SEARCH

SEARCH BY CITATION

Keywords:

  • land/atmosphere interactions;
  • soil moisture

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. PAR Technique
  5. 3. Soil Moisture Data Sets
  6. 4. Experimental Design
  7. 5. Results
  8. 5.3. Global Land Surface Validation
  9. 6. Conclusion
  10. Acknowledgments
  11. References

[1] The goal of this study is to produce a new soil moisture analysis for subseasonal forecasting. An accurate soil moisture initialization is today widely recognized as having a great potential to increase summertime subseasonal forecasting skill. In this study, soil moisture initial conditions are generated using the Precipitation Assimilation Reanalysis (PAR) technique. This technique consists mainly of nudging precipitation in a coupled land-atmosphere model by adjusting the vertical air humidity profile based on the difference between model and observed precipitation rates in a continuous assimilation period of a few months. The effect of the PAR technique on the model soil moisture estimates is evaluated using (1) a benchmark in global soil moisture analysis produced by the Global Soil Wetness Project Phase 2 (GSWP-2) and (2) in-situ observations from the state of Illinois. Using PAR, we find that the coupled land-atmosphere FSU/COAPS model reproduces with a reasonable accuracy the seasonal cycle and the temporal variability of monthly soil moisture anomalies produced by GSWP-2 across most of the globe. The temporal structure of monthly soil moisture anomalies match also fairly well with in-situ observations from the Illinois state deep into the soil. Finally, overall the PAR technique shows better results than the NCEP/DOE Reanalysis 2. Therefore, in this study we have developed a new soil moisture analysis product that (1) is physically consistent with the atmospheric physics of the coupled land-atmosphere FSU/COAPS model, (2) is globally comparable with GSWP-2 and (3) better fits in-situ observed soil moisture characteristics than R2 over Illinois.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. PAR Technique
  5. 3. Soil Moisture Data Sets
  6. 4. Experimental Design
  7. 5. Results
  8. 5.3. Global Land Surface Validation
  9. 6. Conclusion
  10. Acknowledgments
  11. References

[2] Many numerical studies have shown a significant sensitivity of near-surface climatological variables to soil moisture levels [e.g., Shukla and Mintz, 1982; Rind, 1982; Yeh et al., 1984; Sud and Fennessy, 1984; Fennessy and Shukla, 1999; Koster et al., 2000; Hong and Kalnay, 2000]. More recent numerical studies have shown that, in boreal summer midlatitudes, soil moisture can play a role as importantas that of the sea surface temperature in controlling the continental precipitation variability [Kumar and Hoerling, 1995; Trenberth and Branstator, 1992; Shukla, 1998; Koster and Suarez, 2000; Koster et al., 2004]. All above sensitivity studies have demonstrated that accurate soil moisture initial conditions can potentially improve subseasonal forecasts of near surface variables, particularly during boreal summers. However, most of these studies use extreme values of soil moisture initial conditions (for instance, almost dry or almost saturated). The literature dealing with the use of realistic soil moisture initial conditions is not very extended. Progress in addressing this question has been hampered by the lack of reliable global soil moisture observations to initialize global climate models. Indeed, the heterogeneity of the soil properties (e.g. porosity, permeability), the topography and the land cover types, make a global soil moisture measurement difficult. Today, this variable is sparsely measured in-situ and is not well estimated by satellite remote sensing. Despite their significant advances, the current remote sensing techniques for soil moisture still suffer from issues associated with the shallow depth of the retrieval (less than 5 cm), the absence of retrieval over dense vegetated and frozen areas, and significant uncertainties in the retrieval algorithm. To fill this gap, soil moisture analysis techniques are often used. In an ongoing model intercomparison project named the 2nd phase of the Global Land Atmosphere Coupling Experiment (GLACE-2) [Koster et al., 2011], almost all participants drive their Land Surface Model (LSM) in an offline mode using the GSWP-2 observation-based atmospheric forcing data. The offline land surface state variables are then used to initialize the coupled land-atmosphere model. However, because the offline simulation and the coupled land-atmosphere model have most likely different climatologies, the near surface atmospheric state of the forecasts may undergo a bias adjustment (or spinup). This spinup problem can decrease the short-term to subseasonal forecast skill. To solve this problem, a climatological correction is applied in GLACE-2 to the offline simulations before initializing the forecasts [van den Hurk et al., 2012]. Another alternative to solve the spinup problem is to use a coupled land-atmosphere model for both the initialization and the forecast. In this latter alternative, no climatological correction is required. Starting in 2002, reanalysis products provided by operational centers were among the first to use a land assimilation system using a land-atmosphere model in a coupled mode. For instance, the National Center for Environmental Prediction (NCEP)/Department of Energy (DOE) Reanalysis 2 (R2) adjusts the top 10 cm soil moisture using the difference between the modeled and the 5-day mean of CPC Merged Analysis of Precipitation (CMAP) precipitation rates [Kanamitsu et al., 2003]. However, when the atmospheric physics of the model simulates an error such as a clear sky while a heavy observed rain event is assimilated into the land surface that will, in turn, produce a wet soil moisture analysis. The resulting strong radiative and surface flux adjustments can impair the quality of the soil moisture analysis. This physical inconsistency between the soil moisture analysis and the atmospheric physics of the model reduces the soil moisture predictability. Current efforts have also been put into the NCEP Coupled Forecast System Reanalysis (CFSR) to produce a soil moisture analysis. CFSR performs uncoupled integration of its land surface model driven by the CMAP precipitation data every 24-hours. The offline soil moisture and soil temperature estimates are then used as initial conditions of the CFSR for the following 24-hours. Since it is a similar offline land assimilation approach as that used by GLACE-2, a spinup problem can be encountered (explained above). The aim of the part I of this paper is to produce a new soil moisture analysis using (1) a physically consistent land assimilation system and (2) a land-atmosphere model in coupled mode. In contrast with the offline soil moisture initialization technique used in GLACE-2, a coupled land-atmosphere model is used, and thus no climatological correction is required. The land assimilation system used in this study is called Precipitation Assimilation Reanalysis (PAR) and consists of assimilating high frequency (3-hourly) observation-based precipitation data into the atmospheric component of the model in a continuous assimilation period of a few months. Since precipitation is a diagnostic variable, the precipitation assimilation is performed by adjusting the vertical profile of the atmospheric humidity based on the difference between the model and the observed precipitation rates. A Newtonian nudging of dynamical variables (surface pressure, vorticity, divergence, temperature) toward R2 also is applied to reduce any model drift from the observed large scale atmospheric circulation. The combination of the dynamical nudging and the adjustment of the atmospheric humidity vertical profile not only brings the model precipitation estimates close to the observations but also redistributes the atmospheric heat and moisture, which in turn affects the adiabatic heating and hence the cloudiness. Therefore, unlike in R2, the radiative fluxes (directly affected by the cloudiness) and the surface fluxes remain physically consistent with the soil moisture analysis. This paper is organized as follows. Section 2 describes the PAR technique and the precipitation data sets used to apply and validate the PAR technique. Section 3 introduces the different soil moisture data sets used for the soil moisture analysis evaluation. Section 4 presents the experimental design. Finally, the results and conclusion are provided in sections 5 and 6 respectively.

