Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | , Eltahir and 5 ; 1993 , 1 , R. J. Van der Ent 4 Brubaker et al. , and A. Rammig 7 17480 17479 , H. M. J. Barbosa 3 , G. Sampaio 1 erent datasets and methodologies. Our results show that cascading ff , J. Heinke , C.-F. Schleussner 1,6 1,2 This discussion paper is/has beenand under Physics review (ACP). for Please the refer journal to Atmospheric the Chemistry corresponding final paper in ACP if available. Potsdam Institute for Impact Research (PIK), Telegraphenberg A 31,Department 14473 of Geography, Humboldt Universität zuClimate Berlin, Analytics, Berlin, Berlin, Germany Germany Instituto de Física, Universidade deDepartment São of Paulo, São Management, Paulo, Faculty S.P., Brazil of Civil Engineering and Geosciences,Stockholm Delft Resilience Centre, Stockholm University, KräftriketCenter 2B, for 114 Earth 19 System Stockholm, Science Sweden (CCST), INPE, Cachoeira Paulista, S.P., Brazil Continental moisture , the processtinent by which returns evapotranspiration from as the con- to the continent ( analysis, we reveal the importancea of key the intermediary south-western regionthe part for wet of continental the season. moisture Amazon Our transporta basin results stronger in as impact suggest South on that America downwindously land during rainfed thought. use and change ecosystem in stability this than region previ- might have 1 Introduction studies using di moisture recycling contributes about 10 %and to 17 % the over total the precipitation Lathe over Plata South total basin. America Considering dependency cascading moisture ofabout recycling 25 the % increases from La 23 Plata to 29 basin % on during the moisture wet from season. Using the tools Amazon from complex basin network by cycling”, which describes theatmosphere through exchange precipitation of and re-evaporation moisture cyclescations between on on its the the way continent. between vegetation We twoforced and use lo- by the the precipitation Water from Accounting TRMM Model2001 and until 2-layers (WAM-2layers) 2010 evapotranspiration to from construct MODISdirection moisture for recycling and the networks. period amount These networks ofits describe moisture the destination (precipitation) transported in from Southtal its America. and source Model-based regional calculations (evapotranspiration) recycling of to ratios continen- in the Amazon basin compare well with other existing Continental moisture recycling is atem. crucial Evapotranspiration process of from the thegionally South and Amazon American in climate river the sys- basin LaAmerican Plata contributes moisture river to basin. recycling. Here precipitation We we quantify re- present the an in-depth importance analysis of of “cascading South moisture re- Abstract Atmos. Chem. Phys. Discuss., 14, 17479–17526,www.atmos-chem-phys-discuss.net/14/17479/2014/ 2014 doi:10.5194/acpd-14-17479-2014 © Author(s) 2014. CC Attribution 3.0 License. On the importance of cascadingrecycling moisture in South America D. C. Zemp J. F. Donges 1 Potsdam, Germany 2 3 4 5 University of Technology, Delft,6 the Netherlands 7 Received: 12 May 2014 – Accepted: 8Correspondence June to: 2014 D. – C. Published: Zemp 30 ([email protected]) June 2014 Published by Copernicus Publications on behalf of the European Geosciences Union. 5 25 20 10 15 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | , , ; , , , ). Cox Lean ; 2012 Burde 2006 Bagley , ; , ; 2003 Arraut and , ; 1999 2009 Walker et al. , ), atmospheric , ; ect rainfed agri- Spracklen et al. Sudradjat et al. ff Marengo 2014 ; , Arraut et al. Werth and Avissar 1991 2009 ; ; , , erent types of tracers Bosilovich and Chern 2011 ff ; 2008 Trenberth , , 2010 ; 2009 , , 1999 , Dirmeyer et al. 1994 Oyama and Nobre 1991 , , Hasler et al. Victoria et al. Knox et al. ) (see a review of the methods in ). In most of existing studies using ). Adding di ; ; ; ). Even if regional impact of changes ) the magnitude of rainfall reduction ). The moisture from the Amazon basin Trenberth 2012 Drumond et al. 2001 2012 2008 ; 2011 1991 ; , , , 2012 2014 , , , , , van der Ent et al. 2009 Nobre et al. ; , 1994 17482 17481 ; 2005 , , Eltahir and Bras ) is particularly important for the South American ; 2009 1990 , , Keys et al. Keys et al. ) found that moisture runs on average through more ; ects the (see review in Hirota et al. 2010 ; 1993 ff ; , ) or a posteriory with reanalysis data ( Da Silva et al. Bagley et al. , Marengo Spracklen et al. ; ; Gat and Matsui 1999 2010 ( 2009 Burde and Zangvil Dirmeyer et al. ; 2004 , , 2006 ; ; , 2007 , 2011 Eltahir and Bras , , Shukla et al. ). 1979 ; 2006 Dirmeyer et al. 2013 , , ; , Numaguti 2014 1989 , Brubaker et al. , 2009 Betts et al. van der Ent et al. ), quasi-isentropic back-trajectory method ( ) and tagged water experiments with a general circulation model (GCM) , ; ; Rockström et al. Salati et al. 2006 2014 van der Ent et al. Medvigy et al. Sampaio et al. 2004 Burde et al. 1994 , , what is the importance of cascading moistureWhich recycling are in the South important America? intermediary regions forWhich cascading are moisture the recycling? keyare regions channeled? where the pathways of cascading moisture recycling ; ; ; ). Downwind, e.g., in the La Plata basin, rainfall reduction may a ; , , We perform a tagged water experiment a posteriori with precipitation, evapotran- We call “cascading moisture recycling” the process by which evapotranspiration from Land-use change – in particular in the Amazon basin – impacts the To identify sources and sinks of moisture and to quantify regional and continental 1. 2. 3. Bosilovich and Chern et al. Bras recycling” as the set ofwell locations as including the the starting intermediaries. and the destination locations, as spiration, wind and humidity datasets from reanalysis and satellite products for South a specific location onprecipitates in land another runs one. through Onprocess the one by other which or hand, evapotranspiration is “direct more directly moisturecycle) re-evaporation transported from recycling” cycles (without one refers any location to before on re-evaporation the land it the to location another, where where it re-evaporation cycle precipitates. We is call occurring. “intermediary” We define a “pathway of moisture than two re-evaporation cycles betweenin evaporation northern from Eurasia. the In ocean this study and we precipitation focus for the first time on the following questions: (re-evaporation cycle) and cantributes be to transported precipitation. Such toatmosphere exchanges a of through downwind moisture re-evaporation location between cyclesof the where might moisture vegetation it play and in con- a the thewithin crucial Tropics a role ( GCM, for the transport tagged water experiments, moisture evaporating fromin a time given region from is trackedtion. the forward However, evapotranspiration precipitating until water it can returns be to re-evapotranspirated the in land the surface same as location precipita- ( 2002 for the ecosystem stability in the Amazon rainforests ( van der Ent et al. A resulting reduction in regional moisture supply may have important consequences hydrological cycle. In theis Amazon regionally river recycled ( basin, between 25 and 35 % of the moisture evapotranspiration rate and a et al. Satyamurty Martinez et al. 2006 is also exported outalong of the the Andes and basin,larly contributes transported to during via precipitation the the over wet South the season La American ( Plata Low river Level basin Jet particu- 2002 topes ( and the location ofadvancements in the our most understanding of impactedAmerican the continent regions continental are are moisture needed. recycling still in uncertain. the South Therefore, further moisture recycling in South America, several methods have been used including iso- bulk models ( culture ( 2012 and Warrilow 2009 in precipitation patterns fromtions deforestation has from been atmospheric intensively general studied circulation using models simula- with deforestation scenarios ( et al. 5 5 25 20 15 10 10 25 15 20 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | , ) ). 4 man Tso- ff 2.1.2 Hu van der Ent et al. ). In previous studies, links in 2013 , ). The actual tracking in WAM-2layers is per- 17484 17483 ) and to reveal important regions for propagation Boers et al. b ; , 2014 , 2012 , 2009a , erent locations. In our study, we apply complex network ap- ff Malik et al. van der Ent et al. Donges et al. ; ), 8 days evapotranspiration estimates from Moderate Resolution Imaging 2008 , 2007 , ). WAM-2layers provides the basis for the construction of moisture recycling networks We propose a novel framework to quantify the importance of cascading moisture In Sect. 2.1 we describe the tagged water experiment using the WAM-2layers and we In this study we make2layers use V2.3.01 of ( the Eulerian atmospheric moisture tracking model WAM- 2.1 Building moisture recycling networks 2.1.1 Description of the moisture tracking in WAM-2layers 2 Methods tential of the complex networkhas approach has been been used recently to shown detectnis in large-scale et climate related al. science. climate It anomalies (teleconnections) ( ter Accounting Model-2layers) (van der Ent et al., 2014). recycling for the regionalbasin climate and the in La South Plataimprove basin. America, our In and understanding addition, of we in use moisture particular tools recycling from in pathways complex in the network South theory Amazon America. to The po- America using the numerical atmospheric moisture tracking model WAM-2layers (Wa- The input for WAM-2layers arefall 3 hourly Measuring precipitation Mission estimates from (TRMM)et the based al. Tropical Rain- on the algorithm 3B-42 (version 7) ( formed a posteriori with reanalysis and satellite datasets (see input data in Sect. are used to build twostatic for seasonal the moisture whole recycling season. networks Thismoisture implies is which that tracked are in forward assumed the and to proposed backward analysis, in be for space each but season 2.1.2 not in time. Input of WAM-2layers 2013 analyzed in the following. To this extent,evapotranspiration we for performed each separate forward continental tracking grid runsfor cell of the in the years South 2000–2010. American Thethe domain domain. tagged (Fig. We moisture omitted is the assumedoutput year to 2000 for from be the the gone results years whenaverage because it imports 2001–2010 of leaves model are and spin-up. The aggregated exports first between to all monthly, land then grid to cells. seasonally These seasonal averages Evapotranspiration from atracked in certain the region atmosphere bysisting of applying of water a interest well-mixed balance upper principles ispared and to to lower “tagged” full-3-D each part. tracking, grid The but and cell, was two-layer approach shown con- subsequently to is perform simplified comparably com- well ( of extreme events ( in Sect. 3.4. proach for the first time to a network in whichexplain how the we links use represent it water tomade build fluxes. moisture in recycling the networks. We proposed explainand analysis the 2.4 assumptions in and Sect. we 2.2.the present We continental the develop and complex new regional network recycling measurespresent measures ratios in and in with Sects. discuss Sect. other 2.3 new existing 2.5. results studiesSect. After on in 3.2 comparing Sect. the and 3.1, importance on we complex of networkanalysis cascading analysis of moisture in the recycling Sect. moisture in 3.3. Finally, recycling we between present an the in-depth Amazon basin and the La Plata basin the network were usually builtbetween according time to the series strength in of di the statistically relationships 5 5 25 15 20 10 20 10 15 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | , , ◦ ) as ) and . The is the 4 ii ) and at 2011 m 1965 , , Rozante and 2009 , S, which is the ◦ Mu et al. Kim and Alexander Arraut and Satyamurty ; the number of grid cells in Monteith et al. N ). Precipitation dataset from Franchito et al. ). TRMM precipitation dataset is the amount of evapotranspi- 2004 , ciently well ( ij with 2008 ffi m N 2010 , and the diagonal element ∈ j , , j i } ij ). We downscaled the MODIS evapotran- m { Grimm et al. = 17486 17485 ; 2011 ). M , Rozante et al. 1999 2011 Dee and Uppala ; , , ff 2009 Dee et al. , Ruho ; ). 