Dynamics of the boreal summer African monsoon in the NSIPP1 atmospheric model

Transcription

1 Climte Dynmics (2005) 25: DOI /s x Alessndr Ginnini Æ R. Srvnn Æ Ping Chng Dynmics of the orel summer Africn monsoon in the NSIPP1 tmospheric model Received: 9 Ferury 2005 / Accepted: 15 June 2005 / Pulished online: 13 August 2005 Ó Springer-Verlg 2005 Astrct A nine-memer ensemle of simultions with stte-of-the-rt tmospheric model forced only y the oserved record of se surfce temperture (SST) over is shown to cpture the dominnt ptterns of vriility of orel summer Africn rinfll. One pttern represents vriility long the Gulf of Guine, etween the equtor nd 10 N. It connects rinfll over Afric to the Atlntic mrine Intertropicl Convergence Zone, is controlled y locl, i.e., estern equtoril Atlntic, SSTs, nd is internnul in time scle. The other represents vriility in the semi-rid Shel, etween 10 N nd 20 N. It is continentl pttern, cpturing the essence of the Africn summer monsoon, while t the sme time displying high sensitivity to SSTs in the glol tropics. A lnd tmosphere feedck ssocited with this pttern trnsltes precipittion nomlies into coherent surfce temperture nd evportion nomlies, s highlighted y simultion where soil moisture is held fixed to climtology. As consequence of such feedck, it is shown tht the recent positive trend in surfce temperture is consistent with the ocen-forced negtive trend in precipittion, without the need to invoke the direct effect of the oserved increse in nthropogenic greenhouse gses. We dvnce plusile mechnisms y which the lnce etween lnd ocen temperture contrst nd moisture vilility tht defines the monsoon could hve een ltered in recent decdes, resulting in persistent drought. This A. Ginnini (&) Interntionl Reserch Institute for climte prediction, The Erth Institute t Columi University, P. O. Box 1000, Plisdes, NY , USA E-mil: lesll@iri.columi.edu Tel.: Fx: R. Srvnn Ntionl Center for Atmospheric Reserch, Boulder, CO, USA P. Chng Deprtment of Ocenogrphy, Texs AM University, College Sttion, TX, USA discussion lso serves to illustrte wys in which the monsoon my e pertured, or my lredy hve een pertured, y nthropogenic climte chnge. 1 Introduction One of the outstnding prolems of recent times in tropicl climte dynmics hs een to understnd the cuse(s) of the recurrent droughts tht plgued the Africn Shel during the 1970s nd 1980s. The Shel, the trnsition region etween the humid tropicl rinforest to the south nd the Shr desert to the north, is chrcterized in the climtologicl men y steep meridionl grdient in totl nnul rinfll one order of mgnitude decrese from 1,000 mm ccumulted t its equtorwrd limit to 100 mm ccumulted t its polewrd limit nd y vriility tht increses s the men decreses, with the stndrd devition rising from 10 20% of the men t the southern edge to 50% t the northern edge (Nicholson 1980). The high vriility, hence the potentil for high vulnerility to climte of this semi-rid region is consequence of its geogrphicl loction, t the northern edge of the re influenced y the Africn monsoon in its nnul meridionl migrtion following the sun. The riny seson t the polewrd edge of the Shel is on verge limited to the three-month period immeditely following the northern summer solstice, when the monsoon reches its most northern loction. Filure of the monsoonl circultion to migrte north during this short window of time cn hve serious consequences for n environment tht is lredy dry for the etter prt of yer. Totl nnul rinfll in the Shel ws ove verge during the 1950s nd erly 1960s, then stedily decresed from the lte 1960s through the 1970s nd 1980s. The decline ws punctuted y mjor widespred episodes of drought nd fmine, e.g., in nd (Nicholson 1979, 1980, 1993; lso see, e.g., issues of the Ntionl Geogrphic Mgzine in April 1974, August

2 518 Ginnini et l.: Dynmics of the orel summer Africn monsoon in the NSIPP1 tmospheric model 1975, nd August 1987). The persistence of nomlously dry yer-to-yer conditions in the lte 1960s nd erly 1970s (Glntz nd Ktz 1985) prompted scientists to hypothesize tht something hd gone wry in the environment of tropicl North Afric, nd tht chnge in the form of humn-induced desertifiction, e.g., loss of vegettive cover cused y the frming nd overgrzing of mrginl lnds to stisfy the needs of n everincresing popultion, hd set in (Ottermn 1974; Chrney 1975). The recent success of tmospheric models, of simpler (Zeng et l. 1999) or more complex design (Bder nd Ltif 2003; Ginnini et l. 2003) in replicting the sptio-temporl chrcteristics of oserved Shel rinfll vriility when forced only with the oserved history of se surfce temperture (SST) scores decisive point in fvor of n lterntive hypothesis, tht climte in the Shel is sensitive to glol conditions, more precisely to glol tropicl SST nomlies, s hypothesized y erlier oservtionl nd modeling studies (Lm 1978, 1978; Follnd et l. 1986; Plmer 1986; Rowell et l. 1995; Jnicot et l. 1996; Fontine et l. 1998). Such success hs revitlized the interest in the potentil for ppliction of sesonl climte prediction in the Shel to erly wrning systems in food security nd helth (Thomson et l. 2000; Trore 2003; Plmer et l. 2004). The purpose of this study is to document in detil sttisticl nd dynmicl spects of the simultion of the orel summer Africn monsoon otined with NSIPP1 version 1 of the tmospheric model developed t the Ntionl Aeronutics nd Spce Administrtion (NASA) s Goddrd Spce Flight Center in the frmework of the Sesonl to Internnul Prediction Project. A nine-memer ensemle of integrtions forced only with oserved SST from 1930 to 2000 (Ginnini et l. 2003) is t the core of our investigtion. Additionl integrtions id in the ssessment of the roles of lnd tmosphere interction (the fixed- integrtion), nd of the SST ptterns sttisticlly relted to the recent trend in Shel rinfll (the Shel integrtions). The reder is referred to Section 2 nd Tle 1 for detils on the model nd integrtions. The dominnt ptterns of Africn rinfll vriility during northern summer pper to e united in their common, ocen-forced nture, ut distinct in tht one pttern, representing the Gulf of Guine cost, is inextricly tied to the ocenic Intertropicl Convergence Zone (ITCZ), while the other, representing the Shel, defines the continentl monsoon. Leel et l. (2003) nd Gu nd Adler (2004) hve similrly rgued for dynmicl differentition t intr-sesonl time scles etween the chrcter of convection long the Gulf of Guine, t 5 N, nd cross the Shel, t 15 N, with jump occurring etween the two from the Gulf of Guine polewrd into the Shel in June. While oservtionl studies in the 1970s nd erly 1980s emphsized the role of tropicl Atlntic SSTs (Hstenrth nd Lm 1977; Lm 1978, 1978; Lough 1986), work in the lte 1980s nd 1990s presented more complex picture of Africn rinfll vriility, with incresing evidence for the role of the tropicl Pcific nd Indin Ocens (Wolter 1989; Shinod nd Kwmur 1994; Rowell et l. 1995; Jnicot et l. 1996; Wrd 1998; Jnicot et l. 2001; Rowell 2001). Since the influences of tropicl Atlntic nd Pcific ptterns of climte vriility on the Africn monsoon hve lredy een discussed t length in the literture, we focus our ttention on the Indin Ocen. We re interested not simply in the role of SST nomlies s forcing mechnism for tmospheric nomlies, ut lso, keeping in mind the coupled nture of Indin Ocen climte vriility (Gdgil et l. 1984; Kirtmn nd Shukl 2002; Wester 2003; Krishn Kumr et l. 2005), in the role of persistent deep convective nomlies (Thiw nd Kumr 2001; Gdgil et l. 2003, 2004). In Section 2 we riefly descrie the model nd integrtions performed, nd the oservtionl dt used. In Section 3 we present sttisticl nlysis of the vriility of precipittion, nd covriility of surfce temperture, in oservtions nd in the model. Section 4 contins chrcteriztion of the model s externlly forced nd internl modes of Africn climte vriility. This nlysis is crucil to highlighting fundmentl differences etween the mechnisms tht control rinfll vriility in the Shel nd long the Gulf of Guine cost. Section 5 discusses the role of the Indin Ocen in the context of lrge-scle teleconnections nd climte chnge. Section 6 presents some conclusions. 2 Model set-up nd oservtionl dt 2.1 Model description nd simultion set-up The climte model used is NSIPP1, version 1 of the tmospheric generl circultion model developed t Tle 1 Summry of integrtions nlyzed. Also see text nd the NSIPP wesite, t Ensemle Nme Period Covered Ensemle Size Specifics of Integrtion AMIP forced w/oserved SSTs, 2 lt 2.5 lon resolution fixed fixed to climtology, i.e, lnd-tmosphere disled shel 40 yers climtology climtologicl SSTs positive SST pttern from Fig.4 of Ginnini et l (2003) dded negtive sme SST pttern s ove sutrcted from climtology

