Soil moisture can influence precipitation through a feedback loop with land surface evapotranspiration. A series of numerical simulations, including soil moisture sensitivity experiments, have been performed for the Indian summer monsoon season (ISM). The simulations were carried out with the nonhydrostatic regional climate model Consortium for Small-Scale Modeling (COSMO) in climate mode (COSMO-CLM), driven by lateral boundary conditions derived from the ECMWF Interim reanalysis (ERA-Interim). Positive as well as negative feedback processes through local and remote effects are shown to be important. The regional moisture budget studies have exposed that changes in precipitable water and changes in precipitation efficiency vary in importance, in time, and in space in the simulations for India. Overall, the results show that the premonsoonal soil moisture has a significant influence on the monsoonal precipitation, and thus confirmed that modeling of soil moisture is essential for reliable simulation and forecasting of the ISM.
The earth’s surface plays a key role in weather and climate because of the large exchange of energy and water with the overlying atmosphere (Zhang et al. 2004). There are two important questions: how is the available radiative energy at the surface partitioned into latent and sensible heat fluxes, and how is water reaching the land surface distributed into soil water storage, runoff, or recycled into the atmosphere by evapotranspiration? Soil moisture, orography, and vegetation cover are some of the basic variables that affect the hydrological cycle.
Recently, Jones and Brunsell (2009) have shown that energy balance partitioning is strongly influenced by soil moisture and is important to the atmospheric boundary layer depth. An extended study of Pal and Eltahir (2001) revealed that anomalously high soil moisture tends to increase the convective rainfall via enhancing the latent heat flux at the surface and increasing the moist static energy. These examples emphasize the importance of the land surface and related processes in the earth’s climate system.
Initial disturbances in soil moisture can influence the climate system through different feedback processes. In general, processes that amplify an initial perturbation are called positive feedback processes, and damping processes are called negative feedback processes. In the early 1980s, Shukla and Mintz (1982) found an increase in precipitation amounts in a dry soil simulation experiment over Southeast Asia and the Indian region. They argued that a dry soil condition strengthens the transport of water vapor from the ocean by increasing the ocean–land temperature contrast, which produces more rainfall. This is an example of a negative soil moisture–precipitation (S–P) feedback process. Later on, Barnett et al. (1989) speculated that the negative S–P feedback is actively involved in the Indian summer monsoon season.
But, the S–P feedback processes are complex and act in many different paths. An often-investigated process is called precipitation recycling (Eltahir and Bras 1996; Bosilovich and Chern 2006; Bisselink and Dolman 2009), which involves both positive and negative S–P feedbacks. In precipitation recycling, a “recycling ratio” quantifies the amount of precipitation that has evapotranspirated locally. The recycling ratio is a useful diagnostic measure of the interactions between the land surface hydrology and the regional climate (Marengo 2006).
Schär et al. (1999, hereafter SCH99) categorized the S–P feedback processes into direct and indirect processes. Therein, the direct processes refer to processes changing the locally available precipitable water through evapotranspiration or moisture advection, and the indirect processes refer to processes affecting the efficiency of transforming available atmospheric moisture into precipitation. Thus, the precipitation recycling is a combination of direct and indirect feedback processes since the resulting precipitation amount depends on evapotranspiration and on precipitation efficiency. Numerical experiments with a regional climate model revealed that over Europe, the indirect processes are more important than the direct processes (SCH99).
Nowadays, climate models are useful tools in order to study the land–atmosphere interactions (Hirschi et al. 2011) since regional studies are flawed by the availability of necessary observational data. Especially, regional climate models with advanced physical parameterizations and adequate representation of local features such as the orography, land use, and lakes can provide help in understanding the regionally relevant S–P feedback processes (SCH99; Wei and Fu 1998; Chang et al. 2009).
