You are on page 1of 13

INTERNATIONAL JOURNAL OF CLIMATOLOGY

Int. J. Climatol. (2012)


Published online in Wiley Online Library
(wileyonlinelibrary.com) DOI: 10.1002/joc.3544

The Statistical DownScaling Model: insights from one decade


of application
R. L. Wilbya * and C. W. Dawsonb
aDepartment of Geography, Loughborough University, Leicestershire LE11 3TU, UK
b Department of Computer Science, Loughborough University, Leicestershire LE11 3TU, UK

ABSTRACT: The Statistical DownScaling Model (SDSM) is a freely available tool that produces high resolution climate
change scenarios. The first public version of the software was released in 2001 and since then there have been over 170
documented studies worldwide. This article recounts the underlining conceptual and technical evolution of SDSM, drawing
upon independent assessments of model capabilities. These studies show that SDSM yields reliable estimates of extreme
temperatures, seasonal precipitation totals, areal and inter-site precipitation behaviour. Frequency estimation of extreme
precipitation amounts in dry seasons is less reliable. A meta-analysis of SDSM outputs shows a preponderance of research
in Canada, China and the UK, whereas the United States and Australasia are under-represented. In line with the wider
downscaling community, the most favoured sector of analysis is water and flood risk management which accounts for nearly
half of all output; research in other sectors such as agriculture, built environment and human health is less prominent but
growing. Over 50% of the studies are concerned with production of climate scenarios, comparison or technical refinement
of downscaling methodologies. In contrast, there is relatively little evidence of application to adaptation planning and
climate risk management. We assert that further attention to physically meaningful quantities such as wind speeds, wave
heights, phenological and hazard metrics could improve uptake of downscaled products. Chronic uncertainty in boundary
forcing continues to undermine confidence in downscaled scenarios so these tools are best used for sensitivity testing and
adaptation options appraisal. Copyright  2012 Royal Meteorological Society

KEY WORDS climate change; statistical downscaling; risk assessment; software


Received 18 October 2011; Revised 14 June 2012; Accepted 17 June 2012

1. Introduction spatially and temporally). The Statistical DownScaling


Model (SDSM) is intended to bridge the divide between
Downscaling is a technique by which properties of the accessibility and sophistication. This freely available soft-
free atmosphere are used to predict local meteorologi- ware enables the production of climate change time series
cal conditions. The large-scale information may originate at sites for which there are sufficient daily data for
from systematic weather observations or from climate model calibration, as well as archived General Circulation
model outputs. Either way, the concept of downscal- Model (GCM) output to generate scenarios of the 21st
ing is appealing because the added detail could inform century. SDSM can also be used as a stochastic weather
site-specific assessment and management of climate risk. generator or to infill gaps in meteorological data.
However, the extent to which the downscaling commu- There were over 5100 registered downloads from
nity is delivering these practical outcomes is open to 110 countries between the release of the first version
debate (Wilby and Fowler, 2010). of SDSM in 2001 and June 2012. Approximately 25%
One explanation for limited penetration in adapta- of the registrations originate from Africa, Asia, Cen-
tion planning is that many downscaling models are tral and South America. However, it is recognized that
restricted in their use to specialists and/or research insti- even the modest requirements of SDSM (i.e. location-
tutions. Furthermore, dynamical downscaling models are specific meteorological data or weather-dependent vari-
not widely implemented for all regions, and can incur ables for model calibration, and daily GCM output for
significant computational and technical demands at the future scenario generation) can deter some prospective
outset. On the other hand, public software such as the users, especially those in data sparse regions. Undoubt-
Climate Wizard (http://www.climatewizard.org/)(Girvetz edly, interest in SDSM has been stimulated by cita-
et al., 2009), may be highly accessible, but extracts cli- tions from the United Nations Framework Convention on
mate change scenarios that are relatively coarse (both Climate Change (UNFCCC) (http://unfccc.int/adaptation/
nairobi work programme/knowledge resources and
publications/items/5487.php), the United Nations Devel-
* Correspondence to: R. L. Wilby, Department of Geography, Lough-
borough University, Loughborough, Leicestershire LE11 3TU, UK. opment Programme (Lu, 2006), and the Intergovernmen-
E-mail: r.l.wilby@lboro.ac.uk tal Panel on Climate Change Fourth Assessment Report

Copyright  2012 Royal Meteorological Society


R. L. WILBY AND C. W. DAWSON

(a)

(b)

Figure 1. SDSM-user interfaces (a) version 2.1 (upper) and (b) version 4.2 (lower). This figure is available in colour online at
wileyonlinelibrary.com/journal/joc

(AR4) (Christensen et al., 2007). Another important fac- we conclude with an appraisal of future opportunities,
tor has been the web-delivery of the software and including emergent research themes, and outline plans
downscaling predictor variables via the Canadian Cli- for further software development.
mate Change Scenarios Network (CCCSN) portal (http://
www.cccsn.ec.gc.ca/?page=main&lang=en).
After a decade of research yielding over 170 publi- 2. SDSM concept and evolution
cations, this article reflects on the contribution made by SDSM began life in summer 2000 as a crudely inter-
SDSM. The associated literature provides a sample of the faced Visual Basic programme designed to teach five
themes explored by downscaling researchers. The next elemental steps in statistical downscaling: (1) selection
section explains the original concept and evolution of of predictor variables, (2) calibration of the model
SDSM. This is followed by a full description of the sci- (using observed predictors), (3) validation of the model,
entific algorithm. Section 4 presents a meta-analysis of (4) generation of future scenarios (using climate model
the SDSM publication inventory and draws out the pre- predictors), and (5) analysis of outputs (using a range
ferred regions, sectors and topics of interest. Section 5 of climate diagnostics). Users were led through these
discusses the extent to which SDSM (and downscaling in sequential tasks via a master screen (Figure 1, upper
general) has contributed to development and adaptation panel), a feature that is retained in the present ver-
planning. We also consider a few options for improving sion (Figure 1, lower panel and Figure 2). A proto-
access to climate risk information in some of the most type SDSM was beta-tested at a Canadian Climate
vulnerable and data sparse regions of the world. Finally, Impact Scenarios Workshop on Methods and Techniques

Copyright  2012 Royal Meteorological Society Int. J. Climatol. (2012)


SDSM – ONE DECADE ON

SDSM screen Tasks performed Data requirements

1. Quality control Checking for missing data and outliers Daily predictands

2. Transform data Lagging, normalisation, transformation Daily predictands

3. Screen variables Exploring predictor-predictand relations Re-analysis predictors

4. Calibrate model Deriving statistical transfer functions Predictors & predictands

5. Weather generator Synthesizing observed daily weather Re-analysis predictors

6. Scenario generator Generating climate change scenarios GCM predictors

7. Summary statistics Describing statistical properties (tabular) Any daily time series

8. Compare results Graphing summary statistics (monthly) SDSM output &predictand

9. Frequency analysis Graphing and tabling extreme values Any daily time series

10. Time series Graphing daily and seasonal diagnostics Any daily time series

Figure 2. The main screens, tasks performed and data requirements of SDSM.

