Coupled data assimilation for atmosphere-land surface-subsurface - - PowerPoint PPT Presentation

▶
coupled data assimilation for atmosphere land surface
SMART_READER_LITE
LIVE PREVIEW

Coupled data assimilation for atmosphere-land surface-subsurface - - PowerPoint PPT Presentation

Apr 09, 2023 48 likes •491 views

Coupled data assimilation for atmosphere-land surface-subsurface models Harrie-Jan Hendricks-Franssen 1,2 , Wolfgang Kurtz 1,2 , Hongjuan Zhang 1,2 , Prabhakar Shrestha 3 , Dorina Baatz 1,2 , Clemens Simmer 3 , Stefan Kollet 1,2 , and Harry

slide-1
SLIDE 1

Mitglied der Helmholtz-Gemeinschaft

Coupled data assimilation for atmosphere-land surface-subsurface models

Harrie-Jan Hendricks-Franssen1,2, Wolfgang Kurtz1,2, Hongjuan Zhang1,2, Prabhakar Shrestha3, Dorina Baatz1,2, Clemens Simmer3, Stefan Kollet1,2, and Harry Vereecken1,2

1 Forschungszentrum Jủlich, Agrosphere (IBG 3), Leo-Brandt-Strasse, 52425 Jủlich, Germany 2 HPSC-TerrSys, Geverbund ABC/J, Jülich, Germany 3 Meteorological Institute, Bonn, Germany

slide-2
SLIDE 2
  • 4. Juni 2018

Folie 2 Agrosphere Insitute (IBG-3)

Overview

  • Introduction on coupled data assimilation.
  • Coupled atmosphere-land surface- subsurface model TerrSysMP.
  • Data assimilation framework TerrSysMP-PDAF.
  • Example synthetic and real-world studies.
  • Conclusions and outlook.
slide-3
SLIDE 3
  • 4. Juni 2018

Folie 3 Agrosphere Insitute (IBG-3)

Non-coupled DA

  • Non-coupled DA of hydrological cycle.
slide-4
SLIDE 4
  • 4. Juni 2018

Folie 4 Agrosphere Insitute (IBG-3)

Non-coupled DA

  • Non-coupled DA of hydrological cycle.

Atmospheric DA (e.g., 3D/4DVAR) Land surface DA (e.g., EnKF) Soil hydrology DA (e.g., 1D McMC, PF) Rainfall-runoff DA (e.g., McMC, PF) Groundwater DA (e.g., 3D EnKF)

slide-5
SLIDE 5
  • 4. Juni 2018

Folie 5 Agrosphere Insitute (IBG-3)

Coupled DA

  • Weakly coupled DA
  • DA for individual compartments of terrestrial system.
  • Covariances between states of different compartments not

calculated.

  • Updates for single compartments propagated through coupled model

equations

  • Fully coupled DA
  • DA for multiple compartments of terrestrial system.
  • Covariances between states of different compartments calculated.
  • States of multiple compartments are directly updated by DA.
slide-6
SLIDE 6

Mitglied der Helmholtz-Gemeinschaft

4 June 2018 IBG-3: Agrosphere 6

TerrSysMP

  • 3D Variably saturated subsurface flow

and energy transport (Jones &

Woodward, 2001; Kollet et al., 2009)

  • Integrated overland flow, terrain

following grid (Kollet & Maxwell, 2006;

Maxwell, 2013)

  • Integrated land surface and regional

climate model (Shrestha et al., 2014)

  • External coupling via OASIS3:

Multiple Program Multiple Data Execution Model (Shrestha et al., 2014)

  • Atmospheric downscaling algorithm

(Schomburg et al., 2010)

slide-7
SLIDE 7
  • 4. Juni 2018

Folie 7 Agrosphere Insitute (IBG-3)

TerrSysMP

  • Lateral subsurface transport of water and energy via groundwater
  • PDE-based description of two-way interactions between groundwater,

vadose zone, surface water, vegetation and atmosphere

  • Land surface (CLM3.5) component still has large potential to be improved

(e.g., beta-function for drought stress, photosynthesis types, plant traits)

  • Overland flow process very non-linear → very high spatial resolution

needed

  • In general, many unknown parameters, initial states and forcings → data

assimilation

slide-8
SLIDE 8

Mitglied der Helmholtz-Gemeinschaft

Simulations up to continental scale

  • Groundwater depth calculated over Europe
  • Problem: long spin-ups needed related to slow groundwater dynamics.

Upper Rhine river Tunnel valleys Sweden

Keune et al., 2016

slide-9
SLIDE 9

Mitglied der Helmholtz-Gemeinschaft

Current work: weakly coupled DA

COSMO (Atmosphere) CLM v4.5 (Land surface) ParFlow (Subsurface)

Forecast

GWL COSMO (Atmosphere) CLM v4.5 (Land surface) ParFlow (Subsurface)

Analysis Data

COSMO (Atmosphere) CLM v4.5 (Land surface) ParFlow (Subsurface)

Forecast t+1 Example: Assimilation of land surface and subsurface data, which only update

  • wn compartments and later other compartments.