Previous studies have collected comprehensive suites of S–P feedback processes globally (Koster et al. 2002; Orlowsky and Seneviratne 2010; Wei et al. 2008) as well as regionally (Raman et al. 1998; Chang et al. 2009; Fischer et al. 2007). For example, Koster et al. (2002) compared four atmospheric general circulation models and concluded that the coupling strength between the land surface and the atmosphere are significantly varying in the model simulations. Very recently, based on statistical analysis of multimodel ensembles, Orlowsky and Seneviratne (2010) found a remarkable coupling strength between soil moisture and precipitation for specific regions. They further compared the influence of soil moisture and sea surface temperature (SST), with the result that soil moisture often gives a stronger feedback on summer precipitation than SST. Raman et al. (1998) performed sensitivity simulations for soil moisture for very short time periods (48 h) over the Indian subcontinent. They found that the wet soil condition intensifies the large-scale circulation, which further enhances convective activity and precipitation. Quantitatively, however, our understanding of the S–P feedback processes is still incomplete and uncertain. The current work presents a regionally detailed study on the direct and the indirect soil moisture–monsoonal precipitation feedback processes over the Indian subcontinent.
a. The climate model and data
The simulations are performed with the nonhydrostatic limited-area climate model Consortium for Small-Scale Modeling (COSMO) model in climate mode (COSMO-CLM: Steppeler et al. 2003; Rockel et al. 2008; Dobler and Ahrens 2008; Kothe et al. 2011), version 4.8_clm10. It includes the multilayer soil model TERRA–ML (Schrodin and Heise 2002) and a Kessler-type (Kessler 1969) microphysics scheme including ice-phase processes for cloud water, rain, and snow. The Tiedtke (1989) mass flux approach with equilibrium closure based on moisture convergence is used as convection scheme. At the lateral boundaries of the model domain a relaxation scheme is applied with a buffer zone of eight grid boxes (Davies 1976). More details about the model are given on the community website (www.clm-community.eu).
The initial and lateral boundary conditions are taken from the European Centre for Medium-Range Weather Forecast (ECMWF) Interim reanalysis (ERA-Interim). ERA-Interim is a global reanalysis data product of ECMWF, with a horizontal spectral resolution of T255 with 60 levels in the vertical atmosphere, incorporating surface, upper air, and satellite observations. The data have a temporal resolution of 6 h covering the period from 1989 until near–real time (Simmons et al. 2006, 2007). In the present study, the COSMO-CLM (CCLM) horizontal resolution is set to 0.25° with 32 vertical layers. The reduction of vertical resolution mainly concerns the layers above the atmospheric boundary layer height, which is motivated by the fact that the majority of moisture is concentrated within the lower atmosphere. Hence, this approach mitigates the computational effort while preserving the driving field’s lower vertical structures. The numerical domain encompasses the entire Indian region (see Fig. 1).
b. Simulations’ design
To assess the influence of soil moisture, we have performed several simulation experiments: a reference simulation (CTL) for the period 1989–2008 and perturbed simulations. These latter simulations are initialized each year on the second of April with the CTL state, but with perturbed soil moisture: the WET run is initialized with a soil two times wetter than in the CTL simulation, and the DRY run is initialized two times drier (perturbations are limited by the field capacity and the wilting point of the soil type). The perturbed experiments are driven by the same lateral boundary condition as CTL. Therefore, these experiments explain the soil water initialization impacts on the model simulations at the regional scale as well as at the Indian summer monsoon scale. This setup is motivated by SCH99 and Pielke et al. (1999).
c. Recycling model and water budget
An often-discussed S–P feedback process is precipitation recycling. Lately, various recycling models have been developed for quantifying the recycling ratio. Initially, the recycling models were one dimensional and later extended to two dimensions and consequently accompanied with a bulk approach (Trenberth 1999).
Here the S–P feedback analysis is based on SCH99. They formulated two bulk characteristics: the recycling ratio β = ET/(ET + IN) and the precipitation efficiency χ = P/(ET + IN), where ET is evapotranspiration, IN is influx, and P is precipitation in the analysis regions (introduced in section 3) as depicted in Fig. 1. This recycling model incorporates four assumptions: 1) the atmospheric flow across the region is approximately unidirectional, 2) the water vapor transported across the boundary or evapotranspirated within the region is well mixed, 3) the vertical fluxes P and ET have small variability, and 4) the atmospheric moisture storage is negligible for long averaging time periods (say, monthly). We assume that these aforementioned assumptions are fulfilled in our analysis regions. Note that the definition of the recycling ratio β following SCH99 is based on evapotranspiration within a region and the moisture influx into it. However, multiplying both the numerator (ET) and the denominator (ET + IN) with the precipitation efficiency yields the commonly used definition of the recycling ratio as the ratio of the locally derived precipitation to the total available precipitation.
