1. © Crown copyright Met Office ACRE working group 2: Downscaling David Hein, Richard Jones, Chloe Eagle & Grace Redmond Research funded by

2. © Crown copyright Met Office Overview Relevance Practical requirements Science questions What needs coordinating Simulation of the 1894 floods in southern England

3. © Crown copyright Met Office Relevance – what downscaling linked to ACRE has to offer Reconstruction of past weather events 20CR provides a reconstruction of the evolution of the past 130 years of weather at relatively coarse resolution (2x2 degrees) Downscaling this reconstruction can provide: High resolution reconstructions of specific weather events Good estimates of decadal variability of high resolution climate features Improving statistical downscaling Training statistical downscaling on 20CR data combined with long station records where available would provide more robust statistical transfer functions

4. © Crown copyright Met Office Benefits of Regional Climate Modelling “Teaching Grandmother to suck eggs” Teaching grandmother to suck eggs is an English-language saying, meaning that a person is giving advice to someone else about a subject that they already know about (and probably more than the first person).English-language

5. © Crown copyright Met Office What is a Regional Climate Model (RCM)? Covers a limited area of the Earth’s surface instead of the entire Earth Like GCMs or Observational Re-analyses, RCMs contains representations of the atmosphere, land and surface, and generate weather (and therefore climate)

6. © Crown copyright Met Office Resolution is important (example)

7. © Crown copyright Met Office Resolution is important (example)

8. © Crown copyright Met Office What is a Regional Climate Model (RCM)? Their main advantage is that they allow for higher resolution climate modelling. In most cases, higher resolution = more useful and higher quality information

9. © Crown copyright Met Office Winter precipitation over Great Britain

10. © Crown copyright Met Office The Importance of Decadal Variability Results from a 3-member ensemble projection for Europe using identical models and emissions to sample impact of natural variability Patterns of change locally are often different, especially in summer Differences are due solely to the influence of natural internal variability Run 1, winter Run 2, summer Run 3, winter Run 2, winter Run 1, summer Run 3, summer -80 -40 -20 -10 -5 5 10 20 40 80 Change (%) Change in upper 5% of wet days by 2071-2100 for SRES A2 scenario

11. © Crown copyright Met Office Practical requirements – I Access to 20CR 20CR comprises a 56-member ensemble reconstruction of the past ~130 years at 6-hourly intervals which poses significant data acquisition issues Storage of and making accessible the data requires significant technical and human resources – as does obtaining and manipulating the data Interfacing to dynamical downscaling Dynamical downscaling models require interfaces to be built in order to apply 20CR as driving data for the models

12. © Crown copyright Met Office Practical requirements – I

13. © Crown copyright Met Office Practical requirements – I

14. © Crown copyright Met Office Practical requirements – I

15. © Crown copyright Met Office Practical requirements – I I Ensemble member selection Downscaling the ensemble mean of 20CR does not make sense scientifically – high frequency transients, especially in relatively data-sparse areas, will be smoothed out thus high frequency variability in the downscaled reconstruction will be underestimated Downscaling the full 56 members will be impractical in many applications and thus a methodology and post-processing will be required to select a subset of the ensemble to downscale

16. © Crown copyright Met Office Science questions Estimating high resolution climate variability: Downscaling the full 56 ensemble members could provide estimates of small-scale climate variability: e.g. how predictable is a particular extreme event given a certain large-scale forcing? What is the value of 20CR + downscaling in estimating full spectrum multi-decadal variability? Can this be quantified (and errors estimated) via a validation of a downscaling of 20CR where data are available? Can we deduce the drivers of damaging weather events through an analysis of diagnostics from 20CR downscaled reconstructions of them?

17. © Crown copyright Met Office So what activities related to downscaling need co-ordinating? Three areas would clearly benefit from coordination: Data dissemination and post-processing requirements Guidance on ensemble application and selection Information on scientific findings and research activities We would be interested to hear any other ideas or offers of assistance in any of these

18. © Crown copyright Met Office Some initial activities related to ACRE working group 2 Development of an interface to drive the Hadley Centre regional climate modelling system PRECIS with 20CR Application of PRECIS to study the rainfall associated with the 1894 Thames/Southern England flooding Downscaling of several ensemble members for analysis of issues surrounding predictability and ensemble selection Incorporation of the interface in the next version of PRECIS (V2) to be released to the PRECIS community in a few months

19. © Crown copyright Met Office Simulating the precipitation drivers of the 1894 flood on the Thames

20. © Crown copyright Met Office Overview Thames flood background Relevance of 20CR in providing large-scale context and data for downscaling to simulate the flood drivers Results from the simulations Reproducibility of the events Next steps – simulating Thames river flow

21. © Crown copyright Met Office Thames flood background Precipitation associated with the 1894 floods -- the large-scale situation and relevant weather events

22. The weather in southern Britain in October and November 1894 Very dry weather in September and the first half of October 1894......followed by a three week wet period that lasted until mid November where depressions frequently passed over Britain This resulted in some of the worst flooding ever experienced in many parts of the south of England from Cornwall to Kent Gales also caused widespread damage and flooding around the English coasts © Crown copyright Met Office Source: thamesweb.co.uk

23. The weather in southern Britain in October and November 1894 Source: Symons, British Rainfall, 1894

24. The weather in southern Britain in October and November 1894 In many places over one quarter of the annual mean rainfall was recorded between 20 th October and 14 November (26 days) Totals were in excess of 400mm in parts of Devon South-westerly winds Monthly mean temperature was 0.5-1°C above average for November Source: Brugge (1994), The floods of October and November 1894 in southern Britain. Weather, 49, 383-390