2. PAR Technique

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. PAR Technique
  5. 3. Soil Moisture Data Sets
  6. 4. Experimental Design
  7. 5. Results
  8. 5.3. Global Land Surface Validation
  9. 6. Conclusion
  10. Acknowledgments
  11. References

[3] The goal of the PAR technique is to indirectly assimilate soil moisture by assimilating precipitation into a coupled land-atmosphere model using a nudging approach. The current techniques of land surface data assimilation have faced several difficulties. One difficulty is a lack of subsurface observations. Another difficulty is that LSMs are considerably biased compared with land surface observations. This is due to biases in the atmospheric forcing and deficiencies in the LSM parameterizations which are difficult to validate or improve because of the lack of observations such as surface latent heat fluxes. As a result, LSMs tend to have their own “climatology”. Most LSMs are driven by observation-based near surface atmospheric forcing. Then, the offline LSM simulations are corrected to account for the difference of climatology between the coupled land-atmosphere model and the offline simulations. In contrast with the offline initialization technique, in this study we use a land assimilation system in coupled mode that ensures physical consistency of the surface fluxes between the LSM and the atmospheric variables of the model. Therefore, in this case, there is no need to apply a correction to adjust the model climatology for initialization. Nevertheless, one must keep in mind that for validation purposes against observations, the correction is needed.

[4] The PAR technique is similar to the physical initialization by Krishnamurti et al. [1991] but modified by Nunes and Cocke [2004] and Kim et al. [2007]. An advantage of using the PAR technique relies on the fact that most of the weight is given to precipitation, which is the most important variable to produce a reliable soil moisture analysis. Nudging techniques are often criticized as they do not take observational error characteristics of the various assimilated variables into account. However, for soil moisture initialization we assume that precipitation observations do not have errors, as it is the case with offline methods as well. This is an improvement over currently used offline methods where the proper physical relationship of the atmospheric variables are almost completely ignored.

2.1. Precipitation Nudging

[5] The precipitation nudging is defined by an analytic expression which modifies the humidity vertical profile as a function of the difference between the model and the observed precipitation rates, in such a way that the model precipitation is brought closer to the observations. The analytic expression is a simple vertical structure function:

  • display math

where qm is the modified and q the specific humidity profile before PAR in sigma-coordinates, and Ro and Rp are the observed and model precipitation rates. The precipitation assimilation is not performed for instantaneous precipitation rates less than 10 mm.h−1. The model precipitation are nudged toward a 3-hourly observation-based data set that is described in section 3.3.

2.2. Dynamical Nudging

[6] The large-scale circulation is sensitive to the vertical distribution of the diabatic heating in the tropics. Since the precipitation assimilation modifies the vertical distribution of the diabatic heating, the model state can drift from the observed large-scale circulation. To reduce the model drift, prognostic variables (surface pressure, potential and virtual temperature, divergence and vorticity) are nudged toward the 6-hourly R2 reanalysis that is thought to represent very well the observed large-scale circulation. The dynamical nudging uses a Newtonian relaxation technique, which keeps the model variables close to R2 by adding a nudging term in the prognostic equation, while still allowing the assimilation of precipitation. The Newtonian relaxation can be expressed as follows:

  • display math

where F(ψ) represents the land-atmosphere FSU/COAPS model, α is the nudging term (10−4 s−1 for all dynamical variables), ψ is the model variable and ψa is the variable from R2. The nudging term is applied at each model time step and is interpolated within a 6-hour interval.

2.3. Precipitation Data Sets

2.3.1. GDMF

[7] Here, we describe the precipitation data sets used for assimilation. In order to capture the diurnal cycle of the topsoil moisture state, a precipitation data set with a timescale shorter than daily is necessary. Since there is no global gridded observational data sets with a temporal scale shorter than daily for the study period (i.e. 1986–1995), we choose to assimilate a bias-corrected reanalysis product. The study period was selected to match with that of the GLACE-2 project, in which FSU/COAPS participates by using the coupled land-atmosphere FSU/COAPS model to compare subseasonal forecasting skill attributed due a realistic soil moisture initialization (results given in Part II). The reader is referred to Koster et al. [2011] for more details on GLACE-2.