1 2011 which precipitates in grid cell 681). The non-diagonal element , Liebman et al. i = ). Evapotranspiration is estimated using a recently improved algorithm ) based on the Penman–Monteith equation ( N Franchito et al. 2008 2011 , , Loarie et al. c). This is in agreement with previous studies demonstrating a maintaining of the 4 ). The evapotranspiration is high in the Amazon basin and varies little in time and ). However, it is systematically biased during the dry season in the northeastern We find that, averaged over the full time period, evapotranspiration exceeds precip- Humidity estimation has been improved in the ERA-Interim product in comparison The long term seasonal average of evapotranspiration and precipitation as well as Mu et al. System (SAMS) ( Spectroradiometer (MODIS) based on the MOD16 ET algorithm ( amount of evapotranspiration which precipitatesmoisture). in The the output same of gridcomplex cell WAM-2layers network can with (locally self-interactions, recycled transformed where nodes intoand of a links the network directed between represent nodes and gridbetween represent weighted cells them the (Fig. direction and amount of moisture transported ases in the inputrecycling data ratios as due mentioned tolocal above. precipitation, an As we overestimation this will of exclude might these the lead grid contribution cells to from of2.1.3 a our evapotranspiration bias analysis. to in Output moisture of WAM-2layers as aThe complex output network of WAM-2layers is athe matrix continent ( ration in grid cell the water from the2012). deeper water table (Nepstad et al., 1994; Miguez-Machoitation and Fan, in northeastern Brazil,explanations for in the the imbalance Atacama in Desert these and arid along to the semi-arid Andes. regions Possible are irrigation or bi- 2009 during the dry season due to the deep root system which enables the pumping of space. It exceeds the total precipitationthe in dry the season, southern indicating part that of this(Fig. the region Amazon is basin a during net source ofgreenness moisture of for the the Amazon atmosphere forests (Morton et al., 2014) and the absence of water stress are shown to represent high frequency variability su coast of Brazil, where precipitation is underestimated ( surements ( 2013 the junction of Argentina, Paraguay andCavalcanti Brazil, where it is overestimated ( TRMM are considered reliable overregion South where America and others in products particular perform in the poorly Amazonian due to the lack of ground based mea- ( well as 6 hourly specificInterim humidity reanalysis and product wind ( speed in three dimensions from the ERA- high rainfall in the South Atlantictral Convergence and Zone (including south-eastern the Brazil) Amazonto during basin, cen- the the wet dry season season (December (June to to March) September) compared characterizes the South American Monsoon 0.5 h as requested by the numerical scheme of WAM-2layers. with others reanalysis products ( parametrization of the algorithm.with The 10 evapotranspiration eddy dataset fluxtypes has ( towers been located validated in the Amazonian regionmoisture under flux various divergence (evapotranspiration land – cover precipitation) are shown in Fig. spiration to a 3 hourlytemporal resolution resolution based in onlongitude/latitude the ERA-Interim, and as input WAM-2layers covers data. needs thesouthernmost finer All latitude South covered data American by TRMM is continent product. upscaled In until addition, to 50 all data a is downscaled regular to grid of 1.5 forced by satellite dataity of from the MODIS evapotranspiration dataset and depends meteorological on reanalysis the data. quality of The the qual- input data and the 5 5 20 10 15 15 10 25 20 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | ). for Ω ). The 2004 , : (see also i 2004 Ω Eltahir and , which can be of Savenije ). We then track is the transported gives the regional Ω m Savenije Ω 2.3.1 ). : Ω 2011 , is the tagged fraction of trans- Ω m is the total precipitation in is the tagged fraction of precipitation, i P Ω that is originating from evapotranspiration P i is the total precipitation, 17488 17487 P ). The first condition is usually fulfilled during inter- in the case of a backward tracking and the share of averaged over all grid cells in Ω van der Ent and Savenije that contributes directly to precipitation over Ω ; 2013 in the case of a forward tracking. i ρ , Ω 2006 , quantifies the dependency on direct moisture recycling from Ω , (1) ρ Ω . We note that ). m is the amount of precipitation in , (3) Burde et al. , (2) m ). Considered together, the direct and cascading moisture recycling ratios Ω ; Ω Ω Ω A1 that comes directly from evapotranspiration from i = i is the total evapotranspiration, → ← ← i P E i i i through direct moisture recycling and Ω Goessling and Reick P E P E 2.3.2 E E 1994 Ω , = = = is the tagged fraction of evapotranspiration and i i , , We also define the direct evapotranspiration recycling ratio as the fraction of evapo- This assumption is valid under two conditions: (1), evapotranspiration follows directly Ω Ω Ω Ω P P region. We first consider direct moisture recycling only (Sect. In this section, webackward define in measures space. In to thenation track case (precipitative moisture of sink) from a of a forward evapotranspirationward tracking, from given the tracking, the region measures region. they forward indicate In indicate or the the desti- the case of origin a (evaporative back- source) of precipitation over the 2.3 Moisture recycling ratio early proportional to theof tagged transported fraction moisture: of evapotranspiration and the tagged fraction In order to track moistureany forward shape or and backward scale fromposition (grid a within given cell, region the basin, continent),the surface same. we reservoir This assume and implies that that, the the in atmosphere moisture each grid for com- cell, each the grid tagged fraction cell of remains precipitation is lin- 2.2 Basic assumptions transpiration in each grid cell ε local rainfall. High values indicatesink) locations which from receive moisture (directrecycling precipitative ratio, i.e, theBras fraction of precipitation that is regionally recycled ( We define the direct precipitationgrid recycling cell ratio as the fraction of precipitation in each ρ from Appendix (Sect. provide a full picture of the sources and sinks2.3.1 of moisture from Direct a moisture region recycling of ratios interest. where moisture further forward or backward in space through cascading moisture recycling (in term of tagged fraction) as the atmospheric moisture at the end of the season. However, in seasonal forests with deepduring the rooted dry trees, season the can moisture besecond which hold condition back is is for evaporated one fulfilled or if several the months ( soil water at the beginning has the same composition moisture towards or fromE another grid cell, ported moisture towards orof from the another moisture grid originating cell. from moisture We precipitating call over “tagged fraction” the share after the precipitation event orvoir (2), and the fraction the of atmosphere taggedstate) moisture ( can in be the surface assumed reser- to be temporally constant (i.e., in steady where ception and fast transpiration whichspiration, particularly are important in warm components of and the for total shallow evapotran- rooted plants ( 5 5 20 25 10 15 25 15 10 20 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | ) ) c a), ρ and 2 casc Am con- (see ρ i Am through through casc Am ρ , Table 2). ). ρ Ω Ω indicate re- A2 c 2010 ρ , may be counted ). While being all grid cells through cascading Ω and Ω Ω c casc Pl . This implies that we ε ε Ω with quantifies the dependency Ω ρ casc Ω b). that contributes to precipitation that contributes to precipitation van der Ent et al. ρ 2 , see also Appendix i i quantifies the local contribution to 2 being all grid cells covering the La becomes the continental evapotran- the total evapotranspiration in that is originating from evapotranspi- Ω Ω casc Ω i i ε ε ), we use E 7 with for local rainfall. High values indicate the fi- only (Fig. Ω 17490 17489 ε Ω Ω ). Considered together, the continental precipitation recycling ratios ( which contributes to precipitation over i after one ore more re-evaporation cycles (indirect Ω 2010 ) and ρ Ω ( being all grid cells covering the Amazon basin ( . Am Ω is useful to track moisture from the Amazon basin as Ω ρ ) and Am c with ρ ε quantifies the local contribution to precipitation over casc Ω Ω through cascading moisture recycling. High values indicate loca- ρ ε Ω ). being all grid cells covering the La Plata basin ( ). While that is originating from evapotranspiration from A1 . Pl i Ω through cascading moisture recycling. van der Ent et al. ε , (4) , (5) is the amount of evapotranspiration in is the amount of evapotranspiration in Ω is the amount of precipitation in provides information regarding the spatial heterogeneity of the contribution Ω Ω Ω Ω Ω i i Ω Pl → with ← casc P casc E i → → i ← ε casc casc i i i P through direct moisture recycling and E through cascading moisture recycling. P E E is the entire South American continent, indicate together the precipitative sink region of moisture from the Amazon basin casc Ω provides information regarding the importance of cascading recycling of moisture = = Ω Ω ε Ω i i , , Considered together, the direct and cascading recycling ratios provide a full picture We also define the cascading evapotranspiration recycling ratio as the fraction of The moisture inflow (resp. outflow) that crosses the border of To study the cascading moisture recycling between the Amazon basin and the La To study the direct moisture recycling between the Amazon basin and the La Plata If The direct recycling ratios underestimate the strength of the relationship, in term of casc Am casc Ω casc Pl casc Ω evapotranspiration in each grid cell cascading moisture recycling as: ε ration from on cascading moisture recyclingnal from destination of moistureprecipitative from sink). where over also Appendix where of the origin andρ destination of moisture from a given region. High values of where over precipitation over tions where moisture initiallytributed over originated (indirect evaporative source) before itseveral is times dis- as it isavoid involved this, in we several pathways onlyconsider of track re-evaporation cascading moisture cycles moisture outside that recycling. crosses To the border of that has final destination the La Plata basin (see Fig. Plata basin, we use and siders cascading recycling of moisture originatingε from the Amazon basin (see Fig. spectively sources and sinks ofcontributions continental of moisture. moisture In this incontributions study South to we the America neglect overall from moisture possible budget and are small to ( other continents, since these basin (defined by the red boundaries in Fig. moisture recycling: ρ as defined in 2.3.2 Cascading moisture recycling ratios We define the cascadingeach precipitation recycling grid ratio cell as the fraction of precipitation in direct moisture recycling. Highrect values evaporative indicate source) locations over which distribute moisture (di- spiration recycling ratio ( covering the Amazon basinPlata ( basin ( a whole, inside the Amazon basin. moisture recycling, between two specificnot locations. In only fact, through moisture direct cancling. moisture be exchanged recycling, but also through cascading moisture recy- 5 5 25 10 15 20 10 25 20 15 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | , ). In is the Zemp ; /E 2007 c , E Zemp et al. 2007 ∆ , ) (see Appendix c Fagiolo E ∆ Fagiolo . A grid cell has a high )( ˜ C C1 ). Here, we are interested ). Here, it reveals intermedi- 2002 , 2007 , ) and outflow ( c P ∆ than the 80 percentile (corresponding to Fagiolo Milo et al. /E c 17492 17491 E cient associated with Middleman motifs and be- ∆ ffi highlight evaporative source regions of precipitation is the fraction of total moisture inflow that comes from casc Am /P ρ m P cient associated with Middleman motifs ∆ and ). ffi cient is a measure from complex network analysis which mea- is equal to zero if the grid cell forms no Middleman motif at all. B ffi Am ˜ cient associated with Middleman motifs ( C ρ ffi ) for each grid cell as described in the Appendix is the fraction of precipitation that comes from cascading moisture recycling /P c 2014 ) (see Appendix P ), but are not analyzed here. , m ∆ P if it forms a lot of Middleman motifs and if these motifs contribute largely to relative It is worth to note that the Middleman motif considers three interconnected grid cells The clustering coe In addition, we are interested in the importance of cascading moisture recycling in the ). ∆ ˜ 2.5.1 Clustering coe In complex network theory, motifsinterconnections are that defined occur as in significant the network and ( recurring patterns of We investigate important moisture recyclingplex pathways network using analysis: two clustering measures coe tweenness from centrality. com- 2.5 Complex network analysis intermediary regions for the totalas moisture above. in- and We outflow. We forbiding use from re-evaporation the the in same continent approach the( and we intermediary estimate region the resulting of reduction moisture in originat- total moisture inflow the intermediary region for local rainfall. continent. It quantifies the contributioncycling. of Regions a which specific have location a0.27 to larger during cascading the moisture wet re- regions". season and 0.17 during the dry season) are called "intermediary cascading moisture recycling in theof intermediary precipitation region. that It comes canintermediary from be the region. seen continent as It and the quantifies fraction that the has dependency been on re-evaporated cascading in moisture the recycling in for the total moisture inflowthis, (precipitation) we and forbid outflow re-evaporation of (evapotranspiration). moisture To quantify originatingthe from resulting the continent reduction and in we estimate B total moisture inflow ( on the continent, i.e., thattifies has been the evaporated dependency at on leastfraction twice cascading on of moisture the total recycling continent. evapotranspiration It for thatthe quan- local continent, is rainfall. i.e., involved that in comes cascading from moisture the recycling continent on and that further precipitates over the over the La Plata basin. 