1. Introduction
Whether used for forecasting or predictability research, global weather and climate models need accurate and reliable initial conditions from which to begin integrations (e.g., Beljaars et al. 1996; Fennessy and Shukla 1999; Dirmeyer 2000; Douville et al. 2001; Zhang and Frederiksen 2003). For the state of the atmosphere, initial conditions come either from global operational analyses (Lambert 1988), in the case of operational (real time or near–real time) forecasts, or from global reanalyses (e.g., Kalnay et al. 1996; Gibson et al. 1997) for hindcasts and other retrospective research applications. Similarly, there are near–real time ocean analyses and ocean data assimilation products for the initialization of coupled models (Derber and Rosati 1989; Carton et al. 2000). All such analyses are heavily based on an array of observations, using a realistic geophysical fluid dynamics model to propagate the observationally based information across all regions of the globe. However, for the land surface the choices for observationally based global analyses are limited, particularly in the case of soil wetness.
There is no global in situ observational network for soil wetness (Fennessy and Shukla 1999). There are regional and national networks of varying density and quality [e.g., Soil Climate Analysis Network (SCAN); National Water and Climate Center 2004], but many stations reporting today do not possess long continuous histories of operation. Some networks do exist with histories that span more than a decade (see Robock et al. 2000). However, the problem remains that very little of the earth's land surface area lies within a few hundred kilometers of a measurement site.
An alternative to in situ measurement networks for soil moisture is remote sensing. Sensors in the microwave band are sensitive to water content at the earth's surface. Satellite-based estimates of soil wetness can only sense the water content of the upper few centimeters of soil (Wagner et al. 1999), which represents the fast manifold of the soil moisture reservoir that provides little memory to the climate system beyond day 1, and is thus of little value for the prediction problem. Predictability would be derived from the slow manifold of the subsurface soil wetness reservoir that is important for transpiration and for surface evaporation beyond day 1 (Feddes et al. 2001). Some sort of model is necessary to propagate this information down into the soil. Additionally, the presence of vegetation can corrupt the remote sensing estimate, as plant tissues contain a great deal of water that will also attenuate the microwave signal, and leaves may hold intercepted moisture (dew or rain) that will further mask the signature of the soil below (Wagner et al. 1999).
Data assimilation is the ideal means to incorporate all available soil wetness information into a global analysis. Regional [North American (NLDAS), Mitchell et al. 2004; European (ELDAS), van den Hurk et al. 2002] and global (GLDAS, Rodell et al. 2002): land data assimilation systems (LDASs) are under development, but none are currently operational that ingest the data streams of observed land surface state variables. Because these are new efforts, there will not likely be a retrospective land surface reanalysis from LDAS that incorporates in situ land surface observations for quite some time.
The only available option left for generating global estimates of the soil wetness state is to use a land surface scheme (LSS) driven by observed meteorological forcing. This can be accomplished with the LSS either coupled to an atmospheric general circulation model (GCM), or uncoupled taking global near-surface meteorological analyses as a specified upper boundary condition. The result will be heavily flavored by the character of the chosen LSS, and the quality of the meteorological data to which the LSS is exposed (Dirmeyer et al. 1999). Different LSSs may give very different estimates of soil wetness, even when driven by the same atmospheric data (Entin et al. 1999). However, this approach represents the current state of the art for producing global long-term analyses of soil wetness.
There exist several model-estimated global datasets of soil wetness that cover at least a 20-yr period that can be evaluated for their appropriateness for initialization of weather and climate models. These include soil moisture fields from global atmospheric reanalysis projects, as well as the products of offline calculations following the methodology described above. There also exist at least two satellite-derived products, based on retrievals from microwave sensors, that give global proxies for surface soil wetness over a shorter period (on the order of 10 yr).
An important caveat should be noted for the application of any of these products for the purposes of initialization of soil wetness in prediction models. Because these fields derive from different sources, they may not correspond to soil wetness as it is represented in the target LSS. The lack of consistency of soil wetness representation between models has been clearly shown (Koster and Milly 1997) and must be dealt with carefully in any comparison among multiple models (International GEWEX Project Office 2002). Because different LSS have different operating ranges for soil wetness (i.e., very different mean values, variances, and hydrologically critical thresholds such as the point where evaporation occurs at the potential rate or where surface runoff begins), one cannot simply take the soil wetness field from one product and plug it into an arbitrary LSS as an initial condition without experiencing some sort of initialization shock.
How can one have a consistent soil wetness dataset for initialization of one's LSS? The optimal solution is to run that LSS in its own data assimilation system, or in an offline mode driven by observations, over a period of many years. This is not usually a viable option, particularly outside of operational centers that have ready access to the large quantity of forcing data and observations necessary for such a project. The second, more practical, option is to use one of the existing global soil wetness products, but to renormalize it so as to minimize the statistical differences between its climatology and that of the target LSS. This still requires some knowledge of the climate of the target LSS. Fortunately, most weather or climate models have participated in some sort of multiyear intercomparison project, or at least have performed either long-term integrations or sufficiently large numbers of short integrations. Analysis of these simulations can provide basic statistics on the behavior of soil wetness as produced by the target LSS across the model domain.
In this paper, we examine the characteristics of six model-derived global soil wetness estimates, and two global remote sensing products. We validate their abilities to simulate the phasing of the annual cycle and to accurately represent interannual variability in soil wetness by comparing to in situ measurements. We then apply renormalization among the model-derived products to see how it alters the character of the soil wetness climatologies. Section 2 describes the various global soil wetness products examined in this study. A comparison among the products is presented in section 3. Local validation of the products is conducted in section 4. Section 5 shows the normalization procedure and its results. Conclusions are given in section 6.
2. Data descriptions
a. Model estimates
Six model-derived global soil wetness products are examined. All models are built around physically based parameterizations of the behavior of the land surface water balance. Four of the models also include a full representation of the surface energy balance. Only global model-derived datasets that span at least 20 yr (1980– 99) are included here, to ensure statistical stability and overlap the period of interest to many seasonal climate modeling studies. This choice means we exclude otherwise viable products such as those of Schemm et al. (1992), Schnur and Lettenmaier (1997), and Berg et al. (2003). Table 1 presents a synopsis of the characteristics of the included models, which are described in detail below.
The simplest model of the land surface is that of Willmott and Matsuura (2001, referred to hereafter as W&M). It is based on a simple water-budget procedure (Willmott et al. 1985) with a semiempirical estimation of evaporation and runoff based on monthly mean precipitation and temperature, as well as the level of soil moisture storage within a single-layer bucket-style soil hydrology with a 150-mm capacity. A global digital elevation model is used to impart local variation on runoff and soil wetness. The model also includes a snow-water budget, and winter freezing of soil moisture stores. The dataset is monthly, spans 50 yr (1950–99), and is at a spatial resolution of 0.5°.
