Largescale estimates of changes in terrestrial water storage

1
Large-scale estimates of changes in terrestrial
water storage using streamflow measurements and
ERA-40 reanalysis data Sonia Seneviratne GMAO
Seminar, February 5, 2004
2
Acknowledgements

• Martin Hirschi, Daniel Lüthi, and Christoph Schär
• Atmospheric and Climate Science ETH, ETHZ,
  Zurich, Switzerland
• Pedro Viterbo
• European Centre for Medium-Range Weather
  Forecasts, Reading, UK

3
Outline

• Motivation
• Combined water-balance approach
• Validation study for the Mississippi River basin
• Results for Europe and Asia
• Other applications
• estimation of ET
• possible interactions with projects at NASA/GSFC
• Conclusions

4
Motivation (1)

• What is terrestrial water storage?
• Soil moisture
• Groundwater
• Snow cover
• Land ice, surface water,
• biospheric water,

5
Motivation (2)
Why is it important?

• Risks of droughts and floods agriculture
  freshwater supply

• Importance within climate system
• Critical for NWP, seasonal predictions,
• and long-term climate simulations
• Land-atmosphere feedbacks
• Memory component

6
Motivation (3)

• There are however only very few observations of
  terrestrial water storage and its components

Global Soil Moisture Data Bank (Robock et al.
2000)
7
Motivation (4)

• and the indirect datasets available (reanalysis
  data, model-computed values) do often not agree
  with one another

Seneviratne, 2003
Validation of GSWP1 simulations, Entin et al. 1999
8
Water-Balance Approach (1)

• Terrestrial water balance

• Atmospheric water balance

measured streamflow (RsRg)

• Combined water balance

9
Water-Balance Approach (2)

• Assumptions

• The contributions of the liquid and solid phases
  of atmospheric water are negligible

• The measured streamflow includes both the
  contributions of surface and groundwater runoff

• Limitations

• Atmospheric water balance estimations are
  accurate only for domains gt 105-106 km2
  (Rasmusson 1968, Yeh et al. 1998)

10
Water-Balance Approach (3) Summary

• Changes in terrestrial water storage (dS/dt) in a
  given river basin can be estimated as the sum of
  three terms

Convergence of the vertically integrated
water vapour flux
Reanalysis data
Change in column storage of water vapour
Observations
Measured streamflow

• The estimates depend only on observed or
  assimilated variables (? P,E)

11
Employed Data

• ERA-40
• ECMWF reanalysis data
• (completed in 2003)

• USGS (US Geological Survey)
• GRDC (Global Runoff Data Center)

12
ERA-40 Reanalysis Project (1)

• reanalysis data from 1957-2002

Satellite period (1989-2001)
Pre-satellite period (1958-1972)
ECMWF
13
ERA-40 Reanalysis Project (2)

• Resolution
• horizontal resolution 112 km
• 60 vertical levels
• (high resolution in the lower troposphere)

• Data assimilation
• 3D assimilation system
• 6-hour analysis cycle

Assimilation increment
time (GMT)
14
Case Study Mississippi River Basin

• Seneviratne et al. 2004, J.Climate, in press

• Study Period 1987-1996

• Runoff data USGS

• Validation against
• observations in Illinois
• (soil moisture,
• groundwater and snow)

1) Arkansas-Red (6.105 km2) 2) Missouri (13.105
km2) 3) Upper Mississippi (5.105 km2) 4)
Ohio-Tennessee (5.105 km2) 5) Lower Mississippi
(4.105 km2) 6) Illinois (2.105 km2)
Adapted from Betts et al. 2003
15
Illinois Validation Data (1)

• Soil moisture
• 19 sites, 1-2 measurements per month
• (Illinois State Water Survey)

• Groundwater (ISWS)
• 17 sites, 1 measurement per month
• (Illinois State Water Survey)

• Snow
• 32 stations (less than 10 missing data), daily
  data
• (Midwest Climate Center)

16
Illinois Validation Data (2)
The contributions of the snow cover are
negligible

• The variations in groundwater and soil moisture
  are of similar magnitude and clearly correlated

Terrestrial water storage components mm
Monthly variations mm/d
17
Validation (1) Monthly Variations
Water-balance Estimates

• Extremes are also well captured
• Excellent agreement in most years
• Only significant discrepancy in year with
  highest recycling ratio (Bosilovich and Schubert
  2001)

