1. Introduction
Soil moisture memory, in essence the fact that the soil can “remember” a wet or dry anomaly long after the conditions responsible for the anomaly are forgotten by the atmosphere, is a key aspect of land–atmosphere interactions and has major implications for seasonal forecasting. Indeed, due to its inherent memory, soil moisture is one of the major “slow” drivers of the climate system and possibly the chief source of forecast skill for summer precipitation over land in the midlatitudes (Koster et al. 2000). A detailed understanding of the processes controlling soil moisture memory is therefore necessary for assessing the predictability associated with soil moisture on subseasonal to seasonal time scales, and for characterizing important mechanisms impacting land–atmosphere interactions on these scales.
Various studies have investigated soil moisture memory characteristics either from observations (e.g., Vinnikov and Yeserkepova 1991; Vinnikov et al. 1996; Entin et al. 2000; Wu et al. 2002) or from integrations with land surface models (LSMs) or atmospheric general circulation models (AGCMs; e.g., Delworth and Manabe 1988; Liu and Avissar 1999a; Koster and Suarez 2001; Schlosser and Milly 2002; Mahanama and Koster 2003; Wu and Dickinson 2004). The observational studies, necessarily bound in scope by the limited spatial and temporal availability of the soil moisture measurements and by the lack of observations of evaporation, have generally focused on the analysis of anomaly decay time scales for local soil moisture observations, combined in some cases with inferences from simple analytical models (e.g., Liu and Avissar 1999b; Albertson and Kiely 2001). The modeling studies, taking advantage of the extensive data available from comprehensive model simulations, have focused instead on a more global and detailed understanding of the processes controlling soil moisture memory in the framework of a given AGCM or LSM with the caveat of the possible model dependency of the results obtained. A way to address the issue of model dependency is to analyze such properties for a number of models in a common framework and to thereby establish the effects of specific model characteristics and biases on the simulated soil moisture memory. This is the main purpose of the present study.
This investigation is the first survey examining and contrasting soil moisture memory characteristics across a broad range of AGCMs. The investigated simulations—16-member ensembles spanning 1 June to 31 August for eight different AGCMs—were generated as part of the Global Land–Atmosphere Coupling Experiment (GLACE; Koster et al. 2004a, 2006; Guo et al. 2006). For the analysis of the simulations, we use an approach proposed by Koster and Suarez (2001, hereafter referred to as KS01) that relates soil moisture autocorrelation within a climate model to various climatic and model characteristics. This allows us to relate the simulated soil moisture memory within each AGCM to its representation of specific physical processes and feedback mechanisms.
The structure of the paper is as follows. Section 2 describes the analysis framework of this study (model data and the KS01 soil moisture memory equation). Section 3 gives a brief evaluation of the atmospheric forcing generated in the analyzed simulations, as well as a description of mean climatic and hydrological characteristics of the AGCMs. Section 4 describes the overall soil moisture memory characteristics of the AGCMs, and section 5 presents an analysis of the AGCMs’ soil moisture memory using the KS01 framework. Then, section 6 discusses the dependence of soil moisture memory on the soil moisture regime, and section 7 presents a brief combined evaluation of soil moisture memory and land–atmosphere coupling in the perspective of soil moisture initialization for seasonal forecasting. Finally, the main conclusions are given in section 8.
2. Analysis framework
a. Employed data from GLACE
The analyzed simulations are taken from GLACE, a recent project investigating the strength of land–atmosphere coupling in AGCMs. As part of this experiment, three ensembles of simulations spanning 1 June–31 August were conducted by a number of different AGCMs, using different specifications for the land–atmosphere coupling (full coupling, partial coupling, and no coupling). In the present study, we investigate the soil moisture characteristics of the control experiments (with full coupling, i.e., “ensemble W”) for eight of the participating AGCMs: Community Atmospheric Model version 3 (CAM3), Canadian Centre for Climate Modelling and Analysis (CCCma), Commonwealth Scientific and Industrial Research Organisation Conformal Cubic version 3 (CSIRO-CC3), Center for Ocean–Land–Atmosphere (COLA), AGCM used in the Climate and Radiation Branch at the National Aeronautics and Space Administration Goddard Space Flight Center (NASA GSFC) (GEOS-CRB), Hadley Centre Atmospheric Model version 3 (HadAM3), National Centers for Environmental Prediction (NCEP) Global Forecast System (GFS), and NASA Seasonal-to-Interannual Prediction Project (NSIPP). The main characteristics of these eight AGCMs are detailed in Tables 1 and 2). Some GLACE models are excluded from this study because they did not provide the full complement of necessary data for our calculations.
The analyzed ensemble, consisting of 16 members differing only in their initialization, is essentially a standard set of AGCM simulations with prescribed sea surface temperature (SST). The SST boundary conditions for the integrations are the observed conditions in 1994, a year not characterized by either El Niño or La Niña conditions. Note then that the impact of interannually varying SSTs on soil moisture memory, which can be particularly strong in the Tropics (Koster et al. 2000), is not analyzed in this study. For the initialization of the ensemble simulations, GLACE participants were provided with various approaches that ensured that the initial conditions would not be artificially similar between the ensemble members. The soil wetness fields analyzed here were computed in a consistent way for all AGCMs, as the vertically integrated soil moisture above the wilting point divided by the maximum allowable soil moisture above the wilting point. For a more detailed description of the experimental design and the analysis of AGCMs’ land–atmosphere coupling strength, please refer to Koster et al. (2004a, 2006 and Guo et al. (2006).
b. Soil moisture autocorrelation equation
Delworth and Manabe (1988) pioneered the study of soil moisture memory in AGCMs, using a first-order Markov process model to relate memory to potential evaporation and soil water-holding capacity. KS01 provide a more comprehensive equation that relates soil moisture autocorrelation to several features of the investigated model, including the sensitivity of runoff to soil moisture, land–atmosphere feedbacks, and seasonality. Note that under various simplifying assumptions (e.g., assuming that soil moisture memory is mostly controlled by evapotranspiration) the equation simplifies to the result of Delworth and Manabe (1988). The derivation of the KS01 soil moisture autocorrelation equation is briefly presented in this subsection. Please refer to KS01 for a more detailed description.
