1. Promises and Prospects for Predicting the South Asian Monsoon Jim Kinter COLA George Mason University Special Symposium on the South Asia Monsoon American Meteorological Society Annual Meeting Phoenix, Arizona :: 8 January 2015 Many Thanks To: Rodrigo Bombardi, Paul Dirmeyer, Mike Fennessy, Bohua Huang, Subhadeep Halder, Ed Schneider, J. Shukla, Chul-Su Shin, Bohar Singh Figure credit: National Geographic © 2002

2. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 South Asian Monsoon –One of largest variations of seasonal to interannual climate observed on Earth –Affects lives, property and (largely agrarian) economies of countries inhabited by nearly 25% of human population … –Current state of the science Rudimentary understanding of monsoon dynamics Extremely limited ability to predict monsoon variations

3. South Asian Monsoon –One of largest variations seasonal to interannual climate observed on Earth –Rudimentary understanding of the monsoon remains –Extremely limited ability to predict monsoon variations –Affects lives, property and (largely agrarian) economies of countries inhabited by nearly 20% of human population … either by monsoon flood …

4. …or by monsoon drought

5. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 South Asian Monsoon –One of largest variations seasonal to interannual climate observed on Earth –Affects lives, property and (largely agrarian) economies of countries inhabited by nearly 25% of human population –Current state of the science Rudimentary understanding of monsoon dynamics Demonstrable predictability of monsoon seasonal rainfall Limited ability to actually predict monsoon variations

6. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 F-values for JJAS precip. based on DEMETER data. Gray color indicates not statistically significant at 95% confidence interval (F < 1.4). Thanks to J. Shukla & M. Fennessy F as a Measure of Predictability in DEMETER (5 CGCMs, 9 members, 46 years) Monsoon Rainfall is Predictable

7. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 CFSv2 Reforecasts of JJAS EIMR Precip February IC: Ensemble Mean (24) Bias = -1.48 mm/d ACC = 0.52 RMSE = 0.33 mm/d May IC: Ensemble Mean (24) Bias =-0.78 mm/d ACC = 0.62 RMSE = 0.30 mm/d Monsoon Rainfall Prediction: Only Moderate Skill

8. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 National Monsoon Mission of India and COLA Joint Research to Enhance Monsoon Prediction Ocean-Land-Atmosphere Coupling and Initialization Strategies to Improve CFSv2 and Monsoon Prediction – Sponsored by Ministry of Earth Sciences, Government of India – Improve the coupled ocean-land-atmosphere model (CFSv2) performance. – Improve initialization of ocean and land states in the pre- monsoon season to improve forecasts of onset, monsoon season precipitation. – Improve representation of land-atmosphere feedback in monsoon-dominated regions – Focus on CFSv2 as baseline model

9. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Climate Forecast System version 2 (CFSv2) Global coupled model: Atmosphere, Ocean, Land Surface, Sea Ice Atmosphere: based on the NCEP Global Forecast System (GFS) used for global numerical weather prediction – spectral discretization at T126 resolution (about 100 km grid spacing) – 64 levels in the vertical Ocean: Geophysical Fluid Dynamics Laboratory (GFDL) Modular Ocean Model version 4 (MOM4) – 1/2° horizontal grid spacing; 1/4° meridional grid spacing in the tropics – 40 vertical levels Land Surface: Noah (GFS grid) Sea Ice: a modified version of the GFDL Sea Ice Simulator (MOM4 grid)

10. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 ROLE OF ATMOSPHERIC CONVECTION

11. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Convective Cloud Parameterization: The Simplified Arakawa-Schubert (SAS) Scheme Deep Convection Trigger Mechanism: Level of free convection (LFC) [for parcel with no sub-cloud level entrainment] being within 150- hPa of convection starting point Source: MetEd http://www.meted.ucar.edu/nwp/pcu2/avncp1.htm

12. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Evap Sensible Heated Condensation Framework (HCF) Quantifies how close atmosphere is to moist convection Does not require parcel selection Uses typically measured quantities (needs only q and θ profiles) Is “conserved” diurnally Can be used any time of year or any time of day and interpretation stays the same (Tawfik and Dirmeyer 2014 GRL)

13. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Evap Sensible Threshold Variables: BCL = Buoyant Condensation Level [m] θ BM = Buoyant mixing temperature [K] Heated Condensation Framework (Tawfik and Dirmeyer 2014 GRL) Convection is initiated when: PBL intersects BCL θ 2m reaches θ BM New trigger mechanism for deep convection Original SAS criterion (150-hPa pressure difference).OR. Condensation due to mixing (HCF)

14. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Experiments: Seasonal hindcasts Four member per year (1998 – 2010) From April to October starting on April 1,2,3, and 4 With (HCF) and without (CTRL) the new trigger Thanks to Rodrigo Bombardi Monthly Accumulated Precip. (mm) OBS CTL

15. Thanks to Rodrigo Bombardi JJAS Precipitation

16. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 JJAS Seasonal Precipitation Thanks to Rodrigo Bombardi MeanStDev Improvement of seasonal variability Small but significant improvement of seasonal total

17. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Rainy Season Improved seasonal cycle of precipitation Improved rainy season onset dates. Thanks to Rodrigo Bombardi

18. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Mechanisms The new trigger is an alternative condition Better representation of the background state of convection SAS triggers more frequently Small improvement in the right direction Central India: 16.5-26.5 o N, 74.5-86.5 o E Thanks to Rodrigo Bombardi

19. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 OCEAN INITIALIZATION

20. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Multiple Ocean Initial Conditions (OICs) for CFSv2 1. Four different ocean data sets ECMWF Combine-NV (NEMO) NCEP CFSR (Climate Forecast System Reanalysis) ECMWF ORA-S3 (Ocean Reanalysis System3) NCEP GODAS (Global Ocean data Assimilation System) 2.Variables Monthly mean potential temperature[°C], salinity[g/kg], u[m/s], v[m/s] 3.Pre-Processing Filling up (extrapolation) the land mass to make up potential gaps due to different land-sea masks between each ocean analysis and MOM4 (Note that zonal mean of each variable is initially assigned to grids over the land as an initial guess.) 4. Period: 1979-2009 (Jan. – May)

21. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 OICs Initial month NEMO (4mem*30yrs) CFSR (4mem*31yrs) ORA-S3 (4mem*31yrs) GODAS (4mem*31yrs) May ( 5-month lead) April (6-month lead) March (7-month lead) February (8-month lead) January (9-month lead) * 4 members : 1 st 4 days (00Z) of each month using the atmosphere/land surface conditions from CFSR 600 runs completed 620 runs completed 620 runs completed 620 runs completed CFSv2 Retrospective Forecasts with 4 OICs

22. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 NEMO CFSR ORA-S3 GODAS ES_MEAN JAN ICsFEB ICsMAR ICs CFSv2 Prediction Skill of NINO3.4 (1982-2008) vs. OISST

23. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 NEMO CFSR ORA-S3 GODAS ES_MEAN APR ICsMAY ICs CFSv2 Prediction Skill of NINO3.4 (1982-2008) vs. OISST

24. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Multiple Ocean Analyses Ensemble Mean CFSv2 Prediction Skill of NINO3.4 (1982-2008) vs. OISST

25. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 corresponds to ENSO onset years corresponds to ENSO decay years (1979 – 2008; linear trend removed) 2 Dominant Modes of JJAS mean Precip

26. CFSv2 CMAP 1 st EOF Mode Correlation (1979-2008) lead month NEMO CFSR ORA-S3 GODAS ES_MEAN

27. CFSv2 CMAP 2 nd EOF Mode Correlation (1979-2008) lead month NEMO CFSR ORA-S3 GODAS ES_MEAN

28. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Conclusions ROLE OF CONVECTION – HCF - alternative condition to whether or not deep convection should be initiated - was implemented into the Simplified Arakawa-Schubert scheme in the CFSv2. Better representation of background state of convection Increased frequency at which convective scheme is activated – HCF improves representation of monsoon in CFSv2: Better daily and seasonal precipitation, mean & variability Better seasonal cycle of precipitation, including onset dates ROLE OF OCEAN ICs Substantial dependence on ocean ICs of skill of monsoon rainfall forecasts Multi-analysis initialization results in superior prediction, on average Two main modes of monsoon variability, associated with onset and decay phases of ENSO, are well reproduced and predicted in CFSv2 with most ocean IC data sets

29. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 ABSTRACT There is ample evidence that dynamical coupled models of the Earth's climate system are reasonably good sources of seasonal predictions of the Indian monsoon. There is also evidence that the Indian monsoon is more predictable in theory than can be achieved with current prediction models. There are many sources of this predictability, including the influence of long-lived sea surface temperature anomalies in the tropical Pacific and Indian Oceans and of the soil moisture anomalies in Eurasia that act to alter the circulation and thermodynamic forcing of the atmosphere in the vicinity of south Asia, thereby influencing the precipitation over India during the Asian summer monsoon season. Complex land-atmosphere and ocean-atmosphere interactions, as well as complex interactions of the atmospheric circulation in different regions, all contribute to predictability as well, which is one reason that dynamical models produce predictions that are superior to those obtainable from empirical methods. As part of the Indian National Monsoon Mission, COLA, in collaboration with the Indian Institute of Tropical Meteorology in Pune, India, has conducted experiments with the Climate Forecast System, version 2 (CFSv2) of the U.S. National Centers for Environmental Prediction (NCEP), arguably among the best systems for predicting climate fluctuations on seasonal to interannual time scales. Experiments with the CFSv2 will be described in which the land-atmosphere and ocean-atmosphere interactions, the initialization of the global ocean, and the role of convective clouds are all evaluated for their effects on Indian monsoon prediction skill. It is found that inhibiting the onset of convection, more accurately representing the state of the upper ocean, particularly in the tropics, and correcting biases even in very remote areas (such as the Arctic and North Atlantic) can all significantly improve both sub- seasonal and seasonal mean predictions of the Indian monsoon rainfall, including onset date and geographical distribution of rainfall.

30. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Analysis of Variance: F as a measure of predictability 5 CGCMs, 46 years, 9 ensembles Thanks to J. Shukla

31. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 OBS (CMAP) Mean NCEP CFSv2 Mean JJAS 1982-2010 Precipitation NCEP CFSv2 RMSE NCEP CFSv2 BIAS

32. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 OBSERVATIONS

33. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Tawfik and Dirmeyer (2014) New trigger mechanism for deep convection Original criterion (150-hPa pressure difference).OR. Condensation due to mixing (Heated Condensation Framework - HCF)

34. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Seasonal Precipitation Improvement of seasonal precipitation variability  Small but significant improvement of seasonal totals  Thanks to Rodrigo Bombardi Standard Deviation of JJAS Precipitation Mean JJAS Precipitation

35. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Wind Difference (JJAS) Thanks to Rodrigo Bombardi 200 hPa 850 hPa Shading indicates significant change in precipitation

36. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Central India: 16.5 o N-26.5 o N,74.5 o E-86.5 o E Future Work Thanks to Rodrigo Bombardi

37. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Summary The HCF was implemented into the Simplified Arakawa-Schubert scheme in the CFSv2. It provides an alternative condition to whether or not deep convection should initiate in the model. The HCF trigger resulted in several improvements in the representation of the Indian Monsoon in the CFSv2: Improvement of daily and seasonal precipitation, including seasonal variability Improvement of seasonal cycle of precipitation, including onset dates Mechanisms of how the HCF trigger affects precipitation: 1)Better representation of background state of convection 2)Increased frequency at which convective scheme is activated

38. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Impact of Ocean IC on Prediction of JJAS SST

39. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Impact of Ocean IC on Prediction of JJAS Precip.

40. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 ROLE OF LAND SURFACE

41. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Hot Spots Multi-model results indicate that there are geographic “hot spots” where the atmosphere is responsive to the state of the land surface (soil moisture) = potential predictability. There have been many subsequent studies using models, analyses and observations that corroborate and help explain this result. The Global Land- Atmosphere Coupling Experiment (GLACE) : A joint GEWEX/CLIVAR modeling study Koster, et al., 2004: Science, 1138-. Dirmeyer, et al., 2006: JHM, 1177-. Guo, et al., 2006: JHM, 611-. Koster, et al., 2006: JHM, 590-. Koster, et al., 2004: Science, 1138-. Dirmeyer, et al., 2006: JHM, 1177-. Guo, et al., 2006: JHM, 611-. Koster, et al., 2006: JHM, 590-.

42. Special Symposium on the South Asia Monsoon :: AMS :: Phoenix, AZ :: January 2015 Koster et al., 2011: JHM, 805-. GLACE-2 prediction study (11 GCMs, 1986-1995, forecasts validating during JJA). Places that see the greatest skill impact from realistic land surface initialization have high gauge density (good rainfall data to generate initial soil moisture states) and high land-derived predictability (hot spots). Skill Contributors Land-Derived Predictability GPCP Precipitation Gauge Density 16-30 days 31-45 days 46-60 days Red: Large Improvement Black: No Improvement 2m Temperature Forecast Skill Improvement (Diff in r-squared)