3 Ginnini et l.: Dynmics of the orel summer Africn monsoon in the NSIPP1 tmospheric model 519 NASA s Goddrd Spce Flight Center in the frmework of the Sesonl to Internnul Prediction Project (NSIPP) (Bcmeister et l. 2001; Ginnini et l. 2003; Schuert et l. 2004). The dynmicl core uses finitedifference scheme (Surez nd Tkcs 1995), convection is of the relxed Arkw Schuert type (Moorthi nd Surez 1992), nd the lnd surfce model is Mosic (Koster nd Surez 1992). Detils of the model nd integrtions cn e found t nd in Tle 1. A nine-memer AMIP ensemle, so clled in deference to the Atmospheric Model Intercomprison Project (Gtes et l. 1998), is t the core of our nlysis. It ws integrted with horizontl resolution of 2 in ltitude y 2.5 in longitude, nd 34 levels in the verticl in stndrd r coordinte. Ech ensemle memer ws forced with different initil tmospheric conditions, ut the sme oserved history of SST nd se ice. A medley of Hdley Centre, GISST, nd Reynolds nd Smith products (Ryner et l. 1996, 2003; Reynolds nd Smith 1994) ws used over the su-periods , , nd , respectively. Becuse it cn e shown posteriori tht our nlysis does not suffer from presumed step-wise chnges in the oundry conditions (Ginnini et l. 2003), nd ecuse we re interested in internnul to interdecdl vriility, throughout this study we compute sesonl nomlies with respect to the long-term ( ) men. In the AMIP ensemle tmospheric CO 2 concentrtion is held fixed t 350 ppm, nd vegettion is prescried to e sesonlly vrying, ut internnully fixed. Hence, simultion of the persistent dry conditions of the 1970s nd 1980s in the Shel (Ginnini et l. 2003) is evidence tht the historicl evolution of SST vriility, the only temporl link etween the model nd rel worlds, provided the dominnt forcing. In the sence of n interctive dynmic vegettion model, we cn only evlute the role of vegettion indirectly. Assuming tht in the Shel lnd tmosphere interction nd dynmic vegettion provide successive mplifying steps to the initil, ocen-forced precipittion nomlies (Zeng et l. 1999; Wng et l. 2004), we ssess the role of lnd surfce vriles impcted y rinfll, e.g., surfce temperture nd evportion, y compring the AMIP ensemle to n integrtion where lnd tmosphere interction is disled. This is done in 50-yer ( ) simultion y setting the evportion efficiency, defined s the rtio of evportion to potentil evportion, to its model climtologicl vlue tken from the AMIP ensemle. This integrtion is referred to s fixed- (Koster nd Surez 1995; Koster et l. 2000). Finlly, in light of the results of Ginnini et l. (2003), dditionl integrtions were performed, to test the reproduciility of the reltionship etween the SST pttern ssocited with the interdecdl vriility of Shel rinfll, nd rinfll itself. In this Shel ensemle NSIPP1 ws run for 40 yers in control simultion with climtologicl SSTs, nd for 40 yers ech in two simultions in which the nomly pttern sttisticlly ssocited with the interdecdl vriility of Shel rinfll is superimposed on climtologicl SSTs, once with positive nd once negtive vlues. This pttern (Fig. 4 of Ginnini et l. (2003)) is chrcterized y significnt lodings in the South Atlntic nd Indin Ocens, in contrst to very wek lodings in the Pcific nd North Atlntic Ocens. 2.2 Climte oservtions over tropicl Afric; vlidtion of the precipittion climtology Two precipittion dtsets re used to vlidte model output. The sesonl climtology nd vriility of the model s northern summer (July Septemer, or JAS) rinfll in the tropicl Atlntic sin re compred to the Ntionl Ocenic nd Atmospheric Administrtion (NOAA) s Climte Prediction Center (CPC) Merged Anlysis of Precipittion (CMAP) of Xie nd Arkin (1997), product tht synthesizes stellite nd ground-sed informtion nd therefore hs glol coverge. Becuse CMAP only covers the lst 20 yers or so on record, to vlidte the long-term vriility we revert to sttion records of ccumulted monthly rinfll from the NOAA Glol Historicl Climte Network (GHCN) (Vose et l. 1992; Peterson nd Esterling 1994; Esterling et l. 1996; see the NOAA Ntionl Climtic Dt Center (NCDC) wesite t Fifty seven sttions in tropicl Afric, etween 20 S nd 20 N, 20 W, nd 40 E, were chosen in view of the completeness of their precipittion records during northern summer of Surfce ir temperture dt from the sme tropicl Africn region is riefly used in Section 3.2. This dt ws lso otined from NOAA vi the Interntionl Reserch Institute (IRI) for climte prediction s Dt Lirry ( A smller numer of sttions (44) hs complete sesonlly verged records of surfce ir temperture, nd only for the period. Anlysis in Section 3.2 is limited to these records. In Fig. 1 the CMAP climtology during the July Septemer Africn monsoon seson is compred to the model ensemle men precipittion over the common period, The model reproduces the correct mgnitude nd sptil pttern over Afric in the climtology, though it hs wet is over the Crien/ Centrl Americn region, nd over the southwestern tip of the Arin Peninsul. Climtology nd vriility in the model re underestimted in the two regionl mxim, over the Cmeroon highlnds, nd over the Guinen highlnds stretching west into the Atlntic mrine ITCZ. Vriility in generl is muted in the model, ut overll the model seems to lnce vriility in ocenic nd continentl precipittion in wy tht compres fvorly with oservtions.