From Eq. (5), soil moisture memory is seen to decrease with increasing values of (c
As a final note, the linearizations underlying (2) and (3) necessarily imply simplifications in the analysis. Soil moisture has distinct upper and lower bounds, and the complex control of these bounds on memory are effectively captured here in a simple way by the evaporation and runoff sensitivity terms, which become very large as the lower or upper bound, respectively, is approached. A more complex framework—one that more explicitly treats these bounds and other nonlinearities and thresholds in the soil moisture problem—could be desirable for some regions, as results with stochastic–dynamical models suggest (e.g., Laio et al. 2001; Porporato and D’Odorico 2004). Nonetheless, the framework of (4) does allow its own complex elements of analysis. For example, it explicitly distinguishes several factors contributing to soil moisture memory, and because the linear regressions in (2) and (3) are performed independently at each grid point, it does not presuppose the same dependency of evapotranspiration and runoff on soil moisture at each location. Previous studies (KS01; Mahanama and Koster 2003) have demonstrated that (4) is generally a good approximation of soil moisture autocorrelation in a LSM or AGCM framework, despite the complexity of the problem.
3. Mean climate and biases of the AGCMs
In this section, we present the mean climatic and hydrological characteristics of the analyzed simulations as well as a brief evaluation of the precipitation and net surface radiation fields for the AGCMs’ mean. (A more detailed validation of individual model biases will be provided in section 5c). This examination will be useful for interpreting the results of the soil moisture memory analysis. Note that for all calculations involving evaporation in this study, we use a water-balance estimate of E [derived as Pn,y − Qn,y − Cs(wn+1,y − wn,y)], as the evaporation outputed by some models did not always correspond to soil areas (e.g., due to lakes in the NSIPP model). Note as well that for all computations, the first 8 days of each simulation are disregarded in order to avoid spinup problems in the atmospheric part of the simulations (the spinup of soil moisture is not an issue given the design of the GLACE experiments). For the computation of the AGCMs’ average, the NCEP grid (highest resolution so as not to lose information; see Table 1) was chosen as the common grid and the values of the other AGCMs were interpolated to this common grid.
Figure 1 shows observed [June–August (JJA)] precipitation and net surface radiation (left) and the corresponding simulated fields averaged over the eight analyzed AGCMs (right), for the whole analysis period. The radiation observations correspond to a 7-yr (1984–90) climatology from the Surface Radiation Budget (SRB) dataset; the precipitation observations are taken from the Global Soil Wetness Project (GSWP) forcing data (climatology for 1983–95) and thus reflect a combination of Global Precipitation Climatology Centre (GPCC; Rudolf et al. 1994) and Global Precipitation Climatology Project (GPCP; Huffman et al. 1997) estimates (see Zhao and Dirmeyer 2003 for details). Note that the AGCMs’ precipitation fields are generated for a single set of prescribed SSTs, while the observations cover many years and thus reflect many different sets of SSTs. Thus, some portion of the biases shown may simply reflect SST-induced rainfall variations and not necessarily model error. This is particularly true in the Tropics, where SSTs have a significant impact on interannual rainfall variations.
Precipitation is on average well simulated in the models, with a proper placement of high- (Southeast Asia, equatorial Africa, northern tip of South America) and low- (deserts) precipitation areas. Some bias is found in the northern high latitudes (Alaska, Siberia), where the AGCMs tend on average to overestimate precipitation. Net surface radiation is also relatively well captured, but there are clear positive biases, particularly in the eastern United States, as well as across Eurasia. Thus, in the mean, precipitation and radiation appear to be fairly well simulated in the models. A discussion of individual model biases will be provided in section 5c.
Figure 2 displays further average climatic and hydrological characteristics of the AGCMs over the analysis period. As expected, the mean AGCMs’ soil moisture in the midlatitudes and the Tropics is generally high in the regions of high precipitation. The relatively high values found in the northern high latitudes are likely linked with snowmelt. Regions experiencing large increases in soil moisture over the 3-month simulation period are found in the Tropics and monsoon regions of Asia, Africa, and North America, whereas those experiencing large reductions in soil moisture are generally in the midlatitudes. Evaporation is generally high in regions of high soil moisture content and warm air temperature, and also appears correlated with water-holding capacity. Note that evaporation is high even in regions experiencing an increase in soil moisture over the simulated three-month period (e.g., Southeast Asia, equatorial Africa). Runoff, for its part, is high mostly in Southeast Asia, particularly in the Ganges region.
4. Overall soil moisture memory characteristics of the AGCMs
Soil moisture autocorrelations are computed for each AGCM (from the 16 ensemble simulations) for three 27-day periods delimited at the following time steps: 9, 36, 63, and 90 (the GLACE data being available as daily output). They are then averaged over these three periods. Again, the first 8 days of each simulation are disregarded. The autocorrelations are computed in two ways: (a) directly from the models’ soil moisture values on the four indicated days, and (b) using (4). In this section, we will first analyze the overall soil moisture memory characteristics of the AGCMs [point (a)] and present a brief validation of these results with soil moisture observations. A more detailed analysis using the KS01 framework [point (b)] follows in section 5.
a. Geographical variations in soil moisture memory
The 27-day autocorrelation fields ρ27 computed directly from the soil moisture fields for each model [point (a) above] are averaged to produce the multimodel estimate of memory shown in Fig. 3 (top). Given the lack of ground observations of soil moisture and the potential for biases associated with the analysis of any individual model alone, this represents to a large extent the “best” present global estimate of this quantity, as simulated in current AGCMs. On average, the models are characterized by high soil moisture memory in midlatitude regions and by low soil moisture memory in tropical regions. Interestingly, almost all regions of low soil moisture memory are also characterized by high precipitation (Fig. 1) and high mean soil moisture (Fig. 2), and are therefore typically humid regions. Note the asymmetric distribution of low soil moisture memory regions around the equator, due to the 3-month period under consideration (boreal summer). Regions of low soil moisture memory extend far to the north in East Asia, linked with the monsoon climate of this region. Similarly, the North American monsoon system (NAMS) region (Mexico, southern United States) displays lower soil moisture memory than the rest of North America.
