Atmospheric water balance over oceanic regions as estimated from satellite, merged, and reanalysis data

Authors


Abstract

[1] The column integrated atmospheric water balance over the ocean was examined using satellite-based and merged data sets for the period from 2000 to 2005. The data sets for the components of the atmospheric water balance include evaporation from the HOAPS, GSSTF, and OAFlux and precipitation from the HOAPS, CMAP, and GPCP. The water vapor tendency was derived from water vapor data of HOAPS. The product for water vapor flux convergence estimated using satellite observation data was used. The atmospheric balance components from the MERRA reanalysis data were also examined. Residuals of the atmospheric water balance equation were estimated using nine possible combinations of the data sets over the ocean between 60°N and 60°S. The results showed that there was considerable disagreement in the residual intensities and distributions from the different combinations of the data sets. In particular, the residuals in the estimations of the satellite-based atmospheric budget appear to be large over the oceanic areas with heavy precipitation such as the intertropical convergence zone, South Pacific convergence zone, and monsoon regions. The lack of closure of the atmospheric water cycle may be attributed to the uncertainties in the data sets and approximations in the atmospheric water balance equation. Meanwhile, the anomalies of the residuals from the nine combinations of the data sets are in good agreement with their variability patterns. These results suggest that significant consideration is needed when applying the data sets of water budget components to quantitative water budget studies, while climate variability analysis based on the residuals may produce similar results.

1 Introduction

[2] The atmospheric water cycle is one of the most important components of the global water cycle. Large amounts of water vapor that are evaporated from the ocean are transported to the continents through the atmosphere. The transported water vapor is converted into precipitation that provides vital water for living things on Earth. Precipitation and evaporation over the oceans change the sea surface salinity and help to drive the ocean thermohaline circulation. Changes in the phase of water in the atmosphere involve latent heat exchanges. Latent heat released by condensation is one of the major energy sources driving the general circulation of the atmosphere. Knowledge of the atmospheric water cycle is therefore essential in order to manage water resources and to understand the Earth's weather and climate.

[3] The atmospheric water cycle has been investigated in many regions. The Global Energy and Water Cycle Experiment (GEWEX) initiated by the World Climate Research Program is well known for its scientific studies of the water cycle. The GEWEX aims at observing, understanding, and modeling the hydrological cycle and energy fluxes in order to predict global and regional climate change. Under the missions of the GEWEX, projects including the Regional Hydroclimate Projects (RHPs) (formerly the Continental Scale Experiment) were initiated. The focus of the GEWEX was to solve the problems of closing the balance of water and energy. In addition, several water cycle studies were performed at the atmospheric branch in order to quantify the regional water balance, including the Mackenzie GEWEX Studies [e.g., Stewart et al., 1998 ; Rouse et al., 2003], the Baltic Sea Experiment [e.g., Raschke et al., 2001 ; Ruprecht and Kahl, 2003], the Climate Prediction Program for the Americas [e.g., Roads et al., 2003], and the Murray-Darling Basin [e.g., Draper and Mills, 2008].

[4] One of the major complications of the atmospheric water balance study is the selection of appropriate data sets. The atmospheric water balance equation includes water cycle components such as evaporation, precipitation, water flux convergence, and water tendency. The equation itself is not complicated, but accurate estimation of each of the water cycle components is difficult. Water cycle components have been acquired from direct measurements, numerical weather prediction (NWP) model-derived products, remotely sensed observations, and residual methods using the atmospheric water balance equation. Direct measurements provide relatively reliable data but have limited observation coverage. Earlier water cycle studies were usually performed using radiosonde data [Rasmusson, 1967, 1968]. Despite its limited spatial and temporal coverage, radiosonde data is still used over the regions with dense radiosonde networks [Kanamaru and Salvucci, 2003; Zangvil et al., 2004]. NWP-derived analysis and reanalysis data have been actively employed for the study of the water budget [e.g., Roads et al., 2003; Ruprecht and Kahl, 2003; Turato et al., 2004; Draper and Mills, 2008], because of their ability to minimize known errors and high spatiotemporal resolutions. NWP products, however, are highly dependent on model physics and parameterization. With advancement in satellite instruments and retrieval algorithms, more satellite observation data have become available [e.g., Bakan et al., 2000]. Satellite observations, which have better coverage than in situ observations particularly over the oceans, have been widely used for a complementary purpose or for comparison with other observations. Many satellite-based or merged data sets for water cycle components have been produced using such satellite observations.