A similar global product has recently been developed at the Climate Prediction Center (CPC; Fan and Van den Dool 2004) based on the algorithm of Huang et al. (1996) as it was originally applied over the United States. It is also based on a single-layer bucket hydrology, but with a capacity of 760 mm. It also takes precipitation and temperature as its meteorological inputs and uses a similar approach to W&M to calculate evaporation based on Thornthwaite (1948), but the calculations are performed daily. Also, surface runoff and base flow are separately parameterized, and parameter calibration for runoff and soil moisture is performed using field data from Oklahoma (Huang et al. 1996). These calibration values are then applied globally. The CPC data cover the period from 1948 to the present (the data are updated daily in near–real time) and are available at 0.5° resolution. Daily data have been averaged to monthly for this study.
A more complex procedure was used to produce the Global Offline Land-surface Dataset (GOLD; Dirmeyer and Tan 2001). A modified version of the simplified Simple Biosphere LSS (SSiB; Xue et al. 1991, 1996; Dirmeyer and Zeng 1999) was integrated in an offline mode from 1976 to 1999, driven by hybrid meteorological forcing from the National Centers for Environmental Prediction (NCEP)–National Center for Atmospheric Research (NCAR) reanalysis of Kalnay et al. (1996), rescaled by observations where available, which include downward surface shortwave and longwave radiation; near-surface temperature, humidity, and winds; surface pressure; and two components of precipitation: convective and large scale. This LSS was integrated on a 30-min interval, with 6-hourly meteorological forcing data interpolated to the model time step. This LSS includes complete water and energy budget calculations and parameterizes the impact of vegetation on these budgets explicitly (e.g., the roles of heat and moisture stress on the transpiration component of total evaporation). Global spatially varying datasets of vegetation and soil properties are used to assign many of the parameters in the model. The model has been run at a variety of resolutions—the highest resolution of global data matches the T63 spectral resolution of its host global atmospheric model, which corresponds to a horizontal resolution of 1.875°. There are three soil layers, a surface layer of 50-mm depth, a root zone that is accessible for transpiration, and a deep recharge zone. The depths of the two lower layers vary spatially, but the total soil column is nowhere less than 1 m. Monthly data are used for this study.
The last three model-based soil wetness products used here are from atmospheric reanalysis efforts by operational meteorological centers. These products were produced in a coupled mode, where a global atmospheric circulation model is integrated in tandem with the LSS. The reanalysis schemes employ meteorological data assimilation to produce a long time series of realistic analyses of the atmospheric state. The land surface responds to the constrained atmosphere and can affect the GCM states through the coupled fluxes. Because of the use of analysis increments in the tendency equations to nudge the atmosphere toward observed conditions at regular intervals, the surface energy and water budgets often do not close over land (Roads and Betts 2000). Reanalyses are not immune to systematic errors, which can also manifest themselves in the land surface state variables (Berg et al. 2003).
The European Centre for Medium-Range Weather Forecasts (ECMWF) is finishing production of a global reanalysis (ERA-40; Simmons and Gibson 2000) spanning the period from September 1957 to the present. The model uses a relatively sophisticated LSS referred to as the Tiled ECMWF Scheme for Surface Exchanges over Land (TESSEL; van den Hurk et al. 2000) with four soil levels and subgrid tiling of vegetation and snow cover. The reanalysis is at a spatial resolution of 1.125°. Monthly volumetric soil wetness data are provided. Because there are not routine observations of soil state variables, they are not directly assimilated as part of the analysis cycle. Screen-level temperature and humidity analyses are used to provide an analysis increment to the soil wetness at the beginning of each 6-h cycle, based on regression analysis of the model temperature and humidity errors, thus providing an initial soil wetness state for integration (Douville et al. 2001).
There are two reanalysis products from NCEP. The first is the NCEP–NCAR reanalysis (Kalnay et al. 1996). This estimate uses the Pan and Mahrt (1987) LSS, a model of a level of complexity somewhat lower than for GOLD, but substantially greater than W&M or CPC. The resolution of the reanalysis is T62, or 1.875° in longitude by 1.915° in latitude, and monthly data are available from the late 1940s to present. Two layers of soil wetness are available, a surface layer of 100-mm depth, and a lower layer that reaches 2 m below the surface. The soil wetness is relaxed to a climatology based on Mintz and Walker (1993), in order to avoid excessive climate drift in the soil wetness caused by systematic precipitation errors in the GCM. This relaxation tends to damp the interannual variability of soil wetness in the analysis.
The final reanalysis estimate is from the NCEP–Department of Energy (DOE) reanalysis for the second Atmospheric Model Intercomparison Project (AMIP-II) period (Kanamitsu et al. 2002). This estimate, like the NCEP–NCAR reanalysis, uses the Pan and Mahrt (1987) LSS with the same vertical structure of the soil. The NCEP–DOE reanalysis spans the period 1979–present. It contains a simple rainfall assimilation over land as a means to constrain soil wetness. To correct the infiltration of model rainfall into the soil, it uses the pentad precipitation data of Xie and Arkin (1997), the monthly mean version of which is the basis of the hybrid GOLD precipitation forcing. The goal of this assimilation is to prevent climate drift in soil wetness in a more elegant fashion than in the NCEP–NCAR reanalysis. A thorough description of this correction procedure is given in Kanamitsu et al. (2002). The resolution of the reanalysis is also T62. Daily data are averaged to monthly for this dataset.
b. Satellite estimates
There are two global estimates of soil wetness derived from remote sensing (Table 2). The first is an experimental soil wetness index (SWI), developed by the National Oceanic and Atmospheric Administration/National Environmental Satellite, Data, and Information Service (NOAA/NESDIS) (Basist et al. 1998), which uses data from the Special Sensor Microwave Imager (SSM/I) onboard the Defense Meteorological Satellite Program (DMSP) F-10 and F-13 series of polar-orbiting satellites. The SSM/I is a four-frequency, seven-channel instrument measuring terrestrial radiation at 19.35, 22.235, 37.0, and 85.5 GHz. All measurements are at dual polarizations with the exception of the 22.235 GHz, which is made in vertical polarization only. The SWI uses the difference between the 85- and 19-GHz horizontally polarized radiation from the SSM/I instrument. These differences are selectively scaled to enhance the extremely wet to flooded soil conditions. The monthly product is derived by compositing the daily SWI values in order to identify flooded areas that may otherwise be obscured by precipitation. The product is available for the period 1988–present with missing data from June 1990 through December 1991, due to lack of an SSM/I platform aloft during that time. Some months have swaths of missing data because of instrument or retrieval problems. The spatial resolution of the product is at 0.33°. Because it is designed mainly to assess the areal extent of flooding, it is rather insensitive to variations in the dry range of soil wetness.