An indication of the intermodel differences in soil moisture memory (and thus of the uncertainty of the multimodel average presented in the top plot of Fig. 3) is the standard deviation of ρ27 among the models (Fig. 3, bottom). Strikingly, the largest spread appears in regions displaying low soil moisture memory. This suggests that there is less agreement on soil moisture memory characteristics in humid climates among the models. A more detailed discussion of intermodel differences in soil moisture memory will be provided in section 5c.
b. Comparisons with soil moisture observations
As mentioned earlier, there are only few ground observations of soil moisture available for the analysis of regional soil moisture memory characteristics. It is nonetheless interesting to evaluate to which extent the multimodel estimate of Fig. 3 (top) agrees with the available observations. We compare here regional or basinwide averages of soil moisture observations with the AGCMs’ results. The observations are taken from Hollinger and Isard (1994) for Illinois and from the Global Soil Moisture Data Bank (Robock et al. 2000) for the Amur, Dnepr, Don, Ob, and Volga River basins. The regional and basinwide average values were compiled for use in previous studies (Seneviratne et al. 2004; Hirschi et al. 2006). Note that we compare here values for 27-day periods (simulations) and 30-/31-day periods (observations). One should therefore expect the simulation values to be slightly biased high compared to the monthly observations. Moreover, the observations are taken for a fixed soil depth (down to 1 m for the measurements in the former Soviet Union, and down to 2 m in Illinois), while we consider the total column of available soil water (as defined in section 2) in the AGCMs.
The top part of Fig. 4 displays the boxes used for the analysis, while the bottom part displays histograms of ρ27 for each AGCM (light gray bars), together with the multimodel AGCM estimate (black bar) and the observed monthly autocorrelation value (dark gray bar). In general, the models are close to the observations and are able to distinguish between regions of low or high soil moisture memory. Note that the models’ average value agrees generally well with the observations, particularly in the Amur, Dnepr, Don, and Volga River basins.
This comparison also pinpoints a few outliers among the analyzed AGCMs. The CAM3 AGCM clearly underestimates soil moisture memory in Illinois. In general the COLA and NSIPP AGCMs tend to have low soil moisture memory in many regions, likely due to their comparatively low values of Cs (see section 5d). The CSIRO-CC3 AGCM, which has relatively high values of Cs, tends for its part to have a positive bias in soil moisture memory. These intermodel differences and the relevance of model biases for the simulated soil moisture memory will be discussed in more detail in sections 5c and 5d, respectively.
Despite the presence of outliers, the comparisons with the available observations are encouraging, in particular given the scale discrepancy between the AGCM simulations and the point-scale observations used for the basin averages. These results suggest that in the mean the analyzed AGCMs correctly capture the processes controlling soil moisture memory in the considered regions.
5. Analysis of soil moisture memory with KS01 autocorrelation equation
In this section, we investigate the soil moisture memory characteristics of the simulations using the KS01 soil moisture memory Eq. (4). Again, we analyze soil moisture memory characteristics averaged over three 27-day periods (see section 4). Hence, the 27-day average fluxes and the daily soil moistures at the beginning and end of each 27-day period were used to compute the terms in Eq. (4). We will first compare the equation estimates of soil moisture memory with the actual soil moisture memory of the models, in order to validate the chosen framework. Then, the AGCMs’ mean soil moisture characteristics as well as intermodel differences and geographical variations in soil moisture memory will be investigated.
a. Agreement of autocorrelation equation with simulated soil moisture memory
Figure 5 displays maps of the AGCMs’ 27-day-lagged soil moisture autocorrelation, of the corresponding values of soil moisture autocorrelation computed with Eq. (4), and of their difference. The white areas on the plots correspond to regions covered with snow or permanent ice, or to grid points characterized with soil moisture variance of zero due to peculiarities of the individual models. The limited sample size (48 points) used in the calculation of autocorrelations in this paper necessarily leads to some error in the autocorrelation estimates [in addition to the errors associated with the approach itself, and in particular the linearization assumptions made in (2) and (3)]. Monte Carlo analysis shows that the root-mean-square error (RMSE) of an autocorrelation estimate based on a sample size of 48 ranges from 0.15 for low autocorrelations (0.1–0.2) to 0.13 for high autocorrelations (0.8–0.9). The areas shaded in gray in the plots correspond to errors of below 0.15.
For most regions and models, Eq. (4) represents a very good approximation of the AGCMs’ soil moisture memory. The intermodel and intramodel variations in soil moisture memory are both clearly captured. There are only a few discrepancies, mostly in Southeast Asia (CCCma, CSIRO-CC3, and HadAM3) and in the northwestern part of South America (CCCma, CSIRO-CC3, GEOS-CRB, and HadAM3). The very good agreement between the AGCMs’ simulated soil moisture memory and the autocorrelation values derived with Eq. (4) confirms previous results with the NSIPP AGCM and uncoupled land surface models (KS01; Mahanama and Koster 2003) showing that (4) is an appropriate framework for analyzing model and regional differences in simulated memory.
