Remote Sensing,
Land Surface Modelling and Data Assimilation

Jeffrey Walker

The University of Melbourne

Garry Willgoose and Jetse Kalma

The University of Newcastle

Paul Houser, Rolf Reichle and Randal Koster

NASA Goddard Space Flight Center

http://www.civag.unimelb.edu.au/~jwalker

Importance of Land Surface States
(soil moisture, soil temperature, snow)

Early warning systems

Flood prediction – infiltration, snow melt

Socio-economic activities

Agriculture – yield forecasting, management (pesticides etc), sediment transport

Water management – irrigation

Policy planning

Drought relief

Global change

Weather and climate

Evapotranspiration – latent and sensible heat

Albedo

Soil Moisture vs Sea Surface Temp

Knowledge of soil moisture has a greater impact on the predictability of summertime precipitation over land at mid-latitudes than Sea Surface Temperature (SST).

The Situation

The Problem With LSMs

Same forcing and initial conditions but different predictions of soil moisture!

Importance of Soil Moisture

Soil Moisture Coverage: Veg (Mean PR)

Soil Moisture Coverage: RFI (SD PR)

Data Assimilation Defined

Definition 1: using data to force a model
ie. precipitation and evapotranspiration to force a LSM

Continuous or Sequential DA?

Continuous (ie. variational)

Continuous or Sequential DA?

Sequential (ie. Kalman filter)

Extended or Ensemble KF?

DA as a Spatial Interpolator

Case 1: 1D Synthetic Study

(Walker et al., AWR 2001)

Synthetic Data

Direct Insertion Every Hour

Kalman Filter Update Every Hour

Kalman Filter update Every 5 Days

Case 1: Lessons Learned

Demonstrated the usefulness of near-surface soil moisture (and temperature) measurements.

Require a statistical assimilation scheme
(ie. a scheme which can potentially alter the entire profile).

Require as linear forecasting model as possible to ensure stable updating (ie. q-based model rather than a y-based model).

Case 2: 1D Field Study

(Walker et al., JHM 2001)

Field Instrumentation

Model Calibration/Evaluation

Kalman Filter Update Every 5 Days

Case 2: Lessons Learned

Updating of the model is only as good as the models representation of the soil physics (ie. need an unbiased model).

Porosity and residual soil moisture are the most important parameters as they set the bounds – better to use soil wetness?

Assimilation is most useful during dry-down as the model will reset itself at the extremes.

Case 3: 3D 6ha Field Study

(Walker et al., WRR In Review)

Modified Kalman Filter Application

Catchment Instrumentation

3D Model Calibration

3D Profile Retrieval

Case 3: Lessons Learned

The soil moisture profile cannot be retrieved when the deep soil layer becomes decoupled from the near-surface layer (ie. during extended drying periods).

Simulation results may be degraded slightly if simulation and observation values are already close.

The updating interval is relatively unimportant when using a calibrated model with accurate forcing.

Case 4: 1.5D Continental Synthetic Study

(Part 1: Demonstration)

(Walker and Houser, JGR 2001)

Catchment-based LSM

Catchment Discretisation

Synthetic Demonstration

“Errors” in Assimilated Moisture: 1

“Errors” in Assimilated Moisture: 2

Monthly Evapotranspiration

Monthly Runoff

Case 4.1: Lessons Learned

Demonstrated a methodology for generating soil moisture initialisation states.

Soil moisture assimilation performs best for regions with shallower soils; particularly depths less than 3m.

Important that all of the dominant physical processes in the LSM are identified and included in the covariance forecasting.

Structure of the model effects the retrieval time (ie. layered vs catchment deficit).

Case 4: 1.5D Continental Synthetic Study

(Part 2: Mission Requirements)

(Walker and Houser, WRR In Prep)

Motivation

What are the defensible requirements of a remote sensing mission for measurement of surface soil moisture?

Polarisation – horizontal

Wavelength – L-band

Look Angle – < 50°

Observation Accuracy – ?

Temporal Resolution – ?

Spatial Resolution – ?

Mission Requirements

Spatial Distribution of MC Error

Case 4.2: Lessons Learned

Typically, observations of surface soil moisture
must have an accuracy of better than 5% v/v.

The greatest impact of temporal resolution was for 1 to 5 days, with greater time between observations having a marginal degradation.

Observations at one-quarter to one-half the spatial resolution of the model/application is a good compromise.

