1. © Crown copyright Met Office HadGEM3 monsoon simulation sensitivities using idealised experiments June 2013 SAPRISE workshop, Exeter University Richard Levine, Wilfran Moufouma-Okia, Gill Martin, Andrew Turner, Stephanie Bush

2. © Crown copyright Met Office Introduction Focus of systematic error work is on regional convection characteristics over area covering equatorial Indian Ocean and India Start from scratch, where are main problem areas and how do they interact, what about effect of remote biases? Approach: idealised experiments forcing different regions with realistic conditions using regional nudging techniques and regional model simulations 1. Atmosphere-only simulations with prescribed SST systematic monsoon bias seen in all configurations, need ocean coupling for correct representation of processes, however, coupling results in SST biases with additional detrimental impacts on monsoon 2. Coupled simulations what forces main SST biases

3. © Crown copyright Met Office Atmosphere-only runs with prescribed SST

4. © Crown copyright Met Office Methodology Control: GA3.0 / GA2.0 N96 (~135km) and N216 resolution (~60km) NWP 5-day forecast for 2010 at N512 resolution (~25km) Nudging experiments: Nudging of theta (indirectly affecting moisture) and U,V from model level 4 upwards to ERA interim re-analysis (includes interannual variability) Applied both globally and regionally in separate experiments Regional climate runs: CORDEX WASIA domain ~50Km resolution 6-hourly forcing at boundaries from ERA interim re-analysis

5. © Crown copyright Met Office HadGEM3 JJA precipitation and 850hPa wind biases systematic climate errors: little sensitivity to resolution up to N216 (60km) (enhanced India rainfall in recent (GA5.0) N512 (25km), coincides with substantially enhanced monsoon depression activity) error pattern same in RCM, magnitude reduced: partly locally forced climate time-scale error pattern develops within few days in global NWP

6. © Crown copyright Met Office GA5.0 Monsoon depressions: N512 vs N96 total rainfall in years with depressions trajectories and rainfall contribution total minus depression rainfall ~10x more strong systems fully formed cyclonic systems N512 N96 ~3x more weak systems includes weaker cyclonic disturbances N512 N96

7. © Crown copyright Met Office Seasonal cycle of main precipitation biases Excessive WEIO and Himalayan rainfall biases largest during spring and summer  Coincides with timing of monsoon development and lack of C India rainfall Lack development of WEIO rainfall in RCM coincides with enhanced development of C India rainfall (as also seen with many physics changes)

8. © Crown copyright Met Office Nudging experiments global nudging: reference for best obtainable simulation using regional nudging techniques Global nudging reduces biases: almost eliminates WEIO bias, rainfall moves north from equator, monsoon flow (as expected) near perfect, however Indian dry bias partly remains with preferential rainfall over Himalayas and surrounding ocean Fundamental land / orographic precipitation problem which is insensitive to large-scale circulation

9. © Crown copyright Met Office EIO nudging suppresses local wet bias and feedbacks on Indian dry bias and N India flow bias SASIA nudging maximum impact on Indian rainfall, weaker feedback on WEIO rainfall and convergence Nudging experiments: regional

10. © Crown copyright Met Office TROP PAC nudging: smaller impact on mean-state, improves large-scale monsoon IAV strengthens NW Pacific sub-tropical high, weakens monsoon outflow, enhances convergence over BoB / SCS these changes also seen with other experiments improving Indian rainfall, Indian error develops first, Pacific errors not yet seen in NWP 5-day forecast, so feedback India  W Pacific feedback probably strongest Nudging experiments: regional

11. © Crown copyright Met Office Sensitivity to boundary locations in RCM Small further improvement by removing effect of WEIO bias (already small in RCM, due to boundary constraints on EIO convergence) Large improvement by removing effect of Himalayas bias  Indian bias sensitive to orographic bias