b. Equation terms: Average across models
To assess the relative impact of the four terms of Eq. (4), Fig. 6 displays maps of (σwn/σwn+1), [cov(wn, Fn)/σ2wn], (c
A comparison of Fig. 6 (top) with Fig. 3 reveals that, except for the NAMS region, areas of low soil moisture memory are generally associated with high values of (a
c. Equation terms: Intermodel differences
Figure 7 displays maps of the standard deviation among the AGCMs of the terms (σwn/σwn+1), [cov(wn, Fn)/σ2wn], (c
The standard deviation of the equation terms is in general highest for higher values of the terms. Note that high deviations in the equation terms can sometimes be caused by only one anomalous AGCM. This is for instance the case for (σwn/σwn+1): The high values and high standard deviation of this term in North America and central Asia are mostly caused by very high values of this term for the GEOS-CRB AGCM (not shown). Also, most of the high values in the standard deviation of the [cov(wn, Fn)/σ2wn] term are due to very high values in the CAM3 model, which reflect low variance of the soil wetness term rather than land–atmosphere feedbacks (not shown). In the case of the (c
To investigate such intermodel differences in more detail, we now focus on the analysis of five selected regions with location displayed in Fig. 8: south-central North America (SCNA), equatorial Africa (EQAF), northern India (NIND), Southeast Asia (SEA), and the Volga River basin (VLG). These regions were chosen in order to sample a wide range of soil moisture memory characteristics. Figures 9 and 10 display for each AGCM and the AGCMs’ mean the average values in these five regions of ρ27, ρ27(comp), (σwn/σwn+1), An [see Eq. (4)], [cov(wn, Fn)/σ2wn], (c
For all five regions, ρ27(comp) is a good estimate of ρ27. Despite slight overestimations in the SEA and NIND regions, ρ27(comp) is seen to capture both intermodel and regional differences in ρ27, as previously discussed in section 5a. Note that this is the case despite the fact that the r 2 values of the linear regressions are not extremely high in most regions. Concerning the r 2 values, Monte Carlo simulations show that estimates of a and c (slopes) with an r 2 exceeding 0.08 are significantly different from zero at the 95% level (not shown). This significance test is satisfied for both a and c in all five regions (and indeed across most of the globe; not shown).
Interestingly, intermodel differences in soil moisture memory can be as large as regional differences in this quantity. Water-holding capacity is seen to be an important factor for such intermodel differences, in particular for the NSIPP AGCM, which has low values of Cs and corresponding low values of soil moisture memory in many of the considered regions. Note that the effect of Cs on intermodel variations in soil moisture memory is particularly clear in the SCNA, NIND, and VLG regions.
As previously analyzed for the AGCMs’ mean behavior, the terms (c
d. Importance of AGCMs’ biases for simulated soil moisture memory
An important question is the extent to which the simulated soil moisture memory of the AGCMs is distorted by model biases (e.g., Mahanama and Koster 2005). Here we focus on biases in atmospheric forcing (precipitation, radiation), as well as on water-holding capacity (for which no globally complete observations exist). The AGCMs’ biases in atmospheric forcing are displayed in the left and middle columns of Fig. 11, and the water-holding capacity of the AGCMs is displayed in the right column of Fig. 11. The observational datasets described in section 3 are used for the assessment of the biases in atmospheric forcing.
As for the AGCMs’ mean (section 3), precipitation is generally well represented in the individual models. One should note, however, that the GEOS-CRB AGCM presents a negative precipitation bias in North America, and that various models display positive precipitation biases in Southeast Asia (CCCma, COLA, CSIRO-CC3, GEOS-CRB, and NCEP). Biases in net surface radiation are significant, particularly for certain models: The COLA AGCM exhibits a high positive bias globally, while the GEOS-CRB AGCM has a large positive bias in Eurasia and in the eastern part of North America. At least in the case of the COLA AGCM, this radiation bias appears to be possibly linked with an underestimation of soil moisture memory (Fig. 5).
Water-holding capacity is highly variable among the AGCMs. For the same regions, differences can be as large as 1000 mm for some pairs of models. This is of course of major relevance for simulated soil moisture memory, as previously seen for selected regions (Figs. 9 and 10). Note that for the COLA and NSIPP AGCMs the distribution of ρ27 across the globe (Fig. 5) is highly correlated with variations in water-holding capacity. The importance of water-holding capacity can be easily understood if one considers Eqs. (4) and (5) and the impact of Cs on the terms (c
6. Dependence of soil moisture memory on climate regime
In this section, we address in more detail the dependency of the simulated soil moisture memory on the regional climate regime as characterized by the average soil moisture content.
a. Simulated and computed soil moisture memory
Figure 12 displays plots of ρ27 and ρ27(comp) as functions of soil wetness for each AGCM. (A simple binning procedure produced the curve shown for each model.) As expected, a sharp decrease in soil moisture memory is seen at high soil wetness values. Similar though more muted behavior is also seen for the soil moisture memory estimates derived with (4), as the right panel demonstrates. Interestingly, more than half of the models (CCCma, CSIRO, GEOS-CRB, HadAM3, and NCEP) exhibit a small decrease in soil moisture memory at low soil wetness, in addition to the decrease at high soil wetness. This thus results in a bell-shaped relationship between soil moisture memory and soil moisture content, with highest soil moisture memory values at intermediate soil wetness.