for Constructing Regional Climate Scenarios at the Table I. Candidate predictor variables used by SDSM.
University of Laval, Quebec City, in November 2000.
After incorporating feedback from the workshop, SDSM Predictor Description
Version 2 was completed in September 2001, and dis-
TEMP Mean temperature at 2 m
tributed on request. Meanwhile, SDSM compliant sets
MSLP Mean sea level pressure
of predictor variables were processed by CICS for H850 850 hPa geopotential height
every land grid-cell in the HadCM3 and CGCM1 cli- H500 500 hPa geopotential height
mate models and made available through a web-site USUR Near surface westerly wind
(http://www.cics.uvic.ca/scenarios/index.cgi?Scenarios) U850 Westerly wind at 850 hPa
[now superseded by the CCCSN portal (http://www.cccsn. U500 Westerly wind at 500 hPa
ec.gc.ca/?page=main&lang=en)]. The standard suite of VSUR Near surface southerly wind
SDSM predictor variables is listed in Table I. V850 Southerly wind at 850 hPa
The software and data release were followed by V500 Southerly wind at 500 hPa
journal articles describing SDSM functions (Wilby et al., FSUR Near surface wind strength
F850 Wind strength at 850 hPa
2002) and extension to multi-site applications (Wilby
F500 Wind strength at 500 hPa
et al., 2003). The first user studies began to appear in ZSUR Near surface vorticity
conference proceedings and technical reports (LCCP, Z850 Vorticity at 850 hPa
2002; Cohen and Neale, 2003; Goldstein and Milton, Z500 Vorticity at 500 hPa
2003; Goodess et al., 2003). The first PhD employing DSUR Near surface divergence
SDSM was awarded in the year 2003 (Guangul, 2003). D850 Divergence at 850 hPa
Version 3 was released in May 2004 via the present D500 Divergence at 500 hPa
registration web-site (http://www.sdsm.org.uk/). This ver- QSUR Near surface-specific humidity
sion was more compatible with MS Windows and Q850 Specific humidity at 850 hPa
included a revised screen format whilst retaining the Q500 Specific humidity at 500 hPa
sequential tasking of earlier editions (Figure 2). New RSUR Near surface relative humidity
R850 Relative humidity at 850 hPa
features included quality control procedures, data trans-
R500 Relative humidity at 500 hPa
formations, more flexible graphing and an expanded
suite of diagnostic tests. This enables the interrogation
of climate data for trends, distributions, and covariance
regardless of whether any downscaling is performed. The peer-reviewed article referring to SDSM found that a
initial set of GCMs was also expanded for the UK to simpler and more parsimonious genetic programming
facilitate construction and characterization of uncertainty approach performed better at downscaling extreme tem-
in regional climate change scenarios for the water indus- peratures, at least for the Chute-du-Diable weather station
try (Prudhomme and Davies, 2005; Prudhomme et al., in NE Canada (Coulibaly, 2004). The article also estab-
2005; Vidal and Wade, 2008). The first independent, lished the principle that SDSM could provide a worthy

Copyright  2012 Royal Meteorological Society Int. J. Climatol. (2012)


R. L. WILBY AND C. W. DAWSON

benchmark against which to compare other downscaling studies explored statistical relationships between indices
methods. of atmospheric circulation and local meteorology (Wilby
Version 4 was released in April 2007 with addi- 1994, 1995, 1997, 1998a, 1998b). For example, Conway
tional features for extreme value, time-series analysis and et al. (1996) showed that daily precipitation across the
simulation (e.g. Generalized Extreme Value analysis, a UK could be downscaled from airflow strength, direction
re-sampling algorithm for downscaling using past ana- and vorticity indices that were objectively defined from
logues for future climate, auto-regressive analysis and pressure patterns in the vicinity of the target region. Sub-
modelling, a raft of new summary statistics, and diag- sequent work used these nonlinear, transfer functions to
nostics based on STARDEX indices (Haylock et al., generate scenarios for climate change impact assessments
2006)). These upgrades were required by the Environ- in hydrology (Wilby et al., 1998a; Wilby et al., 1999), or
ment Agency to enable downscaling of tidal surges in to compare performance metrics with other downscaling
the Thames Estuary (Wilby, 2008a). In the same year, techniques including dynamical methods (Wilby et al.,
an automated statistical downscaling model was pub- 1998b; Hay et al., 2000; Wilby et al., 2000).
lished by Hessami et al. (2008) – a clone of SDSM with
automatic predictor selection methods based on back- 3.1. Single site version
ward stepwise regression and partial correlation analysis.
The SDSM algorithm is best described as a condi-
Subsequent work on SDSM (original) has focussed on
tional weather generator because atmospheric circula-
error-trapping and education. To date, SDSM training
tion indices and regional moisture variables (Table I) are
workshops have been held in: Quebec (2000), Victo-
used to estimate time-varying parameters describing daily
ria (2003), Boulder (2004), Casablanca (2007), Toronto
weather at individual sites (e.g. precipitation occurrence
(2010), Xi’an (2010), Loughborough (2011) and Dhaka
or maximum temperatures). The downscaled process is
(2011).
either unconditional (as with wet-day occurrence or air
As noted before, there are other public domain down-
temperature), or is conditional on an event (as with rain-
scaling tools besides SDSM – all intended to make
fall amounts).
climate model products more accessible and relevant
For wet-day occurrence Wi there is a direct linear
to decision-makers. For example, MAGICC-SCENGEN
dependency on n predictor variables Xij on day i:
constructs geographically explicit climate change projec-
tions based on user-specific ensembles of climate model
output from the CMIP3/AR4 archive (http://www.cgd. 
n
Wi = α0 + αj Xij (1)
ucar.edu/cas/wigley/magicc/). LARS-WG is a widely
j =1
used weather generator that produces daily, site-specific
climate scenarios consistent with changes in climate
under the constraint 0 ≤ Wi ≤ 1. Precipitation occurs
model output (http://www.rothamsted.bbsrc.ac.uk/mas-
when the uniform random number r ≤ Wi . The thresh-
models/larswg.php). The Climate Systems Analysis
old (mm) for a wet-day may vary between locations,
Group at the University of Cape Town has a data por-
depending on the definition of trace rainfalls or preci-
tal that provides monthly summary statistics downscaled
sion of measurement. When calibrating the model, it is
to individual meteorological stations in Africa and Asia,
also important to confirm that daily rainfall totals have
based on eight climate models (http://data.csag.uct.ac.
not been bulked over multiple days, thereby understating
za/). The UNFCCC compendium on methods and tools
precipitation frequencies. Predictor variables Xij may be
to evaluate impacts, vulnerability and adaptation to
concurrent and/or lagged to increase persistence of wet-
climate change provides succinct descriptions of the
and dry-spells.
strengths and weaknesses of all these models and more
When a wet-day is returned, the precipitation total Pi
(http://unfccc.int/adaptation/nairobi workprogramme/
is downscaled using:
knowledge resources and publications/items/5457.php).
Faced with a potentially bewildering array it is essen-
tial that prospective users select the most appropriate 
n
Pik = β0 + βj Xij + ei (2)
tool(s) for the task in hand. Accordingly, Wilby et al.
j =1
(2009) provide a checklist of nine sources of climate risk
information with their respective advantages and disad-
where k (typically 0.25) is used to transform daily wet-
vantages.
day amounts to better match the normal distribution.
However, other transformations (such as logarithm or
inverse normal) may also be applied to Pi . Note that
3. SDSM algorithm the same predictor set is used to downscale Wi and Pi
The theoretical origins of SDSM lie in a series of papers and that all predictors vij are standardized with respect
in the 1990s and early 2000s. At this point, the downscal- to their climatological mean V j and standard deviation
ing software was only available in Fortran 90 code and σj :
model testing was being performed using meteorological vij − V j
Xij = (3)
data from the UK, United States, and Japan. These early σj

Copyright  2012 Royal Meteorological Society Int. J. Climatol. (2012)


