Arctic System Reanalysis: Land Surface Parameter Assimilation and Model Updates

1Polar WRF Workshop – 3 November 2011

Arctic System Reanalysis:Land Surface Parameter Assimilation

and Model Updates

Michael BarlageResearch Applications Laboratory (RAL)

National Center for Atmospheric Research

Research funded by NSF (ARC-0733058)

2

− Land surface state spin-up: more consistent initialization, less time for soil states in lower boundary to equilibrate

− Changes to model structure: add more and deeper soil layers, zero-flux lower boundary condition

− Land surface parameter and state assimilation: snow cover and snow depth, albedo, and green vegetation fraction inserted into model daily/weekly

Enhancements/Additions to WRF within ASR

3

− Why is this necessary?

− Land surface models have their own climatology

− Soil layers depths between models may be inconsistent

− Vegetation types, soil types, terrain, etc. are likely different between models

− Land surface equilibrium can take over one year for 4-layer soil with 2 meter depth

− Five year spin-up for 10-layer with 8.5 meter depth

Land Surface State Spin-up

4

− Use High Resolution Land Data Assimilation System (HRLDAS) with atmospheric forcing from reanalysis

− HRLDAS: uses WRF model grid and static fields (land cover, soil type, parameter tables) to run an offline version of the Noah LSM

− Use 6-hourly reanalysis output (precipitation, wind, temperature, pressure, humidity, shortwave and longwave radiation) from ERA-40 (1980 – 1999) and JRA-25 (2000 – 2009)

− Spatially interpolate forcing fields to WRF grid and adjust temperature for terrain height differences between reanalysis and WRF

− Use hourly timestep by linearly interpolating all but solar radiation; the total 24hr radiation is fit to a daily zenith angle curve

− Advantages are that initial fields (especially soil ice/moisture/temperature):

− are already on the WRF grid

− are consistent with terrain, land cover and soil types/levels

− are consistent with WRF land model


5

− August 2008 volumetric soil moisture in top and bottom layer for ERA-I initialization (black) and HRLDAS multi-year simulation (red)

− Region average near 64N, 158E (NE Siberia)

− Land models have their own climatology

− HRLDAS soil moisture is more likely to be in equilibrium for WRF cold start

− Especially important for cycling runs


1.5m

1.5m

5cm

5cm

6

Comparison of HRLDAS Initial Soil Temperature

10 - 40 cm

40 - 100 cm

7

− The default WRF model uses the Noah land surface model with four soil layers that have nodes at 0.05m, 0.25m, 0.7m, and 1.5m and a fixed deep soil (8m/25m) temperature

− It has been suggested that the fixed deep soil temperature is likely too low over much of the Arctic because it is based on annual mean air temperature

− Within the ASR WRF model, the Noah LSM is modified to have 10 soil layers and a free, zero-flux lower boundary condition (3 subroutine + namelist changes)

− The 10 soil layers have interfaces at 0.05m, 0.15m, 0.25m, 0.4m, 0.65m, 1.05m, 1.7m, 2.75m, 4.45m and 7.2m

− For example, below is the 60-70N average bottom 10-layer T vs 4-layer 8m fixed T

Changes to Land Model Structure

4-layer fixed 8m T

10-layer 7.2m T

8

Changes to Land Model Structure

− Difference between lowest layer (7.2m) temperature [K] after a 28 year simulation and the assumed 8m deep soil temperature in standard WRF

− Most of the Arctic region is much warmer in the 10-layer zero-flux model

− Implications for soil temperature/moisture related processes, e.g., permafrost prediction

9

Alaska Measuring Stations

1km2 measurement gridwith 121 points 100m apart

10

HRLDAS Simulation Specifics

27-year (1980-2006) point simulations over CALM measurement sites

Forcing data: ERA-40 (1980-1999); JRA-25 (2000-2006)

layers_control = (/0.05,0.25,0.70,1.5/)layers_zeroflux = layers_controllayers_stagger = (/0.05,0.15,0.25,0.40,0.65,1.05,1.70,2.75,4.45,7.20, \ 11.65,18.85/)layers_constant = (/0.05,0.25,0.70,1.5,2.5,3.5,4.5,5.5,6.5,7.5, \ 8.5,9.5,10.5,11.5,12.5,13.5,14.5,15.5,16.5,17.5/)layers_highres = (/0.01,0.03,0.05,0.07,0.09,0.11,0.13,0.15,0.17,0.19, \ 0.21,0.23,0.25,0.27,0.29,0.31,0.33,0.35,0.37,0.39, \ 0.425,0.475,0.525,0.575,0.625,0.675,0.75,0.85,0.95,1.1, \ 1.3,1.5,1.7,1.9,2.25,2.75,3.25,3.75,4.25,4.75/)layers_organic = layers_highres Organic layer (peat) in the top 12cm