Q LAI SM LST

slide-10
SLIDE 10

Mitglied der Helmholtz-Gemeinschaft

Towards fully coupled DA?

COSMO (Atmosphere) CLM v4.5 (Land surface) ParFlow (Subsurface)

Forecast

COSMO (Atmosphere) CLM v4.5 (Land surface) ParFlow (Subsurface)

Analysis Data

COSMO (Atmosphere) CLM v4.5 (Land surface) ParFlow (Subsurface)

Forecast t+1 Example: Assimilation of atmospheric, land surface and subsurface data; all of them can update all compartments.

PRECIP BL GWL Q LAI SM LST

slide-11
SLIDE 11

Mitglied der Helmholtz-Gemeinschaft

Between weakly and fully coupled DA?

COSMO (Atmosphere) CLM v4.5 (Land surface) ParFlow (Subsurface)

Forecast

COSMO (Atmosphere) CLM v4.5 (Land surface) ParFlow (Subsurface)

Analysis Data

COSMO (Atmosphere) CLM v4.5 (Land surface) ParFlow (Subsurface)

Forecast t+1 Example: Only some of the measurement data are used to update (sensitive) states in other compartments

PRECIP BL GWL Q LAI SM LST

slide-12
SLIDE 12
  • 4. Juni 2018

Folie 12 Agrosphere Insitute (IBG-3)

TerrSysMP-PDAF

  • PDAF (Nerger and Hiller, 2013) was coupled to TerrSysMP
  • COSMO, CLM and ParFlow are parallel, DA in addition also parallel
  • DA system is fully integrated (no I/O, no model reinitializations)
  • Good scalability through effective use of domain decomposition
  • Different DA-algorithms activated (EnKF, local EnKF, LETKF)
  • Multiscale SM, GW levels and river water levels can be assimilated

Kurtz et al. (2016), GMD

slide-13
SLIDE 13
  • 4. Juni 2018

Folie 13 Agrosphere Insitute (IBG-3)

Results feasibility test (synthetic)

In total 2 x 107 states and 2 x 107 parameters are updated with EnKF (Kurtz et al., 2016, GMD)

REFERENCE K INITIAL K FINAL K

slide-14
SLIDE 14
  • 4. Juni 2018

Folie 14 Agrosphere Insitute (IBG-3)

1st Example: Weakly coupled DA

Atmospheric DA (e.g., 3D/4DVAR) Land surface DA (e.g., EnKF) Soil hydrology DA (e.g., 1D McMC, PF) Rainfall-runoff DA (e.g., McMC, PF) Groundwater DA (e.g., 3D EnKF)

slide-15
SLIDE 15
  • 4. Juni 2018

Folie 15 Agrosphere Insitute (IBG-3)

Cosmic ray probe data

Cosmic ray scattering (HydroInnova, 2007)

  • Primary cosmic rays collide with atomic

nuclei

  • Creation of secondary cosmic rays with

lower energy

  • Hydrogen is the most effective neutron

absorber

slide-16
SLIDE 16
  • 4. Juni 2018

Folie 16 Agrosphere Insitute (IBG-3)

Cosmic ray probe data: local effects

Hydrogen pools include:

  • soil water content
  • lattice water
  • aboveground biomass

Neutron counts to be corrected for:

  • incoming cosmic-ray intensity
  • air pressure
  • atmospheric humidity

C R P

r - effective radiusair … up to 200 m z* - effective depthsoil … up to 0.7 m

r~200m z*~0.7m

Non-linear measurement

  • perator
slide-17
SLIDE 17
  • 4. Juni 2018

Folie 17 Agrosphere Insitute (IBG-3)

Cosmic ray probe data: function  

2 1

/ a a N N a

corr grav

   

Equation to calculate soil moisture from cosmic ray counts:

Θgrav – soil water content [g/g] a0, a1, a2 – constants Ncorr – Measured neutrons / hour N0 – Neutron counts under dry soil conditions Fitting curve with a0, a1, a2 a0 = 0.0808 a1 = 0.372 a2 = 0.115 and N0 = 1107

Ref.: Desilets et al. (2010). Nature‘s neutron probe: Land surface hydrology at an elusive scale with cosmic rays. Water Resources Research.

slide-18
SLIDE 18
  • 4. Juni 2018

Folie 18 Agrosphere Insitute (IBG-3)

TERENO observatory Rur catchment

slide-19
SLIDE 19
  • 4. Juni 2018

Folie 19 Agrosphere Insitute (IBG-3)