According to the water balance equation, the associated mass conservation for water vapor in an atmospheric column can be approximated as
where W is the amount of water vapor stored in the atmospheric column. In case of long time averages, changes in the water storage term can assumed to be zero. The moisture convergence −∇ · Q is the net inflow (IN − OUT) of water vapor across the lateral boundary faces.
With the flux integral approach (Bosilovich and Chern 2006), we can calculate the incoming (IN) and outgoing (OUT) fluxes across any arbitrary boundary line. An additional merit of this method is the lowering of a three-dimensional problem to a two-dimensional problem through vertical integration of the vapor flux. Here, the moisture fluxes were calculated from the model’s monthly mean data. However, the covariance between moisture values and wind vector components for single model output time steps is close to zero (not shown). Thus, the average of the moisture fluxes calculated from single model output time steps is close to the moisture fluxes calculated from the monthly mean values. The length of a segment of longitude and latitude on the earth’s surface is calculated as in Snyder (1987). To avoid the interpolation error, we have calculated Q in the model sigma layers following Wang and Yang (2008).
d. The S–P feedback processes
We have, from the definition of precipitation efficiency χ,
For the sensitivity experiments (WET or DRY), Eq. (1) can be written as
where the Δ terms indicate differences between the perturbed and the control simulations. The first term on the right-hand side of the Eq. (3) reflects the precipitation change by direct processes through changed water availability. The second term depicts the indirect contribution through changes in atmospheric precipitation efficiency. The latter process will be called efficiency effect in the following. The direct processes will be called surface and remote effect. Thus, we rewrite Eq. (3) to
3. Results and discussions
A spinup time of 1 year is discarded in the subsequent analyses to avoid initialization errors. To assess the regional water cycle, we have divided the Indian domain into four subregions: east (E), west (W), central (CE), and north (N), as depicted in Fig. 1. One of the problems in the comparison of the water fluxes is that the temporal variability of the water vapor budget components is high (low) in small (large) areas (Zangvil et al. 2010). To avoid the scale dependency of the water vapor components we have chosen regions of equal size of approximately 2 × 105 km2.
a. Spatial distribution
India receives most of its annual precipitation during the monsoon season, which spans the four months from June to September (JJAS). To assess whether the CCLM model is able to realistically reproduce the monsoonal precipitation, Fig. 2 shows the 1990–2007 (JJAS) mean difference between the CCLM CTL precipitation and the precipitation analysis dataset Asian Precipitation Highly-Resolved Observational Data Integration Towards Evaluation (APHRODITE; version 1003R1; Yatagai et al. 2009). The APHRODITE mean (JJAS) precipitation is also shown in the figure. As can be seen, the CCLM simulation captures the monsoonal precipitation reasonably well. We find a significant spatial correlation of 0.8 of the simulation with the APHRODITE dataset and a similar spatial variability (standard deviation of 140 mm). However, there are substantial differences; especially there is overestimation over the western coasts of India and of Myanmar, and underestimation in the northern as well as central parts of India. Recently, these bias patterns have also been observed in other regional climate model (RCM) simulations (Lucas-Picher et al. 2011) as well as in former CCLM model simulations with lower horizontal resolution (Rockel and Geyer 2008; Dobler and Ahrens 2008, 2010). Dobler and Ahrens (2008, 2010), in their CCLM model simulations with coarse grid scale of 0.44°, concluded that the overestimation is mainly due to an excessive moisture transport from the sea to the Ghats mountain range, which is driven by the Somali jet. Lucas-Picher et al. (2011), in their multimodel simulations, found substantial spread over the regions, particularly with respect to the simulated temperature and precipitation variables among the RCMs.