[5] In this study, the atmospheric water balance is examined over oceans using various satellite-based and merged data sets. Reanalysis data sets are also used for comparison with satellite-based data sets for the atmospheric water balances over oceanic regions.

2 Data and Methodology

2.1 Atmospheric Water Balance Over the Ocean

[6] The column integrated atmospheric water balance over the ocean can be expressed as follows [Peixoto and Oort, 1992]:

display math(1)

where E and P are the evaporation and precipitation, respectively. W indicates the column integrated total water vapor, math formula is the horizontal water vapor flux vector, and the subscript c denotes the condensed phase of water. W + Wc and math formula are defined as follows:

display math(2)

and

display math(3)

where g is the acceleration of gravity, ps is the pressure at the surface, p0 is the pressure at the top of the atmosphere, q is the specific humidity, qc is the condensed water mixing ratio, and math formula is the horizontal wind velocity vector.

[7] The terms ∂ Wc/∂ t and math formula are related to the condensed phase water and are usually small enough that both the tendency of liquid and solid water in clouds and their horizontal flux can be ignored in equation (1). Therefore, by averaging equation (1) in the specified temporal and spatial domain of interest, the general atmospheric water balance equation can be simplified to the following:

display math(4)

where the bar indicates the time average and the angular bracket denotes the area average. Equation (4) shows that the excess of evaporation over precipitation is balanced by the local rate change of the water vapor contents and by the horizontal net flux of water vapor.

[8] In order to examine the atmospheric water balance, we calculated the residuals using equation (4) as follows:

display math(5)

where R represents the residual of the atmospheric water balance equation. R was obtained from combinations of the data sets for the components of the water cycle in equation (5). Section 2.2 describes the data sets used for this study.

2.2 Data Sets

[9] The satellite-based and merged data sets for atmospheric water cycle components that are widely used for water cycle study were selected. Table 1 summarizes the characteristics of the data sets for the water cycle components used in this study. For evaporation (E), the monthly fields of three data sets were used. The first data set was the Hamburg Ocean Atmosphere Parameters and Fluxes from Satellite data (HOAPS), and the second data set was the Goddard Satellite-based Surface Turbulent Fluxes (GSSTF). The third data set was the Objectively Analyzed Air-Sea Heat Fluxes (OAFlux) project at the Woods Hole Oceanographic Institution. In HOAPS, evaporation data were obtained from all of the available Special Sensor Microwave Imager (SSM/I) observation data based on the Coupled Ocean-atmosphere Response Experiment (COARE) bulk flux algorithm version 2.6a [Bradley et al., 2000; Fairall et al., 1996] for latent heat flux parameterization. The near-surface wind speed and specific air humidity were retrieved from the SSM/I brightness temperature, and the sea surface temperature data from the Advanced Very High Resolution Radiometer (AVHRR) measurements were used [Andersson et al., 2010]. The GSSTF latent heat flux was retrieved based on the bulk flux model [Chou, 1993], using the SSM/I surface wind, surface air humidity, and near-surface air and sea surface temperatures from the National Centers for Environmental Prediction (NCEP)-National Center for Atmospheric Research (NCAR) reanalysis as the input data [Chou et al., 2003; Shie, 2010]. The OAFlux evaporation data were derived from the blending of the satellite retrievals from the SSM/I, the Quick Scatterometer (QuikSCAT), the AVHRR, the Tropical Rain Measuring Mission Microwave Imager, and the Advanced Microwave Scanning Radiometer Earth Observing System, as well as the NWP reanalysis outputs from the NCEP (e.g., NCEP/NCAR, NCEP/DOE reanalysis) and European Centre for Medium-Range Weather Forecasts (ECMWF) (e.g., ERA40) by objective analysis techniques [Yu and Weller, 2007; Yu et al., 2008]. The OAFlux products were constructed based on the COARE bulk flux algorithm version 3.0 [Fairall et al., 2003].

Table 1. Characteristics of the Data Sets for the Atmospheric Water Balance Componentsa
Data SetsVersionVariablesSpatial ResolutionTemporal ResolutionAvailable Period
  1. a

    The components are evaporation (E), precipitation (P), water vapor flux convergence (WVFC), and water vapor tendency (WVT).