Cosmic Ray Probe Network Rur catchment

x2 CRP

slide-20
SLIDE 20
  • 4. Juni 2018

Folie 20 Agrosphere Insitute (IBG-3)

Weakly coupled DA Land Surface-Subsurface

  • Test value of cosmic ray probe data measured by cosmic ray probe
  • Horizontal model resolution: 500 m (100x162 cells)
  • Vertical resolution: 2cm-136 cm, 30 layers (30 m total thickness)
  • Vegetation classification from MODIS
  • Model forcings from COSMO-DE reanalysis
  • Subsurface properties from European Soil data base
slide-21
SLIDE 21
  • 4. Juni 2018

Folie 21 Agrosphere Insitute (IBG-3)

Set up DA-experiments

  • 128 ensemble members, perturbation of precipitation, incoming short

wave and long wave radiation, air temperature and porosity and log(Ksat).

  • Assimilation period April – September 2013.
  • Assimilation of soil moisture from 8 cosmic ray probes with EnKF.
  • Probe left out in assimilation used for verification (jackknife).
  • Repeated 9 times (all probes once left out).
  • CLM versus ParFlow-CLM assimilation.
slide-22
SLIDE 22
  • 4. Juni 2018

Folie 22 Agrosphere Insitute (IBG-3)

RMSE soil moisture

slide-23
SLIDE 23
  • 4. Juni 2018

Folie 23 Agrosphere Insitute (IBG-3)

2nd Example: Strongly coupled DA

Atmospheric DA (e.g., 3D/4DVAR) Land surface DA (e.g., EnKF) Soil hydrology DA (e.g., 1D McMC, PF) Rainfall-runoff DA (e.g., McMC, PF) Groundwater DA (e.g., 3D EnKF)

slide-24
SLIDE 24
  • 4. Juni 2018

Folie 24 Agrosphere Insitute (IBG-3)

2nd example: GW-level assimilation

  • Soil moisture from satellite: indirect, coarse scale, only upper cm, not

reliable over dense vegetation

24

Good data source:

  • low cost
  • high accuracy
  • widely available

National Groundwater Monitoring Network Sites, e.g. USA

May contain valuable information about root zone soil moisture:

  • Statistical correlations
  • Physically related

Root zone (Source: Modified from Smedema and Rycroft, 1983) Precipitation Transpiration Evaporation

Soil Surface

Irrigation

Capillary Flow of Groundwater Water table

Unconfined Aquifer

Deep Percolation Deep Percolation

slide-25
SLIDE 25
  • 4. Juni 2018

Folie 25 Agrosphere Insitute (IBG-3)

How to assimilate GWL-data?

Water Table (WT) is in the ith layer

WT

. . . . . .

Soil Column

layer i layer i+1 layer 30 . . . . . .

  • bservation

State variable: pressure head. Assimilation of GWL could be done in terms

  • f soil moisture or pressure head:
  • pressure head (P): pressures in

saturated zone (hydrostatic)

  • soil moisture (SM): porosity in

saturated zone

slide-26
SLIDE 26
  • 4. Juni 2018

Folie 26 Agrosphere Insitute (IBG-3)

Five different DA-strategies

Method Observation vector State vector Updated domain P P P strongly coupled P_log log10(100-p) log10(100-p) strongly coupled P_mask P P weakly coupled SWC SWC (=porosity) SWC (=porosity) strongly coupled Mix P P (saturated) SWC (unsaturated) strongly coupled

  • strongly coupled = updating both saturated and unsaturated zone.
  • weakly coupled = groundwater level data do not update unsaturated zone

directly

slide-27
SLIDE 27
  • 4. Juni 2018

Folie 27 Agrosphere Insitute (IBG-3)

Test case

  • 2 x 2 grid cells (and 30 layers). Slope 1%.
  • Spatially homogeneous parameters.
  • Only log(Ksat) uncertain ~ N(0,1); 128 ensemble members.
  • 100 years spin-up, 1 year assimilation.
  • Daily assimilation of GWL (measurement error 0.01m).
slide-28
SLIDE 28
  • 4. Juni 2018

Folie 28 Agrosphere Insitute (IBG-3)

Example results

Strongly skewed (non-Gaussian) local pdf´s during dry conditions cause problems.

slide-29
SLIDE 29
  • 4. Juni 2018

Folie 29 Agrosphere Insitute (IBG-3)

Methodology to incorporate GW-levels

Comparison of five different methodologies to assimilate GW-level data in CLM-ParFlow

Zhang et al. (2018), Adv. Wat. Res.

slide-30
SLIDE 30
  • 4. Juni 2018

Folie 30 Agrosphere Insitute (IBG-3)

Value of GWL for RZSM-characterization

  • Vertically heterogeneous Ksat, α and n parameters.
  • Ksat, α and n uncertain and sampled from multi-normal distribution

(transformed parameters).