To achieve the best results, it is important that the model forcing and observations be unbiased, or the bias modelled as part of the assimilation framework.

Case 4: 1.5D Continental
Synthetic Study

(Part 3: Extended v Ensemble KF)

(Reichle, Walker, Koster and Houser,
JHM Submitted)

Filter Calibration

Extended vs Ensemble KF

Case 4.3: Lessons Learned

EKF and EnKF both provided satisfactory estimates of soil moisture.

EKF cheaper (for 1.5D; equivilant to 4 ensembles), but EnKF more accurate for 6
(or more) ensemble members.

EnKF has the potential to include higher order moments than mean and std dev
(ie. skewness)

EnKF more promising for including horizontal correlations.

Case 5: 1.5D Continental
“Field” Study

(Work in Progress)

Land Surface Initialisation for Seasonal-to-Interannual Prediction

Goal: Provide global land surface initialisation states for the retrospective period of 1979 to present.

A two-tiered approach.

Tier 1: Spin-up the land surface model off-line from the GCM using the best atmospheric forcing data available. Minimises forcing bias!

Tier 2:Assimilate passive microwave remote sensing observations of near-surface soil moisture from SMMR and SSM/I. Minimises model and forcing bias!

Tier 1: Better Forcing

Observational Data Sets

NCAR Northern Hemisphere Sea Level Pressure

01/1899 – present; 5 x 5 degrees

Climate Research Unit (University of East Anglia) Temperature and Precipitation

01/01 - 12/98; 0.5 x 0.5 degrees

Center for Climatic Research (University of Delaware) Terrestrial Temperature and Precipitation

01/50 - 12/96 ; 0.5 x 0.5 degree

Global Precipitation Climatology Project (GPCP)

01/86 - 03/95; 2.5 x 2.5 degree

Langley Eight Year Shortwave and Longwave Surface Radiation Budget

07/83 - 06/91 - in process of being extended; 2.5 x 2.5 degree

Re-Analysis Atmospheric Forcing Data Sets

ECMWF Re-analysis Advanced Global Data

4x/day, 01/79 - 12/93

1.125 degrees (Gaussian)

NCEP/NCAR Re-analysis

4x/day, 01/48 – 12/99

2.5 x 2.5 degrees

Bias Correct Using Monthly Mean Observational Data Sets

Tier 2: Assimilation

SMMR Soil Moisture Observations

Soil Moisture Time Series: Illinois

Animation

Soil Moisture: Lat 50, Lon -100

Evaluation of Assimilation

How can we evaluate the soil moisture assimilation?

Soil moisture – ideal but limited data available. United states has 19 stations in Illinois, 6 stations in Iowa and transect of 89 points in New Mexico for SMMR period.

Analysis increments – only provides a check for systematic biases.

Runoff - data available but assumes that soil moisture is the only reason for poor estimates.

Evapotranspiration – data not available and assumes that soil moisture is the only reason for poor estimates.

Precipitation forecasts – assumes soil moisture is the only reason for poor forecasts.

Other ?

Final Words

Other assimilation work

Complete the global SMMR assimilation – Ni et al.

SMMR/AMSR assimilation Australia – Walker et al.

Continental snow assimilation – Sun et al.

TRMM assimilation – Entin et al.

G-LDAS – Rodell et al.

Runoff assimilation – Walker et al.

Thankyou!


	Jeffrey Walker
	The University of Melbourne

	Garry Willgoose and Jetse Kalma
	The University of Newcastle


	Paul Houser, Rolf Reichle and Randal Koster
	NASA Goddard Space Flight Center

	http://www.civag.unimelb.edu.au/~jwalker


	Early warning systems
		Flood prediction – infiltration, snow melt
	Socio-economic activities
		Agriculture – yield forecasting, management (pesticides etc), sediment transport
		Water management – irrigation
	Policy planning
		Drought relief
		Global change
	Weather and climate
		Evapotranspiration – latent and sensible heat
		Albedo


	Knowledge of soil moisture has a greater impact on the predictability of summertime precipitation over land at mid-latitudes than Sea Surface Temperature (SST).


	Same forcing and initial conditions but different predictions of soil moisture!