SDSM – ONE DECADE ON

For unconditional processes, such as temperature or greater (because of larger signal-to-noise ratio). As with
wind speeds, there is a direct linear relationship between other re-sampling methods, the maximum daily value
the predictand Ui and the chosen predictors Xij : generated cannot exceed the maximum daily amount
in the observations. However, synthetic N-day totals

n
can exceed observed N-day totals if the atmospheric
Ui = γ0 + γj Xij + ei (4) conditioning produces a sequence previously unseen in
j =1
the training set.
The model error ei is assumed to follow a Gaussian dis-
tribution and is stochastically generated from normally 3.3. Evaluation of SDSM
distributed random numbers and added on a daily basis As will be discussed later, SDSM has been deployed
to the deterministic component. This white noise enables in many downscaling comparison studies hence there
closer fit of the variance of the observed and downscaled is good understanding of the model capabilities and
distributions, but can degrade skill at replicating serial limitations. Despite the simplicity of the algorithm,
autocorrelation implicit to daily predictor variables. The SDSM performs favourably in independent evaluations
stochastic process also enables the generation of ensem- of extreme temperatures (Coulibaly et al., 2005; Liu
bles of time series to reflect model uncertainty. et al., 2008); the annual precipitation cycle, seasonal
All downscaling parameters (αj , βj , and γj ) are and annual precipitation totals (Wetterhall et al., 2007a,
obtained by least squares calibration of the local pre- 2007b); extreme areal average precipitation (Hashmi
dictand(s) against regional predictor variables derived et al., 2011a); and inter-site correlation of precipitation
from the National Center for Environmental Prediction amounts (Liu et al., 2011). Indices of rainfall occurrence
(NCEP) re-analysis (Kalnay et al., 1996) using data for are generally more skillfully modelled than intensity,
any period within 1961–2000. Predictands are down- particularly in winter (Haylock et al., 2006).
scaled separately so any covariance must be conveyed by However, SDSM is known to under-estimate variance
common predictor variables and/or correlation between in wet- and dry-spell lengths (Coulibaly et al., 2005; Liu
predictors. Model testing suggests that this is a reason- et al., 2011) as well as heat-wave persistence unless auto-
able assumption (Wilby et al., 1998a). regressive terms or lagged predictors are applied (Wilby,
In common with all downscaling methods, SDSM pre- 2008b). Frequency estimation of extreme precipitation
dictor–predictand relationships are assumed to be unaf- amounts, particularly in summer, is problematic for all
fected by anthropogenic climate change in the calibration downscaling methods and SDSM is no exception (Hay-
period, and transferrable to future decades. In practice, lock et al., 2006; Chen et al., 2010; Liu et al., 2011).
the parameters of all empirical and dynamical downscal- Some studies suggest that alternative modelling tech-
ing models are observed to vary over decadal-time scales niques such as genetic programming may outperform
(Wilby, 1997). Low-frequency variability due to large- SDSM over a range of diagnostics (Coulibaly, 2004;
scale atmospheric changes can be addressed using mix- Hashmi et al., 2011b). Other work has revealed the sen-
tures of stochastic processes (Katz and Parlange, 1996; sitivity of SDSM scenarios to the source of downscaling
Wilby, 1998b) but changes in land-surface feedbacks predictor variables used in model calibration (i.e. choice
are a major unknown affecting all downscaled scenarios of re-analysis product) (Koukidis and Berg, 2009), pre-
(Pielke and Wilby, 2012). dictor suite and domain (Crawford et al., 2007; Mullan
et al., 2012).
3.2. Extension to multiple sites Further testing of SDSM by the authors shows that
Although SDSM is unable to downscale simultaneously the credibility of the model depends on the amount and
to multiple sites, the basic model may be adapted for quality of meteorological data available for calibration,
this purpose (following Wilby et al., 2003). This involves as well as the physical controls of the local meteorology.
two steps. First, a ‘marker’ precipitation series based on This can be challenging in regions where there is high
daily area averages from several sites (or a single key seasonal variability due to monsoon rains, inter-play
site) is generated using predictors Xij . Second, the area between complex terrain and water bodies, or annual
average is disaggregated to observed daily precipitation totals dominated by a few extreme/local convective
totals recorded at the constituent sites. This is achieved events (Wilby, 2008c). These issues are exacerbated
by resampling the multi-site amounts on the date with if meteorological networks are sparse and/or decaying,
observed area average closest to the downscaled area records are broken by civil unrest, or quality controls are
average. not consistently applied.
Since actual patterns of rainfall are being re-sampled Long-term, homogeneous daily meteorological records
both the area average of the marker series and the spatial are particularly rare in the Middle East. Nonetheless,
covariance of the multi-site array are preserved (Wilby if there are sufficient high quality data to calibrate
et al., 2003; Harpham and Wilby, 2005). Area averages SDSM, there is scope for infilling and hindcasting
are favoured over single site marker series because missing data using identified relationships between large-
there is less risk of employing a non-homogeneous or scale predictor variables and local weather. For example,
non-representative record, and predictability is generally Figure 3 shows observed and downscaled daily mean

Copyright  2012 Royal Meteorological Society Int. J. Climatol. (2012)


R. L. WILBY AND C. W. DAWSON

(a) (b)
1000
Daily mean temperature (degC) 35
SDSM Observed 950

Growing Degree Days (GDD)


30
900
25
850
20
800
15 750
10 700
5 650
0 600
SDSM Observed
-5 550
01/94

02/94

03/94

04/94

05/94

06/94

07/94

08/94

09/94

10/94

11/94

12/94

01/95
500
1960 1965 1970 1975 1980 1985 1990 1995 2000

Figure 3. Observed and SDSM hindcast of daily mean temperatures at Kfardane, Lebanon (left panel) and winter growing degree days at Amman,
Jordan (right panel). Source: Wilby (2010a).

temperatures at Kfardane, Lebanon where records began


in 1994 and run to present. Using all available data for
the period 1994–2000 daily temperatures and derived
metrics (such as the winter growing degree days) may be
hindcast for periods that pre-date observations. Similarly,
days with missing data (as in May 1994) were estimated
using the same statistical model. Where multi-decadal
series are reconstructed it is also possible to estimate
historic trends. For example, the SDSM hindcast of
precipitation totals at Kfardane suggests a long-term
decline in annual rainfall (∼125 mm) over the period
1961–2000 (not shown). Figure 4. Annual count of ‘climate downscaling’ publications between
1993 and 2011 identified by the WoS (9 June 2012). Note that the count
combines all types of downscaling study – statistical and dynamical.
4. Meta-analysis of SDSM literature
Since the seminal downscaling research of Wigley et al. Scopus and Web of Science (WoS). By June 2012
(1990) and of von Storch et al. (1993) there have been the inventory comprised of 112 journal articles (65%),
over 1700 publications (∼1500 since 2002) on the sub- 23 technical reports (13%), 22 conference proceedings
ject but the annual volume of output is accelerating (13%), 13 theses (8%), and 2 book chapters (1%). The
(Figure 4). By 2011 more than 300 papers were being list is unlikely to be exhaustive since much down-
published each year by the downscaling community as scaling work is published in the grey literature. Pub-
a whole. Including the work of the authors, more than lications that are not in the English language are
170 SDSM outputs have been identified since the release known to be under-represented too, especially those in
of version 2 (i.e. ∼10% of the total volume). This sec- Chinese.
tion interrogates the SDSM bibliography for underlining
themes. Clearly, the findings represent only a sample 4.1. Regions of interest
of the total downscaling output but the survey provides Given the origins of the tool, it is not surprising that
insights to patterns of research over the last decade. the focus of early SDSM applications was in Europe and
Table II shows the percentage of SDSM publications North America. However, by the year 2012, the volume
by region, country, sector and topic for the top five of output from Asia equalled that of North America
categories in each field. The SDSM bibliography was (30%), and surpassed Europe (22%). The statistics are
compiled using three search engines: Google Scholar, now dominated by Canada (26%), the UK (13%), and

Table II. Percentage of output (n = 172) referring to SDSM by region, country, sector and topic for top five categories since
2002.

Rank Region % Country % Sector % Topic %

1 North America 30 Canada 26 Water 41 Impact 38


2 Asia 30 UK 13 Unspecified 33 Scenario 26
3 Europe 22 China 13 Flood 6 Comparison 20
4 Africa 9 Iran 6 Agriculture 5 Techniques 7
5 Central/South America 6 Ethiopia 4 Urban 5 Management 5
Others 3 Others 38 Others 10 Others 4

Publications were surveyed in June 2012 and are listed in full at: http://co-public.lboro.ac.uk/cocwd/SDSM/Bibliography.pdf.