From Table 1, the CCLM-simulated mean summer monsoonal precipitation over the eastern analysis region is 239 mm, while in the western region it is 35 mm. The western region consists of a predominantly arid to semiarid environment, and is generally characterized by scanty rainfall. On the other hand, the eastern region lies in the humid–subtropical zone, which receives heavy sustained rainfall during the summer monsoon season. These distinct east and west model-simulated rainfall patterns are consistent with and can be explained by the previous studies of Barros et al. (2004) and Chiao and Barros (2007). They showed the existence of a hydrometorological dryline over northwestern India (associated with the region W), which separates the east heavy and west scanty rainfall regions. This dryline is effectively developed by the Thar Desert and the Aravalli mountain range weather regimes. Chiao and Barros (2007) found in their control and idealized regional climate model simulations that the moist weather regime to the east of the dryline is mostly controlled by large-scale synoptic disturbances and monsoon dynamics while the dry west weather regime is strongly influenced by surface heat fluxes.
The sensitivity of summer monsoon rainfall with response to the change in initial soil moisture is shown in Figs. 3a,b. The WET initialization increases precipitation in general. Pronounced changes are over the western and northwestern parts of India. These regions are relatively arid and dry. The mean precipitation change in WET with respect to the CTL experiment is about 6.5% (+14 mm month−1) in the eastern analysis region E, while this is larger than 29% over the western analysis region W. The precipitation changes display a remarkable regional contrast. The change pattern is similar in the DRY experiment, but in general with opposite sign and weaker in magnitude.
Figures 3c,d show the changes of the Bowen ratio (the ratio of sensible to latent heat flux) in the perturbation simulations. Overall the Bowen ratio decreases (increases) in the WET (DRY) experiment. These Bowen ratio results are consistent with the earlier studies of Douville et al. (2001) and Kim and Hong (2007). An anomalous increase in latent heat flux in the DRY simulation in the northern Himalayan foothills and the Aravalli mountain ranges can be linked to the moisture transport from the Arabian Sea and the orographical barrier in the northern Himalayan region (Barros and Hwu 2002).
To further investigate the S–P feedback process, we look at the simulated large-scale circulation. In general, the South Asian summer monsoon is associated with two important synoptic features: 1) the Tibetan high aloft in the upper troposphere over the plateau of Tibet (Flohn 1957), and 2) a heat low over the area extending from Somalia across southern Arabia to Pakistan and northwestern India including the Aravalli mountain ranges (Sajjad 2011; Joshi et al. 1990). In the present study, the two features are found to be sensitive to the premonsoonal soil moisture condition. As one can see in Fig. 4, the wet (dry) soil moisture perturbations cause an increment (decrement) in the surface pressure. As a result, there is a decrement (increment) in the corresponding geopotential height over Pakistan and adjoining areas (at 500 hPa; Figs. 4c,d) and the Tibetan Plateau region (at 200 hPa; not shown). Note that the changes at 500 hPa are more pronounced in the Tibetan Plateau because there the 500 hPa level is very close to the surface and are thus sensitive to changes in soil conditions. These local changes (especially over northwestern India) modify the large-scale circulation as shown by the vertically integrated moisture flux (Figs. 4a,b). An increment (decrement) in the surface pressure (Figs. 4c,d) over northwestern India prompts a decrement (increment) in the transportation of water vapor from the Arabian Sea to the northwest region in the WET (DRY) experiment. It is noteworthy that the drier soil moisture is forcing a strengthening in the circulation over the northern and northwestern regions, and as a result has a significant influence on precipitation (Fig. 3b) in these regions.
The monsoon, which is essentially a seasonal reversal in wind direction, causes most of the rainfall received in India. The ruling factors of the timing of the monsoon onset are still not very well known. To investigate the impact of soil moisture perturbation on the Indian summer monsoon onset, we define the onset date as the earliest date on which a threshold surface moist static energy (MSE) value of 346 kJ kg−1 is exceeded (MSE = gZ + CpT + Lq, where g is the acceleration due to gravity, Z is elevation, Cp is the specific heat capacity at constant pressure, T is temperature, L is the latent heat of vaporization, and q is the water vapor mixing ratio) as was done by Chakraborty et al. (2006). Note that there have been (and are) many onset indices developed, based on different parameters and different threshold values (e.g., Kitoh and Uchiyama 2006; Wang and LinHo 2002; Fasullo and Webster 2003; Haque and Lal 1991). However, since the present study is driven by soil moisture perturbations, the stability-based onset index is a meaningful choice. The remarkable sensitivity of the monsoon onset to the initial soil moisture conditions is illustrated in Fig. 5. The monsoon onset is delayed by up to 11 days in the DRY simulation while in the WET simulation the monsoon onset is earlier by up to 29 days. Interestingly, in the DRY simulation a small portion of the northwestern and northern regions show earlier onset dates than the CTL simulation and in the same places later onset dates in the WET experiment. The regional averages of the onset changes are given in Table 2. The results show that in all subregions, wetter initial soil conditions result in earlier onsets (for instance, 5 days prior to the CTL run in region E), while delayed onsets (4 days in region E) occur in the dry case. The monsoon onset dates have also been tested using a normalized precipitation-based index (Kitoh and Uchiyama 2006), and the results are similar to the stability-based index (not shown).