OAFluxv3E1.0° × 1.0°month1958–2010
GSSTFv2cE1.0° × 1.0°month1987–2008
GPCPv2.1P2.5° × 2.5°month1979–2010
CMAPv1001P2.5° × 2.5°month1979–2009
HOAPSv3E, P, WVT0.5° × 0.5°month1987–2005
XLT08v3WVFC0.5° × 0.5°month1999–2008
MERRA(GEOS-5.2.0)E, P, WVT, WVFC0.5° × 2/3°month1979–2012

[10] The following three monthly precipitation (P) products were used: HOAPS, the Global Precipitation Climatology Project (GPCP), and the Climate Prediction Center Merged Analysis of Precipitation (CMAP). The HOAPS precipitation was retrieved from a neural network algorithm that takes the SSM/I brightness temperature and the precipitation from the ECMWF model as training data [Andersson et al., 2010]. Both the GPCP [Adler et al., 2003; Huffman et al., 2009] and the CMAP [Xie and Arkin, 1997] use multiple satellite and rain gauge data sets with some differences in their input data and merging techniques [Yin et al., 2004].

[11] For water vapor flux convergence (WVFC), the monthly WVFC estimated by Xie et al. [2008, hereinafter referred to as XLT08] was used. The XLT08 algorithm estimates the WVFC based on support vector regression using the surface wind vector from the Quick Scatterometer (QuikSCAT), the cloud drift wind vector from the Multiangle Imaging Spectroradiometer and the NOAA geostationary satellites, and the precipitable water from the SSM/I.

[12] The water vapor tendency (WVT) data were derived in this study using the HOAPS twice-daily total column water vapor (TPW) data. The HOAPS twice-daily TPW were averaged over a pentad of days in order to create daily data (e.g., the 1 January 2005 daily TPW is the averaged value of the 30 December 2004 through 3 January 2005 twice-daily TPW). The monthly WVT was then calculated using the following equation:

display math(6)

where ∆t is the 1 month time interval, and i and f indicate the first and the day of the mth month, respectively.

[13] Modern-Era Retrospective analysis for Research and Applications (MERRA), which is NASA's new reanalysis method, was also used for comparison with the satellite-based and merged data sets. The MERRA is generated using the Goddard Earth Observing System (GEOS) atmospheric model and data assimilation system, version 5.2.0 [Rienecker et al., 2011]. MERRA provides the various quantities for the atmospheric water cycle components. It shows improved aspect for representing the atmospheric branch of the hydrological cycle and provides complete information for budget studies as stated by Rienecker et al. [2011].

[14] Figure 1 is a diagram of the estimated residuals from nine possible combinations of satellite and merged data sets for water cycle components. For example, the residual OGXH indicated that the residual came from the combination of evaporation for OAFlux, precipitation for GPCP, water vapor flux convergence for XLT08, and water vapor tendency for HOAPS. All of the monthly data sets were remapped to have a 5° by 5° latitude-longitude spatial resolution for the period from 2000 to 2005. For each data set, we averaged the data values at each 5° × 5° grid if the number of the missing value did not exceed 50% of the total number of data points. The domain is limited to oceanic regions between 60°N and 60°S. Oceanic regions are defined as the areas where the MERRA's ocean fraction is greater than 0.9.

Figure 1.

A diagram of the nine residuals estimated from possible combinations of the data sets for the atmospheric water cycle components.

3 Results

3.1 Analyses of Atmospheric Water Cycle Components

3.1.1 Evaporation and Precipitation

[15] The time series of the monthly domain averages of the satellite-based and merged data sets for four atmospheric water cycle components during the period from 2000 to 2005 are shown in Figure 2. The reanalysis data MERRA is also displayed for comparison. For evaporation (Figure 2a), the domain averages of the evaporation from GSSTF, HOAPS, OAFlux, and MERRA are 3.88, 3.82, 3.45, and 3.50 mm d−1, respectively. The monthly mean values of evaporation from the satellite-based GSSTF and HOAPS are typically greater than those from the merged and reanalysis data sets. HOAPS and GSSTF are simply paired due to the similarity in using the same satellite data from SSM/I, but their estimation methods are not similar to each other [Chiu et al., 2012]. Correlation coefficients between the evaporation data sets have been also investigated (Table 2). Relatively high correlation coefficients were found between the OAFlux and MERRA data (0.72) and the GSSTF and HOAPS data (0.69), but the correlation coefficients between the HOAPS and MERRA data (0.38) and between the GSSTF and MERRA data (0.29) were much weaker.