[8] The bias-corrected reanalysis product chosen for this study is the 3-hourly, 1.0° Global Dataset of Meteorological Forcing (GDMF) [Sheffield et al., 2006]. This product is constructed by combining the 6-hourly, 2.0° NCEP/National Center for Atmospheric Research (NCEP/NCAR) reanalysis-1 (R1) with global observation-based data sets using a statistical downscaling in time and space [Kalnay et al., 1996]. The R1 precipitation data are (1) spatially downscaled from 2.0° to 1.0° using the 1.0° Global Precipitation Climatology Project (GPCP) [Huffman et al., 2001] daily data set, and (2) temporally downscaled from 6-hourly to 3-hourly using the Tropical Rainfall Measuring Mission (TRMM) [Huffman et al., 2007]. R1 is known to show systematic biases at the monthly timescale. To remove these biases, the monthly totals of R1 are scaled to match that of the monthly precipitation data set of the Climatic Research Unit (CRU).

[9] Another 3-hourly, 1.0° bias-corrected product is available for the study period that is the Second Global Soil Wetness (GSWP-2) atmospheric forcing. This forcing product is used to generate realistic soil moisture initial conditions for climate forecasts by almost all participants in GLACE-2 [Koster et al., 2011]. The reasons why in this study we selected the GDMF over the GSWP-2 forcing are the following: (1) through its spatial downscaling, the GDMF product uses observed diurnal variability statistics based on TRMM observational data; in the GSWP-2 forcing data, the diurnal variability is only from R2 (without being combined with observations) and therefore less reliable; the assimilated precipitation data drive the topsoil moisture diurnal cycle, which, in turn, affects the variability of the deeper soil moisture layers; (2) GDMF adjusts the rain day frequencies to match observed statistics; this adjustment is not applied in the GSWP-2 forcing.

2.3.2. GPCC

[10] To validate the precipitation estimates produced using the PAR technique over the land surface, we use a gauge-based precipitation data set. The Global Precipitation Climatology Centre (GPCC) product, operated by the German Weather Service, holds the largest rain gauge station database in the world with about 65,000–70,000 rain gauge stations worldwide for the period 1986-to-present, collected from the Global Telecommunications Network in real time, supplemented by other worldwide data collections, such as the Monthly Climatic Data for the World. The available rain gauge data are first interpolated into a 0.5° grid and then averaged over a 2.5° grid [Rudolph et al., 1996].

3. Soil Moisture Data Sets

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. PAR Technique
  5. 3. Soil Moisture Data Sets
  6. 4. Experimental Design
  7. 5. Results
  8. 5.3. Global Land Surface Validation
  9. 6. Conclusion
  10. Acknowledgments
  11. References

[11] The different soil moisture data sets used to evaluate the PAR soil moisture analysis are described in this section.

3.1. In-Situ Observations

[12] Only a few in-situ measurements of soil moisture from the Illinois Climate Network are today available for the study period. This data network is provided by the Global Soil Moisture Data Bank (GSMDB) [Robock et al., 2000] and comprises 19stations covering the entire state of Illinois. The soil moisture amount has been measured using neutron probes for 11 layers down to 2 m. A detailed description of the data set and its measurement errors are given in Hollinger and Isard [1994].

3.2. Global Soil Moisture Analysis Products

3.2.1. GSWP-2 Multimodel Analysis

[13] Since there are no reliable global observations of soil moisture, an alternative to evaluating model soil moisture estimates on a global-scale is to use a multimodel analysis product. In this study, we use the Global Soil Wetness Project (GSWP-2) [Dirmeyer et al., 2002] multimodel analysis, which integrates 13 offline state-of-the-art LSMs driven by the same GSWP-2 atmospheric observation-based forcing (discussed in section 3.3.1). Because it is the only available multimodel soil moisture analysis product using state-of-the-art LSMs driven by an atmospheric observation-based forcing, it is the best proxy for global soil moisture observations for the study period. However, the GSWP-2 analysis is a model-based product and thus does not always ensure to be close to the truth. To estimate the uncertainty of a GSWP-2 land surface variable, the standard deviation among LSMs is provided [Dirmeyer et al., 2005]. The GSWP-2 multimodel analysis is also used to evaluate other land hydrological estimates besides soil moisture, such as surface runoff and surface evaporation.

3.2.2. NCEP/DOE Reanalysis (R2)

[14] The soil moisture analysis produced in this study using the PAR technique is compared with that of R2. To generate a soil moisture analysis, R2 uses a coupled land-atmosphere model in a coupled mode (as has been done in this study). However, unlike the PAR technique, the land surface assimilation system of R2 consists of correcting the top 10 cm soil moisture estimates using the difference between the modeled and the observed precipitation (5 day mean) rate [Kanamitsu et al., 2003]. The assimilation is therefore not applied throughout the soil moisture column. In addition, the precipitation observations are directly integrated into the land surface which could generate inconsistencies with atmospheric physical processes and errors could be then introduced and impair the quality of the soil moisture analysis. R2 [Kanamitsu et al., 2002] uses the Oregon State University (OSU) LSM with two soil layers: a very thin top layer (0–10 cm) and a very thick deep layer (10–200 cm) [Pan and Mahrt, 1987; Pan, 1990]. The spatial resolution is T62 (1.915°) with 28 vertical levels.

