Ensemble data assimila!on for soil-‐ vegeta!on

Ensemble data assimila4on for soil-­‐
vegeta4on-­‐atmosphere systems. Tim Hoar: Na#onal Center for Atmospheric Research (NCAR) with a lot of help from: Jeff Anderson: NCAR Andrew Fox: Na#onal Ecological Observatory Network (NEON) Yongfei Zhang: University of Texas Aus#n Rafael Rosolem: University of Bristol/University of Arizona Mo4va4on 1.  The ecological state of the planet is the result of an unknowable history. 2.  Model spinup cannot be counted on to accurately re-­‐
create that history. 3.  Data assimila4on can put the model state more in line with the current state. With that, we can: •  Quan4fy ecological states •  to establish a baseline •  as a preface for ecological forecas4ng •  Be>er understand our models •  Improve our understanding of the underlying processes TJH AMS 2014 pg 2
What is Data Assimila4on? Observa4ons combined with a Model forecast… +
=
[email protected] for more! … to produce an analysis. Overview ar4cle of the Data Assimila4on Research Testbed (DART): Anderson, Jeffrey, T. Hoar, K. Raeder, H. Liu, N. Collins, R. Torn, A. Arellano, 2009: The Data Assimila4on Research Testbed: A Community Facility. Bull. Amer. Meteor. Soc., 90, 1283–1296. doi:10.1175/2009BAMS2618.1 TJH AMS 2014 pg 3
My difficul4es with ensemble land DA: • 
• 
• 
• 
• 
• 
What parts of the model ‘state’ do we update? What is, and how do we get, a proper ini:al ensemble? Represen:ng proper uncertainty in the forcing fields. Model/observa4on bias … probably both wrong … Can models tolerate new assimilated states? Silently fail? Snow (vegeta4on) … depths, layers, characteris4cs, content. –  Destroying easier than crea4ng new •  Forward observa4on operators –  many flux observa4ons are over 4mescales that are inconvenient –  need soil moisture from last month and now … GRACE •  Impact of bounded quan44es on ensemble spread •  Observa:on metadata usually insufficient or hard to use. –  land cover type needed for accurate forward observa4on operators. TJH AMS 2014 pg 4
vegetated landunit.
Models that abstract the gridcell into a “nested gridcell hiearchy of of mul4ple landunits, snow/soil columns, and Plant Func4on Types” are par4cularly troublesome when trying to convert the model state to the expected observa4on value. Loca4on informa4on is contained at this level ONLY! Given a soil temperature observa4on at a specific lat/lon, which PFT did it come from? No way to know! Unless obs have more metadata! Observa4ons occur here! TJH AMS 2014 pg 5
Observa4ons: AMSR-­‐E Tb In collabora4on with Ally Toure, NASA AMSR−E Brightness Temperature
count
15
10
5
0
0
20
40
60
80
Observation Rank (among ensemble members)
Latitude
These are synthe4c results. The only way the rank histogram looks this good is if there is no bias between the obs and the ensemble. Real life interferes here … Innovations in Snow Cover Fraction
0.05
50
Real observa4on converter complete, but 40
… DENSE observa4ons. Superob? Correlated? 30
h() depends on land cover? 0.00
220
TJH AMS 2014 pg 6
240
260
280
Longitude
300
−0.05
In collabora4on with Andy Fox (NEON): An experiment at Niwot Ridge • 
• 
• 
• 
• 
• 
• 
• 
9.7 km east of the Con4nental Divide C-­‐1 is located in a Subalpine Forest (40º 02' 09'' N; 105º 32' 09'' W; 3021 m) One column of Community Land Model (CLM) •  Spun up for 1500 years with site-­‐specific informa4on. 64 ensemble members Forcing from the DART/CAM reanalysis, Assimila4ng tower fluxes of latent heat (LE), sensible heat (H), and net ecosystem produc4on (NEP). Impacts CLM variables: LEAFC, LIVEROOTC, LIVESTEMC, DEADSTEMC, LITR1C, LITR2C, SOIL1C, SOIL2C, SOILLIQ … all of these are unobserved. TJH AMS 2014 pg 7