Figure 2.

Time series of the monthly domain averaged (2) evaporation rates, (b) precipitation rates, (c) water vapor flux convergences, and (d) water vapor tendencies for the period from 2000 to 2005. Units are mm d−1.

Table 2. Correlation Coefficients for Precipitation Data Sets (the Upper Triangular Part) and for Evaporation Data Sets (the Lower Triangular Part)
 GPCPCMAPHOAPSMERRAPrecipitation
OAFlux1.001.000.880.860.42GPCP
HOAPS0.501.001.000.790.51CMAP
GSSTF0.580.691.001.000.52HOAPS
MERRA0.720.380.291.001.00MERRA
EvaporationOAFluxHOAPSGSSTFMERRA  

[16] The time series of the four precipitation data sets are shown in Figure 2b. The period-means of precipitation for CMAP, GPCP, HOAPS, and MERRA are 3.14, 3.01, 2.91, and 3.23 mm d−1, respectively. The correlations between the satellite-based and merged data sets are generally higher than those between the reanalysis MERRA and the other data sets (Table 2). In particular, relatively high correlations are found between the GPCP and CMAP (0.88), GPCP and HOAPS (0.86), and HOAPS and CMAP (0.79) data sets.

[17] The local variances associated with the various data sets can be obtained using the following equations:

display math(7a)
display math(7b)

where σi is the standard deviation of N different data sets of the variable X for the ith month and σm is the mean standard deviation over M months. E(X)i is the average of the X's estimated from N data sets for the ith month. For three different evaporation (GSSTF, HOAPS, and OAFlux) and precipitation (GPCP, CMAP, and HOAPS) data sets, the spatial distributions of σm were computed for each 5° × 5° grid over the 72 month period between 2000 and 2005 (Figures 3a and 3c). Large variances between evaporation data sets exist over some of the oceanic dry regions such as the southeastern Pacific, the subtropics over the west Pacific, and the parts of the south Atlantic near South Africa. For precipitation, significant variances were found over the regions of the intertropical convergence zone (ITCZ) and the south Pacific convergence zone (SPCZ) as well as the East China Sea, South China Sea, and Bay of Bengal portions of the Indian Ocean. The difference in the data sets for precipitation was significantly lower over the southeastern Pacific Ocean and the Atlantic Ocean. The regions with discrepancies between the data sets for precipitation were distributed over a broader area than those for evaporation. The range of σm for precipitation is from 0.05 to 1.65 mm d−1, and the range is from 0.23 to 1.23 mm d−1 for evaporation. Coefficient of variation (CV) which is defined as the ratio of the standard deviation to the average (i.e., σi/E(Xi)) was also computed. The spatial distributions of period mean CVs for evaporation and precipitation data sets are shown in Figures 3b and 3d. Compared with the precipitation, the CVs for evaporation are relatively homogeneous over most areas of the ocean. For precipitation, the CVs are relatively large over oceanic dry regions.

Figure 3.

Spatial distributions of the period-mean standard deviations and the coefficients of variation (CVs) between (a, b) three evaporation data sets (OAFlux, HOAPS, and GSSTF) and (c, d) three precipitation data sets (GPCP, CMAP, and HOAPS). The values in the upper right corners of each panel indicate domain averages. The minimum and maximum are also indicated in parentheses.

3.1.2 Water Vapor Flux Convergence and Water Vapor Tendency

[18] Both of the domain mean time series of the XLT08 and MERRA for WVFC (Figure 2c) indicated that the water vapor flux generally diverges over oceanic areas. Their period mean values were negative (−0.46 mm d−1 for XLT08 and −0.55 mm d−1 for MERRA). For the period between 2000 and 2005, the WVFC of the XLT08 was typically larger than that of the MERRA. The temporal correlation coefficient between the XLT08 and MERRA was 0.76. The period mean seasonal spatial distributions of the WVFCs are illustrated in Figure 4. The red colored areas (positive) indicate water vapor flux convergence, and the blue colored areas (negative) indicate water vapor flux divergence. The major patterns of the WVFCs and their seasonal variations were well matched with those of the precipitation (not shown). We also noted that the XLT08 had relatively higher values than the MERRA over the ITCZ and SPCZ.