4 520 Ginnini et l.: Dynmics of the orel summer Africn monsoon in the NSIPP1 tmospheric model c d Fig. 1 Comprison of the oserved (CMAP, Xie nd Arkin 1997) nd modeled precipittion climtologies for the July Septemer seson ( ). Top pnels re for oservtions, ottom pnels re for model output (in mm/dy). Contour intervl in the left pnels is 2 mm/ dy, in the right pnels it is 0.5 mm/dy 3 A descriptive nlysis of Africn climte vriility ( ) To descrie climte vriility over tropicl Afric (20 S to 20 N, 20 W to 40 E) during the northern summer monsoon seson (July Septemer, or JAS) we use Principl Component Anlysis (PCA; Preisendorfer 1988; Peixoto nd Oort 1992; Von Storch nd Zwiers 1999). This mthemticl procedure is commonly pplied to geophysicl dt to extrct sptilly coherent ptterns, lso known s Empiricl Orthogonl Functions (EOFs) tht sequentilly mximize the frction of totl vriility tht they represent. In this section we compre results of PCA pplied to precipittion (lso see Ginnini et l. 2003) nd to surfce temperture, in oservtions nd in the ensemle men of ll integrtions. In the following section the sme technique will e pplied to intr-ensemle memer vriility in rinfll. 3.1 Vriility in northern summer precipittion Figure 2 compres the two leding sptil structures, or Empiricl Orthogonl Functions (EOFs) of precipittion, in oservtions nd in the model s ensemle men. These ptterns explin 25 nd 15% of the totl vrince in oservtions, 32 nd 21% of the totl vrince in the model s ensemle men. Pnels () nd (d) in Fig. 2 represent the oserved sptil structures, with ech dot depicting the loding t sttion. Pnels () nd (e) represent the model ensemle-men sptil structures. Note tht the two leding model ptterns represent vriility t the northern nd southern edges of the climtologicl precipittion pttern, depicted in contours. EOF1 in oservtions nd EOF2 in the model s ensemle men (Fig. 2, pnels nd ) represent vriility in continentl precipittion etween 10 N nd 20 N. The zonlly elongted pttern of positive lodings centered on 15 N cross northern Afric, from the Atlntic Ocen to the Red Se, defines the Shel. The regression of the ssocited model time series, or Principl Component (PC), with glol precipittion, in Fig. 2c, shows the continentl nture of the Shel pttern. It lso shows coherent pttern of out-of-phse vrition etween precipittion nomlies in the centrl nd estern equtoril Pcific, typicl of the El Nin o- Southern Oscilltion (ENSO) phenomenon, nd in

5 Ginnini et l.: Dynmics of the orel summer Africn monsoon in the NSIPP1 tmospheric model 521 d e c f Fig. 2 The sptil ptterns ssocited with the leding Principl Components (PCs) of northern summer precipittion vriility over tropicl Afric. Anomlies re in mm/dy. Top pnels: EOFs 1 nd 2 from oservtions (sttions from NOAA/GHCN) lue dots represent negtive nomlies, red dots positive nomlies. Middle pnels: EOFs 2 nd 1 of the ensemle men of the NSIPP1 AMIP ensemle verge of nine integrtions red is for positive nomlies, lue for negtive nomlies. The contours, every 2 mm/dy, mny of the world s monsoon regions. We will come ck to this pttern in Section 5. EOF2 in oservtions nd EOF1 in the model (Fig. 2, pnels d nd e) represent rinfll vriility long the Gulf of Guine cost. It should e noted here represent the precipittion climtology ( ). Note tht the model PCA domin is slightly smller in longitudinl rnge. It extends to 35 E, insted of 40 E, to void giving too much weight to the spurious mximum in simulted precipittion etween the Horn of Afric nd the Arin Peninsul. Bottom pnels: regression mps of the model PCs with glol precipittion, in mm/dy, with contour every 0.4 strting t 0.2, nd shding representing sttisticl significnce t the 99.9% confidence level tht northern summer defines reltive minimum in the sesonl cycle of precipittion for this region, cught etween the mxim in northern spring nd fll. Nevertheless, the Gulf of Guine cost eing more humid in the men, it exhiits significnt mount of vriility

6 522 Ginnini et l.: Dynmics of the orel summer Africn monsoon in the NSIPP1 tmospheric model c Fig. 3 The Shel, or monsoon PC. Time series in the top pnel re normlized (see text for dditionl informtion). The verticl lines identify wrm (red) nd cold (lue) ENSO events. Pnels () nd (c) re regression mps with surfce temperture (in C) for the oserved PC1 nd modeled PC2, respectively (the two correlte t 0.74 over ). Contour intervl is every 0.4 C, strting t 0.2 C, solid contours represent positive nomlies, dshed contours negtive nomlies, nd shding represents the 99.9% confidence level even in the reltively dry summer seson, ppropritely cptured y this EOF. As shown in the regression pttern of the ssocited model PC with glol precipittion, in Fig. 2f, this structure connects the Gulf of Guine cost to the tropicl Atlntic Ocen ITCZ, west of our domin of nlysis, covering West Afric in the ltitudinl nd etween the equtor nd 10 N. The fct tht the two leding ptterns pper in reverse order in the nlysis of oserved nd modeled precipittion my e relted to the inhomogeneous sptil distriution of the sttions used. Becuse PCA is computed on the covrince, not on the correltion mtrix, given tht the vrince in oserved precipittion is higher long the Gulf of Guine thn it is cross the Shel (Fig. 1c), Gulf of Guine pttern should dominte. Possily ecuse there re fewer sttions in this region, though, their coherent vriility does not cpture s lrge frction of the totl vriility s it does in the model ensemle men. Creful inspection of Fig. 2 revels the presence of meridionl dipole structure in the oserved Shel pttern, etween the positive lodings in the Shel nd the negtive lodings long the Gulf of Guine cost (Jnicot 1992; Nicholson nd Plo 1993; Rowell et l. 1995). Such pttern is sent in the model ensemle men figure (Fig. 2). Its reproduction hs long een considered enchmrk test for the models ility to correctly simulte vriility in the West Africn monsoon (e.g., Vizy nd Cook 2001, 2002). Is NSIPP1 s filure to reproduce this feture model shortcoming, possily relted to indequte sptil resolution? Awiting conclusive nswer, we present evidence tht my point, insted, to the rtificil nture of the dipolr pttern. Further nlysis revels tht the first EOF of oserved rinfll represents lrge frction of vriility in Shel sttions, not so in Gulf of Guine cost sttions, s mesured y liner correltion. Correltions etween the time series ssocited with the Shel EOF nd ech sttion (not shown) re high nd sttisticlly significnt etween 10 N nd 20 N, not so equtorwrd of 10 N. Furthermore, the correltion etween the verges of Shel nd Gulf of Guine sttions is insignificnt (r= 0.05 over July Septemer ). We will come ck to this distinction etween Shel nd Gulf of Guine cost ptterns lter in this pper, to propose tht it hs dynmicl sustnce; in Section 4 we will show tht the two ptterns ehve differently with regrds to the role tht oundry forcing nd internl noise ply in their definition. The temporl structures of precipittion vriility, or Principl Components (PCs), re shown in pnels () of Figs. 3 nd 4. The two leding PCs re sttisticlly seprte, ccording to North et l. s (1982) rule, in oservtions nd model output. The Shel nd Gulf of Guine PCs in oservtions nd in the model correlte t 0.74 nd 0.62, respectively, over The Shel PC (Fig. 3) trcks the well-known trend : precipittion ws ove-verge in the 1930s, it hovered round the long-term men in the 1940s, ws well ove it in the 1950s nd erly 1960s, nd then progressively declined through the 1970s nd 1980s. In the model time series the reversed trend, of incresing