The other product is a global, multiyear SWI derived from active microwave data acquired by the scatterometer aboard the two European Remote Sensing (ERS) satellites: ERS-1 and ERS-2 (Wagner et al. 1999). The instrument is operated in the C band (5.3 GHz) and has a native spatial resolution of about 50 km. The final scatterometer-derived SWI product is characterized by a grid spacing of approximately 28 km. Such grid spacing results in an irregular grid when transformed to geographic coordinates. The retrieval algorithm is based on a change detection approach that naturally accounts for surface roughness and heterogeneous land cover. More specifically, the change detection technique keeps track of the changes caused by the addition (rainfall) and removal (evaporation, freezing) of liquid water in soil and vegetation. The problem of correctly interpreting backscatter measurements is reduced to the interpretation of temporal changes related to the surface dielectric properties (topsoil moisture, frozen/thawed) and the vegetation phenology. Static surface parameters influencing backscatter—for example, surface roughness and land cover—are indirectly described by reference values established from long backscatter series. Retrievals are not possible over frozen soils, tropical forests, nor over regions of azimuthal anisotropic surfaces (e.g., sandy deserts). Monthly data are available, which represent the value on the last day of the month.
Microwave-based estimates of soil wetness have unique limitations that were noted in the introduction. The measurements are limited to the depth of penetration/emission of the microwave radiation, usually only a few centimeters. To overcome this limitation, the ERS dataset applies a simple infiltration model to estimate soil wetness over a deeper column. Second, moderate to dense vegetation will strongly attenuate the soil surface signal and add its own contribution to emitted radiation (i.e., the measurement includes leaf water content and canopy interception). Thus, remote sensing estimates may be of questionable use in areas of forest or dense ground cover. The ERS product is available with a separate mask for dense vegetation regions, but for comparison purposes we have not applied the mask here.
c. In situ observations
The most complete collection of actual measurements of soil wetness with broad spatial and temporal coverage is the Global Soil Moisture Data Bank (GSMDB) of Robock et al. (2000). This collection of mostly gravimetric station measurements covers regions of North America, Europe, and Asia. Most measurements are in agricultural areas, taken in grassy plots. For purposes of validation, we focus on station data in five regions (Illinois, China, India, Mongolia, and the former Soviet Union). The Soviet data are further divided into two categories, representing winter and spring cereal fields. These datasets span anywhere from 11 to more than 20 yr, although individual stations may have a much shorter record of observations.
d. Preprocessing of the datasets
In order to compare these various gridded global products and in situ observations, some preprocessing has been performed to make them more compatible with one another. Each dataset is converted to a 0–1 scale to represent an SWI. For multilayer models, the uppermost layer is used to represent the surface SWI, and the top two or three layers (extending at least 1 m down) are combined to give a column-available SWI.
The monthly mean soil moisture from GOLD is reported in units of depth in each layer (only the surface and rooting layers are considered here). These values are divided by the local saturated capacity of the surface and rooting layers to give an SWI on a scale of 0–1, where 0 represents complete desiccation, and 1 is complete saturation. The CPC data are likewise reported as a depth of available soil water and are scaled by the total capacity of 760 mm. The NCEP–DOE daily values of 0–10- and 10–200-cm volumetric soil wetness are averaged for each month to get monthly mean soil values. These are then divided by porosity to give an SWI. The ERA-40 and NCEP–NCAR reanalyses are already given as monthly values, so only the scaling by porosity is necessary for these fields. The W&M dataset is handled somewhat differently, because the 150 mm as originally concocted by Manabe (1969) represents the active or middle third of the soil moisture range of a 1-m column of soil with porosity of 0.45. One hundred fifty millimeters are added to the W&M soil moisture, and the result is divided by an assumed capacity of 450 mm. Because of the bucket treatment, the range of soil wetness is limited to 0.33–0.67, representing the assumed wilting point and field capacity of the soil, and the behavior of soil wetness outside of those bounds cannot be tracked.
The SSM/I and ERS remote sensing data are already on a unitary scale and can be treated as SWI directly after conversion from percentage to fractions. However, the ERS uses values greater than 105% to represent frozen ground, so these readings must be set to missing data during preprocessing.
Data from the GSMDB reflect the measurement conventions of each nation and generally have a much higher vertical resolution than the model products listed above. The uppermost layer is used to represent surface SWI, and usually the soil moisture in the column down to 1 m or the deepest observational level, whichever is shallowest, is used to estimate column-available SWI. In some cases, deep layers are unreliable or have incomplete soils information, so the column is truncated to 1-m depth.
For each of the 43 stations in China, gravimetric observations start from January 1981, end in December 1991, and are taken on the 8th, 18th, and 28th of each month. Measurements are taken at 11 levels (5, 10, 20, 30, 40, 50, 60, 70, 80, 90, and 100 cm). Lacking information on the soil characteristics at these stations, we have taken the maximum instantaneous values of soil moisture at each layer over the 11-yr period as the field capacity, and then divided all measurements for that layer by the estimated field capacity to produce an SWI.
The data over Illinois are the most complete and well documented. Observations start from January 1981. There are 19 stations with measurements at 12 levels (10, 30, 50, 70, 90, 100, 110, 130, 150, 170, 190, and 200 cm). The daily samples for the top five layers plus half of the amount from the sixth layer are averaged whenever the data are available to get monthly mean values for column-available soil moisture. Available field capacity data are provided for the upper layers and are used to produce column-available SWI, as well as surface SWI from the 10-cm level.
The data for Mongolia and the former Soviet Union are very similar in character. The Mongolian observations begin from January 1964 and are available through December 1999. There are 42 stations and 11 levels (5, 10, 20, 30, 40, 50, 60, 70, 80, 90, and 100 cm). The observations are sampled on the 7th, 17th, and 27th of each month. The three samples for each month are averaged whenever the data are available to get the monthly mean. All levels are used to calculate the column-available SWI. The Soviet observations start from January 1958. There are 171 stations, but only two layers (2 and 100 cm). The observations are sampled 3 times per month like the Mongolian data.
Maximal overlap between the model and observational data exist for the 20-yr period 1980–99, so this is the period of study we have chosen. The satellite-derived products cover a shorter span; from 1988 to 1999 for the SSM/I product (with no data from June 1990 through December 1991), and from 1992 to 1999 for the ERS product. Where data are not complete for the entire period, we use data where they are available and adjust acceptance criterion and significance thresholds accordingly. Where more than 75% of the observational data for a station are missing during the period of record for that dataset, the station is excluded from the validation exercise. Global comparisons are restricted to land areas north of 60°S, thus neglecting Antarctica.