	Definition 1: using data to force a model ie. precipitation and evapotranspiration to force a LSM


	Demonstrated the usefulness of near-surface soil moisture (and temperature) measurements.
	Require a statistical assimilation scheme (ie. a scheme which can potentially alter the entire profile).
	Require as linear forecasting model as possible to ensure stable updating (ie. q-based model rather than a y-based model).


	Updating of the model is only as good as the models representation of the soil physics (ie. need an unbiased model).
	Porosity and residual soil moisture are the most important parameters as they set the bounds – better to use soil wetness?
	Assimilation is most useful during dry-down as the model will reset itself at the extremes.


	The soil moisture profile cannot be retrieved when the deep soil layer becomes decoupled from the near-surface layer (ie. during extended drying periods).
	Simulation results may be degraded slightly if simulation and observation values are already close.
	The updating interval is relatively unimportant when using a calibrated model with accurate forcing.


	Demonstrated a methodology for generating soil moisture initialisation states.
	Soil moisture assimilation performs best for regions with shallower soils; particularly depths less than 3m.
	Important that all of the dominant physical processes in the LSM are identified and included in the covariance forecasting.
	Structure of the model effects the retrieval time (ie. layered vs catchment deficit).


	What are the defensible requirements of a remote sensing mission for measurement of surface soil moisture?
		Polarisation – horizontal
		Wavelength – L-band
		Look Angle – < 50°
		Observation Accuracy – ?
		Temporal Resolution – ?
		Spatial Resolution – ?


	Typically, observations of surface soil moisture must have an accuracy of better than 5% v/v.
	The greatest impact of temporal resolution was for 1 to 5 days, with greater time between observations having a marginal degradation.
	Observations at one-quarter to one-half the spatial resolution of the model/application is a good compromise.
	To achieve the best results, it is important that the model forcing and observations be unbiased, or the bias modelled as part of the assimilation framework.


	EKF and EnKF both provided satisfactory estimates of soil moisture.
	EKF cheaper (for 1.5D; equivilant to 4 ensembles), but EnKF more accurate for 6 (or more) ensemble members.
	EnKF has the potential to include higher order moments than mean and std dev (ie. skewness)
	EnKF more promising for including horizontal correlations.


	Goal: Provide global land surface initialisation states for the retrospective period of 1979 to present.
	A two-tiered approach.
		Tier 1: Spin-up the land surface model off-line from the GCM using the best atmospheric forcing data available. Minimises forcing bias!
		Tier 2:Assimilate passive microwave remote sensing observations of near-surface soil moisture from SMMR and SSM/I. Minimises model and forcing bias!


Observational Data Sets
	NCAR Northern Hemisphere Sea Level Pressure
		01/1899 – present; 5 x 5 degrees
	Climate Research Unit (University of East Anglia) Temperature and Precipitation
		01/01 - 12/98; 0.5 x 0.5 degrees
	Center for Climatic Research (University of Delaware) Terrestrial Temperature and Precipitation
		01/50 - 12/96 ; 0.5 x 0.5 degree
	Global Precipitation Climatology Project (GPCP)
		01/86 - 03/95; 2.5 x 2.5 degree
	Langley Eight Year Shortwave and Longwave Surface Radiation Budget
		07/83 - 06/91 - in process of being extended; 2.5 x 2.5 degree

Re-Analysis Atmospheric Forcing Data Sets
	ECMWF Re-analysis Advanced Global Data
		4x/day, 01/79 - 12/93
		1.125 degrees (Gaussian)
	NCEP/NCAR Re-analysis
		4x/day, 01/48 – 12/99
		2.5 x 2.5 degrees

Bias Correct Using Monthly Mean Observational Data Sets


	How can we evaluate the soil moisture assimilation?
		Soil moisture – ideal but limited data available. United states has 19 stations in Illinois, 6 stations in Iowa and transect of 89 points in New Mexico for SMMR period.
		Analysis increments – only provides a check for systematic biases.
		Runoff - data available but assumes that soil moisture is the only reason for poor estimates.
		Evapotranspiration – data not available and assumes that soil moisture is the only reason for poor estimates.
		Precipitation forecasts – assumes soil moisture is the only reason for poor forecasts.
		Other ?


	Other assimilation work
		Complete the global SMMR assimilation – Ni et al.
		SMMR/AMSR assimilation Australia – Walker et al.
		Continental snow assimilation – Sun et al.
		TRMM assimilation – Entin et al.
		G-LDAS – Rodell et al.

		Runoff assimilation – Walker et al.