7 Ginnini et l.: Dynmics of the orel summer Africn monsoon in the NSIPP1 tmospheric model 523 precipittion from the lte 1980s onwrds, is ccentuted compred to oservtions. It should e noted here tht the comprison etween oserved nd model ensemle men PCs enefits from stndrdiztion of the time series, i.e., the originl time series re divided y their respective singulr vlues (or, equivlently, y the squre root of their respective eigenvlues). This is ecuse vriility in the model is muted with respect to oservtions, s mentioned in Section 2.2. In the cse of the Shel PCs, the rtio of oserved to model ensemle men singulr vlues is 4. Regression mps of the Shel PC with surfce temperture re displyed in pnels () nd (c) of Fig. 3, for oservtions nd model output, respectively. Regions of highly significnt vlues (the shding in Figs. 3 nd 4 represents sttisticl significnce t the 99.9% level) spn the entire tropicl ocens. In the cse of the model PC (Fig. 3c), the regression includes the simulted, ensemle-men surfce temperture over lnd: significnt regression vlues cn e noted over the Indin sucontinent nd cross the Shel itself. The Gulf of Guine PC is shown in Fig. 4. No trend is pprent in this time series, n indiction tht PCA hs successfully confined it, nd hopefully its physics, to the complementry leding mode. In Fig. 4, pnels () nd (c) disply the regression mps of this PC with surfce temperture, in oservtions nd model, respectively. The very different nture of this pttern, when compred to tht ssocited with the Shel PC, is immeditely pprent: high positive regression vlues re loclized to the estern equtoril Atlntic, in the region of SST vriility ssocited with the Atlntic counterprt to the tropicl Pcific ENSO. 3.2 Co-vriility in surfce temperture c Fig. 4 The Gulf of Guine, or ocenic ITCZ PC. Time series in the top pnel re normlized. Pnels () nd (c) re regression mps with surfce temperture (in C) for the oserved PC2 nd modeled PC1, respectively (the two correlte t 0.62 over ). Contour nd shding s in Fig. 3 The comprison of the leding PCs of the model s ensemle men surfce temperture nd precipittion (Fig. 5), clculted independently over the sme region (20 S to 20 N, 20 W to 35 E), revels the coherence in their vritions. The two leding ptterns of ensemlemen surfce temperture explin 53 nd 17% of the totl vrince, respectively. Though otined independently, they correlte t 0.91 nd 0.81 with the corresponding leding PCs of precipittion, in reverse order: PC1 of model ensemle-men surfce temperture is Shel/continentl pttern, relted to PC2 of precipittion, while PC2 of surfce temperture is Gulf of Guine/ocenic pttern, relted to PC1 of precipittion. Vlues of opposite sign in the sptil ptterns ssocited with the continentl PCs of precipittion nd surfce temperture (pnel of Figs. 2 nd 5, respectively) represent trends of contrsting sign cross the Shel: s precipittion decresed over the pst 50 yers, surfce temperture incresed. The coupling of trends of opposite sign in precipittion nd surfce temperture in the Shel is confirmed y repet of PCA on sttion surfce ir temperture dt (not shown). The first PCs of oserved precipittion nd surfce ir temperture correlte t 0.82 during nlysis is limited to this su-period due to the temporl limittion of the surfce ir temperture records ville from GHCN. PC1 of oserved temperture explins 39% of the totl vrince over Afric, 20 S to 20 N. Agin, high sptil lodings (not shown) re focused on the Shel, 10 N to 20 N.

8 524 Ginnini et l.: Dynmics of the orel summer Africn monsoon in the NSIPP1 tmospheric model Fig. 5 PCA of the ensemlemen surfce temperture in the NSIPP1 AMIP ensemle. Left column: the PCs re normlized. Solid lines re the PCs of surfce temperture, dshed lines the PCs of precipittion. Right column: contour is every 0.2 C, solid contours represent positive nomlies, dshed contours negtive nomlies c d While over the tropicl ocens surfce temperture nd precipittion tend to e positively correlted, i.e., precipittion is usully lrgest over the wrmest SSTs, the opposite is true over lnd. Over lnd the surfce temperture signl is most directly interpretle s consequence of the precipittion signl: incresed precipittion is ssocited with incresed soil moisture, which mens tht given energy flux into the surfce will e used preferentilly to evporte the moisture, rther thn to het the surfce. A negtive short-wve cloud feedck is lso consistent with n out-of phse reltionship etween precipittion nd surfce temperture nomlies: the cloud cover ssocited with rinfll reduces the incoming solr rdition t the surfce, leding to surfce cooling. Regression plots of the Shel PC of precipittion with precipittion, evportion nd verticlly integrted moisture convergence (Lenters nd Cook 1995; Vizy nd Cook 2001) from the AMIP ensemle (left column of Fig. 6) demonstrte tht locl nd lrge-scle processes contriute to the wter udget in comprle prts. Once precipittion nomly, in our cse remotely forced from the ocen, is estlished (Fig. 6), locl moisture recycling, represented y the evportion nomly (Fig. 6c), nd lrge-scle dynmicl constrints, represented y the moisture convergence nomly (Fig. 6d), contriute in eqully significnt terms to precipittion cross Afric. When lnd tmosphere interction is severed, in the fixed- integrtion (right column of Fig. 6), the nomlous precipittion (Fig. 6e) does not result in surfce temperture (Fig. 6f) or evportion (Fig. 6g) nomlies s lrge or significnt s in the AMIP ensemle. Note tht the fct tht surfce temperture nomlies cross the Shel in fixed- re frction of the corresponding AMIP ensemle nomlies implies tht evportion plys dominnt role in defining surfce temperture, compred to cloud-rditive feedcks. The fct tht the surfce temperture nomlies re negtive implies tht the short-wve cloud feedck, i.e., cooling y reduction of incoming solr rdition t the surfce, wins over the long-wve cloud feedck, which would result in more energy eing trpped elow the clouds, hence wrming. The opposite sign in nomlies in precipittion nd lnd surfce temperture is confirmed in the modeled Gulf of Guine ptterns. The correltion etween these EOFs of precipittion nd surfce temperture (compre pnel d in Fig. 5 with pnel e in Fig. 2) reflects the ssocition etween nomlies of the sme sign in precipittion nd Gulf of Guine SST, nd of opposite sign in precipittion nd lnd surfce temperture long the Gulf of Guine cost. For discussion of the dynmics involved in the forcing of n nomlous tmospheric circultion y estern equtoril Atlntic SSTs the reder is referred to the comprehensive modeling studies of Vizy nd Cook (2001, 2002). In rief, the following points summrize wht descried so fr:

9 Ginnini et l.: Dynmics of the orel summer Africn monsoon in the NSIPP1 tmospheric model 525 e f c g d h Fig. 6 Regression mps of the Shel PC with ( e) precipittion, ( f) surfce temperture, (c g) evportion, nd (d h) verticlly integrted moisture convergence, from the AMIP ensemle (left column), nd from the fixed- run (right column). e re in units of mm/dy, ( f) re in C, (c g) nd (d h) re in mm/dy. In ( e), ( f), nd (d h) contour is every 0.4, strting t 0.2. In (c g) contour is every 0.2, strting t 0.1. Solid contours denote positive nomlies, dshed contours negtive nomlies, nd shding indictes the 99.9% confidence level

10 526 Ginnini et l.: Dynmics of the orel summer Africn monsoon in the NSIPP1 tmospheric model 1. When forced with the historicl record of SST only, NSIPP1 skillfully cptures the oserved internnul to interdecdl vriility of northern summer Africn climte. This includes the interdecdl drying trend in the Shel, nd internnul vriility typicl of Gulf of Guine rinfll. 2. The oserved reltionship etween precipittion nd lnd surfce temperture in the Shel, while consistently reproduced in the AMIP ensemle, vnishes in n integrtion where lnd tmosphere interction is disled. This ehvior is suggestive of strong control exercised y precipittion on surfce temperture through lnd tmosphere feedck. As corollry, the positive trend in Shel surfce temperture cn e interpreted s consequence of the negtive trend in precipittion. 4 Signl nd noise in the modeled Africn monsoon Hving shown tht the model is cple of cpturing the dominnt ptterns of the oserved vriility of precipittion, nd tht such ptterns owe their existence primrily to SST forcing, next we exploit the informtion contined in the ensemle of integrtions to further investigte the nture of such ptterns. We re interested in discriminting etween signl nd noise in Africn rinfll vriility s first step towrds quntifying potentil predictility. We define the signl s the oundry-forced component, while the noise is vriility internl to the tmosphere, or to the coupled lnd tmosphere system, which is independent of the prescried ocenic oundry conditions. Assuming linerity, Zwiers (1996), Rowell (1998), nd Koster et l. (2000), mong others, write: r 2 Totl ¼ r2 Forced þ r2 Internl Tle 2 Percent vrinces explined y the leding PCs of the totl, forced, nd internl components of vriility in the AMIP integrtions. X, the coherence reported in the fourth column, is computed using the totl nd forced components of vriility. See text for detils where r 2 is vrince. Forced refers to oundry forcing from the prescried ocenic conditions. In the limit of infinitely lrge ensemle size, the Forced component of vriility is represented y the timevrying ensemle men, while the Internl component of the vrince, i.e., tht due to tmospheric noise nd/ or vriility rising from lnd tmosphere interction, is represented y devitions from it. We pply PCA to totl nd internl vriility of precipittion over Afric, 20 S to20 Nnd 20 W to 35 E, s we did to forced vriility in the previous section. The percent vrinces explined y the leding PCs of forced, totl nd internl vriility re summrized in Tle 2. PCA is first repeted on the totl vriility of precipittion s represented y the stck of devitions of ll the ensemle memers from the ensemle s July Septemer precipittion climtology. Results re displyed in Tle 2 nd in Fig. 7. The two leding sptil ptterns re indistinguishle from those of ensemle-men (Fig. 2, pnels nd e), or forced vriility, hence they re not shown. EOF1 is Gulf of Guine pttern, explining 19% of the totl vrince. EOF2 is Shel pttern, explining 15% of the totl vrince. Note tht the percent vrinces explined y the two leding PCs of totl vriility, in the first column of Tle 2, compre well with the percentges explined y the oserved PCs 1 nd 2 of precipittion, of 25 nd 15%, respectively, over Also note, in Fig. 7, tht ll the ensemle memers reproduce the trend in Shel rinfll. Correltion vlues of the ensemle-men PCs 1 nd 2 with the individul ensemle-memer reliztions rnge etween 0.94 nd 0.96 in PC1, the Gulf of Guine PC, nd etween 0.84 nd 0.89 in PC2, the Shel PC. From visul inspection lone, it is cler tht the sctter etween individul reliztions of the model s Gulf of Guine PC is smller thn tht etween individul reliztions of the Shel PC (compre pnels in Fig. 7). Following Koster et l. (2000), to mesure the tightness of the spred, or coherence mong reliztions we compute: X ¼ Ir2 Forced r2 Totl ði 1Þr 2 Totl r 2 Totl r 2 Forced X r 2 Internl Gulf of Guine PC 19% 32% 0.91 Shel PC 15% 21% % where I is ensemle size. X cn tke vlues etween 0 nd 1. Where the I reliztions to disply vriility completely independently of ech other, then, from the Centrl Limit Theorem it would follow tht r 2 Forced = r 2 Totl/I, hence X =0. Conversely, were the I reliztions to ll disply the sme vriility, then r 2 Forced = r 2 Totl, hence X =1. In oth our PCs, the coherence mong reliztions is high, s seen in vlues of X of 0.91 nd 0.70 in Tle 2. As expected, coherence is higher for the Gulf of Guine PC thn for the Shel PC. Not surprisingly, then, in the repet of PCA on the internl vriility of rinfll, i.e., on the devitions of ech nd every ensemle memer from the time-vrying ensemle men (Tle 2 nd Fig. 8), we find tht the dominnt pttern of forced nd totl vriility, the Gulf of Guine pttern, is no longer preferred pttern. Insted, the dominnt pttern of internl vriility (Fig. 8), explining 10% of the totl vrince of precipittion, is continentl, ering strong resemlnce to the Shel pttern. From sttisticl stndpoint, this finding is consistent with the noise level in the individul ensemle memers of the two leding PCs eing higher for the Shel thn for the Gulf of Guine, s noted ove in reltion to the sctter in the time series in Fig. 7. Similr results hve recently een otined y