Note that the sharp decrease in ρ27 with increasing soil moisture, as well as the anomalous behavior of the CAM3 AGCM, explains the high variability in ρ27 found in humid regions, as identified in Fig. 3 (bottom). One should remark, however, that there are fewer grid points at high soil moisture than at low and intermediate soil moisture, which is responsible for some of the erratic behaviors in the tails; this is particularly relevant for the models with low resolution (see Table 1), as well as for CAM3, which is characterized by dry conditions over most of the globe. Given the parameterization specificities of CAM3, the soil wetness index used here (section 2a) might not be the best estimate of prevailing soil moisture conditions for this model. A better estimate might likely be the “BTRAN” (βt) parameter, which ranges from one when the soil is wet to near zero when the soil is dry and depends on the root distribution of the plant functional type and the soil water potential of each soil layer (Oleson et al. 2004). A recomputation of the soil moisture memory values for CAM3 using the βt parameter yields qualitatively similar results as the ones found for the other AGCMs (not shown).
b. Equation terms
Figure 13 displays the dependency of the equation terms (c
With regards to (σwn/σwn+1) and [cov(wn, Fn)/σ2wn], though these terms were generally found to be less important than (a
In summary, despite the spread among the AGCMs, this analysis shows that common features can be identified in the relationship between soil moisture memory and soil moisture regime. Soil moisture memory is lowest in humid areas, and also decreases at low soil moisture values. Its variations are mostly determined by the evaporation (dry regimes) and runoff (wet regimes) sensitivity terms, with some compensation effects by the seasonality and covariance terms. Overall, soil moisture memory is highest in regions of intermediate soil moisture, where both the evaporation and runoff sensitivity terms are small. Finally, the compensation mechanisms identified between the seasonality and the evaporation and runoff sensitivity terms are found to be important for extreme low and high values of soil wetness and suggest that soil moisture memory cannot be understood simply as a sum of the effects of the four terms of Eq. (4), but that interrelations between these terms also need to be taken into account.
7. Soil moisture memory and land–atmosphere coupling
A main motivation for investigating soil moisture memory characteristics across the globe is the possible use of soil moisture initialization for seasonal forecasting. The potential of soil moisture for seasonal forecasting is likely to be highest in regions where both soil moisture memory and land–atmosphere coupling are important. In this section, we very briefly investigate where such regions are likely to be located.
The yield of an estimate of land–atmosphere coupling across the globe was one of the main aims of the GLACE project (Koster et al. 2004a). Since these estimates are available for the simulations analyzed in this study, we can easily compare them with the average soil moisture memory of the AGCMs. Figure 14 displays in the top panel the average values of Ω(P, S) − Ω(P, W) for the eight models investigated here. For a detailed description of the Ω parameter and the land–atmosphere coupling analysis of the GLACE simulations, please refer to Koster et al. (2004a, 2006) and Guo et al. (2006). For our present analysis, it suffices to see Ω(P, S) − Ω(P, W) as an estimate of the strength of land–atmosphere coupling in the AGCMs and how precipitation is impacted by this coupling. Note that the average Ω(P, S) − Ω(P, W) obtained for the eight AGCMs analyzed here is very close to the values obtained for the whole set of GLACE simulations (12 in total; see Koster et al. 2006). This gives us thus some confidence that the subset of simulations considered in the present paper is representative of the whole GLACE ensemble. As discussed in Koster et al. (2004a, 2006), some regions of high land–atmosphere coupling (or “hot spots”) are easily recognizable from the average Ω(P, S) − Ω(P, W) values: the central Great Plains of North America (also NAMS region), northern India, the Sahel region, as well as equatorial Africa.
The bottom plot of Fig. 14 displays the quantity [Ω(P, S) − Ω(P, W)]*ρ27. This can be considered as an estimate of the “seasonal forecasting potential,” or a combination of strong soil moisture memory and strong land–atmosphere coupling. Interestingly, though this picture is very similar to the one for Ω(P, S) − Ω(P, W) only, there are a few differences. In particular, while the formerly identified hot spots of land–atmosphere coupling are naturally again clearly highlighted, other regions appear to have both a combination of high soil moisture memory and high land–atmosphere coupling: for instance the Mediterranean and Danube regions in Europe, which are known for a propension to extended drought periods. Note as well that the hot spot regions do not all present the same soil moisture memory and thus potential for seasonal forecasting: [Ω(P, S) − Ω(P, W)]*ρ27 in the Sahel and equatorial Africa regions is about as high as in the Great Plains, while Ω(P, S) − Ω(P, W) was much higher in the former than in the latter. This thus confirms that the Great Plains of North America is a region with particularly high potential for using soil moisture initialization in seasonal forecasting (e.g., Koster et al. 2004b).
8. Summary and conclusions
In this study, we have investigated the (near) monthly soil moisture memory characteristics of eight AGCMs, utilizing simulations from the GLACE experiment. This survey is the first of its kind, in that it investigates global soil moisture memory for a range of AGCMs instead of focusing on a single-model analysis. The main value of such a multimodel analysis is that 1) it allows us to determine the extent to which previous results obtained in single-model analyses might apply to a larger number of models; 2) it can help identify causes for intermodel differences in memory behavior; and 3) averaging the memory results across a number of models provides an estimate of the global memory distribution (something that cannot be obtained from observations) that is much less affected by the model-dependent deficiencies or biases that weaken single-model analyses.
The results of this investigation show that the AGCMs present relatively similar global patterns of soil moisture memory. Outliers are generally characterized by anomalous water-holding capacity or biases in radiation forcing. Despite the range in the AGCMs’ behavior, the average soil moisture memory characteristics of the models appear realistic when compared to available in situ soil moisture observations.
We use here the analysis framework proposed by KS01 for analyzing and identifying the main processes controlling soil moisture memory in the models. Equation (4) is obtained through a linearization of the dependency of the evaporation and runoff fractions on soil moisture. While this is of course an important simplification, the results of this multimodel analysis confirm previous results from KS01 and Mahanama and Koster (2003) demonstrating that (4) is a good approximation of simulated soil moisture memory. More detailed analyses comparing these results with, for example, stochastic–dynamical models (e.g., Laio et al. 2001; Porporato and D’Odorico 2004) lie outside the scope of the present study, but could offer useful perspectives for future studies.
The analysis of the four terms of (4) reveals that soil moisture memory in the models is mostly controlled by the evaporation (c
One should of course note a few caveats of our study. First, these results are derived for boreal summer only. The relative importance of the four terms is likely to vary over the course of the year. Second, the GLACE experiments analyzed here use a relatively quiescent (i.e., neither El Niño nor La Niña conditions) year for the prescription of global SSTs (see section 2a). In the Tropics, the impact of interannually varying SSTs on soil moisture memory can be particularly strong (Koster et al. 2000), and thus could also impact the values of the four terms, particularly the covariance term. Impacts in other regions cannot be excluded either. A third point is the fact that we focused on (near) monthly soil moisture autocorrelation, while it is possible that the relative strength of the controls described with (4) could be dependent upon the lag chosen. Such aspects could be the focus of future, more detailed studies on this issue.
