Soil Moisture Assimilation in the NSIPP Land Surface Model

^1,2Jeffrey P. Walker and ²Paul R. Houser

¹Universities Space Research Association

²NASA/Goddard Space Flight Center

Why Assimilate Soil Moisture?

Accurate soil moisture forecasts are necessary for accurate predictions of precipitation.

LSM spin-up does not guarantee correct soil moisture initialization.

Errors in soil moisture forecasts result from:

The memory of errors in LSM soil moisture initialization.

Errors in atmospheric forcing data.

Errors in LSM physics.

Errors in soil and topographic data.

What is Data Assimilation ?

The process of finding the model representation which is most consistent with the observations.

The Kalman-Filter

Forecasting Equations

States:

Covariances:

Observation Equation

Surface Moisture:

Updating Equations

States:

Covariances:

The Catchment-Based LSM

srfexc = srfexc – es – srflow + i + a

rzexc = rzexc – ev + srflow – rzflow + b

catdef = catdef + et – rzflow + baseflow + c

where a, b and c are correction terms to ensure mass balance

Using a first-order Taylor series expansion:

Numerical Experiments

“Truth” Data Set

Spin-up catchment-based LSM for 1987 using ISLSCP forcing data.

Run the simulation for 1987 using ISLSCP forcing data and output surface moisture data as “observations”.

Open Loop Data Set

Degrade the soil moisture prognostic variable spin-up states to make the catchments artificially wet.

Run the simulation for 1987 using ISLSCP forcing data.

Assimilated Data Set

Degrade the soil moisture prognostic variable spin-up states to make the catchments artificially wet (as for the open loop).

Run the simulation for 1987 using ISLSCP forcing data.

Assimilate the surface “observations” once every 3 days using the Kalman-filter.

Soil Moisture: January 30

Soil Moisture: July 31

Soil Moisture: December 29

“Errors” in Assimilated Moisture: 1

“Errors” in Assimilated Moisture: 2

Full covariance forecasting

Partial covariance forecasting

“Errors” in Assimilated Moisture: 3

Full covariance forecasting

Partial covariance forecasting

Monthly Evapotranspiration

Monthly Runoff

Short Term Goals

Identify why the correct soil moisture is retrieved more quickly for some catchments compared to others.

Identify the maximum level of error which surface observations can have before the assimilation is no longer useful.

Identify the minimum frequency of surface observations before there is a significant degradation of the assimilation results.

Longer Term Goals

Assimilate Manfred’s surface soil moisture data over North America and evaluate.

Move towards a global implementation of the assimilation.

Develop a set of global soil moisture initialization states for the start of each month from 1978 to 1994.

Evaluation of Assimilation?

How does NSIPP plan to evaluate the soil moisture assimilation?

Soil moisture - ideal but data not available.

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 ?


	^1,2Jeffrey P. Walker and ²Paul R. Houser

	¹Universities Space Research Association
	²NASA/Goddard Space Flight Center


	Accurate soil moisture forecasts are necessary for accurate predictions of precipitation.
	LSM spin-up does not guarantee correct soil moisture initialization.
	Errors in soil moisture forecasts result from:
		The memory of errors in LSM soil moisture initialization.
		Errors in atmospheric forcing data.
		Errors in LSM physics.
		Errors in soil and topographic data.


	The process of finding the model representation which is most consistent with the observations.


	Forecasting Equations
		States:
		Covariances:

	Observation Equation
		Surface Moisture:

	Updating Equations
		States:
		Covariances:



	srfexc = srfexc – es – srflow + i + a
	rzexc = rzexc – ev + srflow – rzflow + b
	catdef = catdef + et – rzflow + baseflow + c

	where a, b and c are correction terms to ensure mass balance


	Using a first-order Taylor series expansion:


	“Truth” Data Set
		Spin-up catchment-based LSM for 1987 using ISLSCP forcing data.
		Run the simulation for 1987 using ISLSCP forcing data and output surface moisture data as “observations”.
	Open Loop Data Set
		Degrade the soil moisture prognostic variable spin-up states to make the catchments artificially wet.
		Run the simulation for 1987 using ISLSCP forcing data.
	Assimilated Data Set
		Degrade the soil moisture prognostic variable spin-up states to make the catchments artificially wet (as for the open loop).
		Run the simulation for 1987 using ISLSCP forcing data.
		Assimilate the surface “observations” once every 3 days using the Kalman-filter.


	Identify why the correct soil moisture is retrieved more quickly for some catchments compared to others.
	Identify the maximum level of error which surface observations can have before the assimilation is no longer useful.
	Identify the minimum frequency of surface observations before there is a significant degradation of the assimilation results.


	Assimilate Manfred’s surface soil moisture data over North America and evaluate.
	Move towards a global implementation of the assimilation.
	Develop a set of global soil moisture initialization states for the start of each month from 1978 to 1994.


	How does NSIPP plan to evaluate the soil moisture assimilation?
		Soil moisture - ideal but data not available.
		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 ?