12. © Crown copyright Met Office No Tibetan Plateau: “minor” weakening of Himalayan rainfall and flow into N India, only small impact on ΔTT (red vs black lines) No Tibetan Plateau and Himalayas: almost completely removes monsoon from Indian subcontinent, major impact on ΔTT (green) Foiling over Himalayas: substantially weakens Himalayan rainfall and associated flow, enhances C/N India rainfall and monsoon trough flow, large impact on ΔTT (blue) Sensitivity to orography in GCM Monsoon sensitivity in HadGEM3 mainly from steep Himalayan orography, not TP [cf. Boos and Kuang 2010] Barrier effect and elevated heating over Himalayas both important [foiling experiment perhaps could extend further eastwards over rainfall max]

13. © Crown copyright Met Office Sensitivity to steepness of Himalayas original orography smoothing slopes shifts rainfall southwards similar to RCM experiment with change to northern boundary sensitivity both to mean and sub-grid orography  may be issue with dynamics / convection (and coupling) near steep orography, emphasized by strong monsoon feedbacks

14. © Crown copyright Met Office Coupled runs

15. © Crown copyright Met Office HadGEM3 N Indian Ocean cold SST bias (Levine and Turner 2012, Clim Dyn) © Crown copyright Met Office rainfall (colours) and vertically integrated moisture flux (vector) anomalies ~30% reduction in summer monsoon rainfall in coupled model compared to equivalent AMIP run effect of cold Arabian Sea SST bias on local evaporation and moisture fluxes during summer coupled model SST bias

16. © Crown copyright Met Office Development of SST bias bias develops in winter, sustained into (early) monsoon season initialised coupled simulations show bias does not develop with spring-time or later initialisation

17. © Crown copyright Met Office N Arabian Sea SST bias common in CMIP3/CMIP5 (Marathayil et al 2013 ERL, Levine et al 2013 Clim Dyn) In CMIP3/CMIP5 SST bias develops in winter due to excessively strong winter monsoon, related to wide-spread continental cold surface temperature bias and equatorial wet bias

18. © Crown copyright Met Office Methodology Controls: Gregorian (nudging) / 360-day (1.5xENT) version of HadGEM3-AO GA4.0, but with GA3.0 entrainment/detrainment rates (increases sensitivity of equatorial convection to 1.5xENT in AMIP tests) Nudging experiments: nudging theta (indirectly affecting moisture) over subarea for model levels 2-20 (approximately up to ~3000m) to ERA interim re-analysis (includes IAV.. ) 27 year simulations try and reduce cold continental temperature bias 1.5xENT experiments: regionally applied 1.5x convective entrainment over WEIO 30 year simulations try and weaken WEIO convection

19. © Crown copyright Met Office HadGEM3 coupled experiments with regional nudging reducing widespread continental cold surface bias weakens meridional winter monsoon circulation and reduces SST cold bias surface temperature anomalies (colours), 10m wind anomalies (vectors) precipitation anomalies (negative: red contours, positive: blue contours) caveat: orography mismatch with nudging?

20. © Crown copyright Met Office Enhanced equatorial entrainment experiment Aimed at weakening equatorial convection

21. © Crown copyright Met Office 1.5x entrainment weakens WEIO convection, though still large bias remaining, no effect on meridional surface winds BOX 1 BOX 2

22. © Crown copyright Met Office 1.5x entrainment dominant changes to large-scale circulation are zonal iso meridional caveat: max. weakening of WEIO convection is 25%

23. © Crown copyright Met Office Conclusions In AMIP simulations: Indian monsoon systematic error is mainly locally forced in area covering India, N and equatorial Indian Ocean Results suggest Indian land rainfall, WEIO rainfall, and Himalayan rainfall biases would exist on their own, but also feedback on each other Latest configuration shows progressive sensitivity to resolution (N96–N216–N512), with potential role for monsoon depressions Additional dry bias in coupled simulations forced by SST bias, which originally develops due to strong winter monsoon simulations in progress suggest that this is mainly (caveat) due to continental cold surface temperature bias (Arabian peninsula, Iranian plateau, N India, Tibetan plateau) continental T bias develops rapidly in NWP simulations, widespread valley cooling problem?

24. © Crown copyright Met Office The end