Figure 4.

Mean seasonal distributions of WVFC for XLT08 (a) in winter (December–January–February (DJF)) and (c) in summer (June–July–August (JJA)) and the mean seasonal distributions for MERRA (b) in winter and (d) in summer for the period from 2000 to 2005.

[19] The WVT derived in this study was compared with the WVT from the MERRA. The WVT of the MERRA includes the analysis increment tendency of water vapor that is a nonphysically added value during the assimilation process in order to adjust it to the observation data. The domain averaged time series showed that both of the WVTs are in good agreement (Figure 2d) and have a strong correlation (0.90). The domain average estimates of the WVTs were significantly smaller than those of the other components of the water cycle. The mean seasonal and spatial distributions of the WVT shown in Figure 5 reveal that there is a significant seasonal variation between the Northern and Southern Hemispheres. The distinct negative water vapor tendencies are present in the Northern Hemisphere during the winter (from December to February), and apparent positive water vapor tendencies exist in the Northern Hemisphere during the summer (from June to August). Both of the WVTs have relatively strong positive tendencies over the oceanic regions around East Asia and the northeastern Pacific near Mexico during the boreal summer (Figures 5c and 5d) and strong negative tendencies over the Bay of Bengal during the boreal winter (Figures 5a and 5b).

Figure 5.

Mean seasonal distributions of WVT for HOAPS (a) in winter (DJF) and (c) in summer (JJA) and mean seasonal distributions for MERRA (b) in winter and (d) in summer for the period from 2000 to 2005.

3.2 Residual Analysis

3.2.1 Domain Averaged Residual Time Series

[20] The nine residuals estimated from the combinations of satellite-based and merged data sets for atmospheric water cycle components were analyzed by taking domain averages (Figure 6). The monthly time series of the domain averaged residuals (Figure 6a) showed that there is a distinct feature between the residuals in combination with the merged evaporation of the OAFlux and the residuals in combination with the satellite-based evaporation from HOAPS and GSSTF. OAFlux included residuals that fluctuated between small positive and negative values near zero. The period mean values of the OGXH, OCXH, and OHXH residuals were −0.01, −0.14, and 0.08 mm d−1, respectively. The residuals from the HOAPS and GSSTF had relatively high positive values between 0.23 and 0.51 mm d−1.

Figure 6.

Time series of nine monthly domain averaged (a) residuals, (b) residual anomalies, and (c) the averages of the nine residuals and their standard deviations for the period from 2000 to 2005. Residuals from MERRA are also illustrated for comparison.

[21] The residuals of the MERRA were also analyzed in order to make a comparison. MERRA provides closed atmospheric water budget including two unphysical terms [Bosilovich et al., 2011]. One of the terms is the analysis increment of water vapor (ANA) that makes model predicted state variable closer to observations. The other term was a negative filling term (F) in order to ensure positive water vapor content. Since the value of F is small enough to neglect, the MERRA residual can be obtained using the following equation:

display math(8)

[22] ANA reflects the observation effect on the analysis, and it is an important feature of the MERRA system. The budget of water and energy cycles in the model can be studied and evaluated with this quantified term [Robertson et al., 2011; Roberts et al., 2012]. Before defining this term as analysis increment term, several studies considered this quantity a residual term [Roads and Betts, 2000; Roads et al., 2002; Roads et al., 2003]. In order to determine the effects of ANA on the atmospheric water budget in MERRA, we also took into consideration the residual calculated by excluding ANA in equation (8), and we called this residual MERRA (exANA). While the MERRA residual from equation (8) is close to 0, the MERRA (exANA) residual has relatively large negative values (Figure 6a). The period mean values of the MERRA and MERRA (exANA) residuals are 0.01 and −0.29 mm d−1, respectively. The anomalies with respect to the long-term mean of each time series for the nine estimated residuals seem to be in good agreement with each other (Figure 6b). Relatively high correlations exist between the residuals from identical evaporation and precipitation data sets except between OGXH-GHXH (0.80). The highest correlation can be found between GCXH-GGXH (0.92). The correlations between the nine estimated residuals are stronger than the correlations with the MERRA residual except for OHXH-HCXH (0.25) and OHXH-GCXH (0.29). The MERRA (exANA) residual anomaly time series seems to have no relationship with the others. The averaged values of the nine estimated residuals and their standard deviations (error bars) are also shown in Figure 6c.

