assimilation of screen-level observations in the … · assimilation of screen-level observations...

1
Assimilation of Screen-Level Observations in the Canadian Land Data Assimilation System Bernard Bilodeau 1 ,Marco L. Carrera 1 , Sarah Dyck 2 , Nathalie Gauthier 2 , Maria Abrahamowicz 2 and Stéphane Bélair 1 1 Meteorological Research Division, Environment Canada 2 Canadian Meteorological Centre • CaLDAS is being coupled with the atmospheric En-Var data assimilation system • Experiments at intermediate 25-km resolution on global uniform lat-lon grid will soon be conducted for year 2011 • En-Var will provide forcings every hour • In return, CaLDAS will generate w2, T2 and snow depth analyses every 3 hours • The system will be two-way coupled, as opposed to the one-way coupling tests results shown here Final tests (early 2013) to be done on 15-km Yin-Yang global grid • Precipitation forcing: an ensemble of CaPA precipitation analyses is obtained by perturbing observations and by adding a random phase shift to the first guess (which is also applied to the radiative forcing). • Temperature (T) and dewpoint temperature (Td) innovations are randomly perturbed (constant value for whole domain). • Soil moisture and soil temperature analyses are randomly perturbed (again with constant value). • These perturbed forcing, innovations and analyses generate the ensemble spread. Global coverage Yin and Yang independent grids, with minimal overlap 901x301 grid points for each grid • 0.3º resolution Global analysis is recomposed by interpolation The Caldas suite is being converted from SMS to the new CMC operational sequencer, Maestro, in preparation for the upcoming implementation to CMC operations (2013). Sequencing Observation error ISBA LAND-SURFACE MODEL OBS ASSIMILATION x b y (with ensemble Kalman filter approach) xa = xb+ K { y – H(xb) } K = BHT ( HBHT+R)-1 with IN IN OUT OUT Ancillary land surface data Atmospheric forcing Observations Land surface initial conditions for NWP and hydro systems Land surface conditions for atmospheric assimilation systems Current state of land surface conditions for other applications (agriculture, drought, ...) Screen-level (T, Td) Surface stations snow depth L-band passive (SMOS,SMAP) MW passive (AMSR-E) Multispectral (MODIS) Combined products (GlobSnow) T, q, U, V, Pr, SW, LW Orography, vegetation, soils, water fraction, ... YIN-YANG computational grids Global domain, 33 km for Global Deterministic Prediction System (GDPS) Observations: ensemble of screen-level temperature and dewpoint temperature analyses, generated by Optimal Interpolation Control variables: deep soil moisture (w2) and temperature (T2) • 24 members Each member produces a snow depth analysis using Optimal Interpolation Assimilation step: 3 hours From April 2008 to February 2009 YIN YANG BIAS STDE Impact of CaLDAS soil moisture and soil temperature analyses on screen-level forecasts scores (0-120h) of temperature (left) and dewpoint temperature (right) produced by the GDPS for summer 2008 (25 runs starting at 00 UTC). Operational in blue, CaLDAS in red. Warm and dry biases are reduced, and nighttime errors are smaller. Surface scores over Canada Precipitation scores over North America SMS Maestro CaLDAS OPER T Td Upper air scores over North America Impact of CaLDAS soil moisture and soil temperature analyses on upper air 48-hour forecasts produced by the GEM global model for summer 2008 (25 runs starting at 00 UTC). Scores show cooler and wetter forecasts are generated with CaLDAS analyses from surface up to 700 hPa. Impact of CaLDAS soil moisture and soil temperature analyses on precipitation bias and threat scores produced by the GDPS for summer 2008 (25 runs starting at 00 UTC). Verified against synoptic stations. Migration Coupling with the Global Deterministic Prediction System Configuration of the system Perturbations to the ensemble members CaLDAS in a nutshell hours hours hours hours T GZ ES U V P-O 48 hr T Td • Taken from the spread of the ensemble of screen level analyses Example shown for 00 UTC 18 July 2008 The Canadian Land Data Assimilation System (CaLDAS) aims at improving the land surface modeling in environmental prediction systems The high efficiency of the offline modeling system allows the use of the Ensemble Kalman Filter (EnKF) technique The goal of the current project is to replace the current operational pseudo-analysis of soil moisture and soil temperature Instead of statistically correcting soil moisture and soil temperature based on screen level forecast errors, the EnKF does so dynamically STDE BIAS OPERATIONAL CaLDAS Some diagnostics of the system • For a perfectly representative ensemble, the Reduced Centered Random Variable (RCRV) will have a standard normal distribution N(0,1). • The RCRV can be thought of as the mean innovation normalized by its expected variance. ) ( var( ) var( ) ( f f x H O x H O RCRV + - = Candille, G. C. Côté, P.L. Houtekamer and G. Pellerin, 2007: Verification of an Ensemble Prediction System against Observations. MWR, 135, 2688-2699. Mean (left) and standard deviation (right) of time series of RCRV for temperature Same, for dewpoint temperature

Upload: doankhanh

Post on 27-Jul-2018

227 views

Category:

Documents


0 download

TRANSCRIPT

Assimilation of Screen-Level Observations in the Canadian Land Data Assimilation System

Bernard Bilodeau1,Marco L. Carrera1, Sarah Dyck2, Nathalie Gauthier2,Maria Abrahamowicz2 and Stéphane Bélair1

1Meteorological Research Division, Environment Canada2Canadian Meteorological Centre

• CaLDAS is being coupled with the atmospheric En-Var data assimilation system• Experiments at intermediate 25-km resolution on global uniform lat-lon grid will soon be conducted for year 2011• En-Var will provide forcings every hour• In return, CaLDAS will generate w2, T2 and snow depth analyses every 3 hours• The system will be two-way coupled, as opposed to the one-way coupling tests results shown here• Final tests (early 2013) to be done on 15-km Yin-Yang global grid

• Precipitation forcing: an ensemble of CaPA precipitation analyses is obtained by perturbing observations and by adding a random phase shift to the first guess (which is also applied to the radiative forcing).

• Temperature (T) and dewpoint temperature (Td) innovations are randomly perturbed (constant value for whole domain).

• Soil moisture and soil temperature analyses are randomly perturbed (again with constant value).• These perturbed forcing, innovations and analyses generate the ensemble spread.

• Global coverage• Yin and Yang independent grids, with

minimal overlap• 901x301 grid points for each grid• 0.3º resolution• Global analysis is recomposed by

interpolation

The Caldas suite is being converted from SMS to the new CMC operational sequencer, Maestro, in preparation for the upcoming implementation to CMC operations (2013).

Sequencing

Observation error

ISBALAND-SURFACE

MODEL

OBS

ASSIMILATION

xb

y (with ensemble Kalman filter

approach)

xa = xb+ K { y – H(xb) }

K = BHT ( HBHT+R)-1

with

ININ OUTOUT

Ancillary land surface

data

Atmospheric forcing

Observations

Land surface initial conditions for NWP and hydro systems

Land surface conditions for atmospheric

assimilation systems

Current state of land surface conditions for

other applications (agriculture, drought, ...)

Screen-level (T, Td)Surface stations snow depthL-band passive (SMOS,SMAP)MW passive (AMSR-E)Multispectral (MODIS)Combined products (GlobSnow)

T, q, U, V, Pr, SW, LW

Orography, vegetation, soils, water fraction, ...

YIN-YANG computational grids

• Global domain, 33 km for Global Deterministic Prediction System (GDPS)

• Observations: ensemble of screen-level temperature and dewpoint temperature analyses, generated by Optimal Interpolation

• Control variables: deep soil moisture (w2) and temperature (T2)

• 24 members

• Each member produces a snow depth analysis using Optimal Interpolation

• Assimilation step: 3 hours

• From April 2008 to February 2009

YIN

YANG

BIAS

STDE

Impact of CaLDAS soil moisture and soil temperature analyses on screen-level forecasts scores (0-120h) of temperature (left) and dewpoint temperature (right) produced by the GDPS for summer 2008 (25 runs starting at 00 UTC). Operational in blue, CaLDAS in red. Warm and dry biases are reduced, and nighttime errors are smaller.

Surface scores over Canada Precipitation scores over North America

SMS Maestro

CaLDAS

OPER

T Td

Upper air scores overNorth America

Impact of CaLDAS soil moisture and soil temperature analyses on upper air 48-hour forecasts produced by the GEM global model for summer 2008 (25 runs starting at 00 UTC). Scores show cooler andwetter forecasts are generated with CaLDAS analyses from surfaceup to 700 hPa.

Impact of CaLDAS soil moisture and soil temperature analyses on precipitation bias and threat scores produced by the GDPS for summer 2008 (25 runs starting at 00 UTC). Verified against synoptic stations.

Migration

Coupling with theGlobal Deterministic Prediction System

Configuration of the system

Perturbations to the ensemble members

CaLDAS in a nutshell

hours hours

hourshours

TGZ

ES

U VP-O 48 hr

T Td

• Taken from the spread of the ensemble of screen level analyses

• Example shown for 00 UTC 18 July 2008

• The Canadian Land Data Assimilation System (CaLDAS) aims at improving the land surface modeling in environmental prediction systems

• The high efficiency of the offline modeling system allows the use of the Ensemble Kalman Filter (EnKF) technique

• The goal of the current project is to replace the current operational pseudo-analysis of soil moisture and soil temperature

• Instead of statistically correcting soil moisture and soil temperature based on screen level forecast errors, the EnKF does so dynamically

STDE

BIAS

OPERATIONAL

CaLDAS

Some diagnostics of the system• For a perfectly representative ensemble, the Reduced Centered Random

Variable (RCRV) will have a standard normal distribution N(0,1).

• The RCRV can be thought of as the mean innovation normalized by its expected variance.

)(var()var()(

f

f

xHOxHO

RCRV+

−=

Candille, G. C. Côté, P.L. Houtekamer and G. Pellerin, 2007: Verification of an Ensemble Prediction System against Observations. MWR, 135, 2688-2699. Mean (left) and standard deviation (right) of time series of RCRV for temperature Same, for dewpoint temperature