  • Precipitation uncertain: perturbed with multiplicative noise U[0.5, 1.5].
  • Performance evaluated for large number of synthetic experiments (100

cases: 4 soil types x 5 PFT´s x 5 climate types).

slide-31
SLIDE 31
  • 4. Juni 2018

Folie 31 Agrosphere Insitute (IBG-3)

RMSE Reduction vs. GWL

slide-32
SLIDE 32
  • 4. Juni 2018

Folie 32 Agrosphere Insitute (IBG-3)

RMSE Reduction vs. Climate/PFT/Soil

green: mean red: median

If GW-level is not very shallow or very deep, assimilation shows clear benefit Better results for loam soils. Better results for broadleaf trees.

slide-33
SLIDE 33
  • 4. Juni 2018

Folie 33 Agrosphere Insitute (IBG-3)

RMSE Reduction vs. GWL (PFT´s)

Slightly better for broadleaf

slide-34
SLIDE 34
  • 4. Juni 2018

Folie 34 Agrosphere Insitute (IBG-3)

RMSE Reduction vs. GWL (soil types)

Slightly better for loam

slide-35
SLIDE 35
  • 4. Juni 2018

Folie 35 Agrosphere Insitute (IBG-3)

3rd example: atmosphere-land surface DA

  • Homogeneous land surface &

subsurface.

  • 30 x 20 km2 and resolution of 1km.
  • Atmosphere has 50 vertical layers,

with 20m resolution near surface.

  • Subsurface has 30 vertical layers

stretching until 30m.

  • Periodic lateral BCS /impermeable

lower BCS,

slide-36
SLIDE 36
  • 4. Juni 2018

Folie 36 Agrosphere Insitute (IBG-3)

Spin-up

  • 100 days of spin-up (Feb 1- May 7, 2008).
  • Initial ground and vegetation temperature 283 K.
  • Initial groundwater table depth 3m, hydrostatic profile.
  • External forcing by COSMO-DE reanalysis data.
  • Other parameters deterministic.
slide-37
SLIDE 37
  • 4. Juni 2018

Folie 37 Agrosphere Insitute (IBG-3)

Ensemble generation

  • 48 ensemble members.
  • Spatially variable fields of saturated hydraulic conductivity.
  • LAI, soil color, clay percentage, leaf carbon-nitrogen ratio randomly

perturbed (but spatially constant).

  • Turbulent mixing scale parameter.
  • Other parameters deterministic.
slide-38
SLIDE 38
  • 4. Juni 2018

Folie 38 Agrosphere Insitute (IBG-3)

Spatially variable hydraulic conductivity

slide-39
SLIDE 39
  • 4. Juni 2018

Folie 39 Agrosphere Insitute (IBG-3)

DA experiments

  • Atmospheric DA: Atmospheric temperature at 10, 100, 200, 500, 1000,

3000 and 5000m.

  • Land surface DA: Soil temperature at 2, 6, 10, 20, 30, 50, 80 cm depth.
  • Subsurface DA: Soil moisture at 2, 6, 10, 20, 30, 50 and 80 cm depth.
  • Observation variances: 0.60 K2, 0.10 K2 and 0.005.
  • Daily assimilation for 10 locations in space.
slide-40
SLIDE 40
  • 4. Juni 2018

Folie 40 Agrosphere Insitute (IBG-3)

Characterization atmospheric states

  • Only atmospheric

DA improves characterization of boundary layer potential temperature.

slide-41
SLIDE 41
  • 4. Juni 2018

Folie 41 Agrosphere Insitute (IBG-3)

Characterization soil temperature

  • Assimilation of soil

temperature improves soil temperature, but less for upper 20 cm.

  • Assimilation of

atmosperic temperature improves soil temperature for upper layers.

slide-42
SLIDE 42
  • 4. Juni 2018

Folie 42 Agrosphere Insitute (IBG-3)

Characterization soil moisture

  • Assimilation of soil

moisture improves soil moisture.

  • Assimilation of soil

temperature has also an impact on improving soil moisture characterization.

slide-43
SLIDE 43

Mitglied der Helmholtz-Gemeinschaft

4 June 2018 IBG-3: Agrosphere 43

Conclusions and outlook

  • TerrSysMP-PDAF: DA-framework optimally suited for HPC.
  • Cosmic ray probe data very promising for land surface DA.
  • GW-level data have high potential to improve root zone soil moisture

characterization (under certain conditions) using fully coupled DA.

  • Weakly coupled atmospheric-land surface- subsurface DA tested with

different impacts of different observations. New test for drier conditions.

  • DA with fully coupled model does not allow for compute intensive

alternative DA-methods ((iterative) smoothers, PF).

  • Current work: extension of coupled DA. See also poster by Natascha

Brandhorst.

slide-44
SLIDE 44

Mitglied der Helmholtz-Gemeinschaft

Thanks for your attention!