11

Point Simulation: Active Layer Thickness

Black: controlBlue: zerofluxRed: staggerGreen: constantOrange: highresBrown-ish: organic

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

Act

ive

Laye

r Thi

ckne

ss[c

m]

12



Act

ive

Laye

r Thi

ckne

ss[c

m]

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

13

Point Simulation: Temperature Profiles


JulyJanuary

14



JulyJanuary

15

Point Simulation: Snow Depth


Snow depth much too low

Sno

w D

epth

[cm

]

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

16


JulyJanuaryBlack: highresBlue: organicRed: organic_2xGreen: organic_4x

Artificially add precipitation to get deeper snow

17

Point Simulation: Snow Depth

Black: highresBlue: organicRed: snow z0

Green: herb. tundra

Changing zo over snow covered tundra brings model in

line with observations

Sno

w D

epth

[cm

]

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

18


JulyJanuaryBlack: organicBlue: snow zo

Red: herb. tundra

19


control_bias = 57.0highres_bias = 40.7snowz0_bias = 7.1

Black: highresBlue: organicRed: snow z0

Green: herb. tundra

Act

ive

Laye

r Thi

ckne

ss[c

m]

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

20

Slope-Aspect Adjustment for ASR Domain

• Tested slope and aspect adjustment based on terrain

• Bin results based on cardinal directions: North (-45°- 45°), etc.

• Results are consistent with terrain structures in domain

21

Slope-Aspect Adjustment for ASR Domain

• However, if slopes < 1° are masked, the resulting locations where slope-aspect adjustment would make a difference are minimal

• 15km grid is too coarse to necessitate adjustment

22

Data assimilation - infrastructure added to HRLDAS/WRF(+WRF-Var) to include:

- IMS snow cover: daily, 2004 to current at 4km; pre-2004 at 24km; this product is used operationally at NCEP

- SNODEP snow depth: daily, obs/model product; on GFS T382 (~30km) grid; used as guidance to put snow where IMS says snow exists

- MODIS albedo: 8-day 0.05º global; available from Feb 2000; also use MODIS snow cover and cloud cover

- NESDIS vegetation fraction: weekly, 0.144º global; transitioning to use in NCEP operations

- MODIS daily albedo over Greenland: ~1km, available over MODIS period

- Greenland terrain provided by Ohio State

Products are assimilated into the wrfinput file at 00Z of each cycle

Assimilation Products

23

• Product created in near real-time by NESDIS/STAR• Based on smoothed AVHRR NDVI product to remove satellite drift and sensor

degradation• GVF(t) = (NDVI(t) – NDVImin)/(NDVImax – NDVImin)• Available as a 7-day product from 1984 to present• Very similar procedure to existing WRF climatological vegetation so use product

directly after interpolation to WRF grid

Assimilation Procedure: Green Vegetation Fraction

Vegetation Fraction on 0.144° global grid

Use WPS to reproject to WRF grid

Create minimum and maximum file

Interpolate 7-day product to daily

24

Product Comparison: Green Vegetation Fraction

Qualitative comparison to Drought Monitor

August 24, 2004 July 18, 2006

2004 2006 2009• 2004: largest “D2” area• 2006: not significant statewide

but dry in eastern Alaska• 2009: small spike in “D2” but all

concentrated on southern coast; east has no drought

2000

GVF Timeseries for east-central Alaska

25

Albedo highly dependent on snow so how to use MODIS albedo to be consistent with current model state

Assimilation Procedure: MODIS Albedo

MODIS 8-day albedo on 0.05° grid

MODIS 8-day TERRA and AQUA

snow coverMODIS 8-day T/A

cloud cover

Create a snow-free (<1%) and snow-covered(>70%)

climatological dataset (cloud <50%)

Starting with climatology move forward in time replacing with current snow-free or snow-

covered albedo (cloud < 80%); repeat backward in time

Use WPS to reproject MODIS snow-free and snow-covered

albedo to WRF grid

1

2

2

26

Data Generation Procedure: MODIS Albedo

− Develop snow-covered and snow-free albedo based on MODIS albedo and snow cover products