Finally, our analysis shows that water-holding capacity is highly variable among the analyzed AGCMs and is the main factor responsible for intermodel differences in soil moisture memory. While the importance of the water-holding capacity for simulated soil moisture memory is not per se a surprising result (e.g., Delworth and Manabe 1988; Milly and Dunne 1994), of interest is the fact that the AGCMs utilize a wide variety of values for this parameter, with correspondingly strong impacts on simulated memory. This suggests that further research on this topic should first focus on the accurate determination of this parameter before complex process analyses can be performed with confidence, a fortiori when based on single-model experiments. Eventually, an improvement of the soil moisture and evaporation measurements’ networks might help reduce the uncertainty still remaining among present AGCMs.
Acknowledgments
This research project was supported by an NCCR-Climate fellowship funded by the Swiss National Science Foundation. The individual AGCM contributions were supported by the participants’ home institutions and funding agencies. We thank Sarith P. P. Mahanama and Michael Kistler for help with the observational datasets, as well as C. Adam Schlosser, Yogesh C. Sud, Ryan Teuling, and two anonymous reviewers for useful comments on the manuscript. The Ferret program (www.ferret.noaa.gov) was used for the interpolation, analysis, and graphics in this paper.
REFERENCES
Albertson, J. D., and Kiely G. , 2001: On the structure of soil moisture time series in the context of land surface models. J. Hydrol., 243 , 101–119.
Bacmeister, J. T., Pegion P. J. , Schubert S. D. , and Suarez M. J. , 2000: Atlas of seasonal means simulated by the NSIPP1 atmospheric GCM. NASA Tech. Memo. 2000-104606, Vol. 17, 194 pp.
Boer, G. J., McFarlane N. A. , and Lazare M. , 1992: Greenhouse gas–induced climate change simulated with the CCC second-generation general circulation model. J. Climate, 5 , 1045–1077.
Boer, G. J., McFarlane N. A. , and Lazare M. , 1992: Greenhouse gas–induced climate change simulated with the CCC second-generation general circulation model. J. Climate, 5 , 1045–1077.
Bonan, G. B., Oleson K. W. , Vertenstein M. , Levis S. , Zeng Z. , Dai Y. , Dickinson R. E. , and Yang Z-L. , 2002: The land surface climatology of the Community Land Model coupled to the NCAR Community Climate Model. J. Climate, 15 , 3123–3149.
Collins, W. D., and Coauthors, 2004: Description of the NCAR Community Atmosphere Model (CAM 3.0). NCAR Tech. Note NCAR/TN-464+STR, 214 pp. [Available online at http://www.ccsm.ucar.edu/models/atm-cam/docs/description/.].
Conaty, A. L., Jusem J. C. , Takacs L. , Keyser D. , and Atlas R. , 2001: The structure and evolution of extratropical cyclones, fronts, jet streams, and the tropopause in the GEOS general circulation model. Bull. Amer. Meteor. Soc., 82 , 1853–1867.
Cox, P. M., Betts R. A. , Bunton C. B. , Essery R. L. H. , Rowntree P. R. , and Smith J. , 1999: The impact of new land surface physics on the GCM simulation of climate and climate sensitivity. Climate Dyn., 15 , 183–203.
Delworth, T. L., and Manabe S. , 1988: The influence of potential evaporation on the variabilities of simulated soil wetness and climate. J. Climate, 1 , 523–547.
Dirmeyer, P. A., and Zeng F. J. , 1999: An update to the distribution and treatment of vegetation and soil properties in SSiB. COLA Tech. Rep. 78, 25 pp. [Available from the Center for Ocean–Land–Atmosphere Studies, 4041 Powder Mill Road, Suite 302, Calverton, MD 20705.].
Entin, J. K., Robock A. , Vinnikov K. Ya , Hollinger S. E. , Liu S. , and Namkhai A. , 2000: Temporal and spatial scales of observed soil moisture variations in the extratropics. J. Geophys. Res., 105 , D9. 11865–11877.
Essery, R. L. H., Best M. J. , Betts R. A. , Cox P. M. , and Taylor C. M. , 2003: Explicit representation of subgrid heterogeneity in a GCM land surface scheme. J. Hydrometeor., 4 , 530–543.
Guo, Z-C., and Coauthors, 2006: GLACE: The Global Land–Atmosphere Coupling Experiment. Part II: Analysis. J. Hydrometeor., 7 , 611–625.
Hirschi, M., Seneviratne S. I. , and Schär C. , 2006: Seasonal variations in terrestrial water storage for major midlatitude river basins. J. Hydrometeor., 7 , 39–60.
Hollinger, S. E., and Isard S. A. , 1994: A soil moisture climatology of Illinois. J. Climate, 4 , 822–833.
Huffman, G. J., and Coauthors, 1997: The Global Precipitation Climatology Project (GPCP) Combined Precipitation Data Set. Bull. Amer. Meteor. Soc., 78 , 5–20.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437–471.
Kinter, J. L., and Coauthors, 1997: Formulation. Vol. 1. The COLA atmosphere–biosphere general circulation model. COLA Tech. Rep. 51, 46 pp. [Available from the Center for Ocean–Land–Atmosphere Studies, 4041 Powder Mill Road, Suite 302, Calverton, MD 20705.].
Koster, R. D., and Suarez M. J. , 1992: Modeling the land surface boundary in climate models as a composite of independent vegetation stands. J. Geophys. Res., 97 , 2697–2716.