4. Experimental Design

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. PAR Technique
  5. 3. Soil Moisture Data Sets
  6. 4. Experimental Design
  7. 5. Results
  8. 5.3. Global Land Surface Validation
  9. 6. Conclusion
  10. Acknowledgments
  11. References

[15] The coupled land-atmosphere FSU/COAPS model is a global spectral primitive equation model based on an Eulerian semi-implicit scheme [Cocke and LaRow, 2000; Shin et al., 2005]. The spatial resolution is a triangular truncationof 63 waves (1.875° latitude/longitude). The model uses weekly SSTs from Reynolds et al. [2002] at the boundaries. The land surface component of the model is the advanced National Center for Atmospheric Studies (NCAR)/Community Land Model (CLM2) [Bonan et al., 2002]. Each grid cell in the CLM2 is represented by 5 primary subgrid land cover fractions (glacier, lake, wetland, urban, and vegetated). Each vegetated portion of the grid cell is divided into patches of up to 4 Plant Function Types (PFTs). The CLM2 produces prognostic soil moisture fields for 10 layers as opposed to 2 soil moisture layers in R2 or only one layer in the GSWP-2 multimodel analysis (described in section 4.1.1).

[16] Two global numerical simulations are carried out from January 1986 to December 1995. In the first simulation, using the PAR technique (described in section 3), the 3-hourly GDMF precipitation is continuously assimilated into the coupled land-atmosphere FSU/COAPS model (hereafter, PAR). The second simulation is performed without assimilation (hereafter, CONTROL). In other words, CONTROL is a free run. Soil moisture evolves slowly and therefore requires a long spin-up time period to reach equilibrium in the model. Using the Community Land Model Version 3 (CLM3), Du et al. [2006] found that the equilibration state of soil moisture at 1.5 m depth is achieved after at least 20 years. In the coupled land-atmosphere FSU/COAPS model, the total soil moisture depth goes down to 3.4 m. We found that the global average soil moisture of the deepest layer reaches an equilibrium state after 50 years (not shown here). Both simulations thus start after a 50-year spin-up.

5. Results

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. PAR Technique
  5. 3. Soil Moisture Data Sets
  6. 4. Experimental Design
  7. 5. Results
  8. 5.3. Global Land Surface Validation
  9. 6. Conclusion
  10. Acknowledgments
  11. References

5.1. Precipitation Assimilation Validation

[17] Before analyzing the impact of the PAR technique on the model land surface hydrological estimates (in particular soil moisture), we verify whether the coupled land-atmosphere FSU/COAPS model is able to assimilate the GDMF precipitation data set over the land surface. To assess the ability of the model to reproduce observed precipitation patterns regardless of its magnitude, temporal correlations are computed. A commonly used statistical tool for precipitation validation is the Equitable Threat Score (ETS)[Schaefer, 1990]. In contrast with the correlation, the ETS takes the spatial distribution and the magnitude of precipitation into account. A perfect score is equivalent to ETS = 1. In the computation of these two statistical tools, all continental grid points are included for the study period (i.e. 1986–1995). Both the temporal correlation and ETS are computed against GDMF and an observation-based precipitation data set, the GPCC data set.

[18] Figure 1 presents the average continental precipitation over the boreal summer months (June, July, August) and the boreal winter months (December, January, February). For both seasons, considerable differences in the spatial distribution are noted between CONTROL and GDMF. PAR is, not surprisingly, in very good agreement with GDMF. For instance, PAR depicts very well the precipitation maxima belt (South America, Central Africa and Southeast Asia) associated with the seasonal ITCZ path. R2 also does a good job for both seasons in reproducing the spatial distribution and amplitude of GDMF. Figure 2 represents the spatial distribution of monthly precipitation anomaly correlations against two verification data sets, GDMF and GPCC respectively. Since we have found that CONTROL does not provide reliable precipitation estimates, hereafter CONTROL is disregarded. It is clear that PAR obtains the best correlation coefficients (>0.65) across most of the globe with both GDMF and GPCC. Figure 3 shows ETS values computed against GDMF and GPCC respectively. First, one can be surprised that the ETS values of R2 are relatively high (between 0.2 and 0.45 at most thresholds) while low correlation values (<0.4 over most of the globe) were found in Figure 2. This semblance of inconsistency is due the inclusion of all continental grid points in the calculation of the ETS score. Indeed, if a model has a reasonable global precipitation climatology, which is the case for R2, then the ETS may overestimate skill by giving credit to the model for not producing rain deserts. For both seasons, PAR exhibits the highest ETS values (up to 0.52) against both GDMF and GPCC. Thus, the high values of correlation and ETS obtained by the PAR simulation indicate that the coupled land-atmosphere FSU/COAPS model is able to assimilate the GDMF precipitation data and represent very well the precipitation characteristics of an observation-based data set (i.e. GPCC). In addition, we find that the assimilation of GDMF brings the coupled land-atmosphere FSU/COAPS model even closer to GPCC than R2.

image

Figure 1. Spatial distribution of precipitation average (mm/month) over the (a) boreal summer months (Jun, Jul, Aug) and (b) boreal winter months (Dec, Jan, Feb). The study period is Jan 1986 to Dec 1995.

Download figure to PowerPoint

image

Figure 2. Spatial distribution of monthly anomaly correlations of precipitation for (top) R2 and (bottom) PAR against (a) GDMF and (b) GPCC. The study period is Jan 1986 to Dec 1995.

Download figure to PowerPoint

image

Figure 3. Equitable Thread Score (ETS) computed over a range of monthly precipitation thresholds (x-axis) for the (a) boreal summer months against GDMF, (b) boreal summer months against GPCC, (c) boreal winter months (DJF) against GDMF and (d) boreal winter months against GPCC. All land grid points are used to compute the ETS. The study period is Jan 1986 to Dec 1995.

Download figure to PowerPoint

5.2. Local Land Surface Validation

[19] In the previous section, we have verified that by modifying the air humidity vertical profile based on the GDMF data, the PAR technique allows to produce precipitation estimates close to the truth. In the following two sections, the impact of the PAR precipitation estimates on land surface are analyzed.