3. Comparison among global climatologies
Where a product has multiple vertical soil layers, an approximation of a root zone (column available) SWI is calculated and included in the following comparisons. For the NCEP reanalysis products, that includes both layers (down to 2 m). For the ERA-40, the top three layers (down to 1 m) are used. Layer depths below the surface vary in the GOLD product, but always the top two layers are used, which descend to a range of 0.5– 2.5 m. These differences in soil layer depths will lead to somewhat different time scales of variation and degrees of damping of the surface forcing signal among models. Although we do present figures with magnitudes of SWI, all direct comparisons among models and with observations are presented in terms of spatial and temporal variability of monthly mean data, which are less affected by these variations among products.
a. Mean climate
Figure 1 shows the annual mean column-available SWI for the eight products. The different character of the SSM/I product is immediately evident. Even though it is called a soil wetness product, it is more a measure of the saturated fraction of total land area. The other products clearly show the expected large-scale climate features such as the dry deserts, wet forests, and grasslands and savannahs with moderate soil moisture. The narrow range of SWI for the W&M product, as we have defined it, does not impede its ability to show these major climatological features. The ERS product suggests that the wettest conditions exist at high latitudes, whereas the CPC product shows the wettest conditions in the Tropics. The reanalyses, W&M, and GOLD estimates all show the wettest conditions existing at both high and low latitudes. In fact, there is a great deal of resemblance among the three reanalysis products and GOLD (which is driven by surface meteorological data from the NCEP–NCAR reanalysis). Differences in coverage and resolution are also evident among the datasets, including the lack of data over permanent ice sheets in the GOLD product, the inclusion of estimates for small islands in the CPC product, and lack of data over some sandy desert areas for the remote sensing estimates.
Regionally there are many interesting differences. For instance, over the eastern two-thirds of the contiguous United States, the ERA-40, NCEP–DOE, CPC, and GOLD products all show a fanlike gradient of soil wetness, ranging from very wet over the Southeast to dry in the Southwest. W&M and NCEP–NCAR show more of a pure east–west gradient, and ERS shows a strong north–south component, with relatively drier conditions in the Southeast. A similar range of intermodel variations exist over southeastern Asia.
The ERS tends to show a wet tendency in cold regions, possibly because of some lingering snow and ice contamination of the signal, and little signal over tropical forests because of their foliage density. These are acknowledged problems with the ERS data, and it is suggested that the anomalies give a better picture of variations in the local climate state than the mean, and that comparison of values between different regions may not be useful.
Figure 2 shows the mean annual range of SWI (wettest month minus driest month, computed locally) for each product. All products show large annual variations over the southern Amazon and Matto Grosso, monsoon regions of India, Southeast Asia, and off the equator in sub-Saharan Africa. NCEP–NCAR, GOLD, and ERS suggest a stronger annual cycle over Europe than the other products. NCEP—NCAR, W&M, and SSM/I show greater variations at high latitudes than the others, while all products show minimal variability over the deserts of the Eastern Hemisphere. Despite the fact that the maximum range of W&M is limited by definition to 0.33, its annual variability is comparable to the other estimates. As with the annual mean, SSM/I shows a smaller annual cycle than the other products over most regions. ERA-40 has the smallest variation among the model products outside of the high latitudes. Over the United States, NCEP–CPC, W&M, GOLD, ERA-40, and SSM/I have a larger annual range in the east than in the central or western regions.
To compare the phasing of the annual cycle among the various products, we find the month of maximum and minimum mean SWI. These are shown in Figs. 3 and 4. A cyclic color scale is used to convey the months that the extremes take place. For the maximum SWI (Fig. 3), a late-summer peak marks the monsoon regions, generally during September for Asia, North Africa, and Mexico, and February or March over northern Australia and southern Africa. All products show a November maximum in a band just south of the equator in Africa, a March–April apex over the Nordeste region with a peak centered on July over the adjacent coast.
There is discrepancy over other regions. Over the southern Amazon, Matto Grosso, and Pantanal regions, the models range from January to April for the wettest month, while the satellite products both show an early austral summer maximum in the southeast with an April– June peak inland. Over the southeastern United States, every season is represented by at least one of the products. SWI from CPC peaks during November–January over most of the high latitudes of the Northern Hemisphere, while the other products roughly agree on a late-spring maximum, although NCEP–NCAR and GOLD give a winter maximum over much of the Atlantic sector.
Maps of the month of minimum SWI (Fig. 4) show basic agreement among the products in the monsoon regions, and much better agreement over South America than was found for the maximums. The summer minimum over northern China appears robust across the products, but farther north some products show an early-spring minimum while others place the nadir in late summer. There are other areas of disagreement. The remote sensing products imply that the minimums over western Europe occur about two months earlier than suggested by the models. Minimums over southern Africa vary largely among products as well. The January minimum at high latitudes and over western China in the SSM/I data is a reflection of ice and snow contamination of the signal in those regions.
b. Interannual variability
The key to usefulness of any global soil wetness product to improving the initialization of weather and climate forecasts lies in its ability to simulate anomalies. Figures 5 and 6 show the interannual standard deviation of SWI for the eight products during April and October. All products show a general tendency toward low variability in arid regions, and most also have low variability in the deep Tropics. Regions of high variability during April (Fig. 5) in most products include the Nordeste of Brazil, La Plata basin, Mozambique, and eastern Australia. The remote sensing products and NCEP– DOE show a band of high variability across eastern Europe from the Balkans into Kazakhstan, but the other models only hint at that feature. NCEP–NCAR, NCEP– DOE, and W&M show a dipole over North America with high variability in the west and low variability in the east. CPC has the opposite pattern, while ERA-40, GOLD, and the remote sensing products concentrate the variability in the central part of the continent.
In October (Fig. 6), the products generally show high variability in the Asian monsoon region and sub-Saharan Africa, reflecting the anomalies in wet-season precipitation during the preceding months. All products also agree on a band of high variability from western Europe to central Asia, another variable region in east-central North America, and a minimum in the western Amazon basin.
Figure 7 shows the 20-yr trend (1980–99) in column soil wetness for some of the model products. The averages of the interannual trends calculated for each of the four seasons [December–February (DJF), March– May (MAM), June–August (JJA), and September–November (SON)], are shown, with only trends significant at the 95% significance level shown. The spatial patterns of trends for individual seasons tend to strongly resemble one another for all but the GOLD estimates. The NCEP–DOE reanalysis shows the most widespread areas of significant trends, mainly drying, perhaps reflecting a lack of complete spinup of the model soil wetness from the 1979 initial state. The NCEP– CPC product shows drying trends over much of Canada and some moistening over high latitudes in Asia. The Willmott and Matsuuma estimates show drying over the Tibetan Plateau and eastern Canada, with moistening over approximately the same region of Siberia as NCEP–CPC. Trends in the GOLD estimates are very scattered and of small spatial scale. Only over a small region in north-central China do all four products show a significant trend of the same sign in the same location, suggesting that there is no real consensus climate change signal that can be determined from these soil wetness products.
c. Variation among products
It is apparent that the degree of agreement among the products varies in space and time. This is quantified in Fig. 8, which shows the spatial correlation over land among all pairings of the mean annual cycle of SWI of the eight products as a function of month. All data are aggregated onto the lowest-resolution (NCEP reanalyses) grid before the correlations are calculated. For ease of comparison, each correlation appears twice, grouped into eight panels based on the common member. Based on a conservative estimate of the number of spatial degrees for freedom taken for the coarsest resolution products (Dirmeyer 2003), correlations of 0.18 and above are significant at the 95% level.