Copyright  2012 Royal Meteorological Society Int. J. Climatol. (2012)


SDSM – ONE DECADE ON

China (13%) which account for 52% of all the documents. tool, or production of climate scenarios for trend and
Other notable national contributions have been made uncertainty analysis (e.g., Chu et al., 2010; Gachon and
by Iran (6%), Ethiopia (4%), the United States (4%), Dibike, 2007; Souvignet et al., 2010; Tian et al., 2011).
Ireland (3%), and South Korea (3%). Output from Africa According to the WoS, ∼39% of all downscaling stud-
accounts for 9% of the total but includes Masters and PhD ies are concerned with aspects of water, flood, or drought.
theses suggesting that the tool is being used for training The corresponding figures for the other sectors mentioned
and development in the region (mainly in Ethiopia); above are agriculture (11%), ecosystems/environment
a point that has been noted elsewhere (Ziervogel and (6%), urban/built environment (5%), energy (2%), human
Zermoglio, 2009). The most under-represented continents health (1%), and tourism (<1%). Therefore, apart from
are Australasia (just 2% of all outputs), followed by under-representation of agricultural studies (and possibly
Central and South America (6% of outputs). ecosystems) – when allowing for inclusion of more grey
Overall, 39 countries are represented in the SDSM literature and less double counting – the SDSM sam-
bibliography, but apart from Canada, China, and the ple broadly represents the wider community’s sectoral
UK uptake in other regions is at a relatively low emphases.
level. In particular, there are few studies in the United
States (4%) and rest of Europe (8%) – perhaps because 4.3. Topics of interest
alternatives to SDSM are available in both regions. The themes of interest to SDSM-users are harder to
For example, the Bias-Correction Spatial Disaggrega- classify. Nonetheless, five broad topics were identified:
tion technique of Maurer et al. (2007) has been widely regional climate change scenarios; climate change impact
used and cited in the United States. Furthermore, the and vulnerability assessment; managing or adapting to
projections are specifically targeted at the hydrologi- climate risks; downscaling model evaluation; and tech-
cal and ecological modelling communities (http://gdo- nical refinements (Table II). About 42% of the studies
dcp.ucllnl.org/downscaled cmip3 projections/). Likewise, were concerned with climate change vulnerability and
the EU ENSEMBLES portal provides web-access to high impacts, of which two-thirds focused on the water sec-
resolution climate scenarios for Europe derived from tor (see above). Approximately 26% of studies reported
an ensemble of GCMs and downscaling methods [ana- the production of downscaled climate scenarios without
logue, weather typing, regression and neural networks extension to impact or vulnerability assessment (Chaleer-
(http://www.ensembles-eu.org/)]. Furthermore, there are aktrakoon and Punlun, 2010; Qiu and Jiang, 2010). A
over 50 studies worldwide based on the popular weather further 20% undertook comparative work – running mul-
generator LARS-WG (Barrow and Semenov, 1995). tiple downscaling techniques (sometimes with multiple
The WoS survey of all downscaling research returns impact models) to characterize the uncertainty or relative
lower proportions of output for the three leading nations skill attached to different methods (Dibike and Coulibaly,
and a different rank order to the SDSM inventory: United 2005; Khan et al., 2006; Tryhorn and DeGaetano, 2011).
States (32%), Germany (12%), UK (12%), Canada (10%), A few studies investigate the sensitivity of SDSM results
France (6%), Australia (6%), and China (5%). Again, this to model inputs and operation (Mullan et al., 2012); oth-
suggests that the high concentration of SDSM research ers do not use SDSM itself but apply the online predic-
in Canada, UK, and China is atypical of the wider tor sets to alternative downscaling techniques (Chandler
community, and more reflective of the origins of the tool. et al., 2006).
For instance, the SDSM bibliography presently contains As noted above, inter-comparison studies have im-
proved understanding of the capabilities of SDSM and
no items for Germany or France.
downscaling in general. However, continued focus on
these activities is becoming harder to justify given low-
4.2. Sectors of interest levels of uptake of regional climate change scenarios
Table II indicates that water (resource and flood risk for adaptation and resource planning (Fowler and Wilby,
management) has been the most frequently visited sector 2007). This tendency is certainly reflected in the SDSM
by users of SDSM and accounts for 47% of the output outputs: only 5% of the studies approach downscal-
(Scibek and Allen, 2006; Abdo et al., 2009; Futter et al., ing from an adaptation or risk management perspective.
2009; Nyenje and Batelaan, 2009; Yimmer et al., 2009; For example, Coulibaly and Shi (2005), Karamouz et al.
Combalicer et al., 2010; Liu et al., 2011). Other sectors (2011), and Nguyen et al. (2010) consider the conse-
such as agriculture (5%) (Rezaie and Bannayan, 2012), quences of climate change for urban drainage design
urban environment (5%) (Zhou et al., 2009), ecosystems and performance. Whitehead et al. (2006) and Karamouz
(2%) (Adimo et al., 2011), energy/hydropower (2%) et al. (2010) evaluate the effectiveness of different con-
(Noreña et al., 2011), human health/air quality/thermal trol strategies for eutrophication under downscaled cli-
comfort (2%) (Holloway et al., 2008), and tourism/skiing mate scenarios.
(1%) (AGCI, 2006) are represented to a lesser degree.
Some studies address multiple sectors (LCCP, 2002;
Reeves, 2008; Aguilar et al., 2009) but a significant 5. Discussion
fraction (33%) does not specify the sector, concentrating Our meta-analysis of SDSM outputs and patterns of
instead on the technical development or testing of the activity across the wider research community supports

Copyright  2012 Royal Meteorological Society Int. J. Climatol. (2012)


R. L. WILBY AND C. W. DAWSON

the view that downscaling has yet to reach beyond


academia to make significant in-roads to adaptation prac-
tice (Fowler and Wilby, 2007; Wilby and Fowler, 2010).
Nonetheless, there is rapid growth in downscaling output
and the bulk of this literature is focusing on develop-
ment of novel techniques, climate scenario uncertainty
analysis, model inter-comparison, and risk assessment.
Publication trends reveal an overwhelming interest in
water-sector impacts, followed to a lesser extent by agri-
culture. Downscaling for sectors such as human health
and the built environment has received relatively little
attention until recently but these are clearly emerging Figure 5. Number of downscaling publications by region since 1993
topics (Doyon et al., 2008; Holloway et al., 2008; Wilby, identified by the WoS (9 June 2012). Pacific islands include Hawaii,
2008b; Wise, 2009; Zhou et al., 2009; Cueto et al., 2010; Philippines, and Japan.
Chung et al., 2011).
Sectors such as ecosystems/biodiversity and energy
have been largely overlooked by the downscaling com- Second, remotely sensed data such as the Tropical
munity. As far as the authors are aware, there have Rainfall Measuring Mission (TRMM) could bridge or
been only a handful of downscaling studies concerned extend surface records, or provide the predictand (albeit
with potential climate impacts on transport (French at grid-resolution) for downscaling model calibration.
et al., 2010). Nonetheless, downscaling wind speeds and The TRMM multi-satellite precipitation analysis pro-
extreme weather could be useful for assessing future vides 3 hourly, 0.25° × 0.25° latitude–longitude resolu-
energy potential and risks to transport infrastructure and tion data, in near real time since 1998, between latitudes
operations (Sailor et al., 2008; Salameh et al., 2009; de 45° north and south (Huffman et al., 2007). For exam-
Lucena et al., 2010; Traveria, et al. 2010). Aside from ple, Figure 7 compares observed (GPCC), TRMM and
conventional meteorological variables (i.e. temperature, downscaled daily rainfall totals for Khujand in Tajikistan
rainfall, sunshine, and wind speed) and a few unconven- during a 3 year period. Daily rainfall totals at this site are
tional parameters (such as urban heat island, air quality, most strongly correlated with NCEP relative humidity at
or tidal surge) there is little work on other local quantities. 850 hPa (R850).
However, downscaling could be applied to other metrics Climate model uncertainty has become a major focus
that are not represented at the spatial scale of GCMs. for downscaling research: according to both Scopus and
For example, growing degree days, plant phenology, and WoS this topic accounts for over 20% of published output
flowering dates may be linked to regional air tempera- since 1993 (and ∼25% of all SDSM outputs). Studies
tures (Maak and Von Storch, 1997; Avolio et al., 2011); typically quantify uncertainties arising from permutations
wave heights to regional wind climate (Wang et al., 2010; of emission scenario, GCM, downscaling technique,
Reeve et al., 2011); or landslides and slope stability to and impacts model (Wilby and Harris, 2006; Dibike
antecedent/extreme rainfalls (Dehn and Burma, 1999; et al., 2008; Prudhomme and Davies, 2009; Khan and
Chiang and Chang, 2011). Greater attention to indices Coulibaly, 2010; Chen et al., 2011). The trend in down-
that are relevant to hazard management could be one scaling research is towards more exhaustive analysis
way of increasing uptake of downscaled information for of uncertainty (Harding et al., 2012), something which
present climate conditions. presents SDSM-users with a dilemma since only a few
The SDSM and wider statistical downscaling literature forcing scenarios are provided through the CCCSN por-
are geographically biased towards North America and tal. One option is for users to process their own sets of
Europe (Figure 5). Very little peer-reviewed research predictor variables for additional GCMs. However, this
exists for Pacific islands, Polar (Arctic and Antarctic) runs counter to the authors’ view that downscaled scenar-
regions and South America. It can be no coincidence ios are more suited to sensitivity analysis and adaptation
that these are amongst the most data scarce parts of options appraisal than to exhaustive uncertainty analysis
the world – a significant impediment to any statistical (based on imperfect boundary forcing).
downscaling technique. However, this obstacle could be End-to-end analyses typically show that GCMs are
overcome in at least two ways. First, baseline weather the primary source of uncertainty and, although emis-
generator parameters can be estimated from geographic sions and downscaling uncertainty are second-order, their
indices such as longitude, distance from coast or elevation relative magnitude depends on the time-horizon. Future
(Figure 6). Preliminary testing at sites in Africa and the climate may also be investigated alongside other pres-
Middle East suggests that this technique could be used to sures such as land use change (Quilbé et al., 2008).
hindcast and/or infill missing weather data. Furthermore, However, statistical downscaling cannot capture regional
the conditioned weather generator method also yields climate forcing involving land-surface moisture, energy
plausible climate change scenarios even where there are and aerosol feedbacks (Pielke et al., 2011; van der Ent
minimal (paper-based), local meteorological observations et al., 2010). Furthermore, the integrity of all downscaled
(Wilby, 2009). scenarios depends on the extent to which host GCM(s)

