Assimilation of Geostationary Satellite Land Surface Skin - - PowerPoint PPT Presentation

Assimilation of Geostationary Satellite Land Surface Skin Temperature Observations into the GEOS-5 Global Atmospheric Modeling and Assimilation System

Clara Draper1,2, Rolf Reichle2, Gabrielle De Lannoy1,2, and Qing Liu3

• 1. GESTAR, Universities Space Research Association.
• 2. Global Modeling and Assimilation Office, NASA Goddard Space Flight Center.
• 3. Science Systems and Applications, Inc.

6th WMO DA Symposium - October 9, 2013

1 / 16

Project outline

◮ Aim: assimilate geostationary Tskin observations into the land

surface of the GEOS-5 GCM/Atmospheric DA system

◮ Enhance assimilation of surface-sensitive atmospheric radiances ◮ Improve land surface flux forecasts

◮ Coupled GMAO’s EnKF-based land data assimilation system to

the GEOS-5 GCM/ADAS

2 / 16

Tsurf in GEOS-5 Catchment land surface model

◮ Catchment Tsurf is the average temperature of the canopy and

soil surface (represented by an arbitrarily thin layer with minimal heat capacity) Diffusive heat flux b’ween soil layers

RN LH + SH Tsurf

dTsurf dt

=

1 shc (RN − LH − SH − G)

RN = (1 − α)RS↓ + ... ǫ(RL ↓ −σs(Tsurf)4) LH = rL(esat(Tsurf) − eair) SH = rH(Tsurf − Tair)

Surface specific heat capacity (shc): 200 J/K (70,000 J/K for broad-leaf e’green)

G

3 / 16

Geostationary Tskin data

◮ Near-real time geostationary Tskin data set from NASA Langley

Research Center (LaRC)

◮ TIR clear sky observation of the effective radiative temperature of

the land surface

◮ GOES-E, GOES-W, Meteosat-9, MTSAT-2, FengYun-2E ◮ Comparable accuracy to MODIS (vs. in situ Tskin) ◮ Currently 3-hourly (clear sky) at 0.25◦ resolution GOES-E Tskin vs. SGP ARM IRT MODIS Tskin vs. SGP ARM IRT Scarino, B., Minnis, P., Palikonda, R.,Reichle, R., Morstad, D., Yost, C., Shan, B., and Liu, Q. (2013), Retrieving surface skin temperature for NWP applications from global geostationary satellite data, Rem. Sens. 4 / 16

Correcting forecast-observation biases in coupled LA-DAS

5 / 16

Forecast-observation biases

Catchment − GOES−W Tskin August 2012, 18:00 UTC [K] −120° − 80° 40° −10 10 Catchment − GOES−E Tskin August 2012, 18:00 UTC [K] −120° − 80° 40° −10 10

◮ Large forecast-observation biases (ubiquitous in land DA) ◮ Unknown if bias is in forecasts and/or observations (likely both) ◮ Common in land DA to assign f’cast-obs bias to observations

◮ At least ensures that the f’cast and obs are not biased relative to

each other

◮ Usual methods require long data record to estimate forecast and

• bserved climatological statistics

◮ For assimilation into an atmospheric system (frequent model

updates!), do not have a long data record

6 / 16

Observation bias and state estimation

◮ Similar to forecast bias correction of Dee and Todling [2000] ◮ State forecast and update:

x−

k,i = f (x+ k−1,i, qk,i)

x+

k,i = x− k,i + Kk(˜

y o

k − b+ k − Hkx− k,i)

( K is unchanged by inclusion of bias estimate)

◮ Bias forecast and update:

b−

k = b+ k−1

b+

k = b− k + Lk(˜

y o

k − b− k − < Hkx− k >) ◮ Simplify by replacing L with empirical Λ, designed to update bias

more aggressively when observations are available less frequently

7 / 16

The estimated bias

Estimated f’cast-obs bias (K) after one month (30 June 2012)

06 UTC 18 UTC 12 UTC 00 UTC 8 / 16

Offline assimilation results

9 / 16

Comparison to in situ observations

◮ Assimilated 1 year of GOES-E/W over North America into

GEOS-5 land surface model, forced with MERRA atmospheric analyses

1 1.5 2 2.5 1 1.5 2 2.5 Anomaly RMSD (K) to SURFRAD Tskin # 6 Openloop (mean: 1.61K) TIR Tskin assimilation (mean 1.51K) 0.2 0.4 0.6 0.8 1 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

