Energetic and precipitation responses in the Sahel to sea surface temperature perturbations

Energetic and precipitation responses in the Sahel to sea surface temperature perturbations Spencer A. Hill Yi Ming, Isaac Held, Leo Donner, Ming Zhao

Motivations Severe uncertainty in rainfall response to anthropogenic warming Fig. 1a, Park et al 215 Sahel P in RCP8.5 runs

Motivations GFDL AM2.1: uniform 2 K SST warming massive Sahel drying. Plausible? AM2.1 JAS δp in AGCMs in +2K experiments Fig. 5 of Held et al 25 Warm colors=drying. So for AM2.1 at least, full coupled response controlled by atmosphere response to mean SST warming

Energetic and precipitation responses in the Sahel to sea surface temperature perturbations GFDL AM2.1: Mean SST warming dries Sahel via enhanced Sahara-Sahel MSE difference Other models: MSE gradient-based drying mechanism robust & linked to climatological convective depth

9 RAS vs. UW Replacing AM2.1 convection scheme causes drying to disappear entirely RAS (a) 4% 12 15 9 9 Default: Relaxed Arakawa Schubert (RAS) (b) +% 12 15 9 UW 1.5.75.45.15.15.45.75 1.5 dimensionless JAS δp in +2 K runs. Top: default, RAS. Bottom: UW convection. 12 9 Replacement: U. Washington (UW), Bretherton et al. 24 As configured for HiRAM UW designed for shallow convection: more quiescent Whereas RAS very active

9 RAS vs. UW Replacing AM2.1 convection scheme causes drying to disappear entirely (a) 4% RAS 12 15 9 9 (b) +% 12 15 9 UW 12 9 Focus on differences in large-scale control climate Rather than convective processes themselves 1.5.75.45.15.15.45.75 1.5 dimensionless JAS δp in +2 K runs. Top: default, RAS. Bottom: UW convection.

MSE budget Column integral: Energetic forcing balanced by circulation diverging MSE F net { u h } + { ω p h } Canonical tropical convection zone balance: F net { ω p h } Forcing drives deep moist convection, c.f. Neelin and Held 1987 Sahel control simulation: F net { u h } Forcing balanced primarily by northerly advection of dry, low-mse Saharan air

(a) +.9 (a) (a) +.9 +.9 δfnet MSE budget δf δf net net (c) +2. (b) +.9 RAS RAS UW UW +2 K: large RAS (b) (b) +.9 +.9 advection response; less impact on UW (d) +7.2 { } δ v h (c) (c) +2. +2. RAS, horizontal (d) (d) +7.2 +7.2 UW, horizontal { } v h h (e) 15.9 (f) 2.4 δ { ω ph } (e) (e) 15.9 15.9 (f) (f) 2.4 2.4 δ { } ω ph ph (g) 2.8 (h).9 δ {h v } (g) (g) 2.8 2.8 RAS, vertical (h) (h).9.9 UW, vertical {h {h } v } 1 convergence 9 5 1 1 5 9 1 W m 2 divergence MSE import MSE export

MSE budget Sahel-Sahara MSE difference increases, which dries the Sahel u v h/ y (a) (b) (c) (d) Enhances drying influence of Saharan inflow Control Effectively upped-ante +2 K mechanism of Chou & Neelin difference More so and over(e) greater depth in RAS than UW Especially in mid- to upper-troposphere (f) RAS UW Control +2 K difference 15 1 5 5 m s 1 2 1 1 2 4 m s 1.2.15.1.5..5 J kg 1 m 1

Other models Do Sahel drying mechanisms in AM2.1 extend to other models? 7 GFDL model variants AM2.1, AM2.1-UW, AM2.5, AM, c9-am, HiRAM, c48-hiram 1 CMIP5 models Those that ran amip and amip4k Uniform SST perturbation: +2 K for GFDL; +4 K for CMIP5 But still δp still mismatch after normalizing

Other models Do Sahel drying mechanisms in AM2.1 extend to other models? Sahel JAS rainfall reduction in 14 of 17 models! outliers = GFDL variants using UW And northerly dry advection enhanced in all

GFDL (a) (c) v ω h h/p Saharan dry air advection into Sahel increases in all models AM2.1 AM2.5 AM-c9 AM HiRAM AM2.1-UW c48-hiram.2.1..1.2. (c) ω h/p (b) (d) δ(v δ(ω h) h/p) AM2.1 AM2.5 Increased dry advection..4.2..2.4. (d) δ(ω h/p) Colors correspond to Sahel δp : drying wettening AM-c9 AM HiRAM AM2.1-UW c48-hiram AM2.1 AM2.5 AM-c9 AM HiRAM AM2.1-UW c48-hiram

CMIP5 (a) v h (c) v h/y Saharan dry air advection into Sahel increases in all models IPSL-CM5A-LR BCC-CSM1-1 NCAR-CCSM4 IPSL-CM5B-LR MPI-ESM-LR MRI-CGCM MIROC5 MPI-ESM-MR FGOALS-G2 CNRM-CM5 (b) δ(v h) (d) δ(v h/y)..2.1..1.2..15.1.5..5.1.15 (c) v h/y (d) δ(v h/y) Not shown: again largely driven by the increase MSE difference IPSL-CM5A-LR BCC-CSM1-1 NCAR-CCSM4 IPSL-CM5B-LR MPI-ESM-LR MRI-CGCM MIROC5 MPI-ESM-MR FGOALS-G2 CNRM-CM5 IPSL-CM5A-LR BCC-CSM1-1 NCAR-CCSM4 IPSL-CM5B-LR MPI-ESM-LR MRI-CGCM MIROC5 MPI-ESM-MR FGOALS-G2 CNRM-CM5