Among the eight products, the SSM/I product is clearly the outlier. Spatial correlations between it and the other products never exceeds 0.35, and are particularly low during November–April when compared to the reanalysis products. Some of the highest correlations between any two products occur from August to January for the pair of CPC and GOLD. In fact, the GOLD product appears to provide the best consensus, as its correlations with all other products (leaving out SSM/I) never fall below 0.63. All other products show 1-month correlations of 0.55 or lower with at least one other product. Correlations between the models and ERS remote sensing products are generally high (above 0.55) except for the CPC product during May–August. By examining the fluctuations in the envelope of correlations for a particular product (ignoring SSM/I for the moment), one may get an idea of potential large-scale phasing problems in the annual cycle of that product. For example, the correlations between CPC and the other products tend to dip in late spring or early summer, suggesting that there may be a problem in this product during that time of year. The CPC calculation does not take snowmelt infiltration into account, and that may be the cause of the drop in correlations. GOLD shows signs of a dip during early spring, which may reflect problems in the timing of the spring snowmelt in that model. NCEP–DOE and NCEP–NCAR show general peaks in correlation during spring and fall, with minima in summer and winter.
In a similar fashion, Fig. 9 presents a comparison of the spatial correlations of the interannual standard deviations of monthly SWI like those presented in Figs. 5 and 6. Correlations for standard deviation are usually somewhat lower than they are for the mean SWI, with only three instances of correlations above 0.6. Some interesting relationships emerge. The two NCEP reanalysis products appear to be relatively highly correlated throughout the year. This may be a reflection of the use of identical land surface models with the same global vegetation and soil parameters, which can exert a strong control on the spatial pattern of variance. The three offline products (NCEP–CPC, W&M, and GOLD) also appear to cluster with high correlations among them, especially during boreal summer and autumn. This may be a reflection of the similarity of precipitation forcing data used to drive all three models. The SSM/I product still appears to be an outlier, but the ERS product now also shows relatively low correlations with the model estimates. Since the ERS product is the most observationally based, this difference may in fact indicate consistent problems among the models' global distribution of surface parameters, or in the parameterizations themselves.
We attempt to quantify the degrees of freedom among the eight products by calculating at each grid point (again interpolating all datasets to the coarsest grid) the largest set of products for which the correlation between the soil wetness time series among all pairs of products in the set exceeds some threshold. For this calculation, we set a lower bound on the correlation of 0.6. Note that for correlations involving the remote sensing products, the time series will be shorter. Large values of this coherence index (5 to 8) indicate a low number of degrees of freedom among the products, and a low coherence index (1 or 2) indicates a higher number of degrees of freedom.
Figure 10a shows the coherence for the total soil wetness time series (mean annual cycle is not removed). Again, in the monsoon regions there is generally high coherence among the models, with five to seven, and in a few locations even all eight products, showing high correlations across all pairings. Desert and cold winter regions tend to show low coherence. When only anomalies from the mean annual cycle are considered (Fig. 10b), coherence generally drops. The monsoon regions are no longer highlighted as regions of agreement among the products. Now the greatest coherence, showing sets of only three–five members with mutually high correlations, appears to be along the eastern margins of the continents, including much of Europe and nearly all of Australia and the conterminous United States.
4. Local validation
Long-term in situ validation is performed using the station data from the GSMDB (Robock et al. 2000). Validation is performed for surface soil wetness and for column soil wetness in the top 1 m of soil. The surface soil wetness is calculated from the uppermost layer for multilevel models, and from the satellite products as given. For the station observations, we used the shallowest measurement from each station. We assume that the bucket models do not represent surface soil wetness, and exclude them from the comparison. For the 1-m soil wetness, we choose the layers from each model and station that best approximate a 1-m soil depth. The ERS product is designed to represent column soil wetness through a time-filtering technique, so we include it with the model products for this comparison.
Model and satellite estimates of soil wetness are compared to observations over six domains: China (43 stations; 1981–91), Illinois (19 stations; 1981–99), India (10 stations; 1987–98), Mongolia (42 stations; 1980– 98), and two sets of largely collocated agricultural stations in Russia, representing spring cereal fields and winter cereal fields (171 stations total; 1980–99). In many cases only a fraction of the stations are useable for the comparisons. For each station we apply a very liberal threshold of less than 75% missing data during the period of interest that overlaps the period of the dataset. The model time series at the grid box that contains a particular station is compared to the station data. In some cases, the corresponding model grid box for a station is an ocean point, and no analysis is conducted for that model–station combination.
Figure 11 compares the model estimates to available observations for 1-m soil wetness covering the period 1980–99. Because of the missing data issues described above, the percentage of stations used in this comparison are 86% (China), 100% (Illinois), 90% (India), 45% (Mongolia), 50% (Russia—spring cereal), and 46% (Russia—winter cereal). Over Russia and Mongolia in particular, there are few or no data during winter because of frozen ground. Median temporal correlations for each model are plotted for both the monthly means (including the mean annual cycle) and anomalies. Also shown is the percentage of useful stations that show a statistically significant correlation to each model's time series.
Several features are notable. First, there is a tendency for correlations to be higher for the total field than for the anomalies. This is true for all models over Illinois and India. However, there are exceptions over other regions such as for almost all models over Mongolia, and certain models over other regions (e.g., NCEP–CPC). A lower correlation for the total field relative to the anomalies suggests that the product may have some problem in the phasing of the annual cycle of soil wetness in that area. Note that soil wetness measurements are taken gravimetrically in Russia, Mongolia, and China, and no measurements are taken when the ground is frozen.
Second, there are differences in skill between regions. As mentioned above, correlations for the total soil wetness time series are relatively low over Mongolia. India is representative of the monsoon regions of the globe, where we have seen that the products have strong seasonal cycles and are in good agreement with one another. The robust hydrologic forcing in this region is reflected as high correlations for all products in the total field. However, India shows some of the lowest overall correlations for anomalies. Illinois shows some of the highest overall median correlations for both the total field and anomalies, perhaps due to the quality of the meteorological data over that region, which is used to drive the various models. Significance counts are routinely 90%–100% over Illinois for all models as well.
There are also strong distinctions between products. The NCEP–DOE reanalysis appears to perform particularly poorly—consistently worse than its predecessor, the NCEP–NCAR reanalysis. This is especially surprising since a different approach to handling precipitation was applied in the analysis in order to improve its soil moisture characteristics. Lu et al. (2004, manuscript submitted to J. Hydrometeor.) examines the similarities and differences between the soil moisture climatologies of the two NCEP reanalyses in detail. The NCEP–CPC and W&M products do well in warmer climates, but struggle to represent the annual cycle over Russia and Mongolia, perhaps due to their simple bucket hydrology. The GOLD product appears to perform well in the total field over all regions, with ERA-40 close behind. Some products do very well in the total field over some regions, but poorly in others, such as the NCEP–NCAR reanalysis. The NCEP–CPC and GOLD products perform consistently well in terms of anomaly correlations, and ERA-40 is strong except over India.