Copyright  2012 Royal Meteorological Society Int. J. Climatol. (2012)


SDSM – ONE DECADE ON

(a) (b)

(c)

Figure 6. Example relationships between weather generator parameters [the unconditional wet-day probability (PWET); fourth root of mean
wet-day amounts (R4MEAN)] and longitude for sites in Yemen based on data for the period 1971–2000 (upper panel). Comparison of observed
and simulated daily precipitation totals at Addahi, western coastal plain of Yemen based on weather generator parameters derived from site
elevation, latitude and longitude (lower panel). Source: Wilby (2009). This figure is available in colour online at wileyonlinelibrary.com/journal/joc

but it should affect the ways in which downscaling


models are used. Some are advocating ‘scenario neu-
tral’ approaches in which downscaling informs a plausi-
ble range of climates for sensitivity testing (Prudhomme
et al., 2010). Others are using downscaled scenarios to
model performance of alternative adaptation strategies
for water supply-demand (Lopez et al., 2009) or qual-
ity management (Whitehead et al., 2006). These studies
Figure 7. Monthly rainfall totals for Khujand, Tajikistan estimated by show how downscaling could evolve from a scenario-led,
GPCC (gridded observations), TRMM (gridded satellite), and SDSM
(statistical downscaling model, ensemble member 20). Source: Wilby
uncertainty-laden framework to vulnerability-led modes
(2010b). This figure is available in colour online at wileyonlinelibrary. of application.
com/journal/joc Finally, one recent, novel application of SDSM is wor-
thy of special mention. Jia et al. (2012) used SDSM
to downscale two sets of temperature and precipitation
manifest realistic global features such as the El Niño, series for the years 1961–2000 to assess the drivers of
monsoons and storm-tracks, ocean currents, and so forth. increasing water stress in the Hai River basin, China.
Rigorous examination of some of these macro-scale pro- Five explanatory factors were considered in the finger-
cesses shows major shortcomings in the AR4 models print-based attribution analysis: natural climate variabil-
(Annamalai et al., 2007; Paeth et al., 2008; Liepert and ity, anthropogenic climate change, artificial water use,
Previdi, 2012). land use change, and local human activity (encompassing
Given climate model biases and partial boundary forc- both artificial water use and land use change). The two
ing (i.e. atmosphere only), statistical downscaling of downscaled scenarios were derived from GCM output
future scenarios adds spatial precision but of question- with natural only forcing, and natural with greenhouse
able accuracy (Pielke and Wilby, 2012). Unfortunately, gas forcing. The results showed that for this particular
this fundamental caveat is seldom attached to regional basin, the greenhouse gas forcing may be discounted,
impact assessments, whereas uncertainty due to choice whereas local human activity accounts for 60% of the
of GCM is widely acknowledged. This does not imply observed changes in the water resources during the last
that downscaled scenarios of the future are value-less, 40 years of the 20th century.

Copyright  2012 Royal Meteorological Society Int. J. Climatol. (2012)