5.2.1. Soil Moisture

[20] Using in-situ measurements over Illinois, in this section the PAR soil moisture analysis is locally validated by studying its seasonal cycle and temporal anomaly characteristics. The metric used to assess which soil moisture analysis is closer to the observations is the anomaly correlation calculated with in-situ observations at 6 different grid points inside the state of Illinois (Figure 4). The stations inside each model grid cell (dashed boxes) are aggregated.

image

Figure 4. In-situ soil moisture stations (crosses) over Illinois. The dashed lines represent the grid points of the FSU/COAPS model.

Download figure to PowerPoint

5.2.1.1. Climatology

[21] Figure 5 shows the seasonal cycle of the upper 10 cm soil moisture and the deeper soil moisture (10–200 cm) for the period 1986–1995. The grey bars represent the standard deviation among the soil moisture stations. Note that the grid cells 3 and 6 do not possess bars because they correspond to only one station (Figure 4). First, it is clear that in both soil layers CONTROL and PAR have a systematic dry bias. This model error may be attributed to an incorrect partitioning of the evapotranspiration in the CLM2 model (the land surface component of the coupled land-atmosphere FSU/COAPS model) as highlighted by Lawrence et al. [2007]. In this latter study, it was found that several modifications of parameterization, such as increasing transpiration and infiltration and decreasing surface evaporation, greatly reduce the dry soil moisture bias and increase the soil moisture seasonal cycle amplitude. These modifications have been taken into account in the follow-up versions of the community land model that we hope to use in the future.

image

Figure 5. Soil moisture climatology mean of in-situ observations (thick grey), CONTROL (dotted), PAR (dashed) and R2 (semi-dotted) in (a) the top 10 cm layer and (b) the deeper 10–200 cm layer. The 6 locations are in Illinois (see Figure 4). The study period is Jan 1986 to Dec 1995.

Download figure to PowerPoint

[22] In the topsoil layer and at most locations, CONTROL reproduces well the sharp decrease of soil moisture in the summertime but is out of phase in the wintertime (Figure 5a). Despite its weak amplitude and dry bias, PAR best fits the observed soil moisture seasonal cycle. R2 has a nearly constant soil moisture estimate at all locations and therefore fails to capture the observed seasonal cycle in the topsoil layer. This issue may be related to the simplified freezing process used in R2, which assumes that when the air temperature drops to freezing, the precipitation and melted snow cannot infiltrate the soil [Li et al., 2005; Boisserie et al., 2006]. This means that the soil moisture in R2 is not capable of recharging during cold winters. In contrast, R2 follows well the observed seasonal cycle in the deep soil layer (Figure 5b). The deep soil layer of R2 thus lags its topsoil layer. This lag could be due to the combination of two factors: (1) the precipitation forcing applied only at the topsoil moisture layer (0–10 cm); and (2) the large thickness difference between the top (0–10 cm) and the deep soil layer (10–200 cm), which favors the prolongation of the adjustment time of the deep soil moisture layer to the top layer forcing. The behavior of PAR and CONTROL in the deep soil layer is similar to that of the topsoil layer. At all locations, CONTROL is again out of phase in the boreal winter and PAR best fits the observed seasonal cycle.

5.2.1.2. Anomaly

[23] Here, we validate the vertical profile of soil moisture anomalies throughout the soil column down to 2-m depth. Again, CONTROL is disregarded here. Figure 6 displays the vertical profile of the average soil moisture anomalies of in-situ observations, R2 and PAR over Illinois for the time period 1986–1990 and 1991–1995. PAR and the observations are discretized into the same 7 soil layers (0–10, 10–30, 30–50, 50–70, 70–90, 90–110, 110–200 cm), while R2 has only two soil layers (0–10 and 10–200 cm).

image

Figure 6. Vertical profile of soil moisture anomalies averaged over Illinois for the time period (a) 1986–1990 and (b) 1991–1995.

Download figure to PowerPoint

[24] First, it is clear that R2 shows a large discontinuity in soil moisture anomalies between its two layers, which is most likely due to the precipitation forcing applied only at the top layer and the large thickness difference between its two soil moisture layers. Despite this discontinuity, R2 along with PAR reproduces with reasonable accuracy the major dry and wet events occurring in the state of Illinois, such as the 1988 drought and the 1993 flood. However, one can notice that the anomaly amplitude in PAR is most of the time too weak. As previously mentioned, the land surface component of the land-atmosphere FSU/COAPS model has an incorrect evapotranspiration partitioning and produces too much surface evaporation within the canopy. Our speculation is that strong surface evaporation is likely to weaken the soil moisture response to precipitation and hence weaken the amplitude of soil moisture anomalies. The anomaly correlations show that, in both soil layers, PAR has the best correlations with the observations at all 6 grid cells (Figure 7). Therefore, these results indicate that the PAR technique: (1) is able to simulate well the temporal structure of soil moisture anomalies, (2) these anomalies are even closer to the observations than those of R2, but (3) the mean annual cycle and the anomaly magnitude are too weak.

image

Figure 7. Monthly soil moisture anomaly correlation (a) in the top 10 cm layer and (b) in the deep layer 10–200 cm at 6 different locations in Illinois (see Figure 4).

Download figure to PowerPoint

[25] Finally, we evaluate the skill of GSWP-2 against that of PAR and R2. Figure 8 shows the soil moisture seasonal cycle and the temporal anomaly correlation against in-situ observations over Illinois. For comparison purposes, all data sets are scaled to the 1.5 m depth which corresponds to the soil moisture depth of GSWP-2. Although GSWP-2 shows a dry bias, it fits reasonably well the observed seasonal cycle and amplitude (Figure 8a). The high anomaly correlation (0.85) indicates that GSWP-2 reproduces well the observed interannual variability (Figure 8b). PAR also shows a high anomaly correlation (0.78) but remains lower than that of GSWP-2. However, one must keep in mind that the GSWP-2 analysis is the average of 13 LSMs. In general, multimodel ensembles outperform individual models due to the averaging out of random model errors.

image

Figure 8. Climatological (a) mean and (b) temporal anomaly correlation of soil moisture against in-situ observations in the top 1.5 m layer averaged over Illinois. The grey solid line represents the in-situ observations, the dashed line the PAR simulation, the semi-dotted line R2 and the black solid line the GSWP2. The study period is Jan 1986 to Dec 1995.

