Sensi&vity of Surface Climate over the CORDEX-‐SEA Regions to Different Physical
Parameteriza&ons in RegCM4
Fredolin T. Tangang Liew Ju Neng
Ngai Sheau Tieh Chung Jing Xiang Tay Tze Wei
The Na&onal University of Malaysia (UKM)
The 2nd Southeast Asia Regional Climate Downscaling (SEACLID)/CORDEX Southeast Asia Workshop
CORDEX-‐SEA Domain
• East-‐West : ~81.14°E to ~143.86°E
• North-‐South : ~15.04°S to ~ 39.84°N
• Grid : 36 km × 36 km
Revised plan
• East-‐West : ~90°E to ~143.86°E
• North-‐South : ~15.04°S to ~ 27°N
• Grid : 25 km × 25 km
Overview
• Issue on data used for comparison & workaround strategy
• SensiYvity experiments & highlight of results
• Choosing the best combinaYon physical
parameterizaYons for SEACLID/CORDEX-‐SEA
• How do the convecYve parameterizaYons differs?
• Shrinking the domain size while increasing its res.
• Some extras which we have done…
Issue & Workaround Strategy
Annual mean rainfall
(Highest – Lowest) Annual mean temperature
(CRU – APHRO)
One important issue:
VariaYons among the observaYonal products can be large!
Strategy: compare with mulYple gridded products
Annual PrecipitaYon Biases (vs CRU)
Too wet !!!
SYll too wet over the equatorial regions!
N/A
…cont.
Most of the simulaYons produce drier climate over the equatorial regions,
except that used MIT over the land.
PrecipitaYon SpaYal Comparison
SpaYal comparison:
• Generally, the simulaYons have much higher spaYal variaYons.
• CorrelaYon – moderate (~0.5-‐0.7) – complex land/coastal configuraYon.
• Inter-‐model variaYons higher during the winter season.
• Exp14 (Grell(L)/MIT(O)) tends to have smaller RMSE.
Overall, the Grell(L)/MIT(O) experiments seems to have
beker correlaYon and smaller RMSE with CRU
Legend:
PrecipitaYon Temporal Comparison (Annual Cycles)
• Generally, the simulaYons capture the seasonal cycles reasonably well.
Grell/AS, Grell(O)/MIT(L), Kuo
R1 R2 R3 R4
R5 R6 R7 R8 R9 R10 R11
R12 R13 R14
R15 R16 R17 R18 R19 R20
• Zeng (iocnrough=1) usually have smaller RMSEs.
• Generally, pure Grell have smaller RMSE with all 4 observaYonal data.
PrecipitaYon Annual Cycles (vs 4 Obs.): RMSE
3.49 3.21 8.00 6.13 7.64 11.86 5.96 7.77 3.33 2.88 2.85 4.69 3.43 3.98 3.42 4.22 3.75
MIT
MIT(O)/Grell(L) MIT(L)/Grell(O)
Grell/FC Kuo
MIT
MIT(O)/Grell(L)
R1 R2 R3 R4
R5 R6 R7 R8 R9 R10 R11
R12 R13 R14
R15 R16 R17 R18 R19 R20
MIT(L)/Grell(O)
Grell/FC Kuo
PrecipitaYon Annual Cycles (vs 4 Obs.): CorrelaYon Coefficient
0.51 0.55 0.72 0.70 0.71 0.31 0.68 0.67 0.50 0.55 0.52 0.58 0.67 0.67 0.66 0.61 0.64
• Generally, all pure MIT experiments have
beker correlaYon with all 4 observaYonal data
While MIT poor in gepng the annual cycle amplitude correct, overall, it out performs the others in simulaYng
the annual cycle shape correctly
For precipitaYon…
• In terms of spaYal: Grell(L)/MIT(O)
• In terms of temporal: MIT (but with higher RMSE)
Check Point#1
Yet to be able to decide the best
scheme, need to look at other
variables, and study both the
schemes more…
• Generally BOTH MIT and Grell(L)/MIT(O) have stronger interannual variability.
• Grell(L)/MIT(O) has very strong inter-‐annual variability over the Eastern Indian Ocean.
Coefficient of variaYons of the annual precipitaYon (1989-‐2008)
Exp05 -‐ MIT Exp14 – Grell(L)/MIT(O) ERA Interim
TRMM
Annual Mean Temperature Biases (vs CRU)
N/A
…cont. Inter-‐model variaYon is small
Mean Temperature SpaYal Comparison
SpaYal comparison:
• All the simulated mean temperature spaYal pakerns show high fidelity and consistency.
• Higher spaYal correlaYons between the modeled and the observed values.
Legend:
Mean Temperature Temporal Comparison (Annual Cycles)
• Generally, the simulaYons capture the seasonal cycles well, but with notable cold biases.
Grell/AS, Grell(O)/MIT(L)
R1 R2 R3 R4
R5 R6 R7 R8 R9 R10 R11
R12 R13 R14
R15 R16 R17 R18 R19 R20
• Less sensiYve across the different ocean flux
treatments.
• MIT scheme has beker skills.
Mean Temperature Annual Cycles (vs 2 Obs.): RMSE
2.82 2.96 1.21 1.44 1.34 2.12 2.20 2.17 2.79 2.93 2.90 1.65 1.77 1.78 2.29 2.26 2.34
MIT
MIT(O)/Grell(L) MIT(L)/Grell(O)
Grell/FC Kuo
MIT
MIT(O)/Grell(L)
R1 R2 R3 R4
R5 R6 R7 R8 R9 R10 R11
R12 R13 R14
R15 R16 R17 R18 R19 R20
MIT(L)/Grell(O)
Grell/FC Kuo
Mean Temperature Annual Cycles (vs 2 Obs.): CorrelaYon Coefficient
0.88 0.88 0.85 0.86 0.86 0.77 0.88 0.88 0.64 0.67 0.72 0.89 0.90 0.90 0.90 0.90 0.90
• Generally, the simulaYons capture the shape of the annual cycle well, notably the MIT, MIT(O)/Grell(L) and Kuo.
