Some slides from a recent overview of the CRU climate scenario generator 'ClimGen' of which I am a co-developer. EU HELIX, ERMITAGE & TOPDAD FP7 & NERC projects have all recently supported development rounds.
2. History:
ClimGen: Update 4/2016
Tim Mitchell & TO (PhD 2002)
TO & CW (TYN Proj)
- ClimGen v1.0 incarnated
- added precipitation i.v. ability
TO & IH (QUEST-GSI Proj) - CMIP3 GCMs added
- QUMP models added ( HadCM3 variants)
TO & CW & MS (ERMITAGE Proj) - GENIE ESM added;
- abilitity to disaggregate monthly precip -> daily (all GCMs)
- write NetCDF
- skip adding obs (V bar)
TO & TM & IH ( TOPDAD Proj ) -Added CMIP5 GCMs v1.5
-HDD/CDD module
-PET module
- RCP-specific patterns
TO & CW (HELIX, 2016) Suitability of patterns across emission scenarios?
CW & TO (NERC Innovation (WQ - SD),
2016)
Ability to disaggregate monthly T -> daily T
3. ClimGen: Update 4/2016
ClimGen v1.5: Capability / functionality
0.5 x 0.5 grid
To 2100, monthly:
- Precip total
- Tas
- Tas Min
- Tas Max
- Cloudiness
- vap
- Precip i.v. (gamma)
CMIP5 GCM patterns
RCPall
RCP2.6
RCP4.5
RCP6.0
RCP8.5
- daily precip
- HDD / CDD
- P.E.T.
(simplified Penman Montieth)
GlobalT
-RCP2.6
-RCP4.5
-RCP6.0
-RCP8.5
*
4. ClimGen: Update 4/2016
Errors: Pattern scaled T (using RCPall pattern) – actual T in the GCM RCP8.5 simulation
at global warming of 4 C (or the year 2066)
HadGEM2-ES ClimGen validation: EU HELIX
5. ClimGen: Update 4/2016
Errors: square the gridded errors, then globally mean them (y axis)
Plot the global errors as function of deltaglobalT
HadGEM2-ES ClimGen validation
error v an individual
HadGEM2-ES run (we had 4 of
them)
error v the ensemble-mean
of all HadGEM2-ES RCP8.5 runs
7. ClimGen: Update 4/2016
NERC Innovation fund (6 months): towards daily T data.
Precip: Samples from gamma:
- shape
- scale
..but fixed by obs
8. ClimGen: Update 4/2016
NERC Innovation fund (6 months): towards daily T data.
Precip: Samples from gamma:
- shape
- scale
..but fixed by obs
Daily T: samples from skew normal
- skewness paramater (<1 or >1) ‘xi’
- variance parameter ‘v’
- location parameter (it’s actually the mean month T)
..big advantage: we allow parameters
to change. We pattern scale them.
9. ClimGen: Update 4/2016
Why a skewnormal distribution?
There is no method of moments solution for sn
and it is computationally slow to find the maximum likelihood estimates....but:
Normal
not normal
not normal