Download figure to PowerPoint

5.2.2. Land Surface Water Budget

[26] The PAR technique not only affects soil moisture but also other land surface water budget components, such as surface runoff and surface evaporation. The evaluation of the two latter variables is difficult due to a lack of observations. Here, we evaluate these variables against GSWP-2 by comparing the seasonal cycle and the temporal anomaly correlation over Illinois. Since GSWP-2 corresponds to the mean of 13 LSMs, the bars represent the variance among LSMs as a measure of uncertainty of this product. However, one must be cautious when interpreting these bars. A small variance does not necessarily mean that GSWP-2 is close to the observations. However, a large variance can be a good indicator that GSWP-2 is not reliable. A brief evaluation of precipitation estimates against the GDMF (i.e. the assimilated precipitation data) is also shown here.

[27] Figure 9 presents the climatological mean, ± one standard deviation around the climatological mean and the temporal anomaly correlation of precipitation, surface runoff and surface evaporation over Illinois for the study period. For precipitation, PAR agrees, not surprisingly, very well with GDMF and exhibits the best anomaly correlation (0.82). These results simply corroborate the success of the GDMF precipitation assimilation into the coupled land-atmosphere FSU/COAPS model found in section 5.1. R2 also does a good job in following the observed seasonal cycle and obtains a reasonably high anomaly correlation (0.67).

image

Figure 9. (left) The climatological mean, (middle) +/−1 standard deviation around the climatological mean and (right) the anomaly correlation of the (a) precipitation, (b) surface runoff and (c) surface evaporation monthly mean averaged across Illinois (1986–1995.

Download figure to PowerPoint

[28] The comparison of surface runoff estimates shows that all data sets are inconsistent with the GSWP-2 seasonal cycle due to the large scatter of the GSWP-2 models. However, the temporal anomaly correlation of surface runoff is significantly improved in PAR over CONTROL. The PAR anomaly correlation is even higher than that of R2. All data sets have a surface evaporation maximum during the warm season. The summer peak of PAR and R2 occurring in July best matches with that of GSWP-2. The relatively small bars indicate that GSWP-2 surface evaporation estimate is most likely reliable. As far as the temporal anomaly correlations are concerned, again PAR shows a great improvement over CONTROL and outperforms R2.

[29] To conclude, using the PAR technique, the coupled land-atmosphere FSU/COAPS model reproduces with reasonable accuracy not only the observed soil moisture characteristics but also those of surface evaporation. In addition, PAR shows most of the time better results than R2 which indicates that the PAR technique may be more efficient in producing realistic soil moisture initial conditions than the land assimilation system used in R2. However, a local validation over Illinois is not sufficient to fully validate the global soil moisture analysis.

5.3. Global Land Surface Validation

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. PAR Technique
  5. 3. Soil Moisture Data Sets
  6. 4. Experimental Design
  7. 5. Results
  8. 5.3. Global Land Surface Validation
  9. 6. Conclusion
  10. Acknowledgments
  11. References

[30] Here, we evaluate the effect of the PAR technique over the globe using GSWP-2 analysis (described in section 4.1.1). For this purpose, the temporal anomaly correlations against the GSWP-2 analysis is analyzed. Since the GSWP-2 analysis possesses only one soil moisture layer (1.5 m depth), the soil moisture layers of PAR are combined to match that of GSWP-2.

[31] Figure 10 shows the spatial distribution of temporal anomaly correlations with respect to GSWP-2 for each land surface hydrological estimates (soil moisture, surface runoff and surface evaporation). Between −0.4 and 0.4, the temporal anomaly correlation is not statistically significant at a 95% confidence interval. For soil moisture, PAR correlates slightly better than R2 in most regions (South America, Africa, Southeast Asia and Australia), except in the very high latitudes of the Northern Hemisphere. For surface runoff, PAR shows clearly higher anomaly correlations in most regions than R2. The temporal anomaly correlation of surface evaporation is very high across the land surface for both PAR and R2, except over the tropical rain forests (the Amazon and Central Africa).

image

Figure 10. Spatial distribution of monthly anomaly correlations of (a) soil moisture, (b) surface runoff and (c) evaporation between GSWP2 and (top) R2 and (bottom) PAR.

Download figure to PowerPoint

[32] Therefore, we find that the soil moisture analysis produced in this study compares well with a benchmark in global soil moisture analysis that is GSWP-2 across most of the globe. In addition, not only soil moisture but also surface runoff and surface evaporation estimates are comparable with GSWP-2.

