Leakage of continental hydrology in seawater mass change estimations from space in the Mediterranean Sea and Black Sea Second Space for Hydrology Workshop Fenoglio-Marc L. (1), Becker M. (1), Kusche J. (2), Rietbroek R.

(2)

(1) Institut für Physikalische Geodäsie, Technische Universität Darmstadt Petersenstrasse 13, Germany [email protected] [email protected] (2) GeoForschungsZentrum Telegraphenberg 326, Potsdam, Germany [email protected], [email protected] ABSTRACT The leakage of land hydrology has a sizeable effect on the oceanic mass estimated from GRACE in small ocean basins, as the land signal is typically much larger. Its magnitude depends on the dimension of the region, on the smoothing applied to GRACE and the type of the signal. Hydrological models are used to correct for this leakage effect. Departures between the models may cause significant differences in water mass change in the coastal regions. The resulting improvement in the seawater mass estimates is evaluated by comparison with steric-corrected altimeter basin averages. Results for the Mediterranean and the Black Sea are shown. INTRODUCTION GRACE observations from space, when used in conjunction with in situ observations and hydrological modeling, have the potential to significantly improve our understanding of hydrological processes and the global water cycle. The GRACE satellites are measuring gravity variations due to mass changes at the micro-gal level with a spatial resolution of about 300-400 km. Monthly down to weekly models of the actual gravity field are computed using spherical harmonics or other approximation techniques. Mass changes over land are related to hydrology and surface waters, mass changes over the open Sea or in basins are indications of mass transport and sea level changes. In order to obtain sufficient accuracy, filtering within a certain radius around the computation point has to be applied. However the scale of the associated spatial filters causes signal from nearby coastal areas to be absorbed in the ocean estimates. Due to the regional analysis and the small dimension of the basins that we investigate here, it is not possible to proceed as in [1] masking all ocean regions within few hundred km from the coastline. Therefore the hydrological signal due to coastal regions is included in the seawater mass change estimated from GRACE. The leakage of land hydrology has a sizeable effect on the oceanic mass estimated from GRACE, as the amplitude of the land hydrology signal is in general large. In the Mediterranean Sea the annual amplitudes of the leaked land hydrology signal and of the net oceanic mass change estimated from the steric-corrected altimetry are comparable [2]. The annual amplitude and phase of the leaked signal depend on the dimension of the region, smoothing and truncation applied to GRACE monthly coefficients and to the type of the signal (see [3] for a theoretical derivation). The smoothing procedures reduces the GRACE signal and a corrective scaling factor has to be applied ([4], [5], [6]). Within the framework of the STREMP-Project the regional mass transport and mass distribution in the Mediterranean and Black Sea catchments are to be estimated. The study is based on a combination of altimetry, GRACE gravity fields,

oceanography and hydrology. Main topics are the coupling of continental hydrology with oceanography in coastal oceans and the water balance closure at the straits. The GRACE derived time-variable mass change, with spatial accuracy improved by advanced filtering approaches, is regionally compared to estimates of non-steric sea level change from altimeter observations. The joint assessment of mass changes brings all (gravimetric, hydrographic, altimetric and oceanographic) components together to derive the mass balance and the mass exchange for the whole area. In this paper we analyse the leakage of continental hydrology in seawater mass change from GRACE and the differences between three hydrology models

Fig. 1. Study area of the STREMP Project with ocean basins and catchments area DATA AND METHODS SeawaterMass Change from GRACE and Hydrographic Models Release level 4 (RL4) GFZ spherical harmonic models of the gravitational field between February 2003 and December 2006 are used. The non-tidal ocean contribution, removed in the preprocessing, is added back to the GRACE fields on sea (GFZ GAD product). We include degree 1 of the geopotential [7]. and truncate the GRACE fields to degrees 2-70. Spatial averaging is required to produce a reasonable signal to-noise ratio from the GRACE data. We construct an averaging kernel by convolving a 400 km halfwidth Gaussian function with the ocean basin mask (defined as 1 at pointsinside the region and 0 outside). The smoothed averaged time-series of water thickness is obtained as in [2]. The signal from GRACE is thus contaminate by signals outside the ocean basin. To estimate the hydrology leakage we convert the data of three hydrologic models to the spectral domain and compute monthly filtered basin averages in the same way as for GRACE. The land surface model (LSM) used are the LAD, GLDAS and WGHM models [8],[9],[10]. The filtered basin average of the hydrological signal is then removed from the GRACE filtered basin average to obtain the hydrological-corrected seawater mass change from GRACE. Seawater Mass cChange from Altimetry and Oceanographic Models Monthly sea surface height maps are computed from the Jason-1 altimeter mission in 2002-2006. Maps for the steric component of sea surface height variation are derived from the ECCO model in the Mediterranean Sea, in the Black Sea from a local climatology. Filtered basin averages for the two quantities are constructed in the spatial domain with a method equivalent to the GRACE processing described above [2]. The difference of the two basin averages gives the filtered mass change variation. The amplitude damping factor is obtained as in [2] by comparing unfiltered and filtered steric-corrected sea level data. The true amplitudes are then obtained by dividing the filtered results by this factor. In the Mediterranean Sea the amplitude damping factor is 0.56 [2]. Total Water Storage in the Watershed To estimate the total water storage in the watershed from the GRACE data, a similar procedure as described above for the ocean basin has to be applied. We therefore construct an averaging kernel by convolving a 400 km halfwidth Gaussian function with the watershed mask (defined as 1 at pointsinside the region and 0 outside). The signal on the

