1. Regional climate downscaling theory

2. “Unintelligent downscaling”
IPCC Fourth Assessment Report ensemble range for annual precipitation change across Yemen by the 2050s under SRES A2 emissions (left: driest model; right: wettest model). Data source: Climate Wizard

3. What the climate model centres provide…
300km
50km
downscaling
10km 1m
…what (some think) the climate impacts community needs.
Point

4. Justification for downscaling
...studies of the impacts of projected global warming on a regional scale...necessitates the development and application of scenarios to specific problems... Cohen (1990)
...Even if global climate models in the future are run at high resolution there will remain the need to ’downscale’ the results from such models to individual sites or localities for impact studies... DOE (1996)
...‘downscaling’ techniques, [are] commonly used to address the scale mismatch between coarse resolution global climate model (GCM) output and the regional or local catchment scales required for climate change impact assessment and hydrological modelling... Fowler & Wilby (2007)

5. A typology of downscaling methods
Family
Methods
Dynamical
Variable resolution models
Limited Area/ Regional Climate Models (RCMs)
Statistical
Weather pattern classification
Weather generators
Transfer functions

7. NARCCAP RCM domains
Source:

8. Verifying regional climate model skill
Observed (left column) and RegCM3 simulation (right column) of near surface winds, precipitation and surface temperature for summer
Source: Pal et al. (2007)

9. Verifying regional climate model skill
Comparison of observed (UDEL, left panel) and dynamically downscaled (MMFI, right panel) average winter precipitation (mm/day) for
Source:

10. How an RCM sees complex topography
Source:
Ferranti (2007)

11. Heavy rainfall biases (PRUDENCE)
Estimates of return value (in mm) for 1 day, 5 year event for grid cells. Source: Fowler et al. (2007)

12. Uncertainty in projections (PRUDENCE )
Estimates of percent change in the 1-day 5-year and 10-day 5-year return values, respectively, for each RCM and each season
under the SRES A2 2071–2100 emissions scenario for Southeast England (SEE)
Source:
Fowler & Ekstrom (2009)

13. PRECIS: DIY regional downscaling
PRECIS model projections of changes in summer monsoon rainfall by the 2080s, under SRES A2 and B2 emissions scenarios. Source: Kumar et al. (2006)

14. Regional Climate Models
Strengths
Weaknesses
Limited area
Variable resolution
Enhanced spatial and temporal resolution compared with GCMs
Responsive to multiple drivers (atmospheric, land-surface)
Multivariate output across domain and levels in the atmosphere
Generates internally consistent maps of change
Results depend on the quality of GCM inputs
As computationally demanding as GCMs
Results depend on domain location and size
Results depend on method of boundary forcing
Technically demanding to set up and run

15. Statistical downscaling methods
Applicable to:
Sub-grid scales (small islands, point processes)
Complex/ heterogeneous environments
Extreme events
Exotic predictands
Transient change/ ensembles

16. A downscaling “manifesto”
(Wigley et al., 1990)
Key issues
Predictor selection
Local variations in predictability
Stationarity of scaling
Predictor domain
GCM biases

17. Weather classification schemes to condition daily surface variables

18. Hubert Horace Lamb ( )

19. Weather classification: LWT scheme to condition daily rainfall
Conditional probabilities of rainfall and mean intensity in the Cotswolds, UK associated with the seven main Lamb Weather Types (LWT),
Key:
Anticyclonic (A), Westerly (W), Cyclonic (C), Northery (N), North-westerly (NW), Southerly (S) and Easterly (E) patterns.

20. Weather typing methods
Strengths
Weaknesses
Subjective classification
Analogues
Fuzzy clusters
Self organising maps
Monte Carlo
Hybrid methods
Enhanced spatial and temporal resolution compared with GCMs
Yields physically interpretable linkages to surface climate
Can be applied to surface climate, air quality, flooding, soil erosion, etc.
Compositing of selected events such as extremes
Results depend on the quality of GCM inputs
Requires a classification scheme
Circulation patterns can be insensitive to radiative forcing
May not capture intra- type variations in surface weather

21. Key publications reflecting the early development of daily weather generators

22. Precipitation occurrence process
The transition probabilities for Cambridge, UK are as follows
dry-to-wet (p01) = 0.291
wet-to-wet (p11) = 0.654
Therefore it follows (for a two state model) that
dry-to-dry (p00) = 1 - p01 = 0.709
wet-to-dry (p10) = 1 - p11 = 0.346

23. Precipitation amount distributions
Daily precipitation totals at Addis Ababa, Ethiopia modelled using gamma, fourth root and stretched exponential distributions.

24. A “point-n-click” weather generator
EARWIG:
A “point-n-click” weather generator
Example screen for the Environment Agency Rainfall and Weather Impacts Generator (EARWIG). The software is based on the Neyman-Scott Rectangular Pulse (NSRP) weather generator. See: Kilsby et al. (2007)

25. Weather generator methods
Strengths
Weaknesses
Markov chains
Stochastic models
Spell length methods
Neyman-Scott
Mixture models
Enhanced spatial and temporal resolution compared with GCMs
Simultaneous weather generation at multiple sites
Multivariate outputs
Spatial interpolation of model parameters for data sparse regions
Captures variability across different space and time scales
Results depend on the quality of GCM inputs
Arbitrary adjustment of parameters for future climate estimation
Unanticipated effects on secondary variables from changing precipitation parameters

26. Transfer function approaches
Synoptic controls of London’s urban heat island during the summer of 1995
Grid boxes of GCM data available for downscaling to sites across the UK.

27. Validation of modelled nocturnal UHI intensity for the summer of 1995
Grey lines denote observations, red the modelled UHI

28. Validation of modelled ozone concentrations in central London
Downscaled maximum daily ozone concentrations for Russell Square, London. Source: Wilby (2008)

29. Uncertainty in UHI due to GCM output
Twenty-first century nocturnal urban heat island intensity in London downscaled from four GCMs under SRES A2 emissions. Source: Wilby (2008)

30. Transfer function methods
Transfer functions
Strengths
Weaknesses
Linear regression
Artificial neural networks
Canonical correlation
Kriging
Enhanced spatial and temporal resolution compared with GCMs
Relatively straightforward to apply
Useful for exotic predictands
Applicable to a wide range of time and space scales
Results depend on the quality of GCM inputs
Observed variance typically underestimated
May assume linearity or normality of data
Poor representation of extreme events
Assumes stationarity of the predictor-predictand relationship(s)

31. Summary – the six eras of downscaling
Period
Activities
1950s
Origins in numerical weather prediction
1980s
Rationale and proof of concept
1990s
Method refinement and inter-comparison
2000s
Characterising uncertainty
Theory into practice
2010s?
Towards robust adaptation decision-making