• Anom. Corr. (USCRN Tair,Catch Tsurf), 15 UTC, # 60
• Anom. Corr. (Open tile, USCRN), mean: 0.80
• Anom. Corr. (Assm tile, USCRN), mean: 0.82

10 / 16

Comparison to MODIS Tskin: Aqua asc. (∼18UTC)

Anomaly RMSD (K) over JJA 2012

a) Offline GEOS-5 (mean: 2.9 K) c) GOES-E/W bias corrected to GEOS-5 (mean: 3.5 K) b) GOES-E/W (mean: 2.5 K) Improvement from assimilation of bias-corrected GOES (mean: 0.15 K, 65% +ve) 11 / 16

Comparison to MODIS Tskin: Aqua dsc. (∼06UTC)

Anomaly RMSD (K) over JJA 2012

a) Offline GEOS-5 (mean: 1.6 K) c) GOES-E/W bias corrected to GEOS-5 (mean: 1.8 K) b) GOES-E/W (mean: 1.3 K) Improvement from assimilation of bias-corrected GOES (mean: 0.14 K, 78% +ve) 12 / 16

Assimilation into the GEOS-5 atmospheric system

13 / 16

03Z 09Z 15Z 21z PREDICTOR CORRECTOR ATM B’GROUND A-IAU

b b b b b b b b b b b b b b b

L-IAU ATM FORCING

14 / 16

Summary/conclusions

◮ Assimilated TIR skin temperature observations into GEOS-5 land

surface model

◮ Introduced a simple observation bias and state estimation scheme

for use with atmospheric system (does not require long data record)

◮ Offline assimilation of GOES TIR Tskin, with f’cast-obs bias

correction, shows consistent small improvement in model short-term variability

◮ Is correction of short-term variability enough to enhance

atmospheric assimilation / improve land surface fluxes ????

◮ Results with GEOS-5 GCM/ADAS pending...

15 / 16

THANK YOU FOR LISTENING.

16 / 16

Catchment surface energy states and fluxes

∆Z(m) 0.10 0.19 0.38 0.76 1.5 10.0 TP1 TP2 TP3 TP4 TP5 TP6 zero heat flux boundary

Diffusive heat flux TPn =

ghtcntn+icehctn shc(rck)+shc(wtr)+shc(ice)

Water assumed 0.5φ. If no ice: TPn = ghtcnt/(2269050∆Zn)

RN LH + SH T SAT

C

T TRANS

C

T WILT

C

TSURF = w(T SAT

C

,T TRANS

C

,T WILT

C

) ∆(T X

C ) = ∆(W X) shc(sfc)

shc=200 J/K,

• r 70,000 J/K for b-l e’green

dW dt = RN − LH − SH − G

RN = (1 − α)RS↓ + ǫ(RL ↓ −σs(T X

C )4)

LH = RESISTL(esat(T X

C ) − eair)

SH = RESISTH(T X

C − Tair)

G

17 / 16

Two-stage observation bias and state estimation: simplified

◮ Replace L with empirical Λ:

      λ1 λ2 ... λn−1 λn       where λj = 1 − e−∆tj/τj

◮ ∆t is time since last observation ◮ τ is time scale of bias memory (10 days for Tskin assim.) 18 / 16

Enhance impact by relaxing ‘clear-sky’ cloud fraction for TIR Tskin?

−5 5 100 200 300 400 500 600 GOES−E − SURFRAD Tskin anomaly (K) # of 0.25° grid cells (May 2012−Apr 2013) Impact of cloud cover on GOES−E − SURFRAD Tskin anomaly All retrieved GOES−E data Cloud fraction < 5% −5 5 100 200 300 400 500 600 GOES−E − SURFRAD Tskin anomaly (K) # of 0.25° grid cells (May 2012−Apr 2013) Impact of cloud cover on GOES−E − SURFRAD Tskin anomaly All retrieved GOES−E data Cloud fraction < 20%

19 / 16