11 Ginnini et l.: Dynmics of the orel summer Africn monsoon in the NSIPP1 tmospheric model 527 mechnisms of rinfll vriility over West Afric, with distinct sptil nd temporl scles, s seen in Section 3, nd controls. The Gulf of Guine pttern owes its existence entirely to SST vriility, more precisely to tropicl Atlntic SST vriility. The Shel pttern is continentl in nture, nd displys significnt mount of noise which is mplified y lnd tmosphere interction (see previous section nd Fig. 6; lso see Fig. 3 in Ginnini et l. (2003)). Becuse very similr sptil ptterns pper mong the leding modes of oundryforced nd internl vriility, nd ecuse the mplitude exhiited y the oundry-forced Shel pttern is lrger thn tht exhiited y its internl vriility counterprt (the rtio of Shel eigenvlues of oundryforced nd internl vriility is 3), we interpret the Shel pttern s n internl mode of vriility of northern summer Africn precipittion, mode in which lnd tmosphere interction mtters, nd tht is preferentilly excited nd mplified y SST vriility. 5 Dynmics of the Indin Ocen Shel teleconnection Fig. 7 PCA of the totl vrince of July Septemer precipittion over tropicl Afric. Time series for ech ensemle memer ssocited with Gulf of Guine pttern. Time series ssocited with Shel pttern. The mplitude is in units of stndrd devition Tippett nd Ginnini (2005), who nlyzed the predictle components of West Africn rinfll in different tmospheric model. More relevnt is the dynmicl interprettion this ehvior lends itself to; Gulf of Guine nd Shel ptterns represent two different Over the pst 30 yers reserch into the linkges etween the world s ocens nd the persistence of drought in the Shel hs expnded from focus on the tropicl Atlntic (e.g., Hstenrth nd Lm 1977; Lm 1978, 1978) to include the Pcific (e.g., Follnd et l. 1986; Plmer 1986; Wolter 1989; Jnicot et l. 1996), nd Indin Ocens (Shinod nd Kwmur 1994; Jnicot et l. 2001; Rowell 2001). In this section we descrie three plusile dynmicl mechnisms tht my explin the link etween wrming trend in the tropicl ocens nd the drying of the Shel (Bder nd Ltif 2003; Ginnini et l. 2003). These three mechnisms re closely inter-relted, nd could e portryed s three levels of incresing complexity in the tmospheric response to SST nomlies, rther thn three independent, competing mechnisms. The first mechnism involves the therml contrst etween lnd nd ocen. Its considertion follows from study of Chou et l. (2001) on idelized monsoons. The second mechnism reltes to zonlly Fig. 8 PCA of the internl vrince of July Septemer precipittion over tropicl Afric. The leding pttern represents noise in Shel-like pttern. The mplitude for the time series is in units of stndrd devition

12 528 Ginnini et l.: Dynmics of the orel summer Africn monsoon in the NSIPP1 tmospheric model Fig. 9 Regression mps of the high-frequency component of the Shel PC with () 200 hp geopotentil height (contour every 40 gpm) nd winds, () precipittion (contour every 0.4 mm/dy, strting t 0.2), (c) 850 hp winds nd surfce temperture (contour every 0.4 C, strting t 0.2). Solid contours denote positive nomlies, dshed contours negtive nomlies, nd shding indictes the 99.9% confidence level c symmetric tmospheric dynmics involving the glol Hdley circultion. The third mechnism, the eddy response to loclized heting, interprets devitions from the glol Hdley circultion in light of the theory of equtoril wve dynmics (Mtsuno 1966; Wester 1972; Gill 1980). The strting point of our nlysis is the decomposition of the Shel PC of precipittion (Fig. 3) into lowfrequency, or interdecdl (21-yer running men), nd high-frequency, or internnul (its residul) components (Ginnini et l. 2003; see their Fig. 4). This procedure seprtes the influence of the tropicl Pcific nd ENSO, which opertes on the internnul time scle, from tht of the tropicl ocens round Afric. Implicit in our interprettion of this seprtion is the resoning tht the time scle, internnul or interdecdl, is set y the slower component of the ocen tmosphere system, i.e., the ocen, or y ocen tmosphere interction, s in

13 Ginnini et l.: Dynmics of the orel summer Africn monsoon in the NSIPP1 tmospheric model 529 Fig. 10 Composites of the positive-climtology differences in the Shel ensemle. Geopotentil height (contour is every 40 gpm) nd winds t 200 hp (in m/s), () precipittion (in mm/ dy contour is every 0.4 mm/ dy, strting t 0.2 mm/dy), nd (c) 850 hp winds (in m/s) nd SST in C. Contour is every 0.4 C. Shding in (, ) indictes the 99.9% confidence level. The line plots to the right of ech pnel represent the zonl men of the vrile tht is contoured in the pnel. In the cse of surfce temperture the zonl verge is tken over the ocens only Fig. 9, the high-frequency component of the Shel PC is regressed ginst relevnt tmospheric vriles. Positive SST nomlies in the tropicl Pcific (Fig. 9c) re ssocited with positive upper-level geopotentil height nomlies (Fig. 9); in wrm ENSO, the entire tropicl troposphere wrms up (Yulev nd Wllce 1994), nd the verticl profile ecomes more stle (Ching nd Soel 2002; Su et l. 2004). Hence, the regions clim c ENSO. However, t ny given time in the internnul or decdl cycle of these nomlies, the tmosphere responds to the nomlies present, with no regrd to the chrcteristic time scle of their evolution. Before focusing the rest of the discussion on the impct of SST vriility ssocited with the low-frequency component of Shel rinfll, we comment riefly on ENSO s role in shping its internnul vriility. In

