
s /  s
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
M.-L. Segond *, C. Onof, H.S.
Wheater, 2006, J. Hydrology 331,
674– 689
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
Validation methods
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
STATISTICAL MEASURES:
Mean (monthly, seasonal, annual) and standard
deviations
Daily averages (or totals): mean (on wet days) ,
standard deviations, skewness
Minimum, maximum, selected percentiles,
distribution checking
frequency of days with precipitation crossing
selected thresholds
Dry/wet spells
Cold/hot spells
Frequency of days with wind maximum exceeding
thresholds
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
TEMPORAL CONSISTENCY:
trends
Autocorrelations with lag 1 (persistency)
SPATIAL CONSISTENCY:
Anscombe residuals:
Pearson residuals:
M.-L. Segond *, C.
Onof, H.S. Wheater,
2006, J. Hydrology 331,
674– 689
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
Cross – validation principles
Räisänen, Räty, Clim.Dyn., September 2012 online first
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
Mean square error
Continuous ranked probability score
Out of range score
The frequency of cases in which Tver is below the lowest or above the
highest of the all Tproj values
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste
To be continued …..
VALUE - Data and Validation Workshop, 19-20 Sep 2012, ICTP, Trieste