R. L. WILBY AND C. W. DAWSON

6. Conclusions adaptation options appraisal. It is hoped that this critique


provides good examples of the capabilities of SDSM,
The past decade has witnessed growing academic interest and pointers to fruitful areas of research in the coming
in statistical downscaling as evidenced by the volume of decade.
research output. At the beginning of the decade it was
sufficient to construct a model, refine methods, and test
performance against observations and other techniques. Acknowledgements
Downscaled scenarios were then driving environmental
Development of SDSM was supported by the Environ-
models to evaluate potential climate risks to society.
ment Agency of England and Wales, the Canadian Cli-
Latterly, multiple downscaled scenarios are being applied
mate Impact Scenarios Group, and Environment Canada.
to multiple impacts models, and the emphasis is shifting
We also thank the user community and early adopters
to more exhaustive characterization of uncertainties. The
who have so enthusiastically embraced the project, not
typical study has either a North American or European
least Jeanna Goldstein who was the first registrant.
water-sector focus. In comparison, data sparse regions in
Africa, Asia, and South America are under-represented in
the downscaling literature. By the close of the decade, it References
is has become clear that statistical downscaling has raised Abdo KS, Fiseha BM, Rientjes THM, Gieske ASM, Haile AT. 2009.
awareness of climate risks but relatively few studies Assessment of climate change impacts on the hydrology of
Gilgel Abay catchment in Lake Tana basin, Ethiopia. Hydrological
are directly informing adaptation planning – the raison Processes 23: 3661–3669.
d’être. Adimo OA, Bosco NJ, Leonard W. 2011. Net primary productivity
Output by SDSM-users broadly mirrors these trends. response to climate change in the Mount Kenya ecosystem. In
Experiences of Climate Change Adaptation in Africa, Chapter 14,
Thanks to many comparison studies the capabilities of the Filho WL (ed). Springer-Verlag: Berlin, Germany.
model are now well understood. From a core of water- Aguilar MY, Pacheco TR, Tobar JM, Quiñónez JC. 2009. Vulnerabil-
related research in Canada, the UK (and now China), ity and adaptation to climate change of rural inhabitants in the central
coastal plain of El Salvador. Climate Research 40: 187–198.
the tool is gradually reaching other sectors and regions. Annamalai H, Hamilton K, Sperber KR. 2007. The South Asian
Emergent topics of interest include the built environment, summer monsoon and its relationship with ENSO in the IPCC AR4
human health, and energy production. Meanwhile, inter- simulations. Journal of Climate 20: 1071–1092.
Aspen Global Change Institute (AGCI). 2006. Climate Change and
est in ecosystems and agriculture continue at a modest Aspen: An Assessment of Impacts and Potential Responses, Appendix
level. Likewise, there have been relatively few examples B, p107–111. Aspen Global Change Institute: Colorado, 147.
of SDSM use for adaptation planning and, to date, all Avolio E, Orlandi F, Bellecci C, Fornaciari M, Federico S. 2011.
Assessment of the impact of climate change on the olive flowering in
have been water-related (urban drainage, flood risk, and Calabria (southern Italy). Theoretical and Applied Climatology 107:
water quality management). Nonetheless, we have sug- 531–540.
gested that SDSM could perform other functions such as Barrow EM, Semenov MA. 1995. Climate change scenarios with high
spatial and temporal resolution for agricultural applications. Forestry
data-infilling or scenario-production for sensitivity testing 68: 349–360.
of adaptation options. Chaleeraktrakoon C, Punlun P. 2010. Statistical analysis and downscal-
Several technical developments are planned for SDSM. ing for the minimum, average, and maximum daily temperatures of
the Chi and Mun River basins. Thammasat International Journal of
These include: online rather than local operation so the Science and Technology 15: 64–81.
software would be platform independent; incorporation Chandler RE, Isham VS, Wheater HS, Onof CJ, Leith N, Frost AJ,
of optional nonlinear transfer functions such as Artificial Segond M-L. 2006. Spatial-temporal rainfall modelling with climate
change scenarios. R&D Technical Report FD2113/TR. Defra,
Neural Networks; and simplified extraction of ensemble London.
statistics. Research continues into ways of applying sta- Chen J, Brissette FP, Leconte R. 2011. Uncertainty of downscaling
tistical downscaling techniques in data sparse regions. method in quantifying the impact of climate change on hydrology.
Journal of Hydrology 401: 190–202.
New downscaling predictor sets will also need to be pro- Chen ST, Yu PS, Tang YH. 2010. Statistical downscaling of daily
cessed once the Climate Model Intercomparison Project precipitation using support vector machines and multivariate
5 (CMIP5) completes their model projections based on analysis. Journal of Hydrology 385: 13–22.
Chiang S-H, Chang K-T. 2011. The potential impact of climate
the Representative Concentration Pathways. change on typhoon-triggered landslides in Taiwan, 2010–2099.
Finally, it is recognized that scenario tools such as Geomorphology 133: 143–151.
SDSM have an important educational role. It is gratifying Christensen JH, Hewitson B, Busuioc A, Chen A, Gao X, Held I,
Jones R, Kolli RK, Kwon W-T, Laprise R, Magaña Rueda V,
to log the steady output of Masters and PhD theses Means L, Menéndez CG, Räisänen J, Rinke A, Sarr A, Whetton P.
that have implemented the model (13 to date). Climate 2007. Regional climate projections. In Climate Change 2007: The
diagnostics lie at the heart of the software, and the data- Physical Science Basis. Contribution of Working Group I to the
Fourth Assessment Report of the Intergovernmental Panel on Climate
handling functions compel users to think deeply about Change, Solomon S, Quin D, Manning M, Chen Z, Marquis M,
the quality of their own meteorological data, as well as Averyt KB, Tignor M, Miller HL (eds). Cambridge University Press:
the predictability of the regional climate (now and in the Cambridge.
Chu JT, Xia J, Xu CY, Singh VP. 2010. Statistical downscaling of
future). An important aspect of this capacity development daily mean temperature, pan evaporation and precipitation for
is to discern when to apply SDSM and when to use other climate change scenarios in Haihe, China. Theoretical and Applied
tools. In particular, chronic uncertainty in GCM boundary Climatology 99: 149–161.
Chung ES, Park K, Lee KS. 2011. The relative impacts of climate
forcing undermines confidence in downscaled scenarios change and urbanization on the hydrological response of a Korean
so these tools are best used for sensitivity testing and urban watershed. Hydrological Processes 25: 544–560.

Copyright  2012 Royal Meteorological Society Int. J. Climatol. (2012)