Observations (soil moisture groundwatersnow)
drought years
flood year
18
Validation (2) Climatology
19
Validation (3) Correlation
corr 0.84 slope 0.82
20
Estimates Mississippi Climatologies
10-year Mean Water-Balance Components mm/d
Upper Mississippi
Missouri
Arkansas-Red
Illinois
Whole Mississippi
Ohio-Tennessee
21
Estimates Interannual Variability
Monthly Water-Balance Components mm/d
Arkansas-Red
Missouri
Upper Mississippi
Ohio-Tennessee
Illinois
Whole Mississippi
22
Estimates Monthly Variations

• Excellent agreement with observations in Illinois

• The mean climatology and the interrannual
  variability of the computed monthly variations
  dS/dt appear realistic for all subbasins

• Some characteristical regional features are
  recognizable (e.g. Ohio-Tennessee)

23
Temporal Integration (1)

• Is it possible to integrate the estimates in
  order to obtain ?

• Seasonal changes in terrestrial water storage

• Values of absolute terrestrial water storage

24
Temporal Integration (2)

• Seasonal Changes (4-6 months)
• correlation is in general high
• but decreases for longer time ranges

Computed mean (1987-1996) seasonal change in
terrestrial water storage vs observations
(Illinois)
25
Temporal Integration (3)
Observations (Illinois)
Integrated estimates

• Integration over longer time ranges is not
  straightforward due to the presence of small
  systematic imbalances in the monthly estimates

Comparison with imbalances from other
water-balance studies
G97 Gutowski et al. 1997 Y98 Yeh et al.
1998 BR99 Berbery and Rasmuson 1999
26
Temporal Integration (4) Detrending
Absolute terrestrial water storage (observed and
estimated)

• A simple detrending yields good estimates of
  absolute terrestrial water storage

Assumptions - mean annual dS/dt 0 - April
value set to climatological value
27
Summary of Validation Study

• The tested methodology yields excellent estimates
  of monthly changes in terrestrial water storage

• With an appropriate detrending, absolute
  terrestrial water storage can also be estimated

possible dataset of changes in terrestrial water
storage for all major river basins for 1958-2002
28
Application to Northern River Basins

• Hirschi et al. 2004, in preparation
• - whole ERA-40 period (1958-2002)
• - runoff data Global Runoff Data Center (GRDC)

29
Comparisons with soil moisture observations
Neva (1960-91)
Ob (1987-88)
dS/dt (Water-balance estimates)
Hirschi et al. 2004
dSM/dt (Soil moisture observations)
groundwater, snow ?
30
Long-term Imbalances (1)

• The accuracy of the computed water balances
  depends
• both on domain size and on regional
  characteristics

?
Rasmusson (1968) threshold for radiosonde data
(2.106 km2)
Imbalances (mm/d)
Illinois (2 .105 km2)
Europe Western Russia Asia North America
Domain size (km2)
Hirschi et al. 2004
31
Long-term Imbalances (2)

• The accuracy of the computed water balances
• depends critically on the domain size.

• However, the critical domain size might be lower
  
• for reanalysis data (Illinois 2.105 km2)
  than for
• radiosonde data (Rasmusson 1968)

• and might also depend on regional
  characteristics
• (climate, density of radiosonde data,
  topography?).

32
Other applications

• Estimation of Large-scale Evapotranspiration

Atmospheric water balance
Mackenzie GEWEX Study (MAGS)
Louie et al. 2002
33
(monthly) evaporation in Southern Europe
ERA-40 6 hour forecasts

• Aerological estimate of evaporation has a much
  larger interannual variability (and larger values
  in summer) than model evaporation

34
More applications

• Model assessment and validation
• - Catchment models
• - Offline surface model results

• Terrestrial water storage memory
• - Are there some correspondences with soil
  moisture memory computed with GCMs?
• - Slope of Evapotranspiration vs Soil Moisture

• Soil moisture data assimilation
• -Assessment of satellite vs model soil moisture
  data

• Comparison with GRACE data and gravimetry
  measurements at the ground

35
Conclusions

• The combined water-balance approach is a
  promising tool for estimating large-scale changes
  in terrestrial water storage

• Some limitations
• Domain size needs to be at least gt 2.105 km2
• A detrending is needed for the estimation of
  absolute water storage
• Additional validation data would be needed in
  order to test this approach for other regions

• Nonetheless, the possible applications and uses
  are numerous given the dearth of observations of
  terrestrial water storage and its components