Koster, R. D., and Suarez M. J. , 1996: Energy and water balance calculations in the Mosaic LSM. NASA Tech. Memo. 104606, Vol. 9, 59 pp. [Available from NASA Center for Aerospace Information, 800 Elkridge Landing Rd., Linthicum Heights, MD 21090-2934.].
Koster, R. D., and Milly P. C. D. , 1997: The interplay between transpiration and runoff formulations in land surface schemes used with atmospheric models. J. Climate, 10 , 1578–1591.
Koster, R. D., and Suarez M. J. , 2001: Soil moisture memory in climate models. J. Hydrometeor., 2 , 558–570.
Koster, R. D., Suarez M. J. , and Heiser M. , 2000: Variance and predictability of precipitation at seasonal-to-interannual timescales. J. Hydrometeor., 1 , 26–46.
Koster, R. D., and Coauthors, 2004a: Regions of strong coupling between soil moisture and precipitation. Science, 305 , 1138–1140.
Koster, R. D., and Coauthors, 2004b: Realistic initialization of land surface states: Impacts on subseasonal forecast skill. J. Hydrometeor., 5 , 1049–1063.
Koster, R. D., and Coauthors, 2006: GLACE: The Global Land–Atmosphere Coupling Experiment. Part I: Overview. J. Hydrometeor., 7 , 590–610.
Kowalczyk, E. A., Garratt J. R. , and Krummel P. B. , 1994: Implementation of a soil-canopy scheme into the CSIRO GCM—Regional aspects of the model response. CSIRO Atmospheric Research Tech. Paper 32, Aspendale, Australia, 59 pp.
Laio, F., Porporato A. , Ridolfi L. , and Rodriguez-Iturbe I. , 2001: Plants in water-controlled ecosystems: Active role in hydrological processes and response to water stress. II. Probabilistic soil moisture dynamics. Adv. Water Res., 24 , 707–723.
Liu, Y., and Avissar R. , 1999a: A study of persistence in the land–atmosphere system using a general circulation model and observations. J. Climate, 12 , 2139–2153.
Liu, Y., and Avissar R. , 1999b: A study of persistence in the land–atmosphere system with a fourth-order analytical model. J. Climate, 12 , 2154–2167.
Mahanama, S. P. P., and Koster R. D. , 2003: Intercomparison of soil moisture memory in two land surface models. J. Hydrometeor., 4 , 1134–1146.
Mahanama, S. P. P., and Koster R. D. , 2005: AGCM biases in evaporation regime: Impacts on soil moisture memory and land–atmosphere feedback. J. Hydrometeor., 6 , 656–669.
McFarlane, N. A., Boer G. J. , Blanchet J-P. , and Lazare M. , 1992: The Canadian Climate Centre second-generation general circulation model and its equilibrium climate. J. Climate, 5 , 1013–1044.
McGregor, J. L., 1996: Semi-Lagrangian advection on conformal-cubic grids. Mon. Wea. Rev., 124 , 1311–1322.
McGregor, J. L., and Dix M. R. , 2001: The CSIRO conformal-cubic atmospheric GCM. IUTAM Symposium on Advances in Mathematical Modeling of Atmosphere and Ocean Dynamics, P. F. Hodnett, Ed., Kluwer, 197–202.
Milly, P. C. D., and Dunne K. A. , 1994: Sensitivity of the global water cycle to the water-holding capacity of land. J. Climate, 7 , 506–526.
Mocko, D. M., and Sud Y. C. , 2001: Refinements to SSiB with an emphasis on snow-physics: Evaluation and validation with GSWP and Valdai data. Earth Interactions, 5 .[Available online at http://EarthInteractions.org.].
Moorthi, S., Pan H-L. , and Caplan P. , 2001: Changes to the 2001 NCEP Operational MRF/AVN Global Analysis/Forecast System. NWS Technical Procedures Bulletin 484, 14 pp.
Oleson, K. W., and Coauthors, 2004: Technical description of the Community Land Model (CLM). NCAR Tech. Note, NCAR/TN-461+STR, 174 pp. [Available online at http://www.ccsm.ucar.edu/models/ccsm3.0/clm3/index.html.].
Pan, H. L., and Mahrt L. , 1987: Interaction between soil hydrology and boundary-layer development. Bound.-Layer Meteor., 38 , 185–202.
Pope, V. D., Gallani M. L. , Rowntree P. R. , and Stratton R. A. , 2000: The impact of new physical parameterizations in the Hadley Centre climate model: HadAM3. Climate Dyn., 16 , 123–146.
Porporato, A., and D’Odorico P. , 2004: Phase transitions driven by state-dependent Poisson noise. Phys. Rev. Lett., 92 .110601, doi:10.1103/PhysRevLett.92.110601.
Robock, A., Vinnikov K. Y. , Srinivasan G. , Entin J. K. , Hollinger S. E. , Speranskaya N. A. , Liu S. , and Namkhai A. , 2000: The Global Soil Moisture Data Bank. Bull. Amer. Meteor. Soc., 81 , 1281–1299.
Rudolf, B., Hauschild H. , Reuth W. , and Schneider U. , 1994: Terrestrial precipitation analysis: Operational method and required density of point measurements. Global Precipitation and Climate Change, M. Desbois and P. Desalmond, Eds., NATO ASI Series I, Vol. 26, Springer-Verlag, 173–186.
Schlosser, C. A., and Milly P. C. D. , 2002: A model-based investigation of soil moisture predictability and associated climate predictability. J. Hydrometeor., 3 , 483–501.
Seneviratne, S. I., Viterbo P. , Lüthi D. , and Schär C. , 2004: Inferring changes in terrestrial water storage using ERA-40 reanalysis data: The Mississippi River basin. J. Climate, 17 , 2039–2057.
Sud, Y. C., and Walker G. K. , 1999a: Microphysics of clouds with the relaxed Arakawa–Schubert Cumulus Scheme (McRAS). Part I: Design and evaluation with GATE Phase III data. J. Atmos. Sci., 56 , 3221–3240.