GFDL Ascent profile shallows in all models and relates to control convective depth (a) ω (b) δ ω 2 4 hpa 8 1 5 4 2 1 1 hpa day 1 (c) h/ p 15 1 5 5 1 15 2 25 hpa day 1 Sahel (left) control, (right) anomalous ω in GFDL models, same coloring as before 2 AM2.1 AM2.5

CMIP5 Qualitatively the same as for GFDL but with more scatter (a) ω (b) δ ω 2 4 hpa 8 1 5 4 2 1 1 hpa day 1 (c) h/ p 2 15 1 5 5 1 15 2 25 hpa day 1 Sahel (left) control, (right) anomalous ω for CMIP5, same coloring as before IPSL-CM5A-LR BCC-CSM1-1 NCAR-CCSM4

Our claim Deeper convection in RAS enhances Sahel-Sahara MSE difference more RAS UW Schematic courtesy of Yi Ming Notation: m is MSE Ocean warming and moistening communicated to free troposphere by convection Thus Sahel-Sahara MSE increase sensitive to convective depth

Reanalyses hpa 2 4 8 1 Sahel ascent profiles in three reanalyses are predominantly bottom-heavy Sahel JAS ω in reanalyses ERA-I MERRA NCEP-CFSR 5 4 2 1 hpa day 1 Non-negligible scatter; would like to understand better Potential contamination by convection scheme, by our arguments All separated from GFDL and CMIP5 top-heavy outliers And those are among the worst drying models!

AMS request Greatest obs need: better understanding of discrepancies among reanalyses Reanalyses MIP? I.e. run different reanalysis models w/ identical obs. data And run each reanalysis product with the input data of the others (Not sure if this is feasible from technical standpoint)

Where to find this stuff Email Website Sahel AM2.1 δp Sahel models/obs PhD thesis shill@atmos.ucla.edu http://people.atmos.ucla.edu/shill/ In revision, J. Climate Eventual submission somewhere On my website

begin extra slides

2 Wide SST range 1 5 1 15 2 25 Study roles of large-scale circ. vs. physics by varying δsst 45 4 5 25 2 15 1 5.1.5 5 4 2 (g) P vs. { v h } (W m 2 ) 1 2 2 4 8 Sahel (vertical axis) P and (horizontal axis) u h in AM2.1 with uniform δsst from -15 to +1 K. Control and +2 K outlined. (h) P vs. { ω h/ p } (W m 2 ) RAS: P, P conv, P ls, E, and P E decrease monotonically w/ SST Only P shown here UW: P, P conv, and E increase; P ls, and P E decrease w/ SST Not shown 8 4 2 2 4 So P conv is key discrepancy And that UW E increases more rapidly than P (i) P vs. R t (W m 1 2

Ascent RAS: increased horizontal divergence balanced by anomalous subsidence hpa 8 1 (a) (c) ω (b (d Anomalous subsidence drives anomalous MSE convergence by divergent flow I.e. shallows convection and balances increased dry advection hpa 2 2 4 4 8 8 Control +2 K difference theory 1 1 4 2 1 1 2 (c) hpa day 1 1. (d Control +2 K difference 2 4 Control +2 K difference theory

Ascent RAS: increased horizontal divergence balanced by anomalous subsidence Leading order perturbation budget in free troposphere: u δ h + (δω) p h Rearrange: δω u δ h p h Numerator positive all levels; p h = at 5 hpa (not shown) Thus descent above, ascent below 5 hpa

Ascent RAS: more horizontal MSE divergence, less MSE divergence via subsidence hpa 4 8 Dotted curve: δω u δ h p h 1 2 2 4 4 (a) (c) Control +2 K difference theory ω (b (d Sinking in free troposphere, ascent in boundary layer Amounts to major shallowing of ascent profile hpa 8 8 1 1 4 2 1 1 2 (c) hpa day 1 1. (d 2 4 Control +2 K difference theory

Ascent UW: anomalous free tropospheric descent, but more modest than RAS hpa 8 Same qualitative response, despite weaker magnitude Sinking overcome by moistening influences of ocean warming hpa 1 2 4 (c) Control +2 K difference theory (d Diagnostic for δω from RAS doesn t work Neglects forcing term; more important in UW 8 1 4 2 1 1 2 hpa day 1 1.

RAS vs. UW 2 (a) Deeper convection in RAS enhances 8 Sahel-Sahara MSE difference more RAS ω 1 (b) 2 4 (c) UW h/ p Control +2 K difference theory (d) 4 hpa hpa RAS 8 8 1 Control Little UW 2 convection reaches mid-troposphere where most prone +2 K to suppression difference 4 theory Pa (c) 1 Exacerbated by moist static stability effect (d) 4 2 1 1 2 hpa day 1 UW 1.

Combined δω correlated with δp perfectly for GFDL, insignificantly for CMIP5 δp (mm day 1 K 1 ).2..2.4..8 r = -.91 r = -1. r = -.5 1. 2 4 8 1 5 hpa δω (hpa day 1 K 1 ) GFDL models CMIP5 models combined