6. Conclusion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. PAR Technique
  5. 3. Soil Moisture Data Sets
  6. 4. Experimental Design
  7. 5. Results
  8. 5.3. Global Land Surface Validation
  9. 6. Conclusion
  10. Acknowledgments
  11. References

[33] In this study, we evaluate the impact of the Precipitation Assimilation Reanalysis (PAR) technique on model land surface estimates implemented into the coupled land-atmosphere FSU/COAPS model. This evaluation focuses mostly on soil moisture analysis since this variable is thought to be a key variable for improving subseasonal forecasts of near surface atmospheric variables. During both the boreal summertime and wintertime, very high temporal correlations (>0.8 in most regions) and ETS values (up to 0.52) obtained between PAR and GDMF shows the ability of the coupled land-atmosphere FSU/COAPS model to assimilate the GDMF precipitation data set. PAR also is very well correlated with an other observation-based precipitation (i.e. GPCC) and shows better results than R2. Then, using in-situ soil observations of soil moisture and the GSWP-2 multimodel analysis, despite the dry bias in soil moisture, the PAR technique has overall a positive impact on land surface estimates. Over Illinois, we find that the seasonal cycle and the temporal anomaly variability of the soil moisture analysis produced in this study match fairly well with in-situ observations deep into the soil. The PAR soil moisture analysis is also better correlated with in-situ observations from the Illinois state than R2. This may suggest that a physically consistent land assimilation technique is important for generating accurate soil moisture initial conditions. The PAR technique not only improves reasonably well the soil moisture estimates but also other hydrological land surface estimates, such as surface runoff and surface evaporation. The comparison with a proxy for global land surface observations (i.e. GSWP-2 analysis) suggests that the PAR technique is effective not only over Illinois but across the globe. During the boreal summertime and wintertime, the spatial distribution and amplitude of soil moisture, surface runoff and surface evaporation estimates match fairly well with GSWP-2. The soil moisture anomalies in PAR are slightly better correlated with GSWP-2 than R2. However, one should be cautious in interpreting those results. Although GSWP-2 offers the best proxy for land surface observations because it represents the average of 13 state-of-the-art LSMs, its accuracy is not fully known. For instance, GSWP-2 is not most likely to give an accurate estimate when there is a high variability among these LSMs, such as the case for the GSWP-2 surface runoff estimate.

[34] To conclude, we have developed a new soil moisture analysis product that: (1) is physically consistent with the atmospheric physics of the coupled land-atmosphere FSU/COAPS model, (2) better compares with observations over Illinois than R2, and (3) is comparable to a benchmark in soil moisture analysis (i.e. GSWP-2). Therefore, after applying a bias correction, this product can be used in many land surface studies, such as crop modeling, detection of extreme events (drought and flood), water management. In Part II, this data set is used to investigate the contribution of a soil moisture initialization on short-term to subseasonal forecasting skill of near-surface variables. The results of Part II are included in an ongoing international model intercomparison project named GLACE-2.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. PAR Technique
  5. 3. Soil Moisture Data Sets
  6. 4. Experimental Design
  7. 5. Results
  8. 5.3. Global Land Surface Validation
  9. 6. Conclusion
  10. Acknowledgments
  11. References