Sud, Y. C., and Walker G. K. , 1999b: Microphysics of clouds with the relaxed Arakawa–Schubert Cumulus Scheme (McRAS). Part II: Implementation and performance in GEOS II GCM. J. Atmos. Sci., 56 , 3221–3240.
Verseghy, D. L., 1991: CLASS—A Canadian land surface scheme for GCMs: I. Soil model. Int. J. Climatol., 11 , 111–133.
Vinnikov, K. Ya, and Yeserkepova I. B. , 1991: Empirical data and model results. J. Climate, 4 , 66–79.
Vinnikov, K. Ya, Robock A. , Speranskaya N. A. , and Schlosser A. , 1996: Scales of temporal and spatial variability of midlatitude soil moisture. J. Geophys. Res., 101 , D3. 7163–7174.
Wu, W., and Dickinson R. E. , 2004: Time scales of layered soil moisture memory in the context of land–atmosphere interaction. J. Climate, 17 , 2752–2764.
Wu, W., Geller M. A. , and Dickinson R. E. , 2002: The response of soil moisture to long-term variability of precipitation. J. Hydrometeor., 3 , 604–613.
Xue, Y., Sellers P. J. , Kinter J. L. , and Shukla J. , 1991: A simplified biosphere model for global climate studies. J. Climate, 4 , 345–364.
Zhao, M., and Dirmeyer P. , 2003: Production and analysis of GSWP-2 near-surface meteorology data sets. COLA Tech. Rep. 159, 22 pp. [Available from the Center for Ocean–Land–Atmosphere Studies, 4041 Powder Mill Road, Suite 302, Calverton, MD 20705.].
Zobler, L., 1986: A World Soil File for Global Climate Modelling. NASA Tech. Memo. 87802, NASA Goddard Institute for Space Studies, New York, NY, 32 pp.
(top left) Observed rainfall (GSWP forcing data; JJA) (mm day−1); (top right) AGCMs’ average simulated rainfall (mm day−1); (bottom left) observed net radiation (ISLSCP SRB data; JJA) (W m−2); and (bottom right) AGCMs’ average simulated net radiation (W m−2).
Citation: Journal of Hydrometeorology 7, 5; 10.1175/JHM533.1
(top left) AGCMs’ average simulated soil moisture; (top right) AGCMs’ average simulated monthly changes in soil moisture; (middle left) AGCMs’ average simulated evaporation (mm day−1); (middle right) AGCMs’ average simulated runoff (mm day−1); (bottom left) AGCMs’ average simulated air temperature (°C); and (bottom right) AGCMs’ average water-holding capacity Cs.
Citation: Journal of Hydrometeorology 7, 5; 10.1175/JHM533.1
(top) AGCMs’ average values of ρ27; (bottom) std dev of ρ27 among AGCMs.
Citation: Journal of Hydrometeorology 7, 5; 10.1175/JHM533.1
(top) Regional boxes used for comparison with soil moisture observations: Illinois (USILL), Dnepr (DNE), Don, Volga (VLG), Ob, and Amur (AMU); (bottom) histograms of soil moisture autocorrelation for all AGCMs (ρ27; light gray bars), the AGCMs’ average (black bar), and the observations (ρmonthly; dark gray bar) in the regional boxes defined above.
Citation: Journal of Hydrometeorology 7, 5; 10.1175/JHM533.1
(left) Maps of simulated 27-day-lagged autocorrelation of total profile soil moisture ρ27 in the eight AGCMs. (middle) Corresponding maps of ρ27(comp) as estimated with Eq. (4). (right) Differences, i.e., estimated autocorrelations minus simulated autocorrelations.
Citation: Journal of Hydrometeorology 7, 5; 10.1175/JHM533.1
Maps of mean AGCMs’ values of (top left) (σwn/σwn+1); (top right) [cov(wn, Fn)/σ2wn]; (bottom left) (c
Citation: Journal of Hydrometeorology 7, 5; 10.1175/JHM533.1
Maps of standard deviation among the AGCMs of (top left) (σwn/σwn+1); (top right) [cov(wn, Fn)/σ2wn]; (bottom left) (c
Citation: Journal of Hydrometeorology 7, 5; 10.1175/JHM533.1
Boxes used for the regional analysis: south-central North America (SCNA), equatorialAfrica (EQAF), northern India (NIND), Southeast Asia (SEA), and Volga (VLG).
Citation: Journal of Hydrometeorology 7, 5; 10.1175/JHM533.1
Regional analyses in the boxes defined in Fig. 8 (SCNA, EQAF, and NIND).
Citation: Journal of Hydrometeorology 7, 5; 10.1175/JHM533.1
Regional analyses in the boxes defined in Fig. 8 (SEA and VLG).
Citation: Journal of Hydrometeorology 7, 5; 10.1175/JHM533.1
(left) Biases in net surface radiation, (middle) biases in precipitation, and (right) water-holding capacity of the AGCMs (Cs).
Citation: Journal of Hydrometeorology 7, 5; 10.1175/JHM533.1
Distribution of ρ27 and ρ27(comp) as functions of soil moisture (computed as bins for given soil moisture ranges).
Citation: Journal of Hydrometeorology 7, 5; 10.1175/JHM533.1
Distribution of (c
Citation: Journal of Hydrometeorology 7, 5; 10.1175/JHM533.1
(top) Mean Ω(P, S) − Ω(P, W) of the eight AGCMs investigated. (bottom) Mean Ω(P, S) − Ω(P, W) of the eight AGCMs multiplied by the average ρ27 values of the models.
Citation: Journal of Hydrometeorology 7, 5; 10.1175/JHM533.1
AGCM–LSM combinations corresponding to the simulations analyzed in this paper.
LSM characteristics of the AGCMs analyzed in this paper; Cs refers to the water-holding capacity.
For the present study, y corresponds to each ensemble member.