SDSM – ONE DECADE ON

Cohen S, Neale T (eds). 2003. Expanding the Dialogue on Climate Hashmi MZ, Shamseldin AY, Melville BW. 2011a. Comparison of
Change and Water Management in the Okanagan Basin, British SDSM and LARS-WG for simulation and downscaling of extreme
Columbia. Environment: Canada. precipitation events in a watershed. Stochastic Environmental
Combalicer EA, Cruz RVO, Lee S, Im S. 2010. Assessing climate Research and Risk Assessment 25: 475–484.
change impacts on water balance in the Mount Makiling forest, Hashmi MZ, Shamseldin A, Melville BW. 2011b. Statistical downscal-
Philippines. Journal of Earth System Science 119: 265–283. ing of watershed precipitation using Gene Expression Programming
Conway D, Wilby RL, Jones PD. 1996. Precipitation and airflow (GEP). Environmental Modelling and Software 26: 1637–1646.
indices over the British Isles. Climate Research 7: 169–183. Hay LE, Wilby RL, Leavesley GH. 2000. A comparison of delta
Coulibaly P. 2004. Downscaling daily extreme temperatures with change and downscaled GCM scenarios for three mountainous basins
genetic programming. Geophysical Research Letters 31: L16203. in the United States. Journal of the American Water Resources
Coulibaly P, Shi X. 2005. Identification of the Effect of Climate Change Association 36: 387–397.
on Future Design Standards of Drainage Infrastructure in Ontario. Haylock MR, Cawley GC, Harpham C, Wilby RL, Goodess CM.
Ministry of Transportation of Ontario: Canada. 2006. Downscaling heavy precipitation over the UK: a comparison
Coulibaly P, Dibike YB, Anctil F. 2005. Downscaling precipitation of dynamical and statistical methods and their future scenarios.
and temperature with temporal neural networks. Journal of International Journal of Climatology 26: 1397–1415.
Hydrometeorology 6: 483–496. Hessami M, Gachon P, Ouarda TBMJ, St-Hilaire A. 2008. Automated
Crawford T, Betts NL, Favis-Mortlock DT. 2007. GCM grid box regression-based statistical downscaling tool. Environmental Mod-
choice and predictor selection associated with statistical downscaling elling and Software 23: 813–834.
of daily precipitation over Northern Ireland. Climate Research 34: Holloway T, Spak SN, Barker D, Bretl M, Moberg C, Hayhoe K,
145–160. Van Dorn J, Wuebbles D. 2008. Change in ozone air pollution
Cueto ROG, Martinez AT, Ostos EJ. 2010. Heat waves and heat days over Chicago associated with global climate change. Journal of
in an arid city in the northwest of Mexico: current trends and in Geophysical Research-Atmospheres 113: D22306.
climate change scenarios. International Journal of Biometeorology Huffman GJ, Adler RF, Bolvin DT, Gu G, Nelkin EJ, Bowman KP,
54: 335–345. Hong Y, Stocker EF, Wolff DB. 2007. The TRMM multi-satellite
de Lucena AFP, Szklo AS, Schaeffer R, Dutra RM. 2010. The precipitation analysis: Quasi-global, multi-year, combined-sensor
vulnerability of wind power to climate change in Brazil. Renewable precipitation estimates at fine scale. Journal of Hydrometeorology
Energy 35: 904–912. 8: 38–55.
Dehn M, Burma J. 1999. Modelling future landslide activity base on Jia Y, Ding X, Wang H, Zhou Z, Qiu Y, Niu C. 2012. Attribution of
general circulation models. Geomorphology 30: 175–187. water resources evolution in the highly water-stressed Hai River
Basin of China. Water Resources Research 48: W02513.
Dibike YB, Coulibaly P. 2005. Hydrologic impact of climate change
in the Saguenay watershed: comparison of downscaling methods and Kalnay E, Kanamitsu M, Kistler R, Collins W, Deaven D, Gandin L,
hydrologic models. Journal of Hydrology 307: 145–163. Iredell M, Saha S, White G, Woollen J, Zhu Y, Leetmaa A,
Reynolds R, Chelliah M, Ebisuzaki W, Higgins W, Janowiak D,
Dibike YB, Gachon P, St-Hilaire A, Ouarda TBMJ, Nguyen VTV. Mo KC, Ropelewski C, Wang J, Jenne R, Joseph D. 1996. The
2008. Uncertainty analysis of statistically downscaled temperature NCEP/NCAR 40 year Reanalysis project. Bulletin of the American
and precipitation regimes in Northern Canada. Theoretical and Meteorological Society 77: 437–471.
Applied Climatology 91: 149–170.
Karamouz M, Noori N, Moridi A, Ahmadi A. 2011. Evaluation of
Doyon B, Bélanger D, Gosselin P. 2008. The potential impact of floodplain variability considering impacts of climate change.
climate change on annual and seasonal mortality for three cities in Hydrological Processes 25: 90–103.
Québec, Canada. International Journal of Health Geographics 7: 23.
Karamouz M, Taheriyoun M, Baghvand A, Tavakolifar H, Emami F.
Fowler HJ, Wilby RL. 2007. Editorial: Beyond the downscaling 2010. Optimization of watershed control strategies for reservoir
comparison study. International Journal of Climatology 27: eutrophication management. Journal of Irrigation and Drainage
1543–1545. Engineering 136: 847–861.
French HK, Eggestad HO, Ovstedal J, Jahren PE. 2010. Climate Katz RW, Parlange MB. 1996. Mixtures of stochastic processes:
conditions and consequences for de-icing operations as exemplified application of statistical downscaling. Climate Research 7: 185–193.
by the situation on a motorway and airport at Gardemoen, Norway. Khan MS, Coulibaly P. 2010. Assessing hydrologic impact of climate
Hydrology Research 41: 269–281. change with uncertainty estimates: Bayesian neural network
Futter MN, Helliwell RC, Hutchins M, Aherne J. 2009. Modelling approach. Journal of Hydrometeorology 11: 482–495.
the effects of changing climate and nitrogen deposition on nitrate Khan MS, Coulibaly P, Dibike Y. 2006. Uncertainty analysis of
dynamics in a Scottish mountain catchment. Hydrology Research statistical downscaling methods. Journal of Hydrology 319:
40: 153–166. 357–382.
Gachon P, Dibike Y. 2007. Temperature change signals in northern Koukidis EN, Berg AA. 2009. Sensitivity of the Statistical DownScal-
Canada: convergence of statistical downscaling results using two ing Model (SDSM) to reanalysis products. Atmosphere-Ocean 47:
driving GCMs. International Journal of Climatology 27: 1623–1641. 1–18.
Girvetz EH, Zganjar C, Raber GT, Maurer EP, Kareiva P, Lawler JJ. Liepert BG, Previdi M. 2012. Inter-model variability and biases of the
2009. Applied climate-change analysis: the climate wizard tool. Plos global water cycle in CMIP3 coupled climate models. Environmental
One 4: e8320. Research Letters 7: 014006, DOI: 10.1088/1748-9326/7/1/014006.
Goldstein J, Milton J. 2003. Regional climate scenarios set devel- Liu XL, Coulibaly P, Evora N. 2008. Comparison of data-driven
opment for hydrological impact studies. Proceedings of the 14th methods for downscaling ensemble weather forecasts. Hydrology and
Symposium on Global Change and Climate Variations, American Earth System Sciences 12: 615–624.
Meteorological Society, Long Beach, California. Liu Z, Xu Z, Charles SP, Fu G, Liu L. 2011. Evaluation of two
statistical downscaling models for daily precipitation over an arid
Goodess C, Osborn T, Hulme M. 2003. The identification and
basin in China. International Journal of Climatology 31: 2006–2020.
evaluation of suitable scenario development methods for the
London Climate Change Partnership (LCCP). 2002. A climate change
estimation of future probabilities of extreme weather events. Tyndall
impacts in London evaluation study. Final Technical Report, Entec
Centre for Climate Change Research, Technical Report 4.
UK Ltd.
Guangul SG. 2003. Modelling the effect of climate and land- Lopez A, Fung F, New M, Watts G, Weston A, Wilby RL. 2009. From
use changes on hydrological processes: An integrated GIS and climate model ensembles to climate change impacts: a case study of
distributed modelling approach, PhD thesis, Vrije Universiteit, water resource management in the South West of England. Water
Brussels, Belgium. Resources Research 45: W08419.
Harding BL, Wood AW, Prairie JR. 2012. The implications of climate Lu X. 2006. Guidance on the Development of Climate Scenarios
change scenario selection for future streamflow projection in the within the Framework of National Communications from Parties not
Upper Colorado River Basin. Hydrology and Earth System Sciences Included in Annex I (NAI) to the United Nations Framework Con-
9: 847–894. vention on Climate Change (UNFCCC). National Communications
Harpham C, Wilby RL. 2005. Multi-site downscaling of heavy daily Support Programme (NCSP), UNDP-UNEP-GEF.
precipitation occurrence and amounts. Journal of Hydrology 312: Maak K, Von Storch H. 1997. Statistical downscaling of monthly mean
235–255. air temperature to the beginning of flowering of Galanthus nivalis L.

Copyright  2012 Royal Meteorological Society Int. J. Climatol. (2012)