2.4 Quantifying cascading moisture recycling In this section, we are interested in the importance of cascading moisture recycling and high values of tifs formed by three2014 or more grid cells linked by moisture recycling exist ( which corresponds to cascadingevaporation cycle. moisture These recycling pathways pathwaystween contribute two involving usually locations. only most In one to fact,decreases the re- moisture with amount the transport of number be- of moisture re-evaporation transported cycles in involved in a the pathway pathway. Other typically mo- sures the tendency to form aary particular locations motif in ( cascading moisturerecycling recycling of pathways, moisture as between the sources alternativethe and to sinks. clustering the We coe compute direct theet weighted al. version of C moisture transport. in a particularour pattern study, of a directed gridpathway triangles: cell to forms the the direct a Middleman transport Middleman motif of motif moisture ( if between two is other an grid intermediary cells (Fig. on 3). an alternative 5 5 15 20 10 25 25 15 20 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | ˜ B C erent d and ff 4 , p. 36). ) and the 1977 , if many optimal ). The southern 2 B 2.5.2 Freeman (Sect. relies on particular motifs B ˜ C b and g), this observation indi- 4 e and j and Table 4 ). A grid cell has a high C2 17494 17493 refers to cascading moisture recycling pathways ˜ are complementary measures. C , all number of cycles are possible in the pathways. ˜ C B ) also found a maximum contribution of moisture from ) aims to highlight nodes in the network with central B . erences between the betweenness centrality and the 2014 ( ff B and the B is based on the optimal pathways, the B , only cascading moisture recycling pathways with one re-evaporation ). In fact, in the La Plata basin, 42 % of the precipitation during the wet ˜ C 2 Drumond et al. ). In fact, both measures reveal important intermediary grid cells in cascad- shows locations where cascading moisture recycling pathways are channeled. Middleman motif 2.5.1 existing studies In the cycle are considered. In the An implication of (1) isas that the alternative to the directB transport of moisture between two locations, while the Moisture recycling pathways involving long-rangethe transport calculation are of not the considered in While the formed by three connected grid cells. The main sink regions of moisture originating from the continent are the western part To compute it, we first identify for each pair of grid cells the moisture recycling path- 3. 4. 2. 1. The main continental sourcewith of large precipitation heterogeneity in in time South and America space is (Fig. the Amazon basin, For these reasons, the 3 Results and discussion 3.1 Comparison of continental and regional moisture recycling ratios with other The betweenness centralityposition ( “to theimpede degree or that bias they the transmission stand of between messages” in others the and network can ( therefore facilitate, 2.5.2 Betweenness centrality of the Amazon basin duringthe the wet dry season season, the and south-westerni the part La and of Plata Table the basin basin during especially during the wet season (Fig. cates that during the drythe Amazon season, basin a is high advected amountsion out of model, of moisture the continent. from Using thethe a southern Amazon Lagrangian part basin particle of to disper- the ocean during this period. part of the Amazonseason basin but contributes only 80 40 % %basin during to is the continental dry high precipitation season. and during As varies the the little wet evapotranspiration in in space the Amazon and time (Fig. than 15 geographical degrees. Thesults choice qualitatively. Once of optimal this pathways are thresholdthat identified, does they we have not find in influence intermediary common the grid (see re- cells Appendix ways with theinclude greatest any throughput, number called ofdirect “optimal one re-evaporation pathways”. (without cycles. These any Asthat re-evaporation pathways the cycle), the can optimal we optimal pathways first pathway involve hadfrom cascading is the to moisture network recycling. usually modify all To the the do long-range network so, moisture we such transport, removed i.e, occurring over distances larger We expect similar spatial patterns in the results of the Here, we use itpathways to are channeled. reveal intermediary grid cells where cascading moisture recycling pathways pass through it,equal i.e., to moisture 0 often if cascadesthe none grid of in cell. these the pathways grid pass cell, through and it, has i.e.,2.5.3 moisture a never cascades Similarities in and di ing moisture recycling pathways. However, theseconcepts two and measures methods. are based on di (Sect. 5 5 20 15 10 10 20 15 25 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | , ); 2008 erence ( ff Drumond Bosilovich Martinez et al. ). Because the ; erent from those ff erences in the ver- ff 2011 2012 Drumond et al. c, and g, a and e) are ). However, some other , , 7 , the Amazon region is 2014 , 1993 ). The inverse situation is ob- ). The spatial patterns of con- 2 , 1 Keys et al. ; Lewis et al. , Moisture Sources by Basin). ) compares well with previous studies ; 2 2010 ). However, this latter observation should , 2008 Drumond et al. a and c), one of the most vulnerable biodi- , ; 5 d, i, e and j) are slightly di 17496 17495 2000 Brubaker et al. 4 , 2009 , Figs. 3 and 4) due to the di , a and c) are also dominant sinks of moisture from the 2010 5 ). The importance of continental moisture recycling in the , Myers et al. http://www.iges.org/wcr/ van der Ent et al. Marengo et al. 2004 , , we reveal the evaporative source of cascading moisture which see ). Despite this importance, we find that the ocean remains the main ), this might influence the output of the atmospheric moisture track- 3.4 ). The maps of direct recycling ratios (Fig. d and i). 2009 ). We now quantify this importance and identify intermediary and sink 2014 4 , ( 2014 4 , 2006 Arraut and Satyamurty ( ; van der Ent et al. Marengo et al. ). In Sect. 2 2008 ). , The share of cascading moisture on total moisture inflow reaches up to 50 % in the We note that cascading moisture contributes more to the precipitation over the Ama- We note that our analysis period from 2001–2010 includes two major droughts There are uncertainties in the moisture recycling ratios depending on the quality of Bagley et al. ( land–atmosphere coupling on the hydrological cycles increases during drought years between seasons is explainedwith by the a subtropical weaker Atlantic transport highLevel and of Jet by oceanic an (SALLJ) moisture intensificationseason associated which of ( the transports South American moisture Low- in the meridional direction during this season and 35 % during the dry season evaporated from the continent. This di might lead to anestimation over-estimation of of the importance the of regional cascading recycling moisture recycling. process and thus an over- precipitates over the La Plata basin and we understand thiseastern seasonal side variability. of theversity hotspots central on Andes Earth (Fig. be ( considered with caution due to the imbalance of the water cycle in this area which served in the Ladry Plata season and basin, 17 % where(Table during on the average wet season 14 % comes from of cascading the moisture precipitation recycling during the South American continent. Regionscycling which for are local dependent rainfallcontinent on (Fig. (Fig. cascading moisture re- zon basin during thecompared dry to the season wet (11 % season on (8 % average, on up average) to (Table 25 % in the western part) regions of cascading moisture recycling. 3.2.1 The dependency on cascading moistureThe recycling share of cascading moisture on total moisture inflow is on average 10 % in the ing model used in this study. Analyzing these periods3.2 separately is ongoing research. Importance of cascading moisture recycling Continental moisture recycling is of crucialpatterns importance (Fig. for South American precipitation precipitation recycling ratio inHemisphere the summer Amazon is basin compatibleand with reaching Chern the 30 % estimation during of the 36.4 % Southern found by using other datasets andtinental methodologies moisture (See recycling Table ratiosfound (Fig. by ( sions of the model (here we use WAM-2layers) and the datasets used. The continental Dirmeyer et al. in good agreement withies maps (Eltahir of and regional Bras, recycling 1994, ratio Figs. presented 4 in and 6 previous stud- and Burde et al., 2006, Figs. 2 and 8 and represented by a rectangle).tion Considering recycling ratio these in uncertainties, the the Amazon basin regional (Table precipita- in the Amazon basin ( source of moisture in theet La al. Plata basin in agreement with previous studies ( Martinez et al. La Plata basin during the wet season has been emphasized in the datasets used, theto assumptions define made the in domain the (for methods example and in the boundaries used studies estimated a higherover contribution the of moisture La from Plata2014 the basin continent ( to precipitation 5 5 20 25 15 10 20 25 15 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | b ). 5 6 ). In the 4 a and c). This 6 ect of the acceler- ff ) and 23.9 % found coincide with regions B cient associated with the ) and the high precipitation ffi 2006 ) allowing for an intensive local , 4 a). ). These results mean that the south- b and d). 5 2 5 ). This result suggests that the impact of c), but not during the dry season (Fig. Vera et al. 6 http://www.iges.org/wcr/ 17498 17497 2004 , , see , as we have shown that the incoming moisture over B ) reveals intermediary regions where cascading mois- ). As mentioned above, this might be explained by the 2009 B ). Using the clustering coe 2 ( 5 a) increases the dependency from 23 to 29 % during the 2 ). This is compatible with the yearly average estimates of 2 ). Considering cascading moisture recycling once moisture has Marengo et al. ), we are able to identify intermediary locations involved in cas- ˜ C h and Table 2014 7 ( Dirmeyer et al. during the wet season (Fig. ˜ C d). This observation may be explained by the combined e 6 c and g and Table Plata basin 7 Martinez et al. During the wet season, cascading moisture recycling pathways are channeled in the In the La Plata basin, 23 % of the precipitation during the wet season and 25 % during The betweenness centrality ( In order to quantify the importance of these intermediary regions for regional rainfall south-western part of the Amazondes basin (Fig. and a narrow band east of the subtropical An- the La Plata basin duringlong-range) the moisture dry recycling season is pathways (Fig. mainly transported through direct (and thus pathways in the calculating of the The contribution of evapotranspirationrecycling from in the the Amazon continent basin reachesof to up the to cascading Amazon 25 basin moisture % and during 35 the % dry during season the wet in season the in central its part southwestern part (Fig. 3.2.2 The contribution of intermediary regions to cascading moisture recycling deforestation in the Amazonianmight forest be on larger thebasins than are moisture expected considered. supply if in only the direct La transport Plata of basin moisture between the two left the Amazon basin (Fig. wet season (Fig. high evapotranspiration and precipitation allowingway, downwind for an of exchange the ofthis Amazon moisture time basin, on of and the the by year the ( intensification of the SALLJ during This non-overlap is probably explained by the cutting of long-range moisture recycling ation of the South American Low Level Jet ( following, we further investigate the transport mechanism between the twothe basins. dry season(Fig. originated from the Amazon23 % basin found through in directin moisture recycling and evapotranspiration during the wet season (Fig. We have shown the importancetinental of moisture the and Amazon the basin La as Plata the basin dominant source as of a con- central sink region (see Fig. exchange of moisture between the vegetation and the atmosphere. 3.4 Direct and cascading moisture recycling from the Amazon basin to the La ture recycling pathwayswhich are high channeled. Regions with high cading moisture recycling asThese alternative regions pathways are to the central thesouth-western part direct part of transport the of of Amazon theis basin moisture. Amazon during in basin the good dry agreement during with seasonregions the and other to wet the measures cascading season quantifying moisture (Fig. the recycling contribution (Fig. of intermediary western part of the Amazontransported basin towards is the an La important Plata intermediary basin for during cascading the moisture wet3.3 season. Complex network analysis We have shown themoisture importance transport of (see cascadingMiddleman moisture Fig. recycling motif for ( South American over the La Platabasin basin, that has we cascaded quantify in the4 these % regions. share This during of share the is thetotal 8 dry moisture % moisture during season. inflow inflow the in These that wetand the season estimations has and La one represent cascaded Plata in almost third the half during entire of the continent the dry during share the season of wet (Table season and d). These regionspathways. are important intermediaries in cascading moisture recycling 5 5 25 20 15 10 25 20 10 15 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | , f). 7 Hirota et al. ). Figueroa and 3.4 e). This finding is in agree- 7 a and ). This result means that during the 7 ). In the eastern side of the central 2 2012 b), the entire Amazon basin becomes the , 2 17499 17500 ected regions, important impacts would be ob- ff Keys et al. ) who found that the southern part of the basin is an ; 2014 2009 ( , ). Martinez et al. Rockström et al. 1990 , ) and in the La Plata basin which is dependent on incoming rainfall for the agri- Land cover change in the southern part of the Amazon basin might weaken continen- The La Plata basin is highly dependent on moisture from the Amazon basin during Using the atmospheric moisture tracking model WAM-2layers (WAM-2 layers) forced The southern part of the Amazon basin is a direct source of precipitation over the La both seasons, aswet 23 % season and of 25 % the duringthese direct the total dependencies, dry 6 precipitation season % of comes overif the directly precipitation the from cascading during the moisture La the Amazon wet recycling basin. Plata seasondry outside To can basin be the season, added Amazon during the basin the southern main are part considered. source During of of the theAmazon continental Amazon basin moisture basin. is over During not thedistributes the moisture only La wet from a Plata season, the the source basinchange entire southern in region is Amazon these part but the regions, basin of which is into include the also the the an arc La of intermediary Plata deforestation, basin. may region weaken Land which moisture use recycling pathways is typicallysphere, smaller the contribution than by the the oneIn ensemble of fact, transported cascading 10 % directly pathways can of in notover the the be the total neglected. La precipitation atmo- Plata over basin South comes America from and cascading 17 % moisture of recycling. the precipitation re-evaporation cycles along the way.portance We of have cascading proposed moisture measures recyclingsinks to and of quantify to moisture reveal the for direct im- a andapproach given indirect region. to sources We and have identify usedways. intermediary for the regions first time in a the complex network cascading moistureby recycling precipitation from path- TRMM andhave evapotranspiration from shown MODIS that in even South if America, we the amount of water transported through cascading moisture In this work we investigatedatmosphere the on exchange of the moisture way betweenAmerica. between the We sources have vegetation introduced and and the the sinksmoisture concept recycling of of “cascading between continental moisture two moisture recycling” locations in to on refer South the to continent which involve one or more 4 Conclusions Plata basin, with a direct contributiondry reaching season 15 and % 35 of % its during evapotranspiration the during wet the season (Figs. Andes, the impact ofreduced an since upwind precipitation weakening in of thisNobre cascading region moisture is recycling insured by might orographic be lifting ( ing intermediary locations involved in cascading moisture recycling (see3.5 Sect. 3.2). Possible impact of land coverThe change southern in the part intermediary of regions the the La Amazon Plata basin basin. is Ityear a is round a key and source region it of for is moisture moistureoriginating in for transport from addition precipitation towards the over an the intermediary entire La Amazon region Plata basin for basin cascading during all moisture the recycling wet season (Sect. culture ( wet season, the southern partalso of an the intermediary Amazon basin region is whereon not moisture its only originating way a to from source the the of La entire Plata moisture basin basin. but This cascades finding is in agreement with other measures show- tal moisture recycling and mightlocally lead to and an downwind. important decrease Amongserved in in the the particular total a precipitation ina the high south-western probability part to experience of2011 a the critical Amazon transition basin from which forest to has savanna already ( The indirect contribution representsthe on Amazon average basin 7 % during of the wet the season total (Table evapotranspiration in evaporative source of moisture over the La Plata basin during the wet season (Fig. ment with quasi-permanent direct source of moistureing moisture for the recycling are La considered Plata (Fig. basin. However, if cascad- 5 5 25 10 15 25 20 20 10 15 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | ff ). is j ers A1 P ff and j (Sect. that comes j Ω ). which precipitates is calculated as: through cascading 100, while the con- i Ω = 2014 Ω ( n through cascading moisture and precipitates in re-evaporation cycles is: i Ω n Zemp et al. that comes from that comes directly from i j ), the fraction of precipitation in that comes directly from ) which add up to the total cascading moisture 17502 17501 that comes from } A3 j . j n i , ... 1 re-evaporation cycles. The total fraction of precipi- 1, through cascading moisture recycling is the sum of all − ∈ { n Ω k , (A4) , (A3) 1) i i . The fraction of evapotranspiration in grid cell − , , j n ( Ω Ω ρ ρ · · , (A1) , (A2) j ij ji ij ij j P is the fraction of precipitation in P is calculated as: m m m m i j is the fraction of precipitation in 1) that comes from Ω Ω Ω is the amount of moisture which evaporates in Ω P Ω i E − j , is the evapotranspiration in 6∈ 6∈ ∈ ∈ i i i j n j ij Ω ( Ω i through cascading moisture recycling involving ρ P P P ρ E P m Ω = = = re-evaporation cycles ( = j j j i ) , , , , k In addition, we showed that the eastern flank of the subtropical Andes – located in n Ω ( (1) Ω Ω Ω recycling involving one re-evaporation cycle is: ρ tribution of pathways with numbermoisture of forward cycles in larger space, than weon 3 have the to are way take close during into to thetracking account zero. cascading in If that recycling. space we moisture This because track is is wearrives lost not quantify at as the the destination. runo case remaining for amount moisture of backward moisture thatA2.1 actually Cascading precipitation recycling ratio The fraction of precipitation in grid cell where To calculate the cascading moistureculate recycling the individual ratios contributions as of definedof cascading in moisture Sect. recycling pathways 2.3.2, consisting werecycling cal- contribution. We chose a maximum number of cycles A2 Cascading moisture recycling ratio the water cycle needed in a context of and climateAppendix change. A: Moisture recycling ratios In these measures thethe irregular grid sizes cells of are the taken portion into account of as the described Earth’s in surface covered by natural ecosystems regionally and downwind as previously thought. the pathway of the Southtinental American moisture Low recycling Level Jet as – itnew plays methods channels an to many important improve cascading our role pathways. understanding in This of the vegetation study con- and o atmosphere feedback on recycling processes and may have stronger consequences for rainfed agriculture and where moisture recycling involving tation in from ρ Following the same principle as in Eq. ( where the precipitation in directly in ε A1 Direct moisture recycling ratio The fraction of precipitation in grid cell ρ where 5 5 20 10 15 20 15 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | ). Ω Ω ij (A8) (B3a) (B4a) continent < m through ): ← j 1 Ω P that precip- ocean − i j ← which precipi- P ij i = m re-evaporation cycles ocean n ← j P ( i that precipitates over which precipitates directly over j originating from the re-evaporation j j that results from evaporation of mois- that contributes to precipitation over . j through cascading moisture recycling is i Ω Ω can be seen as the evapotranspiration of and 1 re-evaporation cycles. The total fraction of o , i − Ω 17504 17503 n → i E .). Using the same assumption, we get the moisture ij m , (A7) , (B2) . Similarly, the fraction of evapotranspiration in . Similarly, j j , (A6) , (B4b) for each grid cell: can be interpreted as the evapotranspiration in α j Ω , (B1) . (A5) · o that precipitates over /P . (B3b) , o j α , ) j , i which precipitates over continent Ω ) i ocean · , Ω 1) n Ω ∈ n ( Ω E i j j i : ( Ω − ← , , → i ρ ← n i ocean ocean ε ( Ω Ω = j Ω ocean P ocean E E + P ← ε ε ← j i · + i ← · · ← i is the precipitation from oceanic origin in = ij α − E P − ij i ij ij E ij ij · through cascading moisture recycling involving ...... m Ω is the fraction of evapotranspiration in E Ω i m i m m m m + → is the fraction of evapotranspiration in + P ← Ω E Ω i Ω 1) j j Ω Ω i ocean , = ∈ , j j ∈ − , , E 6∈ 6∈ P ) and j X and which has been evaporated from the ocean before that ( i X (1) Ω = n j j (1) Ω ← continent Ω ( Ω i j ρ = ε can be interpreted as the precipitation in = = = ε P ε P P ← A1 j o Ω o o = Ω = , ocean , , P = = Ω ocean → Ω Ω ← ← j i i i i j ) , , , , or outflow from Thus, are able to derive the corresponding reduction in total moisture inflow towards In a second step, we derive the corresponding moisture in- and outflow from or to- → ← ij ← ← n P E i i casc Ω Ω casc Ω ( (1) Ω j j Fig. B1). To achieve this,oceanic we origin derive (i.e., for that each has grid been cell last the evaporated evaporation over of the moisture ocean) as from in Eq. ( we inhibit all cascading recycling of moisture from continental origin in the network (see To quantify the contributionoutflow, of we remove cascading the moisture re-evaporation of recycling moisture to from continental total origin. moisture By in- doing so, and where cascading moisture recycling involving evapotranspiration in the sum of the individual contribution of cascadingε recycling: Appendix B: Quantifying cascading moisture recycling is: ε The fraction of evapotranspiration inthrough grid cascading moisture cell recycling involving one re-evaporation cycle is: ε where (Sect. itates over ρ A2.2 Cascading evapotranspiration recycling ratio individual contributions of cascading recycling: At this stage, tates in wards a given region P E P of oceanic moisture in oceanic moisture in Ω ∆ ∆ transport between each pair ofture grid from cells oceanic origin only: m E where and 5 5 10 20 15 25 15 20 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | . . 0 ). = j Ω be- = P Ω 2.4 with with i Ω → ), the ij E which ˜ ii j ( C a ) ← i E / j P P 2 ij T 2007 ∆ and m ( and PP j cient and the ( = ffi Ω ij = i p ˜ t Fagiolo and pointing towards is the total number of i i T ) as defined in Sect. m is: E i : i ∆ originating from is the number of arrows pointing j in and pointing towards i forms and (resp. k i i c ) can be derived as E C2 ∆ ) is: which contributes to precipitation over where of the grid cell i C1 C out i 17506 17505 cient is derived from the methodology of a pre- k (i.e., the sum of the weights of the Middleman ffi in i i t k becomes . The contribution of this pathway is defined as the cient j = ffi Ω i → T i is the re-evaporation of continental moisture in E ∆ Ω ) is → , Table 1) and has been corrected in order to account for the i C2 ). is the total precipitation in E i to grid cell cient associated with Middleman motifs is the weight of the arrow originating from ) and ∆ i ij 2007 m ffi ). The numerator of Eq. ( ij m , P p . the number of arrows originating from Ω ) be the intermediary grid cells in a cascading moisture recycling ∆ and ∈ n Ω i t could have formed according to its number of incoming and outgoing 2014 out i , Ω and P i , k is the precipitation in j originating from the re-evaporation of continen- N = Fagiolo ... , (C3b) 1 if there is an arrow originating from ∈ Ω (resp. , , (C3a) j ij , Ω 2 i c ji ← = and t a is the total evapotranspiration in j } is the number of Middleman motifs that ← , P a i i 3 j P ij 1 i i ij 6= / ∆ the weighted counterpart of , (C2) , (C1) t P t a j ∆ 6= 1 ij X is the entire South American continent (resp. the intermediary region), i i i i j i m ˜ X t t T T ˜ p t = { Ω Ω = Zemp et al. = = ∈ Let ( The calculation of the clustering coe If j = i i in out i i . Here, in order to avoid a strong correlation between the clustering coe ˜ in ( irregular sizes of the portion of the Earth’s surface covered by the grid cells as explained comes P Thus, tal moisture in precipitates over where As this pathway iswe not will necessarily call the it “optimal shortest pathway” one to in avoid confusion. termpathway of from geographical grid distance, cell In complex network theory,ness) many are based centrality on measures theas concept (e.g. of the a closeness pathway shortest between and path. nodesas between- The which the shortest has pathway path which the is contributes minimum most usually cost. to defined the In moisture this transport work, between two it grid is cells. defined where otherwise. In order tothe compare maximum the observed value results for for each the network. twoC2 seasons, we normalize Optimal pathways and betweenness C2.1 Optimal pathway k k P j mean evapotranspiration and precipitation, we chose this weightThe to denominator be of Eq. ( towards with motifs that is formed by vious study ( C weight of a motif isarrows. defined The as weighted the counterpart geometric of mean Eq. of ( the weights of the three involved Mathematically, the clustering coe C where that motif that arrows. To give moreof weight moisture, to we assign a a motif weight involved to in each the motif. transport In agreement of with a larger amount Appendix C: Complex network analysis C1 Clustering coe 5 5 20 15 10 20 15 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | = . , it 1 0 + l (C5) (C4) t W l t ) is the C2.1 i w ( jk σ in the package ). In this method, , and k through cascading i 2001 in the package iGraph ( and as calculated in iGraph as defined above. reaches values between j j : i . The contribution of each W B . i B1 Newman shortest_paths ) to grid cell j i n j betweenness t P is the fraction of the number of optimal m i log( − 17508 17507 ) ) the number of grid cells. To calculate it, we used 1 j + n 1 j l t + P t N l l m t t · P j ) m j n 1 n t j t 1 + l P + t w l l 2 with m t t log( P · / + 1 m that comes from evapotranspiration in 1 1 1 1 ( − = 2) ) + j + l 1 1 n X l l 1 j t + t , − l + l l = t t n − t n l t P , ) N w m Q 1 ... 3 1 1 · 1 , 1 1 it 1 t 1 − This paper was developed within the scope of the IRTG 1740/TRP = t − 1 − l = P , n X t 1 l Y i m n t 1 2 · P + m W N 1 1 ( 1 t it t log( 1 ) = P i m w − log( log( ( ). The cost of a pathway from grid cell  jk 1 1 = = = = = jk σ 1 + − l 2 j j σ + is the number of optimal pathways between grid cells t , , l l N t n n t t t P k , ,  , jk m j X ...... σ , , 1 1 t t = , , 0 To find the optimal pathway, we use the method i i log( i Mathematically, betweenness of the gridpathways cell between any pair of grid cells which pass through B C2.2 Betweenness centrality with number of these pathways that pass through0 the and grid cell the directed and weighted version offor Python. the method The choice of the weights used in this method is explained in Sect. moisture recycling: W fraction of precipitation in the network schemes. 2011/50151-0, fundedStordalen by Foundation and the BMBFthank DFG/FAPESP. (project K. GLUES) J. Thonicke and for Donges R. comments J. on van acknowledges the der manuscript funding Ent and from P. Manceaux from NWO/ALW. for We his the help on designing Acknowledgements. project together with J. Heinke and supervised it together with A. Rammig. to adapt the method to− our purpose, we chosebecomes: the weight of the arrows as W Because the optimal pathwaycorresponds is to defined the as pathway the with pathway the with maximum contribution the minimum cost Author contribution J. F. Donges, H. M. J. Barbosa,sis. C.-F. Schleussner R. and J. D. Van C. der Zemp Ent,the developed the performed mask analy- the simulation of ofthe the WAM-2layers. G. manuscript Sampaio La with provided contributions Plata from all basin. co-authors. D. C.-F. Schleussner C. conceived the Zemp performed the analysis and prepared existing pathway is calculated between anypathway pair is of grid the cells path in with the the network. maximum The contribution. optimal iGraph for Python based on anthe algorithm cost proposed by of a pathway is calculated as the sum of the weight of its arrows. In order An example of pathway contributions is provided in Fig. 5 5 20 15 15 10 20 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | , er- ff 17481 , 2010. 17481 17516 , 17483 17516 17498 , , 10.5194/hessd-11-1023-2014 17485 17481 , 2009b. 17495 17495 , , 17482 17516 17505 10.1029/2009JG001179 , 17481 , 17481 , 17495 17488 17492 , , 17510 17509 17480 17496 17481 , 17481 , 2007. 17483 , 17481 17481 17495 , 2008. 17480 , 10.1209/0295-5075/87/48007 17483 17516 17495 , , 17486 , 17495 17516 17494 , , , 17481 17488 17482 17481 , , 10.1103/PhysRevE.76.026107 10.1029/2007JD009547 Soc., 120, 861–880, 1994. doi: the Amazon basin wet-season climate, J. Climate, 21, 1153–1170,orandum, 2008. ECMWF, Shinfield Park, Reading, England, 2008. McNally, A. P., Monge-Sanz, B.Tavolato, M., C., Morcrette, Thépaut, J.-J., J.-N., Park,performance and B.-K., of Vitart, Peubey, the C., F.: data de The17485 assimilation Rosnay, P., system, ERA-Interim Q. reanalysis: J. configuration Roy. Meteor. and Soc., 137, 553–597,between 2011. nations, J. Hydrol., 365, 11–22, 2009. 174, 157–179, 2009a. phys. Lett., 87, 48007, doi: 2014. Balmaseda, M. A., Balsamo, G.,Bidlot, Bauer, J., P., Bechtold, Bormann, P., Beljaars, N.,Healy, A. S. Delsol, C. B., C., M., Hersbach, van Dragani, H., de Hólm, R., Berg, E. Fuentes, L., V., M., Isaksen, L., Geer, Kållberg, A. P., J., Köhler, M., Haimberger, Matricardi, L., M., – Comparing linear and nonlinear network construction methods, Eur. Phys. J.-Spec. Top., zon basin moisture on theysis, atmospheric Hydrol. branch Earth of the Syst. hydrological Sci. cycle: a Discuss., Lagrangian 11, anal- 1023–1046, doi: nian forest dieback under climate-carbonClimatol., cycle 78, projections 137–156, for 2004. the 21st century, Theor. Appl. sources of moisture over Centraldoi: Brazil and La Plata basin, J. Geophys. Res., 113, D14128, looking at large-scalerainfall moisture in South transport America, and J. Climate, its 25, relation 543–556, 2012. toestation: has Amazonia land and cover change toClimate, influenced 27, recent subtropical 345–361, precipitation 2014. extremes in the Amazon?, J. sphere with emphasis on1912, 2009. the South American region, J. Appl. Meteorol. Clim., 48, 1902– atmosphere interactions in simulatedunder Amazonian global precipitation climate decrease warming, and Theor. Appl. forest Climatol., dieback 78, 157–175,spatial 2004. patterns of extreme rainfall eventsRes. of the Lett., South 40, American 4386–4392, Monsoon 2013. System, Geophys. 17481 cycling, J. Climate, 6, 1077–1089, 1993. Aguiar, R. G., andtropical de rain Araújo, forest A. evapotranspiration: are C.:ent?, the Atmospheric J. wet Geophys. versus and Res.-Biogeo., vegetation 115, seasonally controls G0402, dry doi: rain of forests Amazonian any di the MacKenzie, Mississippi, and Amazon River basins, J. Hydrometeorol., 7, 312–329, 2006. mixing. Part II: Precipitation recycling in17481 the Amazon basin, J. Climate, 19, 1473–1489, 2006. recycling models, J. Climate, 14, 2497–2508, 2001. Arraut, J. M. and Satyamurty, P.: Precipitation and water vapor transport in the Southern Hemi- References Eltahir, E. A. B. and Bras, R. L.: Precipitation recycling in the Amazon basin, Q. J. Roy. Meteor. Fagiolo, G.: Clustering in complex directed networks, Phys. Rev. E, 76, 026107, Dirmeyer, P. A., Brubaker, K. L., and DelSole, T.: Import and export of atmospheric water vapor Da Silva, R. R., Werth, D., and Avissar, R.:Dee, Regional D. and impacts Uppala, of S.: future Variational land-cover bias changes correction on in ERA-Interim, no. 575 in Technical Mem- Donges, J. F., Zou, Y., Marwan, N., and Kurths, J.:Donges, Complex J. F., Zou, networks Y., Marwan, in N., climate and Kurths, dynamics J.: The backbone of the climate network, Euro- Dee, D. P., Uppala, S. M., Simmons, A. J., Berrisford, P., Poli, P., Kobayashi, S., Andrae, U., Drumond, A., Marengo, J., Ambrizzi, T., Nieto, R., Moreira, L., and Gimeno, L.: The role of Ama- Cox, P. M., Betts, R. A., Collins, M., Harris, P. P., Huntingford, C., and Jones, C. D.: Amazo- Drumond, A., Nieto, R., Gimeno, L., and Ambrizzi, T.: A Lagrangian identification of major Arraut, J. M., Nobre, C., Barbosa, H. M., Obregon, G., and Marengo, J.: Aerial riversBagley, and J. lakes: E., Desai, A. R., Harding, K. J., Snyder, P. K., and Foley, J. A.: Drought and defor- Boers, N., Bookhagen, B., Marwan, N., Kurths, J., and Marengo, J.: Complex networks identify Costa, M. H., Biajoli, M. C., Sanches, L., Malhado, A. C. M., Hutyra, L. R., da Rocha, H. R., Betts, R., Cox, P., Collins, M., Harris, P., Huntingford, C., and Jones, C.: The role of ecosystem- Bosilovich, M. G. and Chern, J.-D.: Simulation of water sources and precipitationBrubaker, recycling K. for L., Entekhabi, D., and Eagleson, P. S.: Estimation of continental precipitation re- Burde, G. I., Gandush, C., and Bayarjargal, Y.: Bulk recycling models with incomplete vertical Burde, G. I. and Zangvil, A.: The estimation of regional precipitation recycling. Part I: Review of 5 5 30 10 20 15 25 15 20 30 10 25 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | , 17498 17499 , , 17485 17495 17495 , , 17492 17481 17487 17499 , 17481 17481 17482 17481 17481 , 17481 17481 , 2013. 17481 , 2012. 17498 , in press, 2014. , er, M.: Global resilience of tropical forest 17485 ff 17495 , H., Nikoli, R., and Savenije, H. H. G.: An- ff 17499 17495 ects of deforestation on spatiotemporal distribu- ff , 2009. 17481 17512 17511 ects of tropical deforestation on global hydroclimate: 17495 17495 ff 10.5194/hess-17-4133-2013 17486 10.5194/bg-9-733-2012 17481 , 2012. 10.1175/JCLI-D-14-00022.1 17486 10.1029/2007JD009580 , D. B.: The TRMM Multisatellite Precipitation Analysis (TMPA): quasi-global, ff 17484 17483 17493 17485 10.1029/2012JD017540 man, G. J., Adler, R. F., Bolvin, D. T., Gu, G., Nelkin, E. J., Bowman, K. P.,Hong, Y., Stocker, ff state-of-the-art, Revista Brasileira de Meteorologia, 21, 1–19, 2006. of the Andesvariability, as J. Climate, derived 17, from 2261–2280, the 2004. NCEP-NCAR reanalyses: Characteristics and temporal Budget, Clim. Dynam., 24, 11–22, 2005. Basin, J. Climate, doi: simple building blocks of complex networks, Science, 298, 824–827, 2002. 2005, J. Climate, 21, 495–516, 2008. fluence on seasonal soildoi: moisture and evapotranspiration, J. Geophys. Res., 117, D15114, monsoonal rainfall over South2012. Asia using complex networks, Clim. Dynam., 39, 971–987, De Oliveira, R., Camargo, H., Alves, L. M., and Brown, I. F.: The drought17499 of Amazonia in De Oliveira, R., Camargo,2005, J. H., Climate, Alves, 21, L. 495–516, 2008. M., and Brown,tions I. of precipitation F.: The in South drought America, of J. Climate, Amazonia 24, in 2147–2163, 2011. of sugar-cane expansion in Brazil, Nature Climate Change, 1, 105–109, 2011. tive variability over South America1877–1891, and 1999. the South Atlantic convergence zone, J. Climate, 12, Nature, 342, 411–413, 1989. amazon drought, Science, 331, 554–554, 2011. spiration model, J. Geophys. Res.-Atmos., 96, 13179–13188, 1991. transition in southwestern Amazonia, J. Climate, 24, 2368–2377, 2011. South America, Climanalise, 5, 36–45, 1990. 114, D02105, doi: Third International Workshop on Monsoons, World111–129, Meteorological 2–6 Organizations, November Hangzhou, 2004. torial waves on submonthly time scales2013. in reanalyses and TRMM, J. Climate, 26, 3013–3030, TRMM precipitation radar monthly rainfall estimates over Brazil, J. Geophys. Res.-Atmos., 1977. Earth Syst. Sci., 17, 4133–4142, doi: a multimodel ensemble analysis, J. Climate, 22, 1124–1141, 2009. alyzing precipitationsheds to understand thegeosciences, 9, vulnerability 733–746, of doi: rainfall dependent regions, Bio- E. F., and Wol multiyear, combined-sensor precipitation estimates at55, fine 2007. scales, J. Hydrometeorol., 8, 38– and savanna to critical transitions, Science, 334, 232–235, 2011. Marengo, J. A.: On the hydrological cycle of the Amazon basin: a historical review and current Marengo, J. A., Soares, W. R., Saulo, C., and Nicolini, M.: of the low-level jet east Marengo, J. A.: Characteristics and spatio-temporal variability of the Amazon River Basin Water Martinez, J. A., and Dominguez, F.: Sources of Atmospheric Moisture for the La Plata River Milo, R., Shen-Orr, S., Itzkovitz, S., Kashtan, N., Chklovskii, D., and Alon, U.: Network motifs: Marengo, J. A., Nobre, C. A., Tomasella, J., Oyama, M. D., Sampaio de Oliveira, G., Miguez-Macho, G. and Fan, Y.: The role of groundwater in the Amazon water cycle. 2. In- Malik, N., Bookhagen, B., Marwan, N., and Kurths, J.: Analysis of spatial and temporal extreme Marengo, J. A., Nobre, C. A., Tomasella, J., Oyama, M. D., Sampaio de Oliveira, G., Medvigy, D., Walko, R. L., and Avissar, R.: E Loarie, S. R., Lobell, D. B., Asner, G. P.,Mu, Q., and Field, C. B.: Direct impacts on local climate Liebman, B., Kiladis, G. N., Marengo, J. A., Ambrizzi, T., and Glick, J. D. : Submonthly convec- Lean, J. and Warrilow, D. A.: Simulation of the regional climatic impact of Amazon deforestation, Lewis, S. L., Brando, P. M., Phillips, O. L., van der Heijden, G. M., and Nepstad, D.: The 2010 Knox, R., Bisht, G., Wang, J., and Bras, R.: Precipitation variability over the forest-to-nonforest Figueroa, S. N. and Nobre, C. A.: Precipitation distribution over central and western tropical Gat, J. and Matsui, E.: Atmospheric water balance in the Amazon basin: an isotopic evapotran- Franchito, S. H., Rao, V. B., Vasques, A. C., Santo, C. M. E., and Conforte, J. C.: Validation of Goessling, H. F. and Reick, C. H.: Continental moisture recycling asGrimm, a A. Poisson M., process, Vera, Hydrol. C. S., and Mechoso, C. R.: The South American Monsoon System, in: The Kim, J.-E. and Alexander, M. J.: Tropical precipitation variability and convectively coupled equa- Hasler, N., Werth, D., and Avissar, R.: E Freeman, L. C.: A set of measures of centrality based on betweenness, Sociometry, 40, 35–41, Keys, P. W., van der Ent, R. J., Gordon, L. J., Ho Hirota, M., Holmgren, M., Van Nes, E. H., andHu Sche 5 5 15 20 10 30 25 30 10 15 25 20 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 17488 , 1979. , 2008. , 2009. 17485 17481 17507 , 2011. 17481 17482 17487 17485 , , 2007. , 2001. , 2003. 17485 17481 10.5194/hessd-10-6723- 17516 , 17496 10.1029/2007JD009566 17495 10.1029/2007WR006767 , 10.1029/WR015i005p01250 17489 , 10.5194/acp-11-1853-2011 10.1029/2007GL030612 , H., Rost, S., and Gerten, D.: Future water ff 17482 10.1029/2003GL018600 17482 , 17483 17514 17513 10.1103/PhysRevE.64.016132 17516 17481 , 17481 17481 17485 , 2010. 17484 , 17482 17481 17482 17499 , , 2013. , A.: Predicting evapotranspiration in tropical biomes using MODIS remote sensing data, 10.1029/2010WR009127 ff Should we use a simpletracking?, or Hydrol. complex model Earth for2013 Syst. moisture recycling Sci. and Discuss., atmospheric moisture 10, 6723–6764, doi: to global change,17481 Water Resour. Res., 45, W00A12, doi: development, J. Geophys. Res.-Atmos.,17485 113, D17106, doi: and fate ofdoi: atmospheric moisture over continents, Water Resour. Res., 46, W09525, availability for global food production: the potential of green water for increasing resilience climate, J. Climate, 21, 2990–3001, 2008. cycling, Atmos. Chem. Phys., 11, 1853–1863, doi: experiments using an atmospheric general1957–1972, circulation 1999. model, J. Geophys. Res.-Atmos., 104, America, Geophys. Res. Lett., 30, 2199, doi: Ph.D. thesis, Federal University of Rio Grande do Sul, Porto Alegre, 2011. change, J. Climate, 4, 957–988, 1991. 17481 surface observations of precipitation: techniqueForecast., and 25, validation 885–894, over South 2010. America, Weather isotopic study, Water Resour. Res., 15, 1250–1258, doi: Climate, 12, 1368–1381, 1999. tion and climate change in a coupled model simulation, J. Climate, 22, 5686–5697, 2009. Regional climate change overexpansion, eastern Geophys. Amazonia Res. caused Lett., by 34, L17709, pasture doi: and soybean cropland preceded by air passage over forests, Nature, 489, 282–285, 2012. spiration algorithm, Remote Sens. Environ., 115, 1781–1800, 2011. 1994. centrality, Phys. Rev. E, 64, 016132, doi: 17481 transpiration from our vocabulary, Hydrol. Process., 18, 1507–1511, 2004. 1322–1325, 1990. Francisco, California, 6–10 December 2002. Harding, D. J., and North,greenness P. R. during J.: the Amazon dry forests season, maintain Nature, consistent 506, canopy 221–224, structure 2014. and hotspots for conservation priorities, Nature, 403, 853–858, 2000. da Silva, E. D.,hydrological Stone, and T. carbon A., cycles Trumbore, S. of E., Amazonian and forests and Vieira, pastures, S.: Nature, The 372, role 666–669, of deep roots in the the Amazon and La Plata River basins, in: AGU Fall Meeting Abstracts, vol. 1, p. 0830, San van der Ent, R. J., Tuinenburg, O. A., Knoche, H.-R., Kunstmann, H., and Savenije, H. H. G.: Rockström, J., Falkenmark, M., Karlberg, L., Ho Rozante, J. R. and Cavalcanti, I. F. A.: Regional Eta model experiments: SALLJEX and MCS van der Ent, R. J., Savenije, H. H. G., Schaefli, B., and Steele-Dunne, S. C.: Origin van der Ent, R. J. and Savenije, H. H. G.: Length and time scales of atmospheric moisture re- Tsonis, A. A., Swanson, K. L., and Wang, G.: On the role of atmospheric teleconnections in Oyama, M. D. and Nobre, C. A.: A new climate-vegetation equilibrium state for tropical South Numaguti, A.: Origin and recycling processes of precipitating water over the Eurasian continent: Ruho Nobre, C. A., Sellers, P. J., and Shukla, J.: Amazonian deforestation and regional climate Rozante, J. R., Moreira, D. S., de Goncalves, L. G. G., and Vila, D. A.: Combining TRMM and Nobre, P., Malagutti, M., Urbano, D. F., de Almeida, R. A., and Giarolla, E.: Amazon deforesta- Salati, E., Dall’Olio, A., Matsui, E., and Gat, J. R.: Recycling of water in the Amazon basin: an Trenberth, K. E.: Atmospheric moisture recycling: role of advection and local evaporation, J. Spracklen, D. V., Arnold, S. R., and Taylor, C. M.: Observations of increased tropical rainfall Sampaio, G., Nobre, C., Costa, M. H., Satyamurty, P., Soares-Filho, B. S., and Cardoso, M.: Mu, Q., Zhao, M., and Running, S. W.: Improvements to a MODIS global terrestrial evapotran- Newman, M. E. J.: Scientific collaboration networks. II. Shortest paths, weighted networks, and Morton, D. C., Nagol, J., Carabajal, C. C., Rosette, J., Palace, M., Cook, B. D., Vermote, E. F., Nepstad, D. C., de Carvalho, C. R., Davidson, E. A., Jipp, P.H., Lefebvre, P.A., Negreiros, G. H., Savenije, H. H. G.: The importance of interceptionShukla, and J., why Nobre, we C., and should Sellers, delete P.: Amazon the deforestation term and evapo- climate change, Science, 247, Sudradjat, A., Brubaker, K., and Dirmeyer, P.: Precipitation source/sink connections between Myers, N., Mittermeier, R. A., Mittermeier, C. G., Da Fonseca, G. A. B., and Kent, J.: Monteith, J.: Evaporation and environment, Sym. Soc. Exp. Biol., 19, 205–234, 1965. 5 5 30 30 25 25 20 15 20 10 15 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | , 2 km 6 26 41 10.5194/esdd-5-203-2014 17481 17506 , 17501 , ects of Amazon deforestation, J. Geophys. ff 17492 17515 17516 ce (DAO); Integral Moisture Balance (IMB) model; GFDLIC from the modelDAODOE 1948–1997 reanalysis 27.2 during OND 1963–1973ERA-Interim reanalysis 35 TRMM and MODIS 1981–1993 31 1999–2008 1979–2003 28 10.8 for area 10 2001–2010 28 ffi 17481 tracers (general) Atmospheric Bulk model (Budyko model) tracking model Atmospheric Bulk model (IMB) back-trajectory method tracking model ) AGCM with water vapor ) Atmospheric moisture 2006 ) Atmospheric Bulk model ECMWF reanalysis 1985–1990 25 17481 , ) Atmospheric Bulk model GFDL and NCAR 1963–1973 24 ) Quasi-isentropic 2010 ) Atmospheric Bulk model 1994 , , 1993 2009 ) Atmospheric Bulk model CMAP and NCEP-NCAR reanalysis 1979–95 34 , , 2006 , Overview of regional precipitation recycling ratio in the Amazon basin as found in 1999 17484 17498 , sures for complex networks withture directed recycling, and Europhys. weighted Lett., edges revised, for 2014. studying continental mois- and transpiration inmosphere the Model, hydrological Earth cycle Syst.2014. – Dynam. Discuss., Part 5, 1: 203–279, doi: Simple Terrestrial EvaporationRes.-Atmos., to 107, At- 1322–1325, 2002. 10586, 2009. the Amazon basin: isotopic insights, Ambio, 20, 384–387, 1991. Bohrer, C.: Protecting the Amazon with protected areas, P.Natl. Acad. Sci. USA, 106, 10582– Paegle, J., Paegle, J., Penalba,Zipser, O., E.: Salio, The P., Saulo, South C., American2006. Silva low-level Dias, jet M. experiment, A., B. Silva Am. Dias, Meteorol. P., and Soc., 87, 63–77, of interception and transpiration inSyst. the Dynam. hydrological Discuss., cycle 5, 281–326, – 2014. Part 2: Moisture recycling, Earth Brubaker et al. Eltahir and Bras Trenberth Bosilovich and Chern Burde et al. Dirmeyer et al. van der Ent et al. ( Study( ( ( ( MethodZemp Dataset et al. (this study) Atmospheric moisture Period Precipitation recycling ratio ( ( Table 1. many studies. Abbreviations:(ECMW); the Geophysical European FluidCenter Centre Dynamics Merged for Laboratory Analysis of Medium-Range PrecipitationDecember Precipitation (GFDL); Weather (OND); (CMAP); Climate Forecasts Initinal DataNCEP conditions Prediction Assimilation – (IC); O Department October-November- ofby Energy the (DOE); National World Center Monthly for Surface Atmospheric Station Research Climatology (NCAR). distributed Werth, D. and Avissar, R.: The local andZemp, global D. C., e Wiedermann, M., Kurths, J., Rammig, A., and Donges, J. F.: Node-weighted mea- van der Ent, R. J., Gordon, L. J., and Savenije, H. H. G.: Contrasting roles of interception Walker, R., Moore, N. J., Arima, E., Perz, S., Simmons, C., Caldas, M., Vergara, D., and Victoria, R. L., Martinelli, L. A., Mortatti, J., and Richey, J.: Mechanisms of water recycling in Vera, C., Baez, J., Douglas, M., Emmanuel, C. B., Marengo, J., Meitin, J., Nicolini, M., Nogues- van der Ent, R. J., Wang-Erlandsson L., Keys P. W., and Savenije H. H. G.: Contrasting roles 5 20 15 10 i Ω (2) (3) that i (see also (see also i i as the frac- that is origi- i through direct mois- through direct mois- that is originating from Ω Ω through direct moisture i Ω averaged over all grid cells in through direct moisture recycling: Ω Ω ρ for local rainfall. High values indicate is the total precipitation in direct evapotranspiration recycling ra- i Ω P quantifies the dependency on direct mois- quantifies the local contribution to precip- the total evapotranspiration in Ω Ω i ρ ε E direct precipitation recycling ratio through direct moisture recycling. High values is the amount of evapotranspiration in , , is the amount of precipitation in Ω Ω Ω Ω Ω i i → ← → i P ← i E i i . We note that P E E P Ω = = as the fraction of evapotranspiration in each grid cell ,i ,i

Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | We also define the Ω Ω gives the regional recycling ratio,itation i.e, that the fraction is of regionallyBurde precip- et recycled al., 2006; (Eltahir van and der Ent Bras, and 1994; Savenije, 2011). ture recycling and Appendix A1). tio ture recycling: ε tion of precipitation in each grid cell ture recycling from locations which receive moisturefrom (direct precipitative sink) that contributes to precipitation over (Savenije, 2004). The second conditionwater is at fulfilled the if beginning the has soil tagged the fraction) same as the composition atmospheric (in moistureseason. term at of the end of the 2.3 Moisture recycling ratio In this section, we define measuresor to backward track moisture in forward spaceforward from tracking, a the measures given indicate region.cipitative the In sink) destination the (pre- of case evaporationof of from a a the backward tracking, region. theysource) In indicate the the of origin case precipitation (evaporative direct over moisture the recycling region. onlymoisture We (Sect. further first forward 2.3.1). consider or We backwardcading then moisture in recycling track space (Sect. through 2.3.2). Considered cas- the together, direct and cascading moisturefull picture recycling of ratios the sources provide and a of sinks of interest. moisture from a region 2.3.1 Direct moisture recycling ratios We define the evapotranspiration from where nating from evapotranspiration from otranspiration, particularly in warm climatesrooted and for plants shallow (Savenije, 2004). However,with in deep seasonal rooted trees, forests the moistureing which the is dry evaporated season dur- can be hold back for one or several months ρ where contributes to precipitation over recycling and Appendix A1). itation over 320 315 325 290 295 300 305 310 330 22 erent grid (1) ff precipitates m is the 24 m Ω E precipitates is the tagged 23 Ω to the total precipitation in 2 m m 12 is the total precipi- m precipitates in 3, in the case of a for- P 23 Ω m is locally recycled. 22 – – – 7 3 5 4 1 3 6 2 4 – – – 11 68 8 4 6 4 6 5 4 4 4 32 12 26 16 7 11 15 7 12 43 16 35 77 45 65 56 31 47 11 9 9 11 24 12 13 15 10 42 3523 41 25 30 24 35 26 32 30 30 28 29 18 31 21 20 17 14 17 8 11 10 10 9 10 m to the total precipitation in 2 wet dry year wet dry year wet dry year 12 17518 17517 m contributes together with 22 m , Ω in the case of a backward tracking and the ) splits up in three branches: m ) splits up in three branches: 2 m 2 Ω E is the tagged fraction of precipitation, E = Ω precipitates in 4 and which can be of any shape and scale (grid cell, is the transported moisture towards or from another P is the total evapotranspiration, Ω represent different grid cells and arrows indicate the E 24 Ω E m 4 E , m is locally recycled. 3 = ). , Scheme of the moisture recycling network. Nodes 1,2,3,4 represent di Fraction of evapotranspiration that contributes to precipitationthrough cascading over moisture recycling the La Plata basin Fraction of evapotranspiration that contributes to precipitationthrough direct over moisture recycling the La Plata basin Fraction of evapotranspiration that contributes to precipitation over the continent Fraction of evapotranspiration that isin involved cascading moisture recycling Fraction of precipitation originating from evap- otranspiration from the continent Fraction of precipitation originating from evap- otranspiration from the Amazon basin through direct moisture recycling Fraction of precipitation originating from evap- otranspiration from the Amazon basin through cascading moisture recycling Fraction of precipitation that comescading from moisture cas- recycling on the continent Fraction of precipitation that comescading from moisture cas- recycling inregion the intermediary 22 This assumption is valid under two conditions: (1), evapo- 2 2 Ω Results (in %) averaged for the La Plata basin, the Amazon basin and for the South m , P P P fraction of transported moisture towards orcell. from We another call grid “tagged fraction” theinating share of from the moisture orig- share of moisture precipitatingward over tracking. transpiration follows directly after the(2), precipitation the event fraction or ofand tagged the atmosphere moisture can be in assumed the to(i.e., be surface in temporally constant reservoir steady state)condition (Goessling is and usually fulfilled Reick, during interception 2013). andspiration The fast which tran- first are important components of the total evap- contributes together with basin, continent), we assumewithin that the the surface reservoir moisture and composition thecell atmosphere remains for the each same. grid This impliestagged that, fraction in of each precipitation grid is cell, linearlytagged the proportional fraction to of the evapotranspiration and theof tagged transported fraction moisture: where ( 2.2 Basic assumptions In order to track moistureregion forward or backward from a given tation, grid cell, Fig. 1: Scheme1 of the moisture recycling network. Nodes in 3, tagged fraction of evapotranspiration and direction and amounttransporation of in moisture the originating sourcetation from cell in evapo- the and target contributing cell.ration o For in example, precipi- 2 the ( total evapotranspi- 4 Zemp et al.: Cascading moisture recycling /P /E /P c c m E P P c Am casc Am casc Pl Pl c ). ε ε ∆ ε ∆ Notation Descriptionρ ρ ρ La Plata basin Amazon basin South America ∆ 2 P 275 280 285 265 270 ( in 4 and Figure 1. cells and arrows indicateration the in direction the and source amount cellevapotranspiration of and in contributing moisture 2 to originating precipitation ( from in evapotranspo- the target cell. For example, the total American continent during the wet season (DJFM), the dry season (JJAS) and all year round. Table 2. if many optimal path- B equal to 0 if none of these ) aims to highlight nodes in B B ) (Fagiolo, 2007; Zemp et al., 2014) for ˜ C others and can therefore facilitate, impede or if it fulfills two conditions: (1) it is the inter- ˜ C ness Centrality and the Middleman motif is equal to zero if the grid cell forms no Middleman between ˜ C To compute it, we first identify for each pair of grid cells Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion PaperIt is| worth to Discussion note that the Middleman motif considers Paper | We expect similar spatialtweenness patterns centrality in (Sect. the 2.5.2)cient results and associated of with the the the clustering Middlemanfact, coeffi- be- motifs both (Sect. measures 2.5.1). reveal In important intermediary grid cells ternative to the direct recyclingand of moisture sinks. between We sources compute theing weighted coefficient version ( of the cluster- the network withstand central position “tobias the the degree transmission of that messages” in1977, they the p. network 36). (Freeman, Here, we usewhere cascading it moisture to recycling reveal pathways intermediary are grid channeled. cells the moisture recycling pathwaysput, with called the ”optimal pathways”. greatest Theseany through- pathways can number include of re-evaporationway cycles. is usually As the the direct onecle), optimal we (without path- first any had re-evaporation cy- to modifypathways the involve cascading network moisture such recycling. that To the doremoved optimal so, from we the network all long-rangei.e, moisture occurring transport, over distances largergrees. than The 15 choice geographical of de- thissults threshold qualitatively. does Once not optimal influence pathways thefind are re- identified, intermediary we grid cellsAppendix that C2). A they grid cell have has inways a high pass common through (see it andpathways has pass a through it, i.e., moisturegrid never cell. cascades in the 2.5.3 Similarities and differences between the Between- motif at all. In our study,ing regions coefficient which than have a the larger 80regions”. cluster- percentile are called “Middleman three interconnected gridcading cells moisture which recycling pathways correspondsevaporation cycle. involving These to only pathways contribute cas- one usually most re- moisture to transport between two locations. Inof fact, moisture the transported amount in a pathwaythe typically number decreases with of re-evaporation cycles involvedOther in the motifs pathway. formed bymoisture recycling three exist or (Zemp et more al.,lyzed grid 2014), here. but cells are linked not ana- by 2.5.2 Betweenness Centrality The betweenness centrality ( each grid cell ashas described a in high the Appendixmediary C1. location A in a grid high cell (2) number of each Middleman arrow motifs and involvedput. in the motifs has a great through- in 23 495 470 475 480 485 490 455 460 465 450 Ω m /P casc Am casc Am c ρ ρ ) and P . with ∆ 2 in (b)). casc Am and and only (Fig. ). The grid cell 1 is ρ /P . 12 casc Ω in (b)) and 3 (b) Ω 12 ρ m . This implies Am Am in 32 ρ m ρ Ω 31 12 · and m (b) m m uses two alterna- 2 31 23 3 m m or is the fraction of total and (a) in (a) or 2 for moisture originating from the Amazon in in (a) or /P 13 (a) 23 ) (see Appendix B). m m ) is a measure from complex 13 c m 21 P C E m m ∆ 17520 17519 ∆ 21 m (dark gray) receives and distributes 1 considers cascading moisture recycling is an intermediary on an alternative pathway casc Am ρ 1 (a) being all grid cells covering the La Plata basin ) and outflow ( c Ω P is the fraction of total evapotranspiration that is in- ) and the cascading pathway ( for moisture that has final destination the La Plata basin. In both figures, the ∆ ) (see Appendix B). (b) . m ). While provides information regarding the importance of cas- with Scheme of two possible patterns in the Middleman motif from the perspective of 2 (b) /E in Scheme of cascading moisture recycling P The clustering coefficient ( In addition, we are interested in the importance of cascad- c /P 32 12 ∆ Considered together, the direct and cascading recycling ra- To study the cascading moisture recycling between the E casc P l m moisture inflow that comescling in from the cascading Middleman region. moistureof It recy- precipitation can be that seen comes asbeen from re-evaporated the in fraction the the Middleman continent region. anddependency It quantifies on that the cascading has moisture recyclingman in region for the local Middle- rainfall. 2.5 Complex network analysis We investigate important moisture recycling pathwaystwo using measures from complex network analysis:efficient clustering associated co- with Middleman motifs andcentrality. betweenness 2.5.1 Middleman motifs and clustering coefficient In complex network theory, motifsand are recurring defined as patterns significant ofnetwork interconnections (Milo that et occur al.,dleman in 2002). motif the Here, (Fagiolo, a 2007) if gridternative is pathway cell to an forms the intermediary direct a on transporttwo Mid- an of other moisture al- grid between cells (Fig. 3). network analysis which measures the tendencyticular to form motif a par- (Fagiolo, 2007).locations in Here, cascading moisture it recycling pathways, reveals as the intermediary al- The grid cell to the direct transport of moisture between that comes from theover the continent continent. and It quantifies thatlocation the further to contribution cascading of precipitates moisture a recycling. specific ing moisture recycling in2.5.1) for the the Middleman total moisture region in-approach and (see as outflow. We Sect. above. use We the forbid same man re-evaporation region in the of Middle- moisturewe originating estimate from the the resulting( continent reduction in and total moisture inflow the cascading pathway ( moisture from and toalso grid exchange moisture cells such 2 thatThe and there exchange 3 of is moisture (light no between cyclic gray) relation. which Fig. 3: Scheme ofmotif. two The possible grid patterns cell in the Middleman tive pathways: the direct one ( 6 Zemp et al.: Cascading moisture recycling m casc Ω casc P l ε · or ( cading moisture recycling for moisture thattion has the final La destina- Plata basin (see Fig. 2b). tios provide a fullmoisture from picture a given of region. the High values origin of and destination of pathways of cascading moisture recycling.only To track avoid moisture this, that we crosses thethat border of we consider re-evaporation cycles outside being all grid cellsε covering the Amazon basin ( for moisture originating from the Amazon basinε (see Fig. 2a), is the fractionmoisture of recycling precipitation on the that continent,orated comes i.e., at that from has least been cascading twicependency evap- on on cascading the moisture continent. recycling∆ for It local quantifies rainfall. the de- volved in cascading moisture recycling on the continent, i.e., Fig. 2: Scheme of cascading moistureture recycling originating (a) from for the mois- Amazonthat basin has and (b) final for destination moisture thethe La amount Plata of basin. precipitation Infrom in evapotranspiration both grid in figures, grid cell cel 3 1 that is is originating 2, see also Appendix A2). Amazon basin and the La Plata basin, we use indicate together thefrom precipitative the sink Amazon region basinhighlight of and evaporative source high moisture regions of values precipitationLa of over Plata the basin. 2.4 Quantifying cascading moisture recycling In this section, we are interesteding in moisture the recycling importance for of the cascad- tation) total and moisture outflow inflow (evapotranspiration). To (precipi- quantify this,forbid we re-evaporation of moisture originating fromnent and the we conti- estimate the resultinginflow reduction ( in total moisture 23 (a) Figure 3. grid cell 1. The2 grid and cell 3 1 (dark (lightexchange gray) of gray) receives moisture which and between also distributes 2 exchange moisture and moisture from 3 and such uses to that two grid alternative there cells pathways: is the no direct cyclic one ( relation. The an intermediary on an alternative pathway to the direct transport of moisture between 2 and 3. Figure 2. basin and amount of precipitation inm grid cell 3 that is originating from evapotranspiration in grid cell 1 is 430 435 440 425 445 420 390 395 410 415 385 400 405 c Ω . ρ ρ (5) (4) Ω that pro- i as the and P l ) and ε c

after one becomes c ε ε

Ω Ω is useful to that is orig- ) as defined ε c i ρ

Am

ρ

that is originating

through cascading

through cascading

being all grid cells i

Ω Ω

Ω through cascading mois- for local rainfall. High val- with ). While being all grid cells covering Ω through cascading moisture Ω Ω P l Ω ε ε Ω quantifies the dependency on cas- with ) and Ω ρ casc Ω Am quantifies the local contribution to pre- cascading evapotranspiration recycling ρ ρ . Ω through cascading moisture recycling. High casc Ω ε , Ω , cascading precipitation recycling ratio is the amount of evapotranspiration in is the amount of precipitation in Ω Ω i i Ω Ω → ← casc casc P i i E → ← casc i casc P i E is the entire South American continent, E P = as the fraction of evapotranspiration in each grid cell = may be counted several times as it is involved in several Ω ,i Ω ,i The moisture inflow (resp. outflow) that crosses the border To study the direct moisture recycling between the Ama- We also define the If The direct recycling ratios underestimate the strength of which contributes to precipitation in casc Ω casc Ω of cipitation over values indicate locations where moisture(indirect initially evaporative source) originated before it is distributed over ture recycling. where contributes to precipitation over moisture recycling as: ε where ratio i indicate respectively sources andture. sinks In of this continental study mois- weture neglect possible in contributions South of mois- Americathese contributions from to and the to(van overall der other moisture Ent budget continents, et are al., since 2010, small Table 2). zon basin and the Laaries in Plata Fig. basin 7), we (defined use bythe the Amazon red basin bound- ( covering the La Platatrack basin moisture from ( the Amazon basin as a whole, inating from evapotranspiration from moisture recycling. cading moisture recycling from ues indicate the final destination ofore moisture from more re-evaporation cycles (indirect precipitative sink) . the relationship, in term ofspecific moisture locations. recycling, between In two only fact, through moisture direct moisture can recycling,cading but be also moisture through recycling. exchanged cas- not 2.3.2 Cascading moisture recycling ratios We define the from evapotranspiration from recycling: ρ Zemp et al.: Cascading moisture recyclingindicate locations which distribute moisturetive (direct source) evapora- over the continental evapotranspiration recycling ratio ( the continental precipitation recyclingin ratios van ( der Ent et al. (2010). Considered together, vides information regarding thecontribution spatial inside heterogeneity the of Amazon basin. the 5 fraction of precipitation in each grid cell 380 375 370 340 345 365 355 360 335 350 ˜ C B B indicate interme- (b) (d) ˜ C Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | ) of /P 3 c P Dry season (JJAS) ∆ Wet season (DJFM) , precip- indicate regions where pathways of cascad- ˜ ˜ C C B (b,d). While high values of (a) (c) (b, g) c B c ε ε (j) (e) and the hatches ) (e,j) and continental c Fig. 6: Complex networkassociated analysis. with Clustering the coefficient motif MiddlemanCentrality (a,c) and Betweenness diary locations where cascadingfor alternative moisture pathways recycling to allows high the values direct of transport ofing moisture, moisture recycling arethe channeled. dry Results season (upper are row) given and for the wet season (lower row). (Table 42). These results implyare important that intermediary the regions involved Middleman in the regions pathways alternative to the directzon transport basin of to moisture the from La Plata the basin. Ama- 3.3.2 Betweenness Centrality Cascading moisture recycling pathways aresouth-western channeled in part the of theeast Amazon of basin the subtropical and Andes a (Figs. narrow 6b and band 6d). Regions with ρ month · 2 640 645 m ( in- ) associated with the Middleman c c /P ˜ C ρ ρ c indicate P (i) (d) ∆ /E , evapotranspiration /E /E c and continental evapotranspiration c c indicate intermediary regions which ofmoisture/ E E E 3 ∆ m ∆ ∆ /E (a, f) (b) (d) c (d, i) E c ) (a,c) and fraction of to- ∆ ρ cascading moisture recycling might be reducedtation since in precipi- this region is insuredand by orographic Nobre, lifting 1990). (Figueroa 3.3 Complex network analysis 3.3.1 Middleman motif Cascading moisture recyclingSouth plays American an moisture important transportclustering (see role coefficient Fig. for ( 5).motif, Using we are the able to identifyin intermediary locations cascading involved moisture recyclingthe direct as transport alternative of pathwaysdleman moisture. to regions) These are regions the central (called andzon Mid- western basin part during of the the Ama- dry season and the south-western part /P c P 615 620 625 ∆ Dry season Wet season 17522 17521 indicate regions which are dependent on cascading (c) Evap. - Precip. (h) Evap. - Precip. Dry season (JJAS) Wet season (DJFM) /P ) (b,d). While high values of c /P /P P c c /E P P c ∆ for the period 2001–2010 as calculated from TRMM and ∆ ∆ E ∆ (c) (a) (d,i) showing respectively sinks and sources of continental moisture. Here and in the c ε (g) Evap. (b) Evap. (c, h) In the La Plata basin, the fraction of precipitation that Fig. 5: Fraction ofcading total moisture recycling precipitation ( originatingtal from evapotranspiration cas- that is involvedrecycling in ( cascading moisture Zemp et al.: Cascading moisture recycling 9 dicate regions which arerecycling for dependent local of rainfall,regions cascading high which moisture values contribute to of cascading moisture recycling. of the Amazon basin duringagreement the with wet other season. measures This quantifyingof is the evapotranspiration in contribution good within(Figs. cascading 5b and moisture 5d). recycling comes from cascading moisture recyclingregion in is the 8% Middleman duringseason. the This wet represents season 47% andthat of 5% comes the during from fraction the cascading ofnent dry precipitation moisture during recycling the in wet the season conti- and 35% during the dry season ) and the hatches represent grid cells where annual mean evapo- 1 − showing respectively sinks and sources of continental moisture. Here 630 635 (e, j) month c × ε . While high values of 2 (f) Precip. (a) Precip. − Long term seasonal mean of precipitation m Fraction of total precipitation originating from cascading moisture recycling ( ¨ om et al., 2009; KeysThe et al., share 2012). of cascading moisture on total moisture inflow (b, d) × evapotranspiration recycling ratio following figures, the vectors indicate the horizontal moisture flux field in represent grid cells where annual meanseason evapotranspiration (JJAS) exceeds (upper mean row) annual and precipitation. the Results wet are season given (DJFM) during (lower the row). dry portant impacts would be observed inwestern particular part in of the south- theprobability Amazon basin to which experience has asavanna already critical (Hirota a et transition high al., from 2011) forestis and in to dependent the of La incoming Platastr rainfall basin for which the agriculture (Rock- reaches up to 50%(Figs. in 5a the and eastern 5c),hotspots side on of one Earth the of (Myers central theobservation et Andes should most be al., considered vulnerable 2000). with However, caution biodiversity balance this due of to latter the the im- wateran cycle over-estimation of in the this regional area recyclingan process which over-estimation and might of thus the lead importancerecycling. to In of addition, cascading the impact moisture of an upwind weakening of Fig. 4: Long term seasonalperiod mean 2001-2010 of precipitation as (a,f), calculated evapotranspiration (b,g), from precipitation TRMM - and evapotranspiration (c,h) MODIS. for Continental the precipitation recycling ratio 8 Zemp et al.: Cascading moisture recycling ) and fraction of total evapotranspiration that is involved in cascading moisture recycling 600 610 605 /E c E ∆ Figure 5. Figure 4. itation – evapotranspiration MODIS. Continental precipitation recycling ratio moisture and in the following figures, the vectors indicate the horizontal moisture flux field (in m recycling ratio transpiration exceeds mean annual(upper precipitation. row) Results and are the given wet for season the (DJFM) (lower dry row). season (JJAS) moisture recycling for local rainfall, high values of contribute to cascading moisture recycling. ( (a, c) Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | b ). , h casc Pl and ε d (A1) . The , j casc Am ρ casc Am casc Am ρ ρ indicate intermedi- that comes directly , c,g) and cascading j (d) (h) ˜ C Am ρ , a,e) and cascading moisture is the precipitation in j P l ε ˜ P C associated with the motif Mid- and j ˜ C , ij B B m Am Am j Ω ρ ρ is the amount of moisture which evaporates in P indicate interme- ∈ (b) (d) cient i ij ˜ (c) (g) C ffi is calculated as: P m Ω = . While high values of ,j ) and cascading moisture recycling ( show source regions of precipitation in the La Plata basin and precipitates in Ω e Appendix A Moisture recycling ratios A1 Direct moisture recycling ratio The fraction of precipitation in gridfrom cell ρ where i casc P l ε 760 765 (b, d) and and 17524 17523 a B P l , Dry season (JJAS) ) and cascading moisture recycling ( ε Dry season (JJAS) Wet season (DJFM) Wet season (DJFM) show sink regions of evapotranspiration from the La Plata basin (defined g Pl indicate regions where pathways of cascading moisture ε indicate regions where pathways of cascad- case P l ˜ ˜ casc Am casc P l C C ε ε ρ B B (b,d). While high values of and (c) (a) (f) (b) show source regions of precipitation over the La Plata basin B and c , Am casc Pl ρ Am ε ρ Fig. 6: Complex network analysis. Clustering coefficient associated with the motif MiddlemanCentrality (a,c) and Betweenness diary locations where cascadingfor alternative moisture pathways recycling to allows high the values direct of transport of moisture, ing moisture recycling arethe channeled. dry Results season (upper are row) given and for the wet season (lower row). (Table 42). These results implyare important that intermediary the regions involved Middleman in the regions pathways alternative to the directzon transport basin of to moisture the from La Plata the basin. Ama- 3.3.2 Betweenness Centrality Cascading moisture recycling pathways aresouth-western channeled in part the of theeast Amazon of basin the subtropical and Andes a (Figs. narrow 6b and band 6d). Regions with and 640 645 , d,h). Considered together, Pl ε in- casc Am ρ /P show sink regions of evapotranspiration from the La Plata basin. Results are P l P l c indicate ε ε , b,f) and fraction of precipitation which comes from the Amazon basin through direct ( P (a) (e) ∆ /E /E /E casc Am c casc P l and betweenness centrality c c ε E ρ E E ∆ ∆ ∆ Fraction of evapotranspiration which precipitates over the La Plata basin (defined Complex network analysis. Clustering coe (d) (b) and (a, c) moisture recycling ( (defined by the red boundaries)by and the red boundaries). Results are given for the dry season (JJAS) (upper row) andgionally the and wet downwind. season In addition, (DJFM) we (lowerern showed row). that flank the of east- the subtropical Andesthe - South located American in the Low pathway Levelin of Jet the - continental moisture plays recycling an as importantcading it channels role pathways. many This cas- study offersour new understanding methods of to vegetation and improve atmospherethe feedback water on cycle neededchange. in a context of land use and climate Zemp et al.: Cascading moisture recycling 11 Fig. 7: Fraction of evapotranspirationrecycling which ( precipitates in the La Plata basin through direct ( ) (a,c) and fraction of to- Am ) and fraction of precipitation which comes from the Amazon (defined by the red bound- f ρ /P 755 c P ∆ Figure 7. by the red boundaries) through direct ( given for the dry season (JJAS) (upper row) and the wet season (DJFM) (lower row). and aries) basin through directConsidered ( together, and dleman ary locations where cascadingtransport moisture of recycling moisture, allows high for values alternative of pathways to the direct recycling are channeled. Results are(lower given row). for the dry season (upper row) and the wet season Figure 6. Dry season (JJAS) Wet season (DJFM) ) (b,d). While high values of /P /P c c /E P P c ∆ ∆ E ∆ (c) (a) In the La Plata basin, the fraction of precipitation that of the Amazon basin duringagreement the with wet other season. measures This quantifyingof is the evapotranspiration in contribution good within(Figs. cascading 5b and moisture 5d). recycling comes from cascading moisture recyclingregion in is the 8% Middleman duringseason. the This wet represents season 47% andthat of 5% comes the during from fraction the cascading ofnent dry precipitation moisture during recycling the in wet the season conti- and 35% during the dry season regions which contribute to cascading moisture recycling. Fig. 5: Fraction ofcading total moisture recycling precipitation ( originating from cas- dicate regions which arerecycling for dependent local of rainfall, cascading high moisture values of Zemp et al.: Cascading moisture recycling 9 tal evapotranspiration that is involvedrecycling in ( cascading moisture 630 635 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 4. = = /P 14 4 4 , , m 3 3 · , , Originally, the 3 2 1 , (a) 1 C /P C 13 ) and can contribute is m · is 3 2 ocean 3 ← /P i E 12 ( m i and = 4 and the contribution of the path 2 /P 14 1,2,3,4 W m · 3 4, the contribution of the path involving one , the contribution of the path /P . The contribution of the di- 4 /P . 4 13 4 14 m ) is evaporated in 17526 17525 m /P = = /P 14 ) and some part of it contributes to precipitation in the ocean i 1,4 1,3,4 14 E ← m i ( W W P i m = and the contribution of the path involv- ∗ ) is composed by oceanic and continental moisture. The total 4 i This paper was developed within the scope of , ). By doing so, we remove cascading recycling of continental 3 4 P ( 1 to grid cell i C 1 ocean /P /P ← ij 13 14 m ( m m If we forbid the re-evaporation of continental precipitation, only the precip- j ∗ ∗ (b) 2 3 erent cascading moisture recycling pathways from grid cell 1 to grid cell 4. The ff ). ij Di Scheme explaining the removal of cascading moisture recycling. /P /P that has oceanic origin ( m ( i Transport in theSouth Southern American Hemisphere Region., with J.1912, Emphasis 2009. Appl. on Meteorol. Clim., the 48, 1902– Marengo, J.: Aerial RiversMoisture and Transport and Lakes: Its Relation Looking to Amazoniaical at and Rainfall to Large-Scale in Subtrop- South America, J. Climate, 25, 543–556, 2012. J. A.: Drought andenced Deforestation: recent Has precipitation land extremes cover in27, change the 345–361, Amazon?, influ- 2014. J. Climate, Jones, C.: The role of ecosystem-atmosphere interactions in sim- j 12 13 Appendix D Author contribution J.F.D., H.M.J.B., C.-F.S. and D.C.Z.R.J.E. developed the performed analysis. the simulation ofvided WAM-2layers. the G.S. mask pro- ofanalysis the and La prepared the Plata manuscript basin. with contributionsall D.C.Z. from performed co-authors. the C.-F. S.J.H. conceived and the supervised it project together together with A.R. with Acknowledgements. Arraut, J. M., Nobre, C., Barbosa, H. M., Obregon,Bagley, J. G., E., Desai, and A. R., Harding, K. J., Snyder, P. K., and Foley, Betts, R., Cox, P., Collins, M., Harris, P., Huntingford, C., and the IRTG 1740 / TRP 2011/50151-0,J.F. funded Donges by acknowledges the DFG funding / fromand FAPESP. the BMBF Stordalen (project Foundation GLUES)ALW. We thank and K. R.J. Thonicke van forP. Manceaux comments der for on his Ent the help from manuscript on and designing NWO the / network schemes. References Arraut, J. M. and Satyamurty, P.: Precipitation and Water Vapor m ing re-evaporation cycles in grid cells m involving one re-evaporation cycle in grid cell rect pathway is Fig. C1: Differentfrom cascading grid moisture cell recycling pathways contribution of the direct pathway is Figure C1. re-evaporation cycle in grid cell 3 is involving re-evaporation cycles in grid cells 2 and 3 is moisture from the network. to precipitation in Figure B1. precipitation in the grid cell incoming moisture is evaporated in grid cell itation in 955 960 965 970 975 980 ) j = n j t (C5) (C4) P +1 l m t to grid l t i w ) log( j − n j

is the fraction t ) P

i m

+1

l ·

+1

t

l l

)

the number of grid t t

P j j

+1

m

n

N

n

j

l

+1

in the package iGraph

t

t

t

l

l P

t t w

m reaches values between 0 P

· log( 1

+ m with

1 B ( 1 − . 2 =1 +1 ) +1 i l l X n l / − +1 t t =1 l ,j l n l l t t t − n is the number of these pathways P ) as defined above. Q : w m i + 2) · ) 1 1 1 1 i 1 ,...,t C t it 1 ( − =1 1 − N t 1 =1 l X n P t l Y 1 n , it corresponds to the pathway with the 3 m in the package iGraph for Python based jk i,t 0 betweenness · P σ + m C − C 1 1 1 2 t t it log( 1 P . The cost of a pathway from grid cell ) N m i − w ) ( , and paths jk k = log( = log( = = = +1 = ( jk σ l +1 t l σ is the number of optimal pathways between grid  ,j ,j l t t 1 n n P and − 2 jk m j,k as calculated in iGraph becomes: X N j σ ,...,t ,...,t j through cascading moisture recycling: 1 1 = i To find the optimal pathway, we use the method log( 0 i i,t i,t for Python. The choice ofexplained the in weights Sect. used C2.1. in this method is cells. To calculate it, wesion used the of directed the and weighted method ver- that pass through the gridand cell with cells Because the optimal pathway isthe defined minimum as cost the pathway with maximum contribution C2.2 Betweenness centrality Mathematically, betweenness of the grid cell of the number of optimalcells pathways which between pass any through pair of grid B on anmethod, algorithm the proposed cost ofthe by a weight Newman pathway of isour (2001). its calculated purpose, arrows. we as In chose In the the this order weight sum of to of the arrows adapt as the method to cell C 14in C − Zemp et al.: Cascading moisture recycling An example of pathway contributions isThe provided contribution in of Fig. each C1. existingany path pair is of calculated grid between cellspath in with the the network. maximum The contribution. optimal path is the shortest 950 940 945 935 930 920 925