14 530 Ginnini et l.: Dynmics of the orel summer Africn monsoon in the NSIPP1 tmospheric model c Fig. 11 As in Fig. 10, except for the zonl eddy circultion. Contour intervls re: () every 20 gpm, () every 0.4 mm/dy, strting t 0.2 mm/dy, (c) every 0.4 C, strting t 0.2 C tologiclly ffected y deep convection ll suffer reduction in rinfll, visile in Fig. 9 cross Afric, South Asi, nd Centrl Americ. Consistently, s seen in Section 3.2, s precipittion decreses, lnd surfce temperture increses. To descrie the impct ssocited with the low-frequency component of Shel rinfll we nlyze the Shel ensemle. We designed this set of integrtions specificlly s test for the dynmicl consistency of the sttisticl reltionship identified y Ginnini et l. (2003), etween lrge-scle ptterns of SST nd the longterm trend of Shel rinfll. We compre pir of 40- yer integrtions, referred to s positive nd negtive Shel, to 40-yer climtologicl SST integrtion. The positive nd negtive Shel simultions re otined y forcing the tmospheric model with the pttern sttisticlly relted to the low-frequency vriility of Shel rinfll dded or sutrcted, respectively, to climtologicl SSTs. Linerity of the response to SST is tested, y compring positive minus climtology nd negtive minus climtology differences. Since the response is found to e liner, we will only discuss the positive minus climtology cse. The SST nomlies in the positive Shel integrtion (see Fig. 4 of Ginnini et l. (2003) or Fig. 10c here) cpture the net wrming of the South Atlntic nd Indin Ocens. The ltter, ccording to Levitus et l. (2000), dtes ck to the mid-1960s. The contriution from the tropicl Pcific is negligile, hving een isolted into the internnul component. Also of note is the sence of significnt nomlies in the North Atlntic, hence of n Atlntic interhemispheric grdient in SST, which hs een rgued to contriute to longterm vriility of rinfll in the Shel (Follnd et l. 1986; Rotstyn nd Lohmnn 2002; Hoerling et l. 2004). Before we go on to interpret the role of Indin Ocen SSTs in the Shel simultions, we note tht the vlidity of the AMIP frmework hs een clled into question y oservtionl nd modeling studies which qulify the rekdown of the ssocition etween wrm SST nd ove-verge precipittion in the Indin Ocen during the southwesterly monsoon seson (Ro nd Goswmi 1988; Kirtmn nd Shukl 2002; Gdgil et l. 2003; Krishn Kumr et l. 2005). In relity, wrm SSTs in the Indin Ocen during northern summer re more consistently interpreted s the consequence of reduced deep convection nd cloud cover, nd incresed incoming solr rdition t the surfce, thn s the cuse for incresed precipittion. Hence, forcing model with SSTs cn crete n unrelistic source of energy nd moisture, which leds to n unrelistic precipittion response. This could e, indeed, wht our model simultes. Assume, on the other hnd, tht we could seprte two components in Indin Ocen SST vriility. One component would e rising from locl, coupled ocen tmosphere dynmics with strong signture on the internnul time scle. The other would e externlly forced, e.g., from the rditive effect of incresed tmospheric greenhouse gs concentrtions, or from decdl or longer-term chnges in ocen circultion, nd would mnifest itself in the trend. By ttriuting the equtoril Indin Ocen wrming to the ltter, externlly forced mechnism, we could rightfully interpret the response in Shel rinfll s consequence, vi the tmospheric response descried ove, of this externlly forced trend component in the vriility of Indin Ocen SSTs.

15 Ginnini et l.: Dynmics of the orel summer Africn monsoon in the NSIPP1 tmospheric model The role of lnd-ocen temperture contrst In their idelized study, Chou et l. (2001) show how, in regions dominted y the monsoon, precipittion over lnd cn e fvored t the expense of ocenic precipittion. The lnce ultimtely involves temperture grdients, set up y the differences in het cpcity of lnd nd ocen, nd moisture vilility. At equinox conditions, if the lnd surfce is rtificilly sturted in moisture, there is no difference etween lnd nd ocen equtoril convection extends cross lnd nd ocen surfces. When soil moisture is not prescried to e sturted, nd is, insted, llowed to interct with the tmospheric circultion, ocenic convection is fvored t the expense of continentl convection, presumly ecuse moisture is redily ville over the ocen. Finlly, when simple prmeteriztion of the meridionl ocen het trnsport is dded, which effectively cools the tropicl ocens y exporting het to mid-ltitudes, the continentl convergence zone gins the upper hnd. In our simultions, the positive minus climtology difference equtes to wrming of the tropicl ocens. Composites of this difference, i.e., the difference etween the July Septemer 40-yer verge in the positive Shel nd climtology integrtions, in 200 hp geopotentil height nd winds, precipittion, nd surfce temperture nd 850 hp winds re displyed top to ottom in Fig. 10. The line plots to the right of ech pnel represent the zonl men of the quntities contoured in the corresponding sptil plot. These nomly ptterns re virtully indistinguishle from regression plots (not shown) of the low-frequency component of the Shel PC on the sme vriles tken from the AMIP ensemle men. If we focus our ttention on the precipittion nomlies in the positive-climtology difference (Fig. 10), we find tht concomitnt with the wrming of the Indin nd South Atlntic Ocens, nd with the drying of the Shel, precipittion incresed in the equtoril Indin Ocen. Unfortuntely, due to the sence of criticl sttion dt from ville rchives, we re unle to confirm the reltionship etween longterm vriility of rinfll in the Shel nd in the equtoril Indin Ocen. For exmple, the record from the one sttion in GHCN representtive of the western equtoril Indin Ocen, Mhé in the Seychelles, is incomplete dt is missing for the crucil yers which signled the trnsition etween wet nd dry epochs in the Shel, i.e., However, recent study y Hurrell et l. (2004) finds sin-wide consistency etween oserved nd modeled trends in precipittion in the Indin Ocen. The ssocition etween elow-verge rinfll in the Shel nd ove-verge precipittion in the equtoril Indin Ocen fits the mechnism descried ove: wrming of the tropicl ocens, s in the idelized simultion of Chou et l. (2001) where meridionl ocen het trnsport is disled, shifts convection from lnd to ocen. This interprettion then rings to our ttention the question of ttriution why hve the tropicl ocens consistently wrmed up in the pst few decdes? 5.2 The response of the Hdley circultion to n equtoril SST wrming The precipittion nomlies in the equtoril Indin Ocen (Fig. 10) re colocted with, nd of the sme sign s, the SST nomlies (Fig. 10c), nd re most conspicuous just to the north nd to the south of the equtor etween 50 E nd 90 E. Precipittion shows decrese ll round the gloe t ltitudes etween 10 N nd 20 N noticele in the zonl-men plot to the side of Fig. 10. The decrese is mrked cross the monsoonl regions of Centrl Americ nd the Crien, in the Shel, nd over the western tropicl North Pcific. In contrst, precipittion increses in the equtoril Indin Ocen, etween 50 E nd 110 E, in the By of Bengl, nd in the western equtoril Pcific. These chnges imprint the zonl men picture, representtive of the glol Hdley circultion. We hypothesize tht wrming not only of tropicl SSTs, ut more specificlly of equtoril SSTs, wrming which is especilly mrked in the Indin Ocen, my hve fvored more equtoril position of the scending rnch of the glol Hdley circultion, hence of the locus of deep convection. Ner-surfce nd upper-level winds would then hve djusted to the newly found equilirium etween the glol tmospheric circultion nd the ditic heting nomlies tht pertured it. Positive 200 hp geopotentil height nomlies (Fig. 10) not unlike those ssocited with the response to the equtorwrd shift of convection typicl of wrm ENSO event (see Fig. 9) re consistent with n increse in equtoril deep convection in the Indin nd western Pcific Ocens, with mrked center in the Indin Ocen etween 60 E nd 100 E. Awy from the equtor, the nomlies in 200 hp winds re consistent with qusi-geostrophic flow long the geopotentil height nomly contours. Of prticulr interest here is the wekening of the tropicl esterly jet in the region etween South Asi nd Afric, in the ltitudinl nd etween 10 N nd 20 N weker thn norml upperlevel jet is hllmrk sign of yers of reduced Shel precipittion (Newell nd Kidson 1984; Nicholson nd Grist 2001). 5.3 The zonl eddy response nd equtoril wve dynmics An interesting degree of complexity is dded when one considers devitions from the zonl men in the tmospheric circultion (Fig. 11). Then, the eddy component of the upper-level nd ner-surfce flows in the Africn Asin sector cn e interpreted s the dynmicl response of the tmosphere to the superposition of two competing ditic het sources, one equtoril nd