watershed is contaminated by the land outside the watershed and by ocean signals. As the hydrology signal is large with respect to the ocean signal, we neglect here the leakage of ocean mass change in the water storage change measured by GRACE in our area. We then analyse the difference between the hydrology models computing unfiltered and filtered regional averages over the watershed and compare the filtered basin averages to GRACE water storage estimations over the same region. RESULTS Seawater Mass Change In the Mediterranean Sea, the amplitude of the land hydrology signal that leakages in the Mediterranean Sea watermass change is comparable to the water mass signal in the Mediterranean basin. Using a Gaussian filter with 400 km halfwidth, in the interval from April 2002 to July 2004, its annual amplitude is 29 +/-3 mm, similar to the annual amplitude of the filtered mass change from steric-corrected altimetry [6]. Fig. 2 shows the contamination in the Mediterranean Sea produced by the LAD model for two given months. The two composite time-series, the filtered steric-corrected sea level and the hydrology-corrected GRACE CSR RL3 fields , have an RMS difference of 28 mm and a correlation of 0.8 between April 2002 and July 2004 [5]. With GRACE GFZ RL4 the correlation is 0.8 and the RMS 14 mm from June 2002 to July 2006 before rescaling. The results are divided by the amplitude damping factor 0.56 to obtain the true amplitude [2]. In the Mediterranean coastal regions the RMS difference between the filtered monthly basin averages of the various land surface models (LSM) is smaller than 11 mm. The true signal difference is then around 20 mm.

Fig. 2. Contribution of hydrography to smoothed basin average

In the Black Sea, using the same Gaussian filter, the contamination of land hydrology is bigger than in the Mediterranean Sea. (Fig. 3). The annual amplitude of the filtered hydrology leakage is between 41 and 66 mm depending on the land surface model (LSM) used, the smaller value corresponding to the WGHM and the bigger value to the LAD model (Table 1). The amplitude damping factor is in this case 0.31, therefore the true signal that leakages in the Black Sea watermass change estimates is 3 times bigger than in the Mediterranean Sea. A similar value is given in [11] for the Caspian Sea. The difference in the annual phases is smaller than 13 days. The RMS difference between the filtered signals in coastal regions is below 13 mm, the true signal is therefore 42 mm. The composite time series of steric-corrected altimetry and hydrology-corrected GRACE in the Black Sea (Fig. 4) present the smallest RMS (27 mm) when GLDAS is used and the highest correlation (0.55) when WGHM is used (Table 1).

Fig. 3. Filtered basin average of total water storage that leakages in the Black Sea GRACE water mass estimation. Model used: LAD (diamonds), GLDAS (triangle), WGHM (circle)

Fig. 4. Comparison of seawater mass change in Black Sea from and steric-corrected altimetry and hydrology-corrected GRACE with Hydrological Models LAD (top), GLDAS (middle) and WGHM (bottom)

Total Water Storage in the Watershed In the Black Sea watershed the filtered and unfiltered basin average of the hydrological signal are similar. The RMS difference between the filtered basin averages of ´different models is below 10 mm. The time series of filtered GRACE and filtered hydrology present the smallest RMS differences (26 mm) and the highest correlation (0.89) when GLDAS is used (Table 1).

Table 1. Correlation and RMS of the composite time-series (steric-corrected sea level and hydrology-corrected GRACE)in Black Sea and correlation with GRACE in the watershed. A Gaussian filter of 400 km radius is used. Field LAD GLDAS WGHM

Black Sea Correlation 0.43 0.42 0.55

Black Sea RMS (mm) 33 27 29

Watershed BS Correlation 0.84 0.89 0.81

Watershed BS RMS (mmm) 30 26 33

The differences in annual amplitude of the filtered basin averages is less than 7 mm, in annual phase are smaller than 9 days. Compared to the GRACE basin averages, the difference in amplitude is small, the maximum phase difference (25 days) is obtained with the WGHM model. The best agreement in both amplitude (2 mm) and phase (16 days) is realised by the GLDAS model (Table 2) . The water storage from GRACE has a strong trend, that is not present in any of the hydrology models (Fig. 5). Table 2. Annual amplitude and phase of the filtered basin average of the hydrology leakage in the Black Sea from three land surface models. A Gaussian filter of 400 km radius is used. Field LAD GLDAS WGHM GRACE 1-70