Figure 12 shows the same 1-m soil wetness statistics as Fig. 11 for the period beginning in 1992, and includes the ERS remote sensing product. This period is after the record for the Chinese station data, and the Mongolian data are largely missing during the 1990s, so there are no statistics for those regions. Nevertheless, we see that the ERS estimates provide competitive correlations to station data for both the total field and anomalies. The performance of the ERA-40 product also appears to be better during this period than for the entire 20-yr span, and the correlations of the total fields for the bucket models are also much higher.
Comparison of surface soil wetness (Fig. 13) during the 1992–99 period shows weaknesses in the NCEP– DOE and SSM/I estimates over most regions. The GOLD and NCEP–NCAR estimates do the best job of capturing the variation in soil wetness anomalies over India, but in other regions the ERA-40 and ERS outperform their peers. All four of those products perform comparably well in simulating the total soil wetness time series.
The poor performance of the SSM/I product may be attributable to the different nature of its measure (open surface water) and the problems the product has with signal contamination, especially by ice and snow. The weakness in the NCEP–DOE product is more mysterious. Figure 14 shows the complete time series of 1‐m soil wetness for all models and the observations corresponding to one of the stations in Illinois. The observations are highlighted as the black curve near the top of the figure, and the NCEP–DOE reanalysis is the bold blue curve below it.
The NCEP–DOE reanalysis exhibits multiyear trends and short-term jumps that do not appear in the observational record nor in the other models. For instance, that model's soil wetness is unusually high in 1986, trends down to 1990, and then back up to 1994. Also, some years have an extremely weak annual cycle in the NCEP–DOE reanalysis, such as 1995. These aberrations appear in other locations as well and appear to diminish the correlations for this product.
This time series also illustrates some of the other unique features of the various products. For instance, the limited capacity of the bucket hydrology used by W&M clearly limits its ability to simulate the complete annual cycle of soil wetness in this and other locations. The small range of variability seen globally in the ERA-40 product in Fig. 2 is also evident here at the station level. Only the NCEP–NCAR and GOLD products approach the high soil wetnesses shown in the observations, but both products get significantly drier during the dry season. Note that the clear differences in the mean soil wetness (all products appear drier than the observations) have no impact on the skill as measured by correlations.
5. Soil wetness normalization
A principal reason to examine so many global soil wetness products is to determine which are best suited to using as a proxy for global observations for the purpose of initializing weather or climate models. In particular, with the burgeoning number of multimodel experiments and intercomparisons, there needs to be a consistent and realistic way to initialize the land surface state across many models. As mentioned in the introduction, soil moisture is not a true state variable in weather and climate models, and is not directly transferable among them, because different models have different operating ranges and different sensitivities of evaporation and runoff (Koster and Milly 1997). Transferability requires a method of translating soil wetness to a given model from another model or data source. Here we examine one method, based on normalizing a global gridded proxy of observed soil wetness, which we refer to as the baseline product, so as to match the statistical characteristics of multiple target models.
Figure 15 shows a portion of the time series from Fig. 14, focusing on only two products for clarity. The products in this example are from the NCEP–CPC and ERA-40 column (1 m) soil wetness estimates. The bold red and green curves represent the original NCEP–CPC and ERA-40 products, respectively. Notice that the different characteristics of the two products: the NCEP– CPC soil wetness is generally higher, and has greater range of variation, but is smoother than the ERA-40 product. To illustrate the impact of hybridization, we have applied the technique each way, using the NCEP– CPC as the baseline for anomalies that are scaled to match the statistics of the ERA-40 target (thin orange curve), and vice versa (blue). Thus, the orange curve, for example, represents how the NCEP–CPC soil wetness data might be used as a proxy for truth to initialize the ECMWF land surface model. The orange curve largely follows the trajectory of the NCEP–CPC estimate (red) but has a mean and variance that mimic the ERA-40 (green). On the other hand, the blue curve has a higher mean and larger range like the NCEP–CPC product (red), but tracks the higher-frequency variability that appears in the ERA-40 product (green). The correlation between the original time series from NCEP– CPC and ERA-40 is 0.70, as is the correlation between the two hybrid products. Each hybrid product correlates more highly with its baseline contributor (about 0.88) than with its target (0.81). Similar relationships are generally seen among the other products, except that hybrid products that use the NCEP–NCAR reanalysis as the target typically correlate very highly with the original NCEP–NCAR time series. This is because the NCEP– NCAR reanalysis time series is very strongly dominated by the annual cycle in most locations, with weak interannual variability.
We have tested the hybridization approach with the eight soil wetness products in this study by examining every combination of target and baseline products, even using the remote sensing products as if they were climatologies from models. We illustrate global results from only one example, where the GOLD product is used as the baseline. Figure 16a shows the average temporal correlation of the GOLD dataset at each point with all of the other products. Figures 16b and 16c show the impact of hybridization using the GOLD dataset as the baseline product, and each of the other products as the target. The average correlation between the GOLD product and each of the hybrid products is shown in Fig. 16b. Figure 16c shows the average correlation between each of the target products and the corresponding hybrid. Comparison of Figs. 16a and 16c reveals that hybridization appears to retain much of the information from the target product globally. The increase in correlation from Fig. 16a to Fig. 16b suggests that hybridization also contributes signal from the baseline product to the hybridized product. These general characteristics are seen in all of the combinations.
Obviously, situations can arise where there are unusual results from hybridization, such as when the variance from one product is substantially lower than the other. Care must be taken to ensure that hybrid soil wetnesses do not exceed unrealistic values when compared to the target model. Also, one must be careful when mapping data between products with different numbers of layers and depths. Mismatch between land– sea masks can also complicate the technique. Nevertheless, hybridization appears to be a viable method for transferring the information from a baseline global gridded soil wetness product to a set of initial conditions that are consistent with the target model.
6. Conclusions
Eight global gridded soil wetness products, three from offline land surface model simulations driven by observationally based near-surface meteorology, three from global atmospheric reanalyses, and two derived from microwave remote sensing, have been compared and validated in terms of their ability to simulate seasonal–interannual variability of soil wetness. The period of interest for model products is the common period that they all span: 1980–99. The remote sensing products are limited to the later years of that period. Our motivation is to determine the validity of using a single soil wetness product as a source of initial conditions for multiple weather and climate models. This assessment necessarily also involves normalization of any baseline soil wetness product so that it is consistent with the climatology of the land surface scheme within each target weather and climate model.