16 532 Ginnini et l.: Dynmics of the orel summer Africn monsoon in the NSIPP1 tmospheric model nother off-equtoril (Wester 1972; Gill 1980). The equtoril het source is, gin, one ssocited with nomlously strong convection in the Indin Ocen (Fig. 11). Consistent with Gill s (1980) description, to the west of the het source re ner-surfce westerly wind nomlies converging towrds it (Fig. 11c), nd upper-level esterly nomlies diverging wy from it (Fig. 11). The Rossy-wve component of the response to this het source lone would drw northesterly ner surfce flow cross the Shel, recurving to westerly t the equtor, synonymous with wekening of the southwesterly monsoon cross Afric. However, the picture is complicted y the presence of off-equtoril het sources, which represent strengthening of the monsoon, nd re centered on the By of Bengl nd in the western North Pcific. Consistently, zonl eddy 200 hp geopotentil height nomlies (Fig. 11) re positive in sector centered on the By of Bengl, from cross the Shel to Est Asi. Upper-level wind nomlies cpture strengthening of the tropicl esterly jet etween Afric nd Indi, while ner-surfce wind nomlies cpture strengthening of the South Asin monsoon. The off-equtoril het source in the By of Bengl would lone drw southwesterly flow to the west of it, strengthening the Africn monsoon cross the Shel. In the lnce, Shel rinfll in the positive-climtology difference of the zonl eddy field (Fig. 11) is negtive, ut not significntly so, compred to the totl (Fig. 10) field. The Africn response to SST nd precipittion in the Asin sector hs potentil vlue for understnding nd predicting new lnce of forces, one tht could develop with climte chnge. If the wrming of the tropicl ocens (Levitus et l. 2000) is sign of nthropogenic climte chnge (Brnett 2005), will it necessrily spell doom for the Shel nd other semi-rid climtes of the Erth (Neelin et l. 2003; Chou nd Neelin 2004)? Or will projected strengthening of the Asin summer monsoon, rought out y the incresed lnd ocen therml contrst ssocited with sped-up melting of snow cover (Meehl nd Wshington 1993; Krishn Kumr et l. 1999; Houghton et l. 2001), e le to counterct the drying trend? 6 Conclusions The comprison of oservtions nd model simultions presented here confirms the dominnt role plyed y the ocens in Africn climte vriility (Follnd et l. 1986; Ginnini et l. 2003), t lest on internnul to interdecdl time scles. The picture tht emerges is tht of climte vriility tht is remotely forced from the ocen, nd mplified y locl lnd tmosphere interction. Vegettion nd dust could conceivly ply role prllel to tht of the lnd surfce, in responding y mplifying the remotely forced precipittion nomlies (Zeng et l. 1999; Wng et l. 2004). In this concluding section we comment on two spects of Africn climte highlighted y this comprison: (1) the reltionship etween Shel nd Gulf of Guine rinfll, or etween the Africn monsoon nd the Atlntic ITCZ (lso see Leel et l. 2003; Bisutti et l. 2004; Gu nd Adler 2004); (2) the role of the Indin Ocen in Africn climte vriility (lso see Bder nd Ltif 2003; Hoerling et l. 2004). Principl Component Anlysis (PCA, in Section 3) pplied to oservtions nd to model output identified two distinct ptterns of rinfll vriility over tropicl Afric during orel summer. These ptterns jointly explin out 40% of oserved nd 50% of the model s ensemle-men vriility. One represents vriility t the northern edge of the sesonl monsoonl precipittion, in the semi-rid Shel, etween 10 N nd 20 N. The other descries vriility t the southern edge, long the Gulf of Guine cost, etween 10 N nd the equtor (see Fig. 2). Further nlysis crried out on the ensemle of simultions indictes tht they re lso dynmiclly distinct. The Gulf of Guine pttern owes its existence to tropicl Atlntic Ocen vriility only, more specificlly to vriility in estern equtoril Atlntic SSTs, nd is inextricly tied to the Atlntic mrine ITCZ. The Shel pttern represents vriility in the Africn monsoon (Bisutti et l. 2004), nd displys non-negligile mount of noise, possily intrinsic to chotic lnd tmosphere interction. At the sme time it is very sensitive to glol, tropicl ocenic conditions (Ginnini et l. 2003). Consequently, the following remrks pper justified: (1) the Atlntic mrine ITCZ is decoupled from the Africn monsoon; (2) vriility in the Shel represents vriility in the ccumulted totl sesonl monsoon rins, s previously rgued y Shinod nd Kwmur (1994), nd not in the meridionl migrtion of the loction of convergence. Long-term vriility in Shel rinfll is sttisticlly relted to vriility in the ocens round Afric, especilly the Indin Ocen. To connect chnges in the Indin Ocen to the continentl climte of the Shel we find vlue in rguments out the contrst in thermodynmic properties etween lnd nd ocen (Chou et l. 2001), s well s in the clssicl description of the zonlly symmetric, or Hdley circultion (Schneider 1977; Held nd Hou 1980; Lindzen nd Hou 1988; Plum nd Hou 1992), nd in concepts relted to equtoril wve dynmics (Mtsuno 1966; Wester 1972; Gill 1980) pplied to the zonlly symmetric circultion (Section 5). Rinfll in the Shel my hve declined over the lst 50 yers s convection migrted from lnd to ocen, or s the zonl men Hdley circultion migrted equtorwrd, ttrcted y wrming of the tropicl ocens most conspicuous in the equtoril Indin Ocen. The questions tht remin to e nswered relte to the dynmicl response of the tmosphere, nd of the Asin Africn monsoon in prticulr, to climte chnge. Since in the simultions considered here the direct rditive effect of nthropogenic greenhouse gses is not tken into ccount, it will e importnt to inter-