Standard DeviaYon of Mean Temperature
• Comparable to observaYon, but lower interannual variabiliYes in the simulaYons, parYcularly Exp05 using MIT scheme.
ERA Interim CRU
Exp05 -‐ MIT Exp14 – Grell(L)/MIT(O)
Monsoon CirculaYon
Comparing the Yme-‐laYtude cross secYon of U and V wind at 850hPa:
Exp04 – 06 (marked ▲) & Exp13 – 15 (marked ▼) simulated monsoon circulaYon closest to ERA-‐Interim.
Searching for the Best CombinaYon Physical ParameterizaYons
PrecipitaYon Temperature Monsoon CirculaYon
Count the frequencies of each of the 18 (17) experiments appeared as the best 3 for each criteria
…cont.
MIT & Grell(L)/MIT(O) performs beker than the others, with MIT out performs Grell(L)/MIT(O).
MIT Grell(L)/MIT(O)
Where are the areas most sensiYve to the convecYve parameterizaYon?
• Winter : Equator region (land, e.g. Borneo, Sumatera)
• Summer : Indian Ocean (Bay of Bengal), Indo China Region, Central of
Borneo, West Philippines
• ConvecYve parameterizaYon affected the mariYme countries.
Variance of the Simulated Seasonal Rainfall Values [Zeng (iocnrough=1)]
Where are the areas most sensiYve to the convecYve parameterizaYon?
• Winter : East of Philippine/Western pacific.
• Summer : over the Indochina regions (land).
• Regions over the equator are generally less sensiYve to convecYve parameterizaYon when simulaYng the temperature.
Variance of the Simulated Seasonal Temperature Values [Zeng (iocnrough=1)]
• PrecipitaYon: choosing the correct convecYve scheme or correct combinaYon of ocean flux treatment and convecYve scheme is crucial.
• Temperature: less sensiYve to ocean flux treatment compare to deep convecYve parameterizaYon.
• MIT scheme produces too much of rainfall over land.
• Grell(L)/MIT(O) improves the rainfall simulaYons but making the Indo-‐China regions a lot cooler.
• The magnitude of the interannual variability -‐ rainfall: stronger compare to observaYons.
-‐ temperature: comparable with the observaYons but weaker
SuggesYon: MIT, with some expert tuning of MIT convecYve scheme or SUBEX moisture scheme.
Check Point#2
• To assess the impact: short (5 years) experiments (Exp05 and Exp14) were ran while coping with Yme and resources
constrain.
Shrinking the Domain Size while Increasing Its ResoluYon
Experiments sepngs:
-‐Domain:
25 km × 25 km;
~90°E to ~143.86°E , ~15.04°S to ~ 27°N
-‐Run length:
2003 – 2007 (5 years)
other sepngs remain the same as Exp05 or Exp14
Ori. Domain/Res. vs New Domain/Res.
vs. CRU: PrecipitaYon Mean Temperature
Ori New Ori New
Exp 05
Exp 14
Exp 05
Exp 14
NO significant deterioraYon of spaYal biases.
Ori. Domain/Res. vs New Domain/Res. (Exp05)
NO significant deterioraYon in
the quality of the simulated seasonal cycle.
0.67 0.67 6.43 6.15
0.85 0.86 1.83 1.78
Corr. Coef. RMSE
Ori. Domain/Res. vs New Domain/Res. (Exp14)
0.65 0.68 3.24 2.89
0.87 0.89 2.17 2.21
NO significant deterioraYon in
the quality of the simulated seasonal cycle.
Corr. Coef. RMSE
1. Tiedtke (1989) convecYve parameterizaYon scheme (labelled as S77).
2. RegCM version 4.4* (labelled as V44E1)
• V44E1 follows MIT scheme + some TUNINGS…
*RegCM 4.4 fix BATSe land surface roughness calculaYon.
Some extras which we have done…
Experiments sepngs:
-‐Domain:
25 km × 25 km;
~90°E to ~143.86°E , ~15.04°S to ~ 27°N
-‐Run length:
2003 – 2007 (5 years)
SpaYal Comparison
vs. CRU
• Tiedtke scheme (marked as magenta ■) show beker spaYal pakern compared to MIT &MIT(O)/Grell(L)…
• SIGNIFICANT improvement over spaYal pakern simulated using MIT scheme when using RegCM4.4 (marked as orange ▼) in precipitaYon.
Seasonal Cycle (Annual PrecipitaYon)
0.67 0.69 0.65 0.67 0.68 0.66 0.69 6.43 7.90 3.24 6.15 2.89 2.57 4.09
0.85 0.86 0.87 0.86 0.89 0.70 0.87 1.83 1.72 2.17 1.78 2.21 1.44 1.32
Improvement in RMSE
Corr. Coef. RMSE
Placing everything on the balance… (MulY criteria sel. w.r.t. V44E1)
• Experiment using RegCM4.4 doing much beker than those using RegCM4.3
• But Tiedtke scheme seems to out performs the others…
BUT…
…cont.
TRMM
S77 V44E1
Tiedtke seems to dry up the ocean!
• Choosing the smaller domain, higher resoluYon will not deteriorate the simulaYon but beneficial (esp. lesser compuYng resources need)
• Will be best if RegCM 4.4 is used in our future simulaYons.
SuggesYon: MIT, with some expert tuning of MIT convecYve scheme or SUBEX moisture scheme.
Check Point#3
Thank You