b. Water budget analysis
Figure 6 shows the recycling ratio, precipitation efficiency, and total precipitation in regions E and W for the years 1990–2008. A substantial interannual variation in the recycling ratio and precipitation efficiency can be seen. In most years the recycling ratio increases in experiment WET, with larger gain on average in region W compared to region E. To know the circumstances from which a high or low recycling ratio originates, Fig. 7 presents the interannual variation of the recycling ratio β together with influx and evapotranspiration variation in all analysis domains. Here, influx and evapotranspiration are standardized (with total mean and standard deviation). It can be seen that the recycling ratios are in the range from 0.01 to 0.22 with larger values for the northern and western (WET experiment) regions. The higher gain in region W compared to region E is as a consequence of a lower moisture influx amount in region W. The results in the region N yield relatively high regional recycling ratios β of around 0.09 (CTL), which arises because the moisture influx is low and the evapotranspiration is high. A possible explanation for the lower moisture influx (compared to W) into the region N is attributed to the attenuation of low-level monsoonal flows when it passes from the Arabian Sea via the Thar Desert.
The effect of perturbed soil water experiments on evapotranspiration can also be seen from the Fig. 7. It shows that the evapotranspiration changes are not of similar size in the two experiments. This is not surprising as they depend on the surface soil moisture condition (Kim and Wang 2007). For example, the regions W and CE are extremely arid and dry during the summertime, where in general soil water is very close to the wilting point, and as a consequence multiplying by a factor of 0.5 has a small influence only.
The precipitation efficiency is higher in region E than in region W (Fig. 6). But, the relative change in efficiency is larger in W than in E. In all experiments the efficiency shows a coherent variation in most years, but in some years (WET: 1990 and 1991 in E and 1993 and 1995 in W; DRY: 1992 and 2001 in E and 1990, 1991, 1999, 2001, and 2005 in W) efficiency decreases with wetter soil initialization and vice versa. This reflects that a negative feedback process is running in those years.
The correlation between the precipitation efficiency and the precipitation found in region W (0.9 in CTL) is higher than in region E (0.6 in CTL). Also, the correlation between the recycling ratio and the precipitation is large in W (0.8 in CTL), but is even negative in E (−0.2 in CTL and −0.3 in WET). This negative correlation is also found in case of precipitation and evapotranspiration in region E (−0.4 in CTL and −0.7 in WET), which reveals that the evapotranspiration is suppressed in regions of frequent rainfall and cloudiness.
We split Eq. (4) in section 2d to be able to differentiate between the processes: precipitation change due to the efficiency, the remote, or the surface effect. Figures 8 and 9 present the time series of precipitation changes due to these different effects. On average, the efficiency effect is largest throughout all regions. The interannual average of the efficiency effect is positive in experiment WET (Table 2).This can be explained by the fact that the increased latent heat flux (Fig. 3c) and decreased planetary boundary layer height (not shown) in the WET experiment have the effect of increasing the moist static energy and thereby increasing the susceptibility to the formation of moist convection. The efficiency effect is negative on average in the regions CE, N, and W, but positive in region E in the DRY experiment. In E, moisture convergence is slightly increasing in both perturbation experiments (1–3 mm month−1) and thus efficiency is increasing (in the model the convective parameterizations is triggered by moisture convergence).