[23] The period-domain means for each atmospheric water budget components and associated residuals are also summarized in Table 3. The relative magnitudes of the residuals compared to corresponding evaporation and precipitation are also calculated in percentage and denoted as R/E and R/P in Table 3. For period-domain mean fields, the relative magnitude of OGXH residual is smaller than other estimated residuals except MERRA residual. The magnitude of OGXH residual is 0.41% of evaporation from OAFlux and 0.47% of precipitation from GPCP.

Table 3. Period-Domain Mean Values for Each Component of the Atmospheric Water Cycle and the Corresponding Residuala
 EPWVFCWVTResidualsR/E (%)R/P (%)
  1. a

    Units are mm d−1. Average of the satellite-based and merged data sets and their standard deviations are denoted as AVGsat and STDsat, respectively. AVGall and STDall also indicate average and standard deviations for all the data sets including the MERRA data set. The ratios of the residual to the corresponding evaporation and precipitation are denoted as R/E and R/P in percentage.

  2. b

    MERRA residual including water vapor tendency analysis increment (ANA) term. The value of ANA term is 0.291.

  3. c

    MERRA residual excluding ANA term (i.e., MERRA (exANA)).

OGXH3.4503.008−0.456−0.000−0.0140.410.47
OCXH3.4503.135−0.456−0.000−0.1414.094.50
OHXH3.4502.910−0.456−0.0000.0842.432.89
HGXH3.8223.008−0.456−0.0000.3589.3711.90
HCXH3.8223.135−0.456−0.0000.2316.047.37
HHXH3.8222.910−0.456−0.0000.45611.9315.67
GGXH3.8773.008−0.456−0.0000.41310.6513.73
GCXH3.8773.135−0.456−0.0000.2867.389.12
GHXH3.8772.910−0.456−0.0000.51113.1817.56
MERRA3.4993.235−0.550−0.000b0.005 c0.2860.140.15
8.178.84
AVGsat3.7163.018−0.456−0.0000.243 (0.224)6,548.05
(STDsat)(0.232)(0.113)
AVGall (STDall)3.662 (0.219)3.072 (0.142)−0.503−0.0000.219 (0.224)5.987.13

3.2.2 Period Mean Residual Distribution

[24] The nine estimated residuals show disagreements in their intensities and patterns for the period mean spatial distributions between 2000 and 2005 (Figure 7). Positive residuals (from green to red colored) indicate that the magnitudes of the source terms (E, WVFC) for water vapor in respect to the atmosphere are larger than the magnitudes of the sink term (P). Negative residuals (from blue to violet colored) indicate the sink terms are larger than the source terms. The residuals from the combinations that include HOAPS and GSSTF evaporation have more positive regions than the residuals that include the OAFlux evaporation. Most of the residuals have positive values due to larger WVFC over some portions of the ITCZ where P is generally larger than E with the exception of that for OHXH. Some of the residuals have widely spread positive values over the southeastern Pacific except the residuals that include OAflux evaporation. Over the SPCZ, there are distinct negative residuals especially in the residuals that included CMAP precipitation. Yin et al. [2004] noted that the CMAP precipitation is often larger than the GPCP over SPCZ regions due to involvement of atoll gauge data. The regions where differences between the nine estimated residuals exist can also be determined using equation (Error! Reference source not found.). The distribution of the large variances in the residuals coincides with those of evaporation and precipitation (Figure 10a).

Figure 7.

Mean spatial distributions of the nine residuals for the period between 2000 and 2005.

[25] The relative magnitudes of residuals to the terms in the water budget were also investigated for each grid with the ratio of period-mean residuals to period-mean values for evaporation and period-mean values for precipitation. As an example, the case of OGXH is included in Figure 8. Over most of the oceanic areas, the period-mean relative magnitudes of OGXH residuals to evaporation from OAFlux are less than about 50%. Near coastline and around 60°S, however, the values become higher (Figure 8a). It is also found that the residuals compared to precipitation from GPCP are usually large over the dry oceanic regions (Figure 8b).