[35] This work was supported by the Applied Research Center, funded by NOAA Office of Global Programs, awarded to James J. O'Brien. The numerical experiments were conducted on the FSU/HPC supercomputer. NCEP/DOE Reanalysis 2 data were obtained at http://www.nomad2.ncep.noaa.gov/. The GSWP-2 analysis can be found at http://www.iges.org/gswp/. We also thank the Global soil Moisture Data Bank for providing us with the Illinois Climate Network of soil moisture.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. PAR Technique
  5. 3. Soil Moisture Data Sets
  6. 4. Experimental Design
  7. 5. Results
  8. 5.3. Global Land Surface Validation
  9. 6. Conclusion
  10. Acknowledgments
  11. References
  • Boisserie, M., D. W. Shin, T. E. LaRow, and S. Cocke (2006), Evaluation of Soil Moisture in the Florida State University climate model–National Center for Atmospheric Research community land model (FSU-CLM) using two reanalyses (R2 and ERA40) and in situ observation, J. Geophys. Res., 111, D08103, doi:10.1029/2005JD006446.
  • Bonan, G. B., K. W. Oleson, M. Vertenstein, S. Levis, X. Zeng, Y. Dai, R. E. Dickinson, and Z.-L. Yang (2002), The land surface climatology of the Community Land Model coupled to the NCAR Community Climate Model, J. Clim., 15, 31233149.
  • Cocke, S., and T. E. LaRow (2000), Seasonal prediction using a regional spectral model embedded within a coupled ocean-atmosphere model, Mon. Weather Rev., 128, 689708.
  • Dirmeyer, P. A., X. Gao, and T. Oki (2002), The Second Global Soil Wetness Project, GSWP-2, IGPO Publ. 37, 65 pp., IGPO, Silver Spring, Md. [Available at http://www.iges.org/gswp/publications.html.]
  • Dirmeyer, P. A., X. Gao, M. Zha, Z. Guo, T. Oki, and N. Hanasaki (2005), The second Global Soil Wetness Project (GSWP-2): Multi-model analysis and implications for our perception of the land surface, COLA Tech. Rep. 185, 46 pp., IGES, Calverton, Md.
  • Du, C., W. Wu, Z. Liu, and W. Gao (2006), Simulation of soil moisture and its variability in East Asia, in Remote Sensing and Modeling of Ecosystems for Sustainability III, Proc. SPIE Int. Soc. Opt. Eng., 6298, F-1F-7.
  • Fennessy, M., and J. Shukla (1999), Impact of initial soil wetness on seasonal atmospheric predictions, J. Clim., 12, 31673180.
  • Hollinger, S. E., and S. A. Isard (1994), A soil moisture climatology of Illinois, J. Clim., 7, 822833.
  • Hong, S.-Y., and E. Kalnay (2000), Role of sea-surface temperature and soil-moisture feedback in the 1998 Oklahoma-Texas drought, Nature, 408, 842844.
  • Huffman, G. J., R. F. Adler, M. Morrissey, D. T. Bolvin, S. Curtis, R. Joyce, B. McGavock, and J. Susskind (2001), Global precipitation at one-degree daily resolution from multi-satellite observations, J. Hydrometeorol., 2, 3650.
  • Huffman, G. J., R. F. Adler, D. T. Bolvin, G. Gu, E. J. Nelkin, K. P. Bowman, E. F. Stocker, and D. B. Wolff (2007), The TRMM multi-satellite precipitation analysis: Quasi-global, multi-year, combined-sensor precipitation estimates at fine scale, J. Hydrometeorol., 8, 3855.
  • Kalnay, E., et al. (1996), The NCEP/NCAR 40-Year Reanalysis Project, Bull. Am. Meteorol. Soc., 77, 437471.
  • Kanamitsu, M., W. Ebisuzaki, J. Woollen, S. K. Yang, J. J. Hnilo, M. Fiorino, and G. L. Potter (2002), NCEP-DOE AMIP-II Reanalysis (R2), Bull. Am. Meteorol. Soc., 83, 16311643.
  • Kanamitsu, M., C.-H. Lu, J. Schemm, and W. Ebisuzaki (2003), The predictability of soil moisture and near surface temperature in hindcasts of NCEP Seasonal Forecast Model, J. Clim., 16, 510521.
  • Kim, B.-M., S. Cocke, G.-H. Lim, and J.-H. Oh (2007), Assimilation of TRMM rain rate into global analysis and its impact on the summer mean circulation over tropics, J. Korean Meteorol. Soc., 43(4), 397409.
  • Koster, R. D., and M. J. Suarez (2000), Soil moisture memory in climate models, J. Hydrometeorol., 2, 558570.
  • Koster, R. D., M. J. Suarez, and M. Heiser (2000), Variance and predictability of precipitation at seasonal-to-interannual timescales, J. Hydrometeorol., 1, 2646.
  • Koster, R. D., et al. (2004), Regions of strong coupling between soil moisture and precipitation, Science, 305, 11381140.
  • Koster, R. D., et al. (2011), The second phase of the Global Land-B-Atmosphere Coupling Experiment: Soil moisture contributions to subseasonal forecast skill, J. Hydrometeorol., 12, 805822, doi:10.1175/2011JHM1365.1.
  • Krishnamurti, T., J. Xue, H. S. Bedi, K. Ingles, and D. Oosterhof (1991), Physical initialization for numerical weather prediction over the tropics, Tellus, Ser. A/B, 43, 5381.
  • Kumar, A., and M. P. Hoerling (1995), Prospects and limitations of seasonal atmospheric GCM predictions, Bull. Am. Meteorol. Soc., 76, 335345.
  • Lawrence, D. M., P. E. Thornton, K. W. Oleson, and G. B. Bonan (2007), The partitioning of evapotranspiration into transpiration, soil evaporation, and canopy evaporation in a GCM: Impacts on land-atmosphere interaction, J. Hydrometeorol., 8, 862880.
  • Li, H., A. Robock, S. Liu, X. Mo, and P. Viterbo (2005), Evaluation of reanalysis soil moisture simulations using updated Chinese soil moisture observations, J. Hydrometeorol., 6, 180191.
  • Nunes, A. M., and S. Cocke (2004), Implementing a physical initialization procedure in a regional spectral model: Impact on the short-range rainfall forecasting over South America, Tellus, Ser. B, 56(2), 125140.
  • Pan, H.-L. (1990), A simple parameterization scheme of evapotranspiration over land for the NMC medium-range forecast model, Mon. Weather Rev., 118, 25002512.
  • Pan, H.-L., and L. Mahrt (1987), Interaction between soil hydrology and boundary-layer development, Boundary Layer Meteorol., 38, 185202.
  • Reynolds, R. W., N. A. Rayner, T. M. Smith, D. C. Stokes, and W. Wang (2002), An improved in situ and satellite SST analysis for climate, J. Clim., 15, 16091625.
  • Rind, D. (1982), The influence of ground moisture conditions in North America on summer climate as modeled in the GISS GCM, Mon. Weather Rev., 110, 14871494.
  • Robock, A., K. Y. Vinnikov, G. Srinivasan, J. K. Entin, S. E. Hollinger, N. A. Speranskaya, S. Liu, and A. Namkhai (2000), The Global Soil Moisture Data Bank, Bull. Am. Meteorol. Soc., 81, 12811299.
  • Schaefer, J. T. (1990), The critical success index as an indicator of warning skill, Weather Forecast., 5, 570575.
  • Sheffield, J., G. Goteti, and E. F. Wood (2006), Development of a 50-year high-resolution global dataset of meteorological forcings for land surface modeling, J. Clim., 19(13), 30883111.
  • Shin, D. W., S. Cocke, T. E. LaRow, and J. J. O'Brien (2005), Seasonal surface air temperature and precipitation in the FSU climate model coupled to the CLM2, J. Clim., 18, 32173228.
  • Shukla, J. (1998), Predictability in the midst of chaos: A scientific basis for climate forecasting, Science, 282, 728731.
  • Shukla, J., and Y. Mintz (1982), Influence of land-surface evapotranspiration on the Earth's climate, Science, 215, 14981500.
  • Sud, Y. C., and M. Fennessy (1984), Influence of evaporation in semi-arid regions on the July circulation: A numerical study, J. Climatol., 4, 383398.
  • Trenberth, K. E., and G. W. Branstator (1992), Issues in establishing causes of the 1988 drought over North America, J. Clim., 5, 159172.
  • vanden Hurk, B., F. Doblas-Reyes, G. Balsamo, R. D. Koster, S. I. Seneviratne, and H. Camargo Jr. (2012), Soil moisture effects on seasonal temperature and precipitation forecast scores in Europe, Clim. Dyn., 1–2, 349362, doi:10.1007/s00382-010-0956-2.
  • Yeh, T. C., R. T. Wetherald, and S. Manabe (1984), The effect of soil moisture on the short-term climate and hydrology change: A numerical experiment, Mon. Weather Rev., 112, 474490.