Black Sea Amplitude (mm) 66 +/- 2 51 +/- 2 41 +/- 2

Black Sea Phase (deg) 62 +/- 2 70 +/- 3 57 +/- 3

Watershed BS Amplitude(mm) 68 +/- 2 68 +/- 2 61 +/- 2 66 +/- 2

Watershed BS Phase (deg) 64 +/- 2 69 +/- 2 60 +/- 2 85 +/- 2

Fig. 5. Filtered basin average of total water storage in the Black Sea watershed. Models used: LAD (diamonds), GLDAS (triangle), WGHM (circle) CONCLUSIONS We have shown that the leakage of land hydrology has a sizeable effect on the oceanic mass estimated from GRACE in small ocean basins. The application of this correction to the GRACE data results in a better agreement with the water mass variation derived from the steric-corrected altimetry. The amplitude damping is obtained by comparing unfiltered and filtered steric-corrected sea level data. Damping factors are 0.56 in the Mediterranean Sea and 0.31 in the Black Sea. The magnitude of the hydrology leakage in the filtered basin average from GRACE depends on the dimension of the region, on the smoothing applied to GRACE and the type of the signal. Its annual amplitude is of same order of magnitude that the filtered oceanic water mass signal in the Mediterranean Sea (29 mm) and bigger in the Black Sea (about 50 mm). When hydrological models are used to correct for this leakage effect, the departures between the models may cause significant differences in water mass change in the coastal regions. In Mediterranean Sea coastal regions differences have a RMS smaller than 22 mm. The filtered total water storage derived from three land surface models appear to be

more different in the Black Sea coastal regions, with RMS up to 30 mm. RMS are 10 mm in the Black Sea watershed. In coastal region the better performing models are GLDAS (lower RMS between composite time-series is 27 mm) and WGH (higher correlation). In the watershed the better agreement with GRACE is obtained with GLDAS. This study shows differences between the land surface models. The GRACE data can be used to validate, tune and improve the hydrological models. Future reprocessing of GRACE data, improved dealiasing products and methods, and tailored analysis methods may also lead to further improvement in the GRACE products. The joint assessment of mass changes brings all (gravimetric, altimetric and oceanographic) components together to derive the mass balance and the mass exchange for the whole area. Our further analysis will concentrate in the validation of the models and separation of the signals. By combining and comparing the independent datasets, we will use their consistency as a validation measure. In the application to coastal areas and ocean basin, tailored methods for signal separation will be developed to separate land and ocean effects. The land-ocean exchange measurement (e.g. runoff) is a further element for the study.. Acknowledgements. We kindly acknowledge C.K. Shum and J. Wahr for providing the GLDAS data. This study was founded by DFG (SPP1257/STREMP) REFERENCES [1] Chambers, D. P. (2006), Observing seasonal steric sea level variations with GRACE and satellite altimetry, J. Geophys. Res., 111, C03010, doi:10.1029/2005JC002914. [2] Fenoglio-Marc L., J. Kusche, M.Becker (2006), Mass variation in the Mediterranean Sea from GRACE and its validation by altimetry, steric and hydrologic fields, Geophys. Res. Lett., 33, L19606, doi.10.1029/2006GL026851. [3] J. Kusche (2006), Approximate decorrelation and non-isotropic smmothing of time-variable GRACE-type gravity field models, J. of Geodesy, doi 10,1007/s00190-007-0143-3 [4] Velicogna I. and J. Wahr (2006), Measurements of Time-Variable Gravity show mass loss in Antarctica, 10.1126/science.1123785 [5] Klees R., E.A. Zapreeva, H.C. Winsemius and H.H.G. Savenije (2007), The bias in GRACE estimates of continental water storage, Hydrology and Earth System Sciences Discussion, in publication [6] Fenoglio-Marc L., J. Kusche, M.Becker, I. Fukumori (2007), Comments on “On the steric and mass-induced contributions to the annual sea level variations in the Mediterranean Sea” by D. Garcia et al., J. of Geophys. Res, in press, doi:101029/2007JC004196 [7] Jansen J., J. Kusche and E.J.O. Schrama (2006). Low degree load harmonic coefficients from combining GRACE, GPS time-series and a-priori dynamics, Proceedings IAG Symposium 2006, Instanbul [8] Milly P.C.D. and A.B. Shmakin (2002), Global modeling of land water and energy balances, Part I: The Land Dynamics (LaD) Model, J. Hydrometerol., 3, 283-299 [9] Rodell M. et al. (2004) The Global data Land Data Assimilation System, Bull. Am. Meteorol. Soc., 85, 381-394 [10] Döll P., F. Kaspar, B. Lehner (2003): A global hydrological model for deriving water availability indicators: model tuning and validation, J. Hydrol. 270, 105-134

[11] Swenson, S., and J. Wahr (2007), Multi-sensor analysis of water storage variations of the Caspian Sea, Geophys. Res. Lett., 34, L16401, doi:10.1029/2007GL030733.