25. The River Thames While it is best known because it flows through London, the river Thames flows alongside several other towns and cities, including Oxford and Reading It is the longest river entirely in England and the second longest river in the UK (346km or 215miles) In total its basin covers an area of about 14,250km 2 The catchment area covers a large part of South Eastern and Western England and the river is fed by over 20 tributaries © Crown copyright Met Office Source: The UK Environment Agency

26. The River Thames Tidal in London with a rise and fall of 7 metres (23 ft), the tides reaching up to Teddington Lock Catchment area of 9950 km 2 above Teddington (non tidal Thames) Diverse in topography, land use and geology Average annual rainfall is about 710mm The river flow varies seasonally - the average January flow is four times that for August

27. The Thames Floods of November 1894 © Crown copyright Met Office Source: Symons and Chatterton (1895), The November Floods of 1894 in the Thames Valley, Quarterly Journal of the Royal Meteorological Society George James Symons (1838-1900) The father of British Rainfall

28. The Thames Floods of November 1894 Highest flow on record (since 1883) of 1060m 3 /s Peak re-assessed at 800m 3 /s (Marsh et al. 2005: The River Thames flood of November 1894 - a reappraisal of the maximum flow) Source: abingdonwalks.co.uk

29. The Thames Floods of November 1894

30. 20CR: large-scale context and data for downscaling 20CR provides regionally complete reconstruction of the evolution of the atmosphere leading up to/during the event Comparison with available observations provides validation Reconstruction at coarse spatial resolution so no detailed weather Six-hourly 3-D atmospheric variables from 20CR are suitable for downscaling to reconstruct detailed historical weather Example here uses the PRECIS regional climate modelling system to downscale the evolution of the weather over 1894 Focus on the precipitation preceding and during the 1894 Thames flood

31. Experimental Design A subselection of the 2x2 degree global land sea mask used in the 20CR. England is represented by a few boxes in red.

32. Experimental Design

33. Experimental Design: Domain 110 grid boxes east-west 100 grid boxes north-south 25 kilometre horizontal resolution (~0.22 ̊ x 0.22 ̊ ) Co-incident subdomain of the EU Ensembles minimum domain

34. Experimental Design: Domain

35. Ideally, a larger domain would have been used to get a ‘better picture’ of what was happening elsewhere in Europe and to allow for observational constraints on the 20CR to affect the output.

36. © Crown copyright Met Office Experimental Design: Spin-up The model takes some time for all of its components to reach full equilibrium The full simulation length must take this into account ~1 year should added to the overall simulation length to account for the spin-up period Soil variables at equilibrium Atmospheric variables at equilibrium Up to 1 week ~1 year START

37. © Crown copyright Met Office Experimental Design: Spin-up + Ensemble runs One ensemble member (37) was run from 00:00z 1 Dec 1893 to 00:00z 1 st Nov 1894 to act as a spin-up period Five 20CR ensemble members (17,25,37,42,51) were then used to drive PRECIS for November 1894 using the “spun-up” 1 st November 1894 exact model state (i.e. the model “dump”) from the Member 37 run acting as initial conditions for each ensemble member. Each of the five members has corresponding six hourly sea surface temperature and sea ice thickness values acting as surface boundary conditions (necessary as PRECIS/HadRM3P is an atmosphere-only model). Member 37 was allowed to run until 31 Dec 1894

38. And for some results…

39. Hourly means on 14 Nov 1894 Units: mm/hour Member 17

40. Hourly means on 14 Nov 1894 Units: mm/hour Member 25

41. Hourly means on 14 Nov 1894 Units: mm/hour Member 37

42. Hourly means on 14 Nov 1894 Units: mm/hour Member 42

43. Hourly means on 14 Nov 1894 Units: mm/hour Member 51

44. Daily precipitation data from 1 st to 15 th November 1894 Units: mm/day Member 37

45. Time series of precipitation Member 17 Member 25 Member 37 Member 42 Member 51

46. Time series of surface temperature Member 17 Member 25 Member 37 Member 42 Member 51

47. Daily pressure data from 1 st to 15 th November 1894 Units: hPa Member 37

48. © Crown copyright Met Office Reproducibility of the events

49. Rainfall accumulation from 20 th October to 14 th November Source: Brugge (1994), The floods of October and November 1894 in southern Britain. Weather, 49, 383-390 Member 37 PRECIS output in mm

50. Rainfall monthly mean November 1894 Units: mm/day Members 17,25,37,42,51

51. Rainfall mean 1 st -14 th November 1894 Units: mm/day Members 17,25,37,42,51

52. BONUS! The North Sea storm of 22 Dec 1894 Units: mm/hour Member 37 Hourly mean Precip for 00z to 08:00

53. © Crown copyright Met Office Next Steps

54. Next steps Simulating with a larger domain (and intercomparison of results) Simulating with more ensemble members Analyse spread of 20CR and PRECIS reconstructions Analyse frequency of simulated precipitation matching the observed (i.e. predictability of the actual events given the large-scale situation) Use precipitation and other PRECIS outputs to drive a river flow model of the Thames

55. © Crown copyright Met Office And now for something completely different...

56. Rob Allan’s desk

57. Modelled Desk Appearance ca. 2020 Source: Hein, 2011

58. Rob Allan Papers Digitisation & Recovery Effort R.A.P.D.R.E.

59. © Crown copyright Met Office Questions Research funded by