MODIS Terra/ Aqua snow cover

MODIS albedo and running

min/max

27

Use IMS daily snow cover to determine snow coverage and SNODEP daily snow depth as guidance for quantity

Assimilation Procedure: Snow

IMS daily 4km/24km snow

cover

Air Force SNODEP 32km snow depth


1. If IMS < 5%, remove snow if present2. If IMS > 40% and SNODEP > 200% model snow or < 50% model

snow, use existing model snow density to increase/decrease model snow by half observation increment

3. If IMS > 40%, don’t let SWE go below 5mm independent of SNODEP

Run both products through a 5-day median smoother to remove snow “flashing”


28

− Seven-month HRLDAS run with land data assimilation

− Region near 69N, 155W (North Slope)

− Model albedo agrees better with MODIS albedo

− SNODEP snow is inconsistent with IMS snow cover in June

− Report snow increments so users can recreate model snow

MODIS Albedo

Datasets

Snow Depth

Results

Albedo Time series

Snow cover and depth

29

Saw some questionable albedo variations in the standard MODIS albedo product over Greenland

High summer albedo and relative low winter albedo is opposite time variation than expected

Greenland MODIS albedo

2001 2002 2003 2004 2005

winter

summer

30

A new daily MODIS-based albedo dataset was provided by Ohio State with higher resolution compared to current MODIS albedo datasets

New Greenland MODIS albedo

31

New MODIS albedo dataset (red) shows a more realistic annual cycle than the original dataset

This dataset is assimilated as the snow covered albedo over Greenland only

Greenland MODIS albedo

32

A new terrain dataset was provided by Ohio State with higher resolution laser altimeter data (Bamber et al 2001) and accuracy compared to standard WRF geogrid data

Note the non-zero terrain outside of Greenland

Two likely causes: different reference geoids and ocean height differences

WRF uses sphere-based definition of sea level

Land terrain also not based from sea level (sphere)

How to use this?

New Greenland Terrain

33

Final product


34

Difference between new and default

Points surrounding Summit (3216m)72.744 -38.6521 3180.68 3195.1172.5384 -38.0021 3181.18 3219.1772.3445 -38.6831 3118.9 3189.7472.1392 -38.0475 3107.8 3200.04


35

Summary

− Land surface state spin-up: use 20+ years of reanalysis data to make land states more consistent with model, land cover, terrain, and soil type

− Changes to model structure− use 10 soil layers instead of the default 4 layers− soil layers go down to ~7m instead of 1.5m− zero-flux lower boundary condition to improve on fixed lower temperature

− Land surface parameter and state assimilation − snow cover (satellite) and snow depth (in situ/model) − albedo (MODIS satellite)− green vegetation fraction (AVHRR satellite) − parameters/states updated daily/weekly

36

Test Simulation

− WRF-3DVAR simulation− 6 hour cycling− 3 hour obs time window− January 2007− 60km− Physics options

− Morrison MP− MYNN− Grell 3D− Noah LSM

− Land surface parameter and state assimilation − snow cover and

snow depth − Albedo max/min

(MODIS satellite)− green vegetation

fraction− Observations

− METAR T2m

− SYNOP T2m

37

Comparison to SYNOP 2-meter Temperature

2.89 3.182.78 3.17

n=10

-0.48 -0.241.01 0.91

n=17

0.11 1.601.03 2.44

n=115

0.09 0.981.24 2.13

n=17

-3.24 -2.36-2.09 -1.50

n=1

0.53 0.470.60 0.42

n=21

1.27 1.250.88 0.87

n=111

0.93 0.34-0.16 -0.50

n=89

3.10 2.212.91 2.02

n=89

5.55 4.595.48 4.36

n=33

4.75 2.364.13 1.69

n=29

2.87 1.131.03 -0.03

n=5

Net positive results: Improved bias in 32 of 48 region/times

00Z 12Z 00Z 12Z 00Z 12Z 00Z 12Z 00Z 12Z 00Z 12Z

Arctic System Reanalysis: Land Surface Parameter Assimilation and Model Updates

Documents

layer soil

soil states

soil icemoisturetemperature

land surface models

noah land surface model

land model structure4layer

land model structuredifference

deeper soil layers