Figure 8.

Spatial distributions of the ratios of period-mean OGXH residual (a) to period-mean evaporation from OAFlux and (b) to period-mean precipitation from GPCP. The period is from 2000 to 2005. Units are %.

[26] The mean spatial distribution of the ANA of MERRA (Figure 9b) had strong negative values over the oceanic regions near Peru. Strong positive values exist over the West Pacific, the Bay of Bengal, subtropical regions over the East Pacific, and inter-American seas. A major feature of the ANA period mean distribution is analogous to ANA climatological mean map represented by Robertson et al. [2011]. The residual MERRA (exANA) had identical patterns with opposite signs as the ANA in its period mean spatial distribution (Figure 9a). The period mean of the MERRA (exANA) residual ranged from −2.64 to 2.69 mm d−1. The MERRA residual derived by taking the ANA into consideration had relatively small values between −0.07 and 0.06 mm d−1 (Figure 9c).

Figure 9.

Mean spatial distributions of (a) the residual excluding the ANA term and (b) the ANA for water vapor as well as (c) the residual including the ANA term for MERRA.

[27] A large quantity of the residual indicates an imbalance in the atmospheric water budget. In order to determine the regions where large imbalances would appear, the mean absolute errors were calculated from the nine estimated residuals for each grid box as follows:

display math(9a)

and

display math(9b)

where MAEi is the mean absolute error from the nine estimated residuals, Ri,j is the jth estimated residual from the satellite-based and merged data sets at month i, and MAEm is the period mean absolute error. Figure 10b shows the spatial distribution of MAEm. Relatively large magnitudes of residuals appeared around some of the coastal areas and over the Arabian Sea. The imbalances were also generally large over the ITCZ, SPCZ, and monsoon area where heavy precipitation occurs.

Figure 10.

Spatial distributions of (a) the standard deviations and (b) the mean absolute errors from the nine estimated residuals.

4 Summary and Conclusions

[28] This study examined column integrated atmospheric water balances based on the analysis of residuals from various combinations of satellite-based and merged data sets for atmospheric water cycle components. Satellite-based data sets for evaporation such as HOAPS and GSSTF were used as well as the satellite-NWP reanalysis merged evaporation data set OAFlux. For precipitation, the satellite-gauge merged GPCP and CMAP data sets as well as the satellite-based HOAPS were used. The water vapor flux convergence by XLT08 and the water vapor tendency derived from the HOAPS TPW were also used. The residuals of the MERRA in regard to analysis increments were also analyzed for comparison.

[29] The mean spatial distribution analysis for the period between 2000 and 2005 over oceanic regions (60°N–60°S) showed that the satellite-based residual distribution and their magnitudes varied with the data sets. The residuals from combinations including HOAPS and GSSTF evaporation had relatively large positive values over the midlatitude oceanic regions due to the relatively high evaporation values of the HOAPS and GSSTF over these areas. The values of the period mean standard deviations between the residuals were significantly larger over the ITCZ, SPCZ, and monsoon regions, and their magnitudes ranged from 0.27 to 1.5 mm d−1. The magnitude of the imbalance in the atmospheric water budget estimated from the satellite-based and merged data sets was also generally larger over the oceanic areas with heavy precipitation such as the ITCZ, SPCZ, and monsoon regions and had values ranging from 0.72 to 4.71 mm d−1 with a domain average value of 1.54 mm d−1. The larger residuals over the ocean may be attributed to errors in the data sets and the approximation of the water budget equation without the condensed phase water component. Meanwhile, the MERRA has closed atmospheric water budget requiring artificially added nonphysical analysis increment term for the water vapor tendency.

[30] Analysis of the residuals from various combinations of data sets in this study indicated that challenges remain in order to obtain an accurate atmospheric water budget. The errors in the individual data sets may be attributed to their own methodologies and algorithms. Therefore, we recommend that careful consideration be used when applying data sets for water budget components to water budget studies. However, similar residual anomalies suggest that climate variability analysis based on the residuals may not be greatly affected by a specific data set.

Acknowledgments

[31] This work was funded by the Korea Meteorological Administration Research and Development Program under Grant CATER 2012–2063.

Ancillary