The different products show gross similarities in the spatial distribution of soil wetness; deserts and humid regions are similarly distributed. However, the products differ in the mean soil wetness for a given location and in the relative wetness between different locations (e.g., one product may show the Tropics to have the highest soil wetness, while another may be wettest at high latitudes). All of the models show a similar annual cycle in strong monsoon regions, but they often disagree on the month of wettest or driest conditions in other areas. There is also often disagreement as to the magnitude of the annual cycle in most locations, and the degree of interannual variability. One might expect the variability of soil wetness in the model products to closely follow the variability of precipitation. This is not always the case, especially when and where soils often saturate, limiting variance.
Skill is measured simply by the correlation between the products and station observations, either in terms of the total series or just the anomalies (mean annual cycle removed). This approach is taken in acknowledgement and acceptance of the fact that different products will inherently differ in their means and variances for the same location. Validation with long-term in situ data shows that no one product is clearly superior in all regions and layers (surface and deep; 1-m or nearest equivalent column total), but there are clear laggards. All products validate well on the mean annual cycle, particularly where there are pronounced wet and dry seasons, except in cold climates where the product does not handle cold-season processes well. Interannual variations are usually less well simulated. There appears to be a consistent geographical relationship among the performance of the products, particularly for the simulation of anomalies. This may be due to the quality of the validation data, or simply the ratio of interannual precipitation variability to the magnitude of the annual cycle. Regions where most of the hydrologic cycle is dominated by a strong annual cycle (e.g., monsoon regions) seem to have poor skill in the simulation of soil wetness anomalies. Regions with a relatively weak annual cycle of rainfall (eastern margins of midlatitude continents) show some of the best skill.
We have described a hybridization process for renormalizing soil wetness estimates from a baseline observational proxy to a target model's climatological statistics. We show that it can preserve the mean and variance of the target model while conveying the seasonal anomalies of the baseline product. Therefore, if a group of modelers can agree on a common land surface dataset for use in initializing the land surface in a multimodel investigation, hybridization may present a means for applying that common dataset to all participating land surface schemes.
Since no soil wetness product is clearly superior in all situations, and skill is high where agreement among products is also high, there may be the opportunity to statistically combine two or more products to produce a combined or consensus product whose performance is superior to any individual product. Such multiproduct ensembling may offer a way to improve global soil wetness analyses with little expense, particularly if the weighting functions found over regions with reliable validation data can be shown to be transferable to other regions.
Acknowledgments
The authors would like to thank the providers of the various soil wetness products. In particular, we wish to thank E. Kalnay, M. Kanamitsu, K. Scipal, W. Wagner, P. Viterbo, Y. Fan, W. Ebisuzaki, and S. Lu for their help in interpreting the preliminary results. We would also like to thank M. Fennessy for helpful comments on this manuscript. This work was supported by National Aeronautics and Space Administration Grant NAG5-11579.
REFERENCES
Basist, A., Grody N. C. , Peterson T. C. , and Williams C. N. , 1998: Using the Special Sensor Microwave Imager to monitor land surface temperatures, wetness, and snow cover. J. Appl. Meteor, 37 , 888–911.
Beljaars, A. C., Viterbo P. , Miler M. J. , and Betts A. K. , 1996: The anomalous rainfall over the United States during July 1993: Sensitivity to land surface parameterization and soil moisture anomalies. Mon. Wea. Rev, 124 , 362–383.
Berg, A. A., Famiglietti J. S. , Walker J. , and Houser P. R. , 2003: Impact of bias correction to reanalysis products on simulation of North American soil moisture and hydrologic fluxes. J. Geophys. Res.,108, 4490, doi:10.1029/2002JD003334.
Carton, J. A., Chepurin G. , Cao X. , and Geise B. , 2000: A simple ocean data assimilation analysis of the global upper ocean 1950– 95. Part I: Methodology. J. Phys. Oceanogr, 30 , 294–309.
Derber, J., and Rosati A. , 1989: A global oceanic data assimilation system. J. Phys. Oceanogr, 19 , 1333–1347.
Dirmeyer, P. A., 2000: Using a global soil wetness dataset to improve seasonal climate simulation. J. Climate, 13 , 2900–2922.
Dirmeyer, P. A., 2003: The role of the land surface background state in climate predictability. J. Hydrometeor, 4 , 599–610.
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.].
Dirmeyer, P. A., and Tan L. , 2001: A multi-decadal global land-surface data set of state variables and fluxes. COLA Tech. Rep. 102, 43 pp. [Available from the Center for Ocean–Land–Atmosphere Studies, 4041 Powder Mill Road, Suite 302, Calverton, MD 20705.].
Dirmeyer, P. A., Dolman A. J. , and Sato N. , 1999: The Global Soil Wetness Project: A pilot project for global land surface modeling and validation. Bull. Amer. Meteor. Soc, 80 , 851–878.
Douville, H., Chauvin F. , and Broqua H. , 2001: Influence of soil moisture on the Asian and African monsoons. Part I: Mean monsoon and daily precipitation. J. Climate, 14 , 2381–2403.
Entin, J. K., Robock A. , Vinnikov K. Ya , Zabelin V. , Liu S. , and Namkhai A. , 1999: Evaluation of Global Soil Wetness Project soil moisture simulations. J. Meteor. Soc. Japan, 77 , 183–198.
Fan, Y., and Van den Dool H. , 2004: The CPC global monthly soil moisture data set at ½ degree resolution for 1948–present. J. Geophys. Res.,109, D10102, doi:10.1029/2003JD004345.
Feddes, R. A., and Coauthors, 2001: Modeling root water uptake in hydrological and climate models. Bull. Amer. Meteor. Soc, 82 , 2797–2809.
Fennessy, M. J., and Shukla J. , 1999: Impact of initial soil wetness on seasonal atmospheric prediction. J. Climate, 12 , 3167–3180.
Gibson, R. K., Kallberg P. , Uppala S. , Hernandez A. , Nomura A. , and Serrano E. , 1997: ERA Description. ERA Tech. Rep. 1, ECMWF, Reading, United Kingdom, 72 pp.
Huang, J., Van den Dool H. M. , and Georgakakos K. P. , 1996: Analysis of model calculated soil moisture over the United States (1931–1993) and applications to long-range temperature forecasts. J. Climate, 9 , 1350–1362.
International GEWEX Project Office, 2002: The Second Global Soil Wetness Project science and implementation plan. IGPO Publication Series 37, 69 pp.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc, 77 , 437–471.
Kanamitsu, M., Ebisuzaki W. , Woollen J. , Yang S-K. , Hnilo J. J. , Fiorino M. , and Potter G. L. , 2002: NCEP–DOE AMIP-II reanalysis (R-2). Bull. Amer. Meteor. Soc, 83 , 1631–1648.
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.
Lambert, S. J., 1988: A comparison of operational global analyses from the European Centre for Medium Range Weather Forecasts (ECMWF) and the National Meteorological Center (NMC). Tellus, 40A , 272–284.
Manabe, S., 1969: Climate and the circulation, I. The atmospheric circulation and the hydrology of the earth's surface. Mon. Wea. Rev, 97 , 739–774.