The remote effect (i.e., the change in moisture influx) is in most years balancing the efficiency effect, but is somewhat smaller than the efficiency effect (Figs. 8 and 9). The remote effect is negative in the regions E and CE in DRY (Table 2). In the regions W and N, a positive remote effect is prominent in case DRY and negative in case WET, respectively. The surface effect is largest in absolute values in region N on average. But, a perusal of the yearly time series shows that the surface effect has a minor impact on annual time scale (Figs. 8 and 9); but, of course, the evapotranspiration change is the trigger of the remote and the efficiency effect.
Soil moisture–precipitation feedback processes were investigated through perturbation simulations with the regional climate model COSMO-CLM. The simulations were analyzed over the Indian subcontinent for the period 1990–2008. The results suggest that the premonsoonal soil moisture has a significant impact on monsoonal precipitation formation. The main results of this study are as follows.
In the WET simulations, where seasonal hindcasts were initialized with soil moisture content increased by a factor of 2, a decrease in the surface Bowen ratio is observed. In consequence, the moisture transport from the Arabian Sea into the northwestern region decreased and the transport from Bay of Bengal into the northeastern region slightly increased. On the other hand, the simulations initialized with decreased soil moisture content (DRY) yield an increase in the sensible heat fluxes, warming of the earth’s surface, and a stronger land–sea contrast in the northwestern parts of India. The effects on the atmospheric dynamics in the DRY (WET) simulations are an intensification (weakening) of moisture advection from the Arabian Sea to the northwestern region compared to the control experiment.
The premonsoonal soil moisture has a large impact on the monsoon onset. The onset is delayed by up to 11 days in the DRY simulation. This may indicate an onset delay in a warmer future climate with dryer premonsoon soils because of the S–P feedbacks.
To study the land–atmosphere interaction quantitatively, we have calculated the recycling ratio β and the precipitation efficiency χ (following SCH99) for different subregions. High values of the precipitation efficiency and recycling ratio were found over the northern (N) region. This reveals that the land–atmosphere interaction is more pronounced in region N than in the central (CE), western (W), and eastern (E) regions of the subcontinent. The influx in region N is modulated by the orography of the Himalayas and the land–sea heat contrast (induced by the soil surface condition), which maintains the high values of χ and β.
Finally, the precipitation changes in the sensitivity experiments WET and DRY are further investigated to determine whether they result from the changes in local evapotranspiration (surface effect), external moisture sources (remote effect), or in the precipitation efficiency. As a first result, the Indian summer monsoon contains all three processes, with the prevailing one varying spatially and temporally. Our analysis reveals that the efficiency and remote effects are of similar importance in changing the precipitation amounts. The surface effect is comparatively small in all regions. But, the surface effect is triggering the S–P feedback processes, and additionally it is not small compared to the combined effect of the often-counteracting remote and efficiency effects.
To summarize, this study indicates that monsoonal precipitation can be strongly influenced by premonsoonal soil moisture perturbation, and therefore the quality of seasonal forecasts with numerical models depends strongly on model initialization. The locally available atmospheric moisture is affected by changes in the surface and the remote effects. Additionally, the resulting monsoonal precipitation is dependent on the complementing changes in precipitation efficiency. Clearly, the herein-proposed hypothesis on S–P feedback mechanisms may not be model independent, as it is drawn from simulations with one model only. For example, the remote effect depends largely on the representation of dynamics in the model while the efficiency depends largely on the model physics. Therefore, a well-balanced model is necessary for a useful description of the S–P feedbacks. Small systematic errors in one part of the model or in the initialization might yield spurious trends in long-term forecasts and climate projections through feedbacks. Thus, more studies with different models, model physics, and dynamics will be necessary to gain more quantitative knowledge on the S–P feedback processes affecting the Indian summer monsoon.
The authors acknowledge funding from the Hessian Initiative for the Development of Scientific and Economic Excellence (LOEWE) through the Biodiversity and Climate Research Centre (BiK-F), Frankfurt am Main. The COSMO-CLM Community supplied access to and support in using COSMO-CLM. The authors thank the Center for Scientific Computing (CSC) of the Goethe University Frankfurt and the German High Performance Computing Centre for Climate and Earth System Research (DKRZ) for supporting part of the calculations.
Current affiliation: Institut für Meteorlogie, Freie Universität Berlin, Berlin, Germany.