FYI: We are assimila4ng flux observa4ons … Sensible heat Net Ecosystem Produc4on Focus on the ensemble means (for clarity) Latent heat The model states are being updated at about 8PM local 4me. TJH AMS 2014 pg 8
Free Run Assim These are all unobserved variables. June 2004 TJH AMS 2014 pg 9
Leaf Area Index Leaf Area Index (m2/m2) This is the result of a synthe4c experiment. Experiments with MODIS LAI observa4ons are in progress. One representa4ve gridcell Carbon pools from all grid cells are in the DART state vector, and are updated through the covariance matrix, propaga#ng informa#on from sites to regions. 1 Oct 2005 1 July 2005 TJH AMS 2014 pg 10
Cosmic Ray-Derived Soil Moisture
410$m$
Rafael Rosolem, U. of Arizona, U. of Bristol
COSMOS Probe Footprint
0.5$m$
600$m$
TJH AMS 2014 pg 11
NOAH-­‐DART Integrated Soil Moisture Daily Averages
(m33 m
SθS (m
m-3)-­‐3) 0.25
0.3
Open Loop
Assim
TDT
0.2
0.15
Posterior Mean of 40 members 0.1
0.05
0.1
07/16/11
07/26/11
08/05/11
08/15/11
08/25/11
09/04/11
09/14/11
0.25
Open Loop
Assim
1:1
0.2
0.15
0.1
0.05
0
0
0.02 0.04 0.06
3 -3 0.08
NOAH S (m3 m -­‐3
)
TDT Sθ (m m ) Open Loop
Assim
TDT
Santa Rita site 0.08
0.06
3
-3
) -­‐3) Δ SS θ (m(mm3 m
0
07/06/11
3 3 -3 -­‐3
TDT S
S (m
)
NOAH ) θ (mm m
0.3
0.04
0.02
0
-0.02
07/06/11
07/16/11
07/26/11
08/05/11
08/15/11
08/25/11
09/04/11
TJH AMS 2014 pg 12
09/14/11
Raphael at Tonzi Ranch 0.1
Assimila:on of the MODIS Snow Cover Frac:on Dataset through the Coupled Data Assimila:on Research Testbed (DART) and the Community Land Model (CLM4)
Yongfei Zhang, Zong-­‐Liang Yang The University of Texas at Aus#n Tim Hoar, Jeffrey Anderson The Na#onal Center for Atmospheric Research Ally Toure, Ma>hew Rodell The Na#onal Aeronau#cs and Space Administra#on
TJH AMS 2014 pg 13
Assimila4on of MODIS snow cover frac4on Snow Cover Frac4on 4 Jan 2003 •  80 member ensemble for onset of NH winter •  assimilate once per day •  Level 3 MODIS – regridded to a daily 1 degree grid •  Observa4ons can impact state variables within 200km •  CLM variable to be updated is the snow water equivalent “H2OSNO” MODIS CLM4 TJH AMS 2014 pg 14
Impact on Snow Water Equivalent 4 Jan 2004 TEASER: Assimila4on of GRACE Total Water Storage & MODIS SCF minus control run Assimila4on of MODIS SCF minus control run 4 Jan 2004 TJH AMS 2014 pg 15
The HARD part is: What do we do when SOME (or none!) of the ensembles have [snow,leaves,precipitaDon, …] and the observaDons indicate otherwise? Corn Snow? New Snow? Sugar Snow? Dry Snow? Wet Snow? Crusty Snow? “Champagne Powder”? Slushy Snow? Old Snow? Dirty Snow? Packed Snow? Early Season Snow? Snow Density? Snow Albedo? The ensemble must have some uncertainty, it cannot use the same value for all. The model expert must provide guidance. It’s even worse for the hundreds of carbon-­‐based quan44es! TJH AMS 2014 pg 16
We are a>acking the problem of data assimila4on for soil-­‐vegeta4on-­‐atmosphere systems with many tools. •  Eddy Covariance fluxes •  Leaf Area Index •  snow cover frac4on •  Soil moisture •  Cosmic ray neutron counts •  Water table depth •  Total water storage (i.e. GRACE) TJH AMS 2014 pg 17
CAM
For more informa4on: GITM
WRF
CLM
POP
AM2
BGRID
COAMPS
www.image.ucar.edu/DAReS/DART NOAH
MPAS_ATM
[email protected] MITgcm_ocean
SQG
NAAPS MPAS_OCN
TIEGCM
NCOMMAS
TJH AMS 2014 pg 18
PE2LYR
COAMPS_nest
PBL_1d
My favorite land use …