Mintz, Y., and Walker G. K. , 1993: Global fields of soil moisture and land surface evapotranspiration derived from observed precipitation and surface air temperature. J. Appl. Meteor, 32 , 1305–1334.
Mitchell, K. E., and Coauthors, 2004: The multi-institution North American Land Data Assimilation System (NLDAS): Utilizing multiple GCIP products and partners in a continental distributed hydrological modeling system. J. Geophys. Res.,109, D07S90, doi:10.1029/2003JD003823.
National Water and Climate Center, cited 2004: Soil Climate Analysis Network. [Available online at http://www.wcc.nrcs.usda.gov/scan/.].
Pan, H-L., and Mahrt L. , 1987: Interaction between soil hydrology and boundary-layer development. Bound.-Layer Meteor, 38 , 185–202.
Roads, J., and Betts A. , 2000: NCEP–NCAR and ECMWF reanalysis surface water and energy budgets for the Mississippi River basin. J. Hydrometeor, 1 , 88–94.
Robock, A., Vinnikov K. Ya , 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.
Rodell, M., Houser P. , Jambor U. , Gottschalck J. , Meng J. , and Arsenault K. , 2002: Status and availability of results from NASA's Global Land Data Assimilation System. Eos, Trans. Amer. Geophys. Union,83 (Spring Meeting Suppl.), Abstract B41A-04.
Schemm, J., Schubert S. , Terry J. , and Bloom S. , 1992: Estimates of monthly mean soil moisture for 1979–1989. NASA Tech. Memo. 104571, Goddard Space Flight Center, Greenbelt, MD, 260 pp.
Schnur, R., and Lettenmaier D. P. , 1997: A global gridded data set of soil moisture for use in general circulation models. Proc. 13th Conf. on Hydrology, Long Beach, CA, Amer. Meteor. Soc., 371– 372.
Simmons, A. J., and Gibson J. K. , 2000: The ERA-40 Project Plan. ERA-40 Project Report Series 1, 63 pp. [Available online at http://www.ecmwf.int/publications/library/ecpublications/_pdf/ERA40_PRS_1.pdf.].
Thornthwaite, C. W., 1948: An approach toward a rational classification of climate. Geogr. Rev, 38 , 55–89.
van den Hurk, B. J. J. M., Viterbo P. , Beljaars A. C. M. , and Betts A. K. , 2000: Offline validation of the ERA40 surface scheme. ECMWF Tech. Memo. 295, 42 pp. [Available from ECMWF, Shinfield Park, Reading, RG2 9AX, United Kingdom.].
van den Hurk, B. J. J. M., and Coauthors, 2002: Overview of the European Land Data Assimilation (ELDAS) project. Eos, Trans. Amer. Geophys. Union,83 (Fall Meeting Suppl.), Abstract H62D-0886.
Wagner, W., Lemoine G. , and Rott H. , 1999: A method for estimating soil moisture from ERS scatterometer and soil data. Remote Sens. Environ, 70 , 191–207.
Willmott, C. J., and Matsuura K. , cited 2001: Terrestrial water budget data archive: Monthly time series (1950–1999). [Available online at http://climate.geog.udel.edu/∼climate/html_pages/README.wb_ts2.html.].
Willmott, C. J., Rowe C. M. , and Mintz Y. , 1985: Climatology of the terrestrial seasonal water cycle. Int. J. Climatol, 5 , 589–606.
Xie, P., and Arkin P. A. , 1997: Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates, and numerical model outputs. Bull. Amer. Meteor. Soc, 78 , 2539–2558.
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.
Xue, Y., Zeng F. J. , and Schlosser C. A. , 1996: SSiB and its sensitivity to soil properties—A case study using HAPEX-Mobilhy data. Global Planet. Change, 13 , 183–194.
Zhang, H., and Frederiksen C. S. , 2003: Local and nonlocal impacts of soil moisture initialization on AGCM seasonal forecasts: A model sensitivity study. J. Climate, 16 , 2117–2137.
Mean annual soil wetness (fraction of saturation) during 1980–99, except where noted
Citation: Journal of Hydrometeorology 5, 6; 10.1175/JHM-388.1
As in Fig. 1, but for mean annual range of soil wetness
Citation: Journal of Hydrometeorology 5, 6; 10.1175/JHM-388.1
As in Fig. 1, but for month of maximum mean soil wetness
Citation: Journal of Hydrometeorology 5, 6; 10.1175/JHM-388.1
As in Fig. 3, but for month of minimum mean soil wetness.
Citation: Journal of Hydrometeorology 5, 6; 10.1175/JHM-388.1
As in Fig. 1, but for the interannual standard deviation of Apr soil wetness
Citation: Journal of Hydrometeorology 5, 6; 10.1175/JHM-388.1
The 20-yr trend in column soil wetness (mm yr−1)
Citation: Journal of Hydrometeorology 5, 6; 10.1175/JHM-388.1
Interannual spatial correlation, as a function of month, of mean SWI between each global product and the remaining seven. The symbol next to each name identifies its curve in the other seven panels
Citation: Journal of Hydrometeorology 5, 6; 10.1175/JHM-388.1
As in Fig. 8, but for spatial correlation of interannual standard deviation of SWI
Citation: Journal of Hydrometeorology 5, 6; 10.1175/JHM-388.1
Index of coherence among the eight soil wetness products (see text for details) for (a) monthly soil wetness and (b) anomalies only
Citation: Journal of Hydrometeorology 5, 6; 10.1175/JHM-388.1
(top) Comparison of temporal correlations during 1980–99 between 1-m soil wetness products and station data over various regions, and (bottom) the fraction of stations for which significant correlations are found for (right) complete time series and (left) anomalies only
Citation: Journal of Hydrometeorology 5, 6; 10.1175/JHM-388.1
As in Fig. 11, but for the period 1992–99
Citation: Journal of Hydrometeorology 5, 6; 10.1175/JHM-388.1
As in Fig. 12, but for surface soil wetness
Citation: Journal of Hydrometeorology 5, 6; 10.1175/JHM-388.1
Time series of 1-m soil wetness for a station in the Illinois State Water Survey, and the collocated grid boxes of the various models
Citation: Journal of Hydrometeorology 5, 6; 10.1175/JHM-388.1
A portion of the time series in Fig. 13 for ERA-40 (green), NCEP–CPC (red), and hybrids using the ERA-40 baseline and NCEP–CPC target (blue), and NCEP–CPC baseline with ERA-40 target (orange)
Citation: Journal of Hydrometeorology 5, 6; 10.1175/JHM-388.1
Average temporal correlation between (a) GOLD vs all other products; (b) all hybrid products with GOLD baseline vs GOLD; (c) each product vs the hybrids using the corresponding target product with GOLD baseline
Citation: Journal of Hydrometeorology 5, 6; 10.1175/JHM-388.1
Synopsis of model-derived soil wetness products
Synopsis of satellite-derived soil wetness products; H is horizontal polarizations, and V is vertical polarization