R. L. WILBY AND C. W. DAWSON

in Northern Germany. International Journal of Biometeorology 41: Traveria M, Escribano A, Palomo P. 2010. Statistical wind forecast for
5–12. Reus airport. Meteorological Applications 17: 485–495.
Maurer EP, Brekke L, Pruitt T, Duffy PB. 2007. Fine-resolution Tryhorn L, DeGaetano A. 2011. A comparison of techniques for
climate projections enhance regional climate change impact studies. downscaling extreme precipitation over the Northeastern United
EOS 88: 504. States. International Journal of Climatology 31: 1975–1989.
Mullan D, Fealy R, Favis-Mortlock D. 2012. Developing site-specific van der Ent RJ, Savenije HG, Schaefli B, Steele-Dunne SC. 2010.
future temperature scenarios for Northern Ireland: addressing key Origin and fate of atmospheric moisture over continents. Water
issues employing a statistical downscaling approach. International Resources Research 46: W09525.
Journal of Climatology DOI: 10.1002/joc.2414. Vidal J-P, Wade SD. 2008. Multimodel projections of catchment-scale
Nguyen VTV, Desramaut N, Nguyen TD. 2010. Optimal rainfall precipitation regime. Journal of Hydrology 353: 143–158.
temporal patterns for urban drainage design in the context of climate von Storch H, Zorita E, Cubasch U. 1993. Downscaling of global
change. Water Science and Technology 62: 1170–1176. climate change estimates to regional scales: An application to Iberian
Noreña JEO, Garcia CG, Conde AC. 2011. A proposal for a rainfall in wintertime. Journal of Climate 6: 1161–1171.
vulnerability index for hydroelectricity generation in the face of Wang XL, Swail VR, Cox A. 2010. Dynamical versus statistical
potential climate change in Columbia. Atmosfera 24: 329–346. downscaling methods for ocean wave heights. International Journal
Nyenje PM, Batelaan O. 2009. Estimating the effects of climate change of Climatology 30: 317–332.
on groundwater recharge and baseflow in the Ssezibwa catchment, Wetterhall F, Bárdossy A, Chen D, Halldin S, Xu C. 2007a. Daily
Uganda. Hydrological Sciences Journal 54: 713–726. precipitation-downscaling techniques in three Chinese regions. Water
Paeth H, Scholten A, Friederichs P, Hense A. 2008. Uncertainties Resources Research 42: W11423, DOI: 10.1029/2005WR004573.
in climate change prediction: El-Nino-Southern Oscillation and Wetterhall F, Halldin S, Xu CY. 2007b. Seasonality properties of four
monsoons. Global and Planetary Change 60: 265–288. statistical-downscaling methods in central Sweden. Theoretical and
Pielke RA Sr, Wilby RL. 2012. Regional climate downscaling – what’s Applied Climatology 87: 123–137.
the point?. EOS 93: 52–53. Whitehead PG, Wilby RL, Butterfield D, Wade AJ. 2006. Impacts of
Pielke RA Sr, Pitman A, Niyogi D, Mahmood R, McAlpine C, climate change on nitrogen in a lowland chalk stream: an appraisal
Hossain F, Goldewijk K, Nair U, Betts R, Fall S, Reichstein M, of adaptation strategies. Science of the Total Environment 365:
Kabat P, de Noblet-Ducoudré N. 2011. Land use/land cover changes 260–273.
and climate: Modeling analysis and observational evidence. Wiley Wigley TML, Jones PD, Briffa KR, Smith G. 1990. Obtaining subgrid
Interdisciplinary Reviews: Climate Change 2: 828–850. scale information from coarse-resolution general circulation model
Prudhomme C, Davies H. 2005. Comparison of different sources of output. Journal of Geohphysical Research 95: 1943–1953.
uncertainty in climate change impact studies in Great Britain. Wilby RL. 1994. Stochastic weather type simulation for regional
Climatic and anthropogenic impacts on water resources variability. climate change impact assessment. Water Resources Research 30:
UNESCO workshop, Montpellier, 22–24 November 2004. 3395–3403.
Prudhomme C, Davies H. 2009. Assessing uncertainties in climate Wilby RL. 1995. Simulation of precipitation by weather pattern and
change impact analyses on the river flow regimes in the UK. Part 2: frontal analysis. Journal of Hydrology 173: 91–109.
future climate. Climatic Change 93: 197–222. Wilby RL. 1997. Nonstationarity in daily precipitation series:
Prudhomme C, Piper B, Osborn T, Davies H. 2005. Climate change implications for GCM downscaling using atmospheric circulation
uncertainty in water resources planning. UKWIR Report 05/CL/04/4, indices. International Journal of Climatology 17: 439–454.
UK Water Industry Research. Wilby RL. 1998a. Modelling low-frequency rainfall events using
Prudhomme C, Wilby RL, Crooks S, Kay AL, Reynard NS. 2010. weather pattern and frontal frequencies. Journal of Hydrology 213:
Scenario-neutral approach to climate change impact studies: 381–392.
application to flood risk. Journal of Hydrology 390: 198–209. Wilby RL. 1998b. Statistical downscaling of daily precipitation using
Qiu B, Jiang J. 2010. Analysis of trends of future temperature in the daily airflow and seasonal teleconnection indices. Climate Research
Bosten Lake Basin based on a statistical downscaling model. Journal 10: 163–178.
of Resources Ecology 1: 268–273. Wilby RL. 2008a. Downscaling future skew surge statistics at
Quilbé R, Rousseau AN, Moquet J-S, Savary S, Ricard S, Gar- Sheerness, Kent. Phase 3 studies – synthesis report. Thames Estuary
bouj MS. 2008. Hydrological responses of a watershed to historical 2100, Environment Agency, 27 pp.
land use evolution and future land use scenarios under cli- Wilby RL. 2008b. Constructing climate change scenarios of urban
mate change conditions. Hydrology and Earth System Sciences 12: heat island intensity and air quality. Environment and Planning B:
101–110. Planning and Design 35: 902–919.
Reeve DE, Chen Y, Pan S, Magar V, Simmonds DJ, Zacharioudaki A. Wilby RL. 2008c. Dealing with uncertainties of future climate: the
2011. An investigation of the impacts of climate change on wave special challenge of semi-arid regions. Proceedings of the Water
energy generation: the Wave Hub, Cornwall, Uk. Renewable Energy Tribune, Expo Zaragoza, Spain.
36: 2404–2413. Wilby RL. 2009. An evaluation of climate data and downscaling
Reeves I. 2008. Climate Change Impacts and Adaptation Saint John, options for Yemen. Report on behalf of the World Bank. 31 pp.
New Brunswick, Canada. Atlantic Coastal Action Plan Saint John: Wilby RL. 2010a. Climate change projections and downscaling for
Canada. Jordan, Lebanon and Syria. Synthesis Report on behalf of the World
Rezaie EE, Bannayan M. 2012. Rainfed wheat yields under Bank. 39 pp.
climate change in northeastern Iran. Meteorological Applications Wilby RL. 2010b. Improving the climate resilience of Tajikistan’s
10.1002/met.268. hydropower sector: Climate variability and change in Tajikistan.
Sailor DJ, Smith M, Hart M. 2008. Climate change implications for Report on behalf of the European Bank for Reconstruction and
wind power resources in the Northwest United States. Renewable Development, London. 30 pp.
Energy 33: 2393–2406. Wilby RL, Fowler HJ. 2010. Regional climate downscaling. In
Salameh T, Drobinski P, Vrac M, Naveau P. 2009. Statistical Modelling the Impact of Climate Change on Water Resources.
downscaling of near-surface wind over complex terrain in southern Fung CF, Lopez A, New M (eds). Wiley-Blackwell Publishing:
France. Meteorology and Atmospheric Physics 103: 253–265. Chichester.
Scibek J, Allen DM. 2006. Modeled impacts of predicted climate Wilby RL, Harris I. 2006. A framework for assessing uncertainties
change on recharge and groundwater levels. Water Resources in climate change impacts: low flow scenarios for the River
Research 42: W11405. Thames, UK. Water Resources Research 42: W02419, DOI:
Souvignet M, Gaese H, Ribbe L, Kretschmer N, Oyarzun R. 2010. 10.1029/2005WR004065.
Statistical downscaling of precipitation and temperature in north- Wilby RL, Dawson CW, Barrow EM. 2002. SDSM – a decision
central Chile: an assessment of possible climate change impacts in support tool for the assessment of regional climate change impacts.
an arid Andean watershed. Hydrological Sciences Journal 55: 41–57. Environmental and Modelling Software 17: 145–157.
Tian P, Tian K, Li J. 2011. Statistical downscaling of future Wilby RL, Hassan H, Hanaki K. 1998a. Statistical downscaling of
temperature and precipitation in upstream of the Ganjiang Basin. hydrometeorological variables using General Circulation Model
Journal of Basic Science and Engineering 19: 57–67. output. Journal of Hydrology 205: 1–19.

Copyright  2012 Royal Meteorological Society Int. J. Climatol. (2012)


SDSM – ONE DECADE ON

Wilby RL, Wigley TML, Conway D, Jones PD, Hewitson BC, Main J, and development planning. International Journal of Climatology 29:
Wilks DS. 1998b. Statistical downscaling of General Circulation 1193–1215.
Model output: a comparison of methods. Water Resources Research Wise K. 2009. Climate-based sensitivity of air quality to climate change
34: 2995–3008. scenarios for the southwestern United States. International Journal
Wilby RL, Hay LE, Leavesley GH. 1999. A comparison of downscaled of Climatology 29: 87–97.
and raw GCM output: implications for climate change scenarios Yimmer G, Jonoski A, Griensven AV. 2009. Hydrological response
in the San Juan River Basin, Colorado. Journal of Hydrology 225: of a catchment to climate change in the upper Beles river basin,
67–91. Upper Blue Nile, Ethiopia. Nile Basin Water Engineering Scientific
Wilby RL, Hay LE, Gutowski WJ, Arritt RW, Takle ES, Leaves-
Magazine 2: 49–59.
ley GH, Clark M. 2000. Hydrological responses to dynamically and
statistically downscaled General Circulation Model output. Geophys- Zhou F, Zhang A, Li R, Hoeve E. 2009. Spatio-temporal simulation
ical Research Letters 27: 1199–1202. of permafrost geothermal response to climate change scenarios in
Wilby RL, Tomlinson OJ, Dawson CW. 2003. Multi-site simulation a building environment. Cold Regions Science and Technology 56:
of precipitation by conditional resampling. Climate Research 23: 141–151.
183–194. Ziervogel G, Zermoglio F. 2009. Climate change scenarios and the
Wilby RL, Troni J, Biot Y, Tedd L, Hewitson BC, Smith DG, development of adaptation strategies in Africa: challenges and
Sutton RT. 2009. A review of climate risk information for adaptation opportunities. Climate Research 40: 133–146.

Copyright  2012 Royal Meteorological Society Int. J. Climatol. (2012)

You might also like