Many studies have suggested an important role of the land surface on atmospheric conditions at a broad range of time scales [Rowntree and Bolton, 1983; Nicholson, 2000; Koster et al., 2004; Seneviratne et al., 2010; Gimeno et al., 2012]. Land-atmosphere feedbacks are particularly important for the intraseasonal variability of precipitation [Beljaars et al., 1996] and extreme events such as droughts or floods [Dirmeyer and Brubaker, 1999; Pal and Eltahir, 2001; Seneviratne et al., 2006; Zaitchik et al., 2006; Fischer et al., 2007]. However, climate models exhibit a large dispersion in the simulation of land-atmosphere feedbacks [Koster et al., 2002, 2004; Lawrence and Slingo, 2005; Koster et al., 2006; Guo et al., 2006; Wei and Dirmeyer, 2010]. It is however difficult to discriminate which model represents land-atmosphere feedbacks in the most realistic way. The motivation underlying this paper is to find observational constraint for land-atmosphere feedbacks. These feedbacks are complex in the sense that precipitation might be either enhanced or decreased when the soil is wetter depending on conditions [Findell and Eltahir, 2003a, 2003b; Ferguson and Wood, 2011].
 Land-atmosphere feedbacks can be either local (sometimes called “direct”) or regional (sometimes called “indirect”).
 Local feedbacks involve the effect of surface fluxes on the local atmospheric conditions. Positive or negative effects of soil moisture on subsequent precipitation are possible depending on large-scale atmospheric conditions [Betts, 1992; de Ridder, 1997; Findell and Eltahir, 2003a, 2003b; Ek and Holtslag, 2004; Santanello et al., 2009, 2011; Tuinenburg et al., 2011; Ferguson and Wood, 2011; Ferguson et al., 2012], on the spatial scale of soil moisture anomalies [Taylor et al., 2011], on the type of convective system [Taylor et al., 2009] or on whether the variable of interest is precipitation intensity or frequency [d'Odorico and Porporato, 2004]. On the one hand, if the soil is wetter, evapo-transpiration increases, which moistens the boundary layer, lowers the condensation level and favors convection [e.g., Betts, 1992; Taylor and Lebel, 1998; Santanello et al., 2009; Lintner et al., 2012]. On the other hand, if the soil is drier, then latent heat fluxes decrease at the expense of sensible heat flux, which warms the boundary layer, leads to more vigorous thermals and higher boundary layer top, and favors convection triggering [e.g., Porporato, 2009; Santanello et al., 2009; Westra et al., 2012; Taylor et al., 2012]. In models, the relative importance of these two effects may additionally depend on the model physics and resolution [Hohenegger et al., 2009]. Local feedbacks can also involve small-scale soil moisture gradients and associated mesoscale circulations [Taylor et al., 2007, 2009, 2011] and radiative effects of clouds [Schär et al., 1999; Betts, 2004; Schlemmer et al., 2011, 2012].
 Regional feedbacks involve the effect of surface fluxes on remote atmospheric conditions and on large-scale circulation. Positive and negative effects are possible depending on conditions. On the one hand, if the soil is wetter, then the evapo-transpiration increases. This moistens the atmosphere and favors convection downstream air mass trajectories [Eltahir and Bras, 1994]. For example, this can contribute to lower precipitation downstream of deforested areas [Spracklen et al., 2012]. This can also contribute to the persistence of droughts [Rodriguez-Iturbe et al., 1991a, 1991b; Entekhabi et al., 1992]. On the other hand, if the soil is drier, the surface temperature increases which favors large-scale convergence of tropospheric humidity, and thus favors convection [Kleidon and Heimann, 2000; Cook et al., 2006; Goessling and Reick, 2011]. For example, this may contribute to lower precipitation over irrigated regions [Lee et al., 2009a; Saeed et al., 2009; Guimberteau et al., 2012]. Storms might also become more intense despite a less frequent triggering [Lee et al., 2012]. In the case of a positive feedback, continental recycling increases; in the case of a negative feedback, it decreases.
 This paper focuses on regional-scale feedbacks. Such feedbacks have been less studied than local feedbacks because it is often thought that the effect of the direct input of water vapor through continental recycling is small, due to the long residence time scale (10 days) of water vapor in the atmosphere [McDonald, 1962]. The importance of continental recycling however strongly depends on the region considered [Koster et al., 1986], on the spatial scale considered [Budyko, 1974; Burde et al., 1996; Trenberth, 1999] and on the methodology used to quantify continental recycling [Eltahir and Bras, 1996]. Progress have been made to more robustly quantify the sources and sinks of precipitation and of water vapor (review by Gimeno et al. ) and underline the importance of continental recycling in some continental regions [Dominguez et al., 2006; Gimeno et al., 2010; van der Ent et al., 2010]. Quantifying the role of continental recycling on precipitation usually involves regional atmospheric water budgets based on reanalyses, on a combination of reanalyses and observations [Eltahir and Bras, 1996; Gong and Eltahir, 1996; Schär et al., 1999; Bosilovich and Schubert, 2002; Dirmeyer and Brubaker, 2007; Dominguez and Kumar, 2008; Dirmeyer et al., 2008, 2009] or on models [Brubaker et al., 1994]. There are, however, two drawbacks to this approach. First, it is difficult for such moisture budgets to accurately take into account the effect of mixing and of subgrid scale water vapor transport, sources, and sinks. Second, reanalyses are model products and their derived budgets are difficult to evaluate observationally. In this paper, to circumvent the first drawback, we use a water tagging approach [e.g., Joussaume et al., 1984; Koster et al., 1986; Numaguti et al., 1999; Yoshimura et al., 2004; Risi et al., 2010b], which accurately tracks the water vapor through each transport, mixing, and phase change process online in a global model. To deal with the second issue, we explore the possibility of using water isotopic measurements.
 The water molecule has several isotopologues. The most common isotopologue is H216O (hereafter called H2O), but heavier isotopologues are also found: HD16O (hereafter call HDO, with D standing for deuterium) and H218O. The water vapor isotopic composition (e.g., the concentration in HDO) is sensitive to the evaporative origin. For example, in the tropics, water evaporated from land surface is more enriched in heavy isotopes than water evaporated from the ocean [Gat, 1996]. Several studies have tried to exploit this property to infer continental recycling or to partition it into evaporation and transpiration, using isotopic measurements in the precipitation [Salati et al., 1979; Gat and Matsui, 1991]. However, precipitation is strongly affected by postcondensational processes [Stewart, 1975; Lee and Fung, 2008; Risi et al., 2010a]. The isotopic composition of water vapor more directly reflects the moisture origin. The development of water vapor isotopic measurements from satellite now offers a unique opportunity to exploit the water isotopic composition as an indication for continental recycling [Risi et al., 2010b].
 The goal of our paper is thus to explore the possibility to use water isotopic measurements from satellites to observationally constrain the role of continental recycling on the intraseasonal variability of precipitation. More specifically, given the close relationship between precipitation and precipitable water (W) in the tropics [Raymond, 2000; Bretherton et al., 2004], we will focus on evaluating the role of continental recycling on the intraseasonal variability of W in the tropics.
1.2. Overview of the Methodology
 To achieve this goal, we use an atmospheric general circulation model (GCM) coupled to a land surface model. Our strategy has three steps. First, we develop a diagnostic for the role of continental recycling variability on the intraseasonal variability of W. This diagnostic is called D1 and is calculated from a GCM simulation in which water vapor from different origins is tagged. We quantify and discuss the role of continental recycling in this simulation. Two sensitivity tests to the land surface physics are also presented in which the role of continental recycling is either larger or weaker than in the control, as quantified by D1.
 Second, we try to find an observation-based proxy for D1 to identify the simulation which has the most realistic role of continental recycling. While D1 is not directly observable, our hypothesis is that water vapor isotopic composition measurements may track continental recycling and its intraseasonal variations. In addition to water tagging, our GCM is also equipped with water isotopic diagnostics. We focus on the ratio of water vapor, expressed in ‰ as anomalies relatively to the ocean surface: , where SMOW is the standard mean ocean water [Dansgaard, 1964]. We show a good relationship between continental recycling and lower tropospheric δD at the intraseasonal time scales over several regions. Based on those results, and on the availability of isotopic observations, we propose an isotope-based, observable proxy for D1, named D1_iso.
 Third, we compare simulated D1_iso with data, to assess to what extent data can help identify the most realistic simulation in terms of D1. To do so, we use two satellite data sets which are sensitive to the isotopic composition of the boundary layer water vapor and which have a good spatio-temporal coverage: the GOSAT (Greenhouse gases Observing SATellite) satellite data set [Frankenberg et al., 2012] and the new version of the TES (Tropospheric Emission Spectrometer) retrievals [Worden et al., 2012a].
 The paper is organized according to these three steps. After a description of the model simulations, of the data sets and of the methodology (section 2), we quantify the role of continental recycling on intraseasonal variability of W in our simulations (section 3). We show the link between continental recycling and water vapor isotopic composition in section 4, and discuss a possible isotopic-based observable